Available online at www.sciencedirect.

com

Environmental Modelling & Software 23 (2008) 569e578 www.elsevier.com/locate/envsoft

Spatial planning for lowland stream basins using a bioeconomic model

Paul van Walsum a,*, John Helming b, Louis Stuyt a, Eric Schouwenberg a, Piet Groenendijk a
a

Alterra, Wageningen University and Research Centre, PO Box 47, 6700 AA Wageningen, The Netherlands b Agricultural Economics Research Institute (LEI), Wageningen University and Research Centre, PO Box 29703, 2502 LS, The Hague, The Netherlands Received 12 April 2006; received in revised form 28 August 2007; accepted 29 August 2007 Available online 23 October 2007

Abstract Most lowland stream drainage-basins have a high population density and the land use is very intensive. The permeable subsoil acts as an integrating medium, thus providing a widespread dispersal of leached nutrients and transmission of water-table lowering. This leads to eutrophication and desiccation of stream ecosystems. For providing suggestions with respect to cost-effective and sustainable spatial planning solutions, the Waterwise bioeconomic model has been developed. It combines the accuracy of simulation models with the versatility of optimization techniques to generate land-use patterns along with the appropriate water management, taking into account the preferences of stakeholders with respect to peak discharges, nutrient loading on groundwater and surface water, the biological value of nature areas, and the revenue from agriculture. Computational experiments with the model show, for instance, that a certain goal for the nitrogen load on surface water can be reached at a 40% lower cost if the measures are tailored to the region instead of using generic-style measures towards the same end. 2007 Elsevier Ltd. All rights reserved.
Keywords: Lowland hydrology; Agriculture; Nature desiccation; Flooding; Nutrients; Combination of simulation and optimization; Spatial planning; Bioeconomic model; Stakeholder; EU Water Framework Directive

Software availability Availability is limited to a reduced version, excluding the DRAM model. Name: Waterwise Developer: P.E.V. van Walsum Contact address: Alterra Wageningen UR, PO Box 47, 6700AA Wageningen, NL; paul.vanwalsum@wur.nl First available: 2007 Minimum hardware requirements: Intel Pentium 4, 1 GHz, 256 Mb

Software required: Windows XP, DASH Xpress Programming language: Mosel Software: freely available from http://www.waterwijs.nl; full version plus demo-version with data set (requires the freely available Student edition of Xpress, obtainable from http://www.dashoptimization.com).

1. Introduction Lowland stream basins have traditionally attracted many dwellers, owing to their easy accessibility and high land-use potential. A high population density and intensive land use are the result. The dense network of channels and the permeable subsoil act as an integrating medium, thus providing

* Corresponding author. Fax 31 317 419000. E-mail address: paul.vanwalsum@wur.nl (P. van Walsum). 1364-8152/$ - see front matter 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.envsoft.2007.08.006

570

P. van Walsum et al. / Environmental Modelling & Software 23 (2008) 569e578

a widespread dispersal of leached nutrients and transmission of water-table lowering. This endangers the drinking water supply and leads to eutrophication and desiccation of stream ecosystems, both the aquatic systems in the streams and the terrestrial systems in the stream valleys. The biological value of the latter is due to the presence of shallow water tables in combination with calcium-enriched upward seepage that provides excellent conditions for vegetation requiring pH-buffered soils. Apart from the degradation of nature areas and pollution of groundwater, climate change is adding extra problems; especially the increase of the ooding hazard is becoming manifest. To combat the deterioration of river basins, the European Community has issued a Framework Directive in the Field of Water Policy (http://europa.eu/scadplus/leg/en/s15005.htm) stipulating the achievement of several water-related environmental objectives by the year 2015. For realising the ambitious goals of the directive, it is clear that in many parts of Europe a substantial reallocation of land use will be needed. This creates a demand for decision support systems that can provide suggestions with respect to cost-effective and sustainable spatial planning solutions. The system described by, for instance, on et al. (2000) cannot fully satisfy that demand because Le the predictive functionality can only be used on a trial-and-error basis: a spatial set of measures is specied through an interactive user interface, and the effects on the objective functions are then evaluated. This is repeated until the desired aspiration levels for the objectives are met. Given the endless possibilities for specifying spatial patterns of fertilization measures and land-use reallocations, the achieved result is bound to be suboptimal, meaning that it will not be cost effective. To achieve the latter, the use of optimization techniques is required, as has been done by for instance Nidumolu et al. (2007). But that approach has the disadvantage that the representation of the physical-biological system has been greatly simplied, without having any direct links to an underlying set of simulation models. Such links to simulation models are available in the RiverWare and WaterWare packages (respectively Zagona et al., 2001; Jamieson and Fedra, 1996a,b), so that the versatility of optimization techniques is combined with the accuracy of simulation models. A similar approach is also followed in Waterwise (Van Walsum et al., 2002a), a model for supporting water and land-use planning in lowland stream basins. It is of the holistic type as described by Cai (2008) and Van Delden et al. (2007). Holistic models are intended for providing decision support in basins where diverse interconnected problems exist that are deeply rooted in the way stakeholders make their living and also depend on environmental services. These problems are often very persistent and require for their solution a high degree of cooperation between the stakeholders. One of the hindrances to nding solutions is that downstream water users tend to only see the negative effects of upstream activities, and take the positive ones for granted. Holistic decision support tools can make the underlying interrelationships between stakeholders explicit and also suggest solutions that are efcient for the stream basin community as a whole. They can thus be helpful in building the water-space partnerships that are needed (Van Walsum et al., 2005a). In this article we give a brief

