(EPA R82-4766-010)

INTEGRATED ECOLOGICAL ECONOMIC MODELING OF THE PATUXENT RIVER WATERSHED, MARYLAND

Executive Summary

Institute for Ecological Economics
University of Maryland Center for Environmental Science,
Box 38, Solomons, MD 20688-0038
E-mail: voinov@cbl.umces.edu
NSF/EPA Partnership for Environmental Research - Water and Watersheds
December 1, 1999

Objective of Research

There has been a major movement over the last decade toward place-based, ecosystem-based, and watershed-based management. To support this effort to effectively manage the complex interactions between human and natural systems at the watershed scale, integrated (across scales and across disciplines) scientific and technical knowledge and models are needed. We have developed an integrated modeling framework aimed at addressing these goals. The approach evolved from work in coastal Louisiana (Costanza et al 1990) and in the Everglades (Fitz et al 1993). Current work is focused on the Patuxent river watershed in Maryland, one of the best studied tributaries of the Chesapeake Bay, and one that has often been used as a model of the entire Bay system. In particular the modeling framework is aimed at addressing the following general questions.

1. What are the quantitative, spatially explicit and dynamic linkages between land use and terrestrial and aquatic ecosystem productivity and health.
2. What are the quantitative effects of various combinations of natural and anthropogenic stressors on ecosystems and how do these effects change with scale.
3. What are useful ways to measure changes in the total value of the landscape including both marketed and non-marketed (natural system) components and how effective are alternative mitigation approaches, management strategies, and policy options toward increasing this value.

The Patuxent Landscape Model (PLM; http://iee.umces.edu/PLM/PLM1.html) was designed to serve as a tool in a systematic analysis of the interactions among physical and biological dynamics of the watershed, conditioned on socioeconomic behavior in the region. A companion socioeconomic model of the region's land use dynamics was developed to link with the PLM and provide a means of capturing the feedbacks between ecological and economic systems. Because of the complex feedbacks and nonlinear dynamics of this watershed, a "systems" approach was necessary. A key part of this process was the development of an integrated, dynamic, spatially explicit simulation model.

In the ecological component of the model, the important processes that affect plant communities are simulated within the varying habitats distributed over the landscape. The principal dynamics modeled are: 1) plant growth in response to available sunlight, temperature, nutrients, and water; 2) flow of water plus dissolved nutrients in three dimensions; 3) decomposition of dead organic material and formation of soil organic matter. Using this approach to incorporating process-based data at a reasonably high spatial, temporal and complexity resolution within the entire watershed, the changing spatial patterns and processes can be analyzed within the context of altered management strategies, such as the use of agricultural Best Management Practices (BMPs) (e.g. reduced tillage).

Accomplishments and Research Results

The modeled landscape is partitioned into a spatial grid of nearly 2,500 square unit cells. The model is hierarchical in structure, incorporating an ecosystem-level "unit" model that is replicated in each of the unit cells representing the landscape. The General Ecosystem Model (GEM) which was developed for the Everglades Landscape Model (ELM) (Fitz et al. 1996), was modified for use within the framework of the PLM. The model was reformulated on a modular basis, with modules representing functional components of the system and capable of being run and calibrated independently (Voinov, et al., 1999).

The same unit model structure runs in each cell. Individual modules are parameterized according to habitat type and georeferenced information for a particular cell. The habitat-dependent information is stored in a parameter database which includes initial conditions, rate parameters, stoichiometric ratios, etc. The habitat type and other location-dependent characteristics are referenced through links to geographic information system (GIS) files. Thus, when run within the spatial framework of the PLM, the landscape response to hydrology and water quality is effectively simulated as material flows between adjacent cells. The independent modules and the full unit model have also been run in the spatial implementation and rigorously tested at the full watershed scale. Sensitivity analysis was used to gain insight about the model dynamics, showing the varying response of plant production to different nutrient requirements, with subsequent changes in the soil water nutrient concentrations and total water head. Changes in the plant canopy structure resulted in differences in transpiration, and consequently water levels and plant production.

