Download la Nette rousse Netta rufina

UNIVERSITE PAUL SABATIER – TOULOUSE III
École Doctorale SEVAB
Sciences Ecologiques Vétérinaires Agronomiques Bioingénierie
THESE DE DOCTORAT
Spécialité : Ecologie
Présentée par :
Pierre DEFOS du RAU
Pour obtenir le titre de docteur de l’Université Toulouse III
Ecologie, démographie et conservation d’une espèce gibier
rare : la Nette rousse Netta rufina
Thèse dirigée par :
Emmanuelle CAM & Sovan LEK
Jury :
Régis Céréghino
Président du Jury
Emmanuelle Cam
Directrice de thèse
James D. Nichols
Rapporteur
Juan A. Amat
Rapporteur
Christophe Barbraud
Examinateur
Jean-Yves Mondain-Monval
Examinateur
Laboratoire Evolution et Diversité Biologique (EDB) – UMR 5174 CNRS/ UPS
Université Paul Sabatier 31062 Toulouse
REMERCIEMENTS
(Où l’on relance good vibration)
Emmanuelle Cam m’a guidé avec soin et générosité. Elle m’a fait découvrir beaucoup de
concepts abstraits, parfois contre-intuitifs et souvent compliqués (exemple : « pour
répondre à une question, il faut des données, beaucoup de données, n’oublie jamais ça,
petit ! »). Elle aura été particulièrement patiente et attentionnée dans son rôle
d’encadrante et aura profondément influencé mon approche du métier de biologiste de
la faune sauvage.
Jean-Yves Mondain-Monval est à l’origine de nombreuses idées développées, testées ou
envisagées dans ce travail qui est donc aussi très largement le sien (à part le coup des
canards sarmatiques qu’il m’envie). Il m’a fourni et appris mon métier, y compris dans
diverses circonstances à haute pénibilité aux Basses-Méjanes, dans le Delta de l’Ebre ou
à table devant une foulque au café arrosée d’un Coto Grindul Lupilor probablement d’une
très mauvaise année…je lui dois beaucoup plus que de l’or pour les faux-frais à bord.
Christophe Barbraud est aussi à l’origine de beaucoup des idées méthodologiques les
plus fouillées de ce travail. Avec une gourmandise roborative pour la modélisation, il m’a
fait découvrir la vraie et plutôt sobre vie de la faune sauvage expliquée (gouvernée) par
Mark, à moi qui croyais qu’elle vivait en liberté.
Juan Amat et Jim Nichols m’ont fait le fastueux honneur de juger ce travail. Ils étaient
et sont toujours des modèles inaccessibles. Ils m’évoquent l’un les grandes étendues
sauvages et méditerranéennes d’Espagne où bégaie l’Engoulevent à collier roux, l’autre le
blues du CCR et les maximum de vraisemblance du CMR d’Amérique où tonitruent Louis
Armstrong, Jimmy Rushing et Jim Mc Clure.
Sovan Lek m’a accueilli au SEVAB avec enthousiasme et énergie tout en me laissant une
grande et agréable liberté et en me faisant une confiance flatteuse qu’il a pu parfois
passagèrement regretter
Régis Céréghino a accepté d’être président du jury avec une immédiate et sympathique
disponibilité
Nicolas Sadoul m’a fait faire des découvertes bouleversantes et souvent rigolotes
comme la magnétique carte des salins, les laro-limicoles, le chocolat devant les V
nations, le douloureux concept de l’imperfection de détection, le jeu télévisuel du fil
rouge en milieu laguno-marin et Christophe Thébaud
Christophe Thébaud a favorisé et accompagné, à 2 reprises et avec tonicité et
enthousiasme, mes premiers pas de doctorant en rendant possible une idée saugrenue
qui devait être l’affaire de 2 ans (tout au plus)
Jean Leduc m’a fait confiance dans le démarrage semi-professionnel de ce travail
Nathalie Hecker, Jean-Laurent Lucchesi, Greg Massez, Nicolas Beck, Cyril Girard,
Matthieu Chambouleyron et tous les amis des amis des Marais du Vigueirat pour l’accès
privilégié à leur replète base de données et à leurs somptueux marais où foisonnent les
nettes rousses et tout plein de biodiversité camarguaise
Mes vieux copaings Yves Kayser, Eric Didner, Marc Thibault, Arnaud Béchet, PierreAndré Crochet, Nico Vincent-Martin, Kiko, Laurent Tatin, Olivier Pineau, Nicolas Sadoul,
Emmanuel Vialet, Tau Bino et tous les ornithos camarguais pour avoir soigneusement
loggé de la toute bonne data malgré l'absence d'objectif, ce qui démontre qu'un
objectif n'est pas toujours nécessaire pour faire de la recherche (c'est peut-être le
résultat le plus intéressant de cette thèse)
Mes co-auteurs roumains, catalans et camarguais: avec eux, tout s’est avéré possibil
Les pionniers Camarguais Alan Johnson, Heinz Hafner, Hubert Kowalski et John
Walmsley pour avoir marqué les esprits, l'histoire camarguaise et les canards
et Heinz pour m'avoir autorisé à en utiliser les données
Luc Hoffmann et Jean-Paul Taris, m’ont ouvert, comme à beaucoup d’autres, les portes
de la Tour du Valat et donc dans une grande mesure de la Camargue, toutes deux, cadre
idéal de formation professionnelle, de birding de classe internationale et de maturation
lente de ce travail.
Marc Thibault, Olivier Pineau et Jean-Marc Sinnassamy m’ont fait découvrir la gestion
des zones humides camarguaises, activité intégrée, participative et surtout compliquée
dévolue au créatif (HAI gardian worldbirder serait un plus) et non pas, comme je le
croyais, à la gravure de mode pour plaquette du Conseil Général tripotant une
martellière ou révélant à une assemblée d’acteurs locaux cynégétiques et médusés
l’existence du cadre logique.
Sébastien Cayuela, Michel Lepley, Julien Travers et Mickaël Veillé auront été de rudes
et autonomes clampins ainsi que d’habiles et zélés récolteurs de données de terrain.
Marc Thibault pour les belles photos et les érudites et, hélas depuis trop longtemps,
stériles précisions sur la phénologie des phylloscopus en Camargue
Maurice Benmergui pour m’avoir généreusement permis de parasiter ses documents
photographiques rares et nombreux sur le parasitisme inter-spécifique de la nette
L’ONCFS et plus particulièrement, Marie Fuziès, François Lamarque, Jean-Yves
Mondain-Monval, Jean Leduc, Gérard Ruven, Jean-Marie Boutin, Pierre Migot et JeanMarc Cugnasse pour avoir soutenu ce travail.
François Lamarque et Jean-Yves Mondain-Monval auront perdu temps précieux, patience
et crédit à défendre mes droits acquis (que je les soupçonne d’appeler entre eux « mes
caprices de diva ») mais ils y auront gagné mon éternelle reconnaissance (les veinards)
Manu Ménoni, incarnation alpino-arapaho du wildlifer à la française, aura été le
compagnon marquant (à vie) de saucissonnades érudites en refuge et de galopades
émerveillées en altitude
Serge Planton de Météo France et Olivier Devineau du CEFE m’auront apporté avec
beaucoup de patience, de gentillesse et une relative et flatteuse exclusivité, des
données cruciales et une expertise précieuse.
Comme pour beaucoup, Jacques Blondel aura été un inspirateur et un modèle.
Enfin, Jean-Dominique Lebreton aura fait preuve d’une indulgence de bonze tibétain
(niveau grand débutant) pour m’aider à passer le délicat obstacle du DEA
J’espère avoir détecté toutes les amis, collègues et parents qui m’ont aidé ; pour en être
certain, il conviendrait de répliquer mon échantillonnage c’est à dire de faire une 2nde
thèse ; les oubliés auront dont l’indulgence de bien vouloir me pardonner et surtout de
me contacter d’un ton acerbe
Mes parents, parfois légèrement incrédules mais surtout formidablement généreux et
attentionnés, m’auront toujours soutenu vers ce métier et ce diplôme d’ornithologue
Lydie, Jeanne et Claire ont pris en charge avec bonne et douce grâce toute la partie
non-doctorante de l’univers
A mes grands-pères explorateurs
A mes grands-mères, femmes d’explorateurs
Summary
CHAPTER 1 - CONSERVATION BIOLOGY AND EXPLOITATION OF A RARE GAME
WATERBIRD: THE RED-CRESTED POCHARD
p.6
CHAPTER 2 - ESTIMATING BREEDING POPULATION SIZE OF THE RED-CRESTED
POCHARD (NETTA RUFINA) IN THE CAMARGUE (SOUTHERN FRANCE) TAKING
INTO ACCOUNT DETECTION PROBABILITY: IMPLICATIONS FOR
CONSERVATION
p.36
CHAPTER 3 - INCORPORATING UNCERTAINTY INTO ANALYSES OF REDCRESTED POCHARD HABITAT SELECTION
p.44
CHAPTER 4 - SOME ASPECTS OF RED-CRESTED POCHARD MACROECOLOGY
AND HABITAT USE IN EUROPE: IMPACT OF DETECTION ISSUES
p.58
CHAPTER 5 - PHYLOGEOGRAPHY OF A GAME SPECIES: THE RED-CRESTED
POCHARD (NETTA RUFINA) AND CONSEQUENCES FOR ITS MANAGEMENT p.108
CHAPTER 6 - DEMOGRAPHY AND HARVESTING SUSTAINABILITY OF A RARE
GAME WATERBIRD: THE RED-CRESTED POCHARD IN EUROPE AND THE
CAMARGUE (FRANCE)
p.120
French summary
p.167
Chapter 1
CONSERVATION BIOLOGY AND EXPLOITATION OF A RARE GAME WATERBIRD:
THE RED-CRESTED POCHARD
The Red-crested Pochard (Netta rufina), is a monotypic Palearctic duck species (Cramp &
Simmons, 1977) of Sarmatic origin (Voous, 1960). It breeds mostly between latitudes 35° and
55° North, in continental, temperate and Mediterranean climatic regions, from the British
Isles to China (Scott & Rose, 1996). It is a moderate-distance migrant, wintering mainly
between latitudes 30° and 50° North (Figure 1). Mayaud (1966), Schneider-Jacoby and Vasic
(1989), and Tamisier (1991) have reviewed the history and biogeography of the species in
Europe.
The central and western Asian population, which concentrates most of the world population
of the species, does not show any clear trend in variation in numbers and is thus assumed to
be stable (Delany & Scott 2006). The Black Sea and Eastern Mediterranean population is
thought to be declining since the 1980’s (Paspaleva et al., 1984; Krivenko, 1994; Scott &
Rose, 1996; Delany & Scott 2006). Among European game waterbirds, the Red-crested
Pochard (RCP) is probably one of the less abundant species and certainly the rarest breeding
game waterbird in southern Europe. Numbers are now considered to be increasing in Europe
(Delany & Scott 2006), after the species has been qualified as having an unfavourable
conservation status in Europe (Tucker & Heath 1994). This concern was based upon the fact
that Red-crested Pochard populations were thought to be declining in several European
countries, especially in Eastern Europe, which previously held some of the largest populations
(Krivenko 1994, Scott & Rose 1996). The Red-crested Pochard is listed on Annex II/2 of the
EU Birds Directive as a species for which hunting is permitted only in France, Portugal and
Spain. The annual hunting bag in the European Union was roughly estimated at 8000 birds by
Shedden in 1986, 700 of which in France and the rest in Spain.
Because of exploitation, the species unclear conservation status and the overall small
population size in Europe compared to other game species of ducks, it was important to assess
potential needs in conservation and management actions. The combined issues of
conservation and exploitation of this species obviously raised various data-demanding
questions. It was necessary to evaluate both (i) the current knowledge available on the species
conservation biology and (ii) the information still lacking to develop a preliminary
conservation and management strategy, at least at the French level.
Information available
The information already available on the species was scanned through published literature ;
we made a tentatively exhaustive survey of the most significant papers published in Western
Europe before 2001 on the species. We identified the main scientific topics of the 87 papers
found (Appendix): they roughly fall into four general categories : behaviour, demography,
ecology and distribution and related topics (Figure 1).
There were several studies of two RCP specific behaviours :
•
Migration and wintering pattern in Western Europe included an Autumn northerly
migration, contrary to general pattern in birds (Szijj 1975, Lucientes 1978, Amat et al.
1987).
•
Relatively high inter- and intra-specific brood parasitism frequency (Hellebrekers &
Voous 1964, Amat 1985, Amat 1991).
As a rare waterbird species, RCP breeding distribution generated a strong interest in Europe,
where breeding remains very localized outside Spain (Figure 2 after Scott & Rose 1996, see
also Berndt 1997, Snow & Perrins 1998). We included atlases, monographs and conservation
reviews in the category of distribution-related papers, hence the quite large number of articles
found.
ecology
demography
behaviour
distributionconservation
0
10
20
30
40
50
Figure 1 : number of articles published in European journals on corresponding topic on RCP
before 2001
Both RCP ecology and demography were the least studied topics.
•
Among the 14 papers on demography, 4 focused on reproductive processes and the
remaining 9 dealt only with regional or local patterns, i.e. trends in numbers over
space and/or time. There was a complete lack of data on survival, as well as on most
components of recruitment processes. Only one paper by Lebreton & Rochette (1965)
marginally dealt with harvest level of RCP in one wetland area in Eastern France.
Some ringing schemes included RCP (Hückler 1966, Johnson 1975, Schlenker 1979)
but were not used to estimate survival nor dispersal rates.
•
Among the 10 papers on ecology, most dealt with feeding ecology or diet and only 1
(Broyer & Daléry 2000) quantitatively focused on other factors of breeding habitat.
More data on breeding habitat characterisation were thus needed because it is a crucial
step (Batt et al. 1992, Elmberg et al. 2005 for a recent European example) in
understanding drivers of reproductive success, hence demographic processes.
Furthermore, in a conservation-oriented context, identifying limiting factors in
breeding habitat is a prerequisite to setting up an efficient habitat management
strategy.
•
Lastly, there was no information at all on RCP population genetics and structure.
Figure 2 : RCP population limits as proposed by Scott & Rose 1996
Information lacking
Information to collect or update on the species concerned mainly habitat ecology, genetics
and demography. Research efforts were then fitted to conservation and management
objectives accordingly. (i) Identifying limiting factors in breeding habitat, (ii) identifying
Western Europe population structure and management unit(s), (iii) and evaluating hunting
impact and sustainability were assessed as the 3 priority research actions to be undertaken to
establish a preliminary basis for RCP conservation and management plan. The first two of
these priorities were in general accordance with Long et al. (2007), who recently stressed the
importance of population size, range size and wetland losses as main drivers of Anseriformes
conservation status and declines.
Identifying limiting factors in breeding habitat and evaluating hunting impact and
sustainability required various types of bird count data. Understanding habitat use patterns
requires comparison of bird occurrence or abundance in time or, in our case, space between
habitat patches or local populations for example (Block & Brennan 1993). Calibrating models
of exploited population dynamics also requires counts to compute population growth or
harvest rates. Quantifying RCP occurrence and abundance was thus a pre-requisite of the next
steps of this work. However this preliminary census question is not a trivial one, as
emphasized by Pollock et al. (2002), MacKenzie et al. (2006), McLoughlin & Messier (2004)
and Barry & Elith (2006).
The first chapter of the present work was thus dedicated to the census question (Defos du Rau
et al. 2003) before considering the three identified research priorities, namely:
•
identifying limiting factors in breeding habitat at both local and continental scales:
chapter 2 (Defos du Rau et al. 2005) and 3,
•
identifying Western Europe population structure and management unit(s) : chapter 4
(Gay et al. 2004),
•
evaluating hunting impact and sustainability : chapter 5.
Here we review most of the methodological options chosen to addresse the above questions
and their limitations. We then provide a summary of the results obtained to these questions
and, in a third step, we assess research and monitoring recommendations for RCP
conservation and sustainable management.
MAIN METHODOLOGIES USED IN THE NEXT CHAPTERS AND THEIR LIMITATIONS
Most methodological tools used in this work to estimate demographic parameters, presence
and abundance of RCP belong to the corpus of Capture-Mark-Recapture models (Williams et
al. 2002). These models rely on some general assumptions:
- representativeness
- absence of mark loss
- independence of fates and identity of rates among individuals within cohorts
- instantaneous and discrete sampling occasions
Except for the last assumption for our dead-recovery model, we had no particular reason to
believe that any of these assumptions was not met in our study or, at worse, less so than in
other similar studies. We performed Goodness-Of-Fit tests whenever available for the models
used. Where necessary, we discuss these general assumptions and some more design-specific
ones under each CMR design used in this work
Double observer (chapter 1 & 2)
This method (Nichols et al. 2000) was used to assess false-absence of RCP broods in the
sampled wetlands in the Camargue in year 2001. We assumed that results from this study
were representative of the overall species-specific detection probability for RCP brood as a
species. Three assumptions are required for this method
- independance of detection performance among observers
- equality of detection performance among observer roles
- control of effect of distance on detection probability
Considering that we sampled only wildfowl broods (5 species) and avian duck-predator (4
species) as members of a relatively poor community, during sampling sessions observers did
not write down observations during point-counts but only after point-count so as not to give
cues to the primary observer and not to weaken performance of secondary observer. This way,
we increased adequation of our design to the first two assumptions. This was only possible
because of the relatively low number of species and individuals to memorize during pointcounts. The third assumption was also taken into consideration because our point-counts only
concerned large birds over water areas, i.e. areas that were delimited in space and relatively
small, averaging 10.3 ha (SD = 16.2).
Resulting detection probability estimate was very high (0.9259 by sampling occasion; SE =
0.1033), which indicates in practice a zero-risk of false-absence (0.04%) in the Camargue. In
contrast another CMR method employed, patch occupancy analysis, estimated a false-absence
probability of 3.5% over our European sample including Ebro, Camargue and Danube deltas.
Deciding whether these results are consistent is complicated by issues of differences in
sampling areas and detectability estimation methods. Double-observer method only estimates
detection probability given availability to detection (Farnsworth et al. 2005); it therefore
provides detection rate estimate that may not be directly comparable to other CMR-based
estimates. It was furthermore likely that very large water areas to be sampled in the Danube
Delta combined with low accessibility of the sampled sites increased mean risk of falseabsence for the European dataset compared to the Camargue dataset alone.
Robust design (chapter 1, 2 & 3)
This method (Kendall et al. 1997) was used to estimate brood-specific detection probability in
the sampled wetlands in the Camargue in year 2001. This individual detection probability was
then used to estimate brood abundance in the Camargue, as such, and as a double sampling
design (Pollock et al. 2002) to predict brood abundance in the two other European deltas. Two
assumptions are required for this method in addition to the general ones :
- sampled population is closed between secondary occasions within each primary occasion
- capture probability variation can be accounted for by a combination of time, behaviour or
individual heterogeneity
To ensure closure of the “populations” of broods within each primary period, time intervals
between secondary occasions were kept below a maximum 45 minutes, which made risk of
brood migration highly unlikely. We assessed the need for accounting for time, behaviour and
individual heterogeneity in capture probability using program CAPTURE (Rexstad and
Burnham 1991). These results were then taken into account to estimate brood abundance
using Robust-design models. As a non-invasive CMR application, no behavioural effect were
to be expected on capture probability but time and individual effects were worth checking.
Constant capture (detection) probability was the model that received the largest support. We
were not interested in questions about individual variation in of capture probability per se, but
this source of variation had to be taken into account to estimate brood abundance using the
Robust Design In addition, model selection provided evidence of time-independence of brood
capture probability.
The assumption of absence of mark loss was a concern because another phenomenon may
have had an effect equivalent to tag loss in our non-invasive design. RCP broods were
individually recognized by the quasi-unique combination of their observed size, estimated age
and location. However there was a non-zero risk of having two broods of the same size and
age on the same wetland (or assessed as having the same size and age by observers): this
would have be equivalent to an undetected loss of individual mark. Difficulties in individual
identification (“marking”) of the different broods may have resulted in underestimation of the
brood abundance estimate.
This would have happened if two broods from the same lake could not be reliably separated
on the basis of their age and size. Broods hatching date and size approximately followed
normal distributions (Kolmogorov-Smirnov d= 0.084 and 0.118, both non significant).
Consequently, difference in hatching date and size of two different broods also followed
normal distributions of known mean and variance. If p1 is the probability for two broods of
having the same age within 2 weeks, p2 the probability for two broods of having the same size
within 2 pulli; then the probability p for two broods of having the same age and size on the
same lake is:
p1=0.397
p2=0.274
p=
p1 * p2
n2
p<0.0001 for n=33
Thus, mistaking one brood for one another was a particularly unlikely event, and we believe
that we did not underestimate brood abundance..
Patch occupancy analysis (chapter 3)
This method (MacKenzie et al. 2002) was used to estimate wetland-specific occupancy rate in
the wetlands sampled in the three European deltas in 2001. Occupancy rate was then
modelled as functions of habitat covariates to understand brood-rearing habitat preference in
RCP. Two assumptions are required for this method in addition to the general ones :
- sampled sites are closed to change in occupancy
- no false presence
To achieve minimum risk of falsely recorded presence, the field survey was performed by a
small number of expert field ornithologists : 1 in Danube Delta, 1 in Ebro Delta and 2 in
Camargue who co-checked observations and skills in a double-observer design. Closure of
sites was of stronger concern because some other studies have shown that duck broods are
mobile and can change wetland during rearing period (e.g. Pöysä & Paasivaara 2006). We
furthermore showed that temporary emigration was high for RCP broods in the Camargue
(Defos du Rau et al. 2003). However, patch occupancy analysis, as developed by MacKenzie
et al. (2002), is robust to violation of the closure assumption, provided that movement of
individuals under study is random and that occupancy can be reinterpreted as use (MacKenzie
et al. 2006). Brood-rearing habitat use was indeed our primary interest in habitat analyses.
This possible occupancy closure violation nevertheless probably remained marginal in our
European dataset because of the following reasons
- it did not concern those wetlands used by a number of broods, where occupancy is probably
permanent despite the fact that some individual broods emigrate
- it did not concern larger or isolated wetlands, emigration from which was unlikely because
impracticable for duck broods
- it only concerned small wetlands with small numbers of broods whose movement was more
likely to correspond to temporary stays in riverine vegetation of the original pool than an
actual change of pool (Defos du Rau et al. 2003).
We validated conclusions of this patch occupancy analysis and ensure their representativeness
using an different data set (test set) of randomly sampled occupancy data.
Multistate (chapter 5)
This model (Williams et al. 2002) was used to estimate RCP brood and duckling survival in
the wetlands surveyed in the Camargue in years 1990, 1991, 2000 and 2001. We modelled the
probability for a brood of size n+1 to make a transition to size n (loss of one chick), as well as
brood size-specific survival probability (where death is the loss of the last chick). These
probabilities were then modelled as function of reproductive traits covariates (e.g., laying
date) to understand brood and duckling survival processes in RCP. These estimates were
ultimately used to model fecundity in a matrix population projection model of RCP. Two
assumptions underly the method we chose (no memory model) in addition to the general
ones :
- homogeneity of survival probability in individuals within each state
- Markovian state transition probability
We tried to account for as many sources of heterogeneity among individual in survival and
transition probability by using several covariates and their quadratic effects, that proved
pertinent in the model selection process. Assumption of Markovian process governing state
transition was assumed fulfilled as this transition corresponded to a one-chick loss. Only
broods of size n+1 can undergo a one-chick loss transition to size n. However, broods of size
n+2 could loose one single chicks twice, with probability of the 2nd chick lost being
influenced by the probability of the 1st chick lost. Such process would probably not qualify as
markovian. The reverse transition of gaining one chick was impossible and thus the
probability of the corresponding event was set at zero.
Dead recoveries (chapter 5)
This model (Williams et al. 2002) was used to estimate RCP age-specific annual survival
rates in the Camargue from 1952 to 1978. Survival probability was then modelled as
functions of individual and time covariates to address sources of variation in annual survival
in RCP. Estimates were ultimately used to model survival in a matrix population projection
model of RCP. Two assumptions are required for this method in addition to the general ones :
- correct determination of age
- absence of banding effect on survival
Band loss was a concern here, but not in the above CMR sampling designs because they were
non-invasive and did not require to capture and tag the individuals. Metal-banding of RCP
probably suffered from the same band loss rates (as well as the same detrimental effect of
metal leg band on survival) than in other similar studies (e.g. Pradel et al. 1997 on teals Anas
crecca banded under the exact same design as RCP)
There was no particular reason to suspect above average non-randomness of trapping or errors
in age determination as the whole field protocol was performed on the long-term by the same
team of expert field-ornithologists.
The assumption of instantaneous and discrete sampling occasions was a major concern in this
dead-recoveries dataset because capture and recovery sampling were almost continuous along
the whole year. Because of the small sample size (41 recoveries among 269 captures), subsampling only data points (capture and recoveries) from a restricted number of sampling
occasions conforming to a discrete sampling scheme was not an option. Instead, for each
recovered bird we calculated the difference in month between the month of the realized
recovery and the month of the initial capture, at the year’s scale. We defined discrete
theoretical recovery occasion for each bird as occuring on the same month as the initial
capture occasion, but of course following the corresponding integer number of years. The
value assigned to each recovered bird corresponds to the difference in months at the year’s
scale between the theoretical recovery and the actual realized recovery. We assumed that this
time-increment individual covariate adjusted for the numerous discrepancies between capture
and recovery occasions in our dataset. The corollary was that this time-increment individual
covariate was expected to have a negative effect on survival, which we checked for using the
dead-recovery models of RCP annual survival with the largest support according to AICc
(Anderson et al. 2001).
A major general limitation in the CMR models listed above was small sample size because of
the general rarity of the species in Europe.
Deciding when to stop collecting data from rare (and sometimes more common) species
depends on a trade-off to achieve between survey costs and efficiency, including reliability of
data-based inference (see Franco et al. 2007 for a preliminary discussion on this vast topic).
The risk is either (i) to base conservation and management actions on fragile, tentative or not
fully corroborated conclusions; this is less costly but leads to higher probability of costly
failure of the actions taken or (ii) to delay conservation and management actions until
sufficient data have been collected, which may often prove economically and politically
costly but lead to lower risk of costly failure. We atttempted to achieve the best costefficiency trade-off in the present study of RCP by relying on the following course of action
-
we aimed in priority at identifying further research needs with increased precision in
order to increase the amount of information available on RCP (lack of information is
one of the current major issues for defining conservation and management
recommendations for RCP)
-
we aimed at specifying low-cost or low-risk conservation and management
recommendations which would be likely to have positive outcomes in terms of
indigenous biodiversity (Koper & Schmiegelow 2006)
In addition, some of our objectives required use of other methodologies relying on specific
assumptions, such as genetics (chapter 4) and meteorology (chapter 5).
Estimate of gene flow between RCP population
The traditional FST estimation of inter-population gene flow relied on the island model
assumptions requiring equal population sizes and symmetric migration rates (Rousset 2001).
This assumption was most probably not met in RCP. To achieve inferences on gene flow, a
combination of several other methods including a coalescence based maximum-likelihood
method, an analysis of ringing recoveries and a morphometric analysis provided a body of
complementary results.
Meteorological-based predictions
Several of the data sets from RCP we used were collected in the past, whereas one of our
objectives was to assess the current state of RCP sub-populations and their viability in the
future under alternative management and conservation scenarii. To do so, we used
meteorological data and results of meteorological models. We addressed the influence of
climatic variables on several demographic parameters using currently available data from
RCP and used the resulting relationships to project population size as functions of climatic
variables available for other periods of time. Models of functional links between
meteorological factors and demographic rates relied on the strong, risky assumption that an
observed correlation allows for a causal/functional interpretation of the involved process. To
maximise reliability of our inferences in this context, we paid a strong attention to the
following steps (Krebs & Berteaux 2006) :
- we avoided large-scale inferences and only used local datasets to draw inferences at the local
scale
- we based our long-term prediction on the best possible meteorological predicted dataset
generated by the French state meteorological agency (Météo-France) under moderate climate
warming ARPEGE scenario
- we made only 3 simple a priori functional hypotheses based on already inferred patterns to
assess demographic-meteorological links:
-
RCP annual survival is positively linked to mean winter temperature (Blums et al.
2002)
-
RCP breeding performance is positively linked to high water levels hence high
precipitation during preceding winter (Defos du Rau et al. 2005)
-
RCP breeding performance is positively linked to abundance of aquatic macrophytes
of temporary wetlands and thus to low precipitation during spring of the previous year
(Defos du Rau et al. 2005)
- we only tested these 3 hypotheses to minimise risk of type-I errors
- these 3 hypotheses proved immediately conclusive so we had no reason to further carry on
hypothesis-testing.
Finally we increased robustness of our inferences through use of external, independent test
datasets and correction for detection imperfection
Test of estimation models with independent data sets
Robustness of the main conclusions was increased by use of external test datasets: We
addressed the different issues in this work using a sequential approach. For each issue, we
first used the empirical data available to estimate relevant parameters (e.g. demographic
parameters, abundance, presence of the species). We addressed the relationship between
environmental or habitat variables and the parameters in question (e.g., climatic variables and
brood survival, habitat variables and abundance, etc.). We then used the results from this step
to model population dynamics, or local abundance, and, where possible, used independent
data sets to check consistency with the predicted population features.
- Camargue habitat-occupancy model was validated by an independent random sample of
sites (chapter 2)
- European habitat-occupancy model was validated by an independent random sample of sites
(chapter 3)
- population phylogenetic structure was corroborated by a morphometric dataset (chapter 4)
- demographic matrix models were validated by two external census datasets (chapter 5)
Imperfect detection of RCP broods
A central feature in this work was the rejection of the a priori assumption of constant, perfect
detection probability of RCP broods. As in the vast majority of ecological studies, we used
measures of occupancy and abundance of RCP broods and addressed their variation in space
and time. However, contrary to many studies, we did not rely on the assumption that our
detection performance was satisfactory and constant. Hence, although this work relies on
several assumptions, we did not rely on one major assumption which is overlooked in most of
other studies, despite the serious repercussions if it is not met (Nichols et al. 2000, MacKenzie
et al. 2002, Pollock et al. 2002). In this context of imperfect detection, we followed the
recommendation by Kéry (2004) and used Capture-Mark-Recapture methodology.
SUMMARY OF THE CONCLUSIONS REACHED IN THIS WORK
The main results obtained concerning the following three topics are reviewed here
- census of RCP breeding population
- identification of limiting factors in breeding habitat at both local and continental scales
- evaluation of hunting impact and sustainability in the Western European management
unit(s), as defined by the preliminary genetic study
Census of RCP breeding population (chapter 1)
According to Dehorter & Rocamora (1999) and Krivenko (1994), RCP breeding populations
in both the Camargue and the Danube Delta are thought to be declining
Camargue breeding population
As a result of cryptic behaviour of broods, RCP breeding population in the Camargue was
found to be largely under-estimated when it was censused without adjustment for
detectability. However, detectability-unadjusted survey results were those taken into account
to draw conservation status and recommendations at national level (Dehorter & Rocamora
1999). When based upon detectability-adjusted census data, conclusions drawn on RCP
conservation status proved less alarming. Hence, in a resource-limited context, conservation
investment of all kind on RCP could be decreased to the benefit of other more threatened
species.
Danube delta breeding population
RCP is believed to have first bred in the Danube Delta around 1955 (Linţia 1955). Although
there might be substantial uncertainty on this date, it is possible to assume that Danube Delta
population have grown at the same rate as the Camargue population, i.e. λ=1.06. With such a
growth rate, the projected RCP population size in Danube Delta would be 46 birds in 2001. In
fact, it is most certainly much higher. Indeed it was estimated at a minimum of 400 breeding
pairs in the same period by Kiss (comm. pers.) and it would be almost at 700 breeding pairs as
calculated according to the habitat area-expansion of the detectability-adjusted density
calculated with our dataset (analysed in chapter 3). Therefore, contrary to what would then
appears to be unfounded statement (Krivenko 1994, Scott & Rose 1996), it is possible that
Danube Delta breeding population of RCP is not declining. It might even be increasing
following appearance of the species as a breeding bird in the middle of the 20th Century.
Long-term representative field census, including aerial surveys are required to assess the
demographic and conservation status of the Danube Delta RCP population.
Conservation perspective on the necessary use of detection probabilities
It is furthermore likely that not only RCP, but also several (if not most) waterbird species are
imperfectly detected by observers during sampling sessions. Therefore, conclusions drawn on
conservation status, trends and recommendations for these species probably suffer from
uncertainty and biases due to unadjustement of their census to detection imperfection. A costeffective solution would be to develop double-sampling design protocoles (Pollock et al.
2002) on threatened waterbird species (see Green 1996 and Long et al. 2007 for reviews) that
are monitored under various schemes including the International Waterbird Census
coordinated by Wetlands International (Delany & Scott 2006).
Objectives of such double-sampling design would be to :
- gain insights into detectability variations in time and space of these threatened species
- increase precision and understanding of waterbird population estimates and trends by
reducing uncertainty in current time-series models
- identify those threatened waterbird species with low detectability, and thus the potential for
substantial underestimation of population size . Such species might qualify to a less
threatened status, thus raise reduced conservation concern and deserve reduced conservation
investment compared to more threatened species
- alternatively, identify those threatened waterbird species with high detectability and thus
satisfactory population size estimation that should indeed be a target of conservation concern
and investment.
Identification of limiting factors in breeding habitat : habitat management for
conservation (chapter 2 and 3)
A number of habitat factors proved to influence on RCP brood occurrence and abundance at
both Camargue and European scales. These habitat analyses provided scope for management
recommendation of RCP breeding habitat. However, the range of identified factors and thus
of potential management recommendations differed whether detection imperfection was taken
into account, or not.
Consequence of adjusting species-habitat models for detection imperfection
Table 1 compares our recommendations for RCP breeding habitat management and those
recommendations that could have been made without having adjusted for detection
imperfection.
Camargue
adjusted occupancy
unadjusted occupancy
Water
favour constant high
favour constant high
favour high water
no conclusion on this
management
water level in spring
water level in spring
level in spring
aspect
Water
management
adjusted abundance unadjusted abundance
avoid regular artificial avoid regular artificial avoid regular artificial no conclusion on this
flooding in summer
habitat
flooding in summer
maintain large
management
habitat
flooding in summer
aspect
no conclusion on this no conclusion on this
reedbeds
maintain large reedbeds
aspect
aspect
avoid wetland
avoid wetland
avoid wetland
no conclusion on this
reclamation/dyking
reclamation/dyking
aspect
fragmentation reclamation/dyking
Europe
adjusted occupancy
unadjusted occupancy
adjusted abundance unadjusted abundance
Water
avoid constant water
avoid constant water
avoid constant water no conclusion on this
level all year long
level all year long
management
Water
management
habitat
level all year long
aspect
avoid regular artificial no conclusion on this avoid regular artificialavoid regular artificial
flooding in summer
aspect
flooding in summer
avoid wetland dyking avoid wetland dyking avoid wetland dyking
management
and enbankment
and enbankment
habitat
avoid habitat
no conclusion on this
management
reclamation
aspect
avoid ecosystem
no conclusion on this
avoid ecosystem
simplication
aspect
simplification
competition
and enbankment
flooding in summer
avoid wetland
fragmentation
reduce swan impact
Table 1 : Main breeding site management recommendations raised from habitat analyses
performed in this work ; discrepancies between detection-adjusted and –unadjusted inferences
are highlighted
Major habitat features of management concern for RCP would not have been identified
without taking detection uncertainty of RCP broods into account. For instance, the
question « is there breeding habitat loss ? » has crucial implications for the RCP breeding
populations of Camargue and Ebro deltas as well as for the associated wetland biodiversity.
This question received a negative answer when detection imperfection was not accounted for
but a positive one when accounting for detection issues. Without accounting for detection, it
might for instance have been concluded that there was no habitat conservation problem but
that some swan removing would be useful to maintain or increase RCP breeding abundance.
These kind of erroneous conclusions based on usual species-habitat modelling methods would
have three main direct consequences
- absence of the desired deterministic effect of the corresponding management action on RCP
abundance
- useless management and evaluation costs
- delay or difficulties, possibly aggravating/detrimental, in identification of factors with a
genuine influence on presence and abundance of RCP
It can be feared that some of the habitat management recommendations produced for the
numerous cryptic species whose ecology has been studied on the basis of detectionunadjusted species-habitat models are ineffective.
Management recommendations for RCP breeding habitat
From table 1, management actions that we recommended for conservation of the RCP
breeding habitat can be summarized in four points
(i) to ban natural habitat destruction or reclamation : destruction or transformation of pools,
lakes, marshes or reedbeds and any other wetland vegetation type would contribute to
simplify and fragment the ecosystem and thus reduce the waterbird community richness and
decrease RCP occupancy or abundance
(ii) to ban embankment and dykes in wetland shaping and management, as it also contributes
to wetland fragmentation and artificial simplification of ecosystem (Aznar et al. 2003)
(iii) to favour some natural hydrological functioning including avoidance of artificial flooding
in summer and artificial drying off in spring
(iv) to evaluate management decision (or to model it through an objective function) using not
only RCP occurrence and abundance but also waterbird species richness as system state
criteria in order to ensure that the implemented management actions are not detrimental to
other taxa in the ecosystem under management (Koper & Schmiegelow 2006). Furthemore,
we showed that waterbird species richness is positively linked with RCP occupancy and
abundance. Measuring this variable as a system state is thus consistent with evaluation of
management actions for both RCP and wetland biodiversity.
Overall these recommendations call for an increased consideration of natural processes and
functioning of Mediterranean wetland ecosystems in their management : preservation of
natural habitat and management of water levels in accordance with natural hydrological
processes including stochastic ones would be the best possible conservation strategy for RCP.
Such a strategy would involve a succession of bad and good hydrological years as it is the
case in naturally evolving and functioning Mediterranean and Central Asian wetlands and
lagoons, depending on stochastic flooding by precipitation, sea storm and rise in river level.
We suggest that RCP is adapted to such high level of environmental stochasticity in
reproductive performance, more precisely in the proportion of breeding birds (breeding
incidence) according to fluctuating flooding conditions. Such a flooding-driven breeding
propensity is a common pattern in ducks (Johnson et al. 1992), including in close
phylogenetic relative (Anderson et al. 2001, Arnold et al. 2002) as well as biogeographical
relative (Almaraz & Amat 2004) of RCP.
Modelling hypothetical population dynamics under fully natural conditions
To assess the assumption of population viability of RCP under strong flooding fluctuation, we
used matrix models developed in Chapter 5 with the following adjustments
- no hunting harvest
- natural adult annual survival set at So= 0.753 as estimated by Devineau et al. (in prep.)
- fecundity under rainfall stochasticity as in chapter 5 except that :
- random adult breeding propensity set at 1 in good years with probability 0.1, 0.5 in average
years with probability 0.72 and 0.2 in bad years of flooding with probability 0.18
- initial population size set at 3 females simulating founding effect
We ran the model 1000 times for 20 time steps.
Under such severe regime of stochasticity but in the absence of hunting, the matrix population
model led to a surprisingly low extinction probability of only 13.4% considering the very
small founding population size of 3 females. The mean stochastic growth rate was slightly
above 1 (1.01; SE= 0.0014). Therefore a small founding or nomadic population of RCP seems
to have a reasonable probability of settling and become viable under highly fluctuating
hydrological conditions naturally prevailing in the Mediterranean or Central Asia. This result
deserves further research under exploitation regime (Beddington & May 1977). Exploited
population dynamics of Mediterranean and Central Asian ducks may appear quite different
from the generally well studied exploited population dynamics of boreal ducks. These
questions are approached in chapter 5 and the conclusions reached are summarized below.
Preliminary approach to RCP population limits and dynamics under exploitation
(chapter 4 and 5)
Understanding population limits and structure allowed us to use corresponding census data to
model the targeted population dynamics. A preliminary approached to RCP population
dynamics reveals some original aspects that may be explained by the species ecology
Understanding RCP population limits and structure
In order to understand and model European population dynamics of RCP, it was first
necessary to gain insight into the population limits and size. In fact, sizes and corresponding
ranges of the different RCP populations were first hypothesized through analyses conducted
by the Waterbird International Census coordinated by Wetlands International (Delany & Scott
2006). In the framework of this international monitoring, RCP global range had been splitted
into 3 population units (Figure 2 after Scott & Rose 1996) based on winter abundance
patterns, as initially proposed by Monval & Pirot (1989). However, their analyses provided
only hypothetical basis to these intraspecific population limits; in particular no genetic studies
had validated the proposed population structure. This was partially done in the genetic study
presented in chapter 4, which confirmed that the Central European and Western
Mediterranean population forms a single management unit (Gay et al. 2004). The
phylogenetic position of the Eastern European population of RCP remains to be investigated.
Modelling RCP population dynamics
The next step was then to address the dynamics of this Central European and Western
Mediterranean population through demographic modelling. This was done using a matrix
population model (chapter 5) which provided satisfactory consistency between observed
census data and expected population abundance. In addition, the following conclusions were
reached.
•
There is evidence that density-dependent harvest compensation operates at the
reproductive level, which is not new but very rarely addressed in ducks. Its effects are
nevertheless probably marginal at the current harvest levels.
•
Current European harvest rate appears to be sustainable and would be increasingly so
in the context of some predicted climate change scenarii. Indeed, a hunting ban
between 1 and 10 February would probably increase the population growth rate only
marginally.
•
However, the above results should be considered cautiously because there is
substantial lack of precision in the estimates of most important demographic
parameters identified through parameter uncertainty analysis (Hunter et al. 2000) :
survival rate, fledging success, survival rate in the absence of hunting and harvest
level. Proper estimation of these parameters is the main priority. Designing a
conservation-oriented monitoring program allowing estimation of population vital
rates would be a good opportunity to develop a semi-experimental international
integrated monitoring as advocated by Elmberg et al. (2006) and as implemented with
success in North America under the label of Adaptive Management (Nichols et al.
1995).
In addition, we provided preliminary estimation of some vital rates of RCP for the first time.
These vital rates may be the key to a somewhat original demographic strategy of the species
that might be hypothesized to be linked to the species particular ecology among a Central
Asian and Mediterranean « Sarmatic » waterfowl guild (Voous, 1960).
Hypothetical links between RCP ecology and demography
Along RCP, four other species can be considered to belong to the Sarmatic guild of Eurasian
wildfowl: Ruddy Shelduck Tadorna ferruginea, Ferruginous duck Aythya nyroca, Marbled
Teal Marmaronetta angustirostris and White-headed Duck Oxyura leucocephala.
Gadwall Anas strepera and Pochard Aythya ferina can be considered as having an
intermediate position between this guild and the well-studied and outnumbering guild of
boreal wildfowl. Overall, like RCP, all four of these Sarmatic species show
- A patchy distribution in Europe and around the Mediterranean, but species are more
widespread in Central Asia where they might originate from (Scott & Rose 1996, Gay et al.
2004, Muñoz-Fuentes et al. 2005)
- an ecological relationship with continental and Mediterranean lagoons ecosystems
- generally unfavourable conservation status (Green 1996, UICN 2006)
We suggest that this Sarmatic wildfowl guild might show some further demographic and
ecological similarities, including:
- high reproductive performance (Green 1998a and Green et al. 1999 for Marbled Teal)
including high juvenile survival (Almaraz & Amat 2004, Defos du Rau et al. in prep)
- substantial breeding habitat plasticity (Boutin 1984 for RCP, Green 1998b for Marbled Teal
and Ferruginous Duck)
- relatively low apparent/local adult survival.
From an evolutionary view point, these peculiarities might be all part of a nomadic strategy in
which these ecological and demographic advantages would trade-off with drawbacks of
irregular reproductive occurrence following fluctuations of breeding habitat factors and
notably flooding conditions.
Again, we stress the fact that these are hypotheses, some of them being also formulated by
Keller (2000a); they remain to be fully addressed, but our results on RCP are consistent with
them. These hypotheses call for stronger international cooperation between research programs
on these 5 Sarmatic species.
CONCLUDING
APPROACH TO THE RESEARCH AND MONITORING NEEDS FOR
RCP
CONSERVATION AND SUSTAINABLE EXPLOITATION
We identified three priority areas for future research and monitoring : genetic, integrated
international monitoring including vital rate estimation, experimental identification of habitat
requirements.
Experimental validation of habitat requirements
Preliminary conclusions on breeding habitat requirements of RCP, notably concerning water
management could be experimentally assessed by comparing RCP occupancy and abundance
on a paired sample of wetlands. Each pair should include (i) one wetland with traditional
current water management with summer flooding and spring drying off and (ii) one wetland
with natural water management simulating flooding only when main river or sea level rise
significantly above average. Most of Western European rivers are nowadays embanked and
would not allow such naturally occurring event. Therefore, an experimental design is
required. Such design might benefit to the understanding of the ecology of some other
flagship species.
Genetics
Genetic characterisation of the Eastern European putative population is a priority in order to
understand whether it belongs to one of either Central Asian or Western European population.
Romania and Turkey are two countries where RCP distribution and abundance is sufficiently
known to facilitate efficient sampling of genetic material. Evidence for existence versus
absence of dispersal between both Eastern and Western population would have important
implications on the species conservation and exploitation management in Western Europe. In
this respect, additional sampling at the major wintering site of Prespa Lake on the AlbanianGreek-Macedonian border (Scott & Rose 1996) would be very interesting as it would allow to
further investigate the hypothetical bird movement between Eastern and Western population.
Research on this site would benefit from both genetic and satellite tracking surveys.
Integrated monitoring of population dynamic
Integrated long-term monitoring of European RCP populations is a high priority in order to
decrease uncertainty in population dynamics understanding and management decisionmaking, and ensure harvest sustainability. Such monitoring scheme would largely benefit
from waterfowl Adaptive Management designs developed in North America (Nichols et al.
1995) and the recently proposed unified wildlife population modelling framework (Thomas et
al. 2005). Features to incorporate in a European integrated monitoring of RCP and other game
ducks would include (Elmberg et al. 2006) :
- winter aerial surveys in Eastern Europe, especially over Black and Caspian seas and the
major river deltas
- harvest surveys in France, Portugal, Romania and Spain (Mondain-Monval et al. 2006)
- demographic vital rate estimation through nasal-saddle marking (Guillemain et al. 2007) and
telemetry CMR schemes in Western Europe. The vital rates to be estimated would be survival
rates under various experimental levels of harvest, hatching and fledging success and
breeding incidence and propensity.
Following recent publication of RCP population trends (Delany & Scott 2006), we believe
that RCP is not of critical conservation concern and that some alarming statements about its
former conservation status have had the potential to locally raise unnecessary efforts for its
conservation. Much of this discrepancy between real and advocated conservation needs
potentially came from the lack of consideration of detection imperfection in census data.
However, we also believe that, as a rare game duck, RCP deserves to be carefully monitored
and that its exploitation for hunting is more complicated and risky than for most, if not all,
other duck species. Therefore, sustainability of its exploitation requires highest level of
warranty on our understanding and predictive ability on the species population dynamics.
Sustainability of RCP exploitation requires to decrease uncertainty to an unparalleled level in
South European game birds.
References
Almaraz, P. & Amat, J.A. 2004: Multi-annual spatial and numeric dynamics of the whiteheaded duck Oxyura leucocephala in southern Europe: seasonality, density
dependence and climatic variability. - Journal of Animal Ecology 73: 1013–1023.
Anderson, D. R., W. A. Link, D. H. Johnson, and K. P. Burnham. 2001. Suggestions for
presenting the results of data analyses. Journal of Wildlife Management 65:373-378.
Anderson, M.G., Lindberg, M.S. & Emery, R.B. 2001: Probability of survival and breeding
for juvenile female Canvasbacks. Journal of Wildlife Management 65: 385-397.
Arnold, T.W., M.G. Anderson, M.D. Sorenson & R.B. Emery. 2002. Survival and philopatry
of female redheads breeding in south-western Manitoba. - Journal of Wildlife
Management 66: 162-169.
Aznar, J.C., Dervieux, A. & Grillas, P. 2003: Association between aquatic vegetation and
landscape indicators of human pressure. Wetlands 23:149-160.
Barry, S.C. & Elith, J. 2006: Error and uncertainty in habitat models. - Journal of Applied
Ecology 43: 413-423.
Batt, B.D.J., Afton, A.D., Anderson, M.G., Ankney, C.D., Johnson, D.H., Kadlec, J.A. &
Krapu, G.L. 1992: Ecology and Management of Breeding Waterfowl. - University of
Minnesota Press, Minneapolis & London, 635pp.
Beddington, J. R. & May, R. M. 1977: Harvesting populations in a randomly fluctuating
environment. - Science 197: 463-465
Block, W.M. & Brennan L.A. 1993: The habitat concept in ornithology: Theory and
applications. - Current Ornithology 11: 35–91.
Blums, P, Nichols, J.D., Hines, J.E. & Mednis, A. 2002: Sources of variation in survival and
breeding site fidelity in three species of European ducks. - Journal of Animal Ecology
71 : 438-450
Cramp, S. & Simmons, K.E.L. (eds) 1977: The Birds of the Western Palearctic. Vol. 1.
Oxford Univ. Press, 722pp.
Defos du Rau, P., Barbraud, C. & Mondain-Monval, J.Y. 2003: Estimating breeding
population size of the red-crested pochard (Netta rufina) in the Camargue (southern
France) taking into account detection probability: implications for conservation. Animal Conservation 6: 379-385.
