Hierarchical Modeling for Integrated Environmental Assessments

The distribution of biotic resources over large spatial extents is often a function of climate, of land-use, and of the demographics of the human population but these different classes of independent variables have different spatial scales for their action. One approach to the integration of these effects across scales is to use hierarchical models that incorporate contingencies and constraints in effects. This project seeks to develop such a modeling paradigm by use of classification and regression trees (CART). The hexagonal grid of the U.S. Environmental Protection Agency\u27s Environmental Monitoring and Assessment Program is used as the spatial grid, with about 12000 grid points within the conterminous U.S. To test the approach, national Breeding Bird Survey data are used as dependent variables (species richness, abundance of individual species). The independent variables are derived from climate models, from remote sensing of land use from Advanced Very High Resolution Radiometry (the Loveland prototype land cover classification), from National Agricultural Statistics Service statistics, and from Bureau of the Census data. The CART modeling then determines the best combination of predictors for each hexagon with bird data (ca. 1200), and the resulting model is used to predict values for all other hexagons. One can then map the regions with unique patterns of environmental effects on birds, can determine the location and degree of species losses associated with the local or regional action of particular stressors, and assess the location and total impact of the overall stressor load on the system. Specific questions that are addressed in the project include the value of different datasets as predictors, the influence of spatial autocorrelation, improvement of model fit and assessment of prediction accuracy, and methods of visualization of the results. The effects of climate, land-use and human demographics on environmental resources are particularly difficult to estimate in an integrated manner because some effects are broad-scale while others are very localized. If one could do so, a national environmental risk assessment would be feasible in a cost-effective manner. Here a new approach to the problem is tested, using a hierarchical model that first adjusts for broad- scale effects and then assesses regional and local effects in turn. As a test, bird data for the 48 conterminous states are modeled against national climate data, remote sensing data on land use, and agricultural and Census data. The resulting national map shows on a 12,000 point grid the extent to which species are lost to particular or to all stressors at each point, in effect a national risk assessment of environmental impacts as experienced by birds in relation to the stressors tested. The approach can readily be extended to other environmental response variables and to smaller or larger scales. This project systematically investigates technical uncertainties that might limit the usefulness of the approach