Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
General enquiries on this form should be made to: Defra, Science Directorate, Management Support and Finance Team, Telephone No. 020 7238 1612 E-mail: [email protected] SID 5 Research Project Final Report Note In line with the Freedom of Information Act 2000, Defra aims to place the results of its completed research projects in the public domain wherever possible. The SID 5 (Research Project Final Report) is designed to capture the information on the results and outputs of Defra-funded research in a format that is easily publishable through the Defra website. A SID 5 must be completed for all projects. 1. Defra Project code 2. Project title This form is in Word format and the boxes may be expanded or reduced, as appropriate. 3. ACCESS TO INFORMATION The information collected on this form will be stored electronically and may be sent to any part of Defra, or to individual researchers or organisations outside Defra for the purposes of reviewing the project. Defra may also disclose the information to any outside organisation acting as an agent authorised by Defra to process final research reports on its behalf. Defra intends to publish this form on its website, unless there are strong reasons not to, which fully comply with exemptions under the Environmental Information Regulations or the Freedom of Information Act 2000. Defra may be required to release information, including personal data and commercial information, on request under the Environmental Information Regulations or the Freedom of Information Act 2000. However, Defra will not permit any unwarranted breach of confidentiality or act in contravention of its obligations under the Data Protection Act 1998. Defra or its appointed agents may use the name, address or other details on your form to contact you in connection with occasional customer research aimed at improving the processes through which Defra works with its contractors. SID 5 (Rev. 3/06) Project identification IF0103 Evaluating ecosystem models as tools for policy development on biodiversity Contractor organisation(s) University of Warwick Warwick HRI Wellesbourne Warwick Warwickshire CV35 9EF 54. Total Defra project costs (agreed fixed price) 5. Project: Page 1 of 23 £ 60,034 start date ................ 01 April 2006 end date ................. 31 March 2007 6. It is Defra’s intention to publish this form. Please confirm your agreement to do so. ................................................................................... YES NO (a) When preparing SID 5s contractors should bear in mind that Defra intends that they be made public. They should be written in a clear and concise manner and represent a full account of the research project which someone not closely associated with the project can follow. Defra recognises that in a small minority of cases there may be information, such as intellectual property or commercially confidential data, used in or generated by the research project, which should not be disclosed. In these cases, such information should be detailed in a separate annex (not to be published) so that the SID 5 can be placed in the public domain. Where it is impossible to complete the Final Report without including references to any sensitive or confidential data, the information should be included and section (b) completed. NB: only in exceptional circumstances will Defra expect contractors to give a "No" answer. In all cases, reasons for withholding information must be fully in line with exemptions under the Environmental Information Regulations or the Freedom of Information Act 2000. (b) If you have answered NO, please explain why the Final report should not be released into public domain Executive Summary 7. The executive summary must not exceed 2 sides in total of A4 and should be understandable to the intelligent non-scientist. It should cover the main objectives, methods and findings of the research, together with any other significant events and options for new work. The aim of this project was to evaluate whether ecosystem scale modelling approaches could be used as tools to develop policy to meet biodiversity targets. The project focussed on producing a framework for the selection and development of models that can be used to predict how plausible changes in land-use might impact on biodiversity. We initially categorised and reviewed existing models and modelling approaches according to trophic level, separately considering models for vegetation, invertebrates and farmland birds. From these reviews we identified that most existing models considered only single species, often at inappropriate spatial scales for the wide-scale modelling of changes in land-use, and that existing models rarely included interactions between trophic levels. Thus there is a need to develop models that include multiple species and trophic levels and the interactions between them, based on quantitative data. To allow the incorporation of multiple species and trophic levels, there is a need to be able to integrate data and models across spatial, temporal and ecological scales. A wide diversity of mathematical and statistical modelling approaches exist to tackle the integration of data in this way, with the specific approach required being dependent on the focus of the modelling exercise and the available data. Modelling approaches used in existing studies of the impact of land-use changes on biodiversity were reviewed, most being based on expert opinion rather than quantitative data, and rarely incorporating data across several spatial scales We therefore needed to ensure that our proposed modelling framework assisted both in the selection of appropriate techniques and the identification of available and necessary data sources, according to the outcome information required to address any particular policy question. Proposed Modelling Framework Our proposed modelling framework, to develop models to determine how plausible changes in land-use might impact on biodiversity, is defined by a set of five key questions: 1) What is the current spatial distribution of the species of interest? 2) What is the potential spatial distribution of the species of interest given the changes in land-use? 3) How much of the potential spatial distribution can the species of interest achieve (within a defined time scale), taking account of habitat and landscape permeability? 4) Are there sufficient resources to allow the species of interest to establish in the new locations (within a defined time period)? 5) Will the species of interest establish, and if so how will it affect the current community? SID 5 (Rev. 3/06) Page 2 of 23 (ecological interactions) The modelling approaches required to answer these questions will need to be selected according to the policy question being asked, the level of spatial detail required, and the level of detail of the available data. The progression through the questions requires increasing detail to be included within the model. Questions 1 and 2 can be answered using quite broad scale data, and for a relatively large geographical region, though models at this scale will only be able to assess potential changes in biodiversity on this large spatial scale. Question 3 requires more localised spatial data, as it requires knowledge of the potential movement of species and will therefore be subject to variation caused by habitat composition (although it should be possible to aggregate the information to provide summaries over wider spatial scales). Questions 4 and 5 certainly require information at a smaller spatial scale, since they involve questions about species interactions with the local environment. Answers to all 5 questions will enable the development of models to provide the most useful and detailed information on how changes in land-use will impact on biodiversity. Predictions from these models of the outcomes associated with ranges of plausible future scenarios can then be used to inform or develop policy on particular issues of interest. Having defined the framework within which ecosystem modelling approaches could be applied to address policy questions, there was a further need to define a clear process within which to apply the modelling framework. This includes the definition of clear, unambiguous targets or hypotheses (under specified land-use scenarios and timescales) to be evaluated using modelling. The key stages in the process are: 1. Define and clarify the policy problem or question to be addressed: a. What is the precise question to be answered, specified both at an overall policy level and at a practical implementation level? b. What definition(s) of biodiversity (species richness, community composition, density of key indicator species) will be used, and how should multiple measures be combined? c. What spatial scale is of interest, over what geographical region? d. Over what time scale should land-use changes and biodiversity impacts be considered? e. How are changes in biodiversity to be measured and presented? 2. Develop plausible future land-use scenarios: a. What are the most likely land-use changes? b. What are the likely spatial distributions of these changes? 3. Select modelling approaches (based on data availability) to answer the framework questions, allowing for any required species interactions: a. What is the current spatial distribution of the species of interest? b. What is the potential spatial distribution of the species of interest after land-use changes? c. How much of the potential spatial distribution can the species achieve (within a defined time period)? d. Are there sufficient resources to allow the species to establish in the new distribution (within a given time period)? e. Will the species establish, and if so how will it affect the current community? 4. Simulate the plausible future scenarios and analyse model outputs. 5. Make recommendations based on analysis of model output. a. Provide list of model assumptions. b. Provide measures of uncertainty on outputs. c. Rank scenarios in terms of being most beneficial (or least harmful) to biodiversity. Together the modelling framework and the above process provide a pragmatic structured approach to producing models to inform and assist in the development of policy on land-use changes, with regards to their impacts on biodiversity. This approach fits well with the ALMaSS modelling approach developed in Denmark, which consists of a landscape and weather simulator, combined with farm management information. Models of individual species are then linked to the landscape simulator, providing a modular approach to developing models of the impacts of landscape change on species abundance, movement and distribution. We recommend that Defra adopt both the framework and modelling process for projects that aim to examine the impact of land-use changes on biodiversity, and embed it into a modular modelling approach. Consultation with Defra provided a number of example policy questions. These questions focussed on specific Public Service Agreement and England Biodiversity Strategy targets. It is possible to use the SID 5 (Rev. 3/06) Page 3 of 23 approach described above to provide potential distributions for individual UK priority species under a range of land-use change scenarios, using existing data (where it is available). It will not be possible to determine achievable distributions unless more data on species dispersal and habitat permeability are collected. We recommend that Defra considers funding research to determine the effect of the permeability of the landscape mosaic on the distribution of species of interest and to collect data on species dispersal through a range of habitat types. For biodiversity conservation, there would almost certainly be a need to consider multiple species, rather than individual species. Although the modelling techniques and tools exist to incorporate species interactions into models, there is a lack of data on interactions between different species. We recommend that Defra considers funding research into the interactions between multiple species in a range of habitats. The availability of appropriate data will be a particular problem that Defra may have to address when considering future projects to model how changes in land-use impact on biodiversity. Where appropriate data are not available, either data will have to be collected as part of the project, or the project will have to develop the model in the absence of data, possibly based on expert opinion. Such projects should, however, also provide clear recommendations on the data that need to be collected to fully parameterise the developed model, and the data collection and subsequent use of the model for decision making should incorporate the concepts of adaptive resource management, allowing iterative improvements in models and decisions as greater information and data become available. We recommend that Defra collate and co-ordinate the collection of data on habitat types and land-use, combined with monitoring of the effects of policy to provide a dataset for use in models to predict the effects of land-use changes on biodiversity. We recommend that Defra revise models and their predictions on a regular basis (annually or bi-annually), using data from monitoring, so that policy can be adjusted to account for the most up-to-date information. Project Report to Defra 8. As a guide this report should be no longer than 20 sides of A4. This report is to provide Defra with details of the outputs of the research project for internal purposes; to meet the terms of the contract; and to allow Defra to publish details of the outputs to meet Environmental Information Regulation or Freedom of Information obligations. This short report to Defra does not preclude contractors from also seeking to publish a full, formal scientific report/paper in an appropriate scientific or other journal/publication. Indeed, Defra actively encourages such publications as part of the contract terms. The report to Defra should include: the scientific objectives as set out in the contract; the extent to which the objectives set out in the contract have been met; details of methods used and the results obtained, including statistical analysis (if appropriate); a discussion of the results and their reliability; the main implications of the findings; possible future work; and any action resulting from the research (e.g. IP, Knowledge Transfer). Evaluating ecosystem models as tools for policy development on biodiversity Aims and objectives The aim of this project was to evaluate whether ecosystem scale modelling approaches could be used as tools to develop policy to meet biodiversity targets. The objectives for the project are listed below: 1. Identify modelling approaches that can be used for predictive ecosystem scale modelling, including food web and vegetation change models. SID 5 (Rev. 3/06) Page 4 of 23 2. Critically review selected ecosystem scale modelling approaches to determine those that can predict policy effects on habitat and biodiversity changes and their impact on farmland birds, using the criteria defined by an expert steering group. 