Integrated modelling and decision-support tools: a Mediterranean example

Environmental Modelling & Software 19 (2004) 999–1010 www.elsevier.com/locate/envsoft

Integrated modelling and decision-support tools: a Mediterranean example T. Oxley a,, B.S. McIntosh b, N. Winder c, M. Mulligan d, G. Engelen e a

Department of Environmental Science and Technology, Imperial College, Prince Consort Road, London SW7 2BP, UK b School of Water Sciences, Cranfield University, Bedford MK43 0AL, UK c School of Historical Studies, Museum of Antiquities, University of Newcastle-upon-Tyne, Newcastle NE1 7RU, UK d Department of Geography, King’s College London, The Strand, London WC2R 2LS, UK e RIKS bv., P.O. Box 463, 6200 AL Maastricht, Netherlands Received 3 January 2003; received in revised form 25 May 2003; accepted 1 November 2003

Abstract A great deal of new knowledge and research material have been generated from research carried out under the auspices of the European Union (EU). However, only a small amount has been made available as practical policy-support tools. In this paper, we describe how EU funded research models and understanding have been integrated into an interactive decision-support system addressing physical, economic and social aspects of land degradation in the Mediterranean. We summarise the 10 constituent models that simulate hydrology, human influences, crops, natural vegetation and climatic conditions. The models operate on very different spatial and temporal scales and utilise different modelling techniques and implementation languages. Many scientific, modelling and technical issues were encountered during the transformation of ‘research’ models into ‘policy’ models. We highlight the differences between each type of model and discuss some of the ontological and technical problems in re-using research models for policy-support, including resolving differences in temporal scale and some of the software engineering aspects of model integration. The involvement of policy-makers, ‘stakeholders’ and other end-users is essential for the specification of relevant decision-making issues and the development of useful interactive support tools. We discuss the problems of identifying both the decision-makers and the issues they perceive as important, their receptivity to such tools, and their roles in the policy-making process. Finally, we note the lessons learned, the resources needed, and the types of end-users, scientists and mediators required to ensure effective communication, technical development and exploitation of spatial modelling tools for integrated environmental decision-making. # 2004 Elsevier Ltd. All rights reserved. Keywords: MODULUS; Decision-support systems; Integrated modelling; Policy-support

Software availability Program title: Developers: Contact address:

Version:

GEONAMICA1 for MODULUS. RIKS bv. in collaboration with the MODULUS Consortium. Guy Engelen, Research Instituut voor KennisSystemen bv., P.O. Box 463, 6200 AL Maastricht, The Netherlands. Demonstration only.

 Corresponding author. Tel.: +44-20-7594-9297; fax: +44-20-75949266. E-mail address: [email protected] (T. Oxley).

1364-8152/$ - see front matter # 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.envsoft.2003.11.003

Hardware required: PC Pentium II, 128 MB RAM or higher. Software required: Windows 95/98/NT/2K. Program language: A combination of C++, Fortran, Power BASIC and SICStus Prolog. Disk requirements: 40 MB.

1. Introduction There are growing pressures from funding bodies to integrate human and environmental issues in a manner that is relevant and accessible to the needs of the policy world (Tress et al., 2003). At the same time, there have

1000

T. Oxley et al. / Environmental Modelling & Software 19 (2004) 999–1010

been calls from academia to better integrate the social and natural sciences to address the issues faced by society today (Winder, 2000; Costanza, 2003). These pressures are being currently reflected in the research community through the emergence of different approaches to the issues of interdisciplinarity and policy-relevance, including integrative research (Winder, 2003), integrated assessment and modelling (Parker et al., 2002; Jakeman and Letcher, 2003), companion modelling (Barreteau et al., 2003) and collaborative learning (Bosch et al., 2003). However, there is no consensus at present on how best to integrate the insights from different disciplines for different purposes or how best to turn these insights into effective policy-support. The European Commission (EC) has, through its successive ‘Framework’ programs, funded a large amount of social and natural science research. To help improve the relevance of this research output to land degradation policy, the MODULUS Project was commissioned to explore the feasibility of directly integrating the results of various Framework funded research projects—EFEDA, ERMES, ModMED, ARCHAEOMEDES, EPPM and MEDALUS. More specifically, MODULUS was designed to integrate the ‘research’ models developed in each of these projects in such a way as to produce a tool to support ‘integrated environmental decision-making’ at a regional scale. As part of this process, MODULUS involved regional policy end-users from the outset to help ensure project relevance and guide the design of the support tool. The MODULUS Project effectively set out to build a flexible, generically applicable decision-support system (DSS) to enable end-users to understand the processes causing and caused by land degradation, and to provide appropriate tools for the design and evaluation of policy options. The system and its models were applied and tested in the Argolida (Greece) and Marina Baixa (Spain) regions in collaboration with local policymakers and researchers with experience in these regions. MODULUS succeeded in bringing together and reusing research material and models in a new context by developing an integrated model embedded within a tailor-made DSS. We learned that it takes more than the 24 months allocated to the project to go though the full development cycle for a DSS of this complexity and indeed that direct integration and policy-use of research results, particularly model-based research, is a non-trivial task. We believe that we produced something rather unique, that the MODULUS DSS represents a ‘proof of concept’ system demonstrating the technical (software) feasibility of integrating diverse research models for policy-support, and that the work undertaken by the project consortium constitutes a valuable ‘narrative’ detailing some potential perils and pitfalls for those engaged in policy-relevant research.