overview of Waterwise, followed by some results demonstrating its potential for supporting the implementation of the EU Water Framework Directive. 2. Models 2.1. Representation of the regional system For predicting effects of measures on a regional hydrologic system and its dependent functions, the following models have been coupled: e SIMGRO (Van Walsum et al., 2004; Veldhuizen et al., 2006) for regional hydrology; SIMGRO is an integrated regional hydrologic model with a timestep-by-timestep two-way coupling of submodels for soil water, groundwater and surface water; e ANIMO (Groenendijk et al., 2005) for leaching of nitrates and phosphates to groundwater and surface water; ANIMO is a process-based simulation model that simulates all relevant components of soil chemistry, including the carbon cycle; for reasons of computational efciency, use has been made of a simplied metamodel based on regression analysis of a large number of computational experiments (Schoumans et al., 2002); e NATLES (Runhaar et al., 1999) for evaluating soil and water site conditions in terms of the potential type of natural vegetation that can develop; e DRAM (Helming, 2005) for the development of agriculture, modelled as a regional farm; DRAM is a regionalized mathematical programming model of agriculture covering the whole of the Netherlands. The coupling mentioned above is of the conventional type; the models are run one after each other. Questions can be answered of the type What is the effect of removing all agricultural drainage on the (potential) value of wet nature areas? The models can, however, not be used for answering questions of the type What is the most cost-effective way to increase the percentage of valuable wet mesotrophic natural grasslands by 10%? To answer such questions a model is needed that is more fully integrated. We have developed such a bioeconomic model using large-scale linear programming (LP, including the use of binary variables) as the integration framework. The LP model is the rst-level model within a multi-level hierarchical system as illustrated by Fig. 1. The idea is to service two modelling objectives (e.g. Orlovski et al., 1986): e achieve a form of model integration that is broad enough for providing the desired decision support; e maintain an acceptable predictive capacity of the (simplied) components. In the terminology of Letcher et al. (2007), at the level of the integrated model the decisions are simulated under conditions of perfect knowledge. If a more accurate estimation of impacts is required, a cycle can be made via the complex

P. van Walsum et al. / Environmental Modelling & Software 23 (2008) 569e578

571

Integrated bioeconomic model simple

complex

hydrology

ecology

economy

Fig. 1. Multi-level modelling with an integrated model connected to complex models.

models at the bottom of the knowledge pyramids in Fig. 1; evaluation via this second level can be seen as expectations based decision making (Letcher et al., 2007). 2.2. Bioeconomic model For obtaining a LP model of a regional hydrologic system and its dependent functions, there broadly are two techniques available, as e.g. explained in Loucks et al. (1981) and Gorelick (1983). The rst technique is that of embedding, which involves the wholesale inclusion of (part of) a model. The second is that of using a repro-function that reproduces the behaviour of the simulation model for a specic type of measure. Both techniques have been used for constructing the bioeconomic model Waterwise. A comprehensive run-down of the model is given (in words) in Appendix A. In the following we will highlight parts of it that we consider to be methodologically interesting. SIMGRO is a complex dynamic simulation model, requiring the repro-function approach for including it in the bioeconomic model. A simpledbut computationally very intensivedmethod for deriving the repro-functions would be to let a land and water management option walk through the study region and then each time do a simulation run to register the effects on the nature areas. Since one simulation run with SIMGRO takes about 10 h, this method is not feasible. To after all arrive at a (realistic) model of spatial interactions we make an intermediate step using an analytical multi-layer
1) measures in agriculture area