3. Determine how to integrate data across a range of geographical scales to allow ecosystem modelling approaches to predict changes at regional and national scales. 4. Prioritise the future research necessary to develop ecosystem scale modelling approaches for policy development on farmland biodiversity. Project progress and outcomes of research Objective 1: Identify modelling approaches that can be used for predictive ecosystem scale modelling, including food web and vegetation change models The aim of this objective was to identify the models and modelling approaches that were available in the scientific literature to provide information on the current status of the modelling of biodiversity and land-use, and to define the scope of the project and identify the definitions to be used for biodiversity. In consultation with the Steering Group, it was decided that the project should consider biodiversity at a range of trophic levels (plants, insects and birds), and that modelling approaches could be developed to address the differing needs of scientists, implementers of policy and policy makers, as each of these user groups would require different degrees of complexity and information within a modelling structure (a potential framework for this is summarised in Appendix 1). The Steering Group also identified the following topics as important areas that would need to be included within ecosystem models for the modelling of biodiversity: Soil type Vegetation type and vegetation structure Spatial aspect between as well as within fields Weather and environmental variables Dispersal of organisms Factors linked to population dynamics, such as food availability, breeding habitat availability, predation Pesticide usage It was noted by the Steering Group that the majority of these factors require detailed biological and environmental information that would be difficult to collect in sufficient detail to incorporate into a model. They would be useful in the context of developing a model that aims to determine the mechanisms which are influencing changes in biodiversity from a scientific viewpoint, but would perhaps be less useful for models aimed at understanding larger scale trends for the development of policy. The Steering Group did however feel that it would be very important to include both multiple species and spatial heterogeneity within any models to be used for the development of policy. The steering group suggested that the key to setting the scope of the project was to talk directly to those people involved with the development of policy to determine the information and level of detail that was required, and the geographical scale at which the model would need to predict. A meeting was arranged with Defra policy makers, and held on 24 th November 2006. A wide range of topics relating to modelling were discussed at the meeting, and it was suggested that although mechanistic models would be helpful from a policy point of view, there were limitations due to the costs and time required for the collection of sufficient data with which to parameterise the models. This meant that the development of mechanistic models was often uneconomic and unfeasible from a policy point of view. It was suggested that there were many existing approaches used to model biodiversity, but these approaches, implemented in projects such as MONARCH and BEETLE, tended to result in a high degree of uncertainty associated with outputs. One useful output of the project could be to examine ways to reduce the uncertainty of models, and to address areas relating to the permeability of landscapes for particular species. It was felt by Defra policy makers that the models should provide generic information that was applicable to a wide geographical area (regional and national). Although knowing what was happening within a farm or field was useful for the delivery of policy, it was the effects at the landscape scale that were of greatest interest in terms of determining policy, particularly in relation to the interactions between multiple species and different trophic levels. The key issues agreed at the meeting, in relation to this project, were: SID 5 (Rev. 3/06) Page 5 of 23 Ecosystem models should capture the effects of land-use changes on biodiversity (defined either as the population dynamics of key indicator species, as species richness or as species dominance). Landscape permeability (i.e. the dispersal capability of species between habitats) is an essential component to be included in any ecosystem modelling approach. Ecological interactions between species should be included in future modelling projects, as this is a major omission that limits existing models. Ecosystem models should generate predictions that provide information that is better than expert opinion. An ecosystem modelling framework should identify crucial information gaps and suggest how they can be filled. An ecosystem modelling framework should determine current data availability and the priorities for further data collection. Following these meetings, the project team finalised the scope and aims of the project. It was felt that a wide range of modelling approaches that could be used for modelling the effects of land-use on biodiversity already existed (described and reviewed in Objective 2), and that these could be integrated and selected to provide the required approaches to solve specific problems. However, selecting the right approach would depend on the specific problem or question that needed to be addressed. It was decided that the project should focus on developing a generic framework to select and develop models to determine the effect of land-use changes on biodiversity. Having provided a clear focus for the project, the literature was examined to determine the existing models and modelling approaches for the three trophic levels, identified at the first steering group meeting, along with issues and modelling approaches for integrating land-use changes with species distributions. The results of the survey of the literature and a critical review of the approaches are described under Objectives 2 and 3 below. Summary of Objective 1 Following discussions with the steering group and with Defra policy makers, the scope of the project was defined as developing a generic framework to select and develop models to determine the effect of land-use changes on biodiversity, where biodiversity can be defined either as species richness, species dominance or in terms of the population dynamics of key indicator species. Objective 2: Critically review selected ecosystem scale modelling approaches to determine those that can predict policy effects on habitat and biodiversity changes and their impact on farmland birds, using the criteria defined by an expert steering group. The aim of this objective was to determine the current state of play with regards to existing models and approaches that are used to model the dynamics of key components of biodiversity. The reviews were done on a trophic level basis, dealing with vegetation, insects and birds, respectively, as these are the major trophic levels within farmland ecosystems (as suggested by the steering group), particularly in relation to the farmland bird indicator species. Vegetation dynamics Published models to describe vegetation dynamics currently fall into three broad categories. 1. Local scale, species specific or life stage focussed models 2. Regional scale species specific models 3. Regional and global generic models Local scale, species-specific or life stage focussed models The first category of models usually describes a particular stage in the life cycle of the species of interest. Life stages may include the germination and emergence response to biotic variables, the growth and competitiveness of a species (either in single or multiple species mixes) or the fecundity of a species. By their very nature these types of vegetation models are very focussed on specific biological or ecological questions. The parameters for these models are often derived from detailed experimentation and can be very specific in their relevance not only to a single species, but sometimes to sub-populations within that species. Several attempts have been made to link together life stage models to develop true population dynamic models for a species of interest. The species may warrant such detailed effort for example because of its damage to a habitat as a threatening invasive species, because of problems (economic, harvesting or contamination) caused within a particular cropping system or even because the species itself has some intrinsic beneficial role within a system or some rarity value worthy of conservation. Linking together detailed life-stage sub-models to develop population dynamic models can entail the inclusion of a great deal of parameters as reviewed by Holst et al. SID 5 (Rev. 3/06) Page 6 of 23 (2007). Whilst these models undoubtedly offer the most detailed insight to the processes within the development of a plant population over time, their complexity is also their downfall. Introducing possible errors within each component sub-model can lead to over-parameterisation and a confounding of errors within the dynamic model (Grundy, 2003). Many detailed population dynamic models remain largely an academic exercise to aid the development of a better understanding of how a species behaves. However, in some cases the development of these models has been specifically triggered by an issue questioning the sustainability of a vegetation management regime and has had industry application and uptake. For example, the response of an economically problematic weed species to a rotational system (Buhler et al. 1997; Heggenstaller & Leibman 2005). Alternatively they may be used to predict the evolution of herbicide resistance within a given control regime and to devise tactics for avoidance – for example the “RIM” model developed for the management of herbicide resistance in Lolium rigidum (Pannell et al., 2004). Detailed population dynamic models for specific species are often targeted at the field scale, but sometimes even within a sub-division of a field (Buckley et al 2002). For these models to be manageable and user friendly, certain assumptions and pragmatism are often necessary where detailed parameterisation is unavailable. However, cases where these detailed, targeted and largely species-specific models have included biotic factors, such as soil type and nutrient status or interactions with other trophic levels and climatic variables are extremely rare within the literature (the only reference found being Buckley et al 2002). Regional scale species-specific models Some plant species are important on not just a local but a regional scale. These species might include invasive species. Vegetation dynamic models may be developed to predict the invasion success of exotic species or the stability of a wetland species within a given landscape. These models usually include some spatial element such as Geographic Information Systems (GIS). (e.g. Townsend Peterson et al., 2003) Two contrasting approaches to these types of models are often seen in the literature. The first approach is to develop vegetation dynamic model predictions based on climatic variables using logistic regression to identify environmental envelopes and hence potential ranges in novel, yet to be colonised, areas. Unfortunately, these reaction-diffusion (R-D) models often fail to consider spatial heterogeneity or other ecological processes other than dispersal (Higgins et al., 1996). The second approach is to develop individual-based models that account for competition between native and introduced species. This approach makes some classification of vegetation using plant functional types to describe behaviour traits. These models tend to place greater emphasis on a wider range of ecological interactions than simply dispersal. Hence they may have some potential in predicting biodiversity outcomes on a trait-based, if not species-specific level (Higgins et al. 1999, Goslee et al. 2006; Grime and Hillier 2000). The spatial grain (scale) of both of the approaches used in these regional scale models is critical to their success. It is imperative that the spatial grain of the models must be compatible with the spatial processes being modelled. In other words, if the information is to develop such a model is only available at a very coarse level, then it is inappropriate to attempt to use this information to make predictions on a much finer spatial scale and vice versa. Regional and global generic