Defos du Rau, P., Barbraud, C. & Mondain-Monval, J.Y. 2005: Incorporating uncertainty into
analyses of Red-crested Pochard habitat selection. - Biological Conservation 125: 355367.
Delany S. & Scott, D., 2006: Waterbird Population. Estimates (ThirdEdition). Wageningen,
Wetlands International, Global Series, n° 12.
Elmberg, J., Nummi, P., Pöysä, H., Gunnarsson, G. & Sjöberg, K. 2005: Early breeding Teal
Anas crecca use the best lakes and have the highest reproductive success. - Annales
Zoologici Fennici 42: 37-43.
Elmberg, J., Nummi, P., Pöysä, H., Sjöberg, K., Gunnarsson, G., Clausen, P., Guillemain, M.,
Rodrigues, D. & Väänänen, V. 2006: The scientific basis for a new and sustainable
management of migratory European ducks. - Wildlife Biology 12: 121-128.
Farnsworth, G.L., Nichols, J.D., Sauer, J.R., Fancy, S.G., Pollock, K.H., Shriner, S.A. &
Simons, T.R. 2005 Statistical approaches to the analysis of point-count data: a little
extra information can go a long way. USDA Forest Service Gen. Tech. Rep. PSWGTR-191 : 736-743.
Franco, A.M.A., Palmeirim, J.M. & Sutherland, W.J. 2007: A method for comparing
effectiveness of research techniques in conservation and applied ecology. - Biological
Conservation 134: 96-105.
Gay, L., Defos du Rau, P., Mondain-Monval, J.Y. & Crochet, P.A. 2004: Phylogeography of
a game species: the red-crested pochard (Netta rufina) and consequences for its
management. - Molecular Ecology 13 : 1035-1045.
Green, A.J. 1996: Analyses of globally threatened Anatidae in relation to threats, distribution,
migration patterns and habitat use. - Conservation Biology 10: 1435-1445.
Green, A.J. 1998a: Clutch size, brood size and brood emergence in the Marbled Teal
Marmaronetta angustirostris in the Marismas del Guadalquivir, southwest Spain. Ibis 140: 670-675.
Green, A.J. 1998b: Habitat selection by the Marbled Teal Marmaronetta angustirostris,
Ferruginous Duck Aythya nyroca and other ducks in the Göksu Delta, Turkey in late
summer. - Revue d’Ecologie, Terre et Vie 53: 225-243.
Green, A.J., Navarro, J.D., Dolz, J.C. & Aragoneses, J. 1999: Brood emergence patterns in a
Mediterranean duck community. - Bird Study 46: 116-118.
Guillemain, M., Poisbleau, M., Denonfoux, L., Lepley, M., Moreau, C., Massez, G., Leray,
G., Caizergues, A., Arzel, C., Rodrigues, D. & Fritz, H. 2007: Multiple tests of the
effect of nasal saddles on dabbling ducks : combining field and aviary approaches. –
Bird Study 54: 35-45.
Hunter, C., Moller, H. & Fletcher, D. 2000: Parameter Uncertainty and Elasticity Analyses of
a Population Model: Setting Research Priorities for Shearwaters. - Ecological
Modelling 134: 299-323.
IUCN 2006 Red List of Threatened Species. <www.iucnredlist.org>. Downloaded on 15 May
2007.
Johnson, A.R. 1975 : Station de baguage de Camargue. Compte-Rendu pour les années 1972
et 1973. - Revue d’Ecologie Terre et Vie 29: 116-130.
Johnson, D.H., Nichols, J. D. & Schwartz, M. D. 1992: Population dynamics of breeding
waterfowl. Pages 446–485 in Batt B. D. J., Afton A. D., Anderson M. G., Ankney C.,
Johnson, D.H., Kadlec, J.A., Krapu, G.L. Ecology and management of breeding
waterfowl. University of Minnesota Press. Minneapolis, USA.
Kendall, W.L., Nichols, J.D. & Hines, J.E., 1997: Estimating temporary emigration and
breeding proportions using capture-recapture data with Pollock's robust-design. Ecology 78: 563-578.
Kery, M. 2004: Extinction rate estimates for plant populations in revisitation studies:
importance of detectability. - Conservation Biology 18: 570–574.
Koper, N. & Schmiegelow, F.K.A. 2006: Effects of habitat management for ducks on target
and nontarget species. - Journal of Wildlife Management 70: 823–834.
Krebs, C.J. & Berteaux, D. 2006: Problems and pitfalls in relating climatic variability and
population dynamics. - Climate Research Special 15, 32: 143-149.
Krivenko, V.G. 1994: Red-crested Pochard in Tucker, G.M. & Heath, M.F. Birds in Europe,
their conservation status. - Birdlife Conservation Series n° 3.
Linţia, D. 1955: Păsările din R.P.R. Ed. Academiei R.P.R. - Bucureşti. 3: 123 – 125.
Long P.R., Szekely, T., Kerhaw, M. & O’Connell, M. 2007: Ecological factors and human
threats both drive wildfowl population declines. - Animal Conservation 10: 183-191.
MacKenzie, D.I., Nichols, J.D., Lachman, G.B., Droege, S., Royle, J.A. & Langtimm, C.A.
2002: Estimating site occupancy rates when detection probabilities are less than one. –
Ecology 83: 2248-2255.
MacKenzie, D.I., Nichols, J.D., Royle, J.A., Pollock, K.H., Hines, J.E. & Bailey, L.L. 2006:
Occupancy estimation and modeling: inferring patterns and dynamics of species
occurrence Elsevier, San Diego, USA, 324 pp.
Mayaud, N., 1966: Contribution à l'histoire de Netta rufina Pallas, la Nette à huppe rousse en
Europe occidentale. - Alauda 34: 191-199.
McLoughlin, P.D. & Messier, F. 2004: Relative contributions of sampling error in initial
population size and vital rates to outcomes of population viability analysis. –
Conservation Biology 18: 1665-1669.
Mondain-Monval, J-Y., Defos du Rau, P., Mathon, N., Olivier, A. & Desnouhes, L. 2006:
The monitoring of hunting bags and hunting effort in the Camargue, France. Waterbirds around the world. Eds. G.C. Boere, C.A. Galbraith & D.A. Stroud. The
Stationery Office, Edinburgh, UK. 862-863.
Monval, J.Y. & Pirot, J.Y. 1989: Results of the IWRB International Waterfowl Census 19671986. - IWRB Spec. Publ. 8, Slimbridge, UK.
Nichols, J.D., Johnson, F.A. & Williams, B.K. 1995: Managing North American Waterfowl in
the Face of Uncertainty Annual Review of Ecology and Systematics 26: 177-199.
Nichols, J.D., Hines, J.E., Sauer, J.R., Fallon, F.W., Fallon, J.E. & Heglund, P.J. 2000: A
Double-observer Approach for Estimating Detection Probability and Abundance from
Point Counts. - Auk 117: 393-408.
Paspaleva, M., Kiss, J.B. & Tălpeanu, M. 1984: Sur la dynamique de quelques espèces
d'oiseaux dominants dans le Delta du Danube. Trav. du Mus. d' Hist. Nat. - "Gr.
Antipa". 25: 312 – 329.
Pollock, K.H., Nichols, J.D., Simons, T.R., Farnsworth, G.L., Bailey, L.L. & Sauer, J.R.
2002 : The design of large scale wildlife monitoring studies. - Environmetrics 13: 105119.
Pöysä, H. & Paasivaara, A. 2006: Movements and mortality of common goldeneye Bucephala
clangula broods in a patchy environment. – Oikos 115: 33-42.
Pradel, R., Rioux, N., Tamisier, A., & Lebreton, J.-D. 1997 : Individual turnover among
wintering teal in Camargue: a mark-recapture study. - Journal of Wildlife Management
61: 816-821.
Rexstad, E., Burnham, K.P. 1991: User's Guide for Interactive Program CAPTURE, Markrecapture population and density estimation program. - Cooperative Fish & Wildlife
Research Unit, Colorado State University, Fort Collins.
Rousset F. 2001: Genetic approaches to the estimation of dispersal rates. In: Dispersal (eds
Clobert, J., Danchin, E., Dhondt, AA. & Nichols, J.D.), Oxford University Press,
Oxford, 452 pp. 18–28.
Schneider-Jacoby, M., Vasic, V. F. 1989: The Red-crested Pochard Netta rufina breeding and
wintering in Yougoslavia. - Wildfowl 40: 39-44.
Scott, D.A., & Rose, P.M. 1996: Atlas of Anatidae Populations in Africa and Western
Eurasia. - Wetlands International Pub. n° 41, Wetlands International, Wagenigen, The
Netherlands, 336 pp.
Shedden, C.B. 1986: Status of European quarry species. - The British Association for
Shooting and Conservation
Snow, D.W. & Perrins, C.M. 1998: The Birds of the Western Palearctic, Vol. 1. Oxford
University Press, Oxford, 1008 pp.
Tamisier, A. 1991: Nette rousse in Yeatman-Berthelot, D. & Jarry, G. Atlas des oiseaux de
France en hiver - Société Ornithologique de France, 116-117.
Thomas, L., Buckland, S.T., Newman, K.B. & Harwood, J. 2005: A unified framework for
modelling wildlife population dynamics. - Australian and New Zealand Journal of
Statistics 47: 19-34.
Tucker, G.M. & Heath, M.F. 1994: Birds in Europe, their conservation status. - Birdlife
Conservation Series n° 3.
Voous, K.H. 1960: - Atlas of European birds. Nelson, Edinburgh.
Williams, B.K., Nichols, J.D., & Conroy, M.J. 2002: Analysis and management of animal
populations. - Academic Press, San Diego, 817 pp.
APPENDIX: LITERATURE REVIEW BEFORE 2001
Allouche, L., Roux, P. & Tamisier, A. 1988 Position trophique des nettes rousses (Netta rufina, Pallas) hivernant en Camargue.
Rev Ecol 43 : 167-175
Amat, J.A. 1982 The nesting biology of ducks in the Marismas of the Guadalquivir, South-Western Spain. Wildfowl, 33 : 94-104
Amat, J.A. 1985 Nest parasitism of Pochard Aythya ferina by Red-crested Pochard Netta rufina. Ibis 127: 255-262
Amat, J.A. 1984. Ecological segregation between Red-crested Pochard Netta rufina and Pochard Aythya ferina in a fluctuating
environment. Ardea 72, 229-233.
Amat, J.A. 1991 Effects of Red-crested Pochard nest parasitism on Mallards. The Wilson bulletin vol. 103, no3, pp. 501-503
Amat, J.A. 1993 Parasitic laying in Red-crested Pochard Netta rufina nests. Ornis Scandinavica 24:11, 65-70
Amat, J. 2000 Courtship feeding, food sharing, or tolerated food theft among paired Red-crested Pochards (Netta rufina)? Journal
of Ornithology 141 : 327-334
Amat, J.A., Lucientes, J. & Ferrer, X. 1987 La migracion de muda del pato colorado (Netta rufina) en Espana. Ardeola, 34 : 79-88
Baatsen, R.G. 1990 Red-crested Pochard Netta rufina in the Cotswold water park. Hobby 16 : 64-67
Badin, T. 1997 Suivi de la reproduction des canards dans le département de la Loire. rapport non publié, Fédération
départementale des Chasseurs de la Loire
Berndt, R.K. 1997 Netta rufina in Hagemeijer, W.J.M. & Blair, M.J. The EBCC Atlas of European Breeding Birds. Their
distribution and abundance. POYSER T & A.D., London
Bersot, E. 1979 Reproduction de la Nette rousse Netta rufina au lac de Neufchâtel en 1978. Nos Oiseaux 35 : 22-24
Boutin J. 1986 Comportement diurne de la Nette rousse, Netta rufina P., pendant son hivernage en Camargue. Rev. Ecol., 41 : 261269
Boutin, J. 1994 Nette rousse in Yeatman-Berthelot D., & Jarry, G., Nouvel Atlas des Oiseaux nicheurs de France 1985-1989.
S.O.F., Paris.
Brichetti, P. 1992 Fistione Turco in Brichetti, P., De Franceschi, P. & Bacetti, N. (Eds.) Fauna d’Italia. Aves. I Gaviidae Phasianidae. Vol XXIX Edizioni Calderini Bologna.
Broyer, J. 1994 The red-crested pochard in France. In The importance of the Mediterranean basin for migratory avifauna,
Carcassonne 22-24 April 1994. Proceedings 32-35
Broyer, J. & Daléry, G. 2000 L’habitat de la Nette rousse Netta rufina sur les étangs piscicoles de l’est de la France en période de
reproduction. Alauda 68 (3) : 185-191
Catry P (1998). Pato-de-bico-vermelho Netta rufina. In: Atlas das Aves Invernantes do Baixo Alentejo.Pp.106-107. Elias GL,
Reino LM, Silva T, Tomé T & Geraldes P (coords.). Sociedade Portuguesa para o Estudo das Aves, Lisboa.
Dehorter, O. & Rocamora, G. 1999 Nette rousse Netta rufina in Rocamora, G. & Yeatman-Berthelot, D. Oiseaux menacés et à
surveiller en France. Listes rouges et recherche de priorités. Populations. Tendances. Menaces. Conservation. Société d’Etudes
Ornithologiques de France/LPO Paris
Dirksen, S., & J. van der Winden. 1996. [Numbers of Red-crested Pochard Netta rufina in the Gouwzee in late summer and
autumn vary with breeding success of Dutch population.] Limosa 69: 131--133
Dobrowolski, K. A., B. Leznicka, & R. Halba. 1996. Natural food of ducks and coots in shallow-macrophyte dominated lake Lake
Luknajno (Masurian Lakeland, Poland). Ekol. Pol. 44: 271--287. (Inst. Zool. Univ., 00--927 Warszawa, Krakowskie Przedmiescie
26/28, Poland.)---Anas platyrhynchos, Aythya ferina, Netta rufina, Fulica atra, and Cygnus olor
Gaillardin, C. 1991 La reproduction de la Nette rousse en Camargue. rapport de stage BEPA cynégétique, Vendôme
GARGIONI A., MUTTI A. & ROSSI A., 1995 - Nidificazione di Fistione turco Netta rufina, sul Lago di Garda (Lombardia).- Riv.
ital. Orn.65: 83-85
Green, A.J., Navarro, J.D., Dolz, J.C. y Aragoneses, J. 1999. Brood emergence patterns in a Mediterranean duck community. Bird
Study 46: 116-118
Hauri, R. 1973 Zum brutvorkommen der Kolbenente Netta rufina in der Aarelandschaft suedlich von Bern. Ornithol. Beob. 70 (2)
: 57-66
Hauri, R. 1983 Zum Vorkommen der Kolbenente Netta rufina in der Aarelandschaft südlich von Bern und am Thunersee, 19731982. Orn. Beob. 80: 119-126
Hauri, R. (1989): Zum Vorkommen und zur Biologie der Kolbenente Netta rufina in der Schweiz 1974-1988, mit besonderer
Berücksichtigung des Thunersees. Ornithol. Beob. 86: 69–87.
Hauri, R. 1995 Bastarde der Kolbenente Netta rufina und der StockenteAnasplatyrhynchos am Thunersee. In: Ornithol. Beob., Bd.
92, 1995, Nr. 4: 481-482
Hauri, R. 1995 Veränderungen im Auftreten der Kolbenente Netta rufina am Thunersee von 1989 bis 1994. Ornithol. Beob., Bd.
92, 1995, Nr. 4, S. 477-480
Heiser, F. 1992 Breeding of the Red-crested Pochard in Swabia, southwest-Bavaria. Ornithol. Anz. 31 (3): 159-161
Hellebrekers, W.P.H. & Voous, K.H. 1964 Nestparasitisme van de Krooneend. Limosa 37: 5-11
Hückler, U. 1966 Ringfunde der Kolbenente (Netta rufina). Auspicium 2 : 248-258
Hudec, K. 1968 Migration of the Czechoslovak populations of the diving duck (gen. Aythya, Netta, Bucephala). Zool. Listy 17: 4962
Jauch, W.A. 1948 die Kolbenente Netta rufina am Bodensee. Ornithol. Beob. 45: 129-134
Jauch, W.A. 1952 Probleme der Kolbenentenforschung. Vögel der Heimat 23: 1-7
Keller, V. 2000a. Winter distribution and population change of red-crested pochard Netta rufina in southwestern and central
Europe. Bird Study 47 : 176-185.
Keller, V. 2000b. Winterbestand und Verbreitung der Kolbenente Netta rufina in der Schweiz und im angrenzenden Ausland.
Ornithol. Beobachter. 97: 175-190.
Keller, V. & M. Antoniazza (2001): L'importance des réserves d'oiseaux d'eau sur le lac de Neuchâtel pour la Nette rousse Netta
rufina et d'autres espèces hivernantes. Nos Oiseaux suppl. 5: 81–90.
Kiss, J.B., Rekasi, J., Sterbetz, I. & Torok, Zs. 1997 Data about feeding activity of some species of Anseriformes in Danube Delta,
North Dobrogea Romania, Acta Cinegetica Romaniae, Studies and Communication 77– 88.
Kraus, M. & Krau¢, W. 2000 Vorkommen und Brutbestand der Kolbenente Netta rufina im Fränkischen Weihergebiet und
weitere neue Brutplätze in Nordbayern. Orn. Anz. 39 : 175-186
Krivenko, V.G. 1994 Red-crested Pochard in Tucker, G.M. & Heath, M.F. Birds in Europe, their conservation status. Birdlife
Conservation Series n° 3.
Kuhk, R. 1951 Herbstliches Massen-Vorkommen der Kolbenente Netta rufina auf dem Untersee (Bodensee) Vogelwarte 16: 82-83
Kux, Z. 1980 Investigations of the Avifauna of the Danube Delta in October, 1979. ACTA MUS. MORAVIAE. Vol. 65, pp. 173190
Lebreton P. & Rochette P. (1965). - Statistiques cynégétiques sur les Anatidés de la Dombes. Alauda, 38 : 84-130.
Lecocq, A. 1997 Observations des stationnements et des comportements alimentaires de la Nette rousse Netta rufina en baie
d’Yvonand VD, lac de Neuchâtel. Du rôle des characées. Nos Oiseaux 44 : 83-95
Léger, F., Lécaille, R. & Thommès, F. 1993 Une nidification de Nette rousse, Netta rufina, en Lorraine. Ciconia 17 (2) 63-70
Lind H. 1962. Zur Analyse des sexuellen Verhaltens der Kolbenente, Netta rufina (Pallas). Zeitschrift für Tierpsychologie. 19:
607–625
Llorente, G.A. & Ruiz, X. 1985 Datos sobre la reproducción del Pato Colorado Netta rufina (Pallas 1773) en el Delta del Ebro.
Misc. Zool. 9 : 315-323
Llorente, G.A., Ruiz, X. & Serra-Cobo, J. 1986 Alimentation automnale de la Nette rousse (Netta rufina, Aves, Anatidae) dans le
delta de l'Ebre, Espagne - Vie Milieu, 36 : 97-107
Llorente, G.A., Farran, A., Ruiz, X. & Albaigés, J. 1987 Accumulation and distribution of hydrocarbons, polychlorobyphenyls, and
DDT in tissues of three species of Anatidae from the Ebro Delta (Spain). Archives of Environmental Contamination and
Toxicology 16 : 563-572
Löppenthin, B. 1960 Die Kolbenente,Netta rufina (Pall.), in Dänemark. Journal of Ornithology. 101 : 126-128
Lucientes, J. 1977 La migración de muda del Pato Colorado en la laguna de Gallocanta. IV Jorn. Ornit. Esp. (Barcelona)
Lucientes, J. 1978 El enigma del Pato Colorado. Trofeo 96: 28-31
MARTIN, M.C. (1995) Première reproduction de la nette rousse Netta rufina en Vaucluse. Faune de Provence 16: 119
MAUFF, G. & R.WEISE (1992): Zur Situation der Kolbenente, Netta rufina Pall.1773, in Thüringen unter besonderer
Berücksichtigung zweier Gewässer in Mühlhausen. Thüring. Ornithol. Mitt. 42, 6-12.
Mayaud, N. 1966 Contribution à l'histoire de Netta rufina Pallas, la Nette à huppe rousse en Europe occidentale. Alauda 34 : 191199
MAYOL, J. & VICENS, P. (1995): "Reintroducció de Porphyrio porphyrio, Netta rufina i Oxyura leucocephala a S'Albufera de
Mallorca". In: S'Albufera de Mallorca. MARTÍNEZ TABERNER, A. & MAYOL, J. (Eds.). Editorial Moll. Palma de Mallorca.
pp. 159-168.
Muntaner, J., Ferrer, X. & A. Martinez-Vilalta (1984). Atlas dels ocells nidifcants de Catalunya I Andorra. Ketres Editora,
Barcelona.
Munteanu, D. (1996) Changes and trends in Anatidae populations in Romania after the second world war in Proceedings of the
Anatidae 2000 Conference, Strasbourg, France, 5-9 December 1994, M. Birkan, J. van Vessem, P. Havet, J. Madsen, B. Trolliet &
M. Moser, eds. Gibier Faune Sauvage, Game Wildl., 13 973-974
Musil, P. & Fuchs, R. 1994 Changes in abundance of water birds species in southern Bohemia (Czech Republic) in the last 10
years. Hydrobiologia Volume 279-280 : 511-519
Musil P. (2000). - Monitoring of water bird breeding populations in the Czech Republic (1988 - 1997). Vogewelt, 120: 253-256
Noll, H. 1929 Biologische Beobachtungen über die Kolbenente,Netta rufina (Pall.), im Gebiete des Untersees (Bodensee). Journal
of Ornithology 77 : 208-220
Olioso, G. 1996. [Data on the breeding of water birds in Vaucluse, South-East France.] Faune de Provence 17: 71--76. (le Mail,
26230 Grignan, France.)---Turdus ruficollis, Parus cristatus, Anas platyrhynchos, Anas querquedula, Anas clypeata, Netta rufina,
Rallus aquaticus, Gallinula chloropus, Fulica atra; 2.5 % in September and October hunting period.
Paspaleva, M., Kiss, J.B. & Tălpeanu, M. 1984 - Sur la dynamique de quelques espèces d'oiseaux dominants dans le Delta du
Danube. Trav. du Mus. d' Hist. Nat. "Gr. Antipa". 25 : 312 – 329.
Rimbert, S. 1990 Eco-éthologie de la Nette rousse en Camargue. rapport de stage BEPA cynégétique, Charleville-Mézières
Rioux, C. 1992 La reproduction de la Nette rousse (Netta rufina) dans la vallée du Rhône. Bulletin Mensuel de l’Office National
de la Chasse 168 : 14-16
Ruiters, P.S., Noordhuis, R. & Van den Berg, M.S. 1994 Kranswieren verlklaren aantalsfluctuaties van Krooneeden Netta rufina
in Nederlands. Limosa 67: 147-158: [Stoneworts account for fluctuations in Red-crested Pochard Netta rufina in The Netherlands.]
Saez-Royuela, R. 1997 Pato Colorado in SEO/BirdLife Atlas de las Aves de España (1975-1995). Lynx Edicions, Barcelona
Schlenker, R. 1979 Ringfunde der Kolbenente (Netta rufina). Auspicium 6 : 417-420
Schneider-Jacoby, M. & Vasic, V.F. 1989 The Red-crested Pochard Netta rufina breeding and wintering in Yougoslavia.
Wildfowl, 40 : 39-44
Schneider-Jacoby, M. 1999 Kolbenente - Netta rufina. In: G. Heine, H. Jacoby, H. Leuzinger & H. Stark (Hrsg.): Die Vögel des
Bodenseegebietes. Ornithol. Jh. Bad.-Württ. 14/15: 273-279.
Schneider-Jacoby, M. 2000 Freizeit und artenschutz am Wasser – Sicherung der Brut- und Rastgebiete von Kolbenente und
Moorenten in Deutschland. Schriftenr. Landschaftspflege Naturschutz 60: 81-93
Sermet, E. 1998 Notes sur la nidification de la Nette rousse Netta rufina à Corcelettes/Grandson (lac de Neuchâtel) en 1995 et
1996. Dans: Nos oiseaux, 1998, n° 452, p. 71-75.
Siblet, J.P. (2000) Premier cas de reproduction de la Nette rousse Netta rufina en Ile-de-France. Alauda 68 (1) 44
Sluiters J.E. 1942. Het broeden van de Krooneend,Netta rufina (Pall.), in Nederland in 1942. Ardea 31: 281-284.
Steinbacher, G. 1957 Zur Biologie des Kolbenente. Vogelwelte 78: 82-88
Szijj, J. 1975 Probleme des Anatidenzuges, dargestellt an den Verlagerungen des europäischen Kolbenentenbestandes. Ardeola 21
: 153-171
Tamisier, A. 1991 Nette rousse in Yeatman-Berthelot, D. & Jarry, G. Atlas des oiseaux de France en hiver - Société
Ornithologique de France, 116-117
Tamisier, A. & Dehorter, O. (1999) Camargue, canards et foulques. Centre Ornithologique du Gard, Nîmes
Timmerman A. 1962. De Krooneend (Netta rufina) in Nederland (i.v.m. het Europese optreden). Limosa 35: 28-39.
Van der Winden, J., Hagemeijer, W., Hustings, F., & Noordhuis, R., 1994 Hoe vergaat het Krooneend Netta rufina in Nederland ?
Limosa 67 : 137-146 :[Vicissitudes of the Dutch Red-crested Pochard Netta rufina population.] ---Long-term trend related to
distribution and abundance of Nitellopsis obtusa (a stonewort), with numbers in autumn 1993 estimated at 120--150 birds.
Van Impe, J. 1985 Contribution à la mue des rémiges chez la Nette à huppe rousse, Netta rufina (Pallas) en Espagne du NordAlaud, 53 : 1-10
van Oort E.D. 1905. Über das Vorkommen von Netta rufina (Pall.) in Holland. Notes from the Leyden Museum XXVI: 196-199.
Voous, K.H. 1943 De Kroonend Netta rufina broedvogel in Nederland. Ardea 32: 1-10
WEISE, R. (1990): Kolbenente (Netta rufina) in Thüringen. Ornith. Mitteilungen 42, 298
WEISE, R. (1992): Zum Brutbestand der Kolbenente (Netta rufina) in Deutschland.- Orn. Mitteil. 44: 115-119.
Weise, R. 1993 Nahrung und Nahrungserwerb der Kolbenente Netta rufina. Der Ornithologische Beobachter 90 67-74
Chapter 2
Animal Conservation (2003) 6, 379–385 © 2003 The Zoological Society of London
DOI:10.1017/S1367943003003457 Printed in the United Kingdom
Estimating breeding population size of the red-crested pochard
(Netta rufina) in the Camargue (southern France) taking into
account detection probability: implications for conservation
Pierre Defos du Rau1, Christophe Barbraud2,3 and Jean-Yves Mondain-Monval1
1
Office National de la Chasse et de la Faune Sauvage, CNERA-Avifaune Migratrice, Le Sambuc, 13200 Arles, France
Station Biologique de la Tour du Valat, Le Sambuc, 13200 Arles, France
3
Centre d’Etudes Biologiques de Chizé, UPR 1934 CNRS, 79360 Villiers-en-Bois, France
2
(Received 17 October 2002; accepted 21 May 2003)
Abstract
The red-crested pochard (Netta rufina), a Eurasian diving duck, has seen its numbers declining and has
received strong conservation concern. Data on population size and rate of decline are required to establish
management plans, none of which is available for this species. Here we present the first population size
estimate taking into account detection probability and habitat use in the Camargue, southern France. A
non-random sample of 33 lakes was used to estimate detection probability from point-counts.
Detectability was low, with only 57.5% of individual broods detected. A random sample of 37 lakes was
then used to estimate brood densities. Adjusted densities (taking into account detection probability) were
0.1106 broods per ha of reedbed. Adjusted densities were extrapolated to the entire surface area of
reedbeds in the Camargue estimated from a GIS to obtain abundance estimates of the brood population.
A minimum estimate of 559 breeding pairs was obtained (95% confidence interval: 436–855). This
estimate is much higher than previous ones (80–100 pairs for the Camargue, 250 pairs for France), and
indicates strong underestimation of the population size when not taking into account detectability. Our
results suggest that the red-crested pochard may require a reassessment of its conservation status for
France and Europe. They further suggest that taking detection probability into account in population
estimates of other cryptic species, and notably those of conservation concern, may help clarify their
conservation status and may even affect the setting of conservation priorities.
INTRODUCTION
Estimates of population size from animal surveys are a
crucial tool for setting conservation priorities (Karanth &
Nichols, 1998; Kéry, 2002; Thompson, 2002). However,
few studies have taken into account the detection
probability (or detectability) of their target species, i.e., the
probability of missing an individual given it is present in
the sampling area, in estimating its abundance (Rosenstock
et al., 2002). Not taking into account the proportion of
individuals missed during surveys may lead to serious
biases in estimates of abundance (Nichols et al., 2000), and
therefore to misleading conservation status and priorities.
The red-crested pochard (Netta rufina) is a migratory
diving duck that breeds in central Asia, around the Black Sea,
and in western Europe. The species is classified as
‘Declining’ in Europe by Krivenko (1994), and is therefore
of strong conservation concern, although it is hunted in
All correspondence to: Pierre Defos du Rau, ONCFS, Délégation
Régionale, 10bis route d’Ax, 31120 Portet/Garonne, France.
Tel: 335 62 20 75 55; Fax: 335 62 20 75 56;
E-mail: [email protected].
France, Portugal, Romania and Spain. Consequently, the
European Commission required a management plan for this
species (Defos du Rau, 2002). In particular, this management
plan stressed the need to update estimates of breeding
population size in Europe. In the Camargue, southern France,
the latest breeding population size estimates of 80–100
breeding pairs (Rimbert, 1990; Gaillardin, 1991) are still in
use in the French Red List of threatened birds, which
classifies the red-crested pochard as ‘Endangered’ on the
basis of a French population size estimated at less than 250
breeding pairs and considered to be strongly declining since
the 1970s (Dehorter & Rocamora, 1999).
Here, we investigate two different aspects of red-crested
pochard broods detectability using two recently developed
methods: the double-observer approach (Nichols et al.,
2000), and a capture–recapture method based on the
Pollock’s robust-design approach (Kendall et al., 1997).
These detection probabilities are then used to estimate
the breeding population size, based on previous findings
on habitat requirements of the red-crested pochard in
the Camargue.
380
P. DEFOS DU RAU ET AL.
METHODS
Study site and species
Red-crested pochards breeding in France are concentrated
on three wetlands along the Rhône river (Boutin, 1994;
Dehorter & Rocamora, 1999). The southernmost of these
strongholds is situated in the Rhône delta, named the
Camargue, a vast wetland area of 145,000 ha. Natural
habitats account for 58,000 ha and mainly consist of
freshwater and brackish marshes, including reedbeds and
temporary flooded salt meadows. These natural areas are
split into 230 protected or hunted estates.
The red-crested pochard has an extensive breeding
season: eggs are laid from late March until early July. The
species’ breeding habitat has been recently described by
P. Defos du Rau, C. Barbraud & J.-Y. Mondain-Monval
(unpubl. data), who confirmed that, in the Camargue,
surface area of reedbeds of Phragmites australis is the
main factor positively governing the species’ reproduction
(see also Llorente & Ruiz, 1985; Schneider-Jacoby &
Vasic, 1989; Heiser, 1992; Snow & Perrins, 1998). As in
many other duck species, the red-crested pochard is rather
cryptic in its breeding behaviour and habits, and broods
mostly come out on open water in the late afternoon and
the evening.
Estimating detectability parameters
Two aspects of detectability were investigated. The
probability of false absence, i.e., the probability of not
detecting any pair of the species on a site given it is
actually raising ducklings, was used to estimate the
probability of missing the presence of one or more redcrested pochard broods during the survey conducted to
estimate the brood population size. The other detectability
parameter that was estimated was the probability of
missing one individually identified brood given it is
present on a site. This second detectability parameter was
used to correct the observed brood densities and to obtain
unbiased abundance estimates.
The red-crested pochard is one of the least abundant
breeding waterbirds in the Camargue. Pairs are not
widespread homogeneously, but are mainly concentrated
in the vast freshwater marshes in the eastern and
northwestern parts of the Camargue. Studying detection
of the species’ broods required performing a specific
detection survey in zones of concentrated use. A nonrandom sample of lakes was therefore necessary to assess
detectability parameters using most densely occupied
sites, since a random sample would have been inadequate
to estimate detectability of such a rare and cryptic
breeding bird.
The species is known or suspected to breed regularly
and densely in 27 estates (P. Defos du Rau, pers. obs.)
when breeding habitat is available, i.e., when flooding and
water management conditions are favourable. In such
estates, disturbance level is generally kept very low on
purpose during the breeding season of the waterfowl. Out
of these 27 estates, only 11 were both access-permitted and
flooded in 2001. Nine of these 11 estates were chosen for
their easy access and the availability of lakes within them.
Lakes of these nine estates constituted the non-random
sample, with a total of 33 lakes used to estimate the
detectability of broods. From 1 June 2001 to 15 August
2001 one point-count was conducted at exactly the same
location every 2 weeks (making a total of five point-counts)
between 1700 pm and 2100 pm on each lake of this nonrandom sample. Point-count localizations were chosen to
maximize visibility of open water, and on larger lakes an
additional point-count was performed on the same five
occasions as the first one but on a different location so as
to cover a remaining part of the lake. Lake area averaged
10.3 ha (SD = 16.2). For all lakes, it was never possible to
survey 100% of open water. Bird counts were only
undertaken under favourable weather conditions, i.e., when
dry, and not or only moderately windy.
Estimating probability of false absence
The probability of false absence of a species can be
estimated using the equation: probability of false absence
= α = (1–ps)N where N is the number of visits to the site,
and ps is the probability of detecting the species presence
(Kéry, 2002). The double-observer approach (Nichols et
al., 2000) was used to estimate ps and the risk of
false absence through estimates of species-specific
brood detection probabilities. Two observers surveyed lakes
from the non-random sample. At each pointcount, a designated ‘primary’ observer indicated to the other
(‘secondary’) observer all broods detected. The secondary
observer recorded all detections of the primary observer as
well as any brood not detected by the primary observer.
Observers alternated primary and secondary roles for a total
of 33 point-counts. Computation of detection probability
was made with program DOBSERV (Hines, 2000).
Estimating individual brood detectability
The robust-design approach (Pollock, 1982; Kendall,
Nichols & Hines, 1997) was used to compute individual
brood detection probability and adjusted densities of
broods. Because of the rarity of red-crested pochard as a
breeding species in the Camargue, the occurrence of two or
more broods of exact same age and size in the same lake
was considered highly unlikely. Each observed brood was
therefore identified (or ‘marked’) by the combination of its
age and size, taking into account that brood size might
decrease when getting older. The five successive pointcounts performed on each lake of the non-random sample
were split into three sub-counts of 15 minutes each, thus
constituting three secondary sessions within five successive
primary sessions, according to the terminology used in
Kendall et al. (1997). Within each site’s primary session,
successive secondary sessions were conducted 20–40
minutes apart from each other. On each of these 15
occasions, successive presence (1) or absence (0) of
identified broods was noted. This capture–recapture design
provides estimates of local survival (S), temporary
emigration (g′′) and 1- temporary immigration (g′, i.e., the
Red-crested pochard detection and abundance
probability that an individual absent during primary session
i is absent during primary session i+1) probabilities, as well
as capture (p) and recapture (c) probabilities which can be
considered here as detection probabilities, since
observations of known broods can be viewed as recapture
events. Detection probability computed from robust-design
thus corresponded to the probability of detecting an
individual brood given it is present in the study area.
Analyses were conducted with program MARK (White &
Burnham, 1999). Following Lebreton et al. (1992) and
Burnham & Anderson (1998), we used the Akaïke
Information Criterion with a correction factor for sample
size (AICc) to select the most parsimonious model. The
model with the lowest AICc is the one to be selected. As a
rule of thumb, two models with a difference in their AICc
< 2 were considered as statistically indistinguishable
(Lebreton et al., 1992). We initiated model selection with
a fully parameterized model without a priori hypothesis on
any parameters of the model.
Estimating adjusted densities
To estimate the breeding population size, a random
sample of lakes, assumed to be representative of the whole
Camargue, was used as a basis for inference on densities
at the Camargue scale, and extrapolation of a population
size. Twenty estates were randomly selected within the 80
largest of the total 230 estates of the Camargue, and two
lakes were randomly selected in 17 out of these 20 estates.
There was only one lake in the remaining three estates so
the total number of lakes of this random sample was 37.
Sampling among the 150 smaller estates would have
required obtaining more access permits to reach a sample
of 37 lakes. A large majority of the total 230 estates are
actually contiguous within a few remnant patches of
natural landscape, and some fragmentation effect
decreasing red-crested pochard densities in the smaller
estates was therefore considered unlikely.
On each of these 37 lakes, three monthly visits were
conducted in May, June and July during daytime,
following traditional design of field surveys for breeding
ducks (e.g,. Amat, 1984; Lillie & Evrard, 1994; Green,
1998; Pöysä et al., 2000; Pöysä, 2001). Lake shorelines
were visited throughout their length in order to maximize
detection of broods. Broods were intensively searched for
through telescopes on the visible water surface and, where
possible, within surrounding vegetation fringes. On each
lake where brood presence was noted, observed peak
number of broods (n) was recorded and was then adjusted
for incomplete detection of individual broods, using the
relation:
n
Nˆ =
pˆ
where N̂ is the estimated adjusted abundance, n is
the observed peak number among the three monthly
visits, and p̂ is the individual detection probability
estimated from the robust design (Barker & Sauer, 1992).
Adjusted brood counts were then expressed as densities
of broods per hectare of reedbeds, since previous work in
381
the Camargue (P. Defos du Rau, C. Barbraud &
J.-Y. Mondain-Monval, unpubl. data) has shown that redcrested pochard breeding occurrence is highly dependent
upon reedbed area. Density computation over a single
habitat type like reedbed permitted stronger inference on
abundance over the whole area because of better
homogeneity of brood density in reedbeds (owing to
causal ecological link) than in any other wetland habitat.
On the basis of estimated density in reedbeds and of
the total reedbed area for the Camargue calculated from
a GIS, an estimate of the brood population size could
be calculated.
For both the non-random and the random sample, surface
areas of Phragmites australis surrounding each lake and
forming islets within each lake were located in the field and
on aerial photographs, and calculated by GIS (Didger, 2000).
Since a unified GIS is not currently available for the entire
Camargue area, we used data sets of habitat-specific areas
from three geographically distinct GIS provided by the Parc
Naturel Régional de Camargue, the Réserve Nationale de
Camargue, the Observatoire des Zones Humides et des
Habitats de Camargue Gardoise, and the Tour du Valat
Biological Station. For two out of the three GIS data sets,
reedbed surface areas were available, but for the last GIS
data set, only the global area including reedbed and water
surface was available. For this particular area, brood
densities were expressed in number of broods per hectare of
reedbed and water and not in number of broods per hectare
of reedbed only. In this case, densities were extrapolated on
the basis of the total area of reedbed plus water surface. This
discrepancy between GIS use was not considered to bias
densities extrapolation severely because water and reedbed
surface are positively correlated, and thus red-crested
pochard breeding occurrence is linked to both water and
reedbed area. Hence, estimating brood densities over water
plus reedbed area was assumed to be as meaningful
biologically as densities over reedbed area singly.
RESULTS
Detectability parameters
Risk of false absence
The probability of detecting the presence of any brood on
a lake given it is present was 0.9259 (SE = 0.1033). With
three visits, the probability of a false absence in the data
set was α = 0.0004. The presence of any brood was
therefore highly unlikely to be undetected with three
successive visits. In other words, three visits only were
necessary to decrease the risk of false absence below
0.1%. Computation of population densities on sampling
sites can therefore be considered with confidence
regarding absence assessment.
Detection probability of individual broods
Starting with the general model where all parameters were
time-dependent, we did not detect significant timedependence in survival, temporary emigration and g′
382
P. DEFOS DU RAU ET AL.
Table 1. Modelling survival, temporary emigration and immigration
and capture and recapture probabilities of red-crested pochard broods
AICc ∆AICc
Model
S(.) g′′(.) g′(.) p(T,.)=c(T,.)
S(.) g′′(.) g′(.) p(.,.)=c(.,.)
S(.) g′′(.) g′(.) p(T,t)=c(T,t)
S(.) g′′(.) g′(.) p(T,t) c(T,t)
S(T) g′′(.) g′(.) p(T,t) c(T,t)
S(T) g′′(T) g′(.) p(T,t) c(T,t)
S(T) g′′(T) g′(T) p(T,t) c(T,t)
130.983
132.755
137.685
169.151
169.361
170.509
177.247
0.00
1.77
6.70
38.17
38.38
39.53
46.26
w
0.691
0.285
0.024
<0.001
<0.001
<0.001
<0.001
np Deviance
10
8
19
29
31
32
33
107.435
114.505
85.346
70.686
58.971
53.709
53.709
Modelling started from the fully parameterised model {S(T) g′′(T) g′(T) p(T,t) c(T,t),
where (T), (t) and (.) respectively indicate primary session-dependent, time-dependent
and constant parameters. For each model, we give AICc, ∆AICc, AICcWeight (w),
number of estimated parameters (np), and deviance. AICcWeights were estimated
following Anderson et al. (2000).
(Table 1). A model where capture and recapture probabilities
were set equal but time-dependent was preferred to a model
where capture and recapture probabilities differed (∆AICc
= 31.466). The two lowest AICc models included either
constant or primary session-dependent capture–recapture
rates (Table 1). Parameter estimates from the lowest AICc
models are shown in Table 2.
Because our main interest was to obtain an estimate of
p with reduced bias and increased precision, we used a
model-averaged estimator of p following Anderson,
Burnham & Thompson (2000). A model-averaging
procedure was run over the two best models to produce
an estimate of p with its associated unconditional standard
error. The estimate was p̂ = 0.5746 (unconditional SE =
0.0978), and its 95% confidence interval was 0.3815 –
0.7473. Thus, the probability of detecting one individually
identified brood on a lake given it is present was on
average 0.5746. This individual detection probability is
then used to adjust abundances.
Density and population size estimates
A total of 42 broods was observed in the non-random
sample, and a total of 14 broods was observed in the
random sample. Total reedbed areas for the non-random
and random samples were 232 ha and 217 ha, respectively.
Observed densities were 0.181 and 0.065 broods per
hectare of reedbed, respectively, and 0.012 broods per
hectare of reedbed plus water.
Table 2. Estimates of survival (S), temporary emigration (g′′), g′, and
capture probabilities (p) of red-crested pochard broods
Parameter
Estimate
SE
S
g”
g′
p session 1
p session 2
p session 3
p session 4
p session 5
0.598
0.734
…
0.588
0.662
0.391
0.708
0.475
0.084
0.100
…
0.111
0.105
0.081
0.102
0.197
Lower 95% CI Upper 95% CI
0.429
0.504
…
0.368
0.439
0.248
0.480
0.161
0.747
0.883
…
0.778
0.831
0.556
0.865
0.810
Estimates are from model S(.) g′′(.) g′(.) p(T,.)=c(T,.). Ellipses indicate
that g′ was not estimable, so SE and lower and upper 95% CI could not
be adequately estimated.
Taking into account detectability of individual broods,
an adjusted total of 42/0.5746 = 73 broods was estimated
to be present in the non-random sample, and an adjusted
total of 14/0.5746 = 24 broods was estimated to be present
in the random sample. Adjusted densities were thus of
0.3147 and 0.1106 broods per hectare of reedbed,
respectively, and 0.0206 broods per hectare of reedbed
plus water.
Adjusted density within the random sample was 0.1106
broods per hectare of reedbed with a 95% confidence
interval of [0.0863–0.1691], and 0.0206 broods per
hectare of reedbed and water [0.0161–0.0315]. Total
reedbed area for both GIS providing details for the
reedbed habitat only was 4502 ha and total reedbed plus
water area for the one GIS providing details for this
particular habitat association was 2964 ha. Total estimated
abundance for red-crested pochard broods in the
Camargue was thus 559 broods with a 95% confidence
interval of [436 – 855].
DISCUSSION
Population size
Our estimates of abundance of red-crested pochard broods
in the Camargue are much higher than previous estimates
of the breeding population in the Camargue and in France,
80–100 and 190–250 pairs, respectively (Boutin, 1994;
Dehorter & Rocamora, 1999). Results from our study
indicate that the detection probability used to estimate
abundance of broods was low, since the probability of
detecting individual broods was only 57%. Not taking
into account detectability in estimating abundance of
breeding red-crested pochards would thus result in major
underestimation. In addition, the robust design approach
allowed us to estimate temporary emigration, which was
high. The probability that a brood present during one
primary session was absent during the next primary
session was 73%. For example, this suggests that if 20
broods are counted during a first session nearly 15 of
these broods will be absent and not observed during the
second session. Furthermore, if ten broods are observed
during the second session, nearly half will be new broods
undetected during the first session. Thus, not taking
into account temporary emigration from one session to
another (as is usually done in ‘traditional’ surveys)
may lead to serious underestimation of abundance in
this species.
The low detection probability and the high temporary
emigration probability are probably a consequence of the
preferred habitat used for breeding, i.e. extensive reedbeds
of Phragmites australis with freshwater (P. Defos du Rau,
C. Barbraud & J.-Y. Mondain-Monval, unpubl. data). The
high temporary emigration found in this study may be
explained by families’ behaviour, being either in areas of
open water or hidden within the surrounding reedbeds
from one primary session to another.
For all the above reasons, we believe that previous
surveys strongly underestimated the number of breeding
pairs of red-crested pochard in the Camargue, and
Red-crested pochard detection and abundance
probably in France and Europe. In addition, and as
opposed to previous studies, our extrapolation of adjusted
densities used three geographic information systems
covering the entire Camargue area, and was based on
precise knowledge of habitat use by broods of red-crested
pochard issued from a companion study (P. Defos du Rau,
C. Barbraud & J.-Y. Mondain-Monval, unpubl. data).
Previous estimates based upon fieldwork by Rimbert
(1990) and Gaillardin (1991) only took into account
observed broods and paired adults to sum up an estimated
breeding population size. Furthermore, as shown by these
authors, the brood-rearing season can last from April until
August; since our observations started in May and ended
in July, some early and late broods may have remained
undetected by our study design, suggesting that our
estimate of the number of broods is a minimum.
Use of a non-random sample within areas of
concentrated use was necessary to attempt maximizing
precision in estimating detection probability and risk of
false absence. Use of a random sample was, of course,
necessary to calculate a brood density that would be
representative of the breeding distribution and density of
the species in the Camargue. Overall, within this random
sample, 1407 ha of wetland habitats were surveyed for an
estimated adjusted abundance of 24 broods. Another
approach would have been to survey the species randomly
in areas of reedbeds as a mean to increase sample size of
detected broods, but access permission to a sufficient
number of sampling sites would have been too difficult to
obtain from landowners.
Bias
Lack of homogeneity of the three GIS may have biased
our extrapolation. Both GIS for the central and western
part of the Camargue have been validated in the field, and
only 15% of the data set for the eastern part has not been
validated, but was only gathered on satellite images. The
eastern part of the Camargue is the smallest of all parts,
accounting for 19.4% of the wetland habitats and 13.4%
of the reedbed area of the whole Camargue. Thus, the
uncertainty on the reedbed surface area estimation through
GIS concerns about 2% of the total surface area, and is
therefore likely to have a negligible impact on our
extrapolation of adjusted densities. Indeed, total reedbed
area for the whole Camargue was calculated for 1984 to
be 5296 ha (Tamisier, 1990), which constitutes a result
very comparable to more up-to-date reedbed area
estimates used in the present study.
Our estimates of adjusted densities may be biased if
model assumptions were not fulfilled. We believe that the
two main assumptions of the double-observer approach
(independency in detection probabilities, equality in
detection distances, Nichols et al., 2000) were fulfilled. A
third important assumption (all broods have the same
probability of being detected) was verified since when we
ran program CAPTURE (Burnham & Overton, 1978;
Rexstad & Burnham, 1991) the model selection criteria
pointed towards model Mo as the most appropriate model
for four out of the five primary sessions, thereby
383
suggesting no heterogeneity in capture rates between
individuals. In addition to the assumptions for the
Cormack–Jolly–Seber model, the assumptions of analysis
under the robust design are (1) within each primary
session, the population is closed, (2) survival is equal for
animals that are in and out of the study area during any
primary session. In our study design, we believe that the
time intervals between the secondary sessions were small
enough to prevent mortality or permanent emigration from
occurring during primary sessions. Because nestlings of
broods that temporarily emigrated from the study area (i.e.
the open water in each sampled lake) could not fly (once
they could fly they probably left the lake and thus were
considered as dead or having permanently emigrated),
they were probably in reedbeds adjacent to the open water
during corresponding point-counts, but mortality risks
were assumed not to differ between broods within a
primary session
Conservation implications
Our estimate of the number of broods produced annually
in the Camargue stands as a minimum estimate of the total
breeding population size, since this estimate does not
include the breeding pairs that failed to hatch their eggs
and remained undetected in our study design. Hatching
success is highly variable from year to year and between
localities in ducks, and there is no reliable estimate of
hatching success for red-crested pochards in the
Camargue. However, our estimate of the brood population
size, combined with a relatively high hatching success of
80–90% as found in some species of the closely related
Aythya genus (Del Hoyo, Elliot & Sargatal, 1992), would
suggest a breeding population size of some 600–700 pairs
in the Camargue. Difficulties in detecting the species
explain this revaluation of the population size estimate
much more probably than a real increase, although such
a rise in the breeding numbers cannot be discarded with
certainty. In fact, the red-crested pochard population from
the western Mediterranean and central Europe is
considered by Wetlands International (2002) to be
increasing. Local declines in wintering numbers, as in the
Camargue or in northern Spain in the past decades, would
actually be due to a major switch in wintering areas from
southwestern to central Europe (Keller, 2000). Moreover,
the only two published breeding population censuses for
the Camargue (Blondel & Isenmann, 1981; Boutin, 1994),
which were used to argue for a national decline, were not
based on comparable field methods and produced only
unadjusted estimates.
