Download Regional assessment of climate change impacts on maize

Reg Environ Change
DOI 10.1007/s10113-007-0039-z
ORIGINAL ARTICLE
Regional assessment of climate change impacts on maize
productivity and associated production risk in Switzerland
Daniele Torriani Æ Pierluigi Calanca Æ Markus Lips Æ
Helmut Ammann Æ Martin Beniston Æ Jürg Fuhrer
Received: 26 April 2007 / Accepted: 29 September 2007
Springer-Verlag 2007
Abstract A simple model of yield was used along with
climate scenarios to assess the impact of climate change on
grain maize productivity and associated economic risk in
Switzerland. In a first application, changes in the precipitation regime alone were shown to affect the distribution of
yield considerably, with shifts not only in the mean but also
in the standard deviation and the skewness. Production risk
was found to respond more markedly to changes in the
long-term mean than in the inter-annual variability of
seasonal precipitation amounts. In a further application,
yield projections were generated with respect to a full
climate scenario, with the emission pathway as specified in
the IPCC A2 scenario. Anticipation of the sowing date was
found to reduce the negative impact of climate change on
yield stability, but was not sufficient to ensure average
productivity levels comparable to those observed at present. We argued that this was caused by the reduction in the
duration of the growing season, which had a stronger
impact than suggested by previous studies. Assuming no
change in price relations, the results also revealed a strong
increase in production risk with climate change, with more
than a doubling in the probability of yield falling short of a
critical threshold as compared to today’s situation.
D. Torriani P. Calanca (&) J. Fuhrer
Air Pollution/Climate Group, Agroscope Research Station ART,
Reckenholzstrasse 191, 8046 Zurich, Switzerland
e-mail: [email protected]
M. Lips H. Ammann
Farm Management Group, Agroscope Research Station ART,
8356 Ettenhausen, Switzerland
M. Beniston
Climate Change and Climate Impacts, University of Geneva,
7 Route de Drize, 1227 Carouge, Switzerland
Keywords Maize productivity Production risk Climate change Climate variability Parametric yield model
Introduction
Climate change as projected by climate models for the
twenty-first century has the potential to significantly alter
the conditions for crop production, with important implications for worldwide food security (Rosenzweig and
Hillel 1998). Referring to the Intergovernmental Panel on
Climate Change Special Report on Emissions Scenarios
(SRES; Nakicenovic and Swart 2000), Parry et al. (2004)
estimated that while global production is likely to remain
stable for most of the century, regional differences could
grow stronger through time, with only developed countries
benefiting from climate change.
Regional differences in the response of crop productivity to climate change are also likely to emerge in Europe.
As reported by Olesen and Bindi (2002), climate change is
expected to have positive impacts only in the northern
countries, implying that areas of crop suitability may
expand northwards (Olesen et al. 2007). Southern areas, on
the other hand, will probably have to face increasing water
shortage and incidence of extreme adverse weather events.
Without suitable adaptation, implications include lower
harvestable yields, higher yield variability and a reduction
in suitable areas for traditional crops.
In Switzerland, increasing temperature and decreasing
summer precipitation could, in the absence of adaptation,
also prove adverse to rainfed production of spring and
summer crops. On the basis of a temperature increase of
5C and a decrease in summer precipitation of roughly
30%, Torriani et al. (2007) calculated that average yields
123
D. Torriani et al.
for grain maize (Zea mays L.) could decrease by 10% even
when positive effects of a doubling of the atmospheric CO2
concentration are taken into account. They also found that
the coefficient of yield variation could increase by as much
as a factor of two, implying depressed yield stability and
thus higher production risks.
Changes in yield behavior in relation to shifts in climate
can become critical for the economy of farms even in areas
profiting from favorable conditions at present. An
increasing probability of low returns as a consequence of
the more frequent occurrence of adverse conditions could
prove dramatic for farmers operating at the limit of economic stress. This could be the case in Switzerland, since
the projected shortage in water availability during the
summer season (Vidale et al. 2007) would more frequently
lower the productivity of rainfed crops. Assessing the
possible impact of climate change on production risks is
therefore necessary to help decision makers and stakeholders identify and implement suitable measures of
adaptation.
The key aspect for quantifying production risk is how
the frequency distribution of yield responds to changes in
the climatic settings (Kaylen and Korom 1991; Park and
Sinclair 1993). There are several ways to infer a distribution of yield from climatic data. A first possibility is to use
process-based models of crop growth along with a set of
daily weather data spanning a reasonable number of years.
This approach is often adopted in regional assessments
(e.g., Tubiello et al. 2000; Torriani et al. 2007) when a
realistic description of crop physiology is of advantage.
A second possibility is to infer the distribution of yield
from the distribution of climate using production functions
(Hexem and Heady 1978). Models of this kind are widely
employed in econometric studies (e.g., Just and Pope 1978;
Moss and Shonkwiler 1993; Chen et al. 2004) and also in
global assessments (e.g., Parry et al. 2004). Despite obvious limitations with respect to crop physiology, production
functions have the advantage of being robust and computationally cheap.
In our study we examined the use of yet another type of
model, which was proposed by Monteith in the 1970s
(Monteith 1972, 1977). It is a parametric model that relates
yield to the average climatic conditions throughout the
growing season using a set of deterministic equations. As
opposed to production functions, the model retains some of
the basics of plant physiology and should have for this
reason a wider range of applicability. Similarly to production functions, however, it cannot accommodate future
management decision taken by farmers on a day to day,
seasonal or annual basis.
Intended as an extension of the investigation of Torriani et al. (2007), our study considered climate change
effects on grain maize production in Switzerland. Maize
123
is one of the most important cereals both for human and
animal consumption (Doorenbos and Kassam 1986).
According to the statistics of the Food and Agriculture
Organization of the United Nations (http://faostat.fao.
org/), in the 25 countries of the European Union maize
production presently accounts for slightly less than 20%
of the total cereal production. In Switzerland, with a
nationwide average yield of 9.7 t ha–1 at present, maize
production shares an 18% of the total cereal production
(statistics of the Swiss Farmers’ Union; http://www.
bauernverband.ch/en/).