models The final broad category of vegetation dynamics models are those developed to predict vegetation patterns on a regional or even global scale. These models are again spatially explicit and often referred to as Dynamic Global Vegetation Models (DGVM). They are appropriate to describe broad-scale patterns in vegetation with for example climate change (Arora & Boer 2006; Hughes et al., 2006). These models frequently use bioclimatic parameters to determine the distribution of plant functional types (Woodward & Lomas, 2004) within climatic specific envelopes. They may use a combination of models to assess changes in ecosystem vegetation types (for example BIOMEUK) with models to assess changes in the distribution of dominant species or species with particular biodiversity interest (for example SPECIES, Berry et al., 2002). SPECIES, and models like it, use neural networks to “train” the model on existing data on species distribution to make predictions. For example, detailed data collected on the distribution of a species in one country could be used as a training data set to then make predictions of the distribution of that species in another country. The advantage over the previous types of models described, is that detailed physiological data on individual species are not required. Because of this they have great potential to make predictions regarding vegetation dynamics on a generic level. The main disadvantages are that these models by default require very large data sets to train and optimise the model before they can be used to make predictions. In addition, neural networks are something of a “black-box” approach, and allow no insight into the importance of the individual variables that are driving the dynamics of the populations. SID 5 (Rev. 3/06) Page 7 of 23 The main advantages and disadvantages of these models and modelling approaches are summarised in the table below. Model categories Local scale species-specific models Advantages Provide an excellent understanding of the biological and ecological processes and interaction driving vegetation Disadvantages Limited by sufficiently detailed data Danger of overparameterisation Specificity of parameters restrict the wider applicability Regional scale species-specific models Provide a level of scale appropriate for looking at how biodiversity may change over a Landscape scale Regional and global generic models Able to help guide global policy decisions Limited by not having data available at correct scale Few account for natural biotic interactions and are mainly driven by dispersal alone Limited by lack of detail for looking at biodiversity No transparency of ecological processes Ideal Generic element to the modelling approach to allow broad application to multiple species. Data available at sufficient level of detail suitable for application at a regional or landscape scale accounting for heterogeneity. Some insight to natural or biotic interactions Avoidance of over parameterisation Insect dynamics There are few models describing insect species interactions in large heterogeneous systems (Brewster & Allen, 1997). Most of the modelling effort has been devoted to understanding the dynamics of insect populations in cropping systems and the majority of these insects have been crop pests. Studies have focused on single species rather than insect assemblages. For example, Cocu et al. (2005) used multiple linear regression to identify the effects of geographical locations, climate and land use on the abundance and phenology of the pest aphid, Myzus persicae. Potting et al. (2005) used simulation modelling to investigate the factors in agro-ecosystem diversification that influence pests. Behavioural factors that influenced the spatial dynamics of herbivore populations were the colonisation pattern, movement speed and sensory mode of finding host plants. Levine & Wetzler (1996) undertook simulation modelling to determine the effect of habitat architecture (host-plant dispersion, density and abundance) on herbivorous pest species with different searching strategies. Bhar & Fahrig (1998) built a stochastic, individual based simulation model of a specialist insect in a landscape of a number of crop fields. Brewster & Allen (1997) developed a framework for a spatiotemporal model that can be used to study insect dynamics in regional (heterogeneous) systems. It integrated ideas from spatially explicit population models, coupled map lattices and integro-difference equations. In the spatially explicit population models, information on the environment of the organisms is incorporated explicitly in the model, which combines an insect population simulation model with a habitat map. There are relatively few studies that have attempted to consider the effects of land management on insect assemblages. Using Principal Component Analysis, Benton et al. (2002) showed that there was linked temporal change between farmland birds, invertebrate numbers (from suction trap data) and agricultural practice (in Scotland). Multivariate measures of arthropod numbers were associated with multivariate measures of farming practice in the current and previous year. Studies explicitly considering the relative roles of plant species composition and habitat area in determining the structure of insect assemblages found in different habitat fragments are relatively recent (Crist et al. 2006). Crist et al (2006) used a multiple regression approach to analyse data on the effects of habitat area, edge and plant community composition on the spatial structure of insect species richness and composition. Collinge & Forman (1998) developed a conceptual model of landscape change that focuses on habitat spatial arrangement and tested its predictions with a field experiment on grassland insects. SID 5 (Rev. 3/06) Page 8 of 23 Walters et al. (2000) used ecosystem modelling to evaluate management policies in the Grand Canyon. They concluded that it was possible to make fairly accurate predictions about some components of ecosystem response to policy change, including insect communities, but that it was less possible for other components of the ecosystem. The main advantages and disadvantages of these models and modelling approaches are summarised in the table below. Model categories Simulation models Advantages Provide an excellent understanding of the biological and ecological processes and interaction driving insect dynamics Disadvantages Limited by sufficiently detailed data Danger of overparameterisation Limited to crop pests Statistical modelling Demonstrate linkages between trophic levels or habitat and environmental variables Difficult to extrapolate beyond data collected Mainly limited to crop pests Lack of suitable data Ideal Generic element to the modelling approach to allow broad application to multiple species. Data available at sufficient level of detail suitable for application at a regional or landscape scale accounting for heterogeneity. Some insight to natural or biotic interactions between species and other trophic levels Avoidance of over parameterisation Need to consider all insects and not just economically important species (insects as consumers of all plants or as food resources for bird species) Farmland bird dynamics The majority of work on farmland birds has been focused on auto-ecological studies, monitoring and experimental work, with only a small number of papers that focus on utilising modelling to integrate and synthesise the large body of data on farmland bird ecology. Underlying this to some extent, is the fact that the majority of work has been done on single species in specific locations, which makes it difficult to generalize the findings to a wider geographical area. Stephens et al (2003), reviews 7 different approaches to the modelling of farmland bird populations. The modelling approaches are split into two categories: phenomological models and behavioural models. The advantages and disadvantages of these two categories of modelling are described below: Phenomological models This modelling approach develops models based on empirical relationships between populations and their environment. As long as the environmental conditions remain within the range observed experimentally then predictions can be made, but these models are unable to make predictions outside the range of data observed. The first type of models in this category are aggregative models that relate the abundance of birds to food density, and can provide reasonable predictions about habitat use for bird species that show a clear aggregative response to food, as has been done for skylarks (Watkinson et al., 2000). The major problem with aggregative models is that it can be difficult to demonstrate aggregative responses empirically due to problems in establishing exact food availability, and ensuring that the dominant food source is actually consumed by the bird species. Aggregative models explain only the presence of a species within a habitat, and provide no link to the demography of the species, which means that these models provide only limited inference about the responses of populations to food availability and cannot be used to predict the effect of a habitat change on the population of a species. The second types of model within this category are population models (Pennycuik, 1969; Potts & Aebischer, 1991). These models are based on experimentally determined rates of fecundity, mortality and, to a limited extent, dispersal. However, for population models to be used to predict the effect of changes to land-use, the models need to link demography to the availability of food, which is not often done, but was done in the case of the grey partridge (Potts & Aebischer, 1991). A crucial limitation to population models is the availability of appropriate data that links demographic parameters to food availability, and Stephens et al. (2003) state that “for farmland birds there are very few published examples of such work”. However the main limitation is that population models cannot be extrapolated outside of the range of data used to parameterize the model with confidence, which means that they should not be used to predict the effect of habitat or land-use changes. Behavioural models SID 5 (Rev. 3/06) Page 9 of 23 Behavioural models are focused around describing the behaviour of individuals within a population and therefore have a much greater potential to be used to make predictions about the effect of changes in habitat or land-use on population dynamics. Depletion models focus on the interactions between individuals and their food supply, and there are many approaches used for this type of model, which gives prediction on the use of habitat (Atkinson, 1998), but does not link habitat use to demography. Due to the need to have a good understanding of the use of resources, behavioural models are best developed for species that use a limited set of resources. The current limitation to the development of these models, which can be extremely useful tools, is the lack of sufficient data to develop and parameterise the models. The main problem with all of the models described in Stephens et al. (2003) is that they are species-specific, describing the population dynamics of a single species, and do not consider the interactions between species, which limits their use within a biodiversity context. In addition the phenomological models, being empirical, may not be representative of the general behaviour of populations, outside of the area for which the data were collected. It is also pointed out by Stephens et al. (2003), that the major limitation for all of the models is that they do not link the demography of the bird populations to the food availability, which makes use of these models to predict the impact of land-use changes on bird populations problematic, as changes in land-use will undoubtedly affect food availability, and possibly also nest site availability. Stephens et al. (2003) clearly state that “Available data severely limit the range of approaches for predicting the response of farmland bird populations to changing food supply.”, which is a major problem for the development of models to predict the effect of land-use changes on biodiversity, as without sufficient robust data to develop the assumptions about movement and resource use, there will be an unacceptable level of uncertainty within the models. The models described above fall into the scientific category of models, in that they are being used to explain behaviour and population dynamics in detail. For the development of policy or the implementation of policy a more simplistic type of modelling could potentially be used. There are very few examples of models being developed for policy development or implementation, but those that do exist are reviewed below. Siriwardena & Vickery – Defra project (BD1618), Chapter 1. This project was concerned with predicting the response of farmland birds to agricultural change, and chapter 1 describes a population modelling approach to predict future population trends of farmland indicator species to determine the changes in population demography required to reduce or reverse the decline of the species. Although chapter 1 deals with all farmland indicator species, a separate model for the grey partridge is provided in chapter 2, and is not considered within this review, which relates solely to chapter 1. The models are simplistic, using multiple regression to develop models to predict the historic changes in populations of each species separately. Because the initial fits did not match the trends appropriately, the models had to be modified to provide better fits. The need for these changes was suggested to be due to biased data from the common bird census, and mis-matched data types. Therefore, the model predictions have a large degree of uncertainty linked to them, which is not stated, meaning