Title: Whole watershed health and restoration: Applying the Patuxent and Gwynns Falls landscape models to designing a sustainable balance between humans and the rest of nature.

Investigators: Robert Costanza (P.I.), Roelof Boumans, Thomas Maxwell, Ferdinando Villa, Alexey Voinov, Helena Voinov, Josh Farley

Institution: University of Maryland, Institute for Ecological Economics, Box 38, Solomons, MD 20688-0038, Phone: (410) 326-7263 Fax: (410) 326-7354 costza@cbl.umces.edu

Project Period: March 1, 1999 - February 28, 2002

Research Category: Water and Watersheds

Background

The Chesapeake Bay watershed has been a model of watershed-based ecosystem restoration (Costanza and Greer 1995). Based on a long history of science, modeling and broad stakeholder participation, various commitments have been made to restore the health of the Chesapeake Bay. In December of 1983, the first Chesapeake Bay Agreement was signed by Maryland, Virginia, Pennsylvania, Washington, D.C. and the EPA. The states acted quickly to begin the restoration process. Maryland, for example, in 1984 approved 34 legislative initiatives, focused into 7 state programs known as the Maryland Chesapeake Bay Program. In 1985 the State of Maryland released its Restoration Plan, aimed at "restoring and maintaining the Bay's ecological integrity, productivity, and beneficial uses and to protect public health." By 1987 a new Chesapeake Bay Agreement was put forward calling for a 40-percent reduction of nitrogen and phosphorus inputs to the Bay, to be achieved by the year 2000. In 1988 the "Year 2020" panel released its visions for the future, detailing the environmental impacts of increased population growth within the watershed and making recommendations to avert further deterioration of the Bay's water quality.

While point sources of nutrients, like industries and sewage treatment plants, were relatively easy to control, it soon became evident that spatially dispersed, nonpoint sources, like residences and agriculture, were responsible for a large part of the problem and were much more difficult to control. The growing population of the watershed itself came to be recognized as a primary cause of the Bay's problems. The Bay is now primed for new approaches to management that can go beyond the traditional command and control methods (which were relatively successful with point sources) to complete the implementation phase on nonpoint sources. Meeting this challenge requires new methods of analyzing, modeling, and managing whole watersheds.

The State of Maryland now has significant experience with watershed-based approaches to water quality protection and restoration. This has come about largely as a part of the multistate federal Chesapeake Bay Program and because the water quality of downstream tidal and estuarine waters has been recognized to be heavily influenced by upstream sources, particularly nonpoint sources. In addition, the nature of the Maryland portion of the Bay watershed lends itself to delineation of discrete tributary watersheds which include tidal rivers. A major focus of the program to restore the water quality of the Bay involves a "tributary strategy" in which the sources of pollutants are estimated for each tributary watershed, fluxes are modeled, loadings are related to ecological conditions and living resources in the receiving subestuary, and goals are set for reduction of contaminants by generating sector (e.g. sewage treatment plants, agriculture, and dispersed residential) and location in the watershed. Thus the focus came to be on watersheds and individual tributaries to the Bay. The Patuxent is one of the most important of these tributaries (Figure 1) and a wealth of data has been collected within and near the river, as described below. The Gwynns Falls watershed in Baltimore (Figure 1) is a highly urbanized watershed that has become the focus of the new NSF funded Baltimore LTER project. It will be the subject of intensive and continuing data collection over at least the next 6 years.

To support and extend these efforts, we have been developing the capability to model watersheds as spatially explicit, integrated, ecological economic systems (http://kabir.cbl.umces.edu/PLM). The proposed research will use these modeling tools to address the issues of whole watershed health assessment, nutrient reduction, and restoration.

The Patuxent Landscape Model (PLM) is designed to serve as a tool in a systematic analysis of the interactions among physical, biological and socioeconomic dynamics of the watershed. (For a complete description of the model and its current status, see our web site: http://kabir.cbl.umces.edu/PLM). In the ecological component of this spatially explicit model, the important processes that shape plant communities are simulated within the varying habitats distributed throughout the landscape. The principal dynamics within the model are: plant growth in response to available sunlight, temperature, nutrients, and water; flow of water plus dissolved nutrients in three dimensions; and succession in the plant community in response to the historical environment. Using a mass balance approach to incorporate process-based data of a reasonably high resolution within the entire watershed, changing spatial patterns and processes can be analyzed within the context of altered management strategies such as the use of Best Management Practices (BMPs). By incorporating high spatial, temporal, and complexity resolution, the model can realistically address large scale management issues within the heterogeneous system of the Patuxent watershed.

