Mapping Detailed Biotic Communities in the Upper San Pedro Valley of Southeastern Arizona Using Landsat 7 ETM+ Data and Supervised Spectral Angle Classifier

Abstract

In this study, we demonstrated that ecologically meaningful, detailed biotic communities can be mapped with high accuracy (85 percent) in a soil dominated semi-arid environment using a simple supervised classification approach with Landsat TM data, if an appropriate classification algorithm and truly representative training signatures are applied. We used the Supervised Spectral Angle Classifier by Sohn and Rebello (2002) to map biotic communities in the Upper San Pedro Valley, Arizona. By choosing an image acquisition date (season) that can maximize the differentiation of feature characteristics associated with each biotic community, and adopting the Supervised Spectral Angle Classifier that is sensitive to the spectral shape pattern, we were able to map biotic communities without the secondary, derived variables such as DEM, soil, and seasonal NDVI, seasonal greenness indices to incorporate environmental factors and phenological changes of vegetation into the classification procedures. Biotic communities and land cover categories mapped in this study include: Chihuahuan desert-scrub, mesquite scrub, Chihuahuan semi-desert grassland ( 30 percent grass-forbs cove density, and disclimax), Madrean encinal woodland, Madrean montane conifer forest, mesquite bosque, riparian gallery, irrigated pasture/golf course, urban/developed, bare soil, and water. Our study reaffirmed that the Spectral Angle Classifier can potentially be one of the most robust and accurate classifiers. Also, according to our study, the current spatial, spectral, and radiometric characteristics of Landsat TM data seem effective for mapping detailed vegetation communities even in a semi-arid environment.