that the models should be used with caution when predicting the changes needed to alter the demography of bird species, to reverse or reduce decline. The model is based on data from a wide range of habitats, and does not take any account of habitat use, or have any link to the availability of food or nesting resources. The lack of geographical consideration is another potential problem, since populations in different geographical locations may be responding in different ways, and by averaging this out in the model, through the use of census data, significant errors could be introduced into the predictions. This model does provide predictions across a wide range of species, but is limited to only describing demography, and would not be suitable for describing the effects of land-use changes on bird populations. Swetnam et al. (2005) – Designing lowland landscapes for farmland birds: scenario testing with GIS This paper develops models to examine habitat variation based on predictor variables, which are then used to determine the effect of changes in land use on territory density. Models were developed for two species, the skylark (Alauda arvensis) and yellowhammer (Emberiza citrinella). For the skylark, the model is based on data recorded in 100 fields over 3 years, and takes the form of a generalised linear model (GLM) for counts, which has been optimised using backwards selection to identify the minimum adequate model. The model for yellowhammer was developed using the similar approach (described in detail in Whittingham et al. (2005)) of binary logistic regression. The models predict either the number of territories expected in a field (skylark) or the probability of finding a territory in a field boundary (yellow hammer). Several potential land-use change scenarios were examined using the models, and the models were able to identify which scenarios would have either a positive or negative impact on the two species concerned. The advantage of these models is that they link use of landscape to the presence of the bird species. However, their major disadvantage is that there is no link between territory occupancy and demography. The models are also SID 5 (Rev. 3/06) Page 10 of 23 based on real data, and so they cannot predict beyond the range of landscape conditions found within the data used to parameterise the model, limiting their ability to be used predictively. In addition, the models predict the potential number of territories, based on habitat conditions and not actual territory occupancy. There are many possible reasons why a suitable habitat would not be occupied, including: inability to disperse to the area, lack of suitable feeding sites (the models use mainly landscape architectural features as predictor variables), and predation pressure. The authors recognize the limitations of the model and state “Ideally, habitat association models would be combined with data on species’ demographic rates (e.g. reproductive success and survival) and their density dependent relationships to create population dynamics models, and with data on dispersal behaviour and response to landscape-scale environmental variation to generate spatially explicit population models.” As mentioned previously, the problem in doing as the authors say, and developing spatially explicit population models, is the lack of suitable data on demography, resource use and dispersal. Butler et al. (2007) This paper is the most recent and most interesting attempt to develop a model that can be used to predict the effect of land-use changes on bird populations. The model uses a matrix of ecological requirements covering diet, foraging habitat and nesting habitat for each species. The assumption is that land-use change will affect a species if it leads to change in food abundance and/or a change in nesting success. In this regard, the model includes the components that are necessary to describe the impacts of land-use changes on a bird species. The output of the model is a risk score for the species, which can be related back to the annual population growth rate, and as such predicts the impacts of land-use change on population growth rate. The ability to predict the impact of land-use change on annual population growth rate is ideal for determining the effects of policy on biodiversity. However, there are some drawbacks to this model. The first is that the risk score is based on expert opinion and ranking, rather than on actual data, and so is effectively a subjective estimate of the effect of a particular change on the bird species. Secondly, the model predicts only for population growth at a national level, and does not consider potential regional differences. The model does not distinguish between different types of habitat, and therefore averages across all habitat types, which could mask potential local and regional differences in population growth rates, which are affected by particular local/regional conditions. As a risk assessment framework, this model is a valuable resource to policy makers, but for predicting the impacts of land-use changes on biodiversity, the model does not fully account for the vital link between habitat and resources, and the fact that the majority of species use a wide range of habitat types. The model could be improved by including greater detail about habitat types and the link between these and resource availability. It would also be beneficial to include more local and regional level predictions. Overall, the current status of modelling of bird populations can be summed up as being severely limited by data availability. Many model approaches exists that could be utilized to develop models of bird biodiversity, and the impact of land-use changes on biodiversity. However, to fully understand how land-use changes may affect bird biodiversity there is a need to move towards models of multiple species that link habitat use to demographic parameters. The most promising approach would be to link several of the models types together to remove the disadvantages of a single model type, but to keep the advantages of the different types so that all factors affecting bird biodiversity are included in the model, e.g. linking habitat association models with population or behavioural models, so that the effect of habitat type on dispersal and movement is explicitly considered within the models. The main advantages and disadvantages of the bird models and modelling approaches are summarised in the table below. Model categories Aggregative models Population models SID 5 (Rev. 3/06) Advantages Link food abundance to demography Based on experimentally determine data (fecundity, dispersal, mortality) Disadvantages Limited by sufficiently detailed data Difficult to prove link between food resource and demography Poor linkage of demography to food availability Cannot be extrapolated outside range of Page 11 of 23 Ideal Generic element to the modelling approach to allow broad application to multiple species. Data available at sufficient level of detail suitable for application at a regional or landscape scale accounting for heterogeneity. Some insight to natural or biotic interactions between species Avoidance of over parameterisation Behavioural models Focusses on individual behaviour Include link between behaviour and resource availability Habitat association models Link landscape factors to potential presence of a species Risk framework Links changes in habitat to demography Provides predictions across a wide spatial scale data Limited data availability Limited links between resource use and demography Lack of sufficient data for parameterisation No inclusion of demography or dispersal Only potential presence, not actual presence Cannot predict beyond range of observed data Based on subjective expert opinion Does not account for landscape heterogeneity No data-based link between habitat use and demography Linkage between habitat use, resource use and demographic parameters Inclusion of dispersal of individuals Summary of Objective 2 From the reviews of vegetation, insect and bird dynamics models, it is clear that there is a lack of suitable and sufficient data across the three trophic levels, which limits the scope and extent of modelling work that can be done. The majority of models are based on single species, and do not consider multiple species interactions, which are important in the context of biodiversity. There is a need to integrate across the trophic levels, and to develop models that are able to integrate the spatial heterogeneity of the landscape, and link this to resource availability, resource usage and the demography of a range of species. A wide diversity of modelling approaches exist to tackle the modelling of specific aspects relating to the different trophic levels, but a framework is needed to ensure that the appropriate approaches are combined and used to determine how land-use affects biodiversity. The choice of modelling approach will depend very much on the question being asked, and the level of detail required to answer the question. To that end the framework must provide a clear set of questions that need to be answered (using appropriate modelling approaches) to determine how land-use affects biodiversity. The integration of modelling approaches and data from a range of scales will be essential in providing a useful framework to address questions relating to land-use and biodiversity, and the potential approaches for this integration are described in Objective 3 below. Objective 3: Determine how to integrate data across a range of geographical scales to allow ecosystem modelling approaches to predict changes at regional and national scales. The aim of this objective was to determine the approaches that exist for integrating either existing models or modelling approaches across a range of geographical scales. In considering this objective we have identified four distinct types of integration that need to be considered in the development of ecosystem models: spatial integration, trophic integration, temporal integration, and data integration. Below we define what we mean by each type, and identify the challenges and problems associated with each. Spatial integration SID 5 (Rev. 3/06) Page 12 of 23 Models describing the population dynamics of a particular plant, invertebrate or bird species (or even a community of different species) will often be limited in their application and appropriateness to a particular location and habitat, usually within a defined physical area (e.g. a single field, a catchment, or a region). Where interest is in the detailed scientific understanding of a particular ecosystem, prediction of the effects of management changes on the biodiversity within this defined area may be all that is required, but for the purposes of assessing the impacts of policy changes or for determining the effects of different implementation approaches, the integration or aggregation of information across a number of adjacent spatial locations will be necessary. Where data and models are available to predict the response of a particular species in each spatial location, then, at a simplistic level, integration to predict the response over a collection of neighbouring spatial locations may just require the aggregating of the responses across the appropriate set of individual locations. However, this does not take account of the effects that neighbouring spatial locations might have on the response in each location – How should we incorporate the effects of the habitat in these neighbouring locations, or of changes in population dynamics of the species being modelled in these neighbouring locations? Over what distance do neighbouring locations have an impact? The answers to these questions will of course depend on the plant, invertebrate or bird species that we are modelling, and also on the spatial scale for which the model has been constructed and is appropriate. Spatial integration of responses becomes more complicated where a community of several species is being considered, and where some summary measure of biodiversity is the response of interest. Here the biodiversity response over the larger spatial area may not be just some simple summary (e.g. arithmetic mean) of the biodiversity responses across each of the constituent areas. One form of integration would require the aggregating of information about each individual species across the individual spatial areas, followed by the recalculation of the biodiversity measure, to provide a single summary for the larger spatial area. But this loses information about the (spatial) variability in biodiversity within the larger area. An alternative approach might be to consider the mean and variance (or some other summary statistics) of the values across the set of individual spatial area, possibly weighted by the size of each area, or taking account of the spatial location of each area. Again, the integration may also need to take account of how neighbouring locations affect the response in a given location. The issue of spatial scale and spatial integration should also influence decisions made in developing the ecosystem model: On what spatial scale should we collect data? This may vary between trophic levels, and also be affected by the purpose of the modelling exercise, the type of model being developed, and the ease of data collection. It will affect the precision of predictions produced by any model developed from the data. On what spatial scale should we develop the models? Again this will be affected by the purpose of the modelling exercise, as well as by the required precision of response, and the availability of data. On what spatial scale should we interpret or predict the response to changing management? This is primarily affected by the purpose of the modelling exercise, but will also be influenced by the availability of data, and