We thus believe that the minimum breeding population
size of red-crested pochards in the Camargue is close to
600 pairs and shows no clear sign of a decline. Based on
a ‘strong decline’ statement and on the previous national
overall estimates thought to be below 250 breeding pairs,
the species was classified as ‘Endangered’ in the French
Red List of threatened birds (Dehorter & Rocamora,
1999). We think such rating may be overpessimistic, but
we do not deny that the French population of red-crested
pochard may still be threatened, at least by habitat loss
384
P. DEFOS DU RAU ET AL.
and degradation that still occurs in the Camargue
(Tamisier & Grillas, 1994; Mathevet & Tamisier, 2002),
and remains therefore vulnerable. Moreover, the impact
of hunting on the breeding population size remains
unknown at present. However, following EU Directive
79/409/EEC on the conservation of wild birds, and for the
first time since it first bred in the Camargue in 1894
(Mayaud, 1966), the red-crested pochard will not be
hunted in February. In fact, the largest part of the annual
red-crested pochard harvest seems to be achieved in
February, accounting, in available data, for 23% of the
annual harvest (ONCFS, unpubl. data). Therefore we
strongly recommend using this change in hunting
legislation as an experimental design of adaptive
management to evaluate the effect of this reduced harvest
at such a critical time in the species’ biological cycle on
the Camargue breeding population size.
Similar surveys taking into account detection
probability need to be undertaken regularly in the future
in order to detect trends in this population. Ideally, such
surveys should be complemented with some estimates of
hatching success. At a larger spatial scale, such surveys
should also be undertaken in the major breeding sites of
the red-crested pochard in France, but also in Europe.
Indeed, we are not aware of any red-crested pochard
population size estimate in other European breeding sites
that took into account detection probability, thereby
suggesting that the present world breeding population size
is probably underestimated. If our results are confirmed
(i.e., low detection probability), the red-crested pochard
may require a reassessment of its conservation status for
France and Europe. More generally, adjusting animal
abundances with detectability is a growing concern in
population biology and conservation (Buckland, Goudie
& Borchers, 2000). Design-based ecological studies (e.g.,
studies of habitat use) and conservation-orientated
surveys, like the present one, are both likely to benefit
from these developing methodologies. In particular, it is
likely that some more or less cryptic species considered
as threatened may have been classified as such on the basis
of unadjusted population size estimates, and that taking
into account their detectability in future field surveys will
lead to a reevaluation of their conservation status. Such
reevaluation should not decrease attention upon these
species, but might rather help to reorient conservation
priorities.
Acknowledgements
We thank all the landowners, Les Marais du Vigueirat and
Domaines Listel who allowed us to use their estates as
study sites. We are most grateful to Claire Lagaye and
Olivier Navarro from the Syndicat Mixte pour la
Protection et la Gestion de la Camargue Gardoise, Gaël
Hemery from the Parc Naturel Régional de Camargue and
Loïc Willm from the Tour du Valat Biological Station for
providing up-to-date GIS data. Thanks are due to
Sébastien Cayuela for highly skilled fieldwork, and to
Matthieu Guillemain, Dr Ken Norris, Alain Tamisier and
an anonymous referee for useful comments.
REFERENCES
Amat, J. A. (1984). Ecological segregation between red-crested
pochard Netta rufina and pochard Aythya farina in a fluctuating
environment. Ardea 72: 229–233.
Anderson, D. R., Burnham, K. P. & Thompson, W. L. (2000). Null
hypothesis testing: problems, prevalence, and an alternative. J.
Wildl. Mgmt. 64: 912–923.
Barker, R. J. & Sauer, J. R. (1992). Modeling population change
from time series data. In: Wildlife 2001: Populations: 182–194.
McCullough, D. R. & Barrett, R. H. (Eds). New York: Elsevier.
Blondel, J. & Isenmann, P. (1981). Guide des oiseaux de Camargue.
Neuchâtel and Paris: Delachaux & Niestlé.
Boutin, J. (1994). Nette rousse. In: Nouvel atlas des oiseaux
nicheurs de France 1985–1989: 142–143. Yeatman-Berthelot, D.
& Jarry, G. (Eds). Paris: Société d’Etudes Ornithologiques de
France.
Buckland, S. T., Goudie, I. B. J. & Borchers, D. L. (2000). Wildlife
population assessment: past developments and future directions.
Biometrics 56: 1–12
Burnham, K. P. & Anderson, D. R. (1998). Model selection and
inference: a practical information theoretic approach. New York:
Academic Press..
Burnham, K. P. & Overton, W. S. (1978). Estimation of the size of
a closed population when capture probabilities vary among
animals. Biometrika 65: 625–633.
Defos du Rau, P. (2002). Elements for a red-crested pochard (Netta
rufina) management plan. Game Wildl. Sci. 19: 89–141.
Dehorter, O. & Rocamora, G. (1999). Nette rousse Netta rufina. In
Oiseaux menacés et à surveiller en France: 376–377. Rocamora,
G. & Yeatman-Berthelot, D. (Eds). Paris: SEOF/LPO.
Del Hoyo, J., Elliot, A. & Sargatal, J. (1992). Handbook of the birds
of the world, vol. 1, Ostrich to ducks. Barcelona: Lynx.
Didger 2.01 (2000). Digitizing software. Golden, CO: Golden
Software Inc.
Gaillardin, C. (1991). La reproduction de la nette rousse en
Camargue. Unpublished report BEPA cynégétique, Vendôme.
Green, A. J. (1998). Comparative feeding behaviour and niche
organization in a Mediterranean duck community. Can. J. Zool.
76: 500–507.
Heiser, F. (1992). Breeding of the red-crested pochard in Swabia,
southwest-Bavaria. Ornithologische Anzeiger 31: 159–161.
Hines, J. E. (2000). DOBSERV: a double-observer approach for
estimating detection probability and abundance from avian point
counts. http://www.mbr-pwrc.usgs.gov/software/dobserv.html.
USGS-PWRC, Laurel, MD.
Karanth, K. U. & Nichols, J. D. (1998). Estimation of tiger densities
in India using photographic captures and recaptures. Ecology 79:
2852–2862.
Keller, V. (2000). Winter distribution and population change of redcrested pochard Netta rufina in southwestern and central Europe.
Bird Study 47: 176–185.
Kendall, W. L., Nichols, J. D. & Hines, J. E. (1997). Estimating
temporary emigration and breeding proportions using
capture–recapture data with Pollock’s robust-design. Ecology 78:
563–578.
Kéry, M. (2002). Inferring the absence of a species – a case study
of snakes. J. Wildl. Mgmt. 66: 330–338.
Krivenko, V. G. (1994). Red-crested pochard. In Birds in Europe,
their conservation status: 120–121. Tucker, G. M. & Heath, M.
F. (Eds). Cambridge: Birdlife International.
Lebreton, J. D., Burnham, K. P., Clobert, J. & Anderson, D. R.
(1992). Modeling survival and testing biological hypotheses using
marked animals: case studies and recent advances. Ecol. Monogr.
62: 67–118.
Lillie, R. A. & Evrard, J. O. (1994). Influence of macroinvertebrates
and macrophytes on waterfowl utilization of wetlands in the
Prairie Pothole Region of northwestern Wisconsin. Hydrobiologia
279/280: 235–246.
Red-crested pochard detection and abundance
Llorente, G. A. & Ruiz, X. (1985). Datos sobre la reproducción del
pato colorado Netta rufina (Pallas 1773) en el Delta del Ebro.
Misc. Zool. 9: 315–323.
Mathevet, R. & Tamisier, A. (2002). Creation of a nature reserve, its
effect on hunting management and waterfowl distribution in the
Camargue (southern France). Biodivers. Conserv. 11: 509–519.
Mayaud, N. (1966). Contribution à l’histoire de Netta rufina Pallas,
la nette à huppe rousse en Europe occidentale. Alauda 34:
191–199.
Nichols, J. D., Hines, J. E., Sauer, J. R., Fallon, J. W., Fallon, J. E.
& Heglund, P. J. (2000). A double-observer approach for
estimating detection probability and abundance from point counts.
Auk 117: 393–408.
Pollock, K. H. (1982). A capture–recapture design robust to unequal
probability of capture. J. Wildl. Mgmt. 46: 757–760.
Pöysä, H. (2001). Dynamics of habitat distribution in breeding
mallards: assessing the applicability of current habitat selection
models. Oikos 94: 365–373.
Pöysä, H., Elmberg, J., Sjöberg, K. & Nummi, P. (2000). Nesting
mallards (Anas platyrhynchos) forecast brood-stage food
limitation when selecting habitat: experimental evidence.
Oecologia 122: 582–586.
Rexstad, E. & Burnham, K. P. (1991). User’s guide for interactive
program CAPTURE. Colorado Cooperative Fish and Wildlife
385
Research Unit, Colorado State University, Fort Collins, Colorado.
Rimbert, S. (1990). Eco-éthologie de la nette rousse en Camargue.
Unpublished report, BEPA cynégétique, Charleville-Mézières.
Rosenstock, S. S., Anderson, D. R., Giesen, K. M., Leukering, T.
& Carter, M. F. (2002). Landbird counting techniques: current
practices and an alternative. Auk 119: 46–53.
Schneider-Jacoby, M. & Vasic, V. F. (1989). The red-crested
pochard Netta rufina breeding and wintering in Yougoslavia.
Wildfowl 40: 39–44.
Snow, D. W. & Perrins, C. M. (1998). The birds of the Western
Palearctic, vol. 1. Oxford: Oxford University Press.
Tamisier, A. (1990). Camargue, milieux et paysages. Évolution
1942–1984. Carte en couleur 1/80000. Arcane, Arles.
Tamisier, A. & Grillas, P. (1994). A review of habitat changes in
the Camargue: an assessment of the effects of the loss of biological
diversity on the wintering waterfowl community. Biol. Conserv.
70: 39–47.
Thompson, W. L. (2002). Towards reliable bird surveys: accounting
for individuals present but not detected. Auk 119: 18–25.
Wetlands International (2002). Waterbird population estimates.
Third Edition. Wageningen: Wetlands International Global Series.
White, G. C. & Burnham, K. P. (1999). Program MARK: survival
estimation from populations of marked animals. Bird Study 46
(Supplement): 120–139.
Chapter 3
BIOLOGICAL
CONSERVATION
Biological Conservation 125 (2005) 355–367
www.elsevier.com/locate/biocon
Incorporating uncertainty into analyses of red-crested
pochard habitat selection
Pierre Defos du Rau
a
a,*
, Christophe Barbraud
b,c
, Jean-Yves Mondain-Monval
a
Office National de la Chasse et de la Faune Sauvage, CNERA-Avifaune Migratrice, Le Sambuc 13200 Arles, France
b
Station Biologique de la Tour du Valat, Le Sambuc 13200 Arles, France
c
Centre dEtudes Biologiques de Chizé UPR 1934 CNRS, 79360 Villiers en Bois, France
Received 6 October 2004
Available online 31 May 2005
Abstract
Studies describing habitat use in animal species need to take into account detectability of individuals in order to reach more
robust conclusions. However, the importance of detectability in habitat selection analyses of rare and cryptic species has received
little investigation, although robust methods for estimating detectability have been made recently available. Understanding habitat
requirements should be an important management tool for the conservation of the red-crested pochard (Netta rufina), a rare duck
species in France and Europe. Three different lake samples (82 lakes in total) were conjointly used in 2000 and 2001 to assess the
species habitat requirements, using its presence, density and hatching dates as response variables. Risk of false absence was estimated using the double-observer approach at less than 0.001. A robust-design approach produced estimates of individual brood
detectability (0.545, SE = 0.053). Observed red-crested pochard densities were adjusted to time dependent detectabilities, and modelled as a function of habitat variables. Habitat fragmentation and low permanent water levels negatively affected brood densities.
Interestingly, these variables were not retained when modelling the unadjusted densities. This analysis showed that investigating
temporal variation in brood detectability was a crucial prerequisite in the study of this rare species habitat requirements. More generally, it strongly suggests that integrating detection probability and its variations in habitat use analyses of cryptic species of conservation concern may be an essential methodological step to reach more valid conclusions on habitat management.
2005 Elsevier Ltd. All rights reserved.
Keywords: Detection probability; Habitat selection; Survey methods; Management recommendations; Netta rufina
1. Introduction
Precise identification of habitat requirements for rare
or endangered species is often a crucial prerequisite for
developing sound conservation strategies. Numerous
studies of habitat use in birds have produced a great deal
of protection measures and management recommendations, notably for ducks and other related game species
*
Corresponding author. Present address: Office National de la
Chasse et de la Faune Sauvage, Délégation Régionale, 10bis route
dAx, 31120 Portet/Garonne, France.
E-mail address: [email protected] (P. Defos du Rau).
0006-3207/$ - see front matter 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.biocon.2005.04.006
(e.g., Kaminski and Weller, 1992; Green and El Hamzaoui, 2000). However, these design based field studies are
dependent upon population size estimation and thus
upon assumptions on detection. Few of these studies
have tested for the detection probability in estimating
abundance of their target species (e.g., Åberg et al.,
2000), or have accounted for the detectability of the target species when comparing its presence or abundance
over time or space (Thompson, 2002; Gu and Swihart,
2003; Freeman et al., 2003). Nichols et al. (2000) have
recently documented significant variations in detection
probabilities and then in numbers estimates among species and observers in point count studies. They highlight
356
P. Defos du Rau et al. / Biological Conservation 125 (2005) 355–367
the need to measure detection probabilities when comparing population sizes over space or time. They proposed the double-observer approach to model
detection probabilities, and suggested that capture probability estimated in capture-recapture models could be
used as a measure of detectability. Here we investigate
two different aspects of the detectability of the broods
of a rare breeding bird, the red-crested pochard, using
two methods: the double-observer approach (Nichols
et al., 2000) and the Pollocks robust-design approach
(Kendall et al., 1997). The detection probabilities obtained from both approaches are then used to obtain
unbiased estimates of brood population size and to identify key variables determining brood habitat use and
choice.
The red-crested pochard (RCP) is a migratory diving
duck that breeds in Central Asia, around the Black Sea
and in Western Europe. The species is listed in appendix
III of the Berne Convention, in appendix II of the Bonn
Convention, in appendix II/2 of the ‘‘Birds’’ Directive
79/409 of the E.E.C and in Annex II of the African-Eurasian Waterbird Agreement. Although it represents a
strong conservation concern, the species is hunted in
France, Portugal, Romania and Spain. A management
plan produced at the request of the European Union
Commission (Defos du Rau, 2002) stresses the need to
identify regional or local factors affecting breeding habitat choice.
Based on previous studies on RCP, we made the following predictions about factors affecting brood habitat
selection at the local and landscape scales. We expect an
increase in density or occupancy probability:
(1) as habitat fragmentation decreases (Stephens
et al., 2003), that is when the number of reclaimed
or degraded lakes remains minimal within wetland
complexes (Weller, 1988),
(2) as abundance of both preferred nest site helophytes and food macrophytes increases, because
breeding RCP is often associated with both reedbeds and extensive macrophytes beds (Snow and
Perrins, 1998),
(3) as variation of water levels decreases and water
depth increases, then decreasing the risk of drying
up, which is high in Mediterranean wetlands
(Kaminski and Weller, 1992),
(4) as predator and inter-specific competitor densities
decrease, or, conversely
(5) as subsequently attracted predator density
increases. Although few studies have succeeded
in demonstrating an impact of intra- or inter-specific competition on brood spacing and space use
(DuBowy, 1991; Anderson and Titman, 1992),
both competition and predation are strongly suspected to influence habitat use by broods (Kaminski and Weller, 1992).
2. Methods
2.1. Study species
The RCP is a rare breeding duck species in France,
with an estimated 190–250 breeding pairs concentrated
on three wetland areas along the Rhône River (Boutin,
1994; Dehorter and Rocamora, 1999). The southernmost of these strongholds is situated in the Rhône River
delta, the Camargue, on the Mediterranean. The breeding season is prolonged: egg laying occurs from late
March until early July. The species breeding habitat
remains poorly documented in France (Broyer and
Daléry, 2000), although it has been reported to change
drastically in the last 30 years in the Camargue from
saltmarshes and saltworks to fresh marshes and lakes
covered with reedbeds. Increasing brood predation by
yellow-legged gull (Larus michahellis) has been suspected to cause this habitat switch (Boutin, 1994). As
with many other duck species, RCP is rather cryptic in
its breeding behaviour and habits, and broods mostly
come out of the vegetation fringe on open water in late
afternoon and in the evening. RCP parasites other ducks
broods (Amat, 1991) but this behaviour remains rare in
the Camargue and is thus supposed to have only marginaly biased brood counts.
2.2. Study site
The Camargue is a vast wetland area of 145,000 ha.
Natural habitats (freshwater and brackish marshes,
including reedbeds and temporary flooded salt meadows)
account for approximately 58,000 ha split between protected and hunting estates, for a total of 230 estates. The
number of RCP was estimated at 80–100 pairs for the
whole Camargue (Boutin, 1994). The species is known
or suspected to regularly breed in 27 estates when breeding habitat is available, i.e. when flooding and water management conditions are adequate. In both protected and
hunting estates, disturbance level is generally kept low
on purpose during the waterfowl breeding season.
2.3. Sampling
Only 9 of these 27 regularly occupied estates were
easily accessible, flooded in 2000 and 2001, and ones
for which we received permission to conduct surveys.
Lakes of these 9 estates constituted consecutive samples
of 34 lakes in 2000 and 40 lakes in 2001. Within these
annual samples, we compared lakes occupied versus
not occupied by RCP broods in both years. Because
they were situated within 9 flooded estates amongst
the 27 favoured ones, unused lakes were then available
as well as accessible to the species (Jones, 2001).
P. Defos du Rau et al. / Biological Conservation 125 (2005) 355–367
For habitat model validation purpose, a further 20
estates were randomly selected within the 80 largest of
the total 230 estates of the Camargue, and 2 lakes were
again randomly selected in 17 of these 20 estates. There
was only 1 lake in the remaining 3 estates, so the number
of lakes of this random sample was 37.
2.4. Bird surveys
Following recommendations by Drapeau et al.
(1999), 6 consecutive point counts were conducted during the 2000 and 2001 breeding seasons on each of the
34 and 40 lakes, respectively (Defos du Rau et al.,
2003). These point counts were conducted between
1700 and 2100 pm during 20 min and by the same observer in summer 2000 and during 30 min and by the same
two observers in summer 2001. Broods of RCP were
intensively searched for through telescopes on all the visible water area and within surrounding vegetation
fringes. Brood presence, abundance, size and age were
recorded. Partially hidden broods were followed within
the vegetation until their size and age could be confidently estimated. The age of broods was estimated in
weeks, based on size of chicks (Office National de la
Chasse, 1982). On each of the 37 randomly selected
lakes, three monthly broods counts were conducted by
the same observer in May, June and July 2001 during
daytime and lake shorelines were visited throughout
their length in order to maximise detection of broods.
All counts were undertaken only under favourable
weather conditions, i.e. when dry and not or only moderately windy.
2.5. Estimation of brood detection probabilities and
abundance
2.5.1. Robust-design approach
Because of the rarity of breeding RCP in the Camargue, the occurrence of 2 or more broods of exactly same
age and size in the same wetland complex was considered highly unlikely. Each observed brood was therefore
identified (or ‘‘marked’’) by the combination of their age
and size. The same capture-recapture design, or Robustdesign (Kendall et al., 1997), used in Defos du Rau et al.
(2003) was used here. It provides estimates of local survival rates (S), temporary emigration (g00 ) and immigration (g 0 ) probabilities and population sizes, as well as
capture (p) and recapture (c) probabilities which can
be considered here as detection probabilities, as observations of known broods can be viewed as recapture
events. The detection probability computed from the robust-design thus corresponded to the probability of
detecting an individually marked brood given its presence in the study area. This detectability parameter
was used to correct the observed brood densities (in
broods/ha) and to obtain unbiased abundance estimates.
357
Analyses were conducted with program MARK (White
and Burnham, 1999).
2.5.2. Double-observer approach
The same double-observer method (Nichols et al.,
2000) used in Defos du Rau et al. (2003) was used here
to provide estimates of species specific and observer specific detection probabilities, as well as population sizes
with program DOBSERV (Hines, 2000). The detection
probability computed from the double-observer approach corresponded to the probability of detecting any
brood given it is present in the study area. This detectability parameter was used to estimate the probability of false
absence, i.e. the probability of not detecting any brood of
the species on a site where the species is actually present.
2.6. Habitat survey
The present survey focused on habitat use at the wetland scale, because this scale is the level of most practical
conservation and management recommendations. The
survey in 2000 was planned as a pilot study. Objectives
of the analyses of the 2000 dataset were not to draw
inferences about habitat use of RCP but to estimate
model order (Mac Nally, 2000), as the number of necessary explanatory variables. Moreover, this analysis was
used to identify which categories of variables would
potentially influence the response variable and would
therefore require more detailed measurements. This preliminary analysis was also dedicated to basic variable
selection as a first mean to lessen multicollinearity and
type-I errors within subsequent regression analyses
(Mac Nally, 2000). A set of 47 predictor variables
(Appendix) was measured for each lake, mainly based
on Kaminski and Weller (1992). These included presence and densities of potential predators and competitors, food resources, water quality and management,
wetland structure and habitat composition:
2.6.1. Wetland structure and habitat types composition
Breeding RCP is associated with Phragmites australis
(Snow and Perrins, 1998), Typha (Broyer and Daléry,
2000) and Juncus maritimus (Llorente and Ruiz, 1985)
beds and salt scrub of Arthrocnemum glaucum (Amat,
1982). Surface areas of patches and islets of Phragmites
australis, Typha spp., Juncus spp., Arthrocnemum spp.,
Scirpus spp. within each lake were located both in the
field and on aerial photographs, and calculated by Geographical Information System (GIS, Didger, 2000). In
addition, wetland mosaic structure was described for
each lake by the distance to the closest lake and the
mean distance to the 5 closest lakes. Distances were
measured by GIS and log-transformed due to unfavourable ratio pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
of mean
to variance. The index
ffi
perimeter=2 p area (shoreline I.) was used to evaluate shoreline length relative to lake area (Joyner, 1980).
358
P. Defos du Rau et al. / Biological Conservation 125 (2005) 355–367
2.6.2. Water quality and management
Water levels and salinity were measured monthly for
each lake during the egg laying period from April to
June and standard deviation of water levels was calculated for each lake. Maximum water level, which rarely
exceeded 60 cm, was also calculated in April from a
sample of 2–6 sampling points according to size of each
lake.
2.6.3. Food resources
The RCP is primarily a herbivore, depending mainly
on Chara spp., Potamogeton spp. and Ruppia spp. beds
(Snow and Perrins, 1998). Submerged macrophytes were
sampled in the entire water column along 5 parallel 1-mlong line-transects (Kent and Coker, 1992). Transects
were 5 m apart on 20-m-long sampling lines chosen at
random both in the centre and perpendicular to the
shores of each lake. Transect sample sizes varied from
10 to 100 depending on lake size. Relative frequencies
of genus Chara, Myriophyllum, Najas, Ludwigia, thickleaved and thin-leaved Potamogeton (Pot.pec frq.) were
calculated for each lake.
2.6.4. Predator and competitor communities
Marsh harrier (Circus aeroginosus) (Opermanis,
2001), black kite (Milvus migrans), yellow-legged gull
(Boutin, 1994) and corvids (Johnson et al., 1989),
including magpie (Pica pica) and crow (Corvus corone)
could be considered as the main avian predators of
RCP clutches and broods in the Camargue. The abundance of individuals was estimated at each point count.
Following Kaminski and Weller (1992), competition
should be integrated as a factor of habitat selection.
Broods of mallard (Anas platyrhynchos), mute swan
(Cygnus olor), coot (Fulica atra) and great-crested grebe
(Podiceps cristatus) were counted at each point counts as
likely competitors because they can use the same trophic
and/or spatial niche as RCP. Detectability issues were
not considered for these widespread, abundant and/or
obtrusive species.
2.7. Habitat use analysis
2.7.1. Model order
We first evaluated the number and level of precision of habitat variables needed in subsequent regression analyses. Data obtained on lakes sampled in 2000
were devoted to this selection (Mac Nally, 2000). In
2000, broods were observed on 9 lakes (classified as
1) versus the remaining 25 lakes where broods were
considered absent (classified as 0). The unfavourable
ratio of the number of observations to the number
of variables required a preselection using forward
stepwise discriminant function analysis to eliminate
variables that did not contribute to differences between occupied and unoccupied lakes. Tolerance
threshold (computed as 1 R2 of the respective variable with all other variables in the model) was set at
0.5 to prevent major redundancy between variables;
inclusion and exclusion F-statistics were set at 1 and
0, respectively. Furthermore, variables were checked
for colinearity between each other and for biological
meaning of their trend relative to the response variable. Five variables were finally retained for logistic
regression (Alldredge et al., 1998) through a generalized linear model (GLM) procedure to identify
through Akaike Information Criteria (AICc) based
exhaustive search (Anderson et al., 2001) the reduced
set of variables that best separated occupied and
unoccupied lakes in 2000 (Table 1).
On the basis of this preliminary analysis, the order of
a logistic regression on RCP breeding site occupancy
would be expected to be a set of 3, or possibly 2 or 4
variables. This preliminary analysis conducted on the
2000 dataset was not used to eliminate some explanatory
variables but to improve measurements within variable
categories of apparently highest explanatory power.
Water management, food resources and predator abundance variables appeared in 7 models of DAICc < 2, justifying subsequent increase in measurement precision of
all variables within these categories in 2001 (Appendix),
unlike wetland structure variables for which no
Table 1
Logistic regression models of habitat variables determining lake occupancy by red-crested pochard in 2000 in the Camargue, France
Variables
gull abund. waterdepth
gull abund. waterdepth
waterdepth water area
gull abund. waterdepth
gull abund. waterdepth
gull abund. waterdepth
gull abund. waterdepth
gull abund. waterdepth
water area
water area shoreline I.
Pot.pec frq.
shoreline I.
Pot.pec frq. water area
Pot.pec frq. water area shoreline I.
np
DAICc
AICc
4
5
3
3
4
4
5
6
0
0.12
0.61
0.66
0.95
1.25
1.59
2.20
39.23
39.34
39.83
39.89
40.18
40.47
40.82
41.43
Only the 7 best models (lowest AICc) are shown. np indicates the number of estimated parameters. Goodness-of-fit : v228 ¼ 31.16, P = 0.31.
P. Defos du Rau et al. / Biological Conservation 125 (2005) 355–367
improvement could be made with respect to the precision level reached in the 2000 dataset.
2.7.2. Water quality and management
In 2001, water levels were surveyed every 2 weeks for
each lake from April to July. Standard deviations and
percent variation of water levels were calculated for each
lake in different periods within the egg laying period.
Maximum water level was also calculated in April on
a larger sample than in 2000, with 2–16 sampling points
depending on lake size.
2.7.3. Food resources
Additional taxa (Ceratophyllum spp., Ranunculus
spp., Scirpus spp., Zannichellia spp. [zanic. freq.], Potamogeton pectinatus, P. fluitans, P. pusillus) of submerged
macrophytes were sampled in 2001 compared to 2000,
and their relative frequencies were obtained for each
lake.
2.7.4. Predator community
In addition to abundance, duration of stay was estimated in individuals.min at each point count in 2001.
These predator abundance indices were log-transformed
owing to unfavourable ratio of mean to variance. Because of reported impact of carnivores on duck nests
(Johnson et al., 1989), presence of terrestrial predators,
such as dogs (Canis canis), foxes (Vulpes vulpes) and
Mustelids, Mustelidae spp., was assessed in 2001 by
counts of faeces performed on 1-km transects and repeated once 2 weeks later. These transects were conducted on the closest track to each lake, and all counts
were performed within the same wetland habitat following a long period of dry weather (Wilson and Delahay,
2001). Rats Rattus spp. are not known as duck nests predators in the Camargue and were thus not considered.
2.7.5. Variable selection
We performed a discriminant analysis using the
same threshold values as in 2000 on the 2001 dataset
to eliminate variables that did not appear to contribute significantly to differences between occupied and
apparently unoccupied lakes. This analysis retained 8
variables that were then checked for colinearity between each other and for ecological meaning of their
trend relative to the response variable. Finally, 3 variables of consistent ecological meaning with respect to
RCP habitat use were intercorrelated: reedbed area
and Zannichellia frequency (r = 0.62), and reedbed
area and water depth (r = 0.50). Indeed, growth of
both reeds Phragmites autralis and Zannichellia spp.
are favoured by the same water regime of annual temporary flooding and drying up (Grillas, 1992; Mesléard and Pérennou, 1996). Also, reedbed surface
should be larger when water surface, and thus water
depth, increases.
359
To avoid statistical multicolinearity and to reduce
the number of parameters of interest in subsequent
GLM, these 3 variables were included in a principal
component analysis. The first component (PC1, total
variance 64.6%) was used as an index of large reedbed
areas, high Zannichellia frequency and, to a lesser extent, high water depth (respective factor loadings
[unrotated]: 0.90, 0.80 and 0.70). The second component (PC2, total variance 24.5%) was used as an index
of low water depth (factor loading [unrotated]: -0.69).
This approach was used for causal inference purpose
only, but colinear variables were retained individually
in predictive models as colinearity does not hamper
predictive power of a model but only its causal inference power.
Thus, 5 variables were retained for AICc based
model exhaustive search: standard deviation of water
levels during egg laying period from April to June
(waterlevelSD), Myriophyllum frequency (myrio. freq.),
mean distance to 5 closest lakes (mdist5lakes), PC1
and PC2.
2.7.6. Regression models
The several logistic regression models were compared
by DAICc. Goodness-of-fit and overdispersion were
checked for by Pearson v2 and variance inflation factor
^c (Anderson et al., 2001). For validation and predictive
purpose, the order-3 model with lowest AICc was run
on the random sample of 37 lakes. Percentage of correct
classification of occupied and unoccupied lakes was
plotted against a gradient of occupancy probability
threshold.
A potential bias for this logistic regression analysis
was false absence. Detection probability p 0 obtained
from the double-observer method, i.e. the probability
to detect brood presence, was used to assess the risk a
of false absence, i.e. of undetected presence, and to evaluate efficiency of the present monitoring of n = 6 point
counts, following Kéry (2002): probability of not detecting a brood on a site given that it is present = probability of false absence = a = (1p 0 )n.
Observed brood densities approximately followed a
Poisson distribution, and were modelled in a Poisson
regression with a log-link function using the same 5 selected variables. Brood densities were then adjusted by
detection probabilities obtained from the lowest AICc
robust-design model and applied to the observed brood
numbers following the relation:
Daj ¼
^j
N
WSj
^j ¼
with N
X nij
;
^pi
i
where Daj is adjusted density for lake j, nij is observed
brood number during period i on lake j, WSj is water
^ j is total adjusted brood number
surface area of lake j, N
360
P. Defos du Rau et al. / Biological Conservation 125 (2005) 355–367
for lake j, ^
pi is detection probability estimated through
robust-design at period i.
Adjusted brood densities were then modelled following the same regression design and with the same dataset
as for observed broods densities.
Fitness consequences of habitat selection were examined using hatching date as a measure of fitness through
brood survival (Blums et al., 2002). Since many broods
were observed irregularly, i.e. often after most duckling
mortality had occurred, hatching date, as inferred from
estimation (in weeks) of each broods age, was therefore
the only response variable measuring consequences of
lake choice in terms of reproductive success. Occupied
lakes of both regular and random lake samples from
2001 were pooled in this analysis and a mean hatching
date was calculated for each lake (n = 22).
Variables were selected and checked for multicolinearity and coherent biological meaning following the
same process as for previous habitat occupancy analyses. As for previous analysis, variables were selected
through discriminant function analysis of the 50% earliest hatching dates versus the 50% latest. Frequencies of
Chara spp. and of Myriophyllum spp., as well as reedbed
area were selected for their consistent and unrelated
influence on breeding success via hatching date. In addition, the 5 variables selected for the previous analyses of
habitat occupancy were expected to influence breeding
success through habitat choice and they were therefore
checked for multicolinearity and consistent biological
meaning with respect to this pooled dataset. All analyses
were performed using Statistica (Statistica, 2000).
3. Results
3.1. Bird surveys
In 2001, 42 broods were observed on 17 out of the 40
lakes of the main sample. In the random sample of 37
lakes, 13 broods were observed on 8 lakes. Only four
lakes occupied in 2000 were again found occupied in
2001.
3.2. Estimation of brood abundance
Because robust-design models generated many
parameters with identifiability problems, we used program CAPTURE (Rexstad and Burnham, 1991) to initiate model selection and to identify a starting model
with either a constant probability of detection, M(o),
or incorporating any combination of variations in time,
M(t), behaviour, M(b), and individual heterogeneity,
M(h), in detection probability.
Model M(o) was selected on 4 out of 5 primary sessions, as the starting basis for choosing the best robust-design model. In particular, capture and
recapture probabilities were set equal as capture events
corresponded to first observations and thus would not
influence subsequent observations probabilities. Two
observers performed the robust-design protocol, 1 on
the first 2 primary sessions, 1 on the last 2 primary sessions, both observers switching with each other on the
third primary session. There were therefore 3 different
possible observer dependent periods likely to affect capture probabilities. Also, we suspected that time had an
impact on brood survival through hatching date (Blums
et al., 2002), and on temporary emigration and immigration rates through changes in water levels. Therefore,
model selection procedure included comparisons of time
and observer based models and we started model selection with model [S(t), g00 (t), g 0 (t), p(t,T) = c(t,T)].
The model with time dependent survival and capture
probabilities differing between 2 observer dependent
periods (primary sessions 1, 2, 4 and 5 versus primary
session 3) had the lowest AICc and was used to estimate
the number of broods at each primary session and their
detection probability (Table 2).
Table 2
Modelling survival, temporary emigration and immigration, and capture probabilities of red-crested pochard broods in 2001, Camargue, France,
under the robust-design protocol
Model
AICc
DAICc
wi
np
Deviance
{S(t) g00 (Æ) g 0 (Æ) p(1 = 3,2) = c(1 = 3,2)}
{S(t) g00 (Æ) g 0 (Æ) p(3) = c(3)}
{S(t) g00 (t) g 0 (t) p(3) = c(3)}
{S(t) g00 (t) g 0 (Æ) p(3) = c(3)}
{S(Æ) g00 (Æ) g 0 (Æ) p(3) = c(3)}
{S(t) g00 (Æ) g 0 (Æ) p(1,2 = 3) = c(1,2 = 3)}
{S(t) g00 (Æ) g 0 (Æ) p(1 = 2,3) = c(1 = 2,3)}
{S(t) g00 (Æ) g 0 (Æ) p(.,.) = c(.,.)}
{S(t) g00 (Æ) g 0 (Æ) p(t,T) = c(t,T)}
116.40
118.92
120.01
120.11
124.53
124.93
126.56
129.29
136.97
0.00
2.52
3.61
3.71
8.13
8.53
10.16
12.83
20.57
0.604
0.171
0.099
0.094
0.010
0.008
0.004
0.001
0.000
9
10
13
13
7
10
11
11
22
95.55
95.38
87.84
87.94
108.81
101.38
100.23
102.90
72.73
(t) and (Æ), respectively, indicate time dependent and constant parameters; p(3) indicates that there are 3 different capture probabilities corresponding
to the 3 observer dependent periods; p(1 = 2,3) indicates that capture probabilities are equal for the first 2 periods; p(1,2 = 3) indicates that capture
probabilities are equal for the last 2 periods; p(1 = 3,2) indicates that capture probabilities are equal for the first and third periods; wi indicates the
AICc weights (Burnham and Anderson, 2002); and np indicates the number of estimated parameters.
P. Defos du Rau et al. / Biological Conservation 125 (2005) 355–367
361
Table 3
Number of red-crested pochard broods counted in 2001, Camargue, France, missed in one detectability design but not in the other, detection
probabilities and estimated number of broods with their 95% confidence intervals from the double-observer and the robust-design approaches
Robust-design third primary session
Double observer
Broods
counted
Broods
missed
Detection
probability p
SE (p)
Estimated
abundance N
SE (N)
95% CI
16
15
1
2
0.359
0.926
0.084
0.103
21
16
4.13
2.14
[17; 36]
[15; 27]
As stated in the method section, detection probabilities estimated by both approaches are not comparable due to different methodologies.
As both double-observer and robust-design approaches were applied to the same lake subset (n = 33)
and as the double-observer protocol was performed on
the third of the 5 primary sessions of the robust-design,
abundance estimates obtained by these 2 methods were
compared (Table 3). Confidence intervals of abundances
estimated from both methods largely overlapped.
3.3. Habitat use analysis
Using the species specific detection probability estimated from the double observer approach, the probability of a false absence in the dataset was: a = (1 p 0 )n with
p 0 = 0.9259 and n = 6 visits, giving a = 1.7 · 107. Brood
presence was therefore highly unlikely to be undetected
considering the present monitoring of 6 successive visits.
3.3.1. Presence
Models of habitat variables affecting lake occupancy by RCP were selected by comparing their AICc
(Table 4). The 4 best models included 2 to 4 habitat
variables indicating that marshes with relatively higher and more stable water levels, with larger reed
and/or Zannichellia beds, and situated closer to other
marshes were preferentially chosen. Parameters of the
order-3 selected model (DAICc = 0) were estimated
(Table 6).
For predictive purposes, we ran models on the 2001
dataset but with the 3 colinear variables included individually (Table 5). Order-3 selected model for this analysis was a combination of water level SD, Zannichellia
Table 4
Logistic regression models of habitat variables (taking into account
colinearity) determining lake occupancy by red-crested pochard in
2001, Camargue, France
Variables
waterlevelSD
waterlevelSD
waterlevelSD
waterlevelSD
waterlevelSD
mdist5lakes PC1
PC1
PC1 PC2
mdist5lakes PC1 PC2
myrio. freq. mdist5lakes PC1
np
AICc
DAICc
4
3
4
5
5
39.65
40.07
40.24
40.36
41.97
0
0.42
0.60
0.72
2.32
Only the 5 best models (lowest AICc) are shown. Goodness-of-fit:
v234 ¼ 35.09, P = 0.42.
Table 5
Predictive logistic regression models of habitat variables determining
lake occupancy by red-crested pochard in 2001, Camargue, France
Variables
np
DAICc
AICc
waterdepth waterlevelSD zanic. freq. reedbed
waterlevelSD zanic. freq. reedbed
waterlevelSD zanic. freq. mdist5lakes reedbed
waterdepth waterlevelSD zanic. freq.
mdist5lakes reedbed
waterlevelSD mdist5lakes reedbed
waterdepth waterlevelSD mdist5lakes reedbed
waterdepth waterlevelSD myrio. freq.
zanic. freq. reedbed
5
4
5
6
0
0.11
0.42
0.99
41.96
42.07
42.38
42.95
4
5
6
1.25
1.54
2.46
43.21
43.50
44.42
Only the 7 best models (lowest AICc) are shown. Goodness-of-fit:
v233 ¼ 34.93, P = 0.38.
Table 6
Parameters estimates and type-3 likelihood ratio tests for red-crested
pochard habitat variables (2001, Camargue, France) from order-3 best
models when colinear variables are included within PCA factors or
singly
Colinear variables
Parameter estimate
SE
v2
P
As PCA factors
Intercept
waterlevelSD
mdist5lakes
PC1
7.53
0.24
1.08
3.36
3.75
0.13
0.68
1.20
4.15
2.90
17.74
0.042
0.089
<0.001
Included singly
Intercept
waterlevelSD
zanic. freq.
reedbed
0.24
0.28
0.09
0.39
0.98
0.16
0.08
0.15
4.32
4.39
11.05
0.038
0.036
<0.001
frequency and reedbed area. This predictive model was
run on the dataset of the random sample of lakes
(n = 37), which allowed for validation of the general
model. Model computed from the 2001 dataset correctly
predicted up to 84% of occupancy of the randomly sampled lakes (Fig. 1), depending on the chosen probability
threshold separating presence and absence. At optimal
occupancy probability threshold (0.05–0.21), distributions of observed and expected occupancy frequencies
were not different (v236 ¼ 2, P > 0.20) and model predictive performance was maximised.
362
P. Defos du Rau et al. / Biological Conservation 125 (2005) 355–367
Table 8
Parameter estimates for the selected model relating hatching date of
red-crested pochard broods with habitat variables in 2001, Camargue,
France (F2,19 = 6.39, P = 0.008)
0.9
0.8
prediction success
0.7
0.6
0.5
0.4
0.3
Variables
Parameter estimate
SE
t
P
Intercept
waterdepth
chara freq.
0.35
0.47
0.18
0.18
7.17
1.93
2.59
<0.001
0.069
0.018
0.2
0.1
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
threshold for probability of lake occupancy
Fig. 1. Prediction success in presence/absence of RCP broods as
predicted by best model selected in Table 5 and run on the random
sample of lakes.
3.3.2. Brood densities
Densities of broods were approximately Poisson distributed and were plotted with a log-link function
against the same explanatory variables as for brood
presence/absence response variable. Goodness-of-fit
(GOF) tests indicated that retaining any explanatory
variables to describe brood densities patterns was unsafe. The selected model (GOF test: Pearson
v238 ¼ 17.09, P = 1) included the mean distance to 5 closest lakes as the only explanatory variable but its effect
was relatively small (Type-3 likelihood test: v2 = 2.45,
P = 0.12).
3.3.3. Hatching date
As expected, brood size was negatively correlated
with hatching date (Type-3 likelihood test: v2 = 5.94,
P = 0.01) and brood age (Type-3 likelihood test:
v2 = 4.22, P = 0.04). Mean hatching date was 10 May,
SD = 19 days, n = 50 broods, and early hatching dates
then indicated higher breeding success. Water depth,
mean distance to 5 closest lakes, frequency of Myriophyllum spp., as well as reedbed area were unrelated
and retained. These 4 variables and the frequency of
Chara spp. (not linear with any other variable) were included in a multiple linear regression with hatching date
as the response variable (Table 7).
The most conservative model included water depth
and frequency of Chara spp. as negative linear predictors of hatching date. This model (Table 8) explained
34% of the variance in hatching date. Thus, lakes in
which RCP bred early and hence had higher breeding
success had significantly deeper water and more Chara
macrophytes.
3.4. Re-analysis of habitat use with adjusted densities
Detection probability estimates obtained from the selected model under the robust-design approach were
used to adjust brood numbers observed during each of
the 2 observer dependent periods. Adjusted densities
of broods were approximately Poisson distributed and
were submitted to the exact same analyses as the unadjusted using the same set of variables (Table 9).
The best models explaining adjusted brood densities
showed better fit ð^c ¼ 0.77Þ than for the observed densities ð^c ¼ 0.41Þ and included mean distance to 5 closest
lakes, PC2 for water shallowness and frequency of
Table 9
Poisson regression models for adjusted densities of red-crested pochard
broods in 2001, Camargue, France
Variables
np
DAICc
AICc
mdist5lakes
mdist5lakes PC2
myrio. freq. mdist5lakes
myrio. freq. mdist5lakes PC2
waterlevelSD myrio. freq. mdist5lakes PC2
waterlevelSD mdist5lakes
waterlevelSD mdist5lakes PC2
myrio. freq.
waterlevelSD myrio. freq. mdist5lakes
2
3
3
4
5
3
4
2
4
0.00
0.28
0.49
0.72
1.33
1.47
1.53
1.59
1.76
53.63
53.91
54.11
54.35
54.96
55.10
55.15
55.21
55.38
Goodness-of-fit: v234 ¼ 26.19, P = 0.83. n = 40.
Table 7
Linear regression models of habitat variables determining mean
hatching date of red-crested pochard broods in 2001, Camargue,
France
Table 10
Parameters estimates and type-3 likelihood ratio tests for habitat
variables from best order-3 model for adjusted broods densities of redcrested pochard in 2001, Camargue, France
Variables
np
AICc
DAICc
Variables
Parameter estimate
SE
v2
P
waterdepth chara freq.
chara freq.
waterdepth chara freq. mdist5lakes
3
2
4
187.91
189.13
190.52
0
1.22
2.61
Intercept
myrio. freq.
mdist5lakes
PC2
3.33
0.02
0.89
0.40
1.94
0.02
0.41
0.27
2.04
4.89
2.24
0.154
0.027
0.134
Only the 3 best models (lowest AICc) are shown. n = 22 lakes.
P. Defos du Rau et al. / Biological Conservation 125 (2005) 355–367
Myriophyllum spp. (Table 10). Thus, densities were
higher in deep water lakes with low frequency of Myriophyllum and situated within wetlands complexes with a
relatively dense network of lakes.
4. Discussion
4.1. Habitat selection
Our results indicate that 7 habitat variables are
important in determining habitat use and selection by
RCP broods (Fig. 2). Predators and competitors did
not appear to influence habitat use and selection by
RCP in the Camargue, but the possible impact of yellow-legged gull should be monitored in a near future.
According to the existing literature on breeding
habitat of the RCP (Boutin, 1994; Defos du Rau,
2002), choice for larger reedbed areas seems the most
consistent and meaningful ecological feature. Water
depth is also a crucial factor, and highlights the effect
of water management on the presence, density and
breeding success of RCP in the Camargue. Preferred
use of deeper lakes was already observed by Amat
(1984) and expected from an opportunistic diving
duck such as the RCP within a dabbling duck community. The positive correlations of Zannichellia and
Chara abundances with presence and breeding success,
respectively, probably indicate preferred use and selection of temporary flooded lakes. Indeed, these macrophyte species are largely dependent on temporary
floods and, consequently, regular drying up, as found
in naturally functioning Mediterranean marshes and
lagoons (Grillas, 1992). Reedbeds too are favoured
by an annual summer drying up (Mesléard and Pérennou, 1996). Discussion of which habitat features be-
Frequency of Zanichellia sp.
+**
+**
Surface of reedbed
-*
Waterlevel variation
Presence
Habitat fragmentation
+**
+
Waterdepth
-*
Density
-
Frequency of
Myriophyllum sp.
+*
+**
Success
Frequency of
Chara sp.
Fig. 2. Summary of the main factors affecting the probability of
presence, the density and the breeding success of RCP broods in the
Camargue. + and , respectively, indicate positive and negative effects.
*P < 0.05; **P < 0.01.
363
tween reedbeds or temporary flooding are preferred
by RCP is beyond the scope of this work, but there
are reasons to believe that both are important. Alleged
importance of temporary flooding is further confirmed
by observed avoidance of lakes with higher Myriophyllum frequency. This probably reflects a favourable
water management rather than a real avoidance of
this macrophyte. Myriophyllum is not a known major
food resource for RCP (Snow and Perrins, 1998), but
it is a highly competitive and productive colonizing
species favoured by low variability of water levels,
which decreases macrophytes species richness (Grillas,
1992), and thus leads to depletion in abundance,
diversity and quality of preferred RCP food. The literature clearly mentions the noted preference for Chara
as a main food resource (Szijj, 1965; Boutin, 1994).
Whether Chara macrophytes are consumed by RCP
broods remains to be confirmed, but it is definitely
consumed by adults, and constitutes an important
determinant of breeding success probably as a food
resource for earliest most successful broods.
4.2. Methodological issues
Assessment of detection probability has been recently highlighted as a crucial issue when monitoring
animal populations (Nichols et al., 2000; Thompson,
2002). Still, the double-observer design cannot be considered to have provided efficient results for the mere
objective of estimating brood abundance, as it clearly
underestimated the real brood population size that is
at least of 17 broods. For rare species with relatively
low individual detectability like the RCP, the double-observer approach may be less well adapted to
accurate abundance estimation than capture-recapture
methods as mentioned by Nichols et al. (2000). The
advantage of such a time extended protocol over point
counts methods like double-observer (Nichols et al.,
2000), Time Species Counts (Freeman et al., 2003),
fixed radius or double-sampling (Bart and Earnst,
2002), is that it constitutes a unified single monitoring
scheme over an extensive breeding season. In comparison, point counts only provides snapshots of the
breeding situation on a given site and will not account
for detectability variation in time if applied on too
few occasions. Nevertheless, adjusted point counts
techniques do not require marking animals and are
equally time replicable during the whole breeding
season.
Detection probability analyses further permitted
inference on habitat use by RCP when applied to observed unadjusted brood densities. A classical study of
abundance-habitat relationship for broods of RCP did
not provide any satisfying explanatory patterns. However, once adjusted, density was described by an order-3 model with a reasonable fit. This suggests that
364
P. Defos du Rau et al. / Biological Conservation 125 (2005) 355–367
without any estimation of detectability and its variation in time, this study would have wrongly concluded
that no apparent limiting factors affected densities of
the species in the Camargue. A simple time restricted
survey of the detection probabilities of the species
broods would equally have produced 1 detectability
estimate that, once applied to point counts adjustment, would have unaltered the results of unadjusted
abundance habitat relationship analysis. It is precisely
the investigation of the variation of detectability with
time that permitted the identification of factors limiting the abundance of RCP broods.