For the PLM, the modeled landscape is partitioned into a spatial grid of nearly 2,500 square unit cells (Costanza, DeBellevue, et al., 1993). The model is hierarchical in structure, incorporating an ecosystem-level "unit" model that is replicated in each of the unit cells representing the landscape (Figure 2). The unit model (Fitz et al., 1996) itself is divided into a set of model sectors that simulate the important ecological dynamics at a daily time step.

The model includes sectors for hydrology, nutrient movement and cycling, terrestrial and estuarine primary productivity, and aggregated consumer dynamics. The hydrology sector of the unit model is a fundamental driving force, simulating water flow vertically within the cell. Phosphorus and nitrogen are cycled through plant uptake and organic matter decomposition, with the latter simulated in another sector that describes the sediment/soil dynamics. The sector for macrophytes includes processes such as growth response to various environmental constraints (including water and nutrient availability), changes in leaf canopy structure (influencing water transpiration), mortality, and other basic plant dynamics. As may be evident from the above linkages, feedbacks among the biological, chemical and physical model components are important structural attributes of the model. While the unit model simulates ecological processes within a unit cell, horizontal fluxes across the landscape occur within the domain of the broader spatial implementation of the unit model to form the PLM. Such fluxes are driven by cell-cell head differences of surface water and of ground water in saturated storage. Within this spatial context, the water fluxes between cells carry dissolved and suspended materials, determining water quality in the landscape.

The same generic unit model structure is run in each cell and there is a database of parameters that serves as input to the model to represent the different habitat types within the landscape. The vegetation communities in the cells respond to changing hydrologic and nutrient regimes via successional algorithms. Thus, when run within the spatial framework of the overall PLM, the landscape evolves to reflect changing hydrology, water quality, and material flows between adjacent cells.

The ecological model is linked to an economic model which predicts a spatial distribution of the probability of land use change within the seven counties of the Patuxent watershed (Bockstael, 1996). Human decisions to develop land are modeled as a function of both economic and ecological spatial variables. The land value (derived from tax assessment data) is used as the dependent variable in the first stage regression model and spatial variation in land prices is explained by an extensive array of features of the location: distance to employment centers, access to public infrastructure (roads, recreational facilities, shopping centers, sewer and water services), and proximity to desirable (e.g. waterfront) and undesirable (e.g. waste dumps) land uses to name a few. Also included are explanatory variables based on spatial pattern of land use that describe the land uses surrounding a parcel. A second stage model predicts the probabilities of land conversion based on the land values in residential land use generated by the first stage model and the costs of conversion. The model generates the relative likelihood of conversion of cells and when used in combination with information about growth pressures, allows maps of predicted new residential development to be developed.

The linked model allows the effects of both direct land use change through human actions and indirect effects through ecological change to be evaluated. Preliminary models of land prices and land use conversion explain the factors that have the most effect on land values in different uses and therefore the factors that affect pressures for land conversion. The physical location of the parcel as well as the spatial pattern of regulations affect the value of a parcel in different uses and therefore the likelihood that a parcel will be developed or kept in a natural state or used in agriculture. The following factors have been investigated for affects on land use pattern: Transportation network (Bell 1997); Public utilities provision (Bell and Bockstael 1997); Competing county zoning and agricultural preservation (Bockstael and Bell 1997).

The unit ecological model has undergone rigorous testing using newly-developed calibration software, which calculates a comprehensive Model Performance Index (MPI). The index integrates an array of variable-specific tests into a single score which expresses the overall fit with data and hypotheses (Figure 2). Each test considers a different aspect of the model’s output, e.g. fit to data, known patterns of autocorrelation, meaningful boundary values, or steady states. For example, calibration data used in the analysis included 10-day maxima of the Normalized Difference Vegetation Index (NDVI), which were supplemented with data on stand characteristics from the Forest Inventory Analysis (FIA) database. The MPI structure is inspired by multi-criteria decision analysis as well as statistical estimation theory and is defined as a weighted average of variables’ partial scores, each weighted according to the importance of the variable for the model’s goals or to the quality of the reference data. MPI values range from 0 to 1, with 1 indicating maximum agreement between the model output and calibration criteria.

An adaptive directed search algorithm was developed to automatically search the parameter space for points yielding the highest MPI values. The combination of the automated search cycles and the formal analysis of the results has allowed us to identify satisfactory parameter combinations which would have otherwise been impossible to find. The unit model was calibrated for two years at a daily timestep and MPI values have reached 0.53 in experiments to date (Figure 3). Documentation on the MPI is available on the Internet at the URL http://kabir.cbl.umces.edu/~villa/svp/svp.html.

