A multivariate analysis of metrics describing landscape pattern and structure

Advances in both the availability of land cover maps from satellite sensors and computing methodology to process such maps means that it is now feasible to monitor some aspects of landscape condition nationwide from remotely sensed imagery. It is therefore desirable to identify a subset of landscape pattern metrics which meet the criteria for monitoring landscape condition and which can be estimated from satellite-derived land cover maps. Previous research revealed many redundancies among landscape metrics and reduced the apparent number of unique dimensions to a half-dozen, but that study used USGS Land Use Data Analysis (LUDA) maps which are qualitatively different from satellite-derived maps. The present study used factor analysis to test the stability of landscape metrics across sets of Landsat Thematic Mapper land cover maps for the Tennessee River and Chesapeake Bay watersheds that differ in resolutions, numbers of attributes, and methods of defining subset boundaries. Diversity and texture and fractal indices were more stable than shape and compaction indices, and they accounted for a significant proportion of the total variance among the land cover maps. Results indicate that individual shape and compaction indices may be too unstable to consistently describe landscape condition across huge data sets, but further research on this suite of metrics might reveal a small set that consistently captures the shape and compaction dimension(s).