4.3. Management recommendations
We suggest the following management recommendations that would theoretically favour breeding and
abundance of the RCP in the Camargue, as well as of
several other mediterranean wildlife and plant species
naturally occuring in the Camargue:
(a) Maintain large reedbed areas around water
bodies, in order to increase the probability of
occupancy.
(b) Maintain relatively high and stable water levels
in spring and early summer, in order to increase
the probability of occupancy, the breeding densities, and the breeding success. This recommendation does not imply necessarily to artificially
maintain high water levels by pumping, but
rather suggests not to artificially drain water
bodies during this period of the year. This recommendation should also favour many other
breeding waterbirds, including the threatened
Purple Heron Ardea purpurea (Kushlan and Hafner, 2000).
(c) Maintain between year variability of flooding conditions (i.e., water levels), including periodic summer drying up, as is typical of naturally
functioning Mediterranean wetlands (Tamisier
and Grillas, 1994). This type of management
should favour the occurrence of macrophytes such
as Phragmites australis, Zannichellia and Chara
species, and should negatively affect the abundance of Myriophyllum species (Grillas, 1992;
Mesléard and Pérennou, 1996).
(d) Maintain wetlands complexes where lakes and
marshes are relatively dense, as brood densities
will be maximised (breeding females may for
instance benefit from the largest possible choice
of brood rearing water bodies [Krapu, 1974]).
In other words, intensity of use and breeding
success of RCP in a wetland patch would
increase with patch size, without apparent major
impact of predators within the range of patch
sizes found in the Camargue. These results need
further specific testing as part of this important
debate (Weller, 1988; Clark and Nudds, 1991).
Nevertheless, among all the tested variables of
predation, competition, food resources, water
management and wetland structure, wetland habitat fragmentation was identified as the main
limiting factor for density of RCP broods. Corresponding management recommendations would
be to avoid increasing distances between lakes or
marshes of a wetland complex by immediately
stopping wetland destruction and fragmentation.
Nearly 30,000 ha of wetland areas have been lost
in the Camargue during the last 30 years (Tamisier and Grillas, 1994), and although many
warnings and recommendations have been produced against wetlands destruction, it is still in
current process in the Camargue (Mathevet and
Tamisier, 2002).
Overall habitat requirements of RCP in terms of
water management seems to be a high and relatively
stable water level through spring before a summer
drying up performed at least for some years, and thus
favouring diverse and Chara-rich macrophyte community. This recommended water management is probably similar to what might have formerly been the
natural functioning of Mediterranean wetlands. Current practices in water management actually favour
permanent flooding. In particular, water levels are
kept artificially high in summer, where they should
naturally be at the lowest, to maximise biomasses of
species poor but highly productive macrophyte communities mainly constituted of Potamogeton pectinatus
and Myriophyllum spicatum (Grillas, 1992; Tamisier
and Grillas, 1994).
However, those management recommendations specially designed to increase abundance of this endangered
breeding duck would have actually been lacking simply
because of the absence of previous knowledge of its
detection patterns. We therefore recommend that, whenever possible, habitat selection surveys involving census
of rare and/or cryptic species should include an analysis
of detectability to raise more reliable inferences for conservation issues.
Acknowledgements
We thank all the Camargue landowners, Les Marais
du Vigueirat and Domaines Listel who allowed us to
use their estates as study sites. Thanks are due to S.
Cayuela and J. Travers for skilled fieldwork, and to T.
Giraud, A. Béchet, T. Boulinier, M. Guillemain and
N. Sadoul, as well as two anonymous reviewers for
comments.
P. Defos du Rau et al. / Biological Conservation 125 (2005) 355–367
365
Appendix
Habitat variables measured for each lake in 2000 and 2001
2000
2001
Response
Brood presence or absence
Brood density
Brood size
Brood age
0 or 1
Broods/ha
# of pulli
In weeks
0 or 1
Broods/ha
# of pulli
In weeks
Competitors
Density of coots
Density of mallards
Density of swans
Density of grebes
Broods/ha
Broods/ha
Broods/ha
Broods/ha
Broods/ha
Broods/ha
Broods/ha
Broods/ha
#
#
#
#
#
# of indiv.min
# of indiv.min
# of indiv.min
# of indiv.min
# of indiv.min
0 or 0.5 or 1
0 or 0.5 or 1
0 or 0.5 or 1
Predators
Abundance of gulls
Abundance of harriers
Abundance of kites
Abundance of magpies
Abundance of crows
Mustelids presence
Dog presence
Fox presence
Water management
Mean salinity
Water depth
Water level SD in April and May
Water level SD from April to June
Water level SD from April to July
Water level variation from 20/04 to 10/05
Water level variation in May
Water level variation from 20/05 to 10/06
Water level variation in June
of
of
of
of
of
indiv.
indiv.
indiv.
indiv.
indiv.
g/l
cm
cm
%
# of species
%
%
%
%
%
%
%
%
%
%
%
# of species
Wetland structure and habitats
Mean distance to 5 closest lakes
Distance to closest lake
Water surface area
Reedbed surface area
Arthrocnemum meadows area
Scirpus beds surface area
Juncus beds surface area
Number of islets
Area of reedbed islets
Area of Arthrocnemum islets
Area of Scirpus islets
Area of Juncus islets
Total wetland area
Shoreline index
m
m
ha
ha
ha
ha
ha
# of islets
ha
ha
ha
ha
ha
Perimeter/2(p · area)1/2
m
m
ha
ha
ha
ha
ha
# of islets
ha
ha
ha
ha
ha
Perimeter/2(p · area)1/2
%
%
%
%
In
In
In
In
In
2001
2001
2001
2001
2001
g/l
cm
cm
cm
cm
%
%
%
%
Food resources
Frequency of Myriophyllum sp.
Frequency of Ceratophyllum sp.
Frequency of Potamogeton pectinatus
Frequency of Potamogeton pusillus
Frequency of Potamogeton fluitans
Frequency of Ranunculus sp.
Frequency of Chara sp.
Frequency of Najas sp.
Frequency of Zannichellia sp.
Frequency of Scirpus sp.
Frequency of Ludwigia sp.
Macrophyte sp. richness
%
log-transformed
in 2000 and 2001
in 2000 and 2001
366
P. Defos du Rau et al. / Biological Conservation 125 (2005) 355–367
References
Åberg, J., Gunnar, J., Swenson, J.E., Mikusinski, G., 2000. Difficulties
in detecting habitat selection by animals in generally suitable areas.
Wildlife Biology 6, 89–99.
Alldredge, J.R., Thomas, D.L., McDonald, L.L., 1998. Survey and
comparisons of Methods for study of resource selection. Journal of
Agricultural, Biological and Environmental Statistics 3, 237–253.
Amat, J.A., 1982. The nesting biology of ducks in the Marismas of the
Guadalquivir, South-Western Spain. Wildfowl 33, 94–104.
Amat, J.A., 1984. Ecological segregation between Red-crested Pochard
Netta rufina and Pochard Aythya ferina in a fluctuating environment. Ardea 72, 229–233.
Amat, J.A., 1991. Effects of Red-crested Pochard nest parasitism on
mallards. Wilson Bulletin 103, 501–503.
Anderson, D.R., Link, W.A., Johnson, D.H., Burnham, K.P., 2001.
Suggestions for presenting the results of data analyses. Journal of
Wildlife Management 65, 373–378.
Anderson, M.G., Titman, R.D., 1992. Spacing patterns. In: Batt,
B.D.J., Afton, A.D., Anderson, M.G., Ankney, C.D., Johnson,
D.H., Kadlec, J.A., Krapu, G.L. (Eds.), Ecology and Management
of Breeding Waterfowl. University of Minnesota Press, Minneapolis, London, pp. 251–289.
Bart, J., Earnst, S., 2002. Double sampling to estimate density and
population trends in birds. Auk 119, 36–45.
Blums, P., Clark, R.G., Mednis, A., 2002. Patterns of reproductive
effort and success in birds: path analyses of long-term data from
European ducks. Journal of Animal Ecology 71, 280–295.
Boutin, J., 1994. Nette rousse. In: Yeatman-Berthelot, D., Jarry, G.
(Eds.), Nouvel Atlas Des Oiseaux Nicheurs De France 1985–1989.
Société Ornithologique de France, Paris, pp. 142–143.
Broyer, J., Daléry, G., 2000. Lhabitat de la Nette rousse Netta rufina
sur les étangs piscicoles de lest de la France en période de
reproduction. Alauda 68, 185–191.
Burnham, K.P., Anderson, D.R., 2002. Model Selection and Inference:
a Practical Information-theoretic Approach, second ed. Springer,
New York.
Clark, R.G., Nudds, T.D., 1991. Habitat patch size and duck nesting
success: the crucial experiments have not been performed. Wildlife
Society Bulletin 19, 534–543.
Defos du Rau, P., 2002. Elements for a red-crested pochard (Netta
rufina) management plan. Game and Wildlife Science 19, 84–141.
Defos du Rau, P., Barbraud, C., Mondain-Monval, J.Y., 2003.
Estimating breeding population size of the red-crested pochard
(Netta rufina) in the Camargue (southern France) taking into
account detection probability: implications for conservation. Animal Conservation 6, 379–385.
Dehorter, O., Rocamora, G., 1999. Nette rousse Netta rufina. In:
Rocamora, G., Yeatman-Berthelot, D. (Eds.), Oiseaux Menacés et
à Surveiller en France, Listes Rouges et Recherche de Priorités,
Populations, Tendances, Menaces, Conservation. Société dEtudes
Ornithologiques de France, Paris, pp. 58–59.
Didger, 2000. Digitizing Software, Version 2.01. Golden Software,
Golden.
Drapeau, P., Leduc, A., McNeil, R., 1999. Refining the use of point
counts at the scale of individual points in studies of bird–habitat
relationships. Journal of Avian Biology 30, 367–382.
DuBowy, P.J., 1991. Seasonality and interspecific competition in
waterfowl guilds: patterns and processes – a reply to Bethke.
Ecology 72, 1159–1161.
Freeman, S.N., Pomeroy, D.E., Tushabe, H., 2003. On the use of
timed species counts to estimate avian abundance indices in speciesrich communities. African Journal of Ecology 41, 337–348.
Green, A.J., El Hamzaoui, M., 2000. Diurnal behaviour and habitat
use of nonbreeding Marbled Teal Marmaronetta angustirostris.
Canadian Journal of Zoology 78, 2112–2118.
Grillas, P., 1992. Les communautés de macrophytes submergées des
marais temporaires oligo-halins de Camargue, étude expérimentale
des causes de la distribution des espèces. PhD thesis, Université
Rennes I, Rennes.
Gu, W., Swihart, R.K., 2003. Absent or undetected? Effects of nondetection of species occurrence on wildlife–habitat models. Biological Conservation 116, 195–203.
Hines, J.E., 2000. DOBSERV: software to compute detection probability and abundance from double-observers point counts. US
Geological Survey-Patuxent Wildlife Research Center. Available
from: <http://www.mbr-pwrc.usgs.gov/software/dobserv.html>.
Johnson, D.H., Sargeant, A.B., Greenwood, R.J., 1989. Importance of
individual species of predators on nesting success of ducks in the
Canadian Prairie Potholes Region. Canadian Journal of Zoology
67, 291–297.
Jones, J., 2001. Habitat selection studies in avian ecology: a critical
review. Auk 118, 557–562.
Joyner, D.E., 1980. Influence of invertebrates on pond selection by
ducks in Ontario. Journal of Wildlife Management 44, 700–705.
Kaminski, R.M., Weller, M.W., 1992. Breeding Habitats of Nearctic
Waterfowl. In: Batt, B.D.J., Afton, A.D., Anderson, M.G.,
Ankney, C.D., Johnson, D.H., Kadlec, J.A., Krapu, G.L. (Eds.),
Ecology and Management of Breeding Waterfowl. University of
Minnesota Press, Minneapolis, London, pp. 251–289.
Kendall, W.L., Nichols, J.D., Hines, J.E., 1997. Estimating temporary
emigration and breeding proportions using capture-recapture data
with Pollocks robust-design. Ecology 78, 563–578.
Kent, M., Coker, P., 1992. Vegetation Description and Analysis, a
Practical Approach. Wiley, New York.
Kéry, M., 2002. Inferring the absence of a species – a case study of
snakes. Journal of Wildlife Management 66, 330–338.
Krapu, G.L., 1974. Foods of breeding pintails in North Dakota.
Journal of Wildlife management 38, 408–417.
Kushlan, J.A., Hafner, H. (Eds.), 2000. Heron Conservation. Academic Press, San Diego.
Llorente, G.A., Ruiz, X., 1985. Datos sobre la reproducción del Pato
Colorado Netta rufina en el Delta del Ebro. Miscellania Zoologica
9, 315–323.
Mac Nally, R., 2000. Regression and model-building in conservation
biology, biogeography and ecology: the distinction between – and
reconciliation of – predictive and explanatory models. Biodiversity and Conservation 9, 655–671.
Mathevet, R., Tamisier, A., 2002. Creation of a nature reserve, its effects
on hunting management and waterfowl distribution in the Camargue (Southern France). Biodiversity and Conservation 11, 509–519.
Mesléard, F., Pérennou, C., 1996. La végétation aquatique émergente :
écologie et gestion. MedWet n 6. Tour du Valat, Arles.
Nichols, J.D., Hines, J.E., Sauer, J.R., Fallon, F.W., Fallon, J.E.,
Heglund, P.J., 2000. A double-observer approach for estimating
detection probability and abundance from point counts. Auk 117,
393–408.
Office National de la Chasse, 1982. Critères de détermination de lâge
du Canard colvert de la naissance à 9 semaines. Fiche technique no.
10, Bulletin mensuel 63.
Opermanis, O., 2001. Marsh Harrier Circus aeroginosus predation on
artificial duck nests: a field experiment. Ornis Fennica 78, 198–203.
Rexstad, E., Burnham, K.P., 1991. Users Guide for Interactive
Program CAPTURE, Mark-recapture Population and Density
Estimation Program. Cooperative Fish & Wildlife Research Unit,
Colorado State University, Fort Collins.
Snow, D.W., Perrins, C.M., 1998. The Birds of the Western Palearctic,
vol. 1. Oxford University Press, Oxford.
Statistica, 2000. Kernel version 5.5 A. StatSoft France, MaisonsAlfort.
Stephens, S.E., Koons, D.N., Rotella, J.J., Willey, D.W., 2003. Effects
of habitat fragmentation on avian nesting success: a review of the
P. Defos du Rau et al. / Biological Conservation 125 (2005) 355–367
evidence at multiple spatial scales. Biological Conservation 115,
101–110.
Szijj, J., 1965. Zur ÖKologie der Tauchenten. Vogelwelt 86, 97–104.
Tamisier, A., Grillas, P., 1994. A review of habitat changes in the
Camargue : an assessment of the effects of the loss of biological
diversity on the wintering waterfowl community. Biological Conservation 70, 39–47.
Thompson, W.L., 2002. Towards reliable bird surveys: accounting for
individuals present but not detected. Auk 119, 18–25.
367
Weller, M.W., 1988. Issues and approaches in assessing cumulative
impacts on waterbird habitat in wetlands. Environmental Management 12, 695–701.
White, G.C., Burnham, K.P., 1999. Program MARK: survival
estimation from populations of marked animals. Bird Study 46
(Suppl.), 120–139.
Wilson, G.J., Delahay, R.J., 2001. A review of methods to estimate the
abundance of terrestrial carnivores using field signs and observation. Wildlife Research 28, 151–164.
Chapter 4
SOME ASPECTS OF RED-CRESTED POCHARD MACROECOLOGY AND HABITAT USE IN EUROPE:
IMPACT OF DETECTION ISSUES
P. Defos du Rau, J.Y. Mondain-Monval, C. Barbraud, F. V. Esquerre, J. Hanganu, J.B. Kiss,
A. Torres, S. Lek, E. Cam
INTRODUCTION
Analyses of habitat requirements over an entire population range (as defined by Krebs 1972)
allows comparisons of high- and low-abundance patches and more generally of
subpopulations exhibiting different dynamics, potentially of different conservation concern
(O’Connel 1999). Apart from the notable exception of large-scale GIS and Atlas datasets
(Gimona & Brewer 2006 for a recent example), habitat surveys over the entire range of a
population distribution are nevertheless rarely performed (Brown 1995), especially on rare or
threatened species (Guisan et al. 2006) because of the amount of fieldwork required.
However, such large scale surveys are potentially useful for macroecology inferences (Brown
1995, Watkinson et al. 2003). In particular, conservation of rare or threatened species can
benefit from large scale surveys, including analyses of spatial variation in habitat use, which
may allow identification of low-quality patches and design of specific management plans for
such patches (Reynolds 2003). Inferences in macroecology as well as in standard habitat use
analyses often involve comparisons of occurrence and abundance among populations and
among sites, respectively (Brown 1995, Block & Brennan, 1993). It is therefore crucial to
base these inferences on unbiased occurrence and abundance estimates among sites and
populations. This is especially true for rare species at large geographical and ecological scales
because there may be variation in detection probability with space, observer, habitat and
mainly, local abundance (Nichols et al. 2000, MacKenzie et al. 2002, Thompson 2002, Gu &
Swihart 2003, Royle & Nichols 2003). Modeling occurrence and abundance for
macroecological or conservation purposes requires statistical inference tools accounting for
detectability variation, as defined by Pollock et al. (2002) and MacKenzie et al. (2006) for
large scale wildlife surveys. Here we estimated occupancy and abundance of broods in a rare
European breeding waterbird, the Red-crested Pochard (RCP) Netta rufina. Defos du Rau et
al. (2005) estimated RCP occupancy and abundance in one major European wetland: the
Camargue (France) and modelled them according to habitat covariates. However, to gain
predictive reliability and generalization power in species-habitat models and management
recommendation over a wider range of situations and locations, an additional larger scale
1
habitat survey with a larger sample was recommended. We modelled occupancy and
occurrence over several locations in Europe. We addressed ecological hypotheses put forward
in the framework of theories of the Hutchinsonian Niche (Block & Brennan, 1993, Brown
1995, Guisan & Thuiller 2005), and on resource selection functions (Manly et al. 2002,
Alldredge & Griswold 2006). Our approach was based on capture-recapture models for
detectability estimation (i.e., models incorporating detection probability; Kendall et al. 1997,
MacKenzie et al. 2002). Resulting occupancy and abundance models were then used to
(i)
identify most important habitat variables for RCP breeding site conservation and
management, and
(ii)
address some simple macroecological hypotheses on Mediterranean population
dynamics of RCP in order to better understand current demographic and
conservation issues for this still rare, possibly threatened, game waterbird.
The Red-crested Pochard is a migratory duck known to breed mainly in Central Asia, from
China to the Black Sea, and more locally in Central and Western Europe (Snow & Perrins
1998). The species is listed in appendix III of the Berne Convention, in appendix II of the
Bonn Convention, in appendix II/2 of the EU Birds Directive 94/24 and in Annex II of the
African-Eurasian Waterbird Agreement. Although it is considered of strong conservation
concern, especially in Eastern Europe which was once supposed to hold some of the largest
European populations (Tucker and Heath 1994), the species is hunted in France, Portugal,
Romania and Spain. In three of these countries, we monitored RCP brood rearing activity and
habitat use in major coastal wetlands, namely Danube delta in Romania, Rhône delta in
France and Ebro delta in Spain.
A priori hypotheses
Based on previous knowledge on RCP breeding ecology and habitat (Tucker and Heath 1994,
Snow & Perrins 1998, Defos du Rau et al. 2005), we made the following a priori hypotheses
about factors affecting habitat use by broods at the local, landscape and regional scales.
1) We expected a higher occupancy probability and/or abundance in Danube delta, which
is considered a traditional, relatively preserved, stronghold of the species (Tucker and Heath
1994), compared to the two other west European deltas. The latter are largely degraded,
fragmented and reclaimed. Furthermore, Danube delta is located closer to the center of the
2
species range than the other two western deltas; we might thus expect this delta to be more
densely occupied (Brown 1995, Watkinson et al. 2003, Guo et al. 2005).
We also expected an increase in occupancy probability and/or abundance:
2) as habitat fragmentation decreases. Low level of fragmentation corresponds to a
minimal number of reclaimed or degraded lakes within wetland complexes like deltas
(Weller, 1988),
3) as abundance of both preferred nest site helophytes and food macrophytes increases.
According to Snow and Perrins (1998), the breeding habitat of RCP is often characterized by
both reedbeds and extensive macrophytes beds,
4) as temporal variation in water level decreases and water depth increases. This should be
associated with decreased risk of drying up, which is high in Mediterranean wetlands
(Kaminski and Weller, 1992),
5) as predator and inter-specific competitor densities decrease.
or, conversely
6) as predator density increases following increased occupancy or abundance of
corresponding prey, i.e. RCP. Although few studies have succeeded in demonstrating an
impact of intra- or inter-specific competition on brood spacing and space use (DuBowy, 1991;
Anderson and Titman, 1992), both competition and predation are strongly suspected to affect
habitat use by broods (Kaminski and Weller, 1992) and species-habitat relationships across
various scales in general (Jones 2001, Guisan & Thuiller 2005, Guisan et al. 2006b).
7) We expected a non-linear increase in abundance as area of preferred habitat increases,
under the assumption of habitat overfilling due to habitat loss (Reynolds 2003).
8) We expected an increase in occupancy probability of low quality patches (sink) as
distance to nearest high-quality patch (source) decreases. This prediction was based on the
hypothesis of dispersal within deltas, and the existence of source and sink subpopulations
(Brown 1995).
Last, we addressed metapopulation dynamics, due to possible source-sink dynamics,
among all or some of the three studied subpopulations, as defined by Gay et al. (2004) for
European RCP population (Watkinson et al. 2003).
3
METHODS
Study Species
The RCP is a rare breeding duck species in Europe, with an estimated 13000 - 25000 breeding
pairs only outside its main breeding strongholds of Central Asia (Snow & Perrins 1998).
Along with Russia, Southeast France, East Romania and Spain are part of its main European
strongholds (Tucker & Heath 1994). The species breeding habitat is poorly documented,
particularly outside France and Spain, but is generally known to include mainly extensive
reedbeds and submerged macrophytes (Snow & Perrins 1998). Increasing brood predation by
yellow-legged gull (Larus michahellis) has been suspected to cause major habitat switch in
the Camargue delta from brackish to fresh marshes and lakes covered with reedbeds (Boutin,
1994). RCP is rather cryptic in its breeding behaviour and habits, and broods mostly come out
of the vegetation fringe on open water in late afternoon and in the early evening.
Study areas (Figure 1)
Figure 1: study areas are deltaic wetlands created by major European rivers
4
All three study areas are extensive wetlands of Southern Europe created by sedimentation at
the mouth of 3 major European rivers. Each of these three Mediterranean deltas is listed in the
Ramsar Convention and under various national conservation designations (Frazier 2002).
They have major conservation value and, in particular, have long been of international
importance as breeding areas for RCP (Bigas & Vidal 2004, Defos du Rau et al. 2003,
Gillissen et al. 2002, Tucker & Heath 1994)
The Danube river mouth is a vast, largely natural delta of 5445 km2 split between Romania
(5007 km²) and Ukraine (438 km²) on the west coast of the Black Sea. Natural habitats
account for approximately 91% of the total delta area due to the largely untouched
hydrological functioning of this characteristic delta ecosystem. However, a large northern part
of the Danube Delta has been reclaimed for agricultural activities in the 1970s and several
canals are currently affecting the natural hydrology of the delta. The number of RCP was
estimated to be comprised between 400-2000 pairs for the whole delta.
The Rhône river mouth, known as the Camargue, is a largely cultivated delta of 1450 km2 on
the west coast of the Mediterranean. Natural habitats (freshwater and brackish marshes,
including reedbeds and temporary flooded salt meadows) account for approximately 40% of
the total delta area. There are 230 estates in the whole delta, which are either protected or
hunting areas. The number of RCP was estimated at approximately 600 pairs for the whole
Camargue. The species is known or suspected to breed regularly in a few estates within the
delta where breeding habitat is available and when flooding conditions are adequate.
The Ebro river mouth forms a relatively small and densely cultivated delta of 330 km2 , also
on the west coast of the Mediterranean. Natural areas remain on 20% of the Delta area,
mainly along the coast, and are composed of sandy beaches, coastal lagoons, brackish, salt
and freshwater marshes surrounded by reedbeds. Most of the land is privately owned. The
number of RCP was estimated at approximately 2000 pairs for the whole delta.
Delta
Area (km²) % natural habitat Sampled sites n
Estimated RCP pairs
Danube
5445
91
36
400-2000
Camargue
1450
40
40
600 (Defos du Rau et al. 2003)
Ebro
330
20
9
1500-2200 (Bigas & Vidal 2004)
Table 1 : study areas characteristics
5
Sampling
To evaluate the performance of our occupancy models at predicting presence of RCP broods
as a function of habitat variables, meeting recommendations from Guisan & Thuiller (2005),
we used a test data set in addition to the training data set. Observed occupancy in the test set
was then compared with occupancy predicted by the model estimated using the training set
(and run on the test set).
- Training data set. Within each of the three deltas, study areas were selected on the
basis of access facilities and presence of the main characteristics of RCP preferred habitat,
including reedbeds and adequate waterdepth. Based on previous knowledge (Tucker and
Heath 1994, Snow & Perrins 1998) of RCP breeding habitat preference for large reedbeds
(confirmed in the Camargue by Defos du Rau et al. 2005), we selected only study areas with
reedbeds and minimal flooding conditions in order to ensure collection of sufficient data on
presence for analysis. Our goal was also to increase our understanding of RCP niche
requirements in reedbed wetland habitat. Within these study areas, we sampled systematically
all wetlands with reedbeds and proper flooding conditions. In Ebro delta, we surveyed 9 of the
11 major wetlands with reedbeds. The remaining two were either dry or forbidden to visit. In
the Camargue, we surveyed 40 wetlands within 7 study areas. In Danube delta, we surveyed
36 wetlands within 5 study areas (Table 1).
We compared habitat variables between lakes occupied versus not occupied by RCP broods.
Unused lakes were situated in immediate vicinity of used lakes and in the same flooding
conditions and were thus available as well as accessible to the species (Jones 2001).
- Test data set. For habitat model validation purpose, a further 37 wetlands were
selected in the Camargue through proportional-stratified random sampling based on a 4x3
cells grid over the whole delta. In each cell, wetlands were randomly drawn proportionally to
estate abundance.
Bird Survey
Following methodology described in Defos du Rau et al. (2005), a minimum of 3 (in Danube
Delta) and a maximum of 9 (in Ebro Delta) consecutive point counts were conducted in the
6
local peak months of the 2001 breeding season (May to July), on each of the 85 lakes.
Because of RCP rarity and apparent occurrence unpredictability, we chose to maximise pointcounts replicates in space rather than in time. These point counts were conducted in the late
afternoon and early evening during 30 min., by two observers in the Camargue and by one
observer in each of the other two deltas. Broods of RCP were intensively searched for using
telescopes on all the visible water area and within surrounding vegetation fringes. Brood
presence, abundance, size and age were recorded. Partially hidden broods were followed
within the vegetation until their size and age could be confidently estimated. The age of
broods was estimated in weeks, based on size of chicks (Office National de la Chasse 1982).
On each wetland of the test data set, three monthly brood counts were conducted by the same
observer in May, June and July 2001 during daytime, including exhaustive visit of the 37
wetlands shoreline in order to maximise detection of broods.
Habitat Survey
We examined the effects of various habitat factors on brood presence and abundance at two
geographic scales which are appropriate for most practical conservation and management
recommendations: the wetland complex scale and the individual wetland scale (Weller 1988).
For models of brood presence and abundance, we considered 2 delta effects (i.e., Delta
identity: Danube and Ebro), and 30 covariates (Appendix) were measured for each wetland,
following recommendations from Weller (1988), Kaminski and Weller (1992), Jones (2001),
Ruth et al. (2003). These included frequencies of potential predators, diversity and densities
of potential competitors, food resources, water management, and wetland landscape structure
(Appendix):
Wetland Structure and Habitat Type Composition
RCP is thought to nest preferentially in Phragmites australis beds (Tucker & Heath 1994,
Snow and Perrins, 1998) or salt scrubs of Arthrocnemum glaucum (Amat, 1982). Surface
areas of vegetation patches and islets of both these abundant habitat type as well as of Scirpus
sp, a likely brood-rearing habitat (Mack & Flake 1980), were measured for each wetland. In
addition, wetland mosaic structure within landscape was described for each lake by the
distance to the closest lake and the mean distance to the 5 closest lakes. The index
perimeter/2 π × area (shore) was used to evaluate shoreline shape from circular/regular to
7
indented (Joyner, 1980). All these geographic variables were computed using standard
Geographical Information System (GIS).
Water Management
Water level was measured monthly for each wetland during the egg laying period from April
to July and standard deviation of water levels during that period was calculated. The
maximum observed water level was used as an index of the wetland depth.
Food Resources
The RCP is primarily herbivorous, known to graze mainly on Chara spp., Potamogeton spp.
and Ruppia spp. beds (Snow and Perrins, 1998). Genus richness of submerged macrophytes
was assessed in each wetland and relative frequencies of two most ubiquist taxa,
Myriophyllum sp and Potamogeton pectinatus, were estimated along a Braun-Blanquet scale
(Kohler 1978). In particular, according to Green (1998), beds of Potamogeton pectinatus are
often preferred foraging habitat for Mediterranean duck communities.
Predator and Competitor Communities
Marsh harrier (Circus aeroginosus) (Opermanis, 2001), yellow-legged gull (Boutin, 1994)
and corvids (Johnson et al., 1989) are predators of duck clutches and broods. We assessed
mean number of individuals seen during the 30min point counts as an index of potential
predation frequency. In addition, competition was considered a potential driver of habitat use
by RCP broods. Brood densities of most abundant competitor, i.e. mallard (Anas
platyrhynchos), gadwall (Anas strepera), mute swan (Cygnus olor), coot (Fulica atra) and
great-crested grebe (Podiceps cristatus) were estimated for each wetland as an index of
potential competition intensity within the same trophic and/or spatial niche as RCP.
Detectability issues were not considered at first for these widespread, abundant and/or
conspicuous species. However, the assumption of perfect detectability of every species of
predator or competitor may not be met. In addition, presence of predators for example may
influence detectability of RCP broods: they may remain more cryptic and investigators may
miss a larger proportion of broods when predators are present compared to situations where
predators are absent. Last, the same habitat variables may influence independently RCP brood
occurrence and predator or competitor occurrence. Not taking RCP brood and predator or
competitor detectability into account may result in an apparent interaction between RCP
brood presence and presence of predators or competitors. Consequently, we re-tested the
8
hypothesis of inter-specific interaction between RCP brood and predator or competitor species
using co-occurrence models for species with imperfect detection (MacKenzie et al. 2004).
These co-occurrence models are implemented in program PRESENCE (available for
download at http://www.mbr-pwrc.usgs.gov/software.html). Species richness of all likely
competitors for food and/or space in the wetland patch was also estimated as a further index
of competition factor potentially influencing RCP brood occurrence and abundance.
Because of its potential major impact on species-habitat models (Segurado et al. 2006, Guisan
et al. 2006b), we addressed spatial autocorrelation in both RCP broods occurrence and
abundance by building omnidirectional variograms using the Software Variowin version 2.21
(Pannatier 1996, Barbraud & Delord 2006).
Occupancy modeling
Because of RCP rarity and cryptic behaviour, it was highly unlikely that all occupied
wetlands be identified during sampling sessions, even after several time-replicated pointcounts. Indeed, imperfect detectability of RCP broods (Pollock et al. 2002) was likely to lead
to false absences. Consequently, we used a mark-recapture approach developed by
MacKenzie et al. (2002) and implemented in program PRESENCE to model occupancy
probability for each wetland and estimate the corrected number of occupied wetlands, on the
basis of RCP brood detection histories and habitat covariates on the surveyed wetland. This
approach relies on the replication of survey occasions through time which can be recorded as
a vector of 1’s (detection) and/or 0’s (non-detection), also allowing for missing data. The 85
wetlands were thus surveyed on 3 to 9 occasions during the breeding season, when RCP
brood population on each wetland was considered demographically closed. Broods are
assumed not to be falsely detected when absent, and to be detected independently of brood
detection on other wetlands.
Detection probability. As highlighted by Gu & Swihart (2003), it was necessary to
model detection-habitat relationships before building a predictive habitat model for brood
occupancy. With the approach developed by MacKenzie et al. (2002), this can be done by
addressing the influence of covariates on detection probability before starting to address
covariates influencing occupancy probability. As recommended by Gu & Swihart (2003),
habitat covariates potentially affecting RCP brood detection probability were first addressed
9
using logistic regression, as implemented in program PRESENCE. For computation purposes
with program PRESENCE, all independent variables were ‘studentized’ before analysis by
subtracting the corresponding mean and then dividing by the standard deviation. Predation
and competition were considered as variables potentially affecting detection of RCP brood
(Appendix), because increased predation and competition may lead to increased brood
tendency to stay under vegetation cover. We also addressed the hypothesis that relative and
absolute extent of open water and vegetated habitat influenced detection of broods. Lastly,
brood detection variation among deltas was also addressed. Prior to incorporating covariates
into models we used pre-defined detection models implemented in program PRESENCE to
assess whether detectability varied in time, i.e. between survey occasions and in space, i.e.
between two or three sites groups.
Occupancy probability. Recommendation to assess the influence of as many habitat
variables as possible on occupancy probability (Kaminski & Weller 1992) was difficult to
follow for such a rare species with obviously unfavourable ratio of sample size to number of
variables. Consequently, we used an approach based on several steps.
First, we conducted a preliminary analysis to select variables that most contributed to
differentiate used and apparently unused wetlands (i.e., data from presence/absence
uncorrected for detection probability). To do so, we used a forward stepwise discriminant
procedure. To account for the specific influence of habitat effects (and not for geographical
variation; i.e., Delta) on apparent occupancy in the discriminant analysis, we centred every
habitat variables on the three delta means according to the corresponding individual wetlands.
Tolerance threshold (computed as 1 – R² of the respective variable with all other variables in
the model) was set at 0.5 to prevent major redundancy between variables; inclusion and
exclusion F-statistics were set at 1 and 0 respectively. Moreover, habitat variables were
checked for colinearity between one another and for coherence of their effect sign with the
corresponding a priori hypothesis.
In a second step, the retained variables were included in logit models of occupancy
probability incorporating detection probability implemented in PRESENCE. The relative
support of the data for various logistic regression models of detection and occupancy (Hirzel
& Guisan 2002, Alldredge & Griswold 2006) was assessed using Akaike Information Criteria
(Anderson et al. 2001). Goodness-of-fit and overdispersion were checked for by Pearson χ²
10
and variance inflation factor ĉ (Anderson et al. 2001, MacKenzie & Bailey 2004). For
comparison purpose with the latter models accounting for imperfect detectability, the retained
variables were also used in an exhaustive search procedure of every possible combinations of
covariates in standard logistic regression models (i.e. assuming perfect detectability).
Both procedures allowed identification of a reduced set of variables that best explained and
predicted RCP broods occupancy among wetlands. We chose not to incorporate interactions
in occupancy models because of the large number of variables considered. Program
PRESENCE computes estimates of proportion of occupied sites and covariate-dependent
occupancy probabilities for individual sites.
Once a first comprehensive occupancy model was identified using PRESENCE (on the basis
of the habitat variables primarily selected using discriminant analysis), predicted proportion
of occupied sites allowed estimation of the number of false absences. We used wetlandspecific occupancy probabilities to identify wetlands where presence of broods had a higher
probability to have remained undetected by investigators: these were wetlands with apparent
absence of brood but with the highest occupancy probabilities.
In a third step these false absence wetlands were converted into presence in a second deltacentred discriminant analysis incorporating only unused habitat variables. This aimed at
identifying habitat variables affecting brood occupancy left undetected in the first
discriminant analysis.
Last, we evaluated predictive performance of best occupancy models on the stratified random
sample test set using receiver operating characteristic (ROC) curves (Liu et al. 2005, Fawcett
2006).
Abundance modeling
- Detection probability. To estimate individual brood detectability, a robust-design
was used as a double-sampling scheme (Pollock et al. 2002) in a subset of 10 Camargue
wetlands with assumed highly favourable habitat and high RCP occupancies. Because of the
rarity of RCP as a breeding bird in the Camargue, the occurrence of 2 broods of identical age
and size in the same wetland complex was highly unlikely. Each observed brood was
11
therefore identified (or « marked ») by the combination of age and size. The same capturerecapture design was used in Defos du Rau et al. (2005).The Robust-design model (Kendall et
al., 1997) provided estimates of local survival rates (S), temporary emigration (g”),
immigration (g’) probabilities and population sizes but was only used to estimate capture (p)
and recapture (c) probabilities of individually identified broods. Both latter probabilities were
considered as detection probabilities, since observations of known broods can be viewed as
(re)capture events. In this view, the detection probability estimated using the robust-design
model corresponded to the probability of detecting an individually marked brood given its
presence in the study area.
This detectability parameter was used to correct brood abundance observed on occupied
wetlands in the three deltas. Detectability was hypothesized to vary in time and space
according to habitat covariates. Consequently, we conducted a Huggins robust-design analysis
allowing incorporation of covariates. We used program MARK (White and Burnham, 1999).
We considered most of the habitat variables assumed to influence RCP brood detection
(Appendix) as covariates, but we pooled densities of all possible competitors (« compdens »).
In addition, we did not use frequency of harrier because this species occurred very rarely on
the double-sampling wetlands subset. We first conducted a model selection procedure by
addressing temporal variation in the different parameters among and within primary sampling
occasions and then by addressing the influence of covariates on capture (p) and recapture (c)
probabilities.
- Abundance. Observed and detectability-adjusted brood abundances were modelled
using Poisson regression with a log-link function. We considered the same covariates as those
selected for occupancy probability. For detectability-adjusted abundances we also included
the covariates selected by both successive discriminant analyses. Observed brood abundances
were adjusted by capture probabilities estimated from the lowest AICc robust-design model as
follows:
Da j =
Nˆ j
n
, with Nˆ j = ∑i ij ,
WS j
pˆ i
where Daj is adjusted density for lake j, nij is observed brood number during period i on lake j,
WSj is water surface area of lake j, N̂ j is total adjusted brood number for lake j, p̂i is detection
probability estimated through robust-design at period i.
12
We used the new proportion of occupied wetlands estimated by the final occupancy model to
estimate the number of false absences among the apparently unoccupied wetlands with the
highest occupancy probabilities. These wetlands were conservatively assigned a minimum
“observed” abundance of one brood. Since we corrected both occupancy probability and
abundance for detection probability, we assumed that remaining absences in the data set were
true ones and we therefore used those in the abundance models. We thus log-transformed the
observed and detectability-adjusted brood counts augmented with one.
Macroecological hypotheses
- Occupancy and abundance should decrease from center (Danube delta) to edge (Ebro
delta) of the species global range (Brown 1995, Watkinson et al. 2003, Guo et al. 2005) and
from preserved to degraded breeding areas. To address these hypotheses, we examined spatial
effects of delta identity in the best occupancy and abundance models and graphically
compared mean predicted occupancy, adjusted abundance and observed abundance (with their
95% confidence intervals) for each delta.
- Habitat loss should lead to habitat overfilling. We addressed the hypothesis of a possible
habitat loss effect among (i) the 3 deltas and (ii) the two most densely occupied and most
degraded deltas by testing linearity of breeding habitat area - brood abundance relationship
(Reynolds 2003, Wiegand et al. 2005). Our reasoning was based on the idea that
proportionality between habitat area and brood abundance reflects an increase in brood
abundance with area but no increase in brood density (i.e., no overfilling/crowding).
Conversely, positive departure from proportionality would indicate an increase of brood
density with habitat area and thus overfilling of larger, most attractive, habitat. All possible
models of brood abundance including quadratic functions of habitat area were compared to
simple linear ones to assess departure from proportionality. In case of non proportionality (i.e.
non-constant/area-dependent density), a positive quadratic parameter of the most supported
order-two polynomial model may provide evidence of crowding. Such habitat overfilling is
potentially indicative of currently occurring habitat loss (Reynolds 2003). We incorporated
interactions of Ebro and Danube deltas with habitat area to account for possible geographical
effects on the relationship between habitat area and brood abundance. Using only non-null
brood abundance (both unadjusted and detectability-adjusted) for this analysis, we compared
all models with ΔAICc below or immediately above 2.
13
We quantified breeding habitat availability using reedbed area because it is an important
nesting habitat of the species (Snow & Perrins 1998, Defos du Rau et al. 2005), it is positively
related to the total wetland area in our sample (r=0,82,p<0,001), and it has high conservation
value (Poulin et al. 2002). For these reasons, we assumed that reedbed area was a highly
relevant habitat variable to address the influence of habitat loss on brood abundance.
- Assuming source-sink dynamics (Pulliam & Danielson 1991, Brown 1995) and
dispersal among sites due to habitat overfilling, occupancy probability of low quality patches
(sink) should increase as distance to nearest high-quality patch (source) decreases. Once the
niche variables that govern local occurrence and abundance have been identified, a possible
effect of dispersal from favorable patches (sources) to less favorable ones (sinks) can be
addressed. Local occupancy rates predicted by traditional logistic regression on observed
presence-absence data or using MacKenzie et al. model (2002) provide different expected
adequacy to observed data that can be evaluated by ROC analysis (Fawcett 2006). Prediction
performance, as well as thresholds of predicted occurrence can be assessed using ROC curves
(Liu et al. 2005). This threshold discriminates between more and less favorable patches. If
dispersal from source to sink habitat occurs, then, in less favorable patches, the occupied ones
should be closer to source (i.e. occupied favorable patches) on average. We therefore
compared models of observed RCP brood occurrence in less favorable patches predicted by
habitat-modeled occupancy and distance to closest source (i.e. occupied favorable patch).
When using detectability-adjusted occupancy, we measured distance to closest source
considering previously identified false absences (i.e. those with highest occupancy
probabilities but where investigators had not detected RCP broods) as sources.
- Assuming source-sink dispersal processes, there may be metapopulation dynamics
among all or some of the three studied subpopulations. A proper assessment of the existence
of
metapopulation dynamics in RCP would require at least time series of RCP brood
occurence or abundance in subpopulations showing extinction and colonization events. But
macroecology theory may provide preliminary insight into the existence of such dynamics, by
addressing the linear relationship between logit-transformed occupancy and log-transformed
local abundance (Watkinson et al. 2003). Positive relationships between abundance and
occupancy may reflect metapopulation dynamics or spatial variation in availability of habitat
14
and resources (Hanski & Gyllenberg 1997, Freckelton et al. 2005). A general model of the
density – occupancy relationship is the following (Hanski & Gyllenberg 1997):
logit(p)=a+b.log(n), where n stands for density and p for occupancy probability.
If there is a metapopulation dynamics in the study system, a density increase occurring at low
abundance will increase patch occupancy more than proportionaly due to colonization
processes (i.e., b will be larger than 1). Conversely, an estimate of b equal to, or smaller than
1, will provide evidence of simple habitat-filling system (Watkinson et al. 2003). We then
estimated slope parameter of the linear regression of logit (occupancy) as a function of
log(density+1), where occupancy and density would be alternatively based upon detectability
adjusted and unadjusted estimates.
RESULTS
A total of 408 broods were observed on 29 wetlands out of the 85 surveyed ones (Table 2). In
the Camargue stratified random sample, 13 broods were observed on 8 wetlands out of the 37
surveyed ones.
Delta
sampled
detected occurences detected broods
Observed densities/ha
sites
Danube
36
6
11
0,005
Camargue
40
17
42
0,175
Ebro
9
6
355
0,236
Table 2: RCP brood survey results in 2001 in the 3 study deltas
Because of low prevalence in Danube delta and low number of total available wetlands in
Ebro delta, RCP broods-habitat relationships within individual delta were addressed only in
the Camargue (Defos du Rau et al. 2005). No clear pattern of spatial autocorrelation was
apparent in any of the three deltas (Figure 2 to 4) ; number of distance lags was adjusted
according to sample size in each delta. Contrary to a general macroecological expectation
(Brown 1995, Watkinson et al. 2003, Guo et al. 2005) observed densities of breeding RCP are
increasing from center (Danube delta) to edge (Ebro delta) of the species global range (Table
2).
15
Figure 2 : omnidirectional variogram (6 distance lags) for RCP brood occurrence in Danube
delta in 2001
Figure 3 : omnidirectional variogram (8 distance lags) for RCP brood occurrence in Camargue
in 2001
Figure 4 : omnidirectional variogram (2 distance lags) for RCP brood occurrence in Ebro
delta in 2001
16
Occupancy modelling
- Detection probability.
Standard pre-defined models implemented in program PRESENCE indicated that
detectability was structured in two rather than one or three groups of sites and varied between
surveys (Table 3). The variable “Time” and only one spatial variable (delta identity) were
therefore incorporated into the model for detection probability, as well as other habitat
variables. AIC-based model comparison provided evidence that detectability increased
linearly through time (tlin), was higher in Ebro delta than in the other two deltas and was
positively related to water surface area (Table 4). Effect of competitor species richness on
detectability, as suggested in second best model (ΔAIC=0.3), was in fact positive, thus not
coherent with our a priori hypothesis. Competitor density may lead to changes in RCP
behaviour. However, estimates made under the lowest-AIC model were consistent with our a
priori hypotheses: this model was therefore retained for simplicity (Table 4).
17
Model
Number of parameters
AIC
ΔAIC
psi(.),p(tlin,Ebro,water)
5
295.55
0
psi(.),p(tlin,Ebro,comp)
5
295.85
0.3
psi(.),p(tlin,Ebro,%water)
5
300.33
4.78
psi(.),p(tlin,Ebro,swan)
5
301.86
6.31
psi(.),p(tlin,Ebro)
4
304.18
8.63
psi(.),p(tlin,Ebro,gull)
5
304.82
9.27
psi(.),p(tlin,Ebro,mallard)
5
305.27
9.72
psi(.),p(tlin,Ebro,harrier)
5
305.60
10.05
psi(.),p(tlin,Ebro,shore)
5
305.97
10.42
psi(.),p(tlin,Camargue)
4
309.35
13.80
2 groups, survey-specific detectability
20
314.70
19.15
psi(.),p(tlin)
3
317.81
22.26
psi(.),p(tlin,Danube)
4
318.84
23.29
2 groups, constant detectability
4
321.19
25.64
3 groups, constant detectability
6
325.34
29.79
psi(.),p(t)
10
325.47
29.92
1 group, constant detectability
2
335.29
39.74
Table 3 : model selection for RCP broods detection probability p in 2001 in the 3 study deltas,
keeping occupancy probability psi constant (pre-defined models proposed in PRESENCE are
in italic)
18
Parameter
structure Estimate
SEˆ
occupancy:
psi
constant -0.356034 0.267197
detection: p
constant -1.275112 0.393632
detection: p
linear time 0.304311 0.098558
detection: p
Ebro
0.532520 0.523699
detection: p
water
0.760683 0.252973
Table 4 : parameter estimates for RCP broods wetlands detectability model with lowest AIC :
psi(.),p(tlin,Ebro,water)
-
Occupancy probability.
Preliminary selection of
habitat variables.
This analysis was performed using
presence/absence data uncorrected for detection probability. In this first step, the set of
candidate habitat variables identified through discriminant analysis included water surface
area, shoreline index, density of swan, frequency of harrier, frequency of Myriophyllum sp.
and macrophyte genus richness. This candidate set was then used for both occupancy analyses
accounting (Table 5) and not accounting (Table 6) for detection probabilities.
Influence of habitat variables on occupancy probability while accounting for detection
probability (Table 5). In this step, we found evidence that the biogeographical factor, wetland
structure, predator frequency and food resource influenced brood occupancy. In spite of very
different observed abundances (Table 2), Camargue and Ebro delta did not differ in terms of
occupancy probability.
19
Model
Number of parameters AIC ΔAIC
psi(Danube,shore,harrier,myrio),p(tlin,Ebro,water)
9
286.33
0
psi(Danube,shore,myrio),p(tlin,Ebro,water)
8
286.87 0.54
psi(Danube,shore,harrier),p(tlin,Ebro,water)
8
287.58 1.25
psi(Danube,shore),p(tlin,Ebro,water)
7
288.20 1.87
psi(Danube,Myrio),p(tlin,Ebro,water)
7
289.07 2.74
psi(Danube,harrier),p(tlin,Ebro,water)
7
290.09 3.76
psi(Danube),p(tlin,Ebro,water)
6
292.06 5.73
psi(Danube,water),p(tlin,Ebro,water)
7
292.27 5.94
psi(Danube,swan),p(tlin,Ebro,water)
7
292.87 6.54
psi(Ebro,Danube),p(tlin,Ebro,water)
7
292.88 6.55
psi(Danube,gen),p(tlin,Ebro,water)
7
293.97 7.64
psi(Ebro),p(tlin,Ebro,water)
6
294.74 8.41
psi(.),p(tlin,Ebro,water)
5
295.55 9.22
psi(Camargue),p(tlin,Ebro,water)
6
296.54 10.21
Table 5 : model selection for RCP broods wetlands occupancy probability psi in 2001 in the 3
study deltas, using best model of detection probability p previously identified, i.e.
psi(.),p(tlin,Ebro,water)
Influence of habitat variables on occupancy probability without accounting for detection
probability. A standard logistic regression with observed presence/absence as a dependant
variable was run using this set of covariates (Table 6). We were only interested in comparing
the lowest-AIC model obtained in this analysis (ignoring detectability) with the lowest-AIC
occupancy models taking detection probability into account. The same habitat variables
influenced brood occupancy whether detection probability was accounted for or not, with the
exception of the Ebro delta location effect. Results provided evidenced of the latter effect only
when not accounting for detection probability (Table 6). However, contrary to standard
logistic regression analysis, analysis in PRESENCE (e.g., using models of occupancy
probability incorporating detection probabilities) allowed us to predict false absence sites.