The specific aims of the study were: (1) to demonstrate
the reliability of the model approach proposed by Monteith
(1972, 1977) in determining the distribution of maize yield
in Switzerland under present climatic conditions; (2) to
study the sensitivity of rainfed maize production and
associated production risk to changes in the precipitation
regime; (3) to study the response of maize yield to climate
change given a particular climate scenario; and (4) to
provide a qualitative measure of how uncertainties in climate projections may affect yield scenarios. The latter
issue is explored in more detail, e.g., by Jones (2000), Luo
et al. (2005) and Minguez et al. (2007).
A word of caution is necessary before proceeding with
the exposition. The specific climate scenario used for the
analysis (see next section) refers to a threefold increase in
the atmospheric CO2 concentration by the end of the century and implies rather drastic changes in the temperature
and precipitation regime of the Alpine region. Moreover, as
stated above, we do not consider adaptation in cropping
practices, nor we take into account changes in the agricultural policy and the regional and global markets, which
ultimately would alleviate the negative effects of climate
change. In this sense, the analysis provides only upper
bounds for what could take place in the real world. It
should therefore be seen as a way to address the vulnerability of the present-day agricultural production of the
Alpine region.
Materials and methods
Yield model
Developed by Monteith (1972, 1977) the model used for
our analysis expresses yield as the product of a radiationdependent potential yield, Ypot, and two limiting factors, gh
and gVPD:
Y ¼ gh gVPD Ypot
ð1Þ
Here, h stands for the soil water content, VPD denotes the
vapor pressure deficit and
Regional assessment of climate change impacts
Ypot ¼ erad SRsGS
ð2Þ
erad being the efficiency of yield formation, SR the average
amount of solar radiation received by the crop throughout
the growing season and sGS the length of the growing
season. Here and in the following, an overbar is used to
designate a seasonal mean value.
The limiting factors account for two of the most
important forms of stress: (1) water stress resulting from
shortage of the water supply to the roots; and (2) heat stress
resulting from an excessive evaporative demand from the
leaves under conditions of elevated levels of the VPD
(Monteith 1977). To parameterize gh we rely on the fact
that soil water shortage reduces biomass accumulation and
evapotranspiration in comparable proportions (Doorenbos
and Kassam 1986). This makes it possible to equate gh with
the evapotranspiration efficiency. At the seasonal time
scale, the latter can be expressed in terms of the ratio
between precipitation, P, and potential evapotranspiration,
ETpot (Budyko 1974; Milly 1993a, b, 2001). Hence:
gh ET
P
¼ tanh c
ETpot
ETpot
ð3Þ
where c is a crop-specific parameter.
To model gVPD, we assume that for values of the VPD in
excess of about 5 hPa (C. Körner, personal communication) the transpiration efficiency of crops becomes nearly
inversely proportional to the VPD (Tanner and Sinclair
1983). This behavior can be generalized as a threshold
response (Thornley and Johnson 2000) as follows:
gVPD ¼ m þ ð1 mÞ
kn
n
kn þ VPD
ð4Þ
where k, m and n are again crop-specific parameters.
Model parameterization
There are five parameters in Eqs. 1–4: erad, c, k, m and n.
Of these, the efficiency of yield formation, which at
350 ppmv atmospheric CO2 concentration is of the order
of 0.07 t ha–1 (W m–2)–1, is considered to be specific not
only for the crop but also for different cultivars, reflecting
among other things disparities in leaf-area development
and harvest index (Muchow et al. 1990). Its value was
therefore specified separately for each application (see
respective sections below) ensuring that computed median
yields matched the median values inferred from
observations.
In contrast, values for the remaining parameters were
determined only once because Eqs. 3 and 4 are assumed to
reflect general responses of maize to changes in
environmental conditions. This was done relying on the
information provided by simulations carried out with
CropSyst (Stöckle et al. 2003), a soil-plant growth simulator that computes biomass accumulation and phenology
for perennial and non-perennial crops at a daily time step.
Inferring the parameters directly from the analysis of field
experiments would have been the preferred alternative.
However, the available data were too scarce to treat the
effects of the individual stress factors separately.
CropSyst results considered to parameterize Eqs. 3 and 4
are essentially those described in Torriani et al. (2007).
Maize yields were simulated for Waedenswil (8410 E,
47130 N, 463 m a.s.l.), a representative location on the
Swiss Plateau, using daily meteorological data for 1981–
2003 provided by the Swiss Federal Office of Meteorology
and Climatology to drive the model.
Given the relatively narrow spectrum of precipitation
amounts and air humidity levels in the daily observations
and the need to ensure the general validity of Eqs. 3 and 4,
these primary simulations were supplemented with additional runs driven with synthetic precipitation and air
humidity data. A first set of synthetic data was generated to
parameterize gh, whereby observed daily rainfall amounts
were reduced in turn by 20, 40 and 60%, leaving all other
variables unchanged relatively to the baseline.
A second set of data was prepared specifying relative
humidity in the range 10–100%. This second set was used
to parameterize gVPD.
For all CropSyst simulations, soil hydraulic properties
were specified with respect to a soil texture with clay, silt
and sand fractions of 38, 36 and 26%, respectively, relative
soil organic matter content of 2.6%, along with optimal
nitrogen fertilization and automatic irrigation. The sowing
date was set at day of the year (DOY) 130 and harvest was
assumed to take place 5 days after maturity. Base temperature of 7C and cut-off temperature of 20C were
specified to calculate growing degree days (GDD) (Stöckle
et al. 2003).
As shown in Fig. 1, in simulations with CropSyst most
of the biomass accumulation takes place during a so-called
linear growth phase (see also Goudriaan and Monteith
1990; Monteith 2000; Yin et al. 2003). The linear growth
phase is easily identified when time is expressed in GDD
units. In practice, start and end of the linear growth phase
were defined in correspondence to GDD thresholds of 400
and 1,250C-days, respectively, coinciding with DOY 182
and 273 at Waedenswil.