steady-state groundwater model. We use it for computing the effect of raising (or lowering) the water table in a spatial planning unit si on the conditions in a planning unit sj. The analytical method is applied for each combination of spatial planning units, which yields the so-called inuence matrix (see for instance Gorelick, 1983; Ahlfeld et al., 2005). The matrix can be used for the superimposition of effects, due to the linear nature of the differential equation describing the steady-state groundwater ow. The inuence matrix is calibrated on the results of sensitivity analysis runs with SIMGRO; in each of these runs a certain measure is uniformly applied to the whole agricultural area in the region. A regression method is used for the calibration. The result is a modelling chain with the following components that are shown in the scheme of Fig. 2: (1) measures in agriculture areas, dened in terms of land use, subsurface drainage, and sprinkling; (2) effects of measures on local water-table conditions in agriculture areas, in terms of effects on the Mean Spring Water table and the Mean Lowest Water table (from the sensitivity analysis runs with SIMGRO); (3) superimposed effects on the steady-state aquifer heads below nature areas, by applying the inuence matrix to the water table effects calculated in step 2); (4) effects on the dynamics of aquifer heads below nature areas, by applying a regression function to the effects calculated in step 3); (5) effects on the water table conditions in nature areas, extracted in the form of tabular functions from the sensitivity analysis runs with SIMGRO; (6) effects of water table conditions in nature areas on the natural vegetation that can develop, extracted in the form of tabular functions from NATLES-evaluations of SIMGRO sensitivity runs. The prediction of head changes in the aquifer below the nature areas is crucial for the validity of the method. An example of a verication (using the SIMGRO model) is given in Fig. 3. The model NATLES for effects on the potential value of vegetation in nature areas requires data with respect to the management (mowing or grazing of grasslands), soil type, and groundwater conditions. The groundwater conditions are

2) water table changes

6) effects in nature area

5) water table changes aquitard aquifer

3) superimposition of effects on aquifer heads

4) X calibration factors

Fig. 2. Modelling chain for calculating hydrologic effects of measures in agriculture areas on nature areas.

572
0.2

P. van Walsum et al. / Environmental Modelling & Software 23 (2008) 569e578

0.1

0.0 y = 0.90x + 0.0006 r = 0.94, r2 = 0.88 -0.1

-0.2 -0.2

-0.1

0.0

0.1

0.2

Ha,SIMGRO (m)
Fig. 3. Verication of the optimization model for water quantity interactions. The predicted changes of the aquifer head as computed by the optimization model (Waterwise) are plotted against those predicted by the simulation model (SIMGRO).

given in terms of the so-called Mean Spring Water table, the Mean Lowest Water table, and the gross seepage to the root zone. (See Van Walsum et al. (2002b) for the coupling between SIMGRO and NATLES; the mean water table refers to over-the-years averaging.) These data are then transformed to suitability maps through a stepwise procedure involving grids in an ArcView software shell (ESRI, 1996). NATLES is incorporated in the bioeconomic model through the reprofunction method in its most simple form: results of sensitivity analyses with the regional hydrologic model SIMGRO are routed through the ArcView shell of NATLES, and are stored in tabular form for use in the bioeconomic model, in step (6) of the procedure given above. A verication example of the combined SIMGRO-NATLES model is given in Table 1 (Van Walsum et al., 2002a; see also Section 3). Results of the NATLES model are presented in terms of nature goal realization W, i.e. the percentage of non-desiccated nature area. The results have been aggregated for nature subareas, each containing several tens of groundwater model nodes; thus the modelling errors shown in Fig. 3 are largely averaged out. As can be seen from the table, the results computed by the simulation models are in nearly all cases better (i.e. have a higher W value) than those computed by the optimization model. These better than differences are due to the fact that the aspiration levels for the nature goal realizations (W

values) are specied by the user; the optimization model then seeks to comply with these constraints. It can happen that strict constraints for some of the nature subareas lead to non-binding constraints for some of the others. This has to do with the proximity of other nature areas with stricter constraints. So the computed better than values are not caused by model defects. To further improve the accuracy of the model, the parameters have also been made dependent on the desired level of goal realization W itself; this yields a higher accuracy than when a general relationship is used. This is especially important for regions where there is a non-linear relationship between the water table effects of step (2) and the effects on the aquifer heads below the nature areas of step (4). Such a non-linearity is usually caused by the presence of surface water channels that become more actively draining due to the higher upward seepage caused by higher aquifer heads. An example involving the mentioned non-linearity is provided in Van Walsum et al. (2006). The model DRAM (Helming, 2005) is a national model for agriculture in the Netherlands. Most of the model equations are in a linear form. That made it possible to realize a style of integration symbolized in Fig. 4: Waterwise overlaps with a substantial part of DRAM. So here the embedding technique has been used. The embedded part DRAM-WW concerns the land balances, the manure balances, and the nutrient balances in terms of the nutritional value of N and P for crops, the balances of the fodder for livestock, and nally the objective function in terms of total revenue. The latter contains terms for the yield of arable land crops, the yield of intensive livestock farming, the yield of dairy farming, and the subsidies on special types of agricultural land-use that are nature friendly. The revenue function also contains terms for the costs of chemical fertilizer, the costs of manure application, the costs of manure export to other regions, the local costs of changing the type of land-use, and the regional costs of expanding a certain type of production. The latter term is derived using the PMP approach (Positive Mathematical Programming, see Howitt, 1995), taking into account the simulated markets at a national scale. Since this part of DRAM is not explicitly included in the bioeconomic model, the functions are delivered to it in the form of quadratic regional cost functions. The resulting convex form of the total revenue function reects the law of decreasing marginal returns on increasing production; the implementation is done using a piece-wise linear function (Loucks et al., 1981). The manner in which the quadratic cost term is derived and included in the model is a typical example of the repro-function method. For relating land and water management measures to peak discharges of the streams, the sensitivity analysis runs with