20
Model
Number of parameters AIC ΔAIC
Danube Ebro shore harrier
5
96.86 0.00
Danube Ebro shore harrier myrio
6
97.51 0.65
Danube water shore harrier myrio
6
97.70 0.85
Danube Ebro shore swan harrier
6
97.81 0.96
Danube water shore harrier
5
97.92 1.06
Danube Ebro shore harrier gen
6
97.92 1.07
Danube shore harrier myrio
5
98.00 1.15
Ebro
4
98.07 1.21
Danube Ebro shore harrier myrio gen
7
98.27 1.41
Danube Ebro water shore harrier
6
98.30 1.44
Danube shore swan harrier myrio
6
98.37 1.51
Danube Ebro shore swan harrier myrio
7
98.44 1.58
Danube water harrier myrio
5
98.45 1.60
Ebro
5
98.54 1.69
Danube water shore swan harrier myrio
7
98.58 1.72
Danube water shore swan harrier
6
98.77 1.91
shore harrier
shore harrier gen
Table 6. Standard logistic regression models of lake occupancy by RCP broods in 2001 in the
3 study deltas. Only models with ΔAIC<2 are shown.
Complementary selection of habitat variables. We first corrected false absence data for sites
with the highest occupancy probabilities among apparent absences in order to assess whether
additional habitat variables that were not selected by discriminant analysis improved the
current occupancy models. We used the overall proportion of occupied sites (0.3809 ; SEˆ
=0.0532) estimated from model psi(Danube,Shore,harrier,myrio), p(tlin,Ebro,water) (Table 5)
to estimate the number of false RCP brood absences from the observed proportion of
occupied site (0.3412). The estimated number of false absences was therefore:
85 x (0.3809-0.3412) = 3.3745. The 3 apparently unoccupied sites with highest occupancy
probability estimated from the same model
(psi(Danube,Shore,harrier,myrio),p(tlin,Ebro,water)) were thus re-assigned as presence in a
second deltas-centered discriminant analysis which excluded the 6 habitat variables already
selected by the first discriminant analysis. The new set of candidate habitat variables
21
identified through this second discriminant analysis included number of islets, competitor
diversity, swan density and frequency of Potamogeton pectinatus.
Complementary analysis of the influence of habitat variables on occupancy probability while
accounting for detection probability. These 4 new candidate variables were incorporated into
lowest AIC model psi(Danube,Shore,harrier,myrio),p(tlin,Ebro,water). The resulting models
are compared in Table 7.
Model
Number of parameters AIC ΔAIC
psi(Danube,shore,harrier,myrio,comp),p(tlin,Ebro,water)
10
284.19
0
psi(Danube,shore,harrier,myrio),p(tlin,Ebro,water)
9
286.33 2.14
psi(Danube,shore,harrier,myrio,swan),p(tlin,Ebro,water)
10
286.58 2.39
psi(Danube,shore,harrier,myrio,islet),p(tlin,Ebro,water)
10
288.25 4.06
psi(Danube,shore,harrier,myrio,pot),p(tlin,Ebro,water)
10
288.29 4.10
Table 7 : model selection for RCP broods wetlands occupancy probability psi in 2001 in the 3
study deltas, using best model of occupancy probability previously identified and new set of
DA-selected habitat variables
Competitor species richness (“ comp”), although not selected in the first discriminant
analysis, provided substantial improvement to occupancy models (Table 7). However, we
have reservations concerning the apparent influence of observed harrier frequency on RCP
brood occupancy because harrier detectability was not taken into account. Assuming that use
of the sampled wetland by harriers foraging around their breeding territories was random, we
used the species co-occurrence model implemented in program PRESENCE (MacKenzie et
al. 2004) to check whether there is a relationship between occupancy probability in the two
species. Starting from the already known detection and occupancy models for RCP broods,
we addressed spatial effect of delta and open water area, as well as time effect on the
probability harrier was present and detected. We addressed the effect of breeding habitat
(reedbed) area, of wetland structure at the landscape scale (mean distance to 5 closest
wetlands) and of RCP competitors as index for harrier prey diversity on wetland use by
harrier. The resulting model was re-run under specified co-use independency assumption
(Table 8). This latter model proved more performance (AIC = 675.77) than the model without
constraint of co-use independency (ΔAIC = 3.73). This suggests that the apparent interaction
22
between RCP broods and harrier is spurious. Indeed, harrier frequency and RCP brood
occupancy were likely to be affected by the same confounding variable (“comp”).
Consequently, we discarded the variable « harrier » from best identified RCP occupancy
models.
For comparison purpose, we also ran the best habitat variable set identified without taking
detectability into account (Table 6), including both delta effects, shore indentation and harrier
frequency. The final RCP brood occupancy model selection is shown in Table 9. Wetland
occupancy probability by RCP broods is lower in Danube delta than in the other two deltas
(Table 10) and decreases as frequency of Myriophyllum sp increases. Occupancy probability
also increases with shore indentation and waterbird species richness. We did not find evidence
of lack of fit of the most global model (using 10000 bootstrap samples : χ² = 396.39, p =
0.53;ĉ=0.63). For validation purpose, the lowest AIC occupancy model was run on the
Camargue test set : its predictive power proved satisfactory, considering that test and training
sample sets were different (Figure 5).
23
Model
Number of parameters
AIC ΔAIC
same model as preceeding but with independent occupancy by harrier and RCP
15
675.77 0.00
psiRCP(Danube,shore,myrio,comp)harrier(comp,mdist),pRCP(tlin,Ebro,water)harrier(tlin,Ebro)
16
679.50 3.73
psiRCP(Danube,shore,myrio,comp)harrier(comp),pRCP(tlin,Ebro,water)harrier(tlin,Ebro)
15
682.44 6.67
psiRCP(Danube,shore,myrio,comp)harrier(.),pRCP(tlin,Ebro,water)harrier(tlin,Ebro)
14
684.76 8.99
psiRCP(Danube,shore,myrio,comp)harrier(mdist),pRCP(tlin,Ebro,water)harrier(tlin,Ebro)
15
684.89 9.12
psiRCP(Danube,shore,myrio,comp)harrier(.),pRCP(tlin,Ebro,water)harrier(tlin,Ebro,Danube)
15
685.45 9.68
psiRCP(Danube,shore,myrio,comp)harrier(Danube),pRCP(tlin,Ebro,water)harrier(tlin,Ebro)
15
685.97 10.20
psiRCP(Danube,shore,myrio,comp)harrier(reed),pRCP(tlin,Ebro,water)harrier(tlin,Ebro)
15
686.03 10.26
psiRCP(Danube,shore,myrio,comp)harrier(Camargue),pRCP(tlin,Ebro,water)harrier(tlin,Ebro)
15
686.05 10.28
psiRCP(Danube,shore,myrio,comp)harrier(.),pRCP(tlin,Ebro,water)harrier(tlin,Ebro,water)
15
686.74 10.97
psiRCP(Danube,shore,myrio,comp)harrier(Ebro),pRCP(tlin,Ebro,water)harrier(tlin,Ebro)
15
686.93 11.16
psiRCP(Danube,shore,myrio,comp)harrier(.),pRCP(tlin,Ebro,water)harrier(tlin,Camargue)
14
697.62 21.85
psiRCP(Danube,shore,myrio,comp)harrier(.),pRCP(tlin,Ebro,water)harrier(tlin)
13
701.82 26.05
psiRCP(Danube,shore,myrio,comp)harrier(.),pRCP(tlin,Ebro,water)harrier(tlin,Danube)
14
703.35 27.58
psiRCP(Danube,shore,myrio,comp)harrier(.),pRCP(tlin,Ebro,water)harrier(.)
12
703.50 27.73
Table 8 : RCP broods and harrier co-occurrence model selection ; probabilities of presence and detection p and of wetland use psi are estimated for
both species in 2001 in the 3 study deltas, with same detection probabilities wether the co-occuring species is present or not (see MacKenzie et al.
2004).
24
Model
Number of parameters AIC ΔAIC
best model after co-occurrence analysis: psi(Danube,shore,myrio,comp),p(tlin,Ebro,water)
9
283.31 0.00
best model after 2nd DA: psi(Danube,shore,harrier,myrio,comp),p(tlin,Ebro,water)
10
284.19 0.88
best model after 1st DA: psi(Danube,shore,harrier,myrio),p(tlin,Ebro,water)
9
286.33 3.02
best model without taking detection probability into account: psi(Danube,Ebro,shore,harrier),p(.)
6
323.03 39.72
Table 9 : final model selection for RCP broods wetlands occupancy probability psi in 2001 in the 3 study deltas. DA = Discriminant Analysis.
25
Parameter
structure
Estimate
SEˆ
occupancy: psi
constant 0.014860 0.352889
occupancy: psi
Danube -1.5605050.657462
occupancy: psi
shore
0.701763 0.361045
occupancy: psi
myrio
-0.4189540.317578
occupancy: psi
comp
0.669753 0.296604
detection: p
constant -1.0564920.379243
detection: p
linear time 0.278029 0.097812
detection: p
Ebro
0.419010 0.513534
detection: p
water
0.770545 0.250053
Table 10 : parameter estimates for RCP broods wetlands occupancy model with lowest AIC :
psi(Danube,shore,myrio,comp),p(tlin,Ebro,water)
1
0,9
true positive rate
0,8
0,7
0,6
0,5
0,4
0,3
0,2
0,1
0
0
0,2
0,4
0,6
0,8
1
false positive rate
Figure 5 : ROC curve for the Camargue test set occupancy predicted using the best model
identified on the 3-deltas training set, i.e. : psi(Danube,shore,myrio,comp),p(tlin,Ebro,water)
26
Abundance modeling
-
Detection probability.
The double-sampling scheme performed on 10 Camargue wetlands produced individual
encounter histories for 29 RCP broods, capture probabilities of which were best modeled with
total densities of likely competitors or habitat surface areas. Because of the small sample size,
we used the lowest AICc model to predict RCP brood detection probabilities (Table 11). The
total surface area (water+vegetation areas) of each sampled wetlands in the three deltas was
therefore used as a covariate of the detectability p of individual RCP brood on the
corresponding wetland the following relationship :
logit(p) = -0.8476429 + 1.7986725*(total wetland surface area)
Model
Number of parameters AICc ΔAICc
{S(t),g''(.),g'(.),p(.,.,water+veg)=c(.,.,water+veg)}
8
238.00 0.00
{S(t),g''(.),g'(.),p(.,.,compdens)=c(.,.,compdens)}
8
238.90 0.89
{S(t),g''(.),g'(.),p(.,.,water)=c(.,.,water)}
8
239.27 1.26
{S(t),g''(.),g'(.),p(.,.)=c(.,.)}
7
239.90 1.90
{S(t),g''(.),g'(.),p(.,.,comp)=c(.,.,comp)}
8
241.32 3.31
{S(t),g''(.),g'(.),p(.,.,gull)=c(.,.,gull)}
8
241.62 3.61
{S(t),g''(.),g'(.),p(.,.,%water)=c(.,.,%water)}
8
241.95 3.94
{S(t),g''(.),g'(.),p(.,.,shore)=c(.,.,shore)}
8
242.25 4.25
{S(t),g''(.),g'(.),p(T,.)=c(T,.)}
11
243.40 5.39
{S(t),g''(t),g'(.),p(T,.)=c(T,.)}
14
244.65 6.65
{S(.),g''(.),g'(.),p(T,.)=c(T,.)}
8
247.53 9.52
{S(t),g''(t),g'(t),p(T,.)=c(T,.)}
16
251.00 12.99
{S(t),g''(t),g'(t),p(T,t)=c(T,t)}
26
272.17 34.17
{S(t),g''(t),g'(t),p(T,t),c(T,t)}
36
328.95 90.94
Table 11 : huggins’ Robust-design modelling of survival, temporary emigration and
immigration, and capture probabilities of RCP broods in 2001 in the Camargue delta. S:
survival rates, g” : temporary emigration, g’ : temporary immigration, p : capture and c :
recapture probabilities.
27
- Abundance. We did not include frequency of harrier as a covariate of the
detectability-adjusted abundances (due to its spurious relationship with brood occupancy; see
above). Since there were several competing models with close AIC values, we only reported
respective weights of the variables composing the models with ΔAIC<3 (Figures 6 to 8).
1,00
0,80
0,60
0,40
0,20
0,00
Danube
water
myrio
Ebro
harrier
shore
swan
gen
Figure 6 : AIC-weights supporting the habitat variables composing the ΔAIC<3 poisson
regression models for RCP brood unadjusted abundances in the three deltas in 2001
We did not find evidence of substantial lack of fit of the most global habitat model for
unadjusted abundances ( ĉ=0.85). The Danube delta was under-populated compared to both
other deltas. The largest abundance was found in the Ebro delta. Water area had a positive
effect on RCP brood abundance and frequency of Myriophyllum sp. a negative one (Table
12). Other variables were of lesser importance for abundance prediction.
28
estimate
parameter
SEˆ Wald stat.
p
intercept
-1.52
0.48
9.90
0.0017
Outside Danube
0.72
0.28
6.73
0.0095
Outside Ebro
-0.58
0.34
2.91
0.0883
water
0.00
0.00
2.95
0.0859
shore
0.18
0.19
0.86
0.3527
swan
-1.27
1.33
0.91
0.3408
harrier
0.54
0.33
2.67
0.1023
myrio
-0.02
0.02
2.25
0.1336
gen
0.14
0.19
0.52
0.4690
Table 12 : parameter estimates for RCP broods unadjusted abundance model.
1
0,8
0,6
0,4
0,2
t
po
at
er
w
ge
n
Eb
ro
et
isl
m
yr
io
sw
an
m
p
co
or
e
sh
Da
n
ub
e
0
Figure 7 : respective AIC-weights supporting the habitat variables composing the ΔAIC<3
poisson regression models for RCP brood adjusted abundances in the three deltas in 2001
Models of detectability-adjusted abundances provided evidence of substantially different
importance of several habitat variables compared to the unadjusted abundance models (Figure
7). Density of swan broods proved to be of major importance as RCP brood abundance
covariate, so we re-addressed the hypothesis of an inter-specific interaction using the cooccurrence models developed by Mackenzie et al. (2004), assuming demographic closure of
29
swan brood populations. We incorporated most of the same habitat variables for swan
occurrence as for RCP broods except that, as a highly territorial species, we expected a
positive influence of competitor diversity on swan detectability, but no influence on wetland
occupancy. Neither did we expect any effect of predators on brood abundance because the
species is of relatively large size. Starting from the already known detection and occupancy
models for RCP broods, we addressed spatial effect of delta, shoreline indentation and open
water area, as well as time effect on swan brood detection. We added islet numbers, as well as
macrophyte frequency as potential covariates of swan brood occupancy. Best resulting model
was re-ran under specified co-use independency assumption (Table 13). This latter model
proved more performance (AIC= 474.31) than the model without constraint of co-use
independency (ΔAIC=2.95). Again, this result suggests that the apparent interaction between
RCP broods and swan broods results from independent habitat selection processes in the two
species. Consequently, we discarded the variable « swan » from RCP brood abundance
analyses (Figure 8).
Model
same model as preceeding but with independent occupancy by swan and RCP
psiRCP(Danube,shore,myrio,comp)swan(Danube),pRCP(tlin,Ebro,water)swan(Ebro,Danube,comp)
psiRCP(Danube,shore,myrio,comp)swan(Danube,water),pRCP(tlin,Ebro,water)B(Ebro,Danube,comp)
psiRCP(Danube,shore,myrio,comp)swan(Ebro),pRCP(tlin,Ebro,water)swan(Ebro,Danube,comp)
psiRCP(Danube,shore,myrio,comp)swan(Danube,myrio),pRCP(tlin,Ebro,water)swan(Ebro,Danube,comp)
psiRCP(Danube,shore,myrio,comp)swan(Danube,pot),pRCP(tlin,Ebro,water)swan(Ebro,Danube,comp)
psiRCP(Danube,shore,myrio,comp)swan(Danube,shore),pRCP(tlin,Ebro,water)swan(Ebro,Danube,comp)
psiRCP(Danube,shore,myrio,comp)swan(Ebro,Danube),pRCP(tlin,Ebro,water)swan(Ebro,Danube,comp)
psiRCP(Danube,shore,myrio,comp)swan(Danube,gen),pRCP(tlin,Ebro,water)swan(Ebro,Danube,comp)
psiRCP(Danube,shore,myrio,comp)swan(Danube,islet),pRCP(tlin,Ebro,water)swan(Ebro,Danube,comp)
psiRCP(Danube,shore,myrio,comp)swan(Camargue),pRCP(tlin,Ebro,water)swan(Ebro,Danube,comp)
psiRCP(Danube,shore,myrio,comp),pRCP(tlin,Ebro,water)swan(Ebro,Danube,comp)
psiRCP(Danube,shore,myrio,comp),pRCP(tlin,Ebro,water)swan(Ebro,Danube,shore)
psiRCP(Danube,shore,myrio,comp),pRCP(tlin,Ebro,water)swan(Ebro,Danube,%water)
psiRCP(Danube,shore,myrio,comp),pRCP(tlin,Ebro,water)swan(Ebro,Danube)
psiRCP(Danube,shore,myrio,comp),pRCP(tlin,Ebro,water)swan(Ebro,Danube,water)
psiRCP(Danube,shore,myrio,comp),pRCP(tlin,Ebro,water)swan(Ebro,Danube,gull)
psiRCP(Danube,shore,myrio,comp),pRCP(tlin,Ebro,water)swan(Ebro)
psiRCP(Danube,shore,myrio,comp),pRCP(tlin,Ebro,water)swan(Camargue)
psiRCP(Danube,shore,myrio,comp),pRCP(tlin,Ebro,water)swan(Danube)
psiRCP(Danube,shore,myrio,comp),pRCP(tlin,Ebro,water)swan(Ebro,Danube,harrier)
psiRCP(Danube,shore,myrio,comp),pRCP(tlin,Ebro,water)swan(.)
psiRCP(Danube,shore,myrio,comp),pRCP(tlin,Ebro,water)swan(tlin)
# parameters
AIC
15
16
17
16
17
17
17
17
17
17
16
15
15
15
14
15
15
13
13
13
15
12
13
474.31
477.26
477.56
478.03
479.17
481.61
481.72
482.37
482.78
486.88
493.85
494.27
497.07
499.91
504.17
505.28
505.54
506.75
506.77
506.82
510.48
510.93
512.04
delta AIC
0.00
2.95
3.25
3.72
4.86
7.30
7.41
8.06
8.47
12.57
19.54
19.96
22.76
25.60
29.86
30.97
31.23
32.44
32.46
32.51
36.17
36.62
37.73
Table 13 : RCP broods and swan brood co-occurrence model selection ; probabilities of
presence and detection p and of wetland use psi are estimated for both species in 2001 in the 3
study deltas, with same detection probabilities wether the co-occurring species is present or
not (see MacKenzie et al. 2004).
The same four variables than in the final occupancy model clearly stood out as abundance
covariates, shore indentation and waterbird guild richness being positive factors, and Danube
30
delta and high Myriophyllum sp frequency corresponding to decreased abundances (Table 14).
The most global habitat model for adjusted abundances showed a good fit to the data, with
only minimal over-dispersion: ĉ=1.15.
1,0
0,8
0,6
0,4
0,2
t
po
le
t
is
n
ge
er
wa
t
ro
Eb
yr
io
m
m
p
co
or
e
sh
D
an
ub
e
0,0
Figure 8 : respective AIC-weights supporting the habitat variables composing the ΔAIC<3
poisson regression models for RCP brood adjusted abundances in the three deltas in 2001
parameter group-level estimate SE Wald stat.
intercept
p
-2.42
0.49
24.46
0.0000
0.98
0.21
20.80
0.0000
shore
0.45
0.16
8.12
0.0044
comp
0.34
0.08
15.58
0.0001
myrio
-0.02
0.01
4.22
0.0399
Danube
0
Table 14 : parameter estimates for most supported model for RCP broods adjusted abundance
Macroecological hypotheses
- Are occupancy and abundance decreasing from center (Danube delta) to edge (Ebro
delta) of the species global range (Brown 1995, Watkinson et al. 2003, Guo et al. 2005) and
from preserved to degraded breeding areas? Both occupancy (Figure 9) and abundance
(Figure 10) models indicated a clear decrease in presence probability and population density
31
of breeding RCP in Danube delta compared to both other deltas. The hypothesis of a decrease
in occupancy and abundance from center to edge of the species range was thus not
corroborated for the RCP breeding area encompassing Western and Eastern Europe. Models
for unadjusted abundance (Figures 6 & 11) suggested an increase in Ebro delta that was not
confirmed when detectability was taken into account (Table 9, Figures 8 & 10).
ajusted occupancy rate
1,0
0,8
0,6
0,4
0,2
0,0
Danube
Camargue
Ebre
Figure 9 : wetlands occupancy rates by RCP in Danube, Rhône and Ebro deltas in 2001
breeding population density(/ha)
predicted by the lowest-AICc detection-occupancy model (Table 9)
0,8
0,6
0,4
0,2
0,0
Danube
Camargue
Ebre
-0,2
Figure 10 : RCP breeding population sizes/ha (and densities confidence interval) in Danube,
Rhône and Ebro deltas in 2001 adjusted for detectability
32
breeding population density(/ha)
0,4
0,2
0,0
Danube
Camargue
Ebre
-0,2
Figure 11 : RCP breeding population sizes/ha (and densities confidence interval) observed in
Danube, Rhône and Ebro deltas in 2001
- Is there habitat overfilling due to habitat loss ? At European scale (Table 15) as well
as at West Mediterranean scale (Table 16), several models accounting for different hypotheses
for the habitat availability-RCP brood abundance relationship were supported by the data :
several polynomial and linear models had ΔAICc<2. However, the hypothesis of a departure
from linearity (i.e., proportionality) in habitat filling by RCP broods was clearly not excluded
for the habitat-abundance model. The positive slope parameter estimate for reedbed² in the
lowest AICc model (0.00310: 95% CI = [0.00163 ; 0.00457]) provided evidence of potential
overfilling of reedbeds habitat by nesting RCP. This may reflect a relative and current loss of
reedbed (i.e., RCP nesting habitat) at European scale.
Model
AICc
ΔAICc
reedbed² Ebro*reedbed
275.08
0.00
reedbed Ebro*reedbed
275.19
0.11
reedbed Ebro*reedbed²
276.56
1.48
Ebro*reedbed²
277.05
1.97
reedbed Danube*reedbed² Ebro*reedbed
277.12
2.04
reedbed Ebro*reedbed
Table 15: best reedbed area linear regression models for adjusted brood abundance in n=32
lakes occupied or predicted to be occupied in the Danube, Ebro and Camargue deltas in 2001.
33
Model
AICc
ΔAICc
reedbed²
Ebro*reedbed
230.25
0.00
reedbed
Ebro*reedbed
230.37
0.12
reedbed
Ebro*reedbed²
230.90
0.65
Ebro*reedbed Ebro*reedbed²
231.01
0.76
reedbed
231.09
0.84
reedbed²
231.19
0.94
reedbed
reedbed²
232.10
1.85
reedbed
Ebro*reedbed Ebro*reedbed²
232.51
2.26
Table 16: best reedbed area linear regression models for adjusted brood abundance in n=26
lakes occupied or predicted to be occupied in the Ebro and Camargue deltas in 2001
Interestingly, however, if we consider unadjusted brood abundance, the influence of reedbed
area would have unambiguously indicated a clear linear habitat filling pattern (Tables 17 &
18). Had we relied on these results to draw inferences about habitat loss, the hypothesis of a
current habitat loss would not have been supported because brood abundance would simply be
proportional to available reedbed area (and broods would not be packing into larger patches).
Using abundances uncorrected for detectability may thus hamper early detection of the
influence of habitat loss on RCP brood abundance.
Model
reedbed
Ebro*reedbed
reedbed Danube*reedbed Ebro*reedbed
AICc
ΔAICc
239.12
0.00
241.78
2.66
Table 17: best reedbed area linear regression models for uncorrected brood abundance in
n=29 lakes observed to be occupied in the Danube, Ebro and Camargue deltas in 2001
Model
reedbed
Ebro*reedbed
reedbed
reedbed²
Ebro*reedbed
AICc
ΔAICc
196.66
0.00
199.61
2.95
Table 18: best reedbed area linear regression models for uncorrected brood abundance in
n=23 lakes observed to be occupied in the Ebro and Camargue deltas in 2001
34
- Assuming source-sink dynamics (Pulliam & Danielson 1991, Brown 1995) and
dispersal among sites due to habitat overfilling, does occupancy probability of low quality
patches (sink) increase as distance to nearest high-quality patch (source) decreases? Best
occurrence threshold identified by ROC curve was 0.4 for both standard and detectabilityadjusted occupancy prediction on the training set. This threshold was used to discriminate
between favorable and unfavorable habitat patches. We addressed the influence of distance
from the nearest high-quality patch on occupancy probability only in the Camargue, where
densities are the highest and where a habitat overfilling effect is thus most likely to occur, as
suggested above. In the Camargue, occupancy of low quality patches (sink habitat) is best
accounted for by distance to nearest high-quality patch (source) (parameter estimate: -2.15
95% CI = [-4.79 ; 0.50]): Likelihood-ratio test : χ 2 = 4.23 ; p = 0.04 ; df = 1 ; ĉ = 1.038 for
best model in table 19. Dispersal from source to sink habitat may explain this pattern.
Model
Number of
parameters AICc ΔAICc
distance to closest source (including false-absences)
2
19.64 0.00
adjusted occupancy
2
21.86 2.22
3
22.03 2.39
adjusted occupancy & distance to closest source (including falseabsences)
Table 19: logistic regression models for RCP brood occurence in n=18 unfavorable patches in
the Camargue in 2001 after accounting for detection errors
Interestingly, when using unadjusted occupancy rates and distances to source patches, RCP
occurrence in unfavorable patches is not well predicted by distance to favorable ones :
Likelihood-ratio test : chi²=0.76 ; p=0.38 ; c-hat=1.126 for lowest AIC-model in table 20.
35
Model
Number of
parameters AICc ΔAICc
distance to closest source (excluding false-absences)
2
30.24 0.00
unadjusted occupancy
2
31.00 0.76
3
32.90 2.65
unadjusted occupancy & distance to closest source (excluding falseabsences)
Table 20: logistic regression models for RCP brood occurence in n=23 unfavorable patches in
the Camargue in 2001 without accounting for detections errors
- Is there a metapopulation dynamics, due to possible source-sink process, in the
Camargue subpopulation? We tried to discriminate between metapopulation and habitatfilling dynamics in the highest-density subpopulation where a source-sink dispersal effect was
furthermore suspected. The linear regression of logit-tranformed adjusted occupancy as a
function of log-transformed adjusted density (r = 0.31 ; p = 0.049) had an estimated slope b̂ =
∧
0.90 ( SE = 0.44). This suggests the existence of a habitat-filling process predominating in the
RCP brood distribution in the Camargue. When using unadjusted occupancy and density, the
linear relationship between those two variables (r = 0.18 ; p = 0.271) had an estimated slope
∧
b̂ =0.97 which was both close to one and not well estimated ( SE = 0.87).
DISCUSSION
There are relatively few large-scale habitat surveys for rare species, notably because of
difficulties in collecting sufficient presence data as well as in obtaining valid/confirmed
absence data (Engler et al. 2004, Guisan et al. 2006). Although making inferences only from
presence data is now a well-developped approach (Engler et al. 2004), using confirmed
absence data is often preferable, especially when false-absence can be identified and
eliminated, and absence data are thus reliable ones (Brotons et al. 2004). Furthermore,
abundance estimated without accounting for detection is likely to be biased. In this respect,
estimating detection errors is a crucial prerequisite of any habitat analysis using absence and
abundance data (Defos du Rau et al. 2005, MacKenzie et al. 2006), as admitted by Décarie et
al. (1995) in a very similar study but assuming perfect detection.
36
Importance of detection issues to identify limiting factors
Not taking imperfect detection into account would have erroneously led us to consider
wetland size and swan density as factors affecting brood abundance (Table 21). The relative
lack of influence of wetland size on brood abundance was counter-intuitive and would not
have been acknowledged without taking detectability into account. Indeed, relationships
between duck reproduction and patch size may not be straightforward (Mack & Flake 1980).
For example, duck nest success was found to be equivalent in small and large patches as a
result of interaction between patch size and predation (Horn et al. 2005).
Without taking detectability into account, our result might also have contributed to
oversimplify the debate on the commonly assumed but not clearly demonstrated spatial
competition impact of mute swan on other breeding ducks (Conover & Kania 1994). Both
these factors might have been erroneously acted upon by managers of RCP wetlands (e.g. by
preventing swan from breeding in specific RCP breeding areas) with probably few costeffective results on RCP reproduction.
Both RCP brood occupancy and abundance analyses taking detectability into account
provided evidence of the importance of the same four variables: positive effects of shore
indentation and waterbird diversity, and negative effects of Danube delta and frequency of
Myriophyllum sp.
-
Danube: Contrary to a general macroecological prediction (Brown 1995, Watkinson et
al. 2003, Guo et al. 2005), breeding RCP occupancy and density increased from the
centre toward the edge of its global distribution area, Red-crested Pochard being
originated from Central Asia (Gay et al. 2004). This was all the more counter-intuitive
as Danube delta was considered a traditional stronghold of the species (Tucker &
Heath 1994) and a much more preserved ecosystem than both other largely reclaimed
and degraded deltas. This pattern was fully demonstrated by both occupancy and
abundance models and suggests that Danube might be in fact at the edge of the Central
Asian or the West European population or situated within an intermediate, possibly
reduced third population (Figure 1). Low densities observed for this Eastern European
and Black Sea population might increase its extinction risk. Further field surveys and
genetic studies are thus needed to identify putative limits of those populations so that a
total population size can be monitored over a known range and thus possibly harvested
in a more sustainable way (Gay et al. 2004). Effect of Ebro delta on occupancy or
37
abundance predictions was weak, suggesting comparable levels of occupancy and
abundance in both Camargue and Ebro deltas, in spite of the much larger numbers
observed in the latter.
-
As expected and known for long by wetland managers, shore indentation positively
influenced brood occupancy and abundance. Atiénzar et al. (2005) have shown that
both presence and abundance of another Mediterranean rare duck, the White-headed
Duck Oxyura leucocephala, during breeding season, were also primarily and
positively affected by shoreline indentation. Naturaly shaped shoreline should thus be
preferred to dikes and embankments of recreated or partially reclaimed water bodies.
Water levels variations should contribute to shoreline irregularities.
-
Frequency of Myriophyllum negatively affected brood occupancy and abundance. This
macrophyte is not known as a preferred food resource (Snow and Perrins, 1998), and
is indicative of relatively constant water level (Grillas 1992), as opposed to temporary
and irregular flooding regime of typical naturally functioning Mediterranean wetlands.
Negative effect of Myriophyllum frequency probably suggests preference for irregular
water regime rather than real avoidance of this macrophyte. Myriophyllum is a highly
competitive and productive colonizing genus favoured by low variability of water
levels, which decreases macrophytes species richness (Grillas, 1992), and thus leads to
decreased diversity and quality of preferred RCP food. Negative effect of
Myriophyllum sp frequency was potentially important at both local (Defos du Rau et
al. 2005) and European scales and in spite of using unadjusted frequency estimates for
this food resource, its impact on RCP occupancy and abundance appears consistent.
-
RCP competitor species richness was of major importance as a covariate of both
occupancy and abundance but would have gone undetected without preliminary
analyses of detection and particularly false absence. Under the terminology of Wilson
(1999), this group of species could have been considered as a spatial distribution Beta
guild of Mediterranean wildfowl or as a Alpha guild of nesting and foraging wetland
birds. We primarily used the latter definition to model competition strengh, as it is a
major driver in the Alpha guild composition (Wilson 1999). However, we did not
observe the expected negative effect on brood occupancy and abundance; on the
contrary, our results provided evidence of a positive effect of competitor species
richness. This probably refers to a spatial distribution Beta guild and may indicate a
tendency of RCP to breed more often and more densely among rich wildfowl guilds,
possibly because of confounding habitat features (Guisan & Thuiller 2005) like e.g.
38
low predation rate (Horn et al. 2005) or complex vegetation structure increasing niche
diversity.
supported after detection
effect
supported wrongly without
adjustment
sign
detection adjustment
sign
composition
shore indentation
+
water area
+
food resources
Myriophyllum sp
-
competitors diversity
+
swan density
-
harrier frequency
+
Ebro
+
water
management
habitat
landscape
competition
predation
delta
Danube
-
range density
cline
Danube<Camargue & Ebro
Camargue<Ebro
possibly in the Camargue
habitat loss
& Ebro
no
possibly in the Camargue
no
habitat filling
?
macroecology
source-sink
dispersal
distribution
dynamic
Table 21: summary of difference in results computed with and without taking detection into
account
We did not find any autocorrelated spatial pattern in occupancy, and our brood occupancy
model appeared to have satisfactory predictive power on a global and regional scale.
However, in the Camargue, distance to closest source patch appeared to be a better occupancy
covariate in unfavorable patches than habitat, which suggests some marginal spatial
autocorrelation at the local landscape scale, contrary to the regional scale. Part of the omission
errors of our predictive model in the Camargue (e.g. Figure 5) may thus be explained by this
39
source-sink dynamics (Guisan & Thuiller 2005), with brood occurring in low-quality patches.
Prediction performance evaluated in the Camargue by area under curve in figure 5, yet
satisfactory, is thus probably an underestimate of the genuine predictive power of our model.
Although inferred from a reduced data set, our results suggest that there might be a current
depletion/loss in RCP breeding habitat in both West-European deltas, which are regional RCP
strongholds. This habitat loss may cause habitat overfilling, which may in turn lead to sourcesink dynamics that we suspected to occur in the Camargue. However, this consistent pattern
requires further studies to be confirmed because it is inferred from a small data set.
Furthermore, this would suggest that the RCP breeding population functions as a
metapopulation, which detectability-unadjusted analysis would have wrongly contributed to
maintain as a competing hypothesis ( b̂ =0.97). However, the metapopulation hypothesis was
not corroborated by our detectability-adjusted analysis based on data from the Camargue. Our
analyses provided stronger support for the hypothesis of Habitat-filling model in this location.
It is important to note that we would not have detected this habitat loss pattern (inducing
source-sink dynamics) if we had used data unadjusted for detectability.
Management recommendations
Natural habitat loss in Mediterranean wetlands remains a well studied issue but still not a well
addressed one : it is urgent to stop wetland destruction and fragmentation. Nearly 30,000 ha
of wetland habitats have been reclaimed in the Camargue during the last 30 years (Tamisier
and Grillas, 1994) and 16% of the Danube delta has been reclaimed, mostly for agriculture
and forestry between 1980 and 1989 (Munteanu 1996). Although many warnings and
recommendations have been issued against wetlands reclamation, it is still currently occurring
in the Camargue (Mathevet and Tamisier, 2002), albeit at a slower pace in recent years.
The 3 habitat factors identified as relevant to wetland use by breeding RCP all point toward a
unique recommandation that
is to
maintain naturally-functioning ecosystems of
Mediterranean wetlands:
-
to maximise shore indentation, original wetland shape must be conserved or restored,
which involves avoiding dikes and embankments. Dikes and embankments are of
increased use in Mediterranean wetlands to increase water control and reduce water
level variability (Aznar et al. 2003).
40
-
to minimise frequency of Myriophyllum sp, water levels can be left uncontrolled or
can mimic naturally varying flooding conditions, including temporary drying-offs in
summer
-
to maximise species richness of wildfowl guild, a minimum disturbance on a
maximum natural habitat area is required during breeding season
More generally, it seems that the best management option would be to minimise management
interventions, i.e. to minimise disturbance or transformation of the natural Mediterranean
wetland ecosystem. A further option for water management of wetlands from within the three
study deltas might be to mimic natural water levels of the river in order to reproduce natural
hydrological conditions including water level variations, temporary flooding and shore
shaping due to successive flooding.
Lastly, low densities observed and modelled in the Danube delta former stronghold are a case
for concern; first estimates of harvest rates and regional population size are urgently required
through field surveys including aerial ones. Genetic analysis quantifying movement of
individuals between this eastern European and Black sea breeding (sub)population to the
West-European or to the Central Asian populations are urgently needed, also to examine
whether this potential Black Sea population is an isolated conservation unit.
Perspectives for approach improvement
As expected at such a large, regional scale, biotic interactions like competition or predation
did not appear to influence breeding RCP occupancy and abundance (Guisan & Thuiller
2005). Only one competition index, guild species richness, was retained in occupancy and
abundance models, but as a positive factor contrary to the a priori hypothesis. This variable
can in fact be used as an index of habitat area (r²=0,28) but also of ecosystem
richness/integrity/complexity and would need to be adjusted to species detectability to be used
as such (Cam et al. 2002), which is beyond the scope of the present work. Indeed, wildfowl
guild species richness may be biased by detection imperfections which may have had
consequences on our brood habitat models. However, we did not correct observed species
richness of the wildfowl guild for species detectability because we assumed that only most
abundant and widespread, thus most visible species (Royle & Nichols 2003) would
potentially be effective competitors of RCP broods for space and food resources. Densities of
these most abundant competitor species were indeed included in our analyses (Appendix)
although we did not find evidence of their effect on either RCP occupancy or abundance.
41
We did not use any recently developed capture-recapture-based abundance models (Royle &
Nichols 2003, Royle 2004) to estimate RCP population size from point-count data on each of
the 85 wetlands because of high abundance (and thus detection) heterogeneity between and
within deltas combined with relatively small sample size. To correct local abundance
estimates for detection errors, we used a simpler approach based on a Pollock’s Robust
Design implemented previously in Camargue (Defos du Rau et al. 2005).
Lastly, improving prediction accuracy should be possible by increasing sampling effort
(Hirzel & Guisan 2002), e.g. in Danube delta and extending it in Volga delta, although our
sampling effort might already be considered satisfactory in space (Stockwell & Peterson
2002) and time (MacKenzie & Royle 2005) and our sampling design met recommendations
for systematic sampling (Hirzel & Guisan 2002) and niche-based design (Guisan et al. 2006).
Conclusion
Studying species-habitat relationship becomes more and more complex as issues like e.g.
spatial autocorrelation or sampling design are considered of major relevance and statistical
inference tools are developed (Guisan & Thuiller 2005 for a review). Species detection
probability is one of these issues that have received a lot of attention since publication of
papers by e.g. Bibby & Buckland (1987) and more recently Nichols et al. (2000), MacKenzie
et al. (2002), Pollock at al. (2002), and Thompson (2002). As shown by several theoretical
and simulation studies (Moilanen 2002, MacKenzie et al. 2006), the detection issue (including
the false-absence one) has the potential to call several conclusions drawn in the past into
question, not only on species monitoring, but also on species-habitat surveys and on species
dynamic and macroecology. The present work is, to our knowledge, one of the first
observational approaches based on data that corroborates and quantifies the impact of
accounting for detection in macroecology analyses (but for inferences on habitat use see
MacKenzie 2006, Defos du Rau et al. 2005, Vojta 2005, and following articles from the
special section of Journal of Wildlife Management 69 on value and utility of presenceabsence data to wildlife monitoring and research). We reached substantially different
conclusions on breeding RCP habitat use and macroecology whether we took detection into
account, or not. Moreover, these results are also relevant for conservation purposes because
they apply to a rare game species. Robust and cost-effective conclusions were thus all the
42
more needed that they relied on important fieldwork efforts and aimed at developing
management recommendations.
REFERENCES
Alldredge, J.R., and J.Griswold. 2006. Design and Analysis of Resource Selection Studies for
Categorical Resource Variables. Journal of Wildlife Management 70:337–346.
Amat, J.A. 1982. The nesting biology of ducks in the Marismas of the Guadalquivir, SouthWestern Spain. Wildfowl 33:94-104.
Amat, J.A., 1984. Ecological segregation between Red-crested Pochard Netta rufina and
Pochard Aythya ferina in a fluctuating environment. Ardea 72:229-233.
Anderson, D. R., W. A. Link, D. H. Johnson, and K. P. Burnham. 2001. Suggestions for
presenting the results of data analyses. Journal of Wildlife Management 65:373-378.
Anderson, M. G., and R. D. Titman. 1992. Spacing patterns. Pages 251-289 in B. D. J. Batt,
A. D. Afton, M. G. Anderson, C. D. Ankney, D. H. Johnson, J. A. Kadlec, and G. L. Krapu,
editors. Ecology and Management of Breeding Waterfowl. University of Minnesota Press,
Minneapolis & London.
Atiénzar, F., J. Larrosa, J.L. Echevarrías and E. Barba. 2005. Area and shoreline complexity
affect white-headed duck distribution and abundance in southeastern Spanish wetlands. Poster
presented at the 5th conference of the European Ornithologists’ Union, Strabourg, France
Aznar, J. C., A. Dervieux, and P. Grillas. 2003. Association between aquatic vegetation and
landscape indicators of human pressure. Wetlands 23:149-160.
Barbraud C., and K. Delord. 2006. Population census of blue petrels Halobaena caerulea at
Mayes Island, Iles Kerguelen. Antarctic Science 18:199-204.
Bibby, C. J., and S. T. Buckland. 1987. Bias of bird census results due to detectability varying
with habitat. Acta OEcologica: OEcol. Gener 8:103-112.
Bigas, D., and F. Vidal. 2004. Cens de cabussons, ànecs I fotges nidificants.
Soldo,
Informatiu del Parc Natural del Delta de l’Ebre 22:4.
Block W. M., and L. A. Brennan. 1993. The habitat concept in ornithology: Theory and
applications. Current Ornithology 11:35–91.
Boutin, J. 1994. Nette rousse. Pages 142-143 in D. Yeatman-Berthelot and G. Jarry, editors.
Nouvel Atlas des Oiseaux nicheurs de France 1985-1989. Société Ornithologique de France,
Paris.
43
Brotons, L., W. Thuiller, M. B. Araújo., and A. H. Hirzel. 2004. Presence-absence vs
presence-only modelling methods for predicting bird habitat suitability. Ecography 27:165172.
Brown, J. 1995. Macroecology. Chicago, University of Chicago Press 269 p.
Cam, E., J. R. Sauer, J. D. Nichols, J. E. Hines, and C. H. Flather. 2000. Geographic analysis
of species richness and community attributes of forest birds from survey data in the MidAtlantic integrated assessment region. Environmental Monitoring and Assessment 63:81-94.
Cam, E., J. D. Nichols, J. E. Hines, J. R. Sauer, R. Alpizar-Jara, and C. H. Flather. 2002.
Disentangling sampling and ecological explanations underlying species-area relationships.
Ecology 83:1118-1130.
Conover, M. R. and G. S. Kania. 1994. Impact of interspecific aggression and herbivory by
mute swans on native waterfowl and aquatic vegetation in new England. - The Auk
111:744-748.
Décarie, R., F. Morneau, D. Lambert, S. Carrière, and J.-P. L. Savard. 1995. Habitat use by
brood-rearing waterfowl in subarctic Québec. Arctic 48: 383–390.
Defos du Rau, P., C. Barbraud, and J.Y. Mondain-Monval. 2003. Estimating breeding
population size of the red-crested pochard (Netta rufina) in the Camargue (southern France)
taking into account detection probability: implications for conservation. Animal Conservation
6:379-385.
Defos du Rau, P., C. Barbraud, and J.Y. Mondain-Monval. 2005; Incorporating uncertainty
into analyses of Red-crested Pochard habitat selection. Biological Conservation. 125:355-367.
DuBowy, P. J. 1991. Seasonality and interspecific competition in waterfowl guilds: patterns
and processes – a reply to Bethke. Ecology 72:1159-1161.
Engler, R., A. Guisan, and L. Rechsteiner. 2004. An improved approach for predicting the
distribution of rare and endangered species from occurrence and pseudo-absence data.
Journal of Applied Ecology 41:263-274.
Fawcett, T. 2006. An introduction to ROC analysis. Pattern Recognition Letters 27:861 – 874.
Frazier, S. (editor) 2002. A Directory of Wetlands of International Importance. Wetlands
International and Ramsar Convention Bureau. Compact Disc.
Freckleton, R. P., J. A Gill, D. Noble, and A. R. Watkinson. 2005. Large-scale population
dynamics, abundance–occupancy relationships and the scaling from local to regional
population size. Journal of Animal Ecology 74:353–364.
44
Gay, L., P. Defos du Rau, J. Y. Mondain-Monval, and P.A. Crochet. 2004. Phylogeography of
a game species: the red-crested pochard (Netta rufina) and consequences for its management.
Molecular Ecology 13:1035-1045.
Gilissen, N, L. Haanstra, S. Delany, G. Boere, and W. Hagemeijer. 2002. Numbers and
distribution of wintering waterbirds in the Western Palearctic and Southwest Asia in 1997,
1998 and 1999. Results from the International Waterbird Census. Wetlands International
Global Series No. 11. Wageningen, The Netherlands.
Gimona, A., and M. J. Brewer. 2006. Local environmental effects and spatial effects in
macroecological studies using mapped abundance classes: the case of the rook Corvus
frugilegus in Scotland. Journal of Animal Ecology 75:1140-1146.
Green, A.J. 1998. Habitat selection by the Marbled Teal Marmaronetta angustirostris,
Ferruginous Duck Aythya nyroca and other ducks in the Göksu Delta, Turkey in late summer.
Revue d’Ecologie, Terre et Vie 53:225-243.
Green, A.J., and M. El Hamzaoui. 2000. Diurnal behaviour and habitat use of nonbreeding
Marbled Teal Marmaronetta angustirostris. Canadian Journal of Zoology 78:2112-2118.
Grillas, P. 1992. Les communautés de macrophytes submergées des marais temporaires oligohalins de Camargue, étude expérimentale des causes de la distribution des espèces. PhD
thesis, Université Rennes I, Rennes.
Gu, W., and R.K Swihart, 2003. Absent or undetected? Effects of non-detection of species
occurrence on wildlife–habitat models. Biological Conservation 116:195-203.
Guisan, A., and W. Thuiller. 2005. Predicting species distributions: offering more than simple
habitat models Ecology Letters 8:993-1009.
Guisan, A., O. Broennimann, R. Engler, N. G. Yoccoz, M. Vust, N. E Zimmermann, and A.
Lehmann. 2006a. Using niche-based models to improve the sampling of rare specie.
Conservation Biology 20:501-511.
Guisan, A., A. Lehmann, S. Ferrier, R Aspinall, R. Overton, M. Austin, and T. Hastie. 2006b
Making better biogeographic predictions of species distribution. Journal of Applied Ecology
43:386–392.
Guo Q, M. L. Taper, M. Schoenberger, and J. Brandl. 2005. Spatial-temporal population
dynamics across species’ range:. from center to margin. Oikos 108:47–57.
Hanski, I., and M. Gyllenberg. 1997. Uniting two general patterns in the distribution of
species. Science 275:397-400.
Hirzel, A.H.and A. Guisan. 2002. Which is the optimal sampling strategy for habitat
suitability modelling. Ecological Modelling 157: 331-341.
45
Horn, D.J., M. L. Phillips, R. R. Koford, W. R. Clark, M. A. Sovada, and R. J. Greenwood.
2005. Landscape composition, patch size, and distance to edges: interactions affecting duck
reproductive success. Ecological Applications 15:1367-1376.
Johnson, D. H., A. B. Sargeant, and R. J. Greenwood. 1989. Importance of individual species
of predators on nesting success of ducks in the Canadian Prairie Potholes Region. Canadian
Journal of Zoology 67:291-297.
Jones, J. 2001. Habitat selection studies in avian ecology: a critical review. Auk 118:557-562.
Joyner, D. E. 1980. Influence of invertebrates on pond selection by ducks in Ontario. Journal
of Wildlife Management 44:700-705.
Kaminski, R.M., and M.W. Weller. 1992. Breeding Habitats of Nearctic Waterfowl. In
Ecology and Management of Breeding Waterfowl, ed. B. D. J. Batt, A. D. Afton, M. G.
Anderson, C. D. Ankney, D. H. Johnson, J. A. Kadlec, and G. L. Krapu, pp. 251-289.
University of Minnesota Press, Minneapolis & London.
Kendall, W. L., J. D. Nichols, and J. E. Hines. 1997. Estimating Temporary Emigration and
Breeding Proportions using Capture-Recapture data with Pollock's Robust-design. Ecology
78:563-578.
Kohler,
A.
1978.
Methoden
der
Kartierung
von
Flora
und
Vegetation
von
SuBwasserbiotopen. Landschaft & Stadt 10:73-85.