The linear growth phase was then used as a reference for
calculating the seasonal mean values of T, SR, VPD and P
needed in Eqs. 1–4. Following Calanca (2004), ETpot was
estimated from the seasonal mean net radiation, NR; using
the Priestley-Taylor (1972) equation:
123
D. Torriani et al.
Fig. 1 Biomass accumulation in the years 1981–2003 as a function
of growing degree days. Simulations with CropSyst for the test site
Waedenswil. The values of 400 and 1,250C-days used to delimit the
linear growth phase are highlighted with vertical lines
ETpot
ðdesat dTÞT NR
¼a
ðdesat dTÞT þ c Lv
ð5Þ
where a = 1.26 is the Priestley-Taylor coefficient,
ðdesat dTÞT the slope of saturation vapor pressure function
of temperature, evaluated at T; c the psychrometric constant
and Lv the latent heat of vaporization. In turn, net radiation
was inferred from solar radiation using a linear relation
(Davies 1967):
NR ¼ a þ bSR
ð6Þ
with coefficients a = –20 W m–2 and b = 0.62 determined
using accurate radiation measurements carried out at Payerne (6570 E, 46490 N, 490 m a.s.l.) within the framework
of the Baseline surface radiation network (BSRN) (Ohmura
et al. 1998).
In summary, based on the results presented in Fig. 2
and assuming for this site an efficiency of yield formation erad = 0.069 t ha–1 (W m–2)–1, values of c = 1.33,
k = 7.494 hPa, m = 0.251 and n = 1.607 were determined
for the parameters appearing in Eqs. 3 and 4.
Model testing
To test the validity of our approach, yield distributions
were computed for two other locations on the Swiss Plateau, Wynau (7470 E, 47150 N, 422 m a.s.l.) and Taenikon
(8540 E, 47290 N, 536 m a.s.l.), and contrasted with distributions inferred from a census of yield data collected
between 1975 and 2001 by the Swiss Federal Research
Station Reckenholz-Tänikon (ART 2002). The nationwide
census refers to a few thousand prototype farms spread all
123
Fig. 2 Relationships between gh and the ratio P=ETpot (top panel)
and between gVPD and VPD (bottom panel). Results of simulations
with CropSyst for Waedenswil
over the Swiss territory and provides information on geographic location, cultivated area, crop productivity and
production costs.
Unfortunately, the number and location of farms substantially varied from year to year. Moreover, there is no
information available in the census concerning cultivars
used and specific management practices. It is therefore
meaningless to compare model results and observations on
a farm unit basis. For this reason, yield data from farms
located within a distance of 15 km from the respective
meteorological stations were spatially aggregated to provide regional mean yields.
For both locations, daily climatic data for 1981–2003
were again obtained from the Federal Office of Meteorology and Climatology. To compute seasonal statistics, the
limits of the linear growth phase were adjusted according
the annual course of GDD. As before, base temperature of
7C and a cut-off temperature of 20C were used to infer
the GDD from temperature. For GDD limits of 400 and
1,250C-days, start and end of the linear growth phase
Regional assessment of climate change impacts
were found to coincide, respectively, with DOY 169 and
254 at Wynau, and DOY 171 and 260 at Taenikon.
Values of the radiation use efficiency of 0.067 and
0.061 t ha–1 (W m–2)–1 were specified for Wynau and
Taenikon, respectively. All other parameters (c, k, m and n)
were assigned the same values as in the Waedenswil runs
and ETpot was again computed using Eq. 5, with seasonal
mean net radiation given by Eq. 6.
Economic framework
The economic framework adopted for the analysis considers production risk as the long-term probability that
actual yield, Y, falls short of a critical threshold, Ycr, viz.:
Risk ¼ ProbfY Ycr g FðYcr Þ
ð7Þ
where F(Y) is the yield distribution function with the
associated mass probability function f(Y).
To define Ycr we assumed that this threshold corresponds
to long-term shut-down decision taking place when the
total revenue becomes equal to the costs of production:
pYcr vðYcr Þ
ð8Þ
where p is the price of grain yield and v(Y) the specific
production costs.
Ycr for today’s conditions in Switzerland was determined
using prices and costs published by Lips and Ammann
(2005). A summary of the most important entries is given
in Table 1. Some of the costs in the economic accountancy
Table 1 Economic
accountancy of maize
production in Switzerland
Yield (t ha–1)
depend on the level of yield. Hence, Ycr was initially
calculated for yield levels of 7.5, 8.5, 9.5, 10.5 and
11.5 t ha–1, resulting in values of 8.05, 8.12, 8.18, 8.21 and
8.24 t ha–1, respectively. A linear relation was then fitted to
these results and used to determine Ycr for the full range of
yields. In the simulations with climate scenarios, market
prices and production costs were left unchanged in view of
the large uncertainties concerning their future evolution.
Again, ignoring adaptation by farmers and policy makers to
the new climatic conditions implies that our results merely
provide an upper bound of the possible impacts.
Climate scenarios
Torriani et al. (2007) calculated yield scenarios for
2071–2100 using climate projections inferred from simulations with HIRHAM4, the regional climate model (RCM)
of the Danish Meteorological Institute (Christensen et al.
1998). With a spatial resolution of 50 km · 50 km, these
regional climate simulations were originally carried out as
a contribution to the PRUDENCE project and are available
though the project webpage (http://prudence.dmi.dk/,
accessed 24/09/2007). Details of the model setup and data
processing can be found in Christensen et al. (1998) and
Christensen and Christensen (2007).
All PRUDENCE experiments included a baseline or
control run for 1961–1990 and a climate scenario for 2071–
2100 (Christensen and Christensen 2007). The emission
scenario adopted for the HIRHAM4 experiment was the
IPCC SRES A2 scenario (Nakicenovic and Swart 2000).
7.5
8.5
9.5
10.5
11.5
Seeds (CHF ha–1)
272
272
272
272
272
Fertilizer (CHF ha–1)
Plant protection (CHF ha–1)
249
217
249
217
249
217
249
217
249
217
Cleaning and drying (CHF ha–1)
805
912
1,019
1,127
1,234
Hail insurance (CHF ha–1)
61
69
77
85
93
Other direct costs (CHF ha–1)
7
7
7
7
7
Costs
Labor costs (CHF ha–1)
764
764
764
764
764
Machinery costs (CHF ha–1)
1,345
1,359
1,368
1,368
1,368
Land value (CHF ha–1)
718
718
718
718
718
Interest rate costs (CHF ha–1)
38
40
43
46
49
Other indirect costs (CHF ha–1)
728
728
728
728
728
Benefits
Data are for the year 2005.