the spatial scale at which we are able to build models. Trophic integration Whilst most existing models only describe the population dynamics of single species, to be able to predict changes in biodiversity ecosystem models must include multiple species and the interactions between them. Some existing studies use empirical modelling approaches to directly model the value of some biodiversity index in terms of a range of potential explanatory variables, but this approach causes serious problems if trying to integrate the results over a larger spatial area (see above). A more ideal, but certainly more time consuming and intensive, approach is to develop detailed models for each individual species, incorporating interactions between species, and then to combine the predictions from these individual models to produce a combined measure of biodiversity. These interactions between species take a number of different forms. Within trophic levels they will often primarily be competitive – plant-plant competition for light, water, nutrients; competition between invertebrate species for food; competition between birds for food and nesting locations – and will often be both within and between species. Other forms of interaction include the predation (or feeding) of one species on another, usually, but not exclusively, between trophic levels. Of course these different interactions operate at different spatial scales, so that the combination of species included in an ecosystem model will influence the choice of spatial scale for data collection, modelling and prediction – ideally the chosen spatial scale for the model needs to encompass the largest spatial interactions between species, whilst being based on data at the smallest possible spatial scale. Temporal integration By their very nature, population dynamic models must consider changes in the modelled population over time. Therefore, in integrating such models to predict the effects of management changes, it is important to consider the time course over which the ecosystem model operates and that over which predictions will be made. Key SID 5 (Rev. 3/06) Page 13 of 23 elements include information about the time course over which management changes are implemented, and the time period during which changes in outcome (biodiversity, species richness, or individual species density) will be considered to be important. Of course not all biological processes operate over the same time scale, though many of the processes of interest will have an annual cycle. Thus it will be important to understand the time course over which each species completes a cycle when combining individual species models and assessing the interactions between species. A second key area of temporal integration relates to the prediction of outcomes, whether biodiversity, species richness or individual species density. Most observations of these measures will be at a single point in time, and so it is likely that constructed models will produce predictions on the same basis. However, these measures are likely to change between different times of the year, and so it will be important to consider some sort of aggregation of the measure of interest over a given period of time (the choice of this period may depend strongly on the question being addressed). Data integration In building models to address policy issues, it is likely that data from a wide range of sources will be used. As already indicated under “spatial integration” above, different data components will often have been recorded at different spatial scales or levels of spatial precision. Incorporating these different data sources will therefore require either the aggregating of data from a smaller spatial scale to provide an appropriate value for modelling at a larger spatial scale, or some assumption about the distribution of a variable across all sub-units contained within the larger unit at which the data were recorded. A second form of data integration becomes apparent when considering the building of large scale models using individual species models as components. Whilst many individual species models have been constructed at a fairly detailed level, it is likely that the incorporation of such models into a more general model will require these models to be simplified and generalised to require fewer estimated parameters. An important process in the development of the large scale models will then be to assess the sensitivity of the overall outcomes to these simplifications. Related to this idea of assessing the sensitivity of overall outcomes to the choice of individual species models, is the concept of including stochastic elements in the model. This may be as simple as drawing individual parameter values from statistical distributions for each simulation run (Monte Carlo simulation), or as complex as allowing a probabilistic choice of the form (mathematical function) used for difference model components. These stochastic model elements are likely to be driven by the analysis of data using empirical modelling approaches, with the incorporated stochasticity reflecting the observed variability in parameter values or input variables, or the range of possible model forms for individual components. Incorporation of such stochasticity allows model outcomes to be presented with associated measures of uncertainty, and hence to allow an assessment of the risks associated with the plausible scenarios evaluated using these models. Modelling approaches Having identified the need for the integration of models and data in these four different ways, a wide diversity of mathematical and statistical modelling approaches were identified that have been used to tackle such integration, with the specific approach required being dependent on the focus of the modelling exercise, and, more importantly, the available data (the lack of appropriate data generally appears to be the major constraint on any model building exercise). There is little value in listing the range of approaches that could be used, though these range from highly empirical statistical modelling methods to techniques based on a detailed knowledge of the mechanisms driving observed processes. Instead we choose to define a modelling framework which would assist both in the selection of appropriate modelling techniques and in the identification of the available and necessary data sources, according to the outcome information required to address any particular policy question. Thee proposed modelling framework, which we believe should form the backbone of any modelling work to determine how plausible changes in land-use might impact on biodiversity, is defined by a set of five key questions: 1) What is the current spatial distribution of the species of interest? 2) What is the potential spatial distribution of the species of interest given the changes in land-use? 3) How much of the potential spatial distribution can the species of interest achieve (within a defined time scale), taking account of habitat and landscape permeability? 4) Are there sufficient resources to allow the species of interest to establish in the new locations (within a defined time period)? 5) Will the species of interest establish, and, if so, how will it affect the current community? (ecological interactions) Whilst the questions above all refer to a single target species of interest, they could equally be applied to a community of species, summarised using some measure of biodiversity. SID 5 (Rev. 3/06) Page 14 of 23 The precise modelling approaches used to answer these questions are not defined, as these will depend very much on the question being asked, and the assumptions that are being made within the model. It is not essential to answer all 5 questions, since the usefulness of each question will depend on the information required to inform policy development. The progression through the questions requires increasing detail to be included within the model. Questions 1 and 2 can be answered using quite broad scale data, and for relatively large geographical regions. Question 3 requires more localised data to provide useful answers, as it is concerned with determining the movement of species and will therefore be subject to variation due to the localised habitat composition (although it may be possible to generalise the approaches to provide information over wider scales). Questions 4 and 5 will certainly require information specific to particular locations, since they involve questions about how a species interacts with the local environment. Providing answers to all 5 questions listed above will provide the most useful and detailed information on how land-use changes will affect biodiversity, which can then be used to inform or develop policy on particular issues. Existing approaches Approaches exist to answer the first two questions of the framework, within the context of conservation biology, and these are described below: Prediction of species distributions The main approach used to predict species distributions is that of a multiple regression using either Generalised Linear Models (GLM) or Generalised Additive Models (GAM), and is described as landscape modelling. A large amount of work on characterising species distributions has been done in Australia, where a number of suitable datasets already exist to allow this form of characterisation. Ferrier et al. (2002b) developed a species-level approach which combined mapped and modelled layers within a Geographical Information System (GIS) describing topography, climate, substrate (soil type), vegetation cover and disturbance. The information was mapped at a range of scales from 1:25000 to 1:100000. This information was combined with floral and faunal surveys. The models used GAM-based logistic regression to describe the species distributions. From the statistical relationships, the distribution could then be extrapolated spatially to regions outside the areas where the surveys were done. The models predicted three probability surfaces within the GIS: the probability of occurrence together with upper and lower 95% confidence limits for this probability. The upper limit is relevant for areas where there is a need to have high confidence that a species exists, whilst the lower limit is relevant for areas where there is a need to have high confidence that a species is absent. This modelling approach is fine where datasets record presence and absence of a species, though it can be difficult to reliably record the absence of a species. If a dataset only consists of presence data, then pseudo absences need to be generated, and these need to be appropriately weighted within the regression to reflect the fact that they are not recorded absences. This basic modelling approach does assume that the presence of a species is driven solely by environmental characteristics or habitat suitability. It does not account for the heterogeneity of the landscape, assuming that two areas of the same habitat type can support the species linked to that habitat type, irrespective of their geographic separation. It should be noted that many species exhibit a spatial pattern in their distribution, not predicted simply by variation in habitat. The models were then adapted to incorporate the spatial configuration of habitat in cells neighbouring the suitable habitat cell, using a set of ‘contextual indices’, with an inverse distance weighting (similar to the approach used to measure habitat neighbourhoods in metapopulation models). The approach provided better estimates of species distribution, but did rely heavily on expert knowledge rather than actual data in determining the effect of neighbouring habitats. This is potentially problematic – Pearce et al. (2001) showed that incorporating habitat indices defined by expert opinion did not improve the predictive accuracy of models of species distributions. They also showed that models based solely on expert opinion were less accurate than models based on the statistical modelling of quantitative data. The approach described above for the species level was then extended to the community level (Ferrier et al., 2002a), the premise being that the community level could be used to derive coarse-filter biodiversity surrogates for use in regional conservation planning. They used two approaches to develop models and predictions of communities, a ‘classification-then-modelling’ approach and a ‘modelling-then-classification’ approach. For the former, the data are classified to derive groups of sites containing similar species or groups of species. These derived data are then modelled using a GLM or GAM approach and the fitted models extrapolated for the wider group of species. The latter approach requires that each individual species distribution is first modelled using a GLM or GAM approach, and then groups of grid cells with similar predicted species or groups of species can be identified. SID 5 (Rev. 3/06) Page 15 of 23 It was shown that species-groups provided better surrogates than site-groups, and that the species-group modelling approach provided similar results to individual species modelling. The approaches described above require large amounts of data, which may not be available within the UK. However, Ferrier et al. (2002a) recognised this potential problem, and suggested that rather than modelling individual species, approaches could focus on modelling collective properties of biodiversity. Species richness is one such property, but has limitations, since conserving areas with the highest species richness may have limited value. Additional information is required in the form of compositional dissimilarity (how dissimilar two areas are in terms of the species they contain). Ferrier et al. (2002a) discuss a matrix regression approach that can be used to predict the dissimilarities between sites, and call the approach Generalised