Table 1 Comparison between the computed indicator for nature goal realization (W in %) by the simplied optimization model and by the SIMGRO-NATLES simulation models Nature area Optimization model SIMGRO-NATLES 1 49 60 (24) 2 15 15 (4) 3 74 77 (33) 4 25 25 (2) 9 48 48 (8) 10 29 31 (7) 11 12 20 (9) 13 21 24 (10) 15 45 62 (30) 17 68 69 (40) 18 91 91 (42) 19 95 93 (34) 20 56 53 (35)

The simulation results for the current state are given in parentheses.

Ha,WATERWISE (m)

P. van Walsum et al. / Environmental Modelling & Software 23 (2008) 569e578

consequences of environmental constraints for agricultural activities

573

DRAM Waterwise DRAMWW

restrictions on total areas of crops regional quadratic cost functions

Fig. 4. Embedding of part (DRAM-WW) of the national agricultural model DRAM in the regional bioeconomic model Waterwise.

SIMGRO are analysed in terms of the incremental ow contribution that each spatial planning unit makes in a period with high rainfall. These contributions are then stored as coefcients of the bioeconomic model. For relating land-use measures to the leaching of nutrients, the results of the sensitivity analyses with SIMGRO are routed through the metamodel of ANIMO (Schoumans et al., 2002) using nutrient surpluses derived from the agricultural model DRAM. Nutrient surpluses are dened in terms of land application minus the crop uptake. The registered effects on the nitrate and phosphate leaching to surface water and groundwater are stored as coefcients. For modelling the spreading of leachates in the aquifers a simple mixing cell model is used as shown in Fig. 5. The nitrate mass-balance equations of the cells (for the equilibrium state) are embedded as equality constraints in the LP model; the concentrations are included as decision variables, i.e. as the unknowns. In order to keep the model linear, the water ows are handled as coefcients that are xed at the beginning of a model run. The equations include a decay term for the denitrication of nitrate under anaerobic conditions. The justication for computing the steady-state equilibrium is that only in this manner can a sustainable planning solution be found. For handling the multiple objective functions the simple constraint method is used. The user/stakeholder supplies aspiration levels for the nature goal realization (reduction of desiccation), the reduction of the peak discharges (taking also into account the possible effects of climate change), and the reduction of the nitrogen leaching. The bioeconomic model rst ascertains whether there is a solution at all, and (if there is one) then nds the land and water use pattern that satises the constraints and at the same time optimizes the revenue from agriculture. The bioeconomic model has been implemented in the Mosel language of the Xpress-mathematical programming package of DASH (2006). The Newton Barrier algorithm is used for solving the resulting linear programming problem. Compared to simplex this interior point method drastically reduces computation times of large-scale problems. 3. Results and discussion The Waterwise model has been applied to the Beerze and Reusel stream basin in the Netherlands. The (twin) basin

covers an area of some 45,000 ha. The nature areas cover roughly 15,000 ha of the area; about 22,000 ha are in use by agriculture. For the bioeconomic model the study region was divided into 4000 spatial units. The implemented model has roughly 200,000 functional decision variables, 60,000 active equations, and about 2 million coefcients in the LP matrix. On a P4, 2.4 GHz PC, the solution time is about 0.5 h. To demonstrate the model we have made a sequence of runs for strategies as listed in Table 2. The objectives for nature goal realization, peak discharges and nitrogen loading have been converted into constraints, and the model was then run to optimize the total economic revenue. It is especially interesting to analyse the results in terms of synergies (or conicts) between the mentioned objective functions that are converted into constraints. For instance, strategy 4 involving the reduction of desiccation and of peak discharges leads to a loss of revenue of 9.7 MV/ year. The total costs of the component strategies 1 and 2 are respectively 2.1 and 1.2, totalling 3.3 MV/year. This yields a synergy of 3.3 9.7 6.4 MV/year. Such a negative value reveals the conict between combating desiccation on the one hand and the reduction of peak ows on the other; to reduce desiccation the eld drainage around the nature areas should be removed. But to reduce peak ows (here with a return period of 10 years) the eld drainage should actually be expanded. The reason for the latter is thatdalthough drainage increases the discharge having a return period of e.g. 0.5 yeardthe drainage increases the storage capacity of the soil for when a really extreme precipitation event occurs. So the drainage lowers the discharge having, e.g. a return period of 10 years by reducing the surface runoff from saturated soils. For many policy makers this result is counterintuitive, and this analysis therefore contributes to understanding the conict between the two objectives. More details of results with respect to desiccation, peak ows and climate change are given in Van Walsum et al. (2005b). The generated land-use pattern for the integrated strategy of run 7 (of Table 2) is shown in Fig. 6. The model includes
Table 2 Optimization results for the indicator total economic revenue, for 7 strategies Strategy 1 2 Description of strategy nature goal realizations as given in Table 1 reduction of peak discharges (with a return period of 10 years) by 20% reduction of nitrogen loading on surface waters by 50%; nature, peak discharges nature, nitrogen loading peak discharges, nitrogen loading nature, peak discharges, nitrogen loading Revenue loss (MV/year) 2.1 1.2 Synergy (MV/year) e e