Unit model calibrations were carried out to test the general model performance for typical forest and agricultural land uses of the Patuxent watershed. Within the macrophyte sector, calculated forest and crop biomass remained within acceptable bounds established from literature, field data, and EPIC model output. The model captures important seasonal dynamics in plant growth. Data from the Patuxent watershed sites provided boundary conditions for biomass, but data on seasonal dynamics to track forest nutrients and primary productivity were not readily available. Data from Coweeta LTER proved useful for comparing general dynamics. Data being developed from satellite imagery (e.g. NDVI) have been used both for calibrating NPP and standing crop biomass in the unit model and the spatial model.

Calibration of the spatial hydrology model at several spatial and complexity scales has improved model robustness and overall performance. The hydrology portion of the landscape model, which serves as the major vector for movement across the landscape, has been calibrated at several spatial extents and the full ecosystem model has been implemented and tested for general conformity to expected variable ranges for the entire Patuxent subwatershed. The hydrology model shows good agreement with measured streamflow data for an initial 2-year testing period in 1980-82. Fig. 4. shows a portion of this comparison. Several nested subwatersheds were used to test model behavior at a range of spatial extents (from 58,905 to 566 cells and resolutions (200x200 m and 1x1 km). The model performs well in describing overall surface and ground water flow at all spatial extents with model predictions generally falling within 10% of daily values, although some large flood peaks deviate to a larger extent. We have run several scenario analyses to investigate the effects of land use changes and other perturbations on various hydrologic variables (Fig.5). (For further details of the model structure and calibration, see: http://kabir.cbl.umces.edu/PLM).

The full ecosystem model is running spatially and displays expected orders of magnitude in ecosystem stock variables, and appropriate seasonal dynamics in plant growth and nutrient cycling. Spatial calibration data include: annual increment to forest biomass (using species-specific tree ring records and spatial distributions in the Patuxent watershed), seasonal and longitudinal dynamic records for phosphorus and nitrogen concentrations in the river, and 10-day maximum NDVI data at the 1 km² spatial resolution derived from AVHRR satellite images.

The Gwynns Falls Landscape Model (GFLM) will be an application and extension of the PLM to a largely urban watershed in Baltimore, as part of the Baltimore LTER project (http://baltimore.umbc.edu/lter/) This will involve both the collection of the relevant data base for the new watershed, and expansion of the "human" component of the model. During year 1 of the LTER project (1998), we have been working with other project participants to apply the basic framework of the PLM model to the Gwynns Falls watershed. This includes discussions about basic data requirements and sampling strategies, and assembly of the relevant existing data bases (see the PLM web site for a more complete description of the model and its data requirements: http://kabir.cbl.umces.edu/PLM/). By the end of year one (1998) we should have assembled most of the data sets necessary to run the PLM model for the Gwynns Falls site. We will also add a human dynamics component to the GEM unit model, including human populations, built infrastructure, and institutions. This will allow the GEM model to represent the full range of habitats, from "natural" ecosystems with little human influence, to "agroecosystems" with intermediate levels of human influence, to "urban ecosystems" with high levels of human influence. By the end of 1998, we expect to have a running model with preliminary calibrations to several sites within the study area.

The models we are developing are part of the set of tools we will use to address appropriate goals in ecosystem management. A basic question in ecosystem restoration is: "restoration to what state?" or "what do we mean by a healthy ecosystem?" The default endpoint has been restoration to a past state in which there was presumably little or no human influence on the ecosystem. For example, the National Research Council’s (1992) definition of restoration as "returning a system to a close approximation of its condition prior to disturbance, with both the structure and function of the system recreated" implies that the state "before disturbance" is the preferred state. This default definition of ecosystem health has proven to be both unrealistic and unworkable (Costanza et al. 1992, Rapport et al. 1998).

Humans have been important components of ecosystems for millennia, and they (like any large and abundant omnivore) have always radically altered the systems of which they have been components (Flannery 1994). For example, the original Australian aborigines caused the extinction of many species of megaherbivores and replaced (in many areas) what was originally a high diversity closed woodland ecosystem which did not burn and where most nutrient recycling was through herbivores, with a lower diversity open woodland ecosystem which recycled nutrients through almost annual fires, which were set and controlled by the aboriginal humans (Flannery 1994). What is the "natural" or "pre-disturbance" system to serve as the restoration endpoint in this case? The pre-aboriginal closed woodland or the post-aboriginal open, fire-adapted woodland which existed for 10,000 years, or some other state? This question is not answerable from a purely "objective" point of view, and must also include consideration of social goals (Costanza et al 1992).