Consequently, there are three main purposes to this paper: i. To highlight the complex scientific, modelling and technical issues involved in the development and exploitation of spatial modelling tools for integrated environmental decision-support. ii. To discuss the problems of identifying decisionmakers and the issues they perceive as important, their receptivity to model-based decision-support tools, and the role of such tools in the policy-making process. iii. To describe some of the issues faced during the research process and document the project as a case-study in policy-relevant research. We feel that covering all three points provides a useful synthesis, combining the technical and social elements that are fundamentally inseparable in policyrelevant research. However, to achieve our objectives in the given space, this paper must necessarily omit some of the finer details of our research. The interested reader is encouraged to follow our references and in particular to read the MODULUS project final reports (Engelen, 2000), all of which are available via the web.

2. Research or policy models Models are ‘a simplified representation of a system (or process or theory) intended to enhance our ability to understand, predict, and possibly control the behaviour of the system’ (Neelamkavil, 1988). However, there are important differences between what we describe here as research models or policy models. We describe research models as those originating in the problem-driven empirical or theoretical domains (Oxley et al., 2003). We describe policy models as those originating in the value-driven perceptual domain. The starting point for development, whether research or policy, will have a significant bearing on the criteria used to develop, evaluate, use and interpret a model, not least by affecting the balance between reality, precision and generality employed (Levins, 1966). Of course, models are developed across a spectrum between these polar extremes for many different purposes. However, for the purpose of explaining our research aims and methods, it is useful to refer to models originally developed for scientific research or policysupport for illustration. Research models tend to be strongly process oriented. Their temporal and spatial scales and level of complexity are solely determined by the characteristics of the process(es) being examined. The model developer aims at a representation that is as accurate as possible, uses the model to test hypotheses and further under-

T. Oxley et al. / Environmental Modelling & Software 19 (2004) 999–1010

standing of the world and tends to make use of scientifically innovative techniques to develop a model that is as complex as necessary. Often this will pose difficulties in validating the resulting model, but in the quest for increased understanding, the development of the model can be a goal in its own right. During the modelling process, new data needed for the model will be gathered as required from field sites or other sources. The processing speed and the interactivity of the model are not typically considered, nor is model transparency or user-friendliness, as the model developer is usually the only user. Policy models are foremost oriented towards addressing practical policy issues. The issues addressed determine the temporal and the spatial resolution at which processes are represented. In addition, the level of detail and degree of complexity are likely to be determined by the availability of data and the constraints imposed by the end-user organisation on data gathering and use in terms of time and cost. Models are interesting in a policy context because they deliver practically useful output, although this may vary in detail from supporting concrete management decisions to prompting the consideration of a previously ignored issue within a decision-taking body (see van Daalen et al., 2002, for a discussion of the role of models in environmental policy). To achieve this, robust, extensively tested methodologies will preferentially be used. The model may be complex, but generally is kept as simple as possible. Such models are not designed for further understanding of processes but to help explore the possible effects of policies and to identify possible policy problems (Winder, 2003). Processing speed and model interactivity may be determining factors for success, particularly if the model is used in participatory exercises involving stakeholders. In addition, transparency and user-friendliness are crucial factors, along with the involvement of the problem-owner during model development. Before a strategy for model integration can be developed, it is important to understand the key differences between research and policy models (see Table 1). The requirements of the two types of model are driven by fundamental differences in the objectives of modelling—to further understand how the world operates or to guide the processes involved in managing it. The task for MODULUS was to transform a selection of disparate research models—through integration, adaptation or rebuilding—into an integrated policy model accessible to the policy-making community.

3. The MODULUS sub-models Many models had been developed in these six EC projects (see above) in the broad context of desertifica-

1001

Table 1 Distinctive contrasting characteristics of models originally developed for scientific research or policy exploration and analysis Research oriented

Policy oriented

Accurate representation of processes Complexity and resolution reflect processes Accurate representation of spatial variability Scientifically innovative Raises more questions than answers Interesting and worthwhile in its own right Process centred Numbers validatable As complex as necessary

Adequate representation of processes Complexity and resolution reflect data Adequate representation (reflects existing data) Scientifically proven Provides simple (?) and definitive (?) answers Interesting and worthwhile only through its output Input/output centred Outcomes validatable As simple as possible

Interfacing issues Model centred

Interface centred

tion in the Mediterranean, each of which was evaluated for its suitability for integration into a policy-relevant decision-support system. The process of model selection, although constrained to models produced by the candidate EC projects, was carried out using a range of end-user, scientific and technical (software) criteria. Full details of the selection and evaluation process can be found in McIntosh et al. (2000). Ten sub-models (including alternative aquifer models) were selected for integration into the MODULUS DSS (see the system diagram in Fig. 1). It is not possible here to describe each model in its entirety, so reference is made to the appropriate literature. More comprehensive summary details of these models can be found elsewhere (Engelen, 2000; Oxley et al., 2000). For each model, we identify the source project, the language it is written in, the spatial and temporal resolution and the processes modelled. 3.1. Climate and weather (Efeda, PatternLITE Weather model, C++, 1 ha.) This model runs daily, calculating the time of sunrise and sunset and the average solar radiation. The temperature per cell is updated monthly. The model generates detailed daily time series for precipitation using a dynamic (‘bucket-tip’) timestep based on data from one or more AWS weather stations. Temperature and precipitation are corrected for climate change as calculated by the HADCM2, GFDL or ECHAM Global Circulation Models (Mulligan, 1996; Mulligan and Reaney, 2000).

1002

T. Oxley et al. / Environmental Modelling & Software 19 (2004) 999–1010

Fig. 1. The MODULUS DSS graphical user interface, showing the system diagram and (clockwise) the following models: Modflow (integrated), irrigation (rebuilt), river flow (integrated), soil hydrology (adapted), land-use (integrated) and plant growth (rebuilt). The adapted climate model dialogue window is also shown, highlighting the daily storm events.