3 4 5 6 7

10.4 9.7 11.2 12.1 17.2

e 6.4 1.3 0.5 3.5

Current situation: 95 MV/year. The synergy is computed for strategies 4e7 by comparing the total of the revenue losses of the component strategies (1, 2, and 3) with the loss of the combined strategy. For instance the synergy of strategies 1 and 2 is computed as 2.1 1.2 9.7 6.4.

574

P. van Walsum et al. / Environmental Modelling & Software 23 (2008) 569e578

CN,s,i

layer 1

CN,g,i,2

layer 2

cell i

layer ..

Fig. 5. Mixing cell scheme for simulating the transport of nitrate in surface water and in groundwater; the nitrate mass balances of the cells are embedded as equality constraints in the bioeconomic model, with the concentrations in surface water (CN,s) and in groundwater (CN,g) as the decision variables (i.e. as the unknowns in the equations).

a penalty term for deviating from the current state; this symbolizes the costs of a transition, and avoids generating model results that are too far removed from the current state to be realistic. It is interesting to note that the model apparently nds it efcient to remove some of the agricultural land and convert it to nature area; this can be seen from the large expanses of new natural grassland and of new forest. The conversion to new natural grassland is used for reducing the desiccation of natural grasslands. This particularly concerns the low-lying spots that in the past have been drained to make them suitable for agricultural production. The next best option for reducing desiccation is by introducing low intensity grassland around the wet natural grasslands, because this option does not involve the use of eld drainage. The low intensity grassland also has a low manuring intensity, and thus also contributes to reducing the nitrogen loading. New forest is popular with the model because it contributes to two of the three objectives: reduction of nitrogen loading and reduction of peak ows. The latter is achieved through the increased evapotranspiration, which creates more storage capacity in the soil; this mechanism is similar to that of eld drainage, as mentioned earlier. But the increased evapotranspiration also has a lowering effect on the water tables, so the model keeps the new forest at a safe distance from the (wet) natural grasslands. Apart from reducing the intensity of (agricultural) land use it is interesting to note that the model also generates locations with the highest intensity of grassland use, involving high levels of manure application. The economic sense of this tendency is that it simply is very efcient for a farmer to get a high return per unit of area, because that reduces a whole array of costs, especially of labour. We will take a closer look at this by making a comparison between these model results and the results for policies currently being considered. In order to reduce nitrogen leaching it has been considered to set a maximum loss of 60 kg N ha1 year1 for each and every location in the Netherlands; this limit refers to the total loss of nitrogen to groundwater and surface water from a certain eld location. When applied to the study area this

generic-style measure reduces the nitrogen load on surface water from 9.5 mg N l1 to 4.5 mg N l1 at the outlet of the basin. Roughly half of this load consists of nitrogen that reaches surface water by surface runoff and by shallow leaching to groundwater and subsequent drainage to surface water at the eld scale. The other half reaches the surface water after deep inltration to groundwater, transport through the groundwater (modelled with the mixing cells of Fig. 5), upward seepage, and nally drainage to surface water in the streams. (The concentrations are in fact loads, and not the real concentrations; nitrogen retention processes in surface water are not modelled). The computed loss of revenue from agriculture is 19%. A computational experiment was made with the bioeconomic model to see whether the same 4.5 mg N l1 could be achieved at a lower cost. The model showed that the 4.5 mg N l 1 could also be achieved at a revenue loss of only 11% (run 3 of Table 2). As can be seen from the comparison of the nitrate concentrations in Fig. 7, the generic-style measures produce a concentration pattern (Fig. 7a) that is much more evenly distributed than if the measures are tailored for a minimum loss of agricultural revenue (Fig. 7b). In the latter case the concentrations in uplands are much higher, because in these parts of the region the bioeconomic model does not remove all of the medium- and high-intensity dairy farming like the generic-style measures do. The reasoning behind this strategy is that if the high nitrogen concentrations are in the uplands, the travel times through the deep subsoil are the longest, and therefore the denitrication of the nitrate can reduce the concentrations by the time the water reaches the surface water system through upward seepage and drainage: in the right hand map (Fig. 7b) the concentrations near the streams are in general lower than in the left-hand map (Fig. 7a) of the generic-style measures. By making use of the groundwater as a denitrication machine the bioeconomic model achieves the same environmental goal (4.5 mg N l1 at the basin outlet) at a 40% lower cost (11% loss of revenue instead of 19%). The model coefcients with respect to nitrogen loading and denitrication