While it is unrealistic to expect developed watersheds to provide as broad a range of support for ecosystem services compared to less developed watersheds, a range of services may be supported that is desirable to protect and maintain. Because more people will come in contact with urban or suburban ecosystems their value to society may be higher than their level of function might first suggest . Societal goals for ecosystem management have come to focus on the concepts of health and sustainability (Lubchenco et al 1991). How do we harvest from, and otherwise utilize ecosystems, while maintaining their health and integrity and the array of non-use services that they also provide (Costanza et al. 1997) into the indefinite future?

Social goals for sustainable ecosystem management are thus centered on maintaining the "ecological health" of the system. The International Society for Ecosystem Health (http://www.uoguelph.ca/~rmoll/) describes ecosystem health as:

While the concept of health applied to the level of ecosystems and landscapes is of relatively recent origin, it has become a guiding framework in many areas, particularly in the evaluation of the large-marine ecosystems (Sherman, 1995), forest ecosystems (Kolb et al, 1994), agroecosystems (Gallopin, 1995), desert ecosystems (Whitford, 1995) and others (Rapport et al. 1995, 1998).

In this project we will address both the sociological and ecological aspects of ecosystem health definition and restoration. We begin by acknowledging that humans are a major component organism in many (if not most) ecosystems today. The human part of the ecosystem includes the humans themselves, their artifacts and manufactured goods (economies), and their institutions and cultures. It is this larger ecosystem (including humans) whose health we need to assess and restore. Based on a survey of health concepts in many fields, Costanza (1992) developed the following three general categories of performance that are usually associated with "well-functioning" in any living system at any scale:

1. The vigor of a system is simply a measure of its activity, metabolism or primary productivity. Examples include gross primary productivity in ecological systems, and gross national product in economic systems.
2. The organization of a system refers to the number and diversity of interactions between the components of the system. Measures of organization are affected by the diversity of species, and also by the number of pathways and patterns of material and information exchange between each component.
3. The resilience of a system refers to its ability to maintain its structure and pattern of behavior in the presence of stress (Holling 1986). A healthy system is one that possess adequate resilience to survive various small scale perturbations. The concept of system resilience has two main components: 1. the length of time it takes a system to recover from stress; and 2. the magnitude of stress from which the system can recover, or the system’s specific thresholds for absorbing various stresses.

The project is structured around the following general and specific questions and hypotheses:

1. How does one most usefully define the "health" of ecosystems, especially at the watershed scale?

   We are developing methods for assessing various aspects of a system’s health and performance for the urban, agricultural, and "natural" components of landscapes, and for the landscapes as a whole (Mageau et al. 1995, Wainger 1998). We plan to include these indicators (along with a host of other indicators which our models can provide) in workshop and ongoing web-based dialogues with stakeholders to assess their usefulness for defining ecosystem health and restoration. A major question is what indicators or combinations of indicators are most useful (both from a socio-economic and an ecological point of view) in defining "well-functioning" in these complex, integrated systems. While we do not believe that this question has a simple or a completely "objective" answer, we hypothesize that a useful consensus can be achieved by combining scientific input (via the landscape models we have developed) with stakeholder preferences in ongoing dialogues. These dialogues will be both face-to-face in a series of workshops and via the web using interactive web sites.

2. What are the impacts of specific spatial patterns of human settlements (urban, suburban and agriculture) on the ecological health of watersheds?

   The spatial patterns of human settlements have been implicated as one of the primary causes of deteriorating ecosystem health. The Maryland Critical Area law, "smart growth" initiatives and various other local zoning ordinances have been enacted to prevent "sprawl" and preserve more contiguous open space. But little is known about the quantitative influence of specific human settlement patterns on the health of ecosystems. The PLM and GFLM have a unique ability to test these impacts.

   The research suggests that when appropriate scales and land use types are considered, landscape pattern metrics can be used to indicate habitat conditions for a variety of species. Using relationships established in the literature and process and empirical models we have developed, we can infer effects from fragmentation on a range of ecosystem processes from nutrient transport to terrestrial and aquatic species. Spatial pattern influences on hydrologic and nutrient processes can be tested through scenarios of land pattern change. Scenarios will be developed by imposing particular structures on the landscape (e.g. extensive riparian buffers vs. minimal riparian buffers) and by creating landscapes to fit spatial characteristics. Land uses may be assigned to the landscape based on random or fractal algorithms which control patch size, fragmentation and land use proportions .

