Great Lakes Assessment: Assessing Landscape Pattern and Structure in Great Lakes Forests

Abstract

A fundamental component of assessing ecological integrity across regional landscapes is an understanding of basic landscape pattern in terms of the types, structure, and spatial relationships of forest patches. Landscape patterns affect animal migration (Weines and Milne 1989), dispersal (Geritz et al. 1987) and species coexistence (Pacala 1987). Spatial analyses in the past, however, have been confounded by a lack of high resolution data, lack of analytical tools, and hardware limitations. Equally important, spatial analyses are extremely scale-dependent, and yet the appropriate spatial scales for answering ecological relevant questions are uncertain. For example, forest birds (Hanowski et al. in press)., stream macroinvertebrates (Richards et al 1996), and large mammals (Johnson et al. 1991) are all affected by factors operating at multiple spatial scales, but the relative importance of local vs regional scale factors is largely unknown. There is therefore a need to improve our ability to analyze and interpret the spatial structures of landscapes, both in terms of analytical techniques well as underlying ecological theory. We have recently been able to improve the classification resolution of interpreted LANDSAT imagery. By using LANDSAT scenes from different seasons, we have used differences in plant phenology (e.g. timing of leaf development and senescence) to discriminate among spectrally- similar forest types, such as red and pin oak, or black ash and other hardwoods (Wolter et al. 1996). The resulting classification consists of 22 cover classes, with many classes at the species level. This technique has since been applied across much of northern Minnesota and northwestern Wisconsin. In addition to a high classification resolution, the spatial resolution of LANDSAT imagery is 28.5 m, resulting in a relatively fine-scale classification with respect to spatial pattern. Concurrently, a number of workstation-based analytical tools to conduct sophisticated analysis of landscape spatial structure have been developed (Boeder et al. 1995; McGarigal and Marks 1993, Turner and Gardner 1991). These techniques generally use raster data files (e.g. ERDAS GIS files) to calculate numerous landscape statistics, such as fractal dimension, Shannon-Weiner diversity, dominance and contagion, and lacunarity indices. This suite of statistics provides the ability to quantify numerous aspects of landscape pattern, and potentially relate these indices to measures of integrity of forest and aquatic ecosystems. While we now have well-developed ecological databases and analytical tools at our disposal, we lack fundamental information on some of the most basic questions on regional-scale landscape structure in the Lake States. Where are the largest contiguous patches of our dominant forest types? How are these distributed within the regional hierarchy? Where are the most highly fragmented and most intact landscapes? The objective of this study was therefore to use the databases and techniques described above to quantify landscape structure in the context of higher- level ecological classification units for the Lake States. Specifically, we will use the recently- developed subsection-level ecological classification of the eastern United States (Keys and Carpenter 1995) as a framework for summarizing and interpreting spatial statistics derived from a multitemporal LANDSAT classification of northern Minnesota and Wisconsin.