3.2. Hillslope hydrology (Efeda, PatternLITE Hillslope model, C++, 1 ha.) This model runs daily, but integrates internally over bucket-tip timesteps. It deals with soil hydraulic properties and calculates the water budget: interception, infiltration, soil moisture, transpiration, soil evaporation, overland flow, surface recharge and erosion (Mulligan, 1994, 1996, 1998; Burke et al., 1998; Reaney and Mulligan, 1999). 3.3. Plant growth (Efeda, PatternLITE Plant model, C++, 1 ha.) This model runs daily. It represents the processes of growth of commercial crops and natural species and calculates the leaf biomass, root biomass, leaf area index (LAI) and the vegetation cover fraction (Mulligan and Reaney, 2000). 3.4. Natural vegetation (ModMED, RBCLM2 community model, Prolog, 25 ha.) This model runs monthly. It represents the processes of growth, succession and decline of the natural vegetation at the community level. It calculates the

LAI, the vegetation cover fraction, and the rooting depth. This model is rule-based, applied to each cell, supplemented with a cellular seed diffusion model (C++), which produces a seed biomass map and links the community level cells at the landscape level (Mazzoleni et al., 1998; McIntosh et al., 2001; McIntosh, 2002). 3.5. Ground water (ARCHAEOMEDES, two user-selectable models: the AUA-ModFlow model (Fortran, 25 ha) and the IERC-Aquifer model (in Power Basic, 1 ha).) These models address the depletion, recharge and pollution of aquifers. They calculate the aquifer water height, salt concentration and the fluxes between cells. ModFlow runs monthly (Giannoulopoulos, 2000; Poulovassilis and Giannoulopoulos, 1999) and the IERC model (Robinson, 1999) runs daily. Both models utilise equivalent mathematical formulations, thus producing identical results, ceteris paribus. However, Robinson’s (1999) model is able to capture external ‘stresses’ which occur within the timestep period required by ModFlow (Oxley et al.. 2002).

T. Oxley et al. / Environmental Modelling & Software 19 (2004) 999–1010

1003

3.6. Surface water

3.9. Land-use

(ERMES, Catchment model, Power Basic.) This model runs daily and represents the river, canal, and water reservoir system, and the water quality of the surface water. It calculates the river flows per stream order, the sinkhole flows, and the catchment recharge flows. The model runs on irregular shaped, natural defined areas: the catchments and sub-catchments (Billen et al., 1990; Billen, 1993; Allen et al., 1996).

(Constrained Cellular Automata model, GEONAMICA1, C++, 1 ha.) This model runs annually. It allocates the land-use dynamics resulting from demographic changes, as well as the dynamics in the agricultural and non-agricultural part of the economy. The allocation methodology takes into consideration the activity-specific attractivity of cells in terms of their suitability, zoning regulations and accessibility to the road transportation infrastructure (Engelen et al., 1997; White and Engelen, 1997). These models are integrated by means of information flows as detailed in Oxley et al. (2000) and McIntosh et al. (2000), with the main flows highlighted in Fig. 2. Maintaining the flows is crucial for the integrity of the DSS; they must, for example, be retained when the user only selects a sub-set of the available models for a simulation. A variety of simulation scenarios can be explored using the DSS. Selections of these scenarios are documented in Engelen (2000) and Oxley et al. (2000), but potentially can include water management practices, crop choice and subsidy change, climate change, economic policy and urban development. Other problems such as planning and land suitability mapping, tourism and water stress, environmental impact assessment, natural vegetation dynamics, desertification and aquifer recharge can also be addressed.

3.7. Crop choice (ARCHAEOMEDES/EPPM, Decision-making model, Power Basic, 1 ha.) This model runs annually. It is a rule-based model representing the crop choices made by farmers as a function of changing physical, socioeconomic and institutional conditions and circumstances. It calculates the crop type, crop water requirements, water source, presence of boreholes, borehole depth, pumping capacity and the total annual exploitation costs (Winder et al., 1998; Winder, 2000; Oxley et al., 2002). 3.8. Irrigation (ARCHAEOMEDES, Power Basic, 1 ha.) This model runs twice daily. It is a rule-based model representing the farmers’ decisions to switch on the water pumps and start the irrigation. It calculates the pump status, volume to be pumped, and extraction from the canal, volume of irrigation/frost water and irrigation/ frost water salt concentrations (Oxley et al., 2000).

4. Modelling and technical issues The MODULUS DSS integrated EU funded research models and gave them a visual interface: a

Fig. 2. MODULUS DSS sub-model information flow table (from McIntosh et al., 2000).

1004

T. Oxley et al. / Environmental Modelling & Software 19 (2004) 999–1010

dynamic map in which hydrological, biological and agronomic landscapes evolve on the screen in real time. The models from which these components were derived were all intended for academic research and their integration on a single platform was a demanding task. Initial calculations suggested that a single run of all the models to be integrated would probably require tens and quite possibly hundreds of processor hours. Simplifications and adaptations were therefore unavoidable. There are three levels at which existing models can be integrated within a DSS for policy application. First, the models can simply be integrated as code or executables close to their original form, providing that the interactions with other models are clearly specified. In this case, unless the models were specifically built for policy application or cover only small parts of the modelled system, it is likely that many of the end-user requirements on speed and responsiveness will not be fully met. Secondly, models can be kept in their original form, but the code adapted in minor ways to meet the end-user requirements more successfully. The kinds of adaptations carried out may include changes to the spatial resolution, scale or timesteps of the model or simplification of key process equations. The third level at which models can be integrated with a DSS is through a complete rebuild of the model for incorporation into the DSS. This has a number of significant advantages over the other methods. In particular, it allows the model to be specifically designed to meet the user requirements. Realistically, for many research models, rebuild is the only way in which the very extensive end-user requirements outlined here can be fully met. At the same time, the key innovations and most significant processes and lessons from the original model can be transferred to the rebuild. In this way, the research model can be rationalised, simplified, modified to the correct scale and rebuilt according to the new objective. Alongside and influencing these questions about whether to integrate, adapt or rebuild models, the additional complexity of multiple spatial scales and different timesteps must be assessed on both a technical (how do you get the models to interact?) and ontologi-