3. What are the impacts of various management and policy options concerning human settlements on the spatial patterns and ecological health of watersheds?

   Various management and policy options have been proposed in order to reduce nutrient inputs and restore the health of watersheds and estuaries. Suggested policies include:

   • Tradeable/transferable development rights
   • Flexible impact fees on new development
   • Taxes/subsidies to encourage wetland, forest, or agricultural preservation and/or adoption of various agricultural and urban "best management practices" (BMPs)

   But little is known about the quantitative influence of specific management and policy options on the evolution of landscapes and the health of ecosystems. The existing economic model can provide information about human responses to policy change since the variables used to predict the values in residential and alternative uses are functions of ecological features, human infrastructure and land use policies. The PLM and GFLM have a unique ability to test these impacts through the links that have been developed and will be further developed to incorporate human decisions rather than imposing expected change external to the model process.

4. How does one restore watersheds by redesigning human settlements and agricultural operations to optimize the health of the overall system?

   The PLM and GFLM can also be used in "design mode." In this mode, a desired future state is envisioned, and the models are used to determine the combination of spatial patterns of human settlement and management options that are necessary to achieve a given state (or if the state is achievable at all). The PLM and GFLM can be used in design mode to help achieve a more "optimal" pattern of human settlement, for example, by maximizing nutrient reductions from specific restoration efforts. It has been shown that in some complex, spatially explicit landscape models, policies are not simply additive (Costanza et al. 1990). For example, policy A may have a positive effect on its own, and policy B may also have a positive effect on its own, but when policy A and B are both implemented simultaneously, it is possible for a negative effect to result. This occurs because of the complexity of interactions in the system. Thus the process of determining which combination of policies will yield the desired result may not simply be one of adding together all positive policies. The landscape model becomes an important tool in testing this question and assessing which combination of policies have the desired net effects.

   Determination of the desired future state itself also requires input from a broad range of stakeholders in workshops designed for this purpose. These will be an integral part of the project. In addition, web-based interaction will broaden and lengthen the possible participation .

The main focus of future model development will be towards improving the representation of humans and their decisions in the landscape and understanding the full range of options between the end points of catastrophic human-induced system degradation and the system’s "natural" state (i.e. with no human presence). The aspects of the model covering agricultural crop growth will be expanded to represent farmer’s economic decisions. The urban habitats will be expanded to include built infrastructure, human population and their decisions and institutions, and the flows of goods and services. In addition, a full range of scenarios will be run using the models, as discussed above. In the process of running these scenarios, we expect to be surprised by the results and to discover problems and missing aspects in the models. This will guide adaptive further development and continuous improvement of the models.

We have developed and begun preliminary testing of methods to assess the health of ecosystems. These methods do not depend on the simplifying assumption that prior states of ecosystems before human interactions are necessarily more healthy (although in many cases this will in fact be the case). Rather, we will quantify the general characteristics of all living systems at all scales which are generally associated with "well-functioning." These include: vigor, organization, and resilience (Costanza 1992, Mageau et al. 1995). At the single ecosystem scale, these characteristics can be applied to "natural" ecosystems, agroecosystems, and urban/suburban ecosystems with equal facility. At the landscape scale, these measures relate to the aggregate productivity of the landscape, its spatial pattern or organization (as measured with various spatial pattern indices , and its resilience (as measured by experimenting with the dynamic landscape models.

The level of change in ecosystem variables which significantly alter function, can be addressed by using the ecosystem simulation model to both develop and calculate ecological indicators. Simulation models offer the ability to detect trends in time and space and to run scenarios into the future to partially assess long-term effects. The calibrated model will be used to examine the relationship between system sinks and flows and long term system behavior in order to suggest promising indicators based on our modeling of ecological interactions.