‘‘Other benefits’’ refers to the
average indemnities received by
the farmers from the hail and
other insurances
Producer benefits (CHF ha–1)
3,375
3,825
4,275
4,725
5,175
Direct payments (CHF ha–1)
1,600
1,600
1,600
1,600
1,600
Other benefits (CHF ha–1)
41
41
41
41
41
Balance (CHF ha–1)
–187
130
453
785
1,116
Critical yield threshold (t ha–1)
8.05
8.12
8.18
8.21
8.24
123
D. Torriani et al.
The corresponding CO2 level was about 800 ppmv by 2100
(three times the pre-industrial value), which provided an
upper bound for the ensemble of projections discussed in
the Third Assessment Report of IPCC (Houghton et al.
2001).
Given the biases in reproducing current climatic conditions with the regional climate models used in
PRUDENCE (Frei et al. 2003; Frei 2007; Jacob et al.
2007), an anomaly approach was chosen to derive a climate
distribution for the time window 2071–2100. The procedure can be outlined as follows.
Let mX and sX be the long-term average and corresponding inter-annual standard deviation of a series of
In addition, let the absolute and
seasonal mean values X:
relative anomalies of a long-term mean be denoted by Dm
and dm, respectively, and the relative anomaly of the interannual standard deviation be denoted by ds. Using superscripts BASE and SCEN to designate baseline and
scenario, these anomalies are defined by:
Dm mSCEN
mBASE
X
X
dm mSCEN
X
mBASE
X
47150 N, 608.98 m a.s.l) closest to the station of
Waedenswil.
Since changes in temperature affect the timing and
length of the growing season, two scenarios were generated. The first one (referred to as SC) was calculated
assuming the same limits for the linear growth phase as in
the baseline. The second one (referred to as AD) was
derived assuming anticipation of the sowing and harvest
dates in response to higher temperatures. For the second
scenario, start and end of the linear growth phase were
recalculated from the annual course of the GDD using
thresholds of 400 and 1,250C-days as in the baseline.
These limits were found in correspondence to DOY 155
and 223, respectively.
The data in Table 2 show that differences between the
two scenarios are appreciable. Compared to SC, the AD
scenario is characterized by a small change in precipitation,
a less severe increase in the seasonal mean VPD and a
more substantial increase in daily mean incoming solar
radiation.
ð9Þ
ð10Þ
Results
Model test
and
sSCEN
ds XBASE
sX
ð11Þ
The next step consists in employing the anomalies to
generate in a consistent way a series of seasonal mean
values XSCEN valid for the scenario from the series of
observed seasonal mean values XOBS : This can be achieved
letting:
ds
XSCEN ¼ mOBS
þ Dm þ XOBS mOBS
X
X
ð12Þ
The results of the model test are presented in Fig. 3
showing that the simulated probability distribution functions closely matched those inferred from the census data.
At Wynau the median yield was of 8.6 t ha–1 and the interquartile range was of 1.0 t ha–1 for the modeled and
1.2 t ha–1 for the observed distribution. At Taenikon
median yield was of 8.2 t ha–1, with simulated and
observed inter-quartile ranges of 1.1 t ha–1 and 1.4 t ha–1,
respectively.
Appreciable differences between simulated and
observed yield distributions were found in the details. For
or
OBS mOBS
XSCEN ¼ mOBS
ds
X dm þ X
X
ð13Þ
depending on whether the shift in the long-term mean is
given in absolute or relatively to the baseline.
Obviously, the first-term on the right-hand side of (12)
and (13) represents the adjustment of the long-term mean,
whereas the second-term reflects the inflation or suppression of inter-annual variability, i.e., of mean departure of
an individual year from the long-term mean. It is easy to
show that Eqs. 12 and 13 satisfy (10) and (11).
Four our analysis, absolute changes in the long-term
mean were assumed for T and VPD; while relative changes
were assumed for P and SR: All anomalies were derived
from the HIRHAM4 results for the grid-point (8350 E and
123
Table 2 Absolute (D), respectively relative anomalies (d) for the
long-term mean value and inter-annual standard deviation of seasonal
mean temperature (T), precipitation (P), solar radiation (SR) and
vapor pressure deficit (VPD)
Parameter
Scenario SC
Scenario AD
Mean
SD
Mean
SD
T
D = + 5.1C
d = 1.15
D = +5.0C
d = 1.25
P
SR
d = 0.77
d = 1.10
d = 0.73
d = 0.60
d = 1.03
d = 1.26
d = 1.01
d = 1.04
VPD
D = +2.3 hPa
d = 2.74
D = +1.6 hPa
d = 1.96
Results inferred from the HIRHAM4 simulations. Values in the first
two columns refer to the scenario without shift in the growing season
(SC); values in the third and fourth column refer to the scenario with
anticipation of sowing and harvest dates (AD)
Regional assessment of climate change impacts
Fig. 3 Histogram (bars), empirical probability density function (thin
continuous line) and modeled probability density function of maize
yield (thick dashed line) at Wynau (top panel) and Taenikon (bottom
panel). The density functions were calculated using a kernel density
estimation with Gaussian kernel
instance, the modeled distribution at Taenikon was negatively skewed, while a symmetric distribution was
suggested by the census data. Since the number of data
available in the census varied from a minimum of 2 in 1997
to a maximum of 13 in 1988 at Wynau, and from a minimum of 4 in 1987 and 2000 to a maximum of 18 in 1989 at
Taenikon, it could not be excluded that the comparison
suffered from a sampling bias in the census.
Denoting by hPi the total amount of rainfall throughout
growing season, long-term averages mhPi between 200 and
500 mm and corresponding inter-annual standard deviations shPi in the range 50–150 mm were specified to
reasonably cover current and future rainfall regimes (see,
e.g., Vidale et al. 2007).
A distribution of climate was obtained for each combination of long-term mean and inter-annual standard
deviation by first defining the distribution of hPi. Since
precipitation amounts are constraint to be non-negative,
hPi was modeled by means of the gamma distribution, with
shape and scale parameters calculated using the moment
estimators (Wilks 2006). The gamma distribution is commonly used in meteorology and climatology to represent
variations in rainfall amounts (e.g., Ison et al. 1971; Stern
and Coe 1984; Wilks 1990).