The spatial model combines the dynamics of the unit model which are calculated at each time step for each cell in the landscape, and adds the spatial fluxes which control the movement of water and materials between cells. Each cell generates stock and flow values which provide input to or accept output from the spatial flux equations.

After the water head in each raster cell is modified by the vertical fluxes controlled in the GEM unit model, the surface water and its dissolved or suspended components move between cells based on one of the two algorithms used. In the first algorithm a certain portion of water is taken out of a cell and added to a cell downstream defined by the link map. This may not be the adjacent cell, but a cell several links down the path of the flow. The length of the flow path is defined by the amount of water fluxed and is calibrated so that the water flow rates match gage data. The other algorithm checks that water movement stops when the water heads in two adjacent cells equilibrate. While the first algorithm works well for the piedmont area with significant elevation gradients, the second one is more appropriate for the coastal plain region where there are significant areas of low relief and tidal forces which permit counterflows.

The ecological model is linked to a companion economic model that predicts the probability of land use change within the seven counties of the Patuxent watershed (Bockstael 1996). The economic model allows human decisions to be modeled as a function of both economic and ecological spatial variables. Based on empirically estimated parameters, spatially heterogeneous probabilities of land conversion are predicted as functions of predicted land values in residential and alternative uses, and costs of conversion. Land value predictions themselves are modeled as functions of local and regional characteristics. The predictive model of land use conversion generates the relative likelihood of conversion of cells, and thus the spatial pattern of greatest development pressure. To predict the absolute amount of new residential development, the probabilistic land use conversion model must be combined with models of regional growth pressure. Linking the ecological and economic models allows the effects of both direct land use change through human actions and indirect effects through ecological change to be evaluated, as well as the feedbacks between the two.

A variety of spatially and temporally disaggregated data is required to develop and calibrate the PLM model. The database we have assembled includes time series, spatial coverages (maps) and parameters. The model data base contains the data which drive the model forcing functions, parameterize equations and provide calibration and verification data for adjusting model parameters and comparing model output to the real system. The data base was developed from extensive data sets collected for the Patuxent watershed by various governmental agencies, academic institutions and research programs. Existing data for the local region were supplemented with broader regional data bases.

To adequately test model behavior and to reduce computational time, we performed the calibration and testing at several time and space scales, and for the unit model independently of the full spatial model. We developed a Model Performance Index (MPI: Villa, 1997) to study the model's response to parameter changes. The MPI framework allows one to develop an error function which can handle the full range of variables and data quality that usually confront complex models. It employs a multi-criteria approach, which allows user weighting of the model variables to reflect their degree of importance and also weighting of the data to reflect its quality. It can deal with both quantitative and semi-quantitative information about the expected behavior of the state variables (like the pattern of temporal autocorrelation, boundaries, steady state, etc.).

Calibrating and running a spatial model of this level of complexity and resolution requires a multi stage approach. We first identified two spatial scales at which to run the model - a 200 m and 1 km cell resolution. We then identified a hierarchy of subwatersheds. The whole watershed has been divided into a set of nested subwatersheds to perform analysis at three scales. The inclusion of plant and nutrient dynamics improved the model’s hydrologic performance in comparison to the output generated with no account for these modules. The spatially explicit representation of plant and nutrient dynamics modifies the evapotranspiration and interception fluxes in the model, making the model performance more realistic. It was also essential for scenario runs that take into account land use and cover changes, in which these changes modify the hydrologic fluxes in the watershed.

The goal of the linked ecological economic model development was to test alternative scenarios of land use management. A wide range of future and historical scenarios may be explored using the calibrated model. We have developed scenarios based on the concerns of county, state and federal government agencies, local stakeholders and researchers. The following set of initial scenarios was considered:

• A group of historical scenarios based on the USGS reconstruction (Buchanan, et al. 1998) of land use in the Patuxent watershed:

• (1) 1700 — pre-development era. Most of the area forested, zero emissions.
• (2) 1850 — agro-development. Almost all the area under agricultural use, traditional fertilizers (marl, river mud, manure, etc.), low emissions.
• (3) 1950 — decline of agriculture, start of reforestation and fast urbanization.
• (4) 1972 — maximal reforestation, intensive agriculture, high emissions.