Krebs, C. J. 1972. Ecology. The Experimental Analysis of Distribution and Abundance.
Harper and Row, New York. 694 pp.
Liu, C., P. M. Berry, T. P. Dawson, and R.G. Pearson. 2005. Selecting thresholds of
occurrence in the prediction of species distributions. Ecography 28:385-393.
Llorente, G. A. and X. Ruiz.1985. Datos sobre la reproducción del Pato Colorado Netta rufina
(Pallas 1773) en el Delta del Ebro. Misc. Zool. 9 : 315-323.
Mack, G. D., and L. D. Flake. 1980. Habitat relationships of waterfowl broods on South
Dakota stock ponds. Journal of Wildlife Management 44:695-700.
MacKenzie, D. I., J. D. Nichols, G. B. Lachman, S. Droege, J. A. Royle, and C. A.
Langtimm. 2002. Estimating site occupancy rates when detection probabilities are less than
one. Ecology 83:2248-2255.
MacKenzie, D. I., and L. L. Bailey. 2004. Assessing the fit of site occupancy models. Journal
of Agricultural, Biological and Ecological Statistics 9:300-318.
MacKenzie, D. I., L. L. Bailey, and J. D. Nichols. 2004. Investigating species co-occurrence
patterns when species are detected imperfectly. Journal of Animal Ecology 73:546-555.
46
MacKenzie, D. I., and J. A. Royle. 2005. Designing efficient occupancy studies: general
advice and tips on allocation of survey effort. Journal of Applied Ecology 42:1105-1114.
MacKenzie, D. I. 2006. Modeling the probability of resource use: the effect of, and dealing
with, detecting a species imperfectly. Journal of Wildlife Management 70:367–374.
MacKenzie, D. I., J. D. Nichols, J. A. Royle, K. H. Pollock, J. E. Hines, and L. L. Bailey.
2006. Occupancy estimation and modeling: inferring patterns and dynamics of species
occurrence Elsevier, San Diego, USA 324p.
Manly, B. F. J., L. L. McDonald, D. L. Thomas, T. L. McDonald, and W. P. Erickson. 2002.
Resource selection by animals, 2nd edn. Kluwer Academic Publishers, Dordrecht, the
Netherlands.
Mathevet, R., and A. Tamisier. 2002. Creation of a nature reserve, its effects on hunting
management and waterfowl distribution in the Camargue (Southern France). Biodiversity and
Conservation 11:509-519.
Moilanen A. 2002. Implications of empirical data quality for metapopulation model parameter
estimation and application. Oikos. 96:516–530.
Munteanu, D. 1996. Changes and trends in Anatidae populations in Romania after the second
world war. In : Proceedings of the Anatidae 2000 Conference, Strasbourg, France, 5-9
December 1994, M. Birkan, J. van Vessem, P. Havet, J. Madsen, B. Trolliet & M. Moser, eds.
Gibier Faune Sauvage, Game Wildl. 13:973-974.
Nichols, J. D., J. E. Hines, J. R. Sauer, F. W. Fallon, J. E. Fallon, and P. J. Heglund. 2000. A
Double-observer Approach for Estimating Detection Probability and Abundance from Point
Counts. Auk 117:393-408.
O’Connell, M. 1999. Research : getting the right balance for wildfowl and wetlands.
Wildfowl 50:1-9.
ONC. 1982. Critères de détermination de l’âge du Canard colvert de la naissance à 9
semaines. Fiche technique n°10, Bulletin mensuel 63.
Opermanis, O. 2001. Marsh Harrier Circus aeroginosus predation on artificial duck nests: a
field experiment. Ornis Fennica 78:198-203.
Pannatier, Y. 1996. VARIOWIN: Software for spatial data analysis in 2D. Berlin: Springer.
Pollock , K. H., J. D. Nichols, T. R. Simons, G. L. Farnsworth, L. L, Bailey, and J.R; Sauer.
2002. The design of large scale wildlife monitoring studies. Environmetrics 13, 105-119.
Poulin B., Lefebvre G., Mauchamp A. 2002 Habitat requirements of passerines and reedbed
management in southern France. Biol Conserv 107 p315-325.
47
Pulliam, H. R., and B. J. Danielson. 1991. Sources, sinks, and habitat selection: a landscape
perspective on population dynamics. Am. Nat. 137:S50-S66.
Reynolds, J. D. 2003. Life histories and extinction risk. In: Macroecology : Concepts and
Consequences (eds. T.M. Blackburn & K.J. Gaston). Blackwell Publishing, Oxford.
Royle, J. A. 2004. N-Mixture models for estimating population size from spatially replicated
counts. Biometrics 60:108-115.
Royle, J. A., and J. D. Nichols. 2003. Estimating abundance from repeated presence-absence
data or point counts. Ecology 84:777-790.
Ruth, J. M., D. R. Petit, J. R. Sauer, M. D. Samuel, F. A. Johnson, M. D. Fornwall, C. E.
Korschgen, and J. P. Bennett. 2003. Science for avian conservation: priorities for the new
millennium. Auk 120:204-211.
Segurado, P., M. B. Araujo, W. E. Kunin. 2006. Consequences of spatial autocorrelation for
niche-based models Journal of Applied Ecology 43:433-444.
Snow, D. W., C. M. Perrins. 1998. The Birds of the Western Palearctic, Vol. 1. Oxford
University Press, Oxford.
Stockwell, D. R. B. and A.T. Peterson. 2002. Effects of sample size on accuracy of species
distribution models. Ecological Modelling, 148:1 –13.
Tamisier, A., and P. Grillas. 1994. A review of habitat changes in the Camargue : an
assessment of the effects of the loss of biological diversity on the wintering waterfowl
community. Biological Conservation 70:39-47.
Thompson, W. L. 2002. Towards reliable bird surveys: accounting for individuals present but
not detected. Auk 119:18-25.
Tucker, G. M. and M. F. Heath,. 1994 Birds in Europe, their conservation status. BirdLife
Conservation Series n° 3. Cambridge, United Kingdom.
Vojta C. D. 2005. Old dog, new tricks: innovations with presence–absence information.
Journal of Wildlife Management. 69:845–848.
Watkinson, A. R., J. A. Gill, and R. P. Freckleton. 2003. Macroecology and microecology:
linking large-scale patterns of abundance to population processes. In: Macroecology :
Concepts and Consequences (eds. T. M. Blackburn and K. J. Gaston). Blackwell Publishing,
Oxford.
Weller, M. W. 1988. Issues and approaches in assessing cumulative impacts on waterbird
habitat in wetlands. Environmental Management 12:695-701.
White, G. C., and K. P. Burnham.1999. Program MARK: survival estimation from
populations of marked animals. Bird Study 46 (supplement):120-139.
48
Wiegand T., E. Revilla, and K. A. Moloney. 2005. Effects of habitat loss and fragmentation
on population dynamics. Conservation Biology 19:108-121.
Wilson, J. B. 1999. Guilds, functional types and ecological groups. Oikos 86:507-522.
Appendix
wetland structure & habitats
food
water managnt. predators competitors
space response
abbreviation
brood presence or absence
potentially affecting detection log-transformed
0 or 1
brood abundance
Danube
Ebre
competitors diversity
density of coot broods
density of mallard broods
density of gadwall broods
density of swan broods
density of grebe broods
frequency of gulls
frequency of harriers
frequency of magpies
frequency of crows
observed maximum water depth
spring water level before May
spring water level in May
spring water level in June
spring water level in July
SD of water level from March to July
frequency of Myriophyllum sp.
frequency of Potamogeton pectinatus
macrophyte genus richness
mean distance to 5 closest lakes
distance to closest lake
water surface area
reedbed surface area
Arthrocnemum beds surface area
Scirpus beds surface area
total vegetation surface area
number of islets
area of islets
% free water area (detection only)
shoreline index
value
# of broods
Danube
Ebro
comp
coot
mallard
swan
grebe
gull
harrier
magpie
crow
myrio
pot
gen
mdist
water
reed
veg
islet
%water
shore
0 or 1
0 or 1
# of species
broods/ha
broods/ha
broods/ha
broods/ha
broods/ha
mean # of indiv/30min
mean # of indiv/30min
mean # of indiv/30min
mean # of indiv/30min
cm.
cm.
cm.
cm.
cm.
cm.
%
%
# of genus
m.
m.
ha.
ha.
ha.
ha.
ha.
# of islets
ha.
perimeter/2(π.area)½.
yes+1
X
X
X
X
X
X
X
X
X
X
X
X
X
49
Chapter 5
Molecular Ecology (2004) 13, 1035– 1045
doi: 10.1111/j.1365-294X.2004.02117.x
Phylogeography of a game species: the red-crested pochard
(Netta rufina) and consequences for its management
Blackwell Publishing, Ltd.
L . G A Y ,* P . D E F O S D U R A U ,† J . - Y . M O N D A I N - M O N V A L † and P . - A . C R O C H E T ‡
*CEFE-CNRS, 1919 route de Mende, F-34293 Montpellier cedex 5, France, †Office National de la Chasse et de la Faune Sauvage,
CNERA Avifaune Migratrice, Le Sambuc, 13200 Arles, France, ‡Laboratoire de Biogéographie et Ecologie des Vertébrés, EPHE, box
94, Université Montpellier II, 34095 Montpellier cedex, France
Abstract
Western European populations of red-crested pochard (Netta rufina) are characterized by
low size and high fragmentation, which accentuate their sensitivity to hunting. Uncertainties regarding the demographic trends of these populations highlight the need for pertinent
hunting regulations. This requires identification of the limits of the populations under
exploitation, i.e. delimiting a management unit. We used the left domain of the mitochondrial control region and seven nuclear loci (four microsatellites and three introns) to assess
the level of genetic structure and demographic independence between the fragmented
Western European and the large Central Asian populations. The second objective was to
investigate the colonization history of the Western European populations. This study
demonstrated that the Western European populations of red-crested pochard constitute a
separate demographic conservation unit relative to the Asian population as a result of very
low female dispersal (mitochondrial DNA: ΦST = 0.152). A morphometric analysis further
suggested that Central Asian and Western European specimens of both sexes originate
from different pools of individuals. Male dispersal seems higher than female dispersal, as
suggested by the lack of clear genetic structure for the nuclear markers at this continental
scale. Genetic data, in conjunction with historical demographic data, indicate that the
current Western European populations probably originate from a recent colonization from
Central Asia. As numbers of red-crested pochards in Western Europe cannot be efficiently
supplemented by immigration from the larger Asian populations, a management plan
regulating the harvest in Western Europe is required.
Keywords: birds, control region, management unit, microsatellites, nuclear introns, phylogeography
Received 16 October 2003; revision received 8 December 2003; accepted 8 December 2003
Introduction
Securing the local persistence of a species requires testing for demographic isolation, which identifies sets of
populations whose dynamics are not significantly affected
by influences from adjacent populations. These units are
often called management units (MUs; Moritz 1994; Fraser
& Bernatchez 2001). This concept is crucial in the case of
exploited populations for which we have to identify those
sets of populations that will be affected by human harvest
and thus evaluate the impact of exploitation (e.g. Ruzzante
et al. 2000; Koljonen 2001). The degree of demographic
isolation is determined by the occurrence of dispersal
Correspondence: Laurène Gay. E-mail: [email protected]
© 2004 Blackwell Publishing Ltd
among populations. Genetic methods can provide an
indirect estimate of dispersal, even if an absence of genetic
differentiation does not necessarily imply a high level
of dispersal (for example in the case of recently isolated
populations, see Whitlock & McCauley 1999). Therefore,
Moritz (1994) advocated recognizing as MUs ‘populations
with significant divergence of allele frequencies at nuclear
or mitochondrial loci’.
The red-crested pochard (Netta rufina, Aves, Anseriforme) is a diving duck with a vast Palearctic range extending from Western Europe to Central Asia (Scott & Rose
1996) (Fig. 1). It is one of the least abundant of the Western
Palearctic waterfowl species (Anatidae) (Scott & Rose 1996;
Dehorter & Rocamora 1999). Hunting is nevertheless allowed
in Spain, Portugal and France, resulting in an annual
1036 L . G A Y E T A L .
Fig. 1 Distribution of the red-crested pochard, Netta rufina, in
Eurasia (after Scott & Rose 1996). Shaded areas indicates the
breeding distribution. Filled circles indicate sampling sites: 1,
Donana; 2, Ebrodelta; 3, Camargue; 4, Dombes; 5, Constance Lake;
6, Volga delta; 7, Kazakhstan. Shaded lines indicate population
boundaries (broken line: uncertain) (outlining the breeding, wintering and migration range).
harvest estimated at 8000 birds (Shedden 1986). The uncertainties regarding the evolution of the population size and
the fragmentation of the distribution range in Western
Europe highlight the need for an international action plan,
including proposals for sustainable hunting (Defos du
Rau 2002). This requires that the impact of hunting be estimated, which necessitates identification of the limits of the
exploited populations.
Based on census data and ringing recoveries, three
distinct population groups have been postulated (Fig. 1)
(Monval & Pirot 1989; Scott & Rose 1996). The first group
(hereafter described as ‘Western European populations’)
occupies the western Mediterranean region and western
and central Europe, with an estimated wintering population size of 50 000 birds (Delany & Scott 2002). The second
group (‘Eastern European populations’) inhabits the area
of the Black Sea and eastern Mediterranean basins, with a
wintering population size of 20 000 – 43 500 birds (Delany
& Scott 2002). The third population group (‘Central Asian
populations’) occupies the steppe areas from the Caspian
Sea to Mongolia and western China and is estimated at
250 000 individuals (Delany & Scott 2002). The Western
European breeding range is highly fragmented and
comprises apparently isolated small-sized subpopulations
(Hagemeijer & Blair 1997), while the distribution in the
eastern part of the global range appears more continuous.
The Western European populations are sedentary or
short-distance migrants (within the Mediterranean basin),
while the eastern populations migrate longer distances.
Wintering areas for the three populations are clearly
separated from each other (Cramps & Simons 1977;
Saez-Royuela 1997).
As for other birds presenting the same type of distribution
(dense populations in Asia and fragmented subpopulations
in Western Europe; Cramps & Simons 1977), the distribution of the red-crested pochard may result from: (i) recent
colonization of Western Europe by individuals from further
east, or (ii) fragmentation of a previously continuous distribution range. Historical data suggest that the species
colonized Western Europe in the late nineteenth century
(Mayaud 1966; Cramps & Simons 1977; Hagemeijer &
Blair 1997). The genetic study of Western European and
Central Asian populations would allow us to evaluate this
hypothesis.
The first objective of this work was to investigate the
amount of gene flow between Western European and
Central Asian populations of red-crested pochard. A
significantly reduced gene flow would indicate a currently
essentially independent demographic functioning of the
Western European populations relative to the much larger
Asian populations. The second objective was to understand
the biogeographical history of the red-crested pochard.
Using genetic data, it is not possible to distinguish ancient
fragmentation from colonization. Thus the analysis is
restricted to the case of recent events. The observed patterns of genetic diversity were compared with the pattern
expected for two possible biogeographical scenarios: recent
colonization of Europe or recent range fragmentation.
Recent colonization should result in a loss of genetic
diversity along the colonization axis and marked genetic
structure (Austerlitz et al. 1997). Positive exponential growth
rate should also be detected. Under the recent rangefragmentation hypothesis, little difference would be expected
in the diversity pattern in both populations and imprints of
a negative exponential growth rate.
The genetic analyses included two classes of molecular
markers, the left domain of the mitochondrial DNA (mtDNA)
control region (hypervariable control region I, CRI) and seven
nuclear markers with biparental inheritance (four microsatellites, three nuclear introns). Two populations were
included: Western Europe and Central Asia. Genetic
analyses were supplemented by a morphometric analysis
because morphology sometimes diverges quicker than
neutral markers as a result of local selection (see Podolsky
& Holtsford 1995; Nice & Shapiro 1999; Chan & Arcese
2003 for bird, insect and plant examples respectively).
Materials and methods
Morphological analyses
The 100 specimens (Western Europe n = 20; Central Europe
n = 13; Central Asia n = 67) held at the Muséum National
d’Histoire Naturelle, Paris, France and the British Museum
of Natural History, Tring, UK were measured by one of the
authors (P.D.D.R.). Each specimen was referenced according
to the locality of collection. The following variables were
measured with a calliper to the nearest mm: wing length,
© 2004 Blackwell Publishing Ltd, Molecular Ecology, 13, 1035–1045
C O N S E R V A T I O N G E N E T I C S O F A D U C K S P E C I E S 1037
Table 1 Sample localities (population and subpopulation) and sample size for mitochondrial sequences, microsatellites (APH11, APH10,
APH1 and SFI4) and nuclear introns (GAPD, OD and RP40)
Microsatellites
Nuclear introns
Population
Sub-population
mtDNA
APH11
APH10
APH1
SFI4
GAPD
OD
RP40
Western Europe
Camargue
Dombes
Ebro delta
Doñana
Constance
Total WE
10
9
2
10
2
33
24
30
7
21
3
85
23
30
4
22
3
82
23
30
5
21
3
82
17
24
0
0
2
43
13
27
6
19
2
67
5
25
5
2
2
39
1
28
6
0
2
37
Central Asia
Kazakhstan
Volga delta
Total CA
27
4
31
58
3
61
54
3
57
56
3
59
54
3
57
57
4
61
53
4
57
43
4
47
tarsus length, culmen length, bill width at base of bill, bill
depth at base of bill, length between bill base and inner
edge of nostril, length between bill tip and outer edge of
nostril. The number of lamellae on the bill was also noted
for each specimen. All characters were standardized (to zero
mean and unit variance) prior to principal components
analysis (PCA). Both sexes were pooled for the PCA.
Differences between populations or sexes (a potentially
confounding effect) in the multivariate morphometric
space were investigated by mean of successive analyses of
variance (anova) on the principal components (PCs).
Samples for the genetic analyses and DNA extraction
Samples for the genetic analyses were either muscles in
alcohol or dried legs (kept at room temperature) collected
from red-crested pochards shot by hunters in different
localities of the distribution range (Table 1, Fig. 1). All
these localities are breeding sites but most samples might
include migrating or wintering birds. Only samples from
Doñana most likely represent breeding birds. Since the
migration pathways and wintering areas seem to be distinct
for Western European and Central Asian populations (see
above), birds caught in Western Europe or Central Asia
would be expected to belong to their respective population.
Total genomic DNA was extracted using mini column
extraction kits (DNeasy Tissue Kit, Qiagen) following the
manufacturer’s instructions.
Mitochondrial DNA sequencing
A 450 base pair (bp) fragment of the first (left) domain of
the mtDNA control region (CR1) was amplified by the
polymerase chain reaction (PCR) with the following
primers: NDLF, 5′-AAA-TAA-GTC-ATT-ATT-CCT-GC-3′
(3′ end at position 253 in the Anas platyrhynchos sequence,
GenBank accession number L22477) and NDLIR, 5′-AACCAG-AGG-CGC-AAA-AAT-GTG-3′ (3′ end at position 821
© 2004 Blackwell Publishing Ltd, Molecular Ecology, 13, 1035–1045
in the same sequence). Primers were designed by aligning
the sequences of waterfowl species found in GenBank. PCR
amplification was performed in a 50-µL reaction volume
containing 2 µL DNA solution (variable concentration),
5 µL 10× buffer (Tris–HCl 100 mm + KCl 500 mm), 6.25 µm
MgCl2, 3.33 µm dNTP, 1.67 µm of each primer and 0.2
units Taq DNA polymerase (Eurogentec). The annealing temperature was 52 °C and annealing duration was 45 s.
Sequencing reactions were conducted with ABI dye terminator chemistry (Applied Biosystem) following the standard
ABI cycle sequencing protocol and were electrophoresed
on an ABI Prism 310 Genetic Analyser following recommended procedures, using NDLIR as the sequencing
primer.
Microsatellites
Between 100 and 146 individuals were genotyped at four
microsatellite loci (see Table 1 for the sample size of each
marker). Primer sequences were found in the literature
(APH11, APH10 and APH1 in the Peking duck Anas
platyrhynchos: Maak et al. 2000; SFI4 in the spectacled eider
Somateria fisheri: Fields & Scribner 1997). PCRs were performed in a total volume of 12 µL containing 2 µL DNA
solution (concentration variable), 1.2 µL 10× buffer (Tris–
HCl 100 mm + KCl 500 mm), 2.08 µm MgCl2, 0.5 µm dNTP,
0.83 µm of forward primer, 0.17 µm of γ[32P]ATP-labelled
reverse primer and 0.1 units Taq DNA polymerase. The
annealing temperature was 52 °C for all marker except APH1
(for which the annealing temperature was 54 °C) and
annealing duration was 30 s. PCR products were resolved
by electrophoresis on 5% denaturing polyacrylamide gels,
exposed for 12–72 h.
Nuclear introns
Three nuclear introns were amplified on 130 individuals
for GAPD, 96 for OD and 84 for RP40 (Table 1) using PCR
1038 L . G A Y E T A L .
(same protocol as for the control region). Primer sequences
were found in the literature (Friesen et al. 1997, 1999):
GAPD (glyceraldehyde-3-phosphate dehydrogenase gene),
RP40 (ribosomal protein 40 gene) and OD (ornithine decarboxylase gene). The annealing temperature was 56 °C for
45 s. Sequence polymorphism was revealed by single-strand
conformation polymorphism (SSCP; Lessa 1992; Palumbi
& Baker 1994) performed using an ABI Prism 310 Genetic
analyser, following the manufacturer’s instructions (Applied
Biosystems).
Data analysis
Within-population analysis. The number of haplotypes (na),
number of polymorphic sites (S), haplotype diversity
(H ± SD) and mean number of pairwise differences (π ± SD)
were estimated on mtDNA data using arlequin version
2.0 (Schneider et al. 2000). The hypothesis of selective
neutrality of the control region fragment sequenced was
tested using the D* and F* tests (Tajima 1989a; Fu 1997)
with the program arlequin. Maximum likelihood estimates
of the exponential growth rate ( g, scaled to the per sequence
mutation rate) were obtained using a coalescence-based
method (fluctuate, Kuhner et al. 1998). For nuclear data,
the mean number of alleles (na), Nei’s unbiased estimates
of expected heterozygosity (HE) and observed heterozygosity
(HO) were calculated using genetix version 4.02 (Belkhir
et al. 2001). FIS was calculated for each locus separately
and for all loci together and significance was tested by
permutation of individuals among populations (1000
permutations). genetix was used to test for linkage
disequilibrium between pairs of loci in each population
(1000 permutations). Levels of significance were adjusted
using sequential Bonferroni corrections (Rice 1989).
Among-population analyses. modeltest version 3.0 (Posada
& Crandall 1998) was used to determine the appropriate
model of substitution for the control region sequences.
The selected model was Tamura–Nei with a gamma distribution of the substitution rates and a proportion of invariable sites. mega version 2.1 (Kumar et al. 2001) was then used
to generate a phylogenetic tree of the mtDNA haplotypes
by the neighbour-joining method based on a Tamura–Nei
distance matrix with the shape parameter of the gamma
distribution determined by modeltest.
The analysis of population structure was based on variance partitioning. Mitochondrial DNA data were analysed
by analysis of molecular variance (amova, Excoffier et al.
1992) with arlequin using a Kimura two-parameter distance for estimating ΦST values. For nuclear markers, the
estimator θ of FST (Weir & Cockerham 1984) was estimated
using genetix. RST was not used for microsatellites because
both sample size and number of loci were too small for RST
to give a better estimation of population structure than
FST (Gaggiotti et al. 1999). The significance of ΦST and FST
estimates was tested by permutations of individuals among
populations (1000 permutations). FST and ΦST estimates
were used to estimate gene flow between the two populations, assuming a finite-island model, with the following
equation at migration/drift equilibrium: FST = 1/(1 + 4Nmα)
and Φ ST = 1/(1 + N fm fα), Nm being the number of
migrants entering a population per generation, Nfmf is
the number of female migrants and α = r/(r − 1) with r
being the number of populations. In this study, r = 2 and
α = 2.
Estimates of gene flow based on FST and its analogues
rely on the island model assumptions requiring equal
population sizes and symmetric migration rates (Rousset
2001). In the case of the red-crested pochard, both assumptions probably do not hold (see Introduction). Therefore,
the gene flow was also estimated using a maximumlikelihood method based on coalescence (Beerli & Felsenstein
1999) implemented in migrate Version 1.5.1. (Beerli 2002),
using mtDNA sequences, microsatellites and nuclear
introns separately, because these markers follow different
mutational models (mtDNA sequences, Felsenstein’s mutation model; nuclear introns, infinite allele model; microsatellites, stepwise-mutation model). For all analyses, the
default settings of migrate were used except that the number of short and long Markov chains and the number of
trees sampled were increased in some runs (20 short chains
with 5000 recorded genealogies and four long chains with
50 000 genealogies). Because convergence problems are
common with Markov chain estimations, the convergence
of the program was tested by using the options ‘heating’
and ‘replication’ and by reiterating the estimations with
different starting values (for example the estimations from
previous runs).
Results
Morphometric analysis
The two first axes of the PCA explain 50% of the total
variance. The first axis of the PCA (PC1) indicates an
obvious morphological differentiation between individuals
from Western Europe and Central Asia, with the Eastern
European population having an intermediate position
(Fig. 2). This is confirmed by the analysis of variance: there
is a strong population effect on PC1 (one-way anova;
F2,97 = 59.28; P < 10−6) but no effect of sex (one-way anova;
F2,97 = 0.37; P = 0.54). This first axis is mainly a size axis,
with most biometric variables having a large contribution,
except the number of lamellae (see Table 2 for the contribution of each variable to the PCs and percentages of
explained variance for PC1 and PC2), Western and Central
European specimens having, on average, smaller biometrics
than individuals from Central Asia.
© 2004 Blackwell Publishing Ltd, Molecular Ecology, 13, 1035–1045
C O N S E R V A T I O N G E N E T I C S O F A D U C K S P E C I E S 1039
Table 2 Eigenvalues, per cent of explained variance and contribution of each variable to the principal components
Eigenvalue
Per cent of explained
variance
Wing
Tarsus
Culmen
Width at bill base
Depth at bill base
Bill base to nostril
Bill tip to nostril
No. of lamellae
Fig. 2 Bivariate plot of PC1 and PC2 scores generated by a PCA on
all specimens (sexes pooled) using the eight morphological variables
(see Materials and methods). , Western European population; +,
Eastern European population; , Central Asian population.
Population genetic diversity
The control region was sequenced in 64 individuals ( Table 1;
GenBank accession numbers AY465764 to AY465827). The
450-bp analysed fragment starts around position 367 of the
Anas plathyrhyncos sequence (GenBank accession number
L22477) and ends around position 821. This segment
proved highly variable, with one haplotype per individual
in Central Asia (31 individuals) and 13 haplotypes in
Western Europe (33 individuals). The sequence exhibited
74 polymorphic sites in Central Asia and 41 in Western
Locus
Allele
Central
Asia
APH11
1
2
3
4
0.2869
0.6557
0.0410
0.0164
APH10
1
2
3
4
5
SFI4
1
2
3
4
5
6
7
6
Western
Europe
Locus
Allele
0.2118
0.6706
0.1176
0.0000
APH1
1
2
3
4
0.6949
0.2288
0.0763
0.0000
0.6585
0.2256
0.1098
0.0061
0.7018
0.1228
0.0439
0.0000
0.0263
0.7195
0.2012
0.0000
0.0183
0.0488
GAPD
1
2
3
4
5
0.2951
0.1475
0.1639
0.0902
0.2213
0.2239
0.2090
0.0597
0.0672
0.2985
0.2105
0.5000
0.0965
0.0702
0.0088
0.0526
0.0614
0.1053
0.2558
0.4651
0.0233
0.0698
0.0000
0.0465
0.1395
0.0122
6
1
2
3
1
2
3
0.0820
0.3830
0.6064
0.0106
0.7544
0.2281
0.0175
0.1418
0.4595
0.5270
0.0135
0.6667
0.3205
0.0128
OD
© 2004 Blackwell Publishing Ltd, Molecular Ecology, 13, 1035–1045
PC2
2.83
35.37
1.14
14.29
0.40
0.64
0.65
0.62
0.78
0.71
0.56
0.07
0.55
− 0.31
− 0.35
− 0.19
− 0.05
0.17
0.35
0.65
Europe. Haplotype diversity was 1.000 ± 0.008 (± SD)
and 0.723 ± 0.080 for Central Asia and Western Europe,
respectively. The mean number of pairwise differences
ranged from 0.027 ± 0.0125 in Central Asia to 0.012 ± 0.006
in Western Europe. Mitochondrial DNA diversity was
thus significantly lower in Western Europe than in Central
Asia (Mann–Whitney test, retaining only independent pairwise comparisons: NCentral Asia = 15, NWestern Europe = 16; U =
54.5; Z = 2.589; P = 0.009).
There were between three and seven alleles per locus
for the four microsatellites and the three nuclear introns
(Tables 3 and 4). Unlike the sequence data, there was no
significant difference of diversity between the two populations at the nuclear loci. The mean number of alleles per
locus was 4.43 in Central Asia and 4.28 in Western Europe.
Expected heterozygosities were similarly high (0.539 ± 0.148
and 0.554 ± 0.133 for Central Asia and Western Europe,
Central
Asia
RP40
PC1
Western
Europe
Table 3 Allele frequencies for seven nuclear
loci (microsatellites: APH11, APH10, APH1
and SFI4; introns: GAPD, RP40 and OD)
for Central Asian and Western European
populations
1040 L . G A Y E T A L .
Table 4 Diversity at seven nuclear loci (microsatellites: APH11,
APH10, APH1 and SFI4; introns: GAPD, RP40 and OD) for Central
Asian and Western European populations
Locus
name
Population
APH11 Central Asia
Western Europe
APH10 Central Asia
Western Europe
APH1 Central Asia
Western Europe
SFI4
Central Asia
Western Europe
GAPD Central Asia
Western Europe
RP40
Central Asia
Western Europe
OD
Central Asia
Western Europe
n
nall HE
HO
61
85
57
82
59
82
57
43
61
67
47
37
57
39
4
3
5
5
3
4
7
6
6
6
3
3
3
3
0.475
0.029 0.317
0.541 − 0.095 0.839
0.421
0.129 0.052
0.463 − 0.050 0.699
0.373
0.196 0.020
0.329
0.351 0.000
0.691 − 0.067 0.789
0.698
0.002 0.410
0.770
0.046 0.170
0.731
0.080 0.075
0.425
0.134 0.127
0.378
0.272 0.031
0.316
0.174 0.056
0.410
0.106 0.192
0.486
0.492
0.479
0.439
0.459
0.503
0.685
0.737
0.800
0.789
0.485
0.511
0.379
0.453
FIS
P
n is the sample size; nall the number of alleles per locus; HE the
gene diversity; HO the observed heterozygosity; FIS Wright’s
inbreeding coefficient and P the probability associated with the
test of Hardy–Weinberg equilibrium.
respectively; Mann–Whitney test on expected heterozygosities for each locus: NCentral Asia = 7; NWestern Europe = 7;
U = 20; Z = −0.575; P = 0.565). There was a significant
deviation from Hardy–Weinberg equilibrium when considering all loci together (Central Asia: FIS = 0.076, P = 0.019;
Western Europe: FIS = 0.093, P = 0.006). However, this
result was predominantly caused by a significant heterozygote deficiency at locus APH1 (Central Asia: FIS = 0.196,
P = 0.02; Western Europe: FIS = 0.351, P < 10−3). The FIS calculated without APH1 is much lower and only marginally
significant (Central Asia: FIS = 0.060, P = 0.05; Western
Europe: FIS = 0.055, P = 0.09). APH1 was not discarded
from further analyses because this heterozygote deficiency
should have limited effect on the estimates of interpopulation
differentiation, as FST is especially intended to separate
deviations from Hardy–Weinberg equilibrium due to geographical structuring, as opposed to other causes. After
Bonferroni correction, no pair of loci was in significant
linkage disequilibrium.
Population genetic structure
The phylogenetic relationships between the control region
haplotypes based on the neighbour-joining method are
presented in Fig. 3. The topology of the tree is poorly
supported, as shown by low bootstrap values. Haplotypes
found in the two populations do not form any distinct
clade. However, a nonrandom distribution of haplotypes
is apparent: the Western European haplotypes are mainly
Fig. 3 Phylogenetic tree obtained by the neighbour-joining method,
based on Tamura–Nei distances with gamma correction for 64
individuals. Each symbol corresponds to one individual, filled
symbol indicate Western European populations while open symbols
indicate Central Asian populations. The symbol corresponds to
the subpopulation: , Dombes; , Camargue; , Ebro Delta;
, Doñana; , Constance; , Volga Delta; , Kazakhstan. Bootstrap
values based on 1000 permutations.
grouped into two clades and several haplotypes are shared
by more than one individual, while some haplotype groups
are not detected in Western Europe. On the contrary, all
haplotypes from Central Asia are present in only one
individual, and all haplotypic groups are present in Central
Asia. In addition, the branches of the tree are much shorter
on average for the Western Europe haplotypes.
This nonrandom distribution of haplotypes was confirmed by the amova, which indicates significant differentiation between Central Asian and Western European
populations (ΦST = 0.152, P < 0.01) explaining 15% of the
total genetic variance. On the contrary, estimates of population structure based on nuclear genes are very low and
not significantly different from zero ( FST = 0.004; P = 0.14).
© 2004 Blackwell Publishing Ltd, Molecular Ecology, 13, 1035–1045
C O N S E R V A T I O N G E N E T I C S O F A D U C K S P E C I E S 1041
Table 5 Number of migrants estimated using F-statistics assuming
a finite-island model for the three different markers
Mitochondrial DNA
Multilocus nuclear markers
Microsatellites only (four loci)
Nuclear introns only (three loci)
FST or ΦST
Nm or Nfmf
estimation
ΦST = 0.152
0.004 (ns)
0.003 (ns)
0.006 (ns)
Nfmf = 1.40
31.20
49.48
21.43
ns, not significantly different form zero.
The corresponding Nm values are about 30 migrants per
generation for nuclear markers and fewer than two female
migrants per generation for mtDNA between Western
Europe and Central Asia (Table 5). No significant structure
was detected among European populations for mtDNA
(amova, P = 0.312) or for nuclear markers (FST not significantly different from zero).
When using migrate on mtDNA sequences, some runs
produced biologically unrealistic values (of population
size for example) suggesting convergence towards local
likelihood maxima, and were excluded. Regarding microsatellites, gene flow and population size estimates were
convergent but some runs gave extremely wide confidence
intervals. For nuclear introns, all estimations were convergent. Estimates of gene flow with migrate provided more
similar results between the different markers than estimation based on the finite-island model. Gene flow from
Central Asia to Western Europe was estimated as 0.64
female migrants (95% confidence interval: 0.249–1.145) for
mtDNA control region sequences (or 1.280 migrants per
generation assuming a balanced sex ratio and no sex-biased
dispersal), 1.706 (1.098 –2811.914) for microsatellites [0.969
(0.944–1.004) if runs with very large confidence intervals
are excluded] and 4.095 (2.680 – 4.976) for nuclear introns
(Fig. 4). The numbers of migrants from Western Europe to
Central Asia were similar (symmetrical gene flow) for nuclear
markers. For mtDNA sequence data though, migrate estimated significantly higher female gene flow from Western
Europe to Central Asia [51.468 (16.550 – 75.247)] than from
Central Asia to Western Europe [0.640 (0.249 –1.145)]. This
could however, be a consequence of the hierarchy of diversities among the two populations: all haplotypes found
in Europe are also present in Asia, so each European
individual could be a potential Asian migrant.
Demographic inferences
Neutrality tests for mtDNA control region sequences
rejected the hypothesis of the neutral equilibrium model
for the Western Europe population with Tajima’s test
© 2004 Blackwell Publishing Ltd, Molecular Ecology, 13, 1035–1045
Fig. 4 Mean effective size (2Nf µ for mitochondrial sequences,
with Nf = females effective size = 0.5 Ne for a balanced sex ratio;
Neµ for nuclear introns and microsatellites) and mean number of
migrants (2Nfmf for mitochondrial sequences, with mf = females
migration rate; Nem for nuclear introns and microsatellites)
estimated by the software migrate on mitochondrial sequences
(a), nuclear introns (b) and microsatellites (c). Values in bold type
indicate the maximum likelihood estimator, and those in italic
indicate 95% confidence intervals.
(D = −1.637; P = 0.0425) and for Central Asia with Fu’s test
(Fs = −24.101; P < 10− 4). In the two other tests, the neutral
hypothesis was not rejected.
Growth rates estimated by fluctuate were positive for
both populations, with a larger growth rate for Central
Asia (g = 409.18; SD = 32.79 for Central Asia and g = 64.03;
SD = 22.70 for Western Europe). Considering a per-site
mutation rate µ for the mtDNA control region comprised
between 10−5 and 10−7 (Quinn 1992; Freeland 1997; Kidd
& Friesen 1998), an absolute growth rate (λ = exp g *µ) and
the time until the populations double were evaluated, as
presented in Table 6. Even if a mutation rate as high as 10 −5
is considered, λ is higher than one for both populations.
1042 L . G A Y E T A L .
Table 6 Estimates of θ (effective population size × mutation rate)
and of g (exponential growth rate) for Central Asian and Western
European populations using fluctuate
g
θ
λ
µ = 10−5
µ = 10−7
td
µ = 10−5
µ = 10−7
Central Asia
Western Europe
409.184 (SD 32.791)
3.177 (SD 0.931)
64.032 (SD 22.701)
0.065 (SD 0.010)
1.004100
(1.004429 –1.003771)
1.000041
(1.000044 –1.000038)
1.000640
(1.000413 –1.000413)
1.000006
(1.000009 –1.000004)
170
16.906
1.083
115.525
The ratio of transitions to transversions was 7.5; θ = 2Nf µ with
Nf being the female effective population size and µ being the
mutation rate. λ is the absolute growth rate [λ = exp(g × µ)] and
td is the time until the population doubles in years. Confidence
intervals are shown in parenthese and SD are indicated.
Thus, neither population shows any sign of demographic
decline.
summary statistics and should thus be more powerful.
However, it is computationally very demanding (Emerson
et al. 2001) and it is practically difficult to find the appropriate
chain length for reaching convergence. As a consequence,
the robustness of these estimations to violations of the
underlying assumptions has still to be tested (Neigel 2002)
and it seems difficult to conclude on the comparison of
FST-based and coalescence-based estimates of gene flow.
The existence of morphological differences between the
Western European and Central Asian populations indicates
that they constitute distinct pools of individuals. Morphological differences are not necessarily linked to genetic
structure, as they could result from phenotypic plasticity
(James 1983). However, if dispersal between Western
Europe and Central Asia was completely random, no
morphological differences would be expected, not even
as a result of plasticity. This morphological analysis thus
further confirms that individuals do not move randomly
between these two distant areas. Moreover, banding data
do not mention any bird banded in Central Asia being
recaptured in Western Europe. Of course, the validity of
that observation depends on the ‘banding effort’ in Central
Asia, but it argues for restricted migration from Central
Asia to Western Europe.
Discussion
Comparison of mitochondrial and nuclear genetic
structure
Population structure
Whereas ΦST indicates a substantial level of mitochondrial
genetic structure, its nuclear analogue F ST is very low
(0.004) and not significantly different from zero. Because of
differences in effective population size of the markers,
structure indices estimated with nuclear or mitochondrial
markers are not directly comparable (see Crochet 2000).
If the red-crested pochard populations were at equilibrium
and without sex-biased dispersal, the observed mitochondrial ΦST of 0.152 would be equivalent to a theoretical
nuclear FST of 0.043, which is approximately 10 times higher
than the observed value for nuclear markers. Two hypotheses
could explain the discrepancy between mitochondrial and
nuclear structure: (i) return time to equilibrium and (ii)
sex-biased dispersal.
In the first of these hypotheses, because the effective
population size of mitochondrial markers is four times
lower than for nuclear DNA in gonochoric species, mtDNA
returns faster to equilibrium than nuclear DNA. In
humans, it has been shown that demographic events
disturbing population equilibrium resulted in different
genetic diversity patterns for mitochondrial and nuclear
DNA (Fay & Wu 1999; Hey & Harris 1999). In the red-crested
pochard, recent demographic events (founding effect or
bottlenecks) could have affected genetic equilibrium (see
below), leading to different nuclear and mitochondrial
genetic structure.
The lack of significant genetic structure among Western
European subpopulations for all markers fits well with
the ringing data (Defos du Rau 2002), which shows
numerous exchanges of individuals between these various
subpopulations. A significant genetic structure exists
between Western Europe and Central Asia for the mtDNA
(ΦST = 0.152, P < 0.01) but not for nuclear loci. Estimates
of female dispersal rates from Central Asia to Western
Europe using mtDNA are consistently low (1.4 females per
generation for FST-derived estimates and 0.64 females per
generation for coalescence-based methods). Considering
the recent history of the species (see below), these results
are probably based on a nonequilibrium situation and
actual levels of dispersal between these populations are
probably even lower (see Whitlock & McCauley 1999
for details on the implications of deviation from the
assumptions of the island model). Estimates of the amount
of nuclear gene flow are more difficult to interpret. On the
one hand, the nonsignificant value of nuclear FST precludes
making any precise estimates of gene flow, but suggests
a high level of gene flow. On the other hand, migrate
estimates suggest a low level of nuclear gene flow, even if
the very large confidence intervals for microsatellites make
this conclusion questionable. The approach employed in
migrate makes full use of data rather than relying on
© 2004 Blackwell Publishing Ltd, Molecular Ecology, 13, 1035–1045
C O N S E R V A T I O N G E N E T I C S O F A D U C K S P E C I E S 1043
Alternatively, the difference of structuring patterns
between mtDNA and nuclear markers could be the consequence of sex-biased dispersal. In Anseriformes (geese
and ducks, Greenwood & Harvey 1982), unlike the prevalent
pattern of male-biased philopatry in avian species, females
often display strong natal- and breeding-site fidelity while
males can migrate long distances (see Avise et al. 1992 for
the snow goose Anser caerulescens and Blums et al. 2002
for three duck species). As a result, a recent study using
nuclear and mitochondrial markers in the spectacled eider
duck Somateria fisheri found the same pattern of strong
mitochondrial structure and low nuclear differentiation as
obtained here (Scribner et al. 2001).
A comparison between male and female relatedness
within populations, as has been performed in the red
grouse Lagopus lagopus scoticus (Piertney et al. 1998) could
allow discrimination between these two hypotheses. If natal
philopatry reduces female-mediated gene flow, females
are expected to be more closely related than males (Luikart
& England 1999, see Prugnolle & De Meeus 2002 for a
review).
Neutral evolution of mtDNA and demographic inferences
The failure of red-crested pochard control region sequences
to pass Tajima’s test (Tajima 1989a) (for Western Europe) and
Fu’s test (Fu 1997) (for Central Asia) does not necessarily
indicate a non-neutral molecular evolution. Deviations
from the assumption of demographic equilibrium (after
demographic expansions or bottlenecks) can lead to the
rejection of neutrality in absence of selection (Tajima
1989b). Indeed, Tajima’s test has been used before to make
demographic inferences (e.g. Fry & Zink 1998 in the song
sparrow Melospiza melodia; Fay & Wu 1999 in humans).
fluctuate also estimates a positive growth rate for each
population, but could be affected by selective processes as
well. However, various studies have indicated that mtDNA
control region polymorphism is usually selectively neutral
(Fry & Zink 1998; Milot et al. 2000; Griswold & Baker 2002),
even if hitchhiking cannot be excluded because of the complete genetic linkage of mitochondrial genes. The results
of Tajima’s and Fu’s tests and fluctuate are thus likely
to reflect demographic fluctuations of red-crested pochard
populations (demographic expansion in Central Asia, founding effect in Western Europe). To reject the selection hypothesis, other neutrality tests discriminating the effect of
demography and selection would be necessary (see Nielsen
2001 for a review).
Phylogeography of the red-crested pochard
Despite low mitochondrial gene flow, the absence of
original mitochondrial lineages in Western Europe and the
lack of reciprocal monophyly between Western Europe
© 2004 Blackwell Publishing Ltd, Molecular Ecology, 13, 1035–1045
and Central Asia indicate a recent isolation of Western
European from Central Asian populations.
The high FST-estimates for mitochondrial sequences are
compatible with an event of colonization (Austerlitz et al.
1997). The ‘star-like’ phylogeny of Western Europe haplotypes contrasts with the deeper branches among Central
Asian lineages, and haplotype diversity is low in Europe
compared to central Asia. All lineages present in Western
Europe are also present in Central Asia. The haplotype
diversity in Western Europe is thus a limited sample of
the Central Asian diversity. This loss of diversity is the
expected outcome of a colonization event by a low number
of founders, with the source population containing the
most divergent haplotypes (Austerlitz et al. 1997; see also
Ingman et al. 2000 for an example in humans). Mitochondrial data are thus compatible with a recent colonization
scenario for Western Europe. Nevertheless, a similar genetic signal would result from a strong decrease in population size in Europe following fragmentation of a formerly
continuous distribution with little or no isolation by distance.
The demographic inferences obtained with fluctuate
and confirmed by the results of Tajima’s test, however,
indicate an increase in population size both in Central Asia
and Western Europe, which does not support the fragmentation scenario. Moreover, historical data date the first
breeding attempts by the red-crested pochard in Western
and Central Europe from the late 1800s (Mayaud 1966;
Cramps & Simons 1977; Hagemeijer & Blair 1997). A recent
colonization of Western Europe by the red-crested pochard
is therefore the most likely scenario.
Conservation of the red-crested pochard in Western
Europe
Analyses of mtDNA and nuclear markers clearly demonstrate that populations of red-crested pochard are not
structured in reciprocally monophyletic clusters for any
marker and that the Western European populations do not
contain any original genetic variants. Western European
populations thus do not constitute an Evolutionarily
Significant Unit (Moritz 1994). However, mitochondrial
allele frequencies are significantly different between
Western Europe and Central Asia, which corresponds to
the definition of the Management Unit sensu Moritz (1994).
Given the low amount of female-mediated gene flow
estimated from FST or migrate, the demographic contribution
of migrant females from Central Asia to Western Europe
can be confidently considered as negligible. Morphological
differences between these populations regardless of the
sex show that males do not disperse freely among these
areas either.
In terms of demography, the low number of female
migrants means that numbers of red-crested pochards in
Western Europe cannot be efficiently supplemented by
1044 L . G A Y E T A L .
immigration from the larger Asian populations. The Western European populations of red-crested pochard thus
need to be managed independently from the large Central
Asian populations. Defining precisely the limits of this
management unit will require more sampling in Eastern
Europe and Turkey, to determine whether Eastern
European populations belong to the same management
unit as the Western European populations or are part of the
Central Asian populations, or even constitute a distinct
unit.
Acknowledgements
We thank all the duck hunters from Camargue and Dombes
(France) and Ebro Delta (Catalonia) as well as from the Chassorbis,
D.H.D. Laïka and Seladang hunting companies who provided us
with the materials used in this work. In particular, we thank Cati
Gerique (Generalitat Valenciana), Christophe Buquet, Fransesc
Vidal Esquerre (Natural Park of Ebro Delta), Jean-Yves Fournier
(ONCFS), Andy Green and Jordi Figuerola (Estación Biológica
de Doñana), Alain Méric Grossi, Dr Botond Kiss (Danube Delta
National Institute), Eric Wacheux (Chassorbis). We are also
grateful to Patricia Sourrouille and Chantal Debain for technical
assistance with the molecular work and to Mark Adams, Frédéric
Jiguet and Jean-Marc Pons for their assistance in the British
Museum of Natural History and the Muséum National d’Histoire
Naturelle, respectively. We thank Philippe Jarne, Nicolas Galtier
and two anonymous referees for helpful comments on the
manuscript.
References
Austerlitz F, Jung-Muller B, Godelle B, Gouyon P-H (1997) Evolution of coalescence times, genetic diversity and structure during
colonisation. Theoretical Population Biology, 51, 148 –164.
Avise JC, Alisauskas RT, Nelson WS, Ankney CD (1992) Matriarchal
population genetic structure in an avian species with female
natal philopatry. Evolution, 46, 1084 –1096.
Beerli P (2002) MIGRATE: Documentation and Program, Part of LAMARK, version 1.5. Available from http://evolution.genetics.washington.edu/
lamark.html.
Beerli P, Felsenstein J (1999) Maximum-likelihood estimation of
migration rates and effective population numbers in two populations using a coalescent approach. Genetics, 152, 763 –773.
Belkhir K, Borsa P, Chikhi L, Raufaste N, Bonhomme F, (1996–2001)
GENETIX version 4.2: Logiciel Sous Windows TM Pour la Génétique
Des Populations. Laboratoire Génome Populations Interactions,
CNRS UMR 5000, Université Montpellier II, Montpellier, France.
Available from http://www.univ-montp2.fr/∼genetix/genetix/
genetix.htm.
Blums P, Nichols JD, Hines JE, Mednis A (2002) Sources of variation in survival and breeding site fidelity in three species of
European ducks. Journal of Animal Ecology, 71, 438 – 450.
Chan Y, Arcese P (2003) Morphological and microsatellite differentiation in Melospiza melodia (Aves) at a microgeographic scale.
Journal of Evolutionary Biology, 16, 939 – 947.
Cramps S, Simons KEL (1977) Handbook of the Birds of Europe, the
Middle East and North Africa. The Birds of the Western Palearctic.