cal (are the models interacting with each other in a way that makes sense with respect to what we know about the world?) level. Fig. 3 highlights the extent of these spatial and temporal differences between the models. Whereas mapping spatial data onto different scales can be relatively straightforward, given the effects and consequences of aggregation and disaggregation, coordination of timesteps can be more problematic. One mechanism for dealing with this is the ‘bucket-tip’ approach of Mulligan and Reaney (2000) discussed below. Two critical issues had to be addressed. The first relates to timesteps and the second relates to spatial self-organisation. Both issues bear on the dynamic sensitivities of the composite system. All of the component models have definite timesteps that determine the simulated times at which system variables are updated. Aquifer levels, for example, are updated relatively infrequently, though water enters the soil in definite precipitation events and may move through the soil very rapidly. Thus, the surface hydrology ‘wants’ to give water to the aquifer on an hour-by-hour basis, while the sub-surface aquifer can only accept it on a much longer timestep. The effect of this is that water may appear to be delivered to the aquifer in enormous and unnatural torrents with potentially significant dynamic impacts. Water abstraction poses similar problems in that irrigation decisions are made on a much shorter time-frame than aquifer level changes. Aquifers can therefore build up huge ‘irrigation debts’ which are paid off instantly at the beginning of a hydrological step. Prigogine (1978) has shown that periodic disturbances of this sort can result in spontaneous selforganisation with the development of complex spatial patterns that could show up on our distribution maps. His work was intended to explain how spontaneous self-organisation could occur in nature, but we are concerned that similar effects could be manifest merely as artefacts of model design. We suggest that under such circumstances it is difficult to know a priori how to separate computational artefacts and patterns that emerge as a spurious consequence of model integration from meaningful dynamics.

Fig. 3. Definition of the diverse spatial resolutions of individual models and the different timesteps required of the processes modelled.

T. Oxley et al. / Environmental Modelling & Software 19 (2004) 999–1010

Reducing timesteps to bring every model into step with the others is not an option, because the increased computational load would increase run-time to an unacceptable level. Increasing timesteps is similarly unacceptable because fine time-scale phenomena like single storm events can have very significant effects. One response to such problems is to use interpolation, and this provided a workable compromise, at least in the hydrological domain. Mulligan and Reaney (2000) developed a simplified version of PATTERN using a novel ‘bucket-tip’ technique such that timesteps responded dynamically to rainfall events. It was not clear, however, that there existed any generic ‘off the shelf’ solution to problems of scale difference between models. Rather, the adaptations made to each model were primarily made in model- and domain-specific ways. Such domain specificity was essential to ensure that model integration made ontological sense and to minimise the chances of implementation artefacts in the finished model. However, in the long run, there is no real alternative to an intensive sensitivity analysis designed to evaluate the likely dynamic knock-on effects of timestep variability. This was not attempted in our proof of concept system and we suspect that to do so might require such substantial investments of time that any saving made by connecting pre-existing models could be nullified. In terms of implementation, two basic problems were encountered—integrating models written in different languages (e.g. Power Basic vs. Prolog) and controlling the order in which variable values are computed across component models. The DSS was built as an integrated model composed of a number of ActiveX components (a self-registering COM component) called Model Building Blocks, each MBB corresponding to one of

1005

the sub-models detailed above. van der Meulen et al. (2000) provide a detailed description of the software framework used for integration in MODULUS. What follows from here is a brief summary intended to provide a strong flavour of our approach. The flexibility requirements of the DSS included exchangeability and the need to maintain a generic model-editor and DSS-shell. ActiveX supports the development of easily exchangeable models because the interface of an ActiveX component is standardised. It is easier to develop a generic model-editor and DSSshell if it uses ActiveX components, because an important part of the design and development work of the required component interfaces has already been done. Moreover, the system developed will be easier to use by and with third parties, because of the standard interfacing employed. The integration of existing models was achieved without having to completely re-code (where otherwise it may have been necessary) through the use of a wrapping technique (see Fig. 4), whereby each sub-model was transformed from its native code into an ActiveX MBB—a more or less complete model with a predefined set of inputs and outputs. The wrapping process was tailored to each component model and involved some minor re-coding. The spatial modelling environment GEONAMICA1 developed by RIKS bv. was used as the core simulation engine and platform for integration. Standard interface definitions were used to integrate each MBB with the GEONAMICA1 system and the Windows OS. A standard interface was defined to permit the simulation engine to run models with different timesteps at the same time and to control variable computation order. Another standard interface was defined to

Fig. 4. The MODULUS simulation engine works with ActiveX Model Components. Existing models are encapsulated in ‘wrappers’ that make them look like ActiveX Components from the outside.