P. van Walsum et al. / Environmental Modelling & Software 23 (2008) 569e578

575

Fig. 6. Generated land-use pattern for the integrated strategy (run 7 of Table 2) involving reduction of desiccation in existing nature areas, reduction of peak discharges, and reduction of nitrogen loading on surface waters.

are, however, very uncertain, because much is not yet known about the underlying processes. But this example does demonstrate how the bioeconomic model can take advantage of the way a regional system functions in order to achieve environmental goals at the lowest possible cost. 4. Concluding remarks A bioeconomic model has been developed for spatial planning of integrated land and water management in lowland stream basins. The technique of linear programming (and its extension involving the use of binary variables) has been used as a framework for the integration of models. The system owes its practical relevance to: e the possibility for the user/stakeholder to specify goals and constraints for the desiccation of nature areas, the nutrient loading on groundwater and surface water, peak discharges, and the revenue from agriculture;

e the predictive accuracy of the simplied submodels incorporated in the bioeconomic model, based on results of simulations with complex dynamic models; simulation models are also used for verication of the spatial solutions found by the bioeconomic model; e the use of state-of-art optimization technology, providing a spatial resolution of 10 ha for basins of up to 50,000 ha, within acceptable computation times that are needed for facilitating a decision-making process. The latter point has been demonstrated by the successful implementation for the Beerze and Reusel basin in the Netherlands. A series of computational experiments demonstrates the potential of the model for revealingdand quantifyingdconicts and synergies between regional objectives. Suggested solutions can be counterintuitive, thus deepening the insight into the regional system functioning. It can for instance be economically efcient in some parts of a basin to relax the environmental constraints on agriculture, in order to reach goals

576

P. van Walsum et al. / Environmental Modelling & Software 23 (2008) 569e578

(a)

(b)

NO3-N(mg/l) 0-2 2-5 5-10 10-25 >25

N 0 2 Kilometers

Fig. 7. Generated patterns of the nitrate concentration in the rst aquifer (second layer of the subsoil schematization) for generic-style measures to decrease the amount of nitrogen losses to the environment (a) and for tailor-made measures using the bioeconomic model (b).

at the stream basin scale. Such a differentiated approach can also help in obtaining stakeholder support. The message here is to not think one-dimensionally about how to achieve environmental goals. Crucial for the success of the simplied modelling in the bioeconomic model is that the submodels are accurate enough to facilitate the computational process of searching for a costeffective solution. For the nal estimation of the goals that actually are achieved the simulation models are needed. The verication with SIMGRO-NATLES for the effects of measures aimed at combating desiccation showed that the results of the simplied submodel were accurate within 10e15% of the simulation models. But the only verication of a simplied optimization model that really counts is whether the model can come up with solutions that make better use of the regional system than is the case with hand-made sets of spatial measures. For obtaining formal proof of that it would be necessary to dene the level of intelligence behind the hand-made alternatives. We have conned ourselves to giving a comparison between the (lack of) cost efciency of generic style measures and the efciency of measures obtained through optimization aimed at reducing the nutrient loading on surface waters. The model is now being further developed (Van Walsum, 2007; Mysiak, 2007) within the scope of the Newater project

(Pahl-Wostl and Kabat, 2004), which is focused on developing new approaches to adaptive water management under uncertainty. For handling the latter, the model is being extended to include multiple events, meaning that the decision simulation will simultaneously take into account different possibilities of external conditions for climate and socio-economy. We expect that in the near future the model will play a role in discussions about the implementation of the EU Water Framework Directive. A combination with decision support systems like MULINO (Giupponi, 2007) and Watersketch (Ulvi et al., 2007) could enhance its application potential by structuring the interaction with the stakeholders.

Acknowledgements This research was funded by the Strategic Expertise Development Fund of the former Directorate for Agricultural Research (DLO) that is now part of the Wageningen University and Research Centre. Use has been made of experience gained during IIASAs Regional Water Policies Project in the period 1983e1985, led by S.A. Orlovski. Research funding for the current model development is from the EU NeWater project (http://www.newater.info).