The landscape model’s dynamic, spatially distributed nature allows us to develop sophisticated indicators of ecosystem health which would be impossible without such a tool. For example, alterations to hydrology are considered important system indicators because hydrologic processes serve to structure ecosystems and therefore may have multiple indirect effects on the system. Bedford and Preston characterized hydrologic alteration as an "action most likely to produce cumulative effects" and therefore especially important to consider when judging impacts on wetlands and landscapes in general. Using the hydrologic components of the landscape model, indicators can be developed to assess impacts to streamflow from land development, considering both direct human needs and stream biota. For example, an adaptation of the hydrologic model, which had previously been calibrated to an urban-suburban coastal plain watershed (Wainger et al., 1996), was used to evaluate hydrologic indicators of land use change. Change in stream baseflow and in peak storm magnitudes were used to evaluate hydrologic impacts from development in hydrologically important zones of the watershed. Sensitivity to land use proportions and pattern decreased sharply above 60% high density residential and commercial uses. Stream buffers in particular lost the ability to mitigate storm peak flows although they were seen to mitigate stream peak flows at low to moderate levels of development (Wainger 1997). High fragmentation of land use was found to increase stream baseflow but did not mitigate peak flow. The hydrologic indicators and methods that were developed will be used to test effects of development in the full landscape model and new indicators will be developed for comparing effects to ecosystems.

Other methods to develop indicators will use model predictions in combination with spatial pattern indices (descriptive statistics used to quantify landscape pattern) to reflect the ability of the landscape to support certain ecosystem functions. A large literature has recently been developed to link landscape fragmentation to plant and animal species diversity and persistence . Other work has shown a relationship between land cover characteristics and water quality and in-stream habitat . The relationship between fragmentation and diversity has been shown to be a complex relationship which may work to both increase and decrease diversity under different conditions. Effects will be scale dependent since the degree of fragmentation experienced by a particular organism will depend on organism mobility . And recovery or resilience can be represented by considering source population distance and natural corridors (represented by various indices) which have been shown to influence recovery rates of both plants and animals following catastrophic events (Detenbeck et al. 1992; Gustafson and Gardner 1996; Hawkins et al. 1988).

Empirical models can link the dynamic spatial processes used in the simulation model to higher trophic functions and integrative indicators. Fish and invertebrate indices are being increasingly used to assess environmental impacts because the biotic community structure responds to both periodic and chronic conditions which can be missed by intermittent chemical sampling. Diversity and other characteristics of the community are thought to reflect the quality of habitat and the ability of the community to respond to perturbation. Combinations of community metrics offer the ability to assess several aspects of fish or macroinvertebrate habitat and allow comparisons across regions and through time. The Maryland Department of Natural Resources (MDNR) has been collecting data throughout the Patuxent watershed to calculate an Index of Biotic Integrity (IBI) for fish community assessment which has been adapted to coastal plain streams . The Maryland IBI, like Karr’s, uses 12 metrics that assess species richness and composition, trophic composition, and fish abundance and condition, and scores each metric relative to an undisturbed watershed with otherwise similar characteristics . Metrics provide information about a range of structural and organizational aspects of the stream ecosystem which are related to physical and chemical habitat quality. The IBI has been defined as a measure of the "ability to support and maintain a balanced, integrated, adaptive community of organisms having a species composition, diversity and functional organization comparable to that of natural habitat in the region" . Because the IBI reflects a multitude of landscape processes which affect water quality and structure habitat, it has been asserted that the IBI represents a range of ecosystem services .

The challenge of selecting appropriate system indicators to judge impacts is simplified by considering the choice in the context of the ecosystem’s potential to provide marketed and life-support services (Costanza et al. 1997). Multicriteria decision analysis offers a method for the users of an ecosystem to have input into the decision-making process and express values and choices. Implementation of MCDA will focus on selecting appropriate indicators given the dominant uses of that system and evaluating indicators in terms of the degree of change that is acceptable to its users.

The steps of a multicriteria decision analysis (MCDA) involve the same steps as any decision, identifying a problem, defining alternative actions, defining criteria upon which to judge those actions and analyzing the alternatives for the best solution. MCDA supports the decision-making process by providing a framework that can compare the relative impacts of alternative actions using a wide variety of criteria. Criteria can be measured in a combination of qualitative and quantitative units, including (but not limited to) monetary cost benefit analysis. All decisions ultimately require, however, that changes be judged according to their relative impacts.

The advantage of MCDA stems primarily from its ability to incorporate a range of criteria which are selected and weighted by a group of stakeholders. Various points of view and levels of uncertainty may be incorporated into the analysis to account for different opinions. Alternative actions are considered from the different points of view and actions may become clear when an option scores highly from all or many points of view. However, much of the time an optimal solution can not be derived with MCDA since any action will be better for some criteria but worse for others. As Munda (1995, p.58) describes, "The main advantage of these models is that they make it possible to consider a large number of data, relations and objectives (often in conflict) which are generally present in a specific real-world decision problem, so that the decision problem at hand can be studied from multiple angles."