Distributions of T; SR and VPD were then generated
taking into account their relation to hPi. It was assumed
that each of T; SR and VPD consists of the sum of a
deterministic term, linearly depending on hPi, and a normally distributed stochastic component with zero mean and
appropriate standard deviation. Deterministic and stochastic components were fitted to the observed climatology
(Fig. 4) and were assumed valid for every other combination of mhPi and shPi. For each of P; T; SR and VPD, a set
of 30,000 realizations was drawn at random from the
respective distributions.
Results of the sensitivity analysis are displayed in
Fig. 5. The distribution of maize yield responded markedly
to changes in mhPi and shPi. Under current conditions (dots
in Fig. 5) mean, standard deviation and skewness of the
distribution were of 8.5 t ha–1, 0.75 t ha–1 and 0.1,
respectively. Decreasing mhPi or increasing shPi depressed
mean yield and yield stability, made the skewness
increasingly negative, and increased the risk of falling short
of the critical yield Ycr.
The sensitivity of the production risk to shifts in shPi
appeared to be less pronounced than for the moments of the
distribution. This reflects the fact that in the framework of
Eqs. 7 and 8 risk was primarily determined by the overall
level of productivity and only secondarily by the spread
and asymmetry of the distribution.
Sensitivity analysis: response to changes
in the precipitation regime
Case study: a maize yield scenario for 2071–2100
In a first application, Eqs. 1–4 were used to study the
response of the yield distribution and associated production
risks to changes in precipitation regime, including shifts in
both the long-term mean and the inter-annual variability.
To provide a possibility for comparison with the results
presented in Torriani et al. (2007), the analysis was conducted with respect to the station of Waedenswil.
Yield distributions calculated for the HIRHAM4 scenarios
are presented in Fig. 6. Despite disparities in the two scenarios (Table 2), the simulated yield distributions were
similar. The median yield was of 7.3 t ha–1 for SC and
7.2 t ha–1 for AD. These values should be contrasted with
the 8.5 t ha–1 obtained for the baseline, which indicates
that climate change is likely to have a negative impact on
123
D. Torriani et al.
Fig. 4 Deterministic (left) and
stochastic (right) components of
seasonal mean air temperature
(top), solar radiation (middle)
and VPD (bottom). Data are for
Waedenswil, valid for 1981–
2003. The panels on the righthand side show both the
empirical distribution of the
residuals (histogram and dashed
line) as well as the assumed
zero-mean normal distribution
(continuous line)
Fig. 5 Nomograms (isolines)
displaying the sensitivity of
long-term mean (top left), interannual standard deviation (top
right), and skewness (bottom
left) of maize yield, as well as
the sensitivity of production risk
(bottom right) to changes in
mean (mhPi) and standard
deviation (shPi) of seasonal
precipitation amounts. Results
are for Waedenswil. Current
climatic conditions are indicated
with a dot. Mean and standard
deviations of yield are in units
of t ha–1, skewness is
dimensionless, and risk is in %
123
Regional assessment of climate change impacts
Fig. 6 Histogram (bars) and modeled probability density function of
maize yield for the baseline climate (BL, thin continuous line), and
modeled probability density functions valid for the two climate
scenarios, without (SC, dashed line) and taking into consideration a
shift in timing and length of the growing period (AD, dotted dashed
line). The density functions were calculated using a kernel density
estimation with Gaussian kernel
yield, irrespective of whether or not an anticipation of the
growing season is taken into account.
Differences between the two scenarios were more pronounced concerning yield stability. Inter-quartile ranges of
2.1 and 1.3 t ha–1 were computed for SC and AD,
respectively. Both values were higher than the inter-quartile range calculated for the baseline (1.0 t ha–1).
In view of the specific climate settings, the fact that the
overall yield levels in SC and AD were comparable is
surprising. A closer look at the individual terms in Eqs. 1
and 2 revealed, however, different backgrounds. Potential
yield was higher in SC (14.1 t ha–1) than in AD (12.0 t
ha–1), thanks to the longer period available for biomass
accumulation (91 against 68 days) and despite lower levels
of solar radiation (204 against 234 W m–2 as an average
throughout the linear growth phase).
Dissimilarities between the two scenarios were also
found with respect to the limiting factors (Fig. 7). In SC,
shifts in climatic conditions resulted in a more severe
limitation of productivity, both with respect to soil water
shortage (median value of gh & 0.89, as opposed to 0.98 in
the baseline) as well as heat stress (median value of
gVPD & 0.59, as opposed to 0.69 in the baseline).
In AD, anticipation of the sowing date was effective in
reducing the impact of decreased summer precipitation on
yield formation. As seen in Fig. 7, the distribution of gh
was very close to the distribution inferred for the baseline
(median value of nearly 0.97 in both cases). On the other
hand, increasing temperature and VPD lowered productivity in AD as well (median value of gVPD & 0.62, as
opposed to 0.69 in the baseline).
Fig. 7 Histograms (bars) and probability density functions for the
two limiting factors, gVPD (top) and gh (bottom). Results are shown
for the baseline (BL, thin continuous line), the scenario without
adaptation (SC, dashed line) and the scenario accounting for
anticipation and shortening of the growing season (AD, dotted dashed
line). The density functions were calculated using a kernel density
estimation with Gaussian kernel
For present-day conditions, production risk was estimated at 28%. At first sight this value may appear high, but
it could be explained by the fact that average productivity
during the period 1981–2003 (8.5 t ha–1) was significantly
lower than in 2005 (9.5 t ha–1), the year taken as the reference for the economic analysis. Production risk increased
considerably as a consequence of climate change, attaining
67% in SC and 79% in AD.
Uncertainties in the projections
There are many sources of uncertainties in climate projections (e.g., Wigley and Raper 2001; Déqué et al. 2007),
and their implications should be considered in impact
studies. With respect to the Alpine region, variations across
the PRUDENCE scenarios have been quantified by Frei
123
D. Torriani et al.
(2007), Christensen and Christensen (2007) and Déqué
et al. (2007) for seasonal mean temperature and precipitation. Despite differences in the specific aims, the number
of scenarios and greenhouse-gases emission pathways
examined, these analyses concurred in indicating that the
average spread of the projections is of the order of 10–30%
around the ensemble mean values. Note that following
Déqué et al. (2007), we considered the inter-model standard deviation as the metric for quantifying the dispersion.