P. van Walsum et al. / Environmental Modelling & Software 23 (2008) 569e578

577

Appendix A. Brief description of Waterwise A.1. Introduction A linear programming (LP) model consists of three types of components: decision variables, constraints, and objective functions. The list of decision variables does not only include variables that directly relate to the actual policy decisions, but also includes a large number of auxiliary variables that are needed for describing the system functioning. The land-use options are described using continuous variables on the interval [0,1] instead of binary ones. Most model results for these decision variables are anyhow equal to zero or to unity; subsequent rounding of those that are not does not signicantly affect the values of objective functions. Strict binary variables are used for representing non-linear aspects of the surface water system. The term constraint is also used for the system equations, i.e. that part of the matrix that describes the functioning of the regional system with all its interdependencies. The method used here for handling the multi-objectivity is that of the simple constraint method, meaning that all but one of the objective functions are converted into constraints using the desired aspiration level on the right-hand side. The regional economic revenue (as far as it is related to the decision variables) is the objective function that gets optimized. A.2. Decision variables All decision variables are non-negative:

e e e e e e e

e e e e e e

Manure balances per planning unit i Manure export/import to/from other regions Nutrient balances per planning unit i and land-use type l Balances of locally produced and consumed animal feedstuffs (maize, grass) Auxiliary equations for constructing piece-wise linear agricultural cost functions Inuences of locally induced water-table changes on regional aquifer heads Constraints on regionally induced aquifer-head changes, to comply with the aspiration levels of the wet nature area indicators for natural grasslands (see objective functions) Summation of peak discharge contributions (of land use and water management options) in downstream direction of the stream network Constraints on peak discharges at key points in the stream network (see objective functions) Summation of nutrient loading (of land use and water management options) on surface water trajectories Summation of nutrient loading in downstream direction of stream network Loading of nitrogen on groundwater Nitrogen mass balances in mixing cells of groundwater and surface water for computing steady-state equilibrium Constraints on concentrations, according to the aspiration levels at key points in the stream system (see objective functions)

A.4. Objective functions e Fractions of land use type l with water management option w in planning unit i e Increments and decrements of land use, as compared to the current state e Chemical fertiliser applications using nutrient n on land use l in planning unit i e Manure applications using manure type m on land use l in planning unit i e Manure transports to other regions e Intensities of animal husbandry type h in planning unit i e Locally induced changes of the mean spring water table and the mean lowest water table in all planning units i e Regionally induced changes of the mean spring aquifer head and mean lowest aquifer head below planning units i that are nature areas e Peak surface water ows through the trajectories o e Concentrations of nitrogen and phosphorus in surface water trajectory o e Concentrations of nitrogen in groundwater of planning unit i and aquifer a e Total regional economic revenue (related to the decision variables), including costs of land and water management practices, costs of changing land use type, subsidies for multifunctional land use e Goal realization in wet natural grasslands e Peak discharge at key points in the stream network, with a return period of 10 years e Nitrogen and phosphorous concentrations at key points in the surface water system References
Ahlfeld, D.P., Barlow, P.M., Mulligan, A.E., 2005. GWM e A Ground-Water Management Process for the US Geological Survey Modular GroundWater Model (MODFLOW-2000). Open-File Report 2005-1072. US Geological Survey. Cai, X., 2008. Implementation of holistic water resources-economic optimization models for river basin management e reective experiences. Environmental Modelling & Software 23, 2e18. DASH, 2006. XPRESS-MP Reference Manual. Dash Associates, Blisworth, UK. ESRI, 1996. Avenue. Customization and Application Development for ArcView. Environmental Systems Research Institute, New York. Giupponi, C., 2007. Decision support systems for implementing the European Water Framework Directive: the MULINO approach. Environmental Modelling & Software 22, 248e258. Gorelick, S.M., 1983. A review of distributed parameter groundwater management modelling methods. Water Resources Research 19 (2), 305e319.

A.3. Constraints e Sum of land use fractions per planning unit i e Diverse interrelationships between land use types, describing the farming system per planning unit i and in the region as a whole