• (5) Baseline scenario. We use 1990 as a baseline to compare the modeling results. The 1990-1991 climatic patterns and nutrient loadings were used.
• (6) 1997 land use pattern. This data set has just recently been released and we used it with the 1990-1991 forcings to estimate the effect of landuse change alone.
• (7) Buildout scenario. With the existing zoning regulations, we assumed that all the possible development in the area occurred. This may be considered as the worst case scenario in terms of urbanization and it’s associated loadings.
• (8) Best Management Practices (BMP) — 1997 land use with lowered fertilizer application and crop rotation. These management practices were also assumed in the remaining scenarios.

• A group of scenarios of change in land use over the 5 years following 1997 (i.e. for 2003) developed based on the Economic Land Use Conversion (ELUC) Model by N. Bockstael:

• (9) Development as usual
• (10) Development with all projected sewer systems in place
• (11) Development with no new sewers but contiguous patches of forest 500 acres and more protected
• (12) Development with all sewers in place and contiguous forest protected

• Another group of hypothetical scenarios to study more dramatic change in land use patterns using the 1997 land use as the starting point:

• (13) Conversion of all agricultural land into residential
• (14) Conversion of all agricultural land into forested
• (15) Conversion of all residential land into forested
• (16) Conversion of all forested land into residential
• (17) Residential clustering — conversion of all low density residential land use into urban around 3 major centers
• (18) Residential sprawl — conversion of all high density urban into residential randomly spread across the watershed.

The scenarios were driven by changes in the Landuse map, the Sewers map, patterns of fertilizer application, amounts of atmospheric deposition, and location and number of dwelling units. We compare the model output in the different scenarios looking at nitrogen concentration in the Patuxent River as an indicator of water quality, changes in the hydrologic flow and changes in the net primary productivity of the landscape.

Table. Some results of scenario runs for the Patuxent Model. The buildout conditions (LUBO) were estimated based on the existing zoning maps and the average population densities for particular land use types. The buildout conditions represent the "worst case" scenario. The ELUC scenarios (LUB1-4) are based on the model by N. Bockstael. The historical scenarios (LU1700 - 1972) are a reconstruction based on estimates done by USGS. Total NPP (g/m²/day) presents the average across the whole watershed productivity of the plant ecosystem. It well represents the approximate proportion of forested and agricultural land use types, which have a larger NPP than the residential ones. N gw.c. is the average concentration on N in the ground water. Since ground water is fairly slow variable in the model and the model had only 1 year of relaxation time in the experiments performed, it is most likely that this parameter does not adapt fast enough to track the changes assumed under different scenarios. Wmax is the total of the 10% of the flow that is maximal over one year period. This represents the peak flow. Wmin is the total of the 50% of flow that is minimal over one year period. This is an indicator of the baseflow.