Dissimilarity Modelling (GDM). This approach may well be useful in assessing how land-use changes could affect species distributions and hence biodiversity, by considering the similarity of the effects of land-use changes at different sites. Virkkala et al. (2004) used a GLM approach to study the effects of landscape composition on birds in a boreal agricultural-forest mosaic in Finland. The model explained 49% of the variation in the total density of farmland birds. They showed that a grid-based approach is effective in modelling bird species distributions as it can take account of variation in habitat structure and other topographical factors. Landscape and environmental variables have been used to predict species distributions in the UK as part of the MONARCH project (Berry et al., 2005). The SPECIES model used in MONARCH uses an artificial neural network (ANN) to characterise bioclimatic variables through a climate-hydrological process model. This is used to identify climatic regions within which species can exist, and to predict changes to their distribution based on these climatic envelopes. A GAM approach is used to model changes in key vegetative land cover variables and the system then uses a cellular automata modelling approach to predict species dispersal based on the climatic and vegetative characteristics of neighbouring cells. This model is then used to predict the probability of occurrence of each species in each location. The main problems with this species modelling approach are the lack of any spatial autocorrelation when considering the dispersal of species and the lack of suitable data, but the approach provides a possible framework for the modelling of the effects of land-use changes on biodiversity. The use of landscape and environmental variables as surrogate ecological indicators is currently being questioned. It is suggested that the surrogate landscape measures lack a generic applicability to a wide range of species and may not have biological or ecological significance (Lindemayer et al., 2002, Austin, 2007). Lindenmayer et al. (2002) suggested that the definition of any landscape surrogate measures should be qualified by an explicit statement about the question being addressed and the biology and ecology of the target organisms being modelled. Austin (2007) presented a similar message, but came up with a set of rules that should be applied to any modelling of species distributions, to make the models as robust as possible. These rules essentially ensure that the assumptions being made are explicitly defined, and that ecological knowledge is used when selecting the predictor variables from landscape and environmental surrogates. Prediction of movement, based on habitat permeability The models described above generally produce potential distributions for individual species or groups of species, but, apart from SPECIES, they take no account of the patterns of dispersal of the different species. To model the effects of land-use changes on biodiversity, it is essential that these dispersal patterns are taken into account, so that the current and potential distributions can be combined to provide realistic estimates of the actual species distributions under a range of plausible scenarios, including the likelihood of a species attaining that distribution within a defined time scale. Currently there is little available information about linking dispersal patterns with current species distributions. One possible approach is based on habitat permeability, defined as the ability of a species to move through a particular habitat. There is a large ecological literature on the permeability of boundaries, but this is largely theoretical, and, although a useful resource, of limited applicability to biodiversity modelling. There are two key published sources that link landscape permeability with species distributions, and these are described below. The first example investigates the occurrence of two amphibian species in Switzerland (Ray et al., 2002). A landscape map was used to derive permeability estimates for each species for a range of habitat types between the breeding ponds of toads and newts. The approach used was to generate a friction map, a matrix of cells that define the energy cost for crossing each cell. A conversion table is then used to assign a migration cost to each habitat type in the landscape. In this example, migration costs were linked to both energy expenditure and the risk of mortality in crossing a habitat. If the organism is given a set budget for crossing habitats, those routes where the cost exceeds the budget can be omitted, and hence migration zones can be defined. Based on these migration zones, the probability that a species can move from its present location to a new location, through the habitat types encountered on the way, can be calculated. This allows a realistic actual distribution to be estimated, with those areas of the potential distribution omitted that could not be reached. Ray et al. (2002) compared the friction map approach with the standard circular migration zone approach, and showed that the friction map approach led to better predictions of the occurrence and dispersal of newts and toads. SID 5 (Rev. 3/06) Page 16 of 23 Another example where potential distributions have been modified by dispersal models is the set of BEETLE (Biological and Environmental Evaluation Tools for Landscape Ecology) tools, developed by Forest Research (www.forestry.gov.uk/fr/infd-69pla5) (Watts et a.l, 2005). This set of tools is aimed at assessing the effect of forest fragmentation on species distributions and currently models ecologically representative generic focal species. The approach uses the habitat area requirements and dispersal characteristics of the species to determine how changes in fragmentation associated with land-use affect the species distribution. The model is based on using functional connectivity, defined as a functional attribute of the landscape related to ecological processes such as movement and dispersal. The functional connectivity is a combination of the dispersal ability of a species and the permeability of the landscape (similar to the friction maps of Ray et al. (2002)). The ALMaSS Model Topping et al. (2003) have developed a modelling approach that addresses many of the problems mentioned under objective 2 and in the section above. The modelling approach uses a modular modelling system that is specifically designed to be “a predictive tool for answering policy questions regarding the effect of changing landscape structure and management on key animal species in the Danish landscape”. The model consists of a core landscape model that provides information to be used by the animal modules included in the model. The data on landscape is stored as landscape type (e.g. arable field, coastline, quarry, road, river, etc) for each gridsquare, and the model can represent an overall area of 10km by 10km at a 1m 2 resolution (100 million grid squares). The model incorporates weather information, and also includes standard management practices for six farm types. The animal models have been developed for five species: roe deer, field vole, skylark, ground beetle and a spider, which represent key ecological types. The animal models are agent-based models of behaviour and use rules for transition between behaviours, based on certain conditions being met (using information from the landscape, weather and farm management modules). The movement of species is incorporated into the model, and the rate of movement is weighted by habitat, so landscape permeability is accounted for within the animal models. The model has been successfully used to model the impact of land-management, through scenarios of reduced pesticides on all the species in the model (Jepsen et al., 2005), producing predictions that illustrate the interactions between land-management and species abundance. For the skylark (Topping, 2005; Topping & Odderskær, 2004), the model predicted that skylark abundance would decline under a reduced pesticide scenario due to associated changes adopted by farmers in the cropping regime, with the alternative crops being less suitable for skylarks. However, the model also showed that the greatest benefit to skylarks was caused by an alteration of the structure of the crop to provide access for nesting and feeding. The approach used in the ALMaSS model is limited by data availability, as with all the other models discussed in this report. Currently ALMaSS does not include interactions between multiple species, due to a lack of suitable data. In addition the models are usually run for time periods of 60 or more years, with the first 20 or so years being discounted as being the time required for populations to reach equilibrium. This effectively means that effects occurring during the transitional time period between different land-uses are not considered, but these may be extremely important in determining the future abundance and distribution of species. Also, the ALMaSS model wraps the boundaries of the geographical region being simulated, so there is no immigration or emigration. These are certainly important processes that would need to be considered when assessing the impacts of landuse changes on biodiversity in a UK context. The use of weather data means that to simulate future behaviour, it is either necessary to use historical weather data, which can be looped if insufficient data is available, or to generate plausible future weather scenarios. Another limitation is the high resolution of the model, which not only constrains the geographical extent of the model, but also means that a high level of detail about the vegetation spatial distribution and the animal behaviour and movement is required. The model does not currently include models for vegetation and the interactions between vegetation and land-use. However, Warren and Topping (1999, 2004) have developed models to examine the effect of environmental heterogeneity on the competition between plants, and these models could easily be included within the ALMaSS modelling approach due to the modular structure used. Despite these limitations, the ALMaSS model remains the most suitable existing approach for modelling the effect of land-use changes on biodiversity, and for providing predictions that can be used to develop policy on land-use changes. The ALMaSS approach could be adapted for use in the UK, subject to sufficient data on vegetation, weather, farm management practices, and animal biology and behaviour being available. Whilst the limited geographical coverage of the model would probably need to be addressed, the key concepts of having a landscape simulator, with farm management and weather modules linked to agent-based models of specific species, could be retained. Adaptive Resource Management SID 5 (Rev. 3/06) Page 17 of 23 Another useful concept is the approach referred to as “Adaptive Management” or, in the area of ecological modelling and monitoring, as “Adaptive Resource Management”. The basic idea behind Adaptive Resource Management is that the implementation of management decisions is adapted or modified as additional data, either independent or from monitoring observations following the earlier implementation of management decisions, is collected and incorporated into the model. With reference to the use of ecosystem modelling approaches to support the development and implementation of policy, the Adaptive Resource Management approach implies the need for continuous dialogue between the model builders and policy makers, and a medium- or long-term commitment to the monitoring of, and collection of data from, sites both associated with and independent of the implementation of particular policy management decisions. It must be realised that any model is only as good as the data on which it is based. Thus as model realisations are observed following the implementation of policy, quantitative observations provide a validation or otherwise of the developed models. Whether or not the observations and models are in agreement, the collected data allows the models to be improved, and so better inform future model predictions and the future development and implementation of policy. Summary Integration will be a major issue when developing models to predict the effect of land-use changes on biodiversity. It will be necessary to have clearly defined spatial and temporal scales, taking account of data availability and model aims, to ensure that appropriate modelling approaches are selected. The aims of the modelling and the hypotheses being tested will also need to be precise and unambiguous to ensure that appropriate techniques are used for integrating data. The modelling framework addresses these issues through a set of structured questions that assist in defining the spatial, temporal and ecological information required within a model, together with identifying the required level of detail of this information,. The questions are structured so that progressively more detail and data on landscape and species biology are required to answer each question. This allows policy makers to decide how much information to include in the model, depending on the question or questions they wish to answer. A lot of research has already been devoted to the use of landscape surrogates to predict species distributions across large areas, and, when this is linked to ecological parameters and landscape permeability, some highly useful and sensible predictions about the effect of land-use changes on species can be made, although it may be necessary to consider incremental changes in land-use over longer timescales, rather than just considering the start and end points. The approaches described above could form a set of tools to be utilised within the framework to assess the effect of land-use changes on