The technique has been applied to a variety of land development and industrial siteing problems and in groundwater management , but has only recently been used with explicit spatial information in GIS form, for example for park planning and sustainable development of tourism . However, resource planners are increasingly modeling both ecological and economic concerns with spatially-explicit models that attempt to evaluate solutions using a variety of means . The spatial databases and models we have developed for the Patuxent and Gwynns Falls watersheds will be employed to provide spatial input to the MCDA process.

We will employ a workshop process that seeks to develop and extend "common ground" rather that accentuating differences and disagreements (Weisbord 1992, Weisbord and Janoff 1995). Face-to-face workshops will be small (<40 participants), 3-day events aimed at involving a range of stakeholders in the process. They will be focused on building consensus about (1) what constitutes ecosystem health; (2) what are the criteria and their relative weights in the MCDA; and (3) what the preferred state of the overall watershed should be. The workshops will have a minimum of presentation time and a maximum of discussion about what constitutes the common ground. A mediator will maintain focus on common ground and will not allow undue divergence into arguments about differences. The workshops will be supplemented by the web-based system discussed below.

We plan to fully exploit the tools available through the World Wide Web to complement our efforts in watershed analysis and watershed management at the regional scale (Voinov and Costanza, submitted). The Patuxent watershed simulation model will serve as a core for watershed management design based on web applications. The model and its data will be made available over the web so that the environmental management community will have the opportunity to use and provide feedback on the model. We further plan to use the Internet to deliver scientific findings and information to stakeholders and to link stakeholders together to provide a forum for collective decision making as an extension of the planned consensus-building workshops.

The distributed nature of the web and it’s ability to display information creatively allows a unique method to provide broad access to a wide range of data and analysis tools and to gather feedback from a broad range of stakeholders. The core of the modeling code is already able to run on a variety of computer platforms and work has begun to translate the interface into Java to allow platform-independent access to the model interface. Animation and display tools are already allowing us to provide access to model results. Further development could provide the ability for managers or students to create scenarios and run the model from a web interface.

To expand the usefulness of dynamic modeling results, model output will either be used directly or in combination with empirical relationships to create indicators with which to judge ecosystem function and overall health, and to evaluate restoration strategies. The definition of ecosystem health and thus the exact set of indicators to use will be the subject of a multi-stakeholder workshop with input from both the scientific community and the local stakeholders. The PLM and GFLM can produce a broad range of indicators for any point throughout the landscape and for several points in the past, present, and future to support this process. A few examples include:

Based on additional empirical models and relationships we are developing, the model can also be used to infer a range of aggregate health indicators, including:

Additionally, model sensitivity analyses will be used to evaluate indicators and to reveal which processes are particularly important to maintaining functioning and the system’s overall resilience. The model allows a comprehensive estimate of conditions through time and allows us to consider a broad range of spatial scales in indicator development. By combining the dynamics of water, nutrients, vegetation, and humans with analyses typically done using non-dynamic GIS techniques, we can more completely characterize the "‘functional combination’ of habitats" for plants described by Noss (1989) and can expand the analyses to the supported fauna in locally distinct environments.

We will emphasize techniques to evaluate conditions across a spectrum of levels of function by considering which processes are relevant in each type of habitat or land cover. By comparing current variability to past variability we can better define variation outside inherent system "norms" and recognize "unhealthy" states of ecosystems.

Preliminary work to select and combine model indicators in a multi-criteria decision analysis (Wainger 1997) will be expanded to incorporate the full ecosystem model and use state of the art methods to create the evaluation matrices (Villa 1996). Indicators used in the decision analysis will be chosen to be relevant to the decision-making process and representative of a variety of societal goals revealed in the workshop process. Indicators will be used to judge relative impacts on system function under a series of development scenarios. The result will be a flexible tool for comparing and communicating effects on ecosystems and economic systems and displaying impacts spatially when suitable. These results will be made available to a wide range of stakeholders via the web.

New economic indicators will also need to be developed to represent human "habitat" concerns such as quality of life issues under different scenarios. Since land is traded in markets, the effect of changing ecological conditions on the private values of the land in different uses can be approximated from our existing economic model. However, many other ecosystem services may be valued but may not be appropriated by the private owners of land and thus will not show up in market values. These services or public goods include the indirect ways that ecosystems support: recreation and aesthetics, hydrologic function, nutrient cycling, wildlife habitat, human health, and others (Costanza et al. 1997). Assessing the effects of human actions on these public goods requires that we be able to link the state of the ecosystem with the existence or quality of these public goods. For instance, can we quantify the effect of particular human habitat patterns with nitrate levels in surface water and fish populations? Or more generally, can we assess the loss of ecosystem resiliency or likelihood of systemic collapse with appropriate ecosystem health indicators? We plan initially to incorporate these feedback effects from the ecosystem to these life support functions in a qualitative manner. Other work will develop quantitative links between ecosystem characteristics and ecological indicators being tracked through monitoring work such as the species diversity of the forest ecosystem.

