Effective management of our natural resources requires an understanding of the direct and indirect effects of human activities and how they interplay with natural variability over long time periods and over large areas. We are developing a linked ecological-economic model based on the idea that modeling natural processes, human decision-making and the feedbacks between the ecology and economy will increase our understanding of how ecosystem processes support economic activities and thus are linked to human values. By modeling the important processes affecting plant growth, nutrient cycling, pollution delivery to waterways and other physical, chemical and biotic processes, we aim to be able to assess the health of the ecosystem, taking into account the effects of changes in land use or management. We expect to understand how perturbations affect the system, at least at a gross level, and we run the model over moderately long time frames (50 years) to analyze long-term effects of current resource use decisions. These models serve as repositories of current understanding of ecosystem processes, allowing us to synthesize field and regional scale information and point to gaps in our understanding.
The project has been progressing on numerous fronts. We continue to refine the ecological simulation model to enhance performance, increase usability and transportability, and enable the model to address resource management questions. Empirical models are being developed from field data which will link ecological model output to habitat quality. The economic model has progressed to the level of generating preliminary maps of the probability of land parcel conversion for input to the ecological model. And a framework in which to link output from the ecological and economic models has been developed which allows consideration of multiple management criteria in land use decisions.
Over the past year, we have created new tools, methods and databases needed for refining the ecological model. The hydrological module has been the main focus of structural efforts to improve spatial model performance since hydrologic fluxes serve as the main transport mechanism of the landscape processes. Previous applications of the model have been predominantly developed for and implemented in wetland regions (Costanza et al. 1990; Fitz et al. 1993), therefore the Patuxent implementation required substantial alterations to be able to handle upland systems. We have expanded the algorithms which flux surface and subsurface water, making them more general and appropriate for a range of landscape types and elevations. The hydrologic module, being part of a larger ecological economic simulation, required innovative approaches, that would be both simple enough to run sufficiently fast over the thousands of cells in the model and sufficiently detailed to handle several hydrologic variables that are necessary to generate meaningful output. Output is needed at a relatively fine scale in order to be useful in the ecological and economic model components.
Work to adapt the Patuxent unit model to specific ecological economic interactions has focused on better characterizing agricultural and urban activities on the landscape. We have added more specific input fluxes and developed better data bases for calibrating the unit model to developed areas. The nitrogen sector has received special attention because of its importance in ecosystem processes and because of the attention nitrogen movement has received in policy goals in the Patuxent watershed and the Chesapeake Bay. Nitrogen cycling is of particular importance in groundwater safety and estuarine productivity. Additional nitrogen flows have been added to address important upland and forested system dynamics which were not present in the initial wetland-dominated model. The addition of atmospheric fluxes of nitrogen provides an important link to economic activity in the watershed and beyond. New Patuxent data on the rate of marsh accumulation of nutrients and new soil parameter data is being incorporated into modeled landscape processes to address the ability of various land use types to provide the ecosystem service of nitrogen uptake and sequestration. Soil classifications were linked to an internationally-recognized classification system (Soil Survey Staff 1975) to provide additional measurements of soil characteristics and better transportability of data to other watersheds. Groundwater nitrogen data were prepared for calibrating nitrogen movements.
Estuarine habitat is being modeled with improved parameters and functions to address important dynamics in that portion of the landscape and better understand effects of processes in the upper watershed on the lower watershed. Data from Multiscale Experimental Ecosystem Research Center (MEERC) experiments (see proposal) and Chesapeake Bay Program monitoring data are being used to calculate the appropriate parameters as required by the macrophyte and consumer sectors. Dr. Roelof Boumans, a new addition to our group, brings a background in marsh and estuarine modeling to this effort.
Calibration of the unit model has moved from the "ballpark" level to the "slalom gate" level, at which, output data are maintained within expected ranges. An interface has been added to the unit model for calibration purposes within the STELLA software package, which provides a graphics window where model output is plotted against monitoring data. The interface allows parameters to be changed while running the model. Expected ranges are being set by evaluating various methods of describing and aggregating field scale data. A more extensive user interface is being developed for running and calibrating the spatial model on the UNIX platform (see below).
The land use-specific database, which contains the values used in model equations for each land cover/habitat, has been restructured to allow data of different scales to be used in the parameter development and calibration process. Land use or habitat classifications have been organized hierarchically, so that data available for a more specific land use or management can be considered when using more aggregated land types. For instance, data on specific crop species have been statistically linked to county-level agricultural data, before being used to create parameters for the more general "agriculture" habitat type used to parameterize model equations. This database structure is allowing us to test effects of cropping patterns, best management practices and forest or wetland species transitions, while still maintaining a lumped land use structure.