biodiversity, and to examine how biodiversity can be conserved at a regional or national level. The only aspect missing from the existing modelling approaches is the impact of a species moving into a new area on those species already resident in that area. To ensure a robust and joined-up approach to biodiversity conservation or protection, there is a need to consider multiple species and their interactions, rather than focussing on individual target or indicator species, since although a land-use change may benefit the target species, it could be detrimental to other species. Consideration of multiple species would require models for the ecological interactions between species. Currently there are few if any modelling tools that cope with interactions between species, although possible modelling approaches have been presented within the ecological modelling literature. There is certainly a paucity of data on the interactions between species, and so a significant research effort is needed to collect data on inter-species interactions. Of the existing approaches, the ALMaSS model (Topping et al., 2003) is the most advanced approach, addressing the majority of the issues raised as limitations in this report. The model does have its own limitations, namely a limited geographical range, and the requirement for a high level of detail on farm management, weather and vegetation distribution as inputs to the model. Despite this, the concept of having a modular set of models, where models for different species can be linked to a landscape simulator, is highly desirable, and reduces the cost and time involved in developing a series of models. We recommend that Defra consider adopting an approach similar to ALMaSS for the development of models to address policy questions about the impacts of land-use changes on biodiversity. We recommend that Defra incorporate adaptive resource management concepts into projects where management decisions and modelling are being integrated, so that improved predictions and decisions can be made as more data becomes available. Objective 4: Prioritise the future research necessary to develop ecosystem scale modelling approaches for policy development on farmland biodiversity. At the meeting with Defra policy makers, described in Objective 1, a set of four key questions were identified that the project should address. These questions were: 1) How will land-use changes associated with biomass/bioenergy crops affect biodiversity? 2) How do ELS incentives need to change to promote biodiversity? 3) What would be the most effective strategy for targeting habitat/resource protection? SID 5 (Rev. 3/06) Page 18 of 23 4) How should habitat creation efforts be focussed? Addressing each of these questions individually is problematic, since there are multiple potential scenarios that could be modelled to assist in answering each question. It is necessary to further refine these questions to provide a set of scenarios and targets/hypotheses (with defined spatial and temporal scales) that can then be evaluated or tested using a modelling approach. To define a model specification for each scenario, complete with estimated costs and data availability would be extremely difficult. Therefore, we provide a generic specification that can be used by Defra to develop projects to answer each of these questions. The generic specification will be applicable to any policy question about the impact of land-use changes on biodiversity. The generic model specification This specification consists of a set of stages that should be followed to develop models to answer policy questions on biodiversity. It does not recommend any particular modelling approach, as that will be dependent upon the question or questions being addressed. It provides a set of questions that need to be considered when developing models to answer questions about the impact of land-use changes on biodiversity. Stage 1: Defining and clarifying the problem/question The first step is to clarify the question. There are several things that need to be considered in this stage: What is the question? What are the essential pieces of information that Defra policy makers require? The question will need to be defined and refined to provide a set of specific and unambiguous questions or hypotheses, with defined outputs, that can be answered using modelling. This will ensure that any model developed will provide useful output. What definition of biodiversity is being used? Is biodiversity being defined as overall UK biodiversity or is it specific to a geographical region or habitat? Is the interest in species richness, community composition, species dominance or the population dynamics of one or more key indicator species? Which trophic levels are of interest: microbes, plants, insects, mammals, birds or some combination of these? Are multiple species being considered, or is the focus on a single species? The aim here is to define the level of detail that will be required within the model, and this will assist with the selection of the modelling tools and techniques to be used. What geographical scale is of interest? Is it sensible to have a case study for a particular geographical region or habitat? Will the land-use changes be applied on a broad-scale or in restricted regions? Here we are determining the level of integration that will be required within the model, and this will aid the selection of the modelling tools required to integrate between different levels of detail within the model. What time scale needs to be considered? Will any land-use changes be immediate or will they be phased in over time? Over what time-scale should the impact of change be simulated – short, medium or long-term? When considering impacts on biodiversity, impacts may be evident over a range of time scales, particularly when there is a need to consider the interactions between species or trophic levels and/or the movement of species. It will therefore be extremely important to ensure that the time-scale over which the impacts need to be determined is explicitly stated, so that the models will provide worthwhile predictions. What format does the output need to be in? The models developed will necessarily be complex, and capable of producing output in a wide variety of formats. It is essential that the precise format of outputs required from the model is defined in advance to ensure that information is communicated in the most useful form for policy development. It may be necessary to have more than one format, so that informed decisions can be made for policy development. In this first stage, consultation between Defra policy makers and the model developers is highly recommended (probably essential) as this will ensure that the models provide the exact type of information that Defra policy makers require to make decisions, rather than relying on the modellers to choose appropriate outputs. Stage 2: Development of plausible futures SID 5 (Rev. 3/06) Page 19 of 23 The aim of this stage is to develop a set of plausible future land-use scenarios, which will form the basis of the model simulations. Plausible future scenarios should be based on clearly stated assumptions about changes that will probably occur in the future. By clearly stating the assumptions used in developing the scenarios, it is possible for policy makers to select those scenarios that have assumptions that most closely fit with what they believe will occur, enabling them to examine the model and select the outputs that they feel will be of greatest value in determining policy. We would recommend that consultation between Defra policy makers and the model developers is maintained during this stage to ensure that the scenarios used fit closely to those envisaged as likely to occur by policy makers, including the most likely best and worse case situations. Stage 3: Selection of modelling approaches, data availability and model development In this stage of the model, the five questions that form the modelling framework need to be considered. Based on the questions being asked, and the plausible future scenarios, the model developers will need to determine which of the framework questions need to be answered, and then the modelling tools and techniques to be used to answer the framework will need to be selected. In selecting the modelling techniques it will be essential to have an overview of the quantity and format of data available to parameterise the models. By combining knowledge on data availability with the selection of modelling techniques, the most appropriate modelling approach can be identified. In addition, knowledge and data gaps will be identified that may have to be filled before a suitable model can be developed. In the absence of suitable or sufficient data, assumptions or expert knowledge can be substituted, if applicable. However we would recommend that, additionally, a plausible data scenario is developed. The plausible data scenario would provide criteria for the format of data that would be required to fully parameterise the model. Using this latter strategy, it is then possible to develop the model in the absence of data, so that once the data have been collected, in the specified format, the model can immediately be used to make predictions. Having selected the techniques and collected/collated the data, the models can then be developed. Stage 4: Simulating the plausible future scenarios and analysis of the outputs The plausible future scenarios would be used to provide inputs into the model, probably in the form of GIS data on land-use, with the current land-use as a baseline for comparison. Once the simulations have been completed, the data would then need to be analysed to determine the impacts of each scenario relative to a baseline scenario of no change. Where several impacts are possible within a particular future scenario, a measure of the uncertainty associated with the occurrence of particular impacts should be provided. This information can then be used to assist in assessing the likelihood of the occurrence of a particular impact, and then used to determine risk levels associated with particular scenarios. Stage 5: Recommendations and presentation of the model output The model developers should present their recommendations based on the model outputs to Defra policy makers, and discuss the assumptions used within the model. The assumptions made in the model, including the implicit assumptions of the individual techniques used to develop the model, should be explicitly stated, along with the implications of the assumptions not being met. This will assist Defra policy makers in their interpretation of the recommendations, and allow them to use their judgement about the usefulness of the model predictions. Can the modelling framework be used to address Public service agreement (PSA) targets? The key PSA target we consider here is: To reverse the long-term decline in the number of farmland birds by 2020, as measured annually against underlying trends. It is possible to use the modelling framework described here to assess the impact of land-use changes on the ability to meet this target. With current distribution and habitat data from the BBS, it will be possible to predict the potential distribution of key indicator species of farmland birds (on an individual species basis) using techniques similar to those used in MONARCH and SPECIES. This is equivalent to answering question 2 in the framework. If there was a need to predict the achievable distribution for key indicator species (questions 3 and 4 of the framework), this is more difficult as there is a lack of data on the dispersal and landscape permeability for bird species, and on the food values of different habitats for farmland bird species. This information could be derived using expert opinion, but this adds a large potential error to the predictions, which would need to be considered when interpreting the model predictions. SID 5 (Rev. 3/06) Page 20 of 23 Predicting the distribution of key indicator species of farmland birds, accounting for interactions with other key indicator species (question 5 of the modelling framework), would be extremely difficult due to lack of data. To include the interactions between species will require a large data collection effort, as expert opinion would probably not be reliable enough if modelling at this level of detail. In addition, it would be useful to revise the models based on annual breeding bird survey data and other research into bird behaviour and life cycles, using the adaptive resource management concepts. This would enable management decisions to be adapted annually to any changes in bird distribution or abundance, subject to the availability of suitable habitat and landscape information. Can the modelling framework be used to address England Biodiversity Strategy (EBS) targets? There is the potential to use the modelling framework to assess how land-use changes will affect individual species targeted in the England Biodiversity Strategy. However, lack of suitable data on the current distribution of many of the species, may limit the applicability of the modelling framework. Also, many of the priority species have very restricted habitat requirements and ranges, which means that examining land-use change on a UK or regional scale may not be appropriate. Also, if land-use changes are examined over a long time scale (15 or more years), it would be necessary to look at how land-use changes annually over this time scale, rather than just considering the start and end points. This would enable the determination of whether species ranges would shift incrementally, allowing the movement of species over time to occupy altered habitat ranges under a range of scenarios. This would require information on the permeability of different habitat types for each individual species, data which do not