Unfortunately, none of the aforementioned studies
considered uncertainties in the projections of other climatic
parameters, nor did they attempt to quantify the uncertainties related to changes in the inter-annual standard
deviation. Based on the results presented by Déqué et al.
(2007) we therefore assumed that uncertainties in both the
absolute and relative anomalies of Table 2 can be represented by a normal distribution with coefficient of variation
of 20%. For precipitation and VPD, the assumed uncertainty is of the same order as the inter-annual variability
under present-day climatic conditions. We sampled each
distribution 10,000 times, assuming statistical independence of the anomalies, and repeated the calculations
reported in the previous section for each of the 10,000
realizations of climate. We then summarized the outcome
with respect to the uncertainty in mean yield and production risk.
As shown in Fig. 8, uncertainties in mean yield were
nearly normally distributed. This reflected an almost linear
mapping of the uncertainties in the climate scenario. At
2.0 t ha–1 the standard deviation of the distribution was
equivalent to a coefficient of variation of 18%, slightly less
than the mean uncertainty level imposed on the climate
anomalies.
Production risk, on the other hand, is constrained in the
interval 0–100%. It follows that the distribution of the
associated uncertainties can become highly asymmetric. As
seen in Fig. 8 this was indeed the case with respect to AD.
Taking a median risk of 81% as a reference, the results of
Fig. 8 implied that the probability of a very low or a very
high risk was not negligible, with a probability of 10% to
have either a risk of less than 30% or a risk higher than
about 90%.
Discussion and conclusions
Regional assessments of the effects of climate change on
crop production are needed at various decision levels, and
they are necessary to quantify the economic impacts at the
farm and regional scale. Of the various aspects related to
climate change, the possible increase in climate variability
has been recognized in recent years as one of the most
critical issues (Mearns et al. 1997; Porter and Semenov
123
Fig. 8 Uncertainties in the projections for mean yield (top) and risk
(bottom) as calculated for the scenario AD. Shown is the empirical
distribution function obtained by assuming normally distributed
uncertainties in the climate anomalies of Table 2 with coefficient of
variation of 20 %
2005). In this paper we demonstrated that a simple model
of crop yield operating at the seasonal time scale can be a
reliable tool for regional impact studies. We also showed
that such a model can be linked to a very basic economic
model, providing means for characterizing, at least in a
qualitative way, the expected effects of climate change on
production risks.
Shifts in yield and yield stability largely depend on
assumptions about future emissions, the climate projections, and the downscaling procedure used to generate the
climatic data at the regional scale typically required as
input to crop models. Olesen et al. (2007) noted that for a
site-based analysis the method used for downscaling is
more crucial than the choice of a specific climate scenario.
They also pointed out that use of climate model outputs
directly as input to the crop is not appropriate. Following
this recommendation, in our study we relied on the application of climate anomalies inferred from a regional
climate scenarios as adjustment factors to an observed
Regional assessment of climate change impacts
climatology, including shifts in both the long-term mean
conditions and the inter-annual variability of climate.
Out of the several scenarios available from the PRUDENCE project, we chose the HIRHAM4 scenario in order
to be consistent with an earlier analysis (Torriani et al.
2007). Comparison of the anomalies given in Table 2 with
the values published by Christensen and Christensen (2007)
and Déqué et al. (2007) shows that this particular scenario
represents the upper limit of the ensemble of simulations
available in PRUDENCE, at least with respect to changes in
summer temperature and precipitation. This needs to be
taken into account when interpreting the results.
In a first application of the yield model, we found that
changes in the precipitation regime alone have important
consequences not only for mean yield but also for the
distribution of yields, with yield stability decreasing with
decreasing mean and increasing variability of seasonal
rainfall, in agreement with studies in other regions (cf.
Fuhrer 2006). Taking into account changes in temperature,
solar radiation and VPD provided a more differentiated
picture. Ruling out adaptation, mean yield was found to
decrease from 8.5 to 7.3 t ha–1, in good agreement with the
results presented in Torriani et al. (2007). A similar
decrease in mean productivity was also found by Tubiello
et al. (2000) for an Italian location, but their results were
obtained assuming irrigated maize production.
Many studies (e.g., Tubiello et al. 2000; Torriani et al.
2007) regard anticipation of sowing date of spring/summer
crops as an effective measure of adaptation to increasing
temperature and decreasing water availability. In these
earlier studies anticipating the date of sowing had indeed a
positive effect on yield levels, but our results indicated that
this might not generally be true. Although average climatic
conditions were more favorable in AD than in SC, shortening of the growing period (–25% relative to the baseline)
resulted in a considerable loss of productivity. Therefore,
to take advantage of more favorable conditions during
spring time, cultivars with higher thermal time requirements would be required (Torriani et al. 2007).
Anticipation of the growing season had, however, a
positive effect on yield stability. While yield variability
more than doubled in SC, as compared to the baseline, the
increase was of only 30% in AD. For a critical yield of
about 8.1 t ha–1, this did not necessarily imply lower production risks. It is clear that quantifying the economic
implications of climate change requires the understanding
of yield behavior in a specific economic context. Accordingly, assessing production risk solely on the basis of shifts
in the distribution of yield can be misleading.
Several aspects were not addressed by our analysis. For
instance we ignored the positive effects of elevated CO2
concentration on maize productivity. Most studies conducted so far assumed that maize yields would improve
roughly by 10–15% for a doubling of CO2 (Sinclair and
Rawlins 1993; Parry et al. 2004). This could have been
considered by adjusting erad (Sinclair and Rawlins 1993)
and by modifying gh and gVPD to reflect acclimation as an
increase in transpiration efficiency (Polley 2002; Fuhrer
2003). However, the extent of the CO2 effect has recently
been questioned (Long et al. 2006).
