Classifying remotely sensed data for use in an agricultural nonpoint-source pollution model

ABSTRACT: Models to predict the magnitude of agricultural nonpoint-source pollution in streams have been developed to meet a growing demand for management information. The Agricultural Nonpoint Source (AGNPS) model requires 20 parameters to calculate potential nonpoint-source pollution for a watershed. Lundsat thematic mapper (TM) data, SPOT multispectral data, and SPOT panchromatic data were tested to determine their ability to provide selected inputs to AGNPS. Each data set was classified using supervised and unsupervised methods, and the accuracy of each classification was evaluated using contingency tables and the kappa statistic. Highest classification accuracies were obtained using merged Landsat TM/SPOT panchromatic digital data, although the most cost-gective method for watersheds larger than 141 km2 (54 square miles) was a supervised classification of SPOT multispectral data. In watersheds less than 141 km2, interpretation of aerial photographs was the most accurate and least expensive method of developing land cover maps.