1006

T. Oxley et al. / Environmental Modelling & Software 19 (2004) 999–1010

retrieve each MBB’s input and output specification, thereby allowing the simulation engine to ‘connect’ one MBB to another in terms of information flow. Each MBB has two graphical representations in GEONAMICA1: (1) a dialogue giving access to the parameters and outputs of the MBB, and (2) a unique graphical object that shows how the MBB relates and is connected to other MBBs in the integrated model. A user could know from this connection scheme where an MBB gets its input(s) from, and where it has to send its output(s) to. The integration of the different parts of the MODULUS DSS was not a trivial task, so a procedure was developed which involved four phases: I. The possibilities and advantages of sub-dividing the original models into smaller, more generic parts were evaluated. The PatternLITE models evolved in this way, where the original model was split up and integrated as a set of separate MBBs. II. The existing software models were then transformed into MBBs. III. The MBBs were tested and run by means of the simulation engine. IV. The exchange of the data between the different MBBs was implemented. Care must be taken in this phase to accommodate differences in the representation of the data and differences in the timing of events. These phases are described in detail in van der Meulen et al. (2000).

5. Policy-makers and end-users As part of its remit, the MODULUS project identified and communicated with a number of local, regional and national policy representatives in both the Argolida and Marina Baixa areas to varying degrees of success. Full details of these representatives, their roles and the way in which the project interacted with them can be found in Blatsou et al. (2000), to which the reader is referred. This section will not repeat the details of our interaction but shall instead offer a discussion of some of the lessons we learned. The view that many scientists have about policies and policy-making in the environmental field is often overly simplistic. Frequently, researchers refer to a rational ‘decision-maker’ as some autonomous individual located at some higher level in an administrative hierarchy. In reality, policy formulation and decisionmaking are complex processes involving many individuals and many different forms of knowledge, and it is difficult, if at all possible, to pinpoint the moment at which, or the people by which, a decision is arrived at. This oversimplification on the scientists’ side is rep-

resentative of a more fundamental source of tension, which resides in the fact that the two communities function in qualitatively different contexts. Many (mostly natural) scientists are often concerned with well-posed problems and, correspondingly, the idea of ‘solutions’ (Winder, 2003). The policy world, on the other hand, exists in a multiple-issue and multipleconstituency world where the agenda is constantly changing and where an environmental issue is only one of many competing for attention. Moreover, scientists and people involved in making and administering policies are subject to different kinds of peer and contextual pressure, they have different time horizons, they speak in many ways a different language. These differences complicate communication between the two groups, but simultaneously make such communication essential if one is to focus research on a community of policy-making end-users. There is a widespread assumption, not least among ourselves at the start of MODULUS, that transparent communication between all the stakeholders involved in environmental problems is necessary. It is worth questioning whether this is in fact always the case and under what circumstances there is anything worthwhile communicating. For example, a local farmer may be more concerned about crop yield than water consumption. In order to be effective, communication between the farmer and a hydrologist working on problems of desertification may not require discussing the effects of excessive water consumption so much as crop choices that do not use so much water. Yet the hydrologist is a specialist on water, not crop choices. Thus, the communication that is needed is not between the hydrologist and the farmer, but between the farmer and an agronomist. In turn, the agronomist need only know that he should advise farmers on crops that require less water consumption; he does not need to know the details of hydrological science. In other words, much depends on the agendas of the participants. The agenda for scientists may be somewhat different to the policy issues at different scales. In turn, policy issues at, say, a national scale may not be the same as those at a regional or local level. There is the additional problem of diagnosis. Scientists and other environmental specialists tend to simplify and abstract from complex situations in a way that enables them to apply their knowledge more generally. We have come across numerous situations where the interpretation of the symptoms of an environmental disorder has been specified as very different ‘problems’ by politicians and specialists, and between specialists. An example is a situation where more water is being consumed than can be sustained in a local environment: in one domain, the problem may be perceived as excessive water use, but in another, it may be perceived as insufficient supply. The policy implications

T. Oxley et al. / Environmental Modelling & Software 19 (2004) 999–1010

of the two perceptions are dramatically different, in that the first would recommend a policy instrument to reduce water consumption, whereas the second would recommend a technical solution that would increase water supply. This mismatch of agendas and ‘problem’ identifications can result in inappropriate research perhaps conducted at an unsuitable scale and not easily connected to decision issues. The knowledge of any individual, group of individuals (at various levels) or institution is derived, negotiated and shared in a particular context. As a result, there are often relatively invisible differences that can lead to serious misunderstandings. One of the areas in which this affects us here is the hampering of communication between people who have, as part of their (formal) education, acquired ‘institutional’ knowledge, and people who have, on the other hand, acquired their knowledge informally, as part of their everyday activities in the area where they live. Such knowledge is often termed ‘local knowledge’. The way in which knowledge is translated into effective use varies at different scales. ‘Institutional’ science tends to address supra-regional, and often supranational, agendas. There is a codification of knowledge in the form of standards and procedures. For instance, water quality standards for potable and recycled water are set on the basis of research commissioned at the national or EU level (and sometimes at a wider international level). At a local and regional level, this knowledge is received in the form of technical standards rather than specific local policy. The ‘communication’ takes place at the institutional level. This creates problems as local issues drive a need for local access to local knowledge. Many of the ‘communication’ issues we have seen at a local and regional scale arise because environmental issues are often unanticipated and highly specific to local circumstances. Institutional knowledge has not been developed with these situations in mind, potentially resulting in a lack of relevant understanding. Sometimes local and regional specialists can be hired to help fill the gap. There is still, however, the difficulty of how scientists interpret situations as problems at this local scale. In addition, there can be very important differences between local and regional areas in terms of effectiveness of communication between scientists and policymakers. Among the factors influencing local communication effectiveness are the extent to which environmental issues are a priority among different constituencies, the social and cultural nature of local networks and the alignment between local politicians and issues. One of the core problems of an exercise like MODULUS is that the same information can mean different things to different people. Communication between participants presumes a certain degree of alignment of