Our efforts to arrive at an optimal scale for the spatial model, were supported by the construction of a workable spatial data set in a GIS format to drive all spatial modeling tasks and which also will be used for spatial analysis based on GIS methods. We have been relying on the public domain GIS software package GRASS, which has been directly linked to the Spatial Modeling Environment (see below). Extensive data sets have been acquired and processed to make them available for modeling and other analytical purposes. These will be made available for other researchers doing work on the Patuxent watershed.
Hierarchical System of Spatial Models As part of our efforts to increase the ability of the model to be implemented in different watersheds, especially those with lower data availability, we are focusing efforts on devel-oping a hierarchical model approach. A hierarchical structure allows the use of various levels of complexity and resolution to address a range of goals. At one end of the hierarch-ical spectrum, we have models of low structural detail operating with spatially detailed data, a model type which can be incorporated into some GIS environments. At the other extreme, we have highly detailed models implemented using low resolution spatial data.
By comparing output from models using input data of various resolution, we are investigating the trade-offs in model structure complexity which are needed to compensate for coarser spatial or temporal resolution. We are currently working with a hydrologic submodel of moderate to high structural detail, using two spatial data resolutions (200x200m and 1x1 km cells) for two spatial areas (the entire 2400 km2 watershed and a 30 km2 subwatershed). The subwatershed is being used to facilitate model calibration and analysis of several alternative methods of water routing. The two spatial resolutions allow us to test methods of model structural aggregation to achieve similar model output. We intend to build a simplified version of the model that could operate at higher resolutions over large areas, thus matching the resolution of the economic component of the model.
The first and primary task of the economics subcomponent of the research is to provide a means of predicting human-induced changes in land use patterns. Considerable progress has been made on this modeling problem. In the seven counties of the Patuxent watershed, the major type of human-induced land use change is conversion of agricultural and forested land into residential use of varying densities. The modeling approach developed is one in which the decision to convert a parcel of land is affected by the value of the parcel in its current use, the predicted value of the parcel should it be converted to residential use, and the costs of conversion. This is represented by a discrete choice (logistic regression) model where observations include any cell in the landscape that could, given current zoning and other restrictions, be converted at a point in time. Model parameters are estimated using historical measures of the above factors and information on which parcels were actually converted over that historical time frame. The value of a parcel, should it be converted to residential use, is predicted with a hedonic model in which price is regressed on attributes. Those values are then input to the discrete choice model independent variables. Estimated parameters in the hedonic model explain the influence of a variety of factors on land prices, including economic and ecological features of the landscape.
The first stage of the modeling process is the acquisition and adaptation of various data necessary to statistically estimate this model of human-induced land use conversion on a disaggregated spatial scale for the seven counties of the Patuxent region. The types of spatially disaggregated data that have been acquired and organized include snapshots of land use/land cover over time, soil types and slopes, zoning, public sewer and water provision, demographic (Census) variables, roads, hydrology, and the location of locally desirable (e.g. parks) or undesirable (e.g. landfills, airports) land uses. A key data source in estimating the hedonic model is the state tax assessment database that includes information on all parcels in the region, including sales prices over the last two decades. Geocoded versions of this data base are now becoming available and are being incorporated into the economics database.
The acquisition and organization of the above data in their preliminary form allowed estimation of preliminary forms of the hedonic price function and the discrete choice land use conversion model. These preliminary estimations have been very successful and have provided support for the modeling approach. The value of land in residential use has been shown to be sensitive to the usual factors - such as distance to major employment centers, waterfront, access to roads - but also to such landscape features as the pattern of land use and land cover surrounding the parcel. Additionally, predicted value in residential use, alternative use, and conversion costs proxies have been shown to be significant in explaining land use conversion in preliminary models of this behavior.
While much data has been accumulated, important data needs remain. We have found that measures of landscape pattern affect the value of land. We are interested in finding other ways in which the ecological system feeds back to the economic system. We are currently collecting data on environmental quality variables that vary spatially to test for other types of ecological-economic feedbacks.
Because of its dependence on spatial data, this modeling approach introduces a number of interesting statistical problems only recently discussed in the economics literature. While methods for handling spatial autocorrelation in continuous models have been applied for a few years, the standard methods cannot be applied in problems such as ours that have several thousand observations. We have made progress in applying new generalized methods of moments estimators to the problem. Additional statistical problems of sample selection and spatial autocorrelation in discrete choice models remain to be solved.
