Quantifying Land Cover Features of the Sabinal River Watershed Using ANFIS

Appraisals of ecological services require quantitative assessments for assigning value to specific contributions. Within ecosystems where resources are limited, such as the semi-arid Edwards Aquifer region of Texas, a quantified interpretation of ecological services change is necessary for effective resource management. The focus of this paper is the development of a supervised neuro-fuzzy sub-pixel classification method to create land cover and land use geographical information datasets. The method has been applied to the Sabinal River watershed, which is located centrally in the Texas counties of Bandera, Real and Uvalde. Through the use of digital elevation model (DEM) data files, an approximate boundary of the Sabinal River watershed was located. Landsat Thematic Mapper (TM) images of the watershed taken in spring and winter of both 1999 and 2000 have been superimposed onto the delineated watershed. Select data from the extracted TM image subset have been coupled with corresponding ground-truth data in order to create an Adaptive Neuro-Fuzzy Inference System (ANFIS) model. After training of the FIS was accomplished, model validation through further ground-truth data comparison was done. Once a reasonable prediction model was developed, the corresponding feature maps were imported into a geographical information system (ArcView) for further analysis.