1007

objectives and perceptions. Such alignment is a very slow and complex process of learning, which, even at the best of times, cannot be accommodated very easily in a project-based agenda, where time is limited and the implicit design objective is one of ‘experts’ giving advice to the ‘non-experts’. During the MODULUS project, workshops were held with the aim of directly involving the regional and local policy stakeholders in the DSS design process. In order to better understand the nature of the reasons why the stakeholders expressed interest in the project, we asked a number of the workshop participants their reasons for being there. Three main reasons were identified by the end-users: prestige, personal or institutional self-interest, and access to a reliable source of EU information. One of the most informative comments arising from the workshops regarding the motivation and involvement of potential end-users was: to quote one key workshop participant ‘‘Finally, we stop being the aboriginals that are studied by civilised people, and from now on we will start collaborating with them at the same level.’’ (Filippucci et al., 2000). This is a crucial point that we feel bears strongly on the way in which we, as scientists, need to structure our approach to providing practical policy-support. Indeed, with respect to the building and maintenance of relationships between researchers and end-users, one of the principal problems was the need to transform an initial relationship based on the ‘commoditisation’ of each group by the other for its internal consumption, into a relationship of mutual trust and respect based on content. In that process, it is essential that contact is frequent, personal and relaxed as well as productive. However, as we saw in the Argolida, the time-frame of researchers and policy-makers is very different, making the relationship building process more difficult. We also observed that the fundamentally dynamic nature of regional change and the ever-changing agendas of local politicians and inhabitants create problems for designing integrated support tools. Such tools, by their very nature, involve fixing the ‘agenda’ to some degree, a feature that is at odds with supporting (dynamic) regional policy. There is, as has been recognised in the systems literature, a real risk that by the time a support tool has been developed to address a set of issues, those issues are no longer salient to the stakeholders concerned (Ackoff, 1991). So not only do lines of communication and good working relationships need to be established that confer trust, tools must be designed to be flexible and to not overly pre-determine or pre-empt the policy agenda. As recognised by van Daalen et al. (2002), environmental policy and support models both have life-cycles and nobody, except perhaps certain extreme sections of the modelling community, believes that support models

1008

T. Oxley et al. / Environmental Modelling & Software 19 (2004) 999–1010

are a decision-making panacea. Rather, they may provide different types of support (‘eye-openers’, argument support, consensus building, management option evaluation) at different times and are very likely only to be temporarily employed within an organisation. Building in a high degree of modelling flexibility to a support tool means also building in the need for additional work to tune the tool for specific issues and contexts. From our experience, it is not clear that regional policy-makers have the desire, need or resources to embark on such an undertaking. We are left then with a genuine quandary—too much flexibility may make a tool prohibitively costly but too little flexibility may make a tool too restrictive and consequently irrelevant to regional policy.

6. Conclusions Re-using and applying models and experience gained in scientific research to provide policy-support is not a trivial problem. There are a number of potentially very serious ontological and technical issues to be solved when integrating different models and it may just not be possible to simply re-use research models for policy purposes. The three way categorisation of ‘integrate, adapt or rebuild’ covers the development options available, whilst our experience highlighted the likely need to adapt or rebuild given the probable mismatch between model formulations and the needs of the research and policy questions. As a team, we decided that if given the chance to do the task again we would opt for a purer rebuild strategy, leaving our modellers to specify their models with respect to policy questions free from software concerns. We would then use our professional software engineering expertise to implement the resulting sub-models from scratch in a single language and unified system to simplify problems of data flow, data management and variable calculation sequencing. This, we believe, would allow us to develop environmental decision-support technology more efficiently and better tailored to the needs of policy questions. Research models almost certainly must be the basis upon which we develop policy-support models, but this basis is probably more fruitfully viewed in terms of basic design and domain understanding than in terms of the provision of actual software components/sub-models. In addition to the various scientific problems encountered, we addressed the needs and concerns of effective communication with the policy problem-owners and stakeholders. We determined that a team consisting of the right kind and the right number of specialists with suitable experience is as essential as a clear, well-planned project design and schedule. We

believe that the following types of people are required in such a team: . Motivated and visionary policy end-users. . ‘Trans-discipline’ and ‘trans-role’ domain specialists/scientists/model developers. . An architect for the integrated model or DSS model base. . Flexible, highly skilled software system developers. . A professional ‘communication’ specialist. . An experienced project manager. We were also hampered by the fact that the initiative for the project came from outside the area for which it was destined. As such, we were not only strangers to the local end-users, but could be seen as ‘imposed’ or at least ‘unsolicited’. This problem, which we have observed in many aspects of our work, placed us at a disadvantage with parts of the community that we wished to reach. However, using the findings of this research and with additional stakeholder involvement, as part of the MedAction and DesertLinks projects (EC, 2000, 2001) this work has continued towards the development of an integrated policy-support system for land degradation in Mediterranean river basins (see, for example, van Delden et al., 2003a, b). Finally, an important lesson concerns the diffuseness of decision-making in policy matters and the nature of the systems to which environmental decision-support technologies like MODULUS are applied. One can simply not speak of ‘a policy-maker’ as the end-user of a DSS. ‘Decision-making’ and ‘policy-making’ are complex, multi-facetted processes in which a host of different actors exert influence and, in a sense, negotiate the decisions and policies to be made and/or instituted. Models like the one we constructed necessarily have to be multi-facetted, and used to support the identification of possible problems within policy domains rather than convincing anyone of ‘the truth’ or finding ‘the solution’ (Winder, 2003). Socio-natural systems like those addressed by MODULUS are not governed by easily enumerable, time invariant and universal rulesets like physical processes. As discussed in Klu¨ver (2002), systems involving a socio-cultural element are evolutionary; they are complex, structurally adaptive wholes dynamically constructed by their participants. Given these characteristics, we believe that decisionsupport for socio-natural systems is more fruitfully concerned with providing the political actors involved a means of exploration than a set of ‘definite’ solutions. Based upon our experience, we feel that it is reasonable to conclude that the direct re-use and integration of research models may not be the best way to do so.