Once estimated, the above model can be used for prediction. We have attempted only a few preliminary prediction scenarios - largely concerned with different patterns of public sewer service provision, but a whole host of regulatory policies could potentially be simulated using these models once they are perfected. A number of public policies are subsumed in this modeling approach, since public policies affect the value of land in residential use, in alternative uses, and the cost of conversion, as well as the pool of land that is available to be converted at any point in time. For example, to the extent that a public policy alters the value of land in an alternative use e.g. agriculture or forestry, the likelihood of conversion is affected and shows up in our land use conversion model. Transportation investments (i.e. new roads) will alter the pattern of residential land use value, as will changes in the provision of other public goods. Of course, land use management controls, in the form of zoning will alter the pattern and density of conversion in ways that are quite complex in that they alter both the pool of land that can legally be converted and the costs of making those conversions. The prediction process takes into account the fact that development can be diverted from one location to another.
Work to integrate the ecological and economic models has encouraged us to find innovative methods for incorporating human concerns and ecological effects. We have identified and added new variables to both the ecological and economic models which will increase the ability of a given model to provide the information needed by the model being linked to it. Within the land transition model, human perceptions must be matched to their underlying ecological cause or significance. The ecological model is being expanded to further link ecosystem processes to environmental services which people recognize as important.
Some technical issues regarding the running of a linked ecological economic model with differing time and space scales have been considered. Spatial resolutions for linked model runs were chosen so they would be compatible and the temporal scale differences will be handled by transferring data between models at the appropriate intervals. For example, the ecological model runs on a daily time step, while the economic model runs on a yearly time step. Data can be passed between the models at the beginning of each yearly time step. The data exchange protocols have been considered and various degrees of model integration have been discussed, although final determinations have not been made.
As the ecological and economic models are linked, we need to ensure that we adequately address the interactions between humans and the ecosystem by identifying quantifiable changes in economic or ecological processes. The level of changes in ecosystem variables which significantly alter function, can be addressed by using ecosystem simulation modeling to both develop and calculate ecological indicators. Simulation models offer the ability to detect trends in time and space and to run a scenario into the future to determine long-term effects. The calibrated model will allow us 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.
Since we are not currently modeling some of the higher-order consumers of the ecosystem (such as fish abundance or invertebrate community structure) we are limited in the indicators which can be calculated directly with model output. As an efficient method of linking the dynamic processes being modeling to other functions, we plan to develop empirical models from monitoring data. We are currently using a smaller coastal plain watershed as a test case of future work in the Patuxent. The Sawmill Creek watershed currently has a longer data set of several indicators of interest and a calibrated spatial model which is a sub-set of the full GEM model. Empirical models are being developed using Sawmill monitoring data and GIS information which model the Index of Biotic Integrity (Karr 1981) as a function of spatial model output, physical habitat descriptors and land use pattern indices.
The indicators will be important as we expand our analysis of economic effects of ecological change and the feedbacks between the ecological and economic systems. Indicators allow us to put a quality level on ecosystem function which is important in making the link between ecosystem function and service. If, for example we understand how nutrient cycles are disrupted by land use change, we can suggest through the spatial model how stream chemistry is likely to be affected and infer from the empirical models the impact on fish resources.
Ecological indicators have the desirable properties of providing information about ecosystem response to a range of stressors and of being applicable across a range of systems. However, indicators are primarily useful for assessing the degree of human influence without identifying when that influence makes the system vulnerable to perturbations, both natural and anthropogenic. Further evaluation of indicators is needed to determine their usefulness in making predictions of irreversible system collapse, but for the present, certain indicators offer important methods for integrating conditions over time. We have focused on using indicators which examine macroinvertebrate and fish community structure which can offer more valuable information than water quality monitoring data which may be too sparse in space and time to show system stresses. By using the simulation model to calculate indicators, we can evaluate stream chemistry fluctuations through time, and the timing of inputs relative to life cycles in order to more fully understand the environment experienced by stream organisms.
