Landscape pattern analysis for assessing ecosystem condition: Development of a multi-resolution method and application to watershed-delineated landscapes in Pennsylvania

Protection of ecological resources requires the study and management of whole landscape-level ecosystems. The subsequent need for characterizing landscape structure has led to a variety of measurements for assessing different aspects of spatial patterns; however, most of these measurements are known to depend on both the spatial extent of a specified landscape and the measurement grain; therefore, multi-scale measurements would be more informative.; In response, a new method is developed for obtaining a multi-resolution characterization of fragmentation patterns in land cover raster maps within a fixed geographic extent. The concept of conditional entropy is applied to quantify landscape fragmentation as one moves from larger “parent” land cover pixels to smaller “child” pixels that are hierarchically nested within the parent pixels. When applied over a range of resolutions, one obtains a “conditional entropy profile” that can be defined by three parameters.; A method for stochastically simulating landscapes is also developed which allows evaluation of the expected behavior of conditional entropy profiles under known landscape generating mechanisms. This modeling approach also allows for determining sample distributions of different landscape measurements via Monte Carlo simulations.; Using an eight-category raster map that was based on 30-meter resolution LANDSAT TM images, a suite of landscape measurements was obtained for each of 102 Pennsylvania watersheds (a complete tessellation of the state). This included conditional entropy profiles based on the random filter for degrading raster map resolutions. For these watersheds, the conditional entropy profiles are quite sensitive to changing pattern, and together with the readily-available marginal land cover proportions, appear to be very valuable for categorizing landscapes with respect to common types. These profiles have the further appeal of presenting multi-scale fragmentation patterns in a way that can be easily visualized graphically.; Watershed-wide assessments of breeding bird-based ecological integrity and surface water pollution all revealed a strong relationship to the landscape variables, as attested to by various analyses. On one hand, the marginal land cover proportions alone provided the strongest predictors of ecosystem condition; however, the addition of spatial pattern measurements substantially strengthened the predictability, and this included the multiscale parameters describing conditional entropy profiles.