Furthermore, we did not attempt to account for technological advances in production, including crop
improvements through breeding (Duvick 2005; Sinclair
and Muchow 2001) or switching to cultivars with higher
thermal time requirements (Torriani et al. 2007). Also, we
disregarded the possibility that farmers can opt among a
variety of crops and reduce the overall economic risks by
choosing those responding more positively or less negatively to climate change. Mendelsohn et al. (1994) and
more recently Mendelsohn and Reinsborough (2007) and
Mendelsohn et al. (2007) showed that allowing for a palette
of crops the effects of global warming on farms’ value are
significantly lower than estimated for a single crop with the
help of traditional production functions.
Finally, we adopted an economic framework in which
critical yield was only dependent on today’s prices and
costs. A more realistic assumption would have been that
such a critical yield threshold reflects the evolution of the
prices and costs relations. For instance, Parry et al. (2004)
considered higher prices in response to lower crop productivity worldwide, but for Switzerland lower crop prices
could be expected in view of the changes in the agricultural
policy prompted by the harmonization with the Common
Agricultural Policy of the European Union or in response
to outcomes of negotiations of the World Trading Organization. In any case, making realistic projections remain a
difficult task. In this sense, the assumption of constant
prices and costs was not more arbitrary than any other
hypothesis.
Acknowledgments This work was supported by the Swiss National
Science Foundation through the National Centre of Competence in
Research on Climate (NCCR Climate). Climatic data were provided
by the Swiss Federal Office of Meteorology and Climatology, while
results of the HIRHAM4 simulations were provided through the
PRUDENCE data archive, funded by the EU through contract EVK2CT2001-00132. The authors would like to thank Prof. C.O. Stöckle
for the internship granted to D.S.T. at Washington State University,
Pullman (USA). Analyses and graphics were produced using the
open-source software package R. Special thanks are due to the R
Development Core Team (URL: http://www.r-project.org/, accessed
24/09/2007). Two anonymous reviews were very helpful for
improving the manuscript.
References
ART (Agroscope Reckenholz-Tänikon) (2002) Ergebnisse der Zentralen Auswertung von Buchhaltungsdaten (Results of the central
123
D. Torriani et al.
accounting assessment). Report available from the Research
Station
Agroscope
Reckenholz-Tänikon,
Ettenhausen,
Switzerland
Budyko MI (1974) Climate and life. Academic, New York
Calanca P (2004) Interannual variability of summer mean soil
moisture conditions in Switzerland during the 20th century: a
look using a stochastic soil moisture model. Water Resour Res
40:W12502. doi:10.1029/2004WR003254
Chen CC, McCarl BA, Schimmelpfenning DE (2004) Yield variability as influenced by climate: a statistical investigation. Clim
Change 66:239–261
Christensen OB, Christensen JH, Machenauer B, Botzet M (1998)
Very high-resolution regional climate simulations over Scandinavia—present climate. J Clim 11:3204–3229
Christensen JH, Christensen OB (2007) A summary of the PRUDENCE model projections of changes in European climate by
the end of this century. Clim Change 81:7–30
Davies JA (1967) A note on the relationship between net radiation
and solar radiation. Quart J Roy Meteor Soc 93:109–115
Déqué M, Rowell DP, Lüthi D, Giorgi F, Christensen JH, Rockel B,
Jacob D, Kjellström E, de Castro M, van den Hurk B (2007) An
intercomparison of regional climate simulations for Europe:
assessing uncertainties in model projections. Clim Change
81:53–70
Doorenbos J, Kassam AH (1986) Yield response to water. FAO
Irrigation and Drainage Paper 33. Food and Agriculture Organization of the United Nations, Rome
Duvick DN (2005) The contribution of breeding to yield advances in
maize (Zea mays L.). Adv Agron 86:83–145
Frei C, Christensen JH, Déqué M, Jacob D, Jones RG, Vidale PL
(2003) Daily precipitation statistics in regional climate models:
evaluation and intercomparison for the European Alps. J
Geophys Res 108. doi:10.1029/2002JD002287
Frei C (2007) Die Klimazukunft der Schweiz (The future of Climate
in Switzerland). Report published by the Swiss Advisory Body
on Climate Change (OcCC), Bern, pp. 12–16. The report is
available at http://www.occc.ch
Fuhrer J (2003) Agroecosystem responses to combinations of elevated
CO2, ozone, and global climate change. Agric Ecosyst Environ
97:1–20
Fuhrer J (2006) Agricultural systems: sensitivity to climate change.
CAB Rev Perspect Agric Vet Sci Nutr Nat Resour 1:052. doi:
10.1079/PAVSNNR20061052
Goudriaan J, Monteith JL (1990) A mathematical function for crop
growth based on light interception and leaf area expansion. Ann
Bot 66:695–701
Hexem RW, Heady EO (1978) Water production functions for
irrigated agriculture. The Iowa State University Press, Ames
Houghton JT, Ding Y, Griggs DJ, Noguer M, van der Linden PJ, Dai
X, Maskell K, Johnson CA (2001) Climate change 2001: the
scientific basis. Contribution of working group I to the third
assessment report of the intergovernmental panel on climate
change. Cambridge University Press, Cambridge
Ison NT, Feyerherm AM, Bark LD (1971) Wet period precipitation
and the gamma distribution. J Appl Meteor 10:658–665
Jacob D, Bärring L, Christensen OB, Christensen JH, de Castro M,
Déqué M, Giorgi F, Hagemann S, Hirschi M, Jones R,
Kjellström E, Lenderink G, Rockel B, Sánchez E, Schär C,
Seneviratne SI, Somot S, van Ulden A, van den Hurk B (2007)
An intercomparison of regional climate models for Europe:
design of the experiments and model performance. Clim Change
81:31–52
Jones RN (2000) Analysing the risk of climate change using an
irrigation demand model. Clim Res 14: 89–100
Just RE, Pope RD (1978) Stochastic specification of production