currently exist. The most valuable way to use the framework for EBS targets would probably be to model local scale land-use changes, examining how habitat creation would need to be targeted to best benefit the species with restricted ranges. This would require the collection of data on habitat permeability to assess whether the species are able to move to the newly created habitats. For both the PSA and EBS targets, the modelling framework could be used with present data to assess shifts in distribution, but would not be able to determine whether species would be able to move to the new distributions. To gain the greatest benefit from the modelling framework, it would be necessary to collect data on the permeability of the landscape for the species of interest. The ALMaSS concepts combined with adaptive resource management could be used to address both PSA and EBS targets, through the development of a single landscape simulator. Individual or multiple species models could then be linked to the landscape simulator to assess how changes in landscape would affect biodiversity, abundance or dispersal. Management decisions could then be made, and the results of these decisions fed back into the models at a later date to improve the predictions and management decisions. The development of an ALMaSS type of landscape simulator will require large-scale detailed information on land use and habitat types, the data for which is limited in England at present. There is a need for a central repository for all landscape and land-use information, and a co-ordination of land-use data collection efforts. This is a role that Defra could have, and ti would ensure that Defra have sufficient land-use information for modelling the impacts of land-use changes on biodiversity and the ability to meet PSA and EBS targets. We recommend that if Defra wish to use the modelling framework to assess the impact of land-use change on the ability to meet PSA and EBS targets, research is funded to develop a modular modelling approach, with a central landscape simulator linked to individual or multiple species models. We recommend that Defra fund research into the movement and dispersal of species, so that the permeability of different habitat types, and the movement of species through these habitats can be determined for use in models assessing the impacts of land use changes on biodiversity. We recommend that Defra co-ordinate the collection of data on land-use and habitat types within the UK, as the availability of suitably detailed data on land-use will be a severe limitation to any modelling of the impacts of land-use change on biodiversity in the UK Summary of Objective 4 The modelling framework described in Objective 3 provides a useful tool for answering questions about the impact of land-use changes on biodiversity. Using this framework to answer policy questions will require consultation between modellers and policy makers to ensure that a clear set of future scenarios and/or targets/hypotheses are developed that can be tested using modelling approaches. To assist in this process we have developed a generic model specification, including a set of questions to assist in ensuring and development of testable scenarios and hypotheses that address the needs of policy makers. The framework will be useful in the development of models to determine the effect of land-use changes on the ability to meet PSA and EBS targets. With existing data it should be possible to predict future potential distributions of individual species, as long as sufficient data on current potential or actual distributions and habitat types for the species of interest are available. However, the modelling framework would only be able to predict potential future distributions, as there is a lack of data on the dispersal of species through the landscape. If robust SID 5 (Rev. 3/06) Page 21 of 23 predictions of the future actual distribution of species are required, then there would need to be research to determine the effect of the landscape mosaic on the movement and re-distribution of the species of interest. For biodiversity conservation, there would almost certainly be a need to consider multiple species and the interactions between them, rather than modelling effects on individual species. Although the modelling techniques and tools exist to incorporate species interactions into models, there is a lack of data on interactions between different species. Therefore, there would need to be research into the interactions between multiple species in a range of habitats. References to published material 9. This section should be used to record links (hypertext links where possible) or references to other published material generated by, or relating to this project. References cited in the report. Vegetation dynamics references Arora, V.K. & Boer, G.J. (2006). Simulating competition and coexistence between plant functional types in a dynamic vegetation model. Earth Interactions 10: Paper 10, 30 pp Berry, P. M., Dawson, T. P., Harrison, P. A. & Pearson, R. G. (2002). Modelling potential impacts of climate change on the bioclimatic envelope of species in Britain and Ireland. Global Ecology and Biogeography 11(6) 453-462. Buckley YM, Briese DT, Rees M. (2003). Demography and management of the invasive plant species Hypericum perforatum. II. Construction and use of an individual-based model to predict population dynamics and the effects of management strategies. Journal of Applied Ecology 40 (3): 494-507 Buhler, D.D., King, R.P., Swinton, S.M., Gunsolus, J.L. & Forcella, F. (1997). Field evaluation of a bioeconomic model for weed management in corn (Zea mays). Weed Science, 45, 158-165. Goslee, S.C., Peters, D.P.C., & Beck, K.G. (2006). Spatial prediction of invasion success across heterogeneous landscapes using an individual-based model. Biological Invasions, 8: 193-200.Grime, J.P. & Hillier, S.H. (2000). The contribution of seedling regeneration to the structure and dynamics of plant communities, ecosystems and larger units of the landscape. In: Seeds: The Ecology & Regeneration of Plant Communities 2 nd Edition. (ed M. Fenner). 361-374. Grundy, A.C. (2003) Predicting weed emergence: a review of approaches and future challenges. Weed Research 43: 1–11. Heggenstaller, A H. & Leibman, M. (2006). Demography of Abutilon theoprasti and Seteria faberi in three crop rotation systems. Weed Research, 46: 138-151. Higgins, S.I., Richardson, D.M. & Cowling, R.M. (1996). Modeling invasive plant spread: the role of plant-environment interactions and model structure. Ecology, 77:2043-2054. Higgins, S.I., Richardson, D.M., Cowling, R.M. and Trinder-Smith, T.H. 1999. Predicting the landscape distribution of invasive alien plants and their threat to native plant diversity. Conservation Biology 13: 303-313 Hughes, J.K., Valdes, P.J. & Betts, R. (2006). Dynamics of a global-scale vegetation model. Ecological Modelling 198:452-462/ Holst, N. Rasmussen, I.A. & Bastiaans, L. (2007). Field weed population dynamics : a review of model approaches and applications. Weed Research, 47: 1-14 Hughes , J.K., Valdes, P.J. Betts, R. (2006) . Dynamics of a global-scale vegetation model. Ecological Modelling. 198: 452462 Pannell, D.J., Stewart, V., Bennett, A., Monjardino, M., Schmidt, C. and Powles, S.B. (2004). RIM: A Bioeconomic Model for Integrated Weed Management of Lolium rigidum in Western Australia. Agricultural Systems 79(3): 305-325. Townsend Peterson, A., Papes, M. & Kluza, D.A. (2003). Predicting the potential invasive distributions of four alien plant species in North America. Weed Science, 51: 863 – 868. Woodward, F.I. & Lomas, M.R. (2004). Vegetation dynamics – simulating responses to climate change. Biological Reviews 79:643-670 Bird dynamics references Atkinson, P. W. (1998) The wintering ecology of the twite Carduelis flavirostris and the consequences of habitat loss. PhD thesis, University of East Anglia, Norwich, UK. British Trust for Ornithology (2005) Predicting the Response of Farmland Birds to Agricultural Change. Defra Report for project BD1618. Butler, S. J., Vickery, J. A. & Norris, K. (2007) Farmland biodiversity and the footprint of agriculture. Science 315: 381-384. Pennycuik, L. (1969) A computer model of the Oxford great tit population. Journal of Theoretical Biology 22: 381 – 400. Potts, G. R. & Aebischer, N. J. (1991) Modelling the population dynamics of the Grey Partridge: conservation and management. In Bird Population Studies. Relevance to Conservation and Management (Eds. C. M. Perrins, J.-D. Lebreton & G. J. M. Harris), pp. 373-390. Oxford University Press, Oxford. Stephens, P. A., Freckleton, R. P., Watkinson, A. R. & Sutherland, W. J. (2003). Predicting the response of farmland bird populations to changing food supplies. Journal of Applied Ecology 40: 970-983. Swetnam, R. D., Wilson, J. D., Whittingham, M. J. & Grice, P. V. (2005) Designing lowland landscapes for farmland birds: scenario testing with GIS. Computers, Environment and Urban Systems 29: 275-296. Watkinson, A. R., Freckleton, R. P., Robinson, R. A. & Sutherland, W. J. (2000) Predictions of biodiversity response to genetically modified herbicide-tolerant crops. Science 289: 1554-1557. Whittingham, M. J., Swetnam, R. D., Wilson, J. D., Chamberlain, D. E. & Freckleton, R. P. (2005) Habitat selection by yellowhammers Emberiza, citrinella on lowland farmland at two spatial scales: implications for conservation manangement. Journal of Applied Ecology 42: 270-280. SID 5 (Rev. 3/06) Page 22 of 23 Insect dynamics references Benton, T.G, Bryant, D.M., Cole, L. & Crick ,H.Q.P (2002). Linking agricultural practice to insect and bird populations: a historical study over three decades. Journal of Applied Ecology 39: 673-687. Bhar R. & Fahrig, L. (1998). Local vs. Landscape effects of woody field borders as barriers to crop pest movement. Conservation Ecology 2, http://www.consecol.org/vol2/iss2/art3/ Brewster, C.C. & Allen, J.C. (1997). Spatiotemporal models for studying insect dynamics in large-scale cropping systems. Environmental Entomology 26: 473-482. Collinge, S.K. & Forma, R.T.T. (1998). A conceptual model of land conversion processes: predictions and evidence from a microlandscape experiment with grassland insects. Oikos 82: 66-84. Crist, T.O., Pradhan-Devare, V. & Summerville, K.S. (2006). Spatial variation in insect community and species responses to habitat loss and plant community composition. Oecologia 147: 510-521. Levine, S.H. & Wetzler, R.E. (1996). Modelling the role of host plant dispersion in the search success of herbivorous insects: Implications for ecological pest management, Ecological Modelling 89: 183-96. Potting, R.P.J, Perry, J.N. & Powell, W. (2005). Insect behavioural ecology and other factors affecting the control efficacy of agro-ecosystem diversification strategies. Ecological Modelling 182: 199-216. Walters, C., Korman, J, Stevens, L.E. & Gold, B. (2000). Ecosystem modelling for evaluation of adaptive management policies in the Grand Canyon. Conservation Ecology 4, article no. 1. http://www.ecologyandsociety.org/vol4/iss2/art1/ Conservation and integration References Austin, M. (2007) Species distribution models and ecological theory: A critical assessment and some possible new approaches. Berry, P.M., Harrison, P. A., Dawson, T. P. & Walmsley, C. A. (2005) Modelling Natural Resource Responses to Climate Change (MONARCH) UKCIP. Ferrier, S., Drielsma, M., Manion, G. & Watson, G. (2002) Extended statistical approaches to modelling spatial pattern in biodiversity in northeast New South Wales. II. Community-level modelling. Biodiveristy and Conservation 11: 2309-2338. Ferrier, S., Watson, G., Pearce, J. & Drielsma, M. (2002) Extended statistical approaches to modelling spatial pattern in biodiversity in northeast New South Wales. I. Species-level modelling. Biodiversity and Conservation 11: 2275-2307. Jepsen, J. U., Topping, C. J., Odderskær, P. & Andersen, P. N. (2005) Evaluating consequences of land-use strategies on wildlife populations using multiple-species predictive scenarios. Agriculture, Ecosystems and Environment 105: 581-594. Lindenmayer, D. B., Cunningham, R. B., Donnelly, C. F. & Lesslie, R. (2002) On the use of landscape surrogates as ecological indicators in fragmented forests. Forest Ecology and Management 159: 203-216. Pearce, J.L., Cherry, M., Drielsma, S. Ferrier, S & Whish, G. (2001) Incorporating expert opinion and fine-scale vegetation mapping into statistical models of faunal distribution. Journal of Applied Ecology 38:412-424. Ray, N., Lehmann, A. & Joly, P. (2002) Modeling spatial distribution of amphibian populations: a GIS approach based on habitat matrix permeability. Biodiversity and Conservation 11: 2143-2165. Topping, C. J., Hansen, T. S., Jensen, T. S., Jepsen, J. U., Nikolajsen, F. & Odderskær, P. (2003) ALMaSS, an agent-based model for animals in temperate European landscapes. Ecological Modelling 167: 65-82. Topping, C. J. (2005) The impact on skylark numbers of reductions in pesticide usage in Denmark. Predictins using a landscape-scale individual-based model. National Environmental Research Institute, Denmark Technical Report No. 527 Virkkala, R., Luoto, M. & Rainio, K. (2004) Effects of landscape composition on farmland and red-listed birds in boreal agricultural-forest mosaics. Ecography 27: 273-284. Warren, J. & Topping, C. (2004) A trait specific model of competition in a spatially structure plant community. Ecological Modelling 180: 477-485. Warren, J. M. & Topping, C. J. (1999) A space ovvupancy model for the vegetation succession that occurs on set-aside. Agriculture, Ecosystems and Environment 72: 119-129. Watts, K., Humphrey, J. W., Griffiths, M., Quine, C. & Ray, D. (2005) Evaluating Biodiversity in Fragemented Landscapes: Principles. Forestry Commission Information Note 73 Forestry Commission, Edinburgh. SID 5 (Rev. 3/06) Page 23 of 23