T. Oxley et al. / Environmental Modelling & Software 19 (2004) 999–1010

Acknowledgements This research was supported by the Commission of the European Union DG-XII Environment (IV) Framework Climatology and Natural Hazards Program Contract ENV4-CT97-0685. We would like to thank Maarten van der Meulen (RIKS bv.), whose technical and programming expertise were crucial to the success of this project. This paper was originally presented at the iEMSs2002 Conference on Integrated Assessment & Decision Support, University of Lugano, Switzerland, 24–27 June 2002.

References Ackoff, R., 1991. The future of operational research is past. In: Flood, R.L., Jackson, M.C. (Eds.), Critical Systems Thinking, Directed Readings. John Wiley and Sons, Chichester, (ISBN 0471-93098-9). Allen, P., Oxley, T., Strathern, M., 1996. A dynamic spatial model of water quality in the Rhone Valley. In: Mouchel, J.M. (Ed.), Coupled ecological and socio-economic modelling in the Rhone basin. Final report, CEE-DGXII, project EV5V-CT94-0547. Barreteau, O., Antona, M., d’Aquino, P., Aubert, S., Boissau, S., Bousquet, F., Dare´, W., Etienne, M., Le Page, C., et al., 2003. Our companion modelling approach. Journal of Artificial Societies and Social Simulation 6 (1), (Forum, http://jasss.soc.surrey.ac.uk/6/2/1.html). Billen, G., 1993. The PHISON River System: a conceptual model of C, N and P transformations in the aquatic continuum from land to sea. In: Wollast, R., Mackenzie, F.T., Chou, L. (Eds.), Interactions of C, N, P and S biogeochemical circles and global change, 521 pp., (NATO ASI Series I4). Billen, G., De Becker, E., Servais, P., Lancelot, Ch., Rousseau, V., 1990. Mode`le mathe´matique de l’estuaire de l’Escaut. Int Rep., Unite´ de Gestion du Mode`le Mathe´matique de la Mer du Nord et de l’Estuaire de L’Escaut. Ref. BH/88/26. Blatsou, C., Mata-Porras, M., Seaton, R.A.F., van der Leeuw, S.E., Winder, N.P. 2000. Policy makers, policies and policy problems seen from the field. In: Engelen, G. (Ed.), MODULUS: A Spatial Modelling Tool for Integrated Environmental Decision Making. Vol. 1, Ch. 2, pp. 29–31. Final report, EU-D9XII (contract ENV4-CT97-0685). Bosch, O.J.H., Ross, A.H., Beeton, R.J.S., 2003. Integrating science and management through collaborative learning and better information management. Systems Research and Behavioural Science 20, 107–118. Burke, S., Mulligan, M., Thornes, J.B., 1998. Key parameters that control recharge at a catchment scale: using sensitivity analysis and path analysis to define homogenous recharge classes. In: Wheater, H.S., Kirby, C. (Eds.), Hydrology in a Changing Environment, vol. II. Wiley, Chichester, pp. 229–243. Costanza, R., 2003. A vision of the future of science: reintegrating the study of humans and the rest of nature. Futures 35, 651–671. EC (European Commission), 2000. MedAction: policies for land-use to combat desertification. (EC-DGXII, EVK2-CT-2000-00085). EC, 2001. Combating desertification in Mediterranean Europe: linking science with stakeholders (DesertLinks). (EVK2-CT-2001-00109). Engelen, G. (Ed.), 2000. MODULUS: a spatial modelling tool for integrated environmental decision making. Final Report, The MODULUS Project, EU-DGXII Environment IV Framework, Climatology & Natural Hazards Programme (contract ENV4CT97-0685), http://www.riks.nl/projects/modulus.