		Forest	Resid	Urban	Agro	Atmos	Fertil	Decomp	Septic	N aver.	N max	N min	Wmax	Wmin	N gw c.	NPP
		number of cells				kg/ha/year				mg/l			m/year		mg/l	kg/m2/d
1	LU1700	2386	0	0	56	2.67	0.00	253.03	0.09	4.14	47.38	0.06	71.934	31.493	1.356	50.308
2	LU1850	348	7	0	2087	5.35	93.15	123.83	3.54	5.65	51.07	0.25	71.849	27.797	1.675	15.739
3	LU1950	911	111	28	1391	96.27	113.31	144.66	9.44	12.03	124.53	0.32	74.112	26.447	1.710	26.586
4	LU1972	1252	223	83	884	85.58	156.21	175.07	9.60	18.29	281.36	0.28	78.711	23.556	1.723	33.568
5	LU1990	1315	311	92	724	80.23	114.58	164.57	15.49	10.82	168.95	0.10	83.081	21.191	1.676	31.670
6	LU1997	1195	460	115	672	80.23	112.75	150.34	22.81	10.28	106.86	0.14	85.312	19.624	1.691	28.974
7	LUBO	312	729	216	1185	85.58	184.34	74.72	26.22	12.02	159.00	0.28	97.957	15.027	1.862	12.713
8	BMP	1195	460	115	672	80.23	61.14	170.03	15.49	6.50	58.47	0.16	83.307	21.694	1.621	30.680
9	LUB1	1129	575	134	604	85.58	97.14	145.45	29.45	8.48	68.30	0.32	85.414	19.631	1.709	26.323
10	LUB2	1147	538	134	623	85.58	98.03	148.19	16.03	8.55	66.52	0.24	85.269	19.667	1.687	26.820
11	LUB3	1129	577	134	602	85.58	96.96	145.37	29.54	8.43	66.72	0.32	85.417	19.654	1.709	26.315
12	LUB4	1133	564	135	610	85.58	97.55	146.12	17.17	8.56	66.62	0.20	85.250	19.494	1.689	26.438
13	S5a2r	1195	1132	115	0	85.58	47.26	127.65	52.70	6.71	52.84	0.29	83.069	13.182	1.651	25.597
14	S6a2f	1867	460	115	0	85.58	19.19	197.63	21.78	7.25	45.66	0.24	84.928	18.857	1.516	39.677
15	S6r2f	1655	0	115	672	85.58	82.14	203.00	5.50	7.82	91.15	0.25	87.805	28.359	1.617	37.289
16	S7f2r	0	1655	115	672	85.58	143.07	30.40	75.99	9.40	170.17	0.29	89.734	10.004	1.922	3.477
17	S8clust	1528	0	276	638	85.58	78.36	187.85	13.02	8.87	91.66	0.39	121.236	24.621	1.633	34.241
18	S9spra	1127	652	0	663	85.58	106.77	147.81	26.61	7.05	35.48	0.30	70.130	20.679	1.718	26.441

Comparing the effect of various land use change scenarios on the water quality in the river shows that there is no obvious connection between the nutrient loading to the watershed and the nutrient concentration in the river. However some conclusions can be drawn. The effects of loadings which are distributed more evenly over the year are much less pronounced than those which occur sporadically . For example, fertilizer applications that occur once or twice a year increase the average nutrient content and especially the maximum nutrient concentration quite significantly, whereas the effect of, say, atmospheric deposition is much more obscure. The difference in atmospheric loading between Scenarios (1) and (3) is almost 2 orders of magnitude, yet the nutrient response is only 5-6 times higher, even though loadings from other sources also increase. Similarly the effect of septic loadings that are occurring continuously is not so large.

The average N concentration is well correlated (corr=0.77) with the total amount of nutrients loaded. The effect of fertilizers is also high (corr=0.74), while the effect of other sources is much less (septic corr=-0.0075; decomposition corr=-0.2267; atmosphere corr=0.49). The fertilizer application defines the maximum nutrient concentrations (corr=0.71), with the total load also playing an important role (corr=0.60), whereas the contributions of individual sources is less pronounced (septic corr=0.0878; decomposition corr=-0.271; atmosphere corr=0.28). Even the groundwater concentrations of nutrients are closely related to the fertilizer applications (corr=0.89), however in this case the septic loadings play a larger role (corr=0.59), even a more important one than the total N loading (corr=0.44).

The hydrologic response is quite strongly driven by the land use patterns. The peak flow (max 10% of flow) is almost entirely determined by urbanization (corr=0.94). The baseflow (min 50% of flow) is somewhat related with the number of forested cells (corr=0.54), but obviously many other factors also influence it.