Oxford University Press, Oxford.
Crochet P-A (2000) Genetic structure of avian populations —
allozymes revisited. Molecular Ecology, 9, 1463–1469.
Defos du Rau P (2002) Elements for a red-crested pochard (Netta
rufina) management plan. Game and Wildlife Science, 19, 89 –141.
Dehorter O, Rocamora G (1999) Nette rousse Netta rufina. In:
Oiseaux Menacés et à Surveiller en France. Listes Rouges et Recherche
de Priorités (ed. Sdéod FL), pp. 58–59. Société d’Études Ornithologiques de France/LPO, Paris.
Delany S, Scott D (2002) Waterbird Population Estimates. Wetlands
International, Wageningen, the Netherlands.
Emerson BC, Paradis E, Thébaud C (2001) Revealing the demographic histories of species using DNA sequences. Trends in
Ecology and Evolution, 16, 707–716.
Excoffier L, Smouse PE, Quattro JM (1992) Analysis of molecular
variance inferred from metric distances among DNA haplotypes:
application to human mitochondrial DNA restriction data.
Genetics, 131, 479–491.
Fay JC, Wu C-I (1999) A human population bottleneck can account
for the discordance between patterns of mitochondrial vs. nuclear
DNA variation. Molecular Biology and Evolution, 16, 1003 –1005.
Fields RL, Scribner KT (1997) Isolation and characterization of
novel waterflow microsatellite loci: cross-species comparisons
and research applications. Molecular Ecology, 6, 199–202.
Fraser DJ, Bernatchez L (2001) Adaptive evolutionary conservation: towards a unified concept for defining conservation units.
Molecular Ecology, 10, 2741–2752.
Freeland J (1997) The Genetic Evolutionary History of the Darwin’s
Finches. Queen’s University, Kingston, Canada.
Friesen VL, Congdon BC, Walsh HE, Birt TP (1997) Intron variation
in marbled murrelets detected using analyses of single-stranded
conformational polymorphisms. Molecular Ecology, 6, 1047–1058.
Friesen VL, Congdon BC, Kidd MG, Birt TP (1999) Polymerase
chain reaction (PCR) primers for the amplification of five nuclear
introns in vertebrates. Molecular Ecology, 8, 2141–2152.
Fry A, Zink RM (1998) Geographic analysis of nucleotide diversity
and song sparrow (Aves: Emberizidae) population history.
Molecular Ecology, 7, 1303–1313.
Fu YX (1997) Statistical tests of neutrality of mutations against
population growth, hitchhiking and background selection.
Genetics, 147, 915–925.
Gaggiotti OE, Lange O, Rassmann K, Gliddon C (1999) A comparison of two indirect methods for estimating average levels
of gene flow using microsatellite data. Molecular Ecology, 8, 1513–
1520.
Greenwood PJ, Harvey PH (1982) The natal and breeding dispersal
of birds. Annual Revue of Ecological Systematics, 13, 1–21.
Griswold CK, Baker AJ (2002) Time to the most recent common
ancestor and divergence times of populations of common
chaffinches Fringilla coelebs) in Europe and North Africa: insights
into pleistocene refugia and current levels of migration. Evolution,
56, 143–153.
Hagemeijer WJM, Blair MJ (1997) The EBCC Atlas of European
Breeding Birds. Their Distribution and Abundance. T. & AD Poyser,
London.
Hey J, Harris E (1999) Population bottlenecks and patterns of
human polymorphism. Molecular Biology and Evolution, 16, 1423–
1426.
Ingman M, Kaessmann H, Paabo S, Gyllensten U (2000) Mitochondrial genome variation and the origin of modern humans.
Nature, 408, 708–713.
James FC (1983) Environmental component of morphological
differences in birds. Science, 221, 184–186.
© 2004 Blackwell Publishing Ltd, Molecular Ecology, 13, 1035–1045
C O N S E R V A T I O N G E N E T I C S O F A D U C K S P E C I E S 1045
Kidd MG, Friesen VL (1998) Analysis of mechanisms of micro
evolutionary change in Cepphus guillemots using patterns of
control region variation. Evolution, 52, 1158 –1168.
Koljonen ML (2001) Conservation goals and fisheries management
units for Atlantic salmon in the Baltic Sea area. Journal of Fish
Biology, 59, 269–288.
Kuhner MK, Yamato J, Felsenstein J (1998) Maximum likelihood
estimation of population growth rates based on the coalescent.
Genetics, 149, 429–434.
Kumar S, Tamura K, Jakobsen IB, Nei M (2001) MEGA2: Molecular
Evolutionary Genetics Analysis Software. Arizona State University,
Tempe, AZ.
Lessa EP (1992) Rapid surveying of DNA sequence variation
in natural populations. Molecular Biology and Evolution, 9, 323–
330.
Luikart G, England PR (1999) Statistical analysis of microsatellite
DNA data. Trends in Ecology and Evolution, 14, 253 – 256.
Maak S, Neumann K, Von Lengerken G, Gattermann R (2000) First
seven microsatellites developed for the Peking duck (Anas
platyrhynchos). Animal Genetics, 31, 228 – 241.
Mayaud N (1966) Contribution à l’histoire de Netta rufina Pallas, la
Nette à huppe rousse en Europe occidentale. Alauda, 34, 191–
199.
Milot E, Gibbs HL, Hobson KA (2000) Phylogeography and
genetic structure of northern populations of the yellow warbler
(Dendroica petechia). Molecular Ecology, 9, 667 – 681.
Monval JY, Pirot JY (1989) Results of the IWRB International Waterfowl
Census 1967–86. IWRB, Slimbridge.
Moritz C (1994) Defining ‘Evolutionarily Significant Units’ for
conservation. Trends in Ecology and Evolution, 9, 373 – 375.
Neigel JE (2002) Is FST obsolete? Conservation Genetics, 3, 167–173.
Nice CC, Shapiro AM (1999) Molecular and morphological divergence in the butterfly genus Lycaeides (Lepidoptera: Lycaenidae)
in North America: evidence of recent speciation. Journal of Evolutionary Biology, 12, 936 – 950.
Nielsen R (2001) Statistical tests of selective neutrality in the age of
genomics. Heredity, 86, 641– 647.
Palumbi SR, Baker CS (1994) Contrasting population structure
from nuclear introns sequences and mtDNA of humpback
whales. Molecular Biology and Evolution, 11, 426 – 435.
Piertney SB, MacColl ADC, Bacon PJ, Dallas JF (1998) Local
genetic structure in red grouse (Lagopus lagopus scoticus):
evidence from microsatellite DNA markers. Molecular Ecology,
7, 1645–1654.
Podolsky RH, Holtsford TP (1995) Population-structure of
morphological traits in Clarkia-Dudleyana. I. Comparison of FST
between allozymes and morphological traits. Genetics, 140, 733–
744.
Posada D, Crandall KA (1998) modeltest: testing the model of
DNA substitution. Bioinformatics, 14, 817 – 818.
© 2004 Blackwell Publishing Ltd, Molecular Ecology, 13, 1035–1045
Prugnolle F, de Meeus T (2002) Inferring sex-biased dispersal from
population genetic tools: a review. Heredity, 88, 161–165.
Quinn TW (1992) The genetic legacy of Mother Goose: phylogeographic patterns of lesser snow goose Chen caerulescens
caerulescens maternal lineages. Molecular Ecology, 1, 105–117.
Rice WR (1989) Analyzing tables of statistical tests. Evolution, 43,
223–225.
Rousset F (2001) Genetic approaches to the estimation of dispersal
rates. In: Dispersal (eds Clobert J, Danchin E, Dhondt AA,
Nichols JD), pp. 18–28. Oxford University Press, Oxford.
Ruzzante DE, Taggart CT, Lang S, Cook D (2000) Mixed-stock
analysis of Atlantic cod near the Gulf of St Lawrence based on
microsatellite DNA. Ecological Applications, 10, 1090–1109.
Saez-Royuela R (1997) Pato Colorado. In: Atlas de Las Aves de España
(1975–95), pp. 80–81. Lynx Edicions, SEO/BirdLife, Barcelona.
Schneider S, Roessli D, Excoffier L (2000) ARLEQUIN, Version 2.000: a
Software for Population Genetics Data Analysis. Genetics and Biometry Laboratory. University of Geneva, Geneva, Switzerland.
Scott DA, Rose PM (1996) Atlas of Anatidae populations in Africa
and western Eurasia. Wetland International, Wagenigen, the
Netherlands.
Scribner KT, Petersen MR, Fields RL et al. (2001) Sex-biased gene
flow in spectacled eiders (anatidae): inferences from molecular
markers with contrasting modes of inheritance. Evolution, 55,
2105–2115.
Shedden CB (1986) Status of European Quarry Species. British Association for Shooting and Conservation, Dunkeld, UK.
Tajima F (1989a) Statistical method for testing the neutral mutation hypothesis by DNA polymorphism. Genetics, 123, 585 – 595.
Tajima F (1989b) The effect of change in population size on DNA
polymorphism. Genetics, 123, 597–601.
Weir BS, Cockerham CC (1984) Estimating F-Statistics for the analysis of population structure. Evolution, 38, 1358–1370.
Whitlock MC, McCauley DE (1999) Indirect measures of gene flow
and migration: FST ≠ 1/(1 + 4Nm). Heredity, 82, 117–125.
This work formed part of the Master of Science degree for L. Gay
who is currently pursuing a doctorate at the CEFE-CNRS in
Montpellier on the impact of selection and hybridization on the
speciation process in large white-headed gulls. P-A. Crochet is a
post-doctorate researcher at the CEFE–CNRS. His main interest is
in the application of molecular methods to the study of natural
populations with an emphasis on questions ranging from dispersal to systematics, phylogeny and biogeography. P. Defos du
Rau and J-Y. Mondain-Monval work for ONCFS (the French Game
and Wildlife Service) and are involved in monitoring surveys
and management plans for various animal species and wetland
sites.
Chapter 6
DEMOGRAPHY
AND HARVESTING SUSTAINABILITY OF A RARE GAME WATERBIRD: THE
RED-
CRESTED POCHARD IN EUROPE AND THE CAMARGUE (FRANCE)
P. Defos du Rau, J.Y. Mondain-Monval, C. Barbraud, A. Olivier, A. Johnson, H. Kowalski, E.
Cam
INTRODUCTION
Sustainability of waterfowl exploitation has long been a major issue in wildlife ecology
(Kalchreuter 1996, Nichols & Johnson 1996), since the middle of 20th Century, mainly in North
America (Nichols 1991, Nichols et al. 1995), but also in Mediterranean Europe (Hoffmann
1963a & b). In response to the uncertainty and questions in waterfowl demography, and
conservation- or exploitation-oriented management, a nation-wide, continental, semiexperimental strategy known as «Adaptive management» has been developed in North America
in the early 1990s. The long-term objective of Adaptive Management, which integrates
monitoring schemes of several demographic and management parameters, is to ensure
sustainability and reduce uncertainty in species exploitation (Nichols et al. 1995). A similar
integrated monitoring is advocated for Europe. Indeed, Europe has a strong tradition of duck
shooting and, locally, relatively high levels of duck exploitation (Nichols et al. 2001).
Sustainability of European duck harvesting is therefore of strong management concern.
Evaluation of hunting harvest sustainability requires integrated analytical tools allowing
optimisation of objective functions and decision making (Conroy & Moore 2001) over entire
biogeographical populations, i.e. at continent scale. However, there are structural obstacles to its
implementation such as data sparseness and heterogeneity, or simply sector-specific lacks
(Elmberg et al. 2006), as well as linguistic and cultural differences.
A common technique allowing assessment of harvest sustainability for exploited species, but
also of conservation-oriented management options for rare or threatened species, is demographic
population projection models (Caswell 1989, Ferrière et al. 1996, Gauthier & Lebreton 2004,
Lebreton 2005). Matrix population models are designed for populations structured by age or
stage; they are well suited for situations where survival probability has been estimated using a
discrete-time model, by capture-recapture methodology for example (Willams et al. 2002). They
may be useful for diagnostic of exploitation sustainability even with minimal knowledge of the
1
species biology and demography, through estimation of integrative population traits like mean
generation time (Gauthier & Lebreton 2004) or demographic invariants (Niel & Lebreton 2005).
The Red-crested Pochard (RCP) is a migratory duck breeding mainly in Central Asia, from
China to the Black Sea, and more locally in Central and Western Europe (Snow & Perrins 1998).
The species is listed in appendix III of the Berne Convention, in appendix II of the Bonn
Convention, in appendix II/2 of the EU Birds Directive 94/24 and in Annex II of the AfricanEurasian Waterbird Agreement. RCP is both exploited and of conservation concern. Although it
is regionally considered scarce or threatened, the species is hunted in France, Portugal, Romania
and Spain. Furthermore, although capture-recapture data sets have been collected on this species
(Johnson 1975, Defos du Rau et al. 2003), which allows preliminary estimation of some vital
rates, existing data are sparse and there is a general lack of knowledge of population dynamics of
the species since seminal studies by Amat (1982) and Llorente & Ruiz (1985). Because of this,
matrix population models seem the only suitable analytical tool to explore the population
dynamics, to provide preliminary insights into exploitation sustainability and to assess
conservation options for RCP in Europe.
Our main objectives were the following. First, we estimated the vital rates of the European
population, we specified a matrix population projection model, and then validated it using
independent data sets from abundance surveys. Second, we addressed the following questions:
(i) - Are there compensatory mechanisms to harvesting? We addressed whether the impact of
harvesting on mortality is compensatory or additive to natural mortality (Lebreton 2005). This is
a major debate in exploited population dynamics theory, relevant to harvesting sustainability. We
also addressed the existence of such compensation on reproductive parameters (e.g., fledging
success; Kalchreuter 1996, Blums et al. 2002a).
(ii) - What is the relationship between harvest levels and RCP population dynamics? We
assessed current growth rate and precision, theoretical maximum sustainable harvest level (Niel
& Lebreton 2005) and the influence of change in harvest on RCP population growth (Gauthier &
Lebreton 2004).
2
(iii) - Which recommendations are prioritized by perturbation analyses? Considering the lack of
data for some vital rates of RCP, we chose to focus on setting research priorities rather than
management priorities and performed some perturbation analyses accordingly, i.e. by giving
more credit to parameter uncertainty analyses (Hunter et al. 2000) than to sensitivity analyses
(Caswell 1989, Reed et al. 2002).
(iv) - What are the projections from different parameterizations of the population matrix model?
We simulated short-term variation in RCP Camargue subpopulation size under different versions
of a preliminary scenario based on environmental stochasticity (following recommendations
from Reed et al. 2002 and Fieberg & Ellner 2001).
We addressed these questions at two geographic scales: (i) the European RCP population (as
defined by Krebs 1972) and (ii) the Camargue RCP subpopulation, where several vital traits
were recently estimated from.
METHODS
Study site and species
The RCP originates from Central Asia (Gay et al. 2004). In France and in Europe in general, it is
considered a rare, locally threatened duck species (Dehorter & Rocamora 1999, Snow & Perrins
1998). The Central Europe and Western Mediterranean population is considered a distinct
demographic unit from the Central Asian one (Gay et al. 2004), and is estimated at 50000
individuals (Delany & Scott 2002). Within this Central Europe and Western Mediterranean
population, the species is hunted in Portugal, France and Spain, despite the fact that the species is
rare compared to most other game duck species (Delany & Scott 2002). The total harvest for this
population was estimated at 8000 birds per year in the mid-80’s (Shedden 1986), 700 of which
on average were harvested in France and the rest in Spain. The French wintering population
(thus, the French harvest) is located almost exclusively in the Rhône river mouth, named the
Camargue, which is a vast natural and cultivated delta of 1450 km2 on the west coast of the
Mediterranean. Natural habitats are freshwater and brackish wetlands, which cover
approximately 40% of the total delta area. The Camargue delta is split between 230 estates, most
of which are privately-owned shooting estates, managed for breeding and wintering wildfowl.
3
The Camargue subpopulation of RCP was estimated at 560 brood-rearing females (95%
confidence interval: 436–855) for the whole Camargue in year 2001 (Defos du Rau et al. 2003).
A two-stage matrix model
As in several other studies of duck demography (Hoekman et al. 2002, Flint et al. 2006), we built
a two-stage matrix model for female RCP (Williams et al. 2002). The model was based on a
post-breeding census, because population estimates time series are generaly available for
wintering birds. The matrix model is the following:
⎡ F1 * S1 F 2 * S 2⎤
,
⎢ S1
S 2 ⎥⎦
⎣
where Fi stands for fecundity at age i and Si for survival probability at age i. Like many duck
species, RCP is supposed to be able to breed at age 1 year (Cramp & Simmons 1977, Johnson et
al. 1992). We implemented the model in software program ULM (Legendre & Clobert 1995,
Ferrière et al. 1996).
We used this model to simulate temporal variation in numbers of both the open Camargue
breeding population and the closed West and Central European population. Survival probabilities
were added an immigration component m to account for the fact that the Camargue breeding
population is an open population. Vital rates used in this model where either estimated using data
from the Camargue breeding and wintering populations, or came from studies of breeding RCP
by Amat (1982), or from the closely related Aythya or Anas duck genus, mainly from Batt et al.
(1992) and Blums et al. (1996). For discussion of costs and benefits of use of data from other
subpopulations or species, respectively, see Frederiksen et al. (2001) and Hunter et al. (2000).
Estimate of variance for deterministic growth rate estimates were calculated following Caswell
(1989).
4
Census and harvest time-series
We used RCP census data for model initialization with realistic initial population estimates, and
for model validation. Three different census data sets were used:
(i) The most restricted one concerned the Camargue breeding population of RCP, which was
estimated at 560 brood-rearing females in 2001, based on a detection probability estimated at
0.575 (Defos du Rau et al. 2003). In addition, RCP was first observed breeding in the Camargue
in year 1894 (Mayaud 1966). Those two population size estimates provided a test set and an
initial population size for simulation of future RCP Camargue population dynamics. Concerning
initial population size, we assumed that detection probability in year 1894 was < 1 (when the
first breeding female was observed). We set the initial population size (number of breeding
females) equal to 3. The test data set was used to assess the degree of agreement between
estimated numbers (560 breeding females in year 2001) and numbers obtained using different
versions of the population projection model (see below).
(ii) - The mid-winter International Waterbird Census coordinated in Europe by Wetlands
International is the basis for published RCP population estimates (Table 1).
Population estimate Year of estimation Expected female population size Reference
20000
1994
10000
Rose & Scott (1994)
25000
1995
12500
Scott & Rose (1996)
27500
1997
13750
Keller (2000)
37000
1999
18500
Gilissen et al. (2002)
50000
2002
25000
Delany & Scott (2002)
Table 1: published population estimates (number of individuals) for the Central and Western
European RCP closed population
Because of the relatively important fluctuations and uncertainties in the European data set for
RCP winter counts prior to 1995 (Scott & Rose 1996), we started population projections in year
1994. We used the Rose & Scott estimates of 1994 (Table 1) as the initial population size value.
Again, this short time-series of RCP European counts provided a test data set for model-based
simulation of RCP European population dynamics.
5
(iii) Also in winter, aerial waterbirds census were conducted every winter months from 1966
onward over the entire wintering and harvesting area of RCP in the Camargue by the Centre
National de la Recherche Scientifique (Tamisier & Dehorter 1999) and the Tour du Valat
(Gauthier-Clerc, Tour du Valat pers. comm.). We used this long-term time-series of RCP winter
counts to compute harvest rates for the corresponding Camargue area (over which both winter
census and harvest data sets were collected).
To estimate harvest rate, we used the long-term duck harvest survey performed over a sample of
Camargue hunting estates by Office National de la Chasse et de la Faune Sauvage, from 1951
onward. RCP harvest data are available by estate and by date. A GIS survey of hunting estates
provided an estimate of total area of hunted wildfowl habitat of 38137ha for the whole
Camargue. Considering potential for sampling errors in harvest rate estimates (Lebreton 2005),
we only used years with the largest samples within this long-term harvest monitoring. Relatively
larger sampling effort started from 1988 onward and encompassed an average 19,6% of the total
hunted wildfowl habitats. Total RCP harvest for the Camargue was then estimated by area
expansion. This harvest estimate was added a 20% crippling loss (Anderson & Burnham 1976)
to account for birds that were shot down but not retrieved or not immediately killed.
An index of harvest rate for the Camargue RCP subpopulation was computed as the ratio of the
estimated total harvest of the corresponding year to an estimate of the total number of birds
available for harvest (i.e., the total number of birds passing through the harvest area during the
hunting season). Such an estimate was computed for Teal Anas crecca by Caizergues et al.
(submitted to Ibis) as twice the maximum number counted. This result was based on a model for
local turn-over combining teal monthly counts and local apparent survival estimates.
Consequently, we used twice the observed maximum winter counts of RCP from aerial surveys
over the Camargue as the total RCP numbers available for harvest.
Overall we used an approach similar to the one used by Gauthier et al. (2001) and assumed that
our estimate of harvest rate H (including crippling loss) was reliable for the following reasons.
First, the aerial census conducted in winter encompassed the whole area of RCP wintering and
harvesting (Tamisier & Dehorter 1999). Second, we used only the most recent (and thus the
6
largest samples) in the harvest survey, which implies that the accuracy of harvest estimates was
increased.
Survival probability estimation
We estimated juvenile and adult annual survival probabilities using the only existing markrecovery data set for RCP in the Camargue (maintained by the Tour du Valat), where 269 known
age RCP were captured, weighted and banded between 1952 and 1979. Because of the scarcity
of data, we pooled records from both sexes. Since male ducks generally have higher survival
probability than female (Johnson et al. 1992, Lake et al. 2006), our final female survival estimate
used in the population projection model was probably too high. Twenty percent of the 54
recoveries were actual live recapture but again because of the scarcity of data, we treated every
recovery as a dead one and only performed a dead recovery analysis, which likely caused a
negative bias in our final survival estimates. We chose to use Seber’s dead recovery model
(Williams et al. 2002) implemented in Program MARK (White and Burnham 1999) because we
wanted to address not only the influence of harvest restriction on survival probability (Sedinger
& Rexstad 1994, Lake et al. 2006), but also the influence of individual- and time varying
covariates on survival. There is no straighforward Goodness-Of-Fit test available for this model
with covariates.
- First, we wanted to address the influence of harvest on survival probability. However, because
sampling of harvest before 1973 was inadequate (see above), we did not use observed harvest
rate. Rather, we used a meteorological covariate as a proxy for harvest. We expected low
temperatures to increase harvest by attracting ducks for longer periods at night on waterbird
foraging habitats that are also hunting sites, which would increase the probability of getting shot
earlier at night or later in the morning. We therefore included mean temperature of the four
coldest months of the hunting season (November to February) in models as a year-specific
covariate for survival probability with the a priori hypothesis of a positive relationship between
mean winter temperature and survival.
- Second, we used weight (wa) as an individual covariate of adult survival probability. Our a
priori hypothesis was a positive relationship between weight and survival (Johnson et al. 1992).
7
- RCP captures were performed almost all year round and hunting season before 1973 was
extensive, encompassing all months between July and March. Consequently, there were no
restricted capture and recovery periods. This raises difficulties for estimation of survival using
discrete-time models because survival probability is defined as the probability of surviving over
a standard time interval of a given length (e.g. a year). In the current data set, an individual may
have been recovered only 6 months after banding. Consequently, we computed a second
individual covariate, namely “recovery time-increment” i, to account for the time lag between
banding (Figure 1) and recovery date (Figure 2). We chose not to truncate the data set in separate
capture occasions, which would increase the number of parameters without taking all the
variation in capture date into account. Rather, we defined “theoretical” recovery periods at
standard time intervals. We assigned each individual the time-lag between observed and
theoretical recovery dates and incorporated this time lag in models of survival probability using
an individual covariate. This time-increment (“i”) was computed as the difference in months, on
a one-year scale, between recovery month and banding month. Non-recovered birds were
affected the mean of the i-values calculated for recovered birds. The larger this time-increment,
the later was recovered, and thus the longer the bird survived, compared to theoretical recovery
period. Hence, we expected a negative relationship between the time-increment covariate and
survival probability.
number of ringed RCP
100
80
60
40
20
0
Jan Feb Mar Apr May Jun
Jul
Aug Sep Oct
Nov Dec
month of ringing
Figure 1: monthly frequency of RCP banding captures by Tour du Valat in the Camargue from
1952 to 1976
8
number of recovered RCP
10
8
6
4
2
0
Jan Feb Mar Apr May Jun
Jul
Aug Sep Oct
Nov Dec
month of recovery
Figure 2: monthly frequency of RCP recoveries
The global model assumed year (t)- and age (a)-specific survival and recovery probabilities
(model {S(a,t)r(a,t)}). S and r denoted local survival and recovery parameters, respectively. We
compared this global model with reduced models that assumed constant recovery probability and
time- and/or individual- varying survival. We conducted a model selection procedure by
addressing temporal variation in recovery probability first and then in survival probability, and
then by addressing the influence of year or individual covariates on survival. The relative support
of the data for the various parameterizations was assessed using an Akaike Information Criterion
for small samples (AICc, Anderson et al. 2001).
Fecundity estimation
Fledging success (fs)
We estimated fledging success in two steps.
First, we estimated brood and ducklings survival using data from 83 individually identified RCP
broods monitored every two weeks on 6 occasions between mid-May and mid-August during
breeding seasons 1990, 1991, 2000 and 2001 in the Camargue. Capture-recapture data from
broods were encoded according to the statum in which they were resighted: stratum 1
corresponded to broods of unchanged size compared to the stratum in which they were
previously resighted, and stratum 2 to broods having lost one duckling. We used a multistate
9
capture-recapture model to estimate brood and duckling survival probability (Schwarz et al.
1993, Williams et al. 2002). Multistate models allowed us to estimate local brood survival (Sbi)
and brood capture probabilities (pi) for each stratum i, as well as two transition probabilities
ψ 12 andψ 21 : ψ 12 corresponded to duckling mortality (i.e., the probability of loosing one
duckling). Reverse transition events were impossible; ψ21 was therefore fixed at zero. Nine
broods out of the initial 83 lost two or more ducklings and were excluded from this analysis. We
used multistate models implemented in software program MARK (White and Burnham 1999).
Because of the relative rarity of RCP as a breeding bird in the Camargue, the occurrence of two
or more broods of exact same age and size in the same lake was considered highly unlikely. Each
observed brood was therefore identified (or « marked ») by the combination of its age, size and
location, taking into account that brood size might decrease when getting older. We surveyed a
total of 21 lakes over the 4 years combined. We addressed the influence of year on stratumspecific brood survival probability, as well as of within-season time, and of two individual
covariates (laying date and initial brood size, not colinear with each other: r=0.14, n=74). Our a
priori hypotheses were the following:
- Each year corresponded to a group covariate with a possible effect on all the 5 parameters (Sb1,
Sb2, p1, p2, ψ 12 ). However, we hypothesized that the most likely interannual variation would
concern capture probability (i.e. in this case detection probability). We expected a difference in
detectability between periods 1990-91 and 2000-01, due to major changes in study areas, their
environment and brood visibility.
- We expected brood and duckling survival to vary with time within the breeding season; we
addressed this hypothesis by incorporating a linear time covariate in models.
- We expected laying date and initial brood size to affect brood and duckling survival
probability. Earlier hatched duck broods generally survive better than later hatched ones
(Johnson et al. 1992), although this is not a universal pattern. In contrast, effect of brood size on
brood or duckling survival varies or is even null among species (Johnson et al. 1992, Gendron &
Clark 2002, Schmidt et al. 2006). There is an obvious systemic link between brood and duckling
survival. Consequently, covariate combinations were always kept identical for both parameters,
including quadratic effects within season. Laying date was assessed from visual estimation of the
brood’s age, based on field criteria defined elsewhere (Office National de la Chasse 1982).
10
We started with a model (Sb(t,g,str),p(t,g,str),ψ(t,g,str)) with state- and time-dependent
parameters (Williams et al. 2002) and compared the various models addressing our a priori
hypotheses using the AICc. Goodness-of-fit of the initial group-, state- and time-dependent
model was assessed using U-CARE (Choquet et al. 2005).
In the second step, we used the resulting model (with the largest support according to AICc) to
simulate brood and duckling multistate capture-recapture histories to estimate fledgling success
over a sample of 365 broods of known age and size observed in the Camargue between 1988 and
2005. Duckling survival probability over two weeks was computed as the complement of
transition probability from state 1 to state 2 (1-ψ12). Both brood and duckling survival between
encounter occasions (i.e., over two weeks) were computed for each of the 365 broods according
to the covariates included in the model with the largest support. The square-root of this value
corresponded to weekly survival probability. The number of weeks before fledging was
calculated as the difference between fledging age, known to be 7 weeks-old (Snow & Perrins
1998), and observed age of the brood in weeks. As transition probabilities are conditional on
survival (Williams et al. 2002), we first simulated survival events for each of the 365 broods and
then simulated duckling survival events, which allowed estimation of fledging success for live
broods only. Capture-recapture histories of each of the 365 broods were simulated 1000 times ;
for each week before fledging of these 1000 histories, weekly survival events was a binomial
trial (with the brood weekly survival probability). For each surviving brood in the 1000
simulated histories, brood size was decreased according to duckling weekly survival, for each
week between observation and fledging. Fledging success of each of the 365 observed broods
was then averaged over the 1000 simulated histories (including brood losses). We then averaged
fledging success for each year from 1988 to 2005.
We then used this simulated data set of reproductive output to address the influence of covariates
on fledging success. Breeding success of RCP in the Camargue had previsouly been shown to be
positively influenced by waterdepth and frequency of Chara macrophytes. The latter are
supposed to be favoured by temporary rather than permanent flooding (Grillas 1992, Defos du
Rau et al. 2005). We thus hypothetized that RCP fledging success would be:
- positively influenced by precipitation during the winter and spring preceding breeding season,
which favours high waterdepth,
11
- negatively influenced by precipitation during the preceding spring and summer, which
enhances flooding permanency during the natural drying-off period (Grillas 1992, Tamisier &
Grillas 1994).
We addressed both hypotheses using a regression model; covariates were selected using an
exhaustive model search based on AICc. For predictive purpose, we included 4 colinear
covariates corresponding to each precipitation pattern. However, we only considered models
with a maximum of two covariates, one for each precipitation pattern. The precipitation variables
corresponded to the sum of winter-spring precipitation from September to April, May, June, July
and the sum of precipitation of preceding spring-summer from April or March to June or July.
In the population projection model, we used a 1:1 sex-ratio (“sr”) for ducklings (Johnson et al.
1992) and a brood hatching success (“bh”) of 53,2% estimated from 62 RCP nests in Spain by
Amat (1982). Following Blums et al. (1996) for Aythya ferina and Anas clypeata, we set the
proportion of breeding juvenile birds (“q1”) at 70%. Flint et al. 2006 provided a very similar
estimate of 71% for Aythya marila (see also Anderson et al. 2001 for Aythya valisineria). These
estimates were not available for RCP. Consequently, we used estimates from closely-related
Aythya species. As a first approximation, we arbitrarily fixed adult laying propensity (“q2”) at
100% (as suggested by Blums et al. 1996). We addressed later uncertainty problems on these
imported vital rates. Brood loss was only estimated for surveyed broods and was thus negatively
biased because some brood losses may have occurred prior to brood field survey. However,
effect, variability and uncertainty of this parameter were confounded with effect, variability and
uncertainty of our brood hatching success estimate, which might have been within lower range
for RCP (Llorente & Ruiz 1985). We considered that double brooding and brood parasitism had
a negligible effect on RCP population dynamics. Brood parasitism is a frequent feature of RCP
reproduction strategy (Amat 1991) but it is not known to be additional to regular brood rearing
by the same female as if it was double brooding. Double brooding itself is only occasional in
most duck species (Johnson et al. 1992, Hoekman et al. 2002). Last, renesting after clutch failure
is not known in RCP and was ignored. However, renesting is known in many other duck species.
Consequently, renesting was accounted for in uncertainty analysis in order to evaluate potential
research need on this vital rate.
12
Age-specific fecundities were finally estimated as:
F2=q2*sr*bh*fs
F1=q1*q2*sr*bh*fs
Compensation of harvest (mortality and fecundity)
Compensation on mortality
A major issue in studies of exploited populations is the detection and modeling of demographic
compensation of harvest (Kalchreuter 1996, Lebreton 2005). Harvest is theoretically supposed to
be compensated by decreased mortality following the relationship : S=So(1-bH) where S =
survival, So = natural survival in absence of hunting, H = harvest rate (including crippling loss)
and b < 1 in case of compensation. In fact, harvest compensation by survival remains seldom
detected. Thus, the effect of harvest is generally assumed additive (Conroy et al. 2002, Lebreton
2005).
We estimated compensation factor b using our estimates of survival and harvest rates for RCP in
the Camargue. We applied Devineau et al.’s (2006 & in prep.) allometric method to derive So, a
natural survival estimate for RCP in absence of hunting. These authors modeled adult annual
survival rates of many Anatidae species as a function of body mass and hunting intensity. Adult
survival in the absence of hunting was modeled as a function of log(body mass) for a large
sample of non hunted Anatidae species or populations. On the basis of the mean adult body mass
calculated from the marked RCP sample from Tour du Valat (943gr., SD = 105gr.), an estimate
of natural survival in the absence of hunting was computed for adult RCP (Devineau et al. in
prep.) as So = 0.753 (SE= 0.138). We assessed potential compensation effect from mortality with
the following inequality :
(1-S/So)/H < 1.
Compensation on reproductive rates
Kalchreuter (1996) hypothesized a possible compensation of harvest through reproductive
performance, although this density-dependent relationship has rarely been demonstrated
(Lebreton 2005, but see Conroy et al. 2002, Gunnarsson et al. 2006 and Zimpfer & Conroy 2006
13
on Anas density-dependent reproductive compensation and Blums et al. 2002a on harvestdependent recruitment in Aythya ferina). If there is density-dependent compensation in
reproductive output, then according to this hypothesis, there should be a positive relationship
between harvest rate and reproductive performance in the following breeding season, assuming
favourable breeding conditions (Kalchreuter 1996). Indeed, many stochastic environmental
factors can mask density-dependent effects on reproductive output (Gunnarsson et al. 2006).
Since reproductive performance is generally related to laying date (Johnson et al. 1992, Flint et
al. 2006), we assessed the effect of harvest rate from the preceding hunting season not only on
mean annual fledging success, but also on mean annual laying date of the subsequent breeding
season. Indeed, the latter trait might be less sensitive to stochastic events during the breeding
season but more sensitive to intraspecific density-dependent behavioural or social factors
(Johnson et al. 1992) like e.g. territorial competition for preferred nesting sites. We included
harvest rates into the precipitation models of mean annual fledging success presented above. For
the laying date – harvest rate relationship, we only incorporated sum of precipitations during the
preceding winter and early spring as an index of potential waterdepth. As for mean annual
fledging success models, we used a regression model and addressed the influence of covariates
on fledging success with an exhaustive model search approach based on AICc model
comparison. We only considered models with a maximum of three covariates in the model
selection process.
Impact of harvest on population growth rate
Demographic invariant method
Because of the scarcity of data for many vital rates of RCP, we used a recently developed
approach to evaluate the impact of harvest on RCP population growth rate. The demographic
invariant method (Niel and Lebreton 2005) provides an estimate of the maximum annual growth
rate (λmax) which allows for comparison between the potential excess growth and the estimated
total number of harvested RCP. The potential maximum harvestable population fraction allowed
by excess growth was estimated following Wade (1998) as P = Nβ (λ max − 1) , where N is the
total population size, currently estimated at 50000 individuals (Delany & Scott 2002) and β is a
correction factor accounting for the effect of density on demographic performance. β was set at
0.5 as recommended by Wade (1998) and Niel and Lebreton (2005). We estimated λmax
following Niel and Lebreton (2005) by solving numerically:
14
λmax=exp([α+So/(λmax-So)]-1)
where So is adult survival probability and α is the average age at first reproduction, both under
optimal growth condition. α was estimated at 1.3 year, assuming 70% birds first breed on their
first year (Blums et al. 1996) and the remaining on the next.
P can be interpreted as the maximum number of RCP that can be harvested by any non-natural
source of mortality, including hunting, without causing decline of the population. However, this
estimate of P only allows diagnostic of unsustainable harvest, but does not provide confirmation
of sustainable exploitation (Niel and Lebreton 2005).
Relationship between generation time and elasticity
The following relationship expresses change in survival (and ultimately in growth rate λ) as a
function of change in harvest rate: S=So(1-bH) (Gauthier & Lebreton 2004, Lebreton 2005).
This relationship takes advantage of the direct link between mean generation time T and
elasticity of λ to fecundity and adult survival (Lebreton & Clobert 1991). Assuming additive
mortality (i.e., b = 1) and equal harvest rates of juveniles and adults, the relationship given by
Gauthier & Lebreton (2004) between changes in harvest rate and in growth rate is the following:
Δλ =
− λ * ΔH
1− H
We used this approach to estimate changes in RCP population growth rate caused by three past
or potential changes in harvest regulation in France:
- wildfowl hunting banned in March from 1979 onward,
- wildfowl hunting banned between February 10 and 28 from 2000 onward,
- wildfowl hunting currently expected to be banned in the whole February.
To quantify harvest decrease due to hunting bans over some time period, we used the overall
distribution of RCP harvest summed over the sampled years, by period of ten days. We deduced
the proportion of banned harvest.
15
Sensitivity analyses
Although controversial (Mills et al. 1999), the comparison of sensitivities of population
asymptotic growth rate to variation in vital rates provides preliminary insight into relevance of
research or management actions; controversy can be avoided if conclusions are carefully
interpreted and recommendations take cost and realism of actions into account (Ferrière et al.
1996, Benton & Grant 1999, Heppell et al. 2000, Ehrlén et al. 2001, Fieberg & Ellner 2001, Link
& Doherty 2002). We performed a sensitivity analysis of the growth rate of the closed European
RCP population to changes in the lower-level demographic parameters. Because we wanted to
estimate sensitivities to variation in both survival and harvest rates, which are expected to have
comparable elasticities but different values, we used arcsin scaled variance-stabilized
sensitivities (Link and Doherty 2002). This allows changes in λ to be partially, yet largely
independent from parameter values. This approach is designed for demographic parameters that
are bounded by 0 and 1, which was the case in our matrix model with the exception of fledging
success, for which we assessed only regular elasticity. In the absence of data on European
harvest levels, we assumed that harvest-driven mortality at the European scale was additive to
natural mortality (i.e., b = 1). H was deduced from So and our dead-recovery estimate of adult
survival.
Parameter uncertainty analysis
Improving evaluation of the research needs on RCP demography in Europe was one of our main
objectives. We conducted a parameter uncertainty analysis based on the methodology
developped by Hunter et al. (2000) using our model for European RCP population. We started
with a female initial population size of 25000 structured according to the stable age distribution
predicted for European population. In order to evaluate research needs concerning every
parameter in a hypothetical population projection model that would be well suited for a
population thoroughly known (Table 2), we increased complexity of our model and incorporated
adult renesting propensity (q3) after clutch loss. After Hoekman et al. (2002), we used the
following formula:F2 = f*sr*p*q2*(1+(1-p)*q3)
16
Parameter
data source
imported or fixed value
adult survival
s2
estimated
See results
fledging success
f
estimated
See results
natural survival
so Devineau et al. in prep.
harvest rate
H
estimated
See results
first year survival
s1
estimated
See results
brood hatching success
p
Amat 1982
0.532
immigration rate
m
arbitrary
0
renesting propensity
q3
Flint et al. 2006
0
adult breeding propensity
q2 Anderson et al. 2001
first year breeding propensity
q1
sex-ratio at birth
sr Blums & Mednis 1996
Blums et al. 1996
0.753
1
0.7
0.5
Table 2: parameters used in RCP demography model
We used parameter values estimated using the above approaches, values from the litterature
when data from RCP were missing, or arbitrary values when nothing was known (e.g. sex ratio at
birth), and ran the model once to obtain a mean value of λ (λmean). Then we ran the model for
each parameter lower and upper limit successively, over 20 time steps. For each parameter, we
obtained a lower and an upper estimate of λ (more precisely, λmin and λmax). We then
computed the following uncertainty coefficient for each parameter: |λmax- λmin|/ λmean. This
coefficient was used as a basis for drawing recommendations for research priorities.
Comparison of population projection scenarii under environmental stochasticity
We built a stochastic matrix population projection model for the Camargue RCP subpopulation
to evaluate alternative scenarii in a variable environment context. To do so, we modeled survival
and fledging success as a function of meteorological covariates (Fieberg & Ellner 2001). This
model had the same stage structure as the deterministic model. For each scenario, we ran models
5000 times for 20 time-steps and recorded the mean stochastic growth rate (the mean of the
observed growth rates of all runs). Starting from the current stable-stage distribution of 560
brood-rearing females, we compared growth rates corresponding to different scenarii:
(i) - The deterministic version of the population projection model,
17
(ii) - The stochastic version of the model incorporating a relationship between meteorological
conditions (winter temperature, precipitation in the current winter-spring and the preceding
breeding season) and demographic parameters (survival, fledging success, respectively).
Meteorological data from 1945 to 2004 have been collected by Météo-France (French state
meteorological agency),
(iii) - The same stochastic model incorporating a harvest compensation process,
(iv) - The stochastic model incorporating a February hunting ban as a possible future harvest
regulation, either as a fixed decrease in harvest rate (set at the mean of the time series of change
in harvest rate) or as a random variable sampled over the same time series,
(v) - The stochastic model incorporating the effect of predicted local climate change by Arpège
climate model of Météo-France (Gibelin & Déqué 2003). Météo-France developed two scenarii
of strong and moderate climate change under the Arpège climate model. We assessed effect of
the most conservative one (i.e., with moderate climate change) on the stochastic growth rate of
the Camargue RCP subpopulation over the next 20 years.
RESULTS
Survival rates estimation
The lowest-AICc model included three covariates (Table 3): individual weight, age, and time lag
between banding and recovery period. Results were consistent with our a priori hypotheses
about the effect of these variables on survival probability (Table 4). Effects of these variables
were non significant but were used as the best predictive model to estimate survival over the
periods of interest.
18
Number of
estimated
Model
AICc
ΔAICc AICc Weights parameters Deviance
S(age,winter temperature,adult weight,i,.)r(.,.)
357.69
0
0,45
6
345.37
S(age,.)r(.,.)
358.91
1.22
0,24
3
352.82
S(age,adult weight,i,.)r(.,.)
359.74
2.05
0,16
5
349.51
S(age,i,.)r(.,.)
359.93
224
0,15
4
351.78
S(age,t)r(.,.)
468.24 110.55
0
55
329.32
S(age,t)r(age,.)
471.39 113.70
0
56
329.28
S(age,t)r(age,t)
674.76 317.07
0
108
311.61
Table 3. Model selection for age-dependent RCP apparent/local survival from the Camargue,
France, 1952 – 1979.
Results provided evidence that mean winter temperature and, for adults, individual weight
positively influenced RCP survival but there remains substantial uncertainty concerning this
influence, as shown by the relatively large estimated standard errors of the parameter estimates.
With this limitation in mind, using a mean adult weight of 943g and a theoretical time-increment
of zero, we then obtained predicted annual adult and juvenile RCP survival probability. Using an
environmental covariate (here mean winter temperature), we obtained model-based estimates of
survival probability not only for the 1952- 1978 period, but also for the more recent period
(following an approach advocated by Fieberg & Ellner 2001). For the sake of simplicity, and
also because of the heterogeneity of our various data sets on RCP, we chose not to use individual
survival probability predicted from individual weight. Rather, we relied on a mean value at the
population level. The former approach may be useful in detailed demographic studies where data
from individual heterogeneity in life history traits are available (Cam et al. 2002).
Model stucture
Estimate
SE
95% CI
S Intercept
-8.449
3.772
-15,841
-1,056
Temperature
0.613
0.332
-0,038
1,263
adult weight
0.004
0.003
-0,001
0,009
time-increment i
-0.006
0.080
-0,163
0,151
S first year
4.877
2.416
0,142
9,612
r Intercept
-1.703
0.170
-2,037
-1,369
Table 4 Slope parameter estimates for the model with the largest support for RCP local survival
from the Camargue, France, 1952 - 1979
19
Estimates of mean survival rates for adults and juveniles, respectively, were the the following for
the 1952-1978 period (SD calculated over the interannual predictions from mean winter
temperature during this period): Sad=0.546 (0.125), Sjuv=0.758 (0.113). Juvenile survival
appeared surprisingly higher than adult’s, which is reverse to the general pattern of other
Anatidae species (Johnson et al. 1992).
Fecundity estimation
The goodness-of-fit test indicated that the multistate model for brood survival fitted the data
satisfactorily for year 1990 (χ² = 0.41, df = 2, P = 0.82), 1991 (χ² = 1.33, df = 4, P = 0.86) and
2001 (χ² = 1.13, df = 5, P = 0.95). GOF test statistics could not be estimated for year 2000 due to
absence of state 2 (i.e. no observed loss of one duckling). However, goodness-of-fit test for
pooled groups of years 2000 and 2001 indicated satisfactory fit (χ² = 1.07, df = 5, P = 0.96). This
pooling of years 2000 and 2001 corresponded to one of our starting a priori hypotheses.
20
Nunber of
estimated
Model
Sb(da,pu,pu²,tlin,.,.),p(.,G,.),ψ(da,pu,pu²,.,.,str)
AICc
ΔAICc AICc Weights parameters
Deviance
189.33
0.00
0.90
11
164.33
Sb(da,da²,pu,pu²,tlin,.,.),p(.,G,.),ψ(da,da²,pu,pu²,.,.,str) 193.97
4.64
0.09
13
163.74
Sb(pu,pu²,tlin,.,.),p(.,G,.),ψ(pu,pu²,.,.,str)
198.88
9.54
0.01
9
178.88
Sb(tlin,.,.),p(.,G,.),ψ(.,.,str)
201.87 12.54
0.00
5
191.23
Sb(tlin,G,.),p(.,G,.),ψ(.,.,str)
203.55 14.22
0.00
6
190.65
Sb(tlin,G,.),p(.,G,.),ψ(tlin,.,str)
203.65 14.32
0.00
7
188.43
Sb(.,G,.),p(.,G,.),ψ(.,.,str)
203.96 14.62
0.00
5
193.32
Sb(.,G,.),p(.,G,.),ψ(tlin,.,str)
204.00 14.67
0.00
6
191.10
Sb(pu,tlin,.,.),p(.,G,.),ψ(pu,.,.,str)
204.91 15.58
0.00
7
189.69
Sb(.,G,str),p(.,G,.),ψ(.,.,str)
207.52 18.19
0.00
7
192.30
Sb(.,G,str),p(.,G,.),ψ(.,G,str)
209.75 20.42
0.00
8
192.17
Sb(.,g,str),p(.,G,.),ψ(.,g,str)
222.29 32.96
0.00
14
189.35
Sb(.,g,str),p(.,g,.),ψ(.,g,str)
225.56 36.23
0.00
16
187.01
Sb(.,g,str),p(.,.,.),ψ(.,g,str)
232.70 43.37
0.00
13
202.47
Sb(.,g,str),p(.,g,str),ψ(.,g,str)
23656
47.22
0.00
20
185.92
Sb(t,g,str),p(t,g,str),ψ(t,g,str)
295,53 106,20
0,00
40
159,94
Table 5 : multistate models for biweekly brood survival with possibility for brood size to stay
constant (stratum 1) or to decrease by one duckling (stratum 2), in the Camargue, France, in
1990, 1991, 2000 and 2001.
Model notation: Sb = bi-weekly brood survival, p = brood resighting probability, ψ = brood
transition probability from size n to n-1, g = year (1990, 1991, 2000,2001), G = year group
(1990&1991 vs 2000&2001), str = stratum, t = time, tlin = linear time, da = laying date, pu =
brood size
One model for brood survival received unambiguously a larger support than all the others, with
an AICc weigth=0.90 (Table 5). As suspected, detection probability differed between earlier
(1990-1991) and more recent (2000-2001) monitoring periods. Brood and duckling survival did
not vary with year. We found evidence that brood survival decreased linearly with time within
season. Both brood and duckling survival varied with laying date, as well as brood size, and in a
similar way. Again, there remains substantial uncertainty concerning these relationships, as
shown by the relatively large SE of the parameter estimates (Table 6). Both brood and duckling
survival (1- ψ12) increased with brood size (pu) and late laying date (da), which is reverse to a
general (but not universal) pattern in ducks (Johnson et al. 1992).
21
Estimate
SE
Sb intercept
7.53
4.53
-1.34
16.40
Sb:pu
-0.12
0.43
-0.96
0.72
Sb:pu²
1.08
0.62
-0.14
2.29
Sb:da
1.80
0.94
-0.04
3.63
Sb: tlin
-1.96
1.09
-4.11
0.18
p intercept
-1.09
0.35
-1.78
-0.40
p: G
1.65
0.67
0.34
2.95
Psi intercept
0.00
0.82
-1.61
1.62
ψ12: pu
-2.67
2.22
-7.02
1.67
ψ12: pu²
-4.13
2.19
-8.43
0.17
ψ12: da
-1.85
1.12
-4.05
0.34
ψ21
95% CI
fixed to zero
Table 6 : parameter estimates of RCP brood survival from the Camargue, France, 1990, 1991,
2000 and 2001. Parameters estimated under the multistate model with the largest support.