578

P. van Walsum et al. / Environmental Modelling & Software 23 (2008) 569e578 land to surface waters by means of a new methodology. Water Science and Technology 45 (9), 177e182. Ulvi, T., Visuri, M., Hellsten, S. (Eds.), 2007. Proceedings of the European Symposium of Spatial Planning Approaches towards Sustainable River Basin Management. Finnish Environment Institute, Helsinki. see also. http:// www.watersketch.net. Van Delden, H., Luja, P., Engelen, G., 2007. Integration of multi-scale dynamic spatial models of socio-economic and physical processes for river basin management. Environmental Modelling & Software 22, 223e238. Van Walsum, P.E.V., Helming, J.F.M., Schouwenburg, E.P.A.G., Stuyt, L.C.P.M., De Bont, C.J.A.M., Vereijken, P.H., Kwakernaak, C., Van Bakel, P.J.T., Van Staalduinen, L.C., Groenendijk, P., Ypma, K.W., 2002a. Waterwijs; plannen met water op regionale schaal [Waterwise; Spatial Planning Based on Water at a Regional Scale]. Report 433. Alterra, Wageningen. download from. http://www.waterwijs.nl. Van Walsum, P.E.V., Verdonschot, P.F.M., Runhaar, J., 2002b. Effects of climate and land-use change on lowland stream ecosystems. Nationaal Onderzoek Programma Mondiale Luchtverontreiniging en Klimaatverandering (NOP). Alterra, Wageningen. Report 523. Van Walsum, P.E.V., Veldhuizen, A.A., Van Bakel, P.J.T., Van der Bolt, F.J.E., Dik, P.E., Groenendijk, P., Querner, E.P., Smit, M.F.R., 2004. SIMGRO 5.0.2, Theory and Model Implementation. Alterra, Wageningen. Report 913.1. Van Walsum, P.E.V., Aerts, J.C.J.H., Krywkow, J., Van der Veen, A., Der Nederlanden, H., Bos, M.Q., Ottow, B.T., 2005a. Framework for integrated design of water and land management systems; towards robust water-space partnerships as a basis for adaptive water management. Report to the NeWater project. USF, Osnabruck. http://www.newater.info. Van Walsum, P.E.V., Runhaar, J., Helming, J.F.M., 2005b. Spatial planning for adapting to climate change. Water Science and Technology 51 (5), 46e52. Van Walsum, P.E.V., Runhaar, J., Veldhuizen, A.A., Jansen, P.C., 2006. Duurzaam waterbeheer Langbroekerwetering; Fase 2: Verkenning van het Gewenste Grond- en Oppervlaktewaterregime met Waterwijs. Alterra, Wageningen. Rapport 1155. Van Walsum, P.E.V., 2007. Waterwise; A Planning Tool for Adaptive Land and Water Management. USF, Osnabruck. http://www.newater.info. Report to the NeWater Project. Veldhuizen, A.A., Van Walsum, P.E.V., Lourens, A., Dik, P.E., 2006. Flexible integrated modeling of groundwater, soil water and surface water. Proceedings of MODFLOW 2006. IGWMC, Golden, CO. Zagona, E.A., Fulp, T., Shane, J.R., Magee, T., Gorano, H.M., 2001. RiverWare: a generalized tool for complex reservoir systems modeling. Journal of the American Water Resources Association (AWRA) 37 (4), 913e929.

Groenendijk, P., Renaud, L.V., Roelsma, J., 2005. Prediction of nitrogen and phosphorus leaching to groundwater and surface waters. Process descriptions of the Animo 4.0 model. Alterra, Wageningen. Report 983. Helming, J.F.M., 2005. A model of Dutch agriculture based on positive mathematical programming with regional and environmental applications. PhD thesis, Wageningen University and Research Centre, Wageningen, The Netherlands. Howitt, R.E., 1995. Positive mathematical programming. American Journal of Agricultural Economics 77, 329e342. Jamieson, D.G., Fedra, K., 1996a. The WaterWare decision-support system for river-basin planning. 1. Conceptual design. Journal of Hydrology 177, 163e175. Jamieson, D.G., Fedra, K., 1996b. The WaterWare decision-support system for river-basin planning. 2. Planning capability. Journal of Hydrology 177, 177e198. on, L.F., Lam, D.C., Swayne, D.A., Farquhar, G.J., Soulis, E.D., 2000. Le Integration of a nonpoint source pollution model with a decision support system. Environmental Modelling & Software 15, 249e255. Letcher, R.A., Croke, B.F.W., Jakeman, A.J., 2007. Integrated assessment modelling for water resource allocation and management: a generalised conceptual framework. Environmental Modelling & Software 22, 733e 742. Loucks, D.P., Stedinger, J.R., Heith, D.A., 1981. Water Resources Systems Planning and Analysis. Prentice-Hall, Englewood Cliffs, NJ. Mysiak, J. (Ed.), 2007. Specication for Enhancing Existing Tools. Report to the NeWater Project, Deliverable 4.2.2. USF, Osnabruck. http:// www.newater.info. Nidumolu, U.B., Van Keulen, H., Lubbers, M., Mapfumo, A., 2007. Combining interactive multiple goal linear programming with an inter-stakeholder communication matrix to generate land use options. Environmental Modelling & Software 22, 73e83. Orlovski, S., Kaden, S., van Walsum, P.E.V., 1986. Decision Support Systems for the Analysis of Regional Water Policies. WP-86e33. International Institute for Applied Systems Analysis, Laxenburg, Austria. Pahl-Wostl, C., Kabat, P., 2004. New approaches to adaptive water management under uncertainty (NEWATER). Integrated Project in PRIORITY 6.3 Global Change and Ecosystems in the 6th EU Framework Programme. Runhaar, J., Boogaard, H.L., Van Delft, S.P.J., Weghorst, S., 1999. Natuurgericht Landevaluatiesysteem (NATLES)[Nature Oriented Land Evaluation System]. SC-DLO Rapport 704. Alterra, Wageningen. Schoumans, O.F., Mol-Dijkstra, J., Akkermans, L.M., Roest, C.W.J., 2002. SIMPLE: assessment of non-point phosphorus pollution from agricultural