In addition to the indicators developed from monitoring data, we are also applying the information being developed from landscape ecology to our analyses. Landscape pattern analysis of land use data has shown promise in addressing how land use pattern may influence population abundance, diversity and resiliency. Work in the Patuxent region has correlated bird abundance and species diversity with land use pattern (Dawson 1996; Flather and Sauer 1996). Others have shown how source population distance and natural corridors can influence recovery of both plants and animals following a catastrophic event (Hawkins et al. 1988; Detenbeck et al. 1992; Gustafson and Gardner 1996). We are calculating several spatial pattern indices using coastal plain watersheds and creating empirical models which link animal population characteristics to spatial pattern. These empirical models will be applied to the Patuxent watershed.
Resource Use Decision Framework The test case implementation in Sawmill Creek is also being used to assess how output from both the ecological and economic models can be combined to address multiple aspects of resource use decisions. Ecological and economic indicators, calculated from the models, will have components sensitive to the spatial arrangement of land. We are in the process of testing the sensitivity of model output to spatial land arrangement in order to assess whether we can identify parcels on the landscape which differ in their relative contribution to ecological or economic functions. For example, selected land parcels will be converted to developed uses to test the relative influence of those parcels to changes in flooding potential, local animal habitat quality, and other environmental services.
A series of maps will be created ranking land areas by the sensitivity of indicator values to the conversion of those land parcels. The economic model will be used to show the magnitude of the effect of land conversions at a particular point on the adjacent land values. The ecological model will be used to generate a series of output maps ranked for their influence on a particular indicator, such as landscape fragmentation. The ranked maps will then be combined using a weighting scheme derived from management goals to produce a composite map which ranks the importance of map regions to multiple management criteria; in effect, producing a map showing land preservation importance given a set of management goals.
This method of combining output from the ecological and economic models provides one method for considering a range of criteria when making land use decisions such as zoning. This scheme allows us to combine different measures of ecological and economic activity without having to convert all variables into the same units. We can avoid some of the time step and spatial scale differences between the models and focus on the spatial analysis of each model. Yet, we still are able to incorporate both ecological and economic goals in setting land use priorities.
Considerable effort was directed into upgrading and testing the software used to translate and run spatial models, the Spatial Modeling Environment (SME). This integrated environment for high performance environmental modeling, links models with several types of databases and handles dynamics within and between model grid cells. The interface offers ease of defining parameters for the model, choosing specifications for running the model and viewing the results in the form of animations, graphs, or tables. The upgraded environment enables students, scientists, policy makers, and stakeholders to participate in the development of high performance spatial models, with minimal knowledge of computers or computer programming.
The current phase of research has focused primarily on developing a system that transparently links icon-based modeling environments (e.g. STELLA, Extend, Vensim) with advanced computing resources (e.g. Sun workstation networks), and provides a uniform interface for different groups of spatial modelers whether they are using parallel or distributed platforms. In this phase we have also created the prototype version of a Modular Modeling Language (MML) to enable the construction of models from reusable components. This paradigm encourages the development of libraries of modules representing model components that are globally available to model builders, enabling users to build an extensive, reusable modeling infrastructure. The adoption of this paradigm should greatly facilitate the application of computer modeling to the study of environmental systems in support of research, education, and policy making.
Gathering of tree ring data was delayed only slightly by the departure of one of the collaborators on this project, Dr. Curtis Bohlen. We will instead contract this work to an experienced tree core analyzer. We are currently developing a robust sampling design for the landscape to maximize the information we acquire for model calibration. This field program will be a reconnaissance study, we expect to take a maximum of 40 samples from trees at least 50 years old throughout the watershed. Tree core analyses will provide data on primary productivity under a variety of weather and water-stress conditions. We further hope to test the influence of adjacent human-dominated landscapes on tree productivity. Historical land use maps developed by the USGS will be used to determine the extent of human activities adjacent to selected sampling sites. Sampling is anticipated in November 1996. Data have been gathered from disparate sources taking full advantage of web sites which offer data for downloading such as the USGS elevation maps and stream gage data. In order to get the best and most up-to-date data, we are forging alliances with a variety of field researchers. The Patuxent Environmental Research Center is conducting extensive research on bird habitat in the Patuxent watershed. The USGS is conducting studies using AVHRR data to monitor plant productivity. And researchers at Battelle National Laboratory are interested in a link between their global climate models and our regional models.
Collaboration with Patuxent researchers was facilitated by a technical workshop we arranged to better coordinate modeling activities and share information with others working in the watershed. We discussed how models under development could be appropriately used and evaluating the status of our understanding of processes in the estuary, river and watershed. The group learned about a variety of methods being used to evaluate eutrophication and toxics issues. A chart comparing models has been posted on our web site.