We then simulated brood survival and size at fledging 1000 times for each of the 365 RCP
broods sampled from 1988 to 2005. This provided an annual estimate of fledging success. Mean
fledging success for this period was 2.99 (SD = 0.19). Furthermore, fledging success varied
according to precipitation. Fledging success was positively influenced by the sum of
precipitation from September to July of the breeding season (slope parameter estimate =0.0027,
F=14.27, p=0.002, df=1) and negatively influenced by the sum of precipitation from March to
June of the preceding year (slope parameter=-0.0084, F=11.74, p=0.004, df=1). Thanks to
models of fledging success as a function of precipitation, it was possible to predict fledging
success (and thus fecundity) over longer periods of time from which data from reproductive
success were not available.
Covariation between vital rates
Covariation between demographic parameters may be a problem in matrix population models
(van Tienderen 1995), especially if traits are involved in trade-offs or vary according to the same
environmental variables (Hoekman et al. 2002). We were obviously concerned with the latter
issue (because we intended to use stochastic versions of the population projection model based
on meteorological time series; see below). Consequently, we checked Pearson correlation
22
coefficients between fecundity and annual survival preceding, encompassing and following
breeding season over the n=58 years for which available meteorological data were available.
Correlation coefficients were clearly small (Table 7), which provided evidence of independence
of the estimated demographic time series.
r
p-value
survival following breeding
-0.12
0.36
survival encompassing breeding
0.08
0.53
survival preceeding breeding
-0.15
0.26
Table 7 : Pearson correlation coefficients between fecundity and survival as estimated under
models including meteorological covariates (n=58).
Model validation
Camargue subpopulation
Using estimated or mean vital rates from the literature, we ran the two-stage matrix model for
108 time steps to simulate numbers in the RCP Camargue breeding population between 1894 and
2001. We set the 1894 settler population in the Camargue at 3 adult females, assuming a
detection probability similar to that of year 2001, and a brood hatching success of 53.2%, as
estimated by Amat (1982) for RCP in South-Western Spain.
Using mean vital rate values, the matrix population model was:
⎡0.4219 0.4341⎤
⎢0.7580 0.5460⎥
⎣
⎦
The projected number of females RCP using the model was equal to 1656 for the Camargue in
2001. The corresponding asymptotic population growth rate was λ=1.06 (SE=0.21).
Using the adult proportion given by the stable stage distribution of 59.55% and a brood hatching
success of 53.2% (Amat 1982), we obtained an estimated brood abundance of 525 for 2001 in
the Camargue. This result was in agreement with the corresponding field survey estimate of 560
broods by Defos du Rau et al. (2003). We also set immigration rate arbitrarily at 0.001 and
23
obtained a brood population size estimate of 586, which was also very close to the actual
estimated breeding females abundance.
Western and Central European population
To further assess our matrix model projection performance, we compared projected population
size to published Western and Central European RCP population estimates over a period of
comprehensive, regular data collection (Table 1). We ran the same matrix model as for the
Camargue subpopulation except that we initiated it with a population size of 10000 females and
used climate-based predictions of survival rates averaged over the corresponding decades, i.e.
1980-2002 (SD calculated over the interannual predictions from mean winter temperature during
this period): Sad= 0.583 (0.139), Sjuv= 0.784 (0.101).
30000
RCP abundance
25000
20000
15000
10000
5000
0
198694
1995
1996
1997
1998
1999
2000
2001
2002
Figure 3 : Central and Western European population trends of RCP projected by the two-stage
matrix model (O) and published estimates based upon mid-winter International Waterbird
Censuses coordinated by Wetlands International (z)
24
As for the Camargue subpopulation, at the European population scale, projected numbers were
close to the corresponding survey-based estimates (Figure 3). Both simulated and observed
demographic trends suggested a marked increase of RCP abundance in Europe over this period:
λ=1.12 (SE=0.22). However, the increase rate was poorly estimated (poor precision on λ).
Compensation of harvest
From 1988 to 2005, the estimated harvest rate of RCP averaged 17.4% with strong interannual
variations (SD = 13.8%).
Compensation on mortality
Using average adult survival and harvest rates from 1988 to 2005, the estimated compensation
factor b = 1.080 suggested additive effect of harvest on RCP natural mortality in the Camargue.
We thereafter approximateed survival rate S as follows: (1-So)H.
Compensation on reproductive rates
As expected, estimated fledging success was negatively influenced by laying date (r = -0.424,
F=78.59, p<0.001, df=1, n = 361). We log-transformed annual harvest rates because of
unfavourable ratio of mean to variance. The model including log(harvest rate) received support
(Table 8, Figure 4) whereas models of precipitation did not (Table 8).
Number of estimated parameters
AICc
ΔAICc
ln(harvest rate)
2
91.72
0.00
ln(harvest rate) sept-jan precip.
3
93.89
2.17
ln(harvest rate) sept-feb precip.
3
94.14
2.43
ln(harvest rate) sept-mar precip.
3
94.15
2.44
ln(harvest rate) sept-apr precip.
3
94.42
2.70
sept-jan precip.
2
113.36 21.64
sept-mar precip.
2
113.82 22.10
sept-feb precip.
2
113.91 22.19
sept-apr precip.
2
114.00 22.29
Table 8 : regression models for mean annual laying date of RCP in the Camargue as a function
of log(harvest rate) and sum of precipitation from preceding winter, from 1988 to 2005.
25
70
mean laying date
60
50
40
30
20
10
0
0
0.1
0.2
0.3
0.4
0.5
0.6
harvest rate
Figure 4 : mean annual laying date of RCP in the Camargue from 1988 to 2005 as a function of
annual harvest rates estimates. Regression line: mean laying date as a linear function of harvest
rate logarithm (r² = 0.763).
The relationship between harvest rate and mean laying date (r² = 0.763, Figure 4) provided
evidence of density-dependent reproductive compensation of harvest in the Camargue population
of RCP. We further addressed compensation in reproductive parameters by addressing effect of
harvest rate on mean annual fledging success, which we had found to be a function of
precipitation sums over the preceeding months. After checking that harvest rate was not colinear
with any precipitation covariate, we compared regression models of fledging success using the
exact same precipitation covariates and logarithm of annual harvest rate (Table 9).
26
Number of
estimated
parameters
sept-jul precip.
mar-jun precip. Year-1
ln(harvest rate)
sept-jul precip.
mar-jun precip. Year-1
sept-jul precip.
ln(harvest rate)
sept-jul precip.
mar-jun precip. Year-1
ln(harvest rate)
mar-jun precip. Year-1
ln(harvest rate)
AICc ΔAICc
3
27.40
0.00
4
28.68
1.28
2
33.38
5.98
3
34.59
7.19
2
35.22
7.81
3
36.41
9.01
2
41.55 14.15
Table 9 : regression models for mean annual fledging success of RCP in the Camargue, from
1988 to 2005 as function of log(harvest rate) and sum of precipitation from March to June of the
preceding year and from September to July of the current breeding season.
Compared to environmental covariates, harvest rate had only a marginal influence on fledging
success. Nevertheless, it was included in a model with substantial support (ΔAICc=1.28, r² =
0.737) in which its effect on fledging success was positive, although non significant (slope =
0.19 95%CI = [-0.11,0.50]). Overall, harvest rate appeared to be compensated for by
reproductive output, but its influence was more pronouced on laying date than in fledging
success. The latter was also dependent on other environmental factors, which may mask harvest
effect.
We did not find evidence of increased harvest when breeding season was successful (large
number of duckling fledged; r = 0.256, F =0 .98, p = 0.339, df = 1, n = 16), which would lead to
a large proportion of vulnerable ducks in early hunting season because of inexperience (Clarck et
al. 1988). Similarly, we did not find evidence of an influence of laying date on harvest rate (r = 0.404, F = 2.72, p = 0.121, df = 1, n = 16); late laying date would also lead to increased
proportion of vulnerable ducks in early hunting season because of poor condition of adults.
Impact of harvest on population growth rate
Demographic invariant method
We estimated λmax = 1.54, which yield in turn an estimate of maximum harvestable population
fraction P = 13500. From 1988 to 2005, the estimated total harvest of RCP in the Camargue was
27
on average 807 birds, with strong interannual variation (SD = 422). Assuming a similar change
in harvest bags in France and Spain compared to the last bag evaluation in the mid-80’s
(Shedden 1986), the total current European RCP bag is approximately 9223 birds. This value
represents 68% of the maximum sustainable harvestable population fraction estimated using the
Demographic invariant method. However, our reasoning is based on assumed change in harvest
bags in Spain, an assumption that may not be realistic.
Relationship between generation time and elasticity
This approach aimed at evaluating impact of harvest on population growth rate in the Camargue
(France). It assumed additive mortality (as shown above). We therefore used the Camargue twostage matrix model with λ=1.060. Corresponding mean generation time was computed using the
ULM software: T = 2.241. We used the previously estimated Camargue harvest rate of 17.4%
and plotted Δλ as a linear function of small to moderate values of ΔH (Figure 5), using the
following equation: Δλ = -λΔH/(1-H). In the same graph, we plotted estimated changes in
harvest rate and corresponding changes in λ following hunting bans in March, 10-28 February
and 1-10 February.
28
corresponding change in lambda
0.14
0.12
0.1
0.08
0.06
0.04
March ban
0.02
0
-0.12
10-28 Feb ban
1-10 Feb ban
-0.10
-0.08
-0.06
-0.04
-0.02
0.00
change in harvest rate
Figure 5 : Change in growth rate of Camargue RCP subpopulation as a function of change in
harvest rate regulation. Predicted impact of 2 past hunting bans (March and 10-28 February) and
one likely further ban (1-10 February) on Camargue subpopulation growth rate.
The hunting ban on March appeared to have been the most effective in increasing the French
RCP subpopulation growth rate: change in λ was Δλm=0.0408. In contrast, a hunting ban from 1
to 10 February would have relatively little influence on population growth rate.
To check this, we re-estimated survival probability in the Camargue after modifying artificially
the data set: we replaced March hunting recoveries by unrecovered histories and reran the deadrecovery capture-recapture model . We obtained a larger estimated mean survival rate Sad =
0.584 (compared to the estimate obtained using the original data set). We interpreted this value
as the hypothetical survival probability under March hunting ban. We used the two survival
probabilities successively in the population projection matrix model. We projected the
population over a 20 year time interval, starting from the Camargue population size previously
obtained for year 1979 using the unmodified population matrix model (njuv=172, nad=253).
Under liberal harvest (hunting allowed in March), λl=1.0609, whereas under restrictive harvest
(hunting allowed in March), λr=1.1017. Interestingly, the difference between the growth rates λr
29
and λl obtained using estimates of survival probability based on dead recovery data matched
exactly to the change in growth rate that would be predicted if we removed the March harvest
fraction from the harvest survey data set: λr - λl = 0.0408 = Δλm .
We believe that our evaluation of growth rate change due to harvest ban in March is robust: we
obtained the same result using two independent data sets.
Sensitivity analyses
Sensitivities calculated on the arcsine scale gave some results different from classical
sensitivities, notably for harvest and first-year survival rates (Table 10). However, both
sensitivities rated sex-ratio at birth and brood hatching success as influential parameters, of
course within the range of their assumed values. Adult survival had the highest rating for both
scales of sensitivities, which is consistent with what we expected .
parameter
sensitivity elasticity sensitivity (arcsin scale) origin of data management opportunities
s2 adult survival
1.074
0.560
0.474
estimated
sr sex-ratio at birth
0.982
0.440
0.440
arbitrary
p brood hatching success
0.923
0.440
0.412
imported
X
f
fledging success
0.164
0.440
estimated
X
so natural survival rate
0.831
0.560
0.321
estimated
H harvest rate
-0.808
-0.163
-0.303
arbitrary
s1 first-year survival
0.626
0.440
0.231
estimated
m1 first-year fecundity
0.345
0.172
0.153
composite
m2 adult fecundity
0.376
0.268
0.136
composite
q1 first-year breeding %
0.274
0.172
0.112
imported
X
Table 10: Sensitivity of growth rate of RCP European population to changes in demographic
parameters. Potentialities for data improvement or management actions are suggested.
Parameter uncertainty analysis
Parameters SE were either estimated or came from the literature (as indicated in Table 11). In
most cases 95% Confidence Interval boundaries were used as parameter lower and upper
uncertainty limits. A few limits were changed to more realistic values when 95%CI were
unrealistic (e.g. survival rate upper boundary > 1 or harvest rate lower boundary < 0). When
30
confidence intervals were not available, such as for immigration probability, we set an arbitrary
upper limit to 20%.
lower upper
estimate SE
limit
limit
sources of limit
parameter uncertainty
estimates
coefficient
adult survival
s2 0.583 0.139 0.311 0.855
estimated
0.525
fledging success
f
estimated
0.460
natural survival
so 0.753 0.138 0.482 0.990 Devineau et al. in prep.
0.379
harvest rate
H
0,.258 0.138 0.000 0.497
estimated
0.361
first year survival
s1 0.784 0.101 0.586 0.982
estimated
0.222
brood hatching success
p
Amat 1982
0.204
immigration rate
m
0
0.000 0.200
arbitrary
0.190
renesting propension
q3
0
0.078 0.000 0.665
Flint et al. 2006
0.078
adult breeding propension
q2
1
0.740 1.000
Anderson et al. 2001
0.074
Blums et al. 1996
0.044
2.989 0.797 1.427 4.551
0.532 0.063 0.408 0.656
first year breeding propension q1
0.7
0.045 0.612 0.788
sex-ratio at birth
0.5
0.010 0.480 0.520 Blums & Mednis 1996
sr
0.035
Table 11 : Parameter uncertainty analysis of RCP European population, including parameters of
unknown importance and not included in the original matrix models. Most lower and upper
limits are CI limits of the corresponding parameter.
Based on uncertainty coefficients, improving estimation of the following parameters appears a
research priority : adult survival probability, fledging success, survival probability in the absence
of hunting and harvest level. This is consistent with the large SE estimates of population growth
rate computed from our matrix models at the Camargue and European scales.
Comparison of scenarii under environmental stochasticity
We used three scenarii of RCP demography in the Camargue
(i) According to previous results, fledging success was found to compensate for harvest mortality
to some extent. Consequently we modelled fledging success as a linear function of the logarithm
of the harvest rate and precipitation covariates (i.e., based on the model with ΔAICc=1.28 in
Table 9).
31
(ii) Under conservative scenarii of the Météo-France Arpège climate model, from 1987 to 2027
there was no trend in precipitation covariates used in models of fledging success either in mean
or in SD values. Conversely, mean winter temperature, a covariate in models of survival
probability, showed a significant increase over the period 1987-2027 ( r²=0,237, p=0,0015,
n=40). Consequently we compared stochastic growth rate of RCP Camargue subpopulation
under current vs predicted increased mean winter temperature. Mean winter temperature values
were drawn at random from a normal distribution parameterized according to the current or
predicted mean winter temperature distributions.
(iii) We calculated annual change (ΔH) in harvest rate by removing RCP killed between
February 1 and 10 from total harvest. The observed ΔH approximately followed a gamma
distribution. We therefore modelled ΔH as drawn at random from a gamma distribution of shape
parameter equal to the observed mean of the harvest change time-serie (0.39%) and a scale
parameter set to 1.
We also considered a scenario where annual ΔH under ban from February 1 to 10 would be
equal to its observed average (0.39%): the corresponding hunting ban would cause a fixed annual
harvest decrease.
Last, we incorporated the following relationship in the matrix model under the expression: ΔS =
-So* ΔH (following Gauthier & Lebreton 2004). ΔS was added to both first-year and adult
survival rates when considering the hunting ban scenario.
Mean stochastic growth rates of the 5 scenarii are shown in Figure 6. We stress the fact that we
were only interested in the comparison between the resulting growth rates and not in the growth
rate estimates themselves (Fieberg & Ellner 2001). The impact of short-term climate change
appeared to affect RCP growth rate in the Camargue in a much stronger way than a hunting ban
in the first ten days of February. Such a hunting ban would have only a marginal impact on RCP
future demography, compared to future climate changes in the Camargue modelled by MétéoFrance France (Arpège climate model; Gibelin & Déqué 2003).
32
mean stochastic growth rate
E
A
B
C
D
Figure 6: stochastic growth rates (averaged over 5000 runs) and their 95%CI for RCP Camargue
subpopulation for different demographic and environmental scenarii over the next 20 years.
Values in Y-axis are not shown : we focused on comparisons of growth rate estimates rather than
on the estimates themselves (see methods). X-axis at λ=1.
A: basic stochastic scenario
B: stochastic scenario with random effect of hunting ban from February 1 to 10
C: stochastic scenario with mean effect of hunting ban from February 1 to 10
D: stochastic scenario with harvest compensation by fecundity
E: stochastic scenario under predicted climate change from 1987 to 2027
DISCUSSION
As a rare game duck, RCP is of strong research and conservation concern. The main objectives
of this study were to increase our understanding of RCP dynamics under exploitation, to assess
the sustainability of RCP hunting, to infere management, and most importantly, research
recommendations, and to make short-term projections of RCP population dynamics while
accounting for environmental covariates.
33
(i) - Are there compensatory mechanisms to harvesting ? As in many other modelling studies of
wildfowl exploitation (Lebreton 2005), we did not find evidence of compensatory mortality.
However, as seems to be less commonly the case (Lebreton 2005), we found substantial support
for the compensatory reproduction hypothesis (Kalchreuter 1996) through a functional link with
laying date. Blums et al. (2002a) formulated a similar hypothesis of reduced competition for
breeding territories following high harvest levels, and indeed they found a similar pattern in the
related species Aythya ferina. However, the indirect influence of harvest level on reproductive
output remains unclear; because of the low precision of the slope parameter and the relatively
weak effect of harvest rate on fecundity, we chose not to incorporate this density-dependent
feedback in the matrix population model. Incorporating potentially spurious density-dependence
might have had strong impact on model outcomes (Reed et al. 2002).
- (ii) - What is the relationship between harvest levels and RCP population dynamics? Because
of large sampling variance of our vital rate estimates, RCP population growth rate estimates were
imprecise at both European and Camargue scales. In spite of relatively high values of λ, we were
not able to confirm the general tendency of strong RCP population increase: both 95%CI of
growth rates included 1. Similarly, the fact that estimated European harvest levels were well
below the predicted maximum sustainable harvest level does not mean that current RCP hunting
is sustainable (Niel & Lebreton 2005), all the more as long as up-to-date harvest data are lacking
for Europe. Unless the consistency between the projected numbers from the population matrix
model and the observed census data at both scales (Figure 3) is a coincidence, such a consistency
showed that our matrix model and initial population values were reasonable and realistic.
Assuming that our model is realistic, results of the modelling exercise supported the hypothesis
of a general increase in RCP populations. Hunting ban in March had probably contributed to this
upward trend through its relatively strong (4%) impact on growth rate. Thus, there is currently no
evidence of unsustainable exploitation of RCP in Europe, although this remains to be confirmed
using more precise parameter estimates notably through implementation of the following
recommendations.
- (iii) - Which recommendations are prioritized by perturbation analyses? Information from
sensitivity analysis and management opportunities suggests that breeding habitat management
may be the best management option to increase RCP population growth rate. In addition, as
expected for a mortality-related factor, results of sensitivity analyses provided evidence that
34
management of harvest levels should influence population growth rate. This result was more
pronounced in arcsine-scaled sensitivity analysis (Link & Doherty 2002) than in traditional
sensitivity analysis. However, results of any class of sensitivity analysis can be used to design
reliable management actions on condition that parameters are estimated with sufficient precision.
The hypothetical influence of variation in demographic parameters on population growth rate
may also reflect the consequences of poor knowledge of the location of such parameters and
range of variation. Thus, given the heterogeneity of data sources we used, we suggest that
parameter uncertainty analyses provided more relevant results than sensitivity analyses :
estimation of adult survival, fledging success, natural survival in the absence of hunting and
harvest rate are research priorities.
We advocate the development of two research projects focusing on demography of harvested
RCP: (i) implementation of a nasal mark-recapture survey to increase precision on survival rates
and harvest rates. (ii) Development of monitoring field surveys to increase precision on
reproductive vital and harvest rates. The latter are particularly relevant to proper RCP population
dynamic modelling. Indeed, our results suggested that harvesting influences survival in an
additive way, and also reproduction in a density-dependent way.
In this context, allometric analyses did not provide enough precision on estimate of natural
survival in the absence of hunting. The predictive performance of allometric models needs to be
improved. In addition, in conjunction with a mark-recapture design, a temporary or spatial partial
experimental ban on hunting would contribute to estimate natural survival in the absence of
hunting. Such large-scale experiment, inspired from the North American Adaptive Management
of waterfowl, would be of great value to reduce uncertainty in our understanding of RCP
demography.
- (iv) - What are the projections from different parameterizations of the population matrix
model? Assuming a persistent relationship between survival rate and mean winter temperature,
the main conclusion from our stochastic modelling analysis was that a scenario of moderate
climate change might overcome any change in management of RCP, including hunting
regulation. That is, such regulation might be useless in a context of temperature rise. However,
readers should keep in mind that the differences among growth rates under various short-term
simulations documented here are only of qualitative interest. For example, it is unlikely that the
relationship between survival and mean winter temperature remains unchanged across a wide
35
range of increased temperature values. Consequently, the quantitative difference between the
climate change scenario and the other scenarii is unlikely to remain unchanged.
Importance of detection issues in census dataset
At the Camargue subpopulation scale, we used breeding pair census to validate results from
population projection models (i.e., to assess the degree of agreement between projected numbers
of RCP and numbers estimated from field data). We used an approach to estimation of female
breeding population size correcting observed breeding pair numbers for detection probability
(Williams et al. 2002). Accounting for imperfect detection of individuals or broods may be
particularly important in species like RCP, where broods have cryptic habits (Defos du Rau et al.
2003). It is interesting to note that the discrepancy between projected and observed population
size would have been equal to 103% if we had used the census data set uncorrected for
detectability (2 females in 1893 and 90 in 1990 (Dehorter & Rocamora 1999)). The discrepancy
was of 6% with census data adjusted for detectability. Would we have used uncorrected census
data, we would have concluded that our matrix model was not reasonable and we would have
modified it to match biased observed breeding numbers. Using estimates of breeding female
population size made with models accounting for detectability also provided a more realistic
initial population size. This is fundamental to demographic modelling. Indeed, PVA models are
highly sensitive to errors in census estimation (McLoughlin & Messier 2004).
At the European scale, we did not validate the population projection model using census but
published population estimates. It is however likely that census of this gregarious and
conspicuous bird in winter suffer from much less detection biases than breeding bird census.
Limitations of our approach
We followed several recommendations put forward by Fieberg & Ellner (2001) and Reed et al.
(2002): parametric matrix as opposed to random transition matrix, explicit relationship between
variation in vital rates and variation of environmental covariates, explicit relationship between
these covariates and management factors (e.g., harvest rate), assessment of the hierarchy in
projected breeding numbers under different scenarii rather than comparison of numbers.
36
Most importantly, we validated our model using two independant data sets (the Camargue and
Western European ones). However, our estimates of some demographic parameters had large
standard errors (i.e., adult or brood survival), which resulted in large imprecision on growth rates
estimates.
Furthermore, the capture-recapture data from banded RCP have been collected under a design
that did not conform to basic recommendations concerning length of capture sessions and time
intervals between sampling occasions. RCPs were captured and marked during several months
each year, and were recovered during several months each year. Consequently, if two individuals
were banded in year i and captured in year i + 2 for example, there may be a large difference
between the number of months the two individuals survived (several months). We tried to
overcome this problem using a time increment between realized and theoretical recapture (i.e.,
the occasion at which the individual would have been recovered if recoveries had occurred at the
same fixed discrete time intervals than capture over the entire study). We used this increment as
an individual covariate but the possible biases of this approach have to be thoroughly addressed.
Some other parameters used in the population projection model had a relatively weak basis,
mostly because their sampling in the field is difficult.
In addition, we assumed maximal breeding propensity in adults, no renesting and no
immigration. None of these assumptions were addressed using empirical data; whether this is
feasible is unclear for the Camargue and European scale. However, we made this null
immigration assumption because breeding habitats in the Camargue had been found overfilled by
RCP (Defos du Rau et al. unpublished/in prep.), which implies that installation of new settlers is
unlikely. Camargue RCP subpopulation was thus assumed to be open to emigration rather than
immigration. Last, the consequences of uncertainty in these parameters was evaluated and found
of lesser importance than uncertainty on other parameters (e.g., survival probability).
Particularities of RCP demography
In this study, we estimated several demographic parameters of RCP for the first time. Several
traits were consistent with patterns generally found in wildfowl, but some other traits were
original:
37
- Higher survival in juveniles than in adults. Contrary to a very general pattern in wildfowl
(Johnson et al 1992, Anderson et al. 2001, Arnold et al. 2002, Lake et al. 2006), our estimate of
survival probability was higher in first-year RCP than in adults. This does not support the
hypothesis of larger dispersal probability of first-year compared to adult wildfowl (Batt et al.
1992), unless this dispersal pattern is reinterpreted as a temporary emigration out from hunted
areas. RCP is hunting over a restricted part of his wintering range. It is thus possible for RCP to
winter in hunting-free areas. If juvenile emigrates temporarily during hunting season but show
natal phylopatry, they should then avoid harvest mortality and thus have increased survival.
Survival estimates can be biased by temporary emigration (Arnold et al. 2002) and by
resident/non-resident status (Kokko & Lundberg 2001). Higher survival in juvenile than in adult
could also be explained by lower natal phylopatry in adult than in juvenile, e.g. due to nomadic
tendancy in adults RCP (see below). Such a higher juvenile survival might be an explanation for
the successfull brood parasitism strategy particularly developed in RCP (Amat 1991). Almaraz &
Amat (2004) indicated an apparent high juvenile survival rate in Spanish white-headed duck
Oxyura leucocephala.
-Density-dependent reproduction. Apparent absence of harvest compensation on mortality but
evidence of compensation on fecundity is not a common pattern among well studied ducks
(Lebreton 2005). However similar regulation processes have been documented in Pochard
Aythyta ferina (Blums et al. 2002a), Mallard Anas platyrhynchos (Gunnarsson 2006) and Black
Duck Anas rubripes (Zimpfer & Conroy 2006). Such a process has been suggested in whiteheaded duck Oxyura leucocephala (Almaraz & Amat 2004). Such regulation might be
particularly adapted in case of strong environmental fluctuations of breeding conditions, when
poor breeding conditions and thus poor reproduction in year t can favour breeding performance
in year t+1 because of relatively lower density of breeding birds. Among the 25 RCP recovered
in the Camargue during the hunting season, 64% had been ringed in the Camargue during the
breeding season, which shows that at least some RCP harvested in the Camargue are local
breeders. This result supports the hypothesis that harvested and breeding fractions overlap within
the Camargue subpopulation, a condition to meet for a functional regulation between density of
breeders and breeding success.
38
- Temperature-dependent survival. Blums et al. (2002b) found evidence of a similar positive
relationship between survival probability and mean winter temperature for the related Pochard
Aythya ferina wintering in Western Europe. However, Blums et al. (2002b) interpreted this
finding by the detrimental effect of cold on survival through increasingly unfavourable trade-offs
between metabolic needs and foraging performance and physical condition. This explanation
may hold in RCP as well, but we favour another hypothesis: an increased harvest level at low
temperature, when metabolic needs force ducks to spend more time foraging during daytime in
hunting areas. Indeed, generally, among European wintering areas the Camargue is not affected
by long period of harsh weather; consequently, we believe that the direct physiological effect of
weather conditions on ducks wintering in the Camargue is small.
- Brood and duckling survival increased non-linearly with brood size. This relationship is not
universal in waterfowl (Johnson et al. 1992, Blums et al. 2002a)
- Brood and duckling survival increased with laying date. This pattern is variable among
waterfowl (Johnson et al. 1992). However, in general, survival decreases with laying date (e.g.
Blums et al. 2002a). Dzus & Clark (1998), Gendron & Clark (2002) and Schmidt et al. (2006)
hypothesized that the decrease in duckling survival with laying date was conditional on habitat
or weather conditions. Such an indirect influence of habitat on brood and duckling survival may
hold in RCP because there are reasons to believe that wetland availability in the Camargue is
more favourable later in the breeding season. This results from hunting management practices
favouring summer vs. spring flooding for instance. This hypothesis obviously deserves further
confirmation upon larger samples.
Perspectives
A relatively low adult survival probability, a demographic potential for adaptation to breeding
conditions through density-dependence and increased duckling survival with brood size and
laying date may be tentatively interpreted as life-history traits characteristic of nomadic, sarmatic
wildfowl. This sarmatic wildfowl guild also includes the white-headed duck Oxyura
leucocephala (Almaraz & Amat 2004) which apparently shares a common feature with RCP : a
high juvenile survival probability. This guild may be adapted to the less predictable flooding
conditions prevailing in Central Asia and Mediterranean lagoons than in the well studied, more
39
predictable boreal wetlands. In unpredictable breeding habitat, condition-dependent habitat
selection tactics based on breeding success might translate into high probability of dispersing in
adults (which would translate into relatively low adult local survival probability). Such tactics
may allow maximization of opportunistic breeding success in favourable breeding conditions.
Intensively managed, wetland habitats in Western Mediterranean Europe making them
predictable may thus provide permanently favourable breeding conditions for RCP, which may
allow managers to artificially reach satisfactory population growth rate (e.g., at least population
persistence), even under moderate to strong harvest level. Such intensive management, however,
could be detrimental to biodiversity of Mediterranean wetland ecosystems (Tamisier & Grillas
1994, Defos du Rau et al. 2005)
References
Almaraz, P. & Amat, J.A. 2004. Multi-annual spatial and numeric dynamics of the white-headed
duck Oxyura leucocephala in southern Europe: seasonality, density dependence and climatic
variability. Journal of Animal Ecology 73 (6), 1013–1023.
Amat J.A. 1982. The nesting biology of ducks in the Marismas of the Guadalquivir, SouthWestern Spain. Wildfowl, 33: 94-104.
Amat, J. 1991. Effects of Red-crested Pochard nest parasitism on Mallards. The Wilson bulletin
vol. 103, no3, pp. 501-503
Anderson, D. R., and K. P. Burnham. 1976. Population ecology of the mallard. VI. The effects of
exploitation on survival. U.S. Fish and Wildlife Service Resource Publication 128.
Anderson, D. R., W. A. Link, D. H. Johnson, and K. P. Burnham. 2001. Suggestions for
presenting the results of data analyses. Journal of Wildlife Management 65:373-378.
Anderson, M.G., Lindberg, M.S. & Emery, R.B. 2001. Probability of survival and breeding for
juvenile female Canvasbacks. Journal of Wildlife Management 65:385-397
Arnold, T.W., M.G. Anderson, M.D. Sorenson & R.B. Emery. 2002. Survival and philopatry of
female redheads breeding in southwestern Manitoba. Journal of Wildlife Management 66: 162169
Batt, B.D.J., Afton, A.D., Anderson, M.G., Ankney, C.D., Johnson, D.H., Kadlec, J.A. & Krapu,
G.L. 1992. Ecology and Management of Breeding Waterfowl. University of Minnesota Press,
Minneapolis & London, 635p.
40
Benton, T. G. and Grant, A. 1999. Elasticity analysis as an. important tool in evolutionary and
population ecology. –. Trends Ecol. Evol. 14: 467–471
Blums P, Mednis A, 1996. Secondary sex ratio in Anatinae. Auk 113:505-511
Blums, P., Mednis, A., Bauga, I., Nichols, J.D. & Hines, J.E. 1996. Age-specific survival and
philopatry in three species of European ducks: a long-term study. The Condor 98 : 61-74
Blums, P., Clark, R.G. & Mednis, A. 2002a. Pattern of reproductive effort and success in birds :
path analyses of long-term data from European ducks. Journal of Animal Ecology 71 : 280-295
Blums, P, Nichols, J.D., Hines, J.E. & Mednis, A. 2002b. Sources of variation in survival and
breeding site fidelity in three species of European ducks. Journal of Animal Ecology 71 : 438450
Cam, E., W.A. link, Cooch, E.G. Monnat, J.Y., Danchin, E. 2002. Individual covariation
between life-history traits: seeing the trees despite the forest. American Naturalist 159: 96-105
Caswell, H. 1989. Matrix population models. Sinauer Associates, Sunderland, Massachusetts,
USA.
Choquet, R., A.-M. Reboulet, J.-D. Lebreton, O. Gimenez, and R. Pradel. 2005. U-CARE 2.2
User’s Manual. CEFE, Montpellier, France. (http://ftp.cefe.cnrs.fr/biom/Soft-CR).
Clark, R. G.; Sugden, L. G.; Brace, R. K.; Nieman, D. J. 1988. The relationship between nesting
chronology and vulnerability to hunting of dabbling ducks. Wildfowl 39 : 137-144
Conroy, M.J. & Moore, C.T. 2001. Simulation models and optimal decision making in natural
resource management. In Modeling in natural resource management: Development,
Interpretation and application. Shenk, T.M. & Franklin, A.B. (Eds) Island Press, Washington
DC, USA.
Conroy, M. J. , M. W. Miller, and J. E. Hines. 2002. Identification and synthetic modeling of
factors affecting American black duck populations. Wildlife Monographs. 150.
Cramp, S. & Simmons, K.E.L. (eds) 1977. The Birds of the Western Palearctic. Vol. 1. Oxford
Univ. Press.
Defos du Rau, P., Barbraud, C. & Mondain-Monval, J.Y. 2003. Estimating breeding population
size of the red-crested pochard (Netta rufina) in the Camargue (southern France) taking into
account detection probability: implications for conservation. Animal Conservation 6: 379-385
Defos du Rau, P., Barbraud, C. & Mondain-Monval, J.Y. 2005. Incorporating uncertainty into
analyses of Red-crested Pochard habitat selection. Biological Conservation 125: 355-367
Dehorter, O., and G. Rocamora. 1999. Nette rousse Netta rufina. Pages 58-59 in G. Rocamora
and D. Yeatman-Berthelot, editors. Oiseaux menacés et à surveiller en France, listes rouges et
41
recherche de priorités, populations, tendances, menaces, conservation. Société d’Etudes
Ornithologiques de France, Paris.
Delany, S. & D.A. Scott 2002. Waterbird population estimates. Wetlands International, Global
Series n° 12, Wageningen (NL), 226 pp.
Devineau, O., Dutheil, J., Guillemain, M. & Lebreton, J.-D. 2006. Variation in waterfowl
survival in relation to hunting and population management. Poster presented at the 4th North
American Duck Symposium, Bismarck, ND, USA, 23-26 Aout 2006.
Dzus, E.H. & Clark, R.G. 1998. Brood survival and recruitment of mallards in relation to
wetland density and hatching date. The Auk 115 : 311-318
Ehrlén, J., Van Groenendael, J. & De Kroon, H. 2001. Reliability of Elasticity Analysis: Reply to
Mills et al.. Conservation Biology 15 (1), 278-280
Elmberg, J., Nummi, P., Pöysä, H., Sjöberg, K., Gunnarsson, G., Clausen, P., Guillemain, M.,
Rodrigues, D., Väänänen, V. 2006. The scientific basis for a new and sustainable management of
migratory European ducks. Wildlife Biology 12 : 121-128
Ferrière R, F Sarrazin, S Legendre & J-P Baron. 1996. Matrix population models applied to
viability analysis and conservation: Theory and practice with ULM software. Acta Oecologica
17:629-656
Fieberg, J. & Ellner, S.P. 2001. Stochastic matrix models for conservation and management: a
comparative review of methods. Ecology Letters 4 (3), 244–266
Flint, P. L., J. B. Grand, T. F. Fondell, and J. A. Morse. 2006. Population dynamics of Greater
Scaup breeding on the Yukon-Kuskokwim Delta, Alaska. Wildlife Monographs 162: 1-21
Frederiksen, M., Lebreton, J.D. & Bregnballe, T. 2001. The interplay between culling and
density-dependence in the great cormorant: a modelling approach. Journal of Applied Ecology
38 (3), 617–627
Gauthier, G., R. Pradel, S. Menu, and J.-D. Lebreton. 2001. Seasonal survival of greater snow
geese and effect of hunting under dependence in sighting probabilities. Ecology 82:3105-3119.
Gauthier, G., Lebreton, J.-D., 2004. Population models for Greater Snow Geese: a comparison of
different approaches to assess potential impacts of harvest Animal Biodiversity and
Conservation, 27(1): 503-514.
Gay, L., Defos du Rau, P., Mondain-Monval, J.Y. & Crochet, P.A. 2004. Phylogeography of a
game species: the red-crested pochard (Netta rufina) and consequences for its management.
Molecular Ecology 13 : 1035-1045
42
Gendron M., R.G. Clark. 2002. Survival of gadwall and mallard ducklings in southcentral
Saskatchewan. Journal of Wildlife Management. 66:170–180
Gibelin, A.L. & Déqué, M., 2003. Anthropogenic climate change over the Mediterranean region
simulated by a global variable resolution model. Climate Dynamics, 20: 327-339
Gilissen, N., Haanstra, L., Delany, S., Boere, G. & Hagemeijer, W. 2002. Numbers and
distribution of wintering waterbirds in the Western Palearctic and Southwest Asia in 1997, 1998
and 1999. Results from the International Waterbird Census. Wetlands International Global Series
N°11, Wageningen, Pays-Bas
Grillas, P., 1992. Les communautés de macrophytes submergées des marais temporaires oligohalins de Camargue, étude expérimentale des causes de la distribution des espèces. PhD thesis,
Université Rennes I, Rennes.
Gunnarsson, G., Elmberg, J., Sjöberg, K., Pöysä, H. & Nummi, P. 2006. Experimental evidence
for density-dependent survival in mallard (Anas platyrhynchos) ducklings. Oecologia 149:203–
213
Heppell, S.S., Caswell, H. & Crowder, L.B. 2000. Life histories and elasticity patterns :
perturbation analysis for species with minimal demographic data. Ecology 81 : 654-665
Hoekman, S.T., L.S. Mills, D.W. Howerter and J.H. Devries, I.J. Ball. 2002. Sensitivity analysis
of the life cycle of mid-continent mallards. Journal of Wildlife Management 66:883-900
Hoffmann, L. 1963a. Shooting pressure on mallard in southern France and its seasonal evolution
as studied by X-ray. Trans 6th Congr Int Union Game Biol p271-273
Hoffmann, L. 1963b. Situation de la sauvagine dans les pays méditerranéens d'Europe. In: 1st
European Meeting on Wildfowl Conservation Proceedings. St Andrews (Scotland) 1618/10/1963. SWIFT J. Nature Conservancy London (GBR) p59-64
Hunter, C., Moller, H. & Fletcher, D. 2000. Parameter Uncertainty and Elasticity Analyses of a
Population Model: Setting Research Priorities for Shearwaters. Ecological Modelling 134: 299323
Johnson, A.R. 1975. Station de baguage de Camargue. Compte-Rendu pour les années 1972 et
1973. Revue d’Ecologie Terre et Vie 29: 116-130
Johnson D. H., Nichols J. D., Schwartz M. D. 1992. Population dynamics of breeding waterfowl.
Pages 446–485 in Batt B. D. J., Afton A. D., Anderson M. G., Ankney C. D., Johnson D. H.,
Kadlec J. A., Krapu G. L. Ecology and management of breeding waterfowl. University of
Minnesota Press. Minneapolis, USA.
Kalchreuter H. 1996. Waterfowl harvest and population dynamics: an overview. Gibier Faune
43
Sauvage / Game Wildl. 13: 991-1008.
Keller, V. 2000. Winter distribution and population change of red-crested pochard Netta rufina
in southwestern and central Europe. Bird Study 47 : 176-185.
Kokko, H. & Lundberg, P. 2001. Dispersal, offspring retention and partial migration in saturated
habitats. - American Naturalist 157: 188-202
Krebs, C.J. 1972. Ecology. The Experimental Analysis of Distribution and Abundance. Harper
and Row, New York. 694 pp.
Lake, B.C., J. Walker, and M.S. Lindberg. 2006. Survival of ducks banded in the boreal forest of
Alaska. Journal of Wildlife Management 70:443-449
Lebreton J.D., Clobert J. 1991. Bird population dynamics, management and conservation : the
role of mathematical modelling. In: Perrins CM, Lebreton J-D, Hirons GJM (eds) Bird
population studies their relevance to conservation and management. Oxford University Press,
Oxford, pp 105-125
Lebreton, J.-D. 2005. "Dynamical and statistical models for exploited populations." Australian
and New Zealand Journal of Statistics 47(1): 49-63
Legendre, S., and J. Clobert. 1995. ULM, a software for conservation and evolutionary
biologists. Journal of Applied Statistics 22:817-834
Link, W.A. and P. F. Doherty, Jr. 2002. Scaling in sensitivity analysis. Ecology 83: 3299–3305
Llorente, G. A. & Ruiz, X. 1985. Datos sobre la reproducción del. pato colorado Netta rufina
(Pallas 1773) en el Delta del Ebro. Misc. Zool. 9: 315–323
McLoughlin, P.D. & Messier, F. 2004. Relative contributions of sampling error in initial
population size and vital rates to outcomes of population viability analysis. Conservation
Biology 18 : 1665-1669
Mayaud, N., 1966. Contribution à l'histoire de Netta rufina Pallas, la Nette à huppe rousse en
Europe occidentale. Alauda 34, 191-199.
Mills, L. S., D. F. Doak, and M. Wisdom. 1999. The reliability of conservation actions based
upon elasticities of matrix models. Conservation Biology 13: 815-829.
Nichols J.D. 1991. Responses of North American duck populations to exploitation. In Perrins
C.M., Lebreton J.-D. and Hirons G.J.M. (Eds.) Bird population studies: relevance to
conservation and management. Oxford University Press, Oxford. Pp 498-525.
Nichols, J.D., Johnson, F.A. & Williams, B.K. 1995. Managing North American Waterfowl in
the Face of Uncertainty Annual Review of Ecology and Systematics 26 : 177-199.
Nichols J.D. and Johnson F.A. 1996. The management of hunting of Anatidae. Gibier Faune
44
Sauvage / Game Wildl. 13: 977-989.
Nichols, J.D., Lancia, R.A. & Lebreton, J.D. 2001. Hunting statistics: What data for what use?
An account of an international workshop. Game and Wildlife Science 18, 185–205.
Niel, C. & Lebreton, J.D. 2005. Using demographic invariants to detect overharvested bird
populations from incomplete data Conservation Biology 19, 826-835
Office National de la Chasse, 1982. Critères de détermination de l’âge du Canard colvert de la
naissance à 9 semaines. Fiche technique n°10, Bulletin mensuel 63.
Reed, J.M., L.S. Mills, J.B. Dunning, E.S. Menges, K.S. Mckelvey, R. Frye, S.R. Beissinger, M.
Anstett, and P. Miller. 2002. Emerging issues in population viability analysis. Conservation
Biology 16: 7–19
Rose P.M. & Scott D.A. 1994. Waterfowl Population Estimates – 1st Edition. . IWRB Pub. 29,
102 p
Schmidt, J.H., E.J. Taylor, and E. Rexstad. 2006. Survival of common goldeneye ducklings in
Interior Alaska. Journal of Wildlife Management 70:792-798
Schwarz, C. J., J. F. Schweigert, and A. N. Arnason. 1993. Estimating migration rates using tagrecovery data. Biometrics 49:177–193.
Scott, D.A. & Rose, P.M. 1996. Atlas of Anatidae Populations in Africa and Western Eurasia.
Wetlands International Pub. n° 41, Wetlands International, Wagenigen, The Netherlands.
Sedinger, J. S., and E. Rexstad. 1994. Do restrictive harvest regulations result in higher survival
rates in mallards? Reply to Smith and Reynolds (1992) Journal of Wildlife Management
58(3):571-577
Shedden, C.B. 1986. Status of European quarry species. The British Association for Shooting
and Conservation
Snow, D.W., Perrins, C.M., 1998. The Birds of the Western Palearctic, Vol. 1. Oxford
University Press, Oxford.
Tamisier, A., and P. Grillas. 1994. A review of habitat changes in the Camargue : an assessment
of the effects of the loss of biological diversity on the wintering waterfowl community.
Biological Conservation 70:39-47.
Tamisier A., Dehorter O. 1999. Camargue, canards et foulques. Fonctionnement et devenir d'un
prestigieux quartier d'hiver. Centre Ornithologique du Gard, Nîmes, 369p
van Tienderen, P.H. 1995. Life cycle trade-offs in matrix population models. Ecology, 76, 24822489
45
Wade, P. R. 1998. Calculating limits to the allowable human-caused mortality of cetaceans and.
pinnipeds. Marine Mammal Science 14:1-37
White, G. C., and K. P. Burnham.1999. Program MARK: survival estimation from populations
of marked animals. Bird Study 46 (supplement):120-139.
Williams, B.K., J.D. Nichols, and M.J. Conroy. 2002. Analysis and management of animal
populations. Academic Press, San Diego . 817pp
Zimpfer, N.L. and M.J. Conroy 2006. Models of production rates in American black duck
populations. J. Wildl. Manage. 70: 947-954
46
Ecologie, démographie et conservation d’une espèce gibier rare : la Nette rousse Netta rufina
résumé de thèse
Pierre Defos du Rau
La Nette rousse Netta rufina est un des canards gibiers migrateurs les plus rares d’Europe. Les noyaux
de populations reproductrices et hivernantes y sont très localisés. La population européenne, estimée à
50000 individus en 2006 est essentiellement concentrée en Espagne, avec des effectifs également
significatifs en Allemagne, en France et en Suisse. En France, le statut de conservation de l’espèce est
considéré comme défavorable. Néanmoins, l’espèce est chassée en France, en Espagne, au Portugal et
en Roumanie. La Camargue est, avec le Forez et les Dombes, le principal noyau de reproduction de
l’espèce en France. Les effectifs camarguais étaient estimés à une centaine de couples depuis 1990.
Toutefois, ni les prélèvements, ni l’importance de la population soumise à ces prélèvements ne sont
connus avec suffisamment de précision pour garantir le caractère pérenne de son exploitation. La
conjonction des enjeux de conservation et de durabilité de l’exploitation concernant cette espèce
constitue un problème complexe qui fait appel à des concepts théoriques de biologie de la
conservation et des méthodologies issues de la génétique des populations, de l’écologie et de la
dynamique des populations.
Les objectifs de ce travail étaient
- d’identifier les exigences d’habitat pour la reproduction afin d’en déduire des recommandations
techniques à l’usage des gestionnaires de zones humides
- d’évaluer la durabilité de l’exploitation cynégétique de l’espèce et les risques éventuels pour la
viabilité de la population présente en Europe
Les pré-requis pour atteindre ces objectifs étaient donc
- de disposer de dénombrements fiables permettant des comparaisons d’effectifs dans l’espace
(écologie de la reproduction) et dans le temps (dynamique de population)
- d’identifier la population présente en Europe, sa distribution et son abondance afin de pouvoir
ensuite en modéliser la dynamique
- de développer un modèle démographique satisfaisant permettant de décrire et de prédire la
dynamique de la population dont la France et d’autres pays europeens participent à l’exploitation.
Dans le cadre de ce travail, les suivis de terrain utilisant des méthodes récentes de capture-recapture
d’estimation de la probabilité de détection ont permis d’estimer l’effectif reproducteur camarguais à
590 couples en 2001, ce qui démontre la sous-évaluation des mêmes effectifs précédemment estimés à
une centaine de couples. Ces méthodes récentes de mesure de l’erreur de détection par capturerecapture ont ensuite été utilisées pour analyser les exigences d’habitat sur la base de comparaisons de
présence et d’abondance tenant compte de ces erreurs de détection dans les deltas de l’Ebre
(Catalogne), du Rhône (France) et du Danube (Roumanie). Les résultats différent sensiblement de
ceux obtenus sans tenir compte de ces erreurs. C’est également le cas pour plusieurs prédictions
macro-écologiques générales. les implications en matière de gestion des habitats ne sont pas
négligeables puisque certaines recommandations de gestion ne peuvent être identifiées qu’à la
condition d’une prise en compte probabiliste des erreurs de détection.
Dans le cadre de ce travail, les analyses génétiques basées sur des échantillons récoltés sur l’ensemble
de l’aire de répartition de l’espèce ont démontré le fonctionnement en unité de gestion indépendante de
la population ouest-européenne. C’est donc la dynamique de cette unité populationnelle exploitée par
la chasse qu’il convient de décrire et de modéliser. L’estimation des paramètres démographiques de
l’espèce s’appuie sur plusieurs anciens jeux de données collectées en Camargue. Ces estimations
permettent la mise en évidence d’un mécanisme de densité-dépendence et d’une stratégie
démographique originale. Elles permettent également le développement d’un modèle matriciel soumis
à des analyses de perturbations et à des simulations sous diverses conditions de climat et de
prélèvements cynégétiques. S’il s’avère que l’exploitation cynégétique de la population ouesteuropéenne ne semble pas mettre en péril sa viabilité, il apparaît également que l’estimation de
plusieurs de ses paramètres démographiques, tels que les taux de survie et de prélèvement doit être
affinée afin de réduire les incertitudes liées à sa gestion globale. Le modèle de gestion adaptative des
gibiers d’eau développé en Amérique du Nord constitue un cadre semi-expérimental d’inférence
scientifique qu’il conviendrait d’appliquer en Europe.