Using the models in "design mode" will allow us to investigate how best to achieve the preferred future state defined via the envisioning workshop. For example, if the future preferred state includes certain minimum water quality conditions, the models can be used to understand how the interactions of hydrology and land use contribute to water quality.

New techniques will be developed to help determine how restoration should be prioritized considering local and regional goals and conditions. Our work will consider such issues as the complex interplay of spatial and temporal influences which mitigate or exacerbate human activities. For example, we can characterize the degree to which the local hydrologic regime controls the degree to which local land cover patterns influence water quality.

The models will also allow us to assess the costs of restoration and techniques to maximize the benefits per restoration dollar within the regional landscape context. For example, for any specific combination of spatial patterns and policies, the models can assess the likely effect on the overall health of the system (as developed in the other project components). We can also assess the relative costs of policies and display all of this information via the web to help stakeholders decide on the preferred restoration strategy.

The project tasks and progress chart are shown diagramatically in figure 6. The five main project components are: (1) Continuing development and applying the landscape models; (2) Ecosystem health indicator development; (3) Multi-criteria decision analysis; (4) workshops to allow broad stakeholder involvement; (5) web-based watershed management system development.

Year One: We will focus on (1) applying the PLM to the Gwynns Falls watershed to create a parallel GWLM; (2) adding urban sectors to the PLM and GFLM; and (3) analysis of spatial pattern and policy scenarios using the PLM and GFWM.

Products: Results will include a range of model-based scenarios of the impact of the range of pattern and policy options discussed above on the future status of the watersheds.

Year Two: We will continue model development, hold the first workshop on ecosystem health definition, and continue development of the web-based management system. We will use the workshop results to develop health indicators in the landscape models. We will compare results from the PLM and the GFLM.

Products: Broad consensus on ecosystem health definitions that represent an integration of science and stakeholder perspectives will be a major product of year two. We will also have implemented the web-based watershed management system to broaden the base of participation, and to serve as a model for other watersheds. Using the PLM and GFLM, we will produce a set of health indicators for each watershed, which will also serve as input to the 2nd workshop on envisioning a preferred future. The intercomparison of the results for the Patuxent and Gwynns Falls watersheds should also prove to be a very interesting product.

Year Three: We will hold the 2nd workshop on envisioning a preferred future watershed and conduct the MCDA using input from the second workshop. We will use the MCDA, the PLM and the GFLM to design strategies to achieve the preferred watershed state. We will complete the implementation of the web-based management system. We will hold the 3rd workshop to present final results.

Products: The 2nd workshop will put the pieces together and build consensus on the preferred state of the watersheds. The landscape models will allow us to determine the necessary steps to achieve any particular restoration state. Product will also include reports and publications on all aspects of the project.

Several ongoing efforts at the University of Maryland contribute to this program, including:

1. Human Settlements as Ecosystems: Metropolitan Baltimore, Maryland, 1797 to 2100. A Long-Term Ecological Research (LTER) study funded by NSF (S. Pickett, lead PI, R. Costanza, co-PI and R. Boumans collaborator) will study the spatial structure and temporal changes of socio-economic, ecological, and physical factors in the urban Baltimore area.

2. An Open Spatial Modeling Environment. A long term (minimum 5-yr, 1998-2004) project (T. Maxwell and R. Costanza, PIs) as part of the NSF National Computational Science Alliance (NCSA) at the U. of Illinois, to develop the Spatial Modeling Environment (SME), a tool for modular, collaborative modeling and distributed computing.

3. Integrated Ecological Economic Modeling and Valuation of Watersheds. A 3-year (1995-1998) NSF/EPA-funded study (R. Costanza, PI) aimed at extending the PLM model to improve our understanding of regional systems, assess potential future impacts of various land uses, development, and agricultural policy options, and better assess the value of ecological systems.

4. Multiscale Experimental Ecosystem Research Center (MEERC). A 10-year (1992-2001), EPA-funded Exploratory Environmental Research Center devoted to investigating the issues of scale and scaling.