1009

Engelen, G., White, R., Uljee, I., 1997. Integrating constrained cellular automata models, GIS and decision support tools for urban planning and policy making. In: Timmermans, H. (Ed.), Decision Support Systems in Urban Planning. Chapman and Hall. Filippucci, P., Mata-Porras, M. & van der Leeuw, S., 2000. The Alicante experience. In: Engelen, G. (Eds), MODULUS: A Spatial Modelling Tool for Integrated Environmental Decision Making. Vol 1, Ch. 2, pp. 87–95. Final report, EU-DGXII (contract ENV4-CT97-0685). Giannoulopoulos, P., 2000. Groundwater flow and mathematical models in the plain of Argos. Ph.D. Thesis, Agricultural University Of Athens. Jakeman, A.J., Letcher, R.A., 2003. Integrated assessment and modelling: features, principles and examples for catchment management. Environmental Modelling and Software 18, 491–501. Klu¨ver, J., 2002. An Essay Concerning Sociocultural Evolution, Theoretical Principles and Mathematical Models. Dordrecht. Levins, R., 1966. The strategy of model building in population biology. American Scientist 54 (4), 421–431. Mazzoleni, S., Legg, C., Strumia, S., Migliozzi, A. (Eds.), 1998. ModMED, Modelling Vegetation Dynamics and Degradation in Mediterranean Ecosystems, (EU-DGXII, ENV4-CT95-0139). McIntosh, B.S., 2002. Rule-based modelling of vegetation dynamics. Ph.D. Thesis, University of Edinburgh. McIntosh, B.S., Mazzoleni, S., & Coppola, A., 2000. Selection and Evaluation of Models, In: Engelen, G. (Ed.), MODULUS: A Spatial Modelling Tool for Integrated Environmental Decision Making. Vol. 1, Ch. 3 pp. 101–134. Final report, EU-DGXII (contract ENV4-CT97-0685). McIntosh, B.S., Legg, C.J., Csontos, P., Arianoutsou, M., Mazzoleni, S., 2001. Rule-based modelling. In: Mazzoleni, S., Legg, C.J. (Eds.), ModMED: Modelling Mediterranean Ecosystem Dynamics, ModMED III Project, (EU-DGXII, ENV4-CT97-0680). Mulligan, M., 1994. Modelling the hydrology of dryland vegetation competition. Annales Geophysicae 12. Mulligan, M., 1996. Modelling the complexity of landscape response to climatic variability in semi arid environments. In: Anderson, M.G.A., Brooks, S.M. (Eds.), Advances in Hillslope Processes. Wiley, pp. 1099–1149. Mulligan, M., 1998. Modelling the geomorphological impact of climatic variability and extreme events in a semi-arid environment. Geomorphology 24 (1), 59–89. Mulligan, M., & Reaney, S., 2000. In: Engelen, G (Ed.), MODULUS: A Spatial Modelling Tool for Integrated Environmental Decision Making. PatternLITE, a policy relevant version of Pattern for MODULUS, Vol. 1, Ch. 4, pp. 145–200, Final report, EU-DGXII (contract ENV4-CT97-0685). Neelamkavil, F., 1988. Computer Simulation and Modelling. John Wiley and Sons, Chichester. Oxley, T., McIntosh, B.S., Mulligan, M., de Roode, R. 2000. Adaptation, integration and application of an integrative modelling framework. In: Engelen, G. (Ed.), MODULUS: A Spatial Modelling Tool for Integrated Environmental Decision Making. Vol. 2, Ch. 6, pp. 291–378, Final report, EU-DGXII (contract ENV4-CT97-0685). Oxley, T., Jeffrey, P., Lemon, M., 2002. Policy relevant modelling: relationships between water, land use and farmer decision processes. Integrated Assessment 3 (1), 30–49. Oxley, T., Lemon, M., Jeffrey, P., 2003. Indicators of socio-natural change: scientific meaning and contextual interpretation. Journal of Environmental Assessment Policy and Management 5 (1), 1–26. Parker, P., Letcher, R., Jakeman, A., Beck, M.B., Harris, G., Argent, R.M., Hare, M., Pahl-Wostl, C., Voinov, A., Janssen, M., et al., 2002. Progress in integrated assessment and modelling. Environmental Modelling and Software 17 (3), 209–217. Poulovassilis, A., Giannoulopoulos, P., 1999. Groundwater Modelling in the Plain of Argos. Agricultural University of Athens, (EU-DGXII, ENV4-CT95-0159).

1010

T. Oxley et al. / Environmental Modelling & Software 19 (2004) 999–1010

Prigogine, I., 1978. Time, structure and fluctuations. Science 201 (4358), 777–785. Reaney, S., Mulligan, M., 1999. Modelling storms from historic and GCM monthly rainfall data for climate impact studies. Geophysical Research Abstracts 1 (2), 293. Robinson, A.V., 1999. Integrated systems modelling of the interaction between water resources and agriculture. Ph.D. Thesis, Cranfield University. Tress, B., Tress, G., van der Valk, A., Fry, G. (Eds.), 2003. Interdisciplinary and Transdisciplinary Landscape Studies: Potential and Limitations, DELTA Series 2, Wageningen, (ISBN 90 80763713). van Daalen, C.E., Dresen, L., Janssen, M.A., 2002. The roles of computer models in the environmental policy life cycle. Environmental Science and Policy 5, 221–231. van Delden, H., van der Meulen, M., Engelen, G., 2003a. An integrated policy support system for land degradation and desertification in Mediterranean river basins. Framing Land Use Dynamics. In: International Conference, Utrecht University, The Netherlands, 16–18 April. van Delden, H., Mulligan, M., Luja, P., Engelen, G., 2003b. Integration of socio-economic and physical processes through a land

use module. Desertification: Our Future, Our Choice. In: The DesertLinks and MedAction Science Conference, Lisbon, Portugal, 16–17 May. van der Meulen, M., Hahn, B., Uljee, I., Engelen, G., White, R., 2000. Technical integration of models. In: Engelen, G. (Ed.), MODULUS: A Spatial Modelling Tool for Integrated Environmental Decision Making. White, R., Engelen, G., 1997. Cellular automata as the basis of integrated dynamic regional modelling. Environment and Planning B 24, 235–246. Winder, N., 2000. The Path-finder as Historian: An Assay of Modernism, of the Emerging Science of Cultural Ecodynamics and the Mathematics of History. SMC, Kiruna, (ISBN 91-973922-0-0). Winder, N., 2003. Successes and problems when conducting interdisciplinary or transdisciplinary (integrative) research. In: Tress, B., Tress, G., van der Valk, A., Fry, G. (Eds.), Interdisciplinary and Transdisciplinary Landscape Studies: Potential and Limitations, DELTA Series 2, Wageningen, (ISBN 90 80763713). Winder, N., Jeffrey, P., Lemon, M., 1998. Simulation of crop choice dynamics: an application of nested master-equation models. Etudes et Recherches sur les Systemes Agraire et le Developpment 31, 175–189.