Different land use patterns result in quite significant variations in the net primary productivity (NPP) of the watershed, both in the temporal and in the spatial domains. The predevelopment 1700 conditions produce the largest NPP, while under Build Out conditions NPP is the lowest. In the latter case the dynamics of NPP are more representative of the agricultural landuse with higher NPP values attained later in the year as crops mature. Interestingly under the BMP scenario with lower fertilizer applications we still get a higher NPP than under reference conditions of 1997, because of the crop rotation and growth of winter wheat that matures earlier in the season than corn.

The major result of the analysis performed thus far is that the model behaves well and produces plausible output under significant variations in forcing functions and land use patterns. It can therefore be instrumental for analysis and comparisons of very diverse environmental conditions that can be formulated as scenarios of change and further studied and refined as additional data and information are obtained. The real power of the model comes from its ability to link hydrology, nutrients, plant dynamics and economic behavior via land use change. The model allows fairly site specific effects to be examined as well as regional impacts so that both local water quality and Chesapeake Bay inputs can be considered. The linked ecological economic model is a potentially important tool for addressing issues of land use change. The model integrates our current understanding of certain ecological and economic processes to give best available estimates of effects of land use or land management change. The model also highlights areas where knowledge is lacking and where further research could be targeted for the most impact.

Publications

• Voinov, A., H. Voinov, R. Costanza. (1999) Landscape modeling of surface water flow: 2.. Patuxent watershed case study. Ecological Modelling: 119, 211-230.
• Maxwell, T., 1999. "A Parsi-Model Approach to Modular Simulation." Environmental Modeling and Software: Special Issue on Modular Modeling , v. 14, pp. 511-517.
• Voinov, A., R. Costanza, L. Wainger, R. Boumans, F. Villa, T. Maxwell and H. Voinov. Patuxent landscape model: integrated ecological economic modeling of a watershed. (1999) Journal of Ecosystem Modelling and Software: 14, 473-491.
• Voinov, A. and R. Costanza. 1999. Watershed management and the web. Journal of Environmental Management. 56:21-245.
• Voinov, A., R. Costanza, L. Wainger, R. Boumans, F. Villa, T. Maxwell, and H. Voinov. 1998. Integrated ecological economic modeling of watersheds. pp. 35-40 in: A Sydow and J-Y Yu (eds.), 1998 Conference on Mission Earth: Modeling and Simulation of the Earth System. The Society for Computer Simulation International, San Diego, CA.
• Voinov, A. and R. Costanza. 1998. Watershed management over the web. pp. 167-168 in: P. A. Fishwick, D. R. C. Hill, and R. Smith (eds.), 1998 International Conference on Web Based Modeling and Simulation: Simulation and Modeling Technology for the Twenty-First Century. The Society for Computer Simulation International, San Diego, CA.
• Maxwell, T. and R. Costanza, 1997. "A Language For Modular Spatio-Temporal Simulation." Ecological Modelling , v. 103, no. 2,3, pp. 105-113.
• Maxwell, T. and R. Costanza, 1997. "An Open Geographic Modelling Environment." Simulation Journal, v. 68, no. 3, pp. 175-185
• Costanza, R., A. Voinov, R. Boumans, T. Maxwell, F. Villa, H. Voinov and Wainger, L. Integrated ecological economic modeling of the Patuxent river watershed, Maryland. In revision for Ecosystems.
• Villa, F. and R. Costanza. 1999. Design of multi-paradigm integrating modeling tools for ecological research. Environmental Modeling and Software (in press)
• Costanza, R. and A. Voinov. 1999. Integrated ecological economic regional modeling: linking consensus building and analysis for synthesis and adaptive management. Chapter in: J. Hobbie (ed.) Estuarine science: a synthetic aproach to research and practice. Island Press (in press)
• Costanza, R. and A. Voinov (eds). Spatially Explicit Landscape Simulation Modeling. Springer, New York (in preparation)
• Villa, R. R. Boumans, and R. Costanza. Design and use of a model performance index (MPI) for the calibration of ecological simulation models. Environmental Modeling and Software (in revision)

Relevant Websites