functions and economic implications. J Econom 7:67–86
123
Kaylen MS, Korom SS (1991) Trend, weather variables, and the
distribution of U.S. corn yields. Rev Agric Econ 13:249–258
Lips M, Ammann H (2005) Vollkostenkalkulationen für Ackerkulturen (full cost calculations for field crops). Agrarforschung
13:228–231
Long SP, Ainsworth EA, Leakey ADB, Nösberger J, Ort DR (2006)
Food for thought: lower-than-expected crop yield stimulation
with rising CO2 concentrations. Science 312:1918–1921
Luo Q, Jones RN, Williams M., Bryan B, Bellotti W (2005)
Probabilistic distributions of regional climate change and their
application in risk analysis of wheat production. Clim Res
29:41–52
Mearns LO, Rosenzweig C, Goldberg R (1997) Mean and variance
change in climate scenarios: method, agricultural applications,
and measures of uncertainty. Clim Change 35:367–396
Mendelsohn R, Nordhaus WD, Shaw D (1994) The impact of global
warming on agriculture: a Ricardian analysis. Amer Econ Rev
84:753–771
Mendelsohn R, Reinsborough M (2007) A Ricardian analysis of US
and Canadian farmland. Clim Change 81:9–17
Mendelsohn R, Basist A, Kurukulasuriya P, Dinar A (2007) Climate
and rural income. Clim Change 81:101–118
Milly PCD (1993a) An analytic solution of the stochastic storage
problem applicable to soil moisture. Water Resour Res 29:3755–
3758
Milly PCD (1993b) A minimalist probabilistic description of root
zone soil water. Water Resour Res 37:457–463
Minguez MI, Ruiz-Ramos M, Diaz-Ambrona CH, Quemada M, Sau F
(2007) First-order impacts on winter and summer crops assessed
with various high-resolution climate models in the Iberian
Peninsula. Clim Change 81:343–355
Monteith JL (1972) Solar radiation and productivity in tropical
ecosystems. J Appl Ecol 9:747–766
Monteith JL (1977) Climate and the efficiency of crop production in
Britain. Phil Trans R Soc Lond B 281:277–294
Monteith JL (2000) Fundamental equations for growth in uniform
stands of vegetation. Agric Forest Meteorol 104:5–11
Moss CB, Shonkwiler JS (1993) Estimating yield distributions with a
stochastic trend and nonnormal errors. Amer J Agr Econ
75:1056–1062
Muchow RC, Sinclair TR, Bennett JM (1990) Temperature and solar
radiation effects on potential maize yield across locations. Agron
J 82:338–343
Nakicenovic N, Swart R (eds) (2000) Special report on emission
scenarios. Intergovernmental panel on climate change. Cambridge University Press, Cambridge
Ohmura A, Gilgen H, Hegner H, Müller G, Wild M, Dutton EG,
Forgan B, Fröhlich C, Philipona R, Heimo A, König-Langlo G,
McArthur B, Pinker R, Whitlock CH, Dehne K (1998) Baseline
surface radiation network (BSRN/WCRC): new precision radiometry for climate research. Bull Amer Meteor Soc 79:2115–
2136
Olesen JE, Bindi M (2002) Consequences of climate change for
European agricultural productivity, land use and policy. Eur J
Agron 16:239–262
Olesen JE, Carter TR, Dı́az-Ambrona CH, Fronzek S, Heidmann T,
Hickler T, Holt T, Minguez MI, Morales P, Palutikof JP, Quemada
M, Ruiz-Ramos M, Rubæk GH, Sau F, Smith B, Sykes MT (2007)
Uncertainties in projected impacts of climate change on European
agriculture and terrestrial ecosystems based on scenarios from
regional climate models. Clim Change 81:123–143
Park WI, Sinclair TR (1993) Consequences of climate and crop yield
limits on the distribution of corn yields. Rev Agric Econ 15:483–
493
Parry ML, Rosenzweig C, Iglesias A, Livermore M, Fischer G (2004)
Effects of climate change on global food production under SRES
Regional assessment of climate change impacts
emissions and socio-economic scenarios. Global Environ
Change 14:53–67
Polley HW (2002) Implications of atmospheric and climate
change for crop yield and water use efficiency. Crop Sci
42:131–140
Porter JR, Semenov MA (2005) Crop responses to climatic variation.
Phil Trans R Soc B 360:2021–2035
Priestley CHB, Taylor RJ (1972) On the assessment of surface heat
flux and evaporation using large-scale parameters. Mon Wea
Rev 100:81–92
Rosenzweig C, Hillel D (1998) Climate change and the global
harvest. Oxford University Press, New York
Sinclair TR, Rawlins SL (1993) Inter-seasonal variation in soybean
and maize yields under global environmental change. Agron J
85:406–409
Sinclair TR, Muchow RC (2001) System analysis of plant traits to
increase grain yield on limited water supplies. Agron J 93:263–
269
Stern RD, Coe R (1984) A model fitting analysis of daily rainfall data.
J Roy Stat Soc A 147:1–34
Stöckle CO, Donatelli M, Nelson R (2003) CropSyst, a cropping
system simulation model. Eur J Agron 18:289–307
Tanner CB, Sinclair TR (1983) Efficient water use in crop production:
research or re-search? In: Taylor HM, Jordan WR, Sinclair TR
(eds) Limitations to efficient water use in crop production.
Madison (USA). Am. Soc. Agronomy, Crop Science Soc. Amer.,
Soil Science Soc. Amer. pp 1–27
Thornley JHM, Johnson IR (2000) Plant and crop modelling. A
mathematical approach to plant and crop physiology. Blackburn,
Caldwell
Torriani DS, Calanca P, Schmid S, Beniston M, Fuhrer J (2007)
Potential effects of changes in mean climate and climate
variability on the yield of winter and spring crops in Switzerland.
Clim Res 34:59–69
Tubiello FN, Donatelli M, Rosenzweig C, Stöckle CO (2000) Effects
of climate change and elevated CO2 on cropping systems: model
predictions at two Italian sites. Eur J Agron 13:179–189
Vidale PL, Lüthi D, Wegmann R, Schär C (2007) European summer
climate variability in a heterogeneous multi-model ensemble.
Clim Change 81:209–232
Wigley TML, Raper SCB (2001) Interpretation of high projections for
global-mean warming. Science 293:451–454
Wilks DS (1990) Maximum likelihood estimation for the gamma
distribution using data containing zeros. J Clim 3:1495–1501
Wilks DS (2006) Statistical methods in the atmospheric sciences. 2nd
edn. Academic, London
Yin X, Goodrian J, Lantinga EA, Vos J, Spiertz HJ (2003) A flexible
sigmoid function of determinate growth. Ann Bot 91:361–371
123