Study on the Trend of Range Cover Changes Using Fuzzy ARTMAP Method and GIS

Abstract

The major aim of processing satellite images is to prepare topical and effective<br />maps. The selection of appropriate classification methods plays an important role. Among<br />various methods existing for image classification, artificial neural network method is of<br />high accuracy. In present study, TM images of 1987, and ETM+ images of 2000 and 2006<br />were analyzed using artificial fuzzy ARTMAP neural network within Mehrgan region,<br />Kermanshah province, Iran, with an area of 5957 ha changes in range cover state in this<br />basin during 3 periods of time from 1987 to 2000 and 2000 to 2006 were examined. In this<br />study, initially, Land sat data for intended years were corrected geometrically and<br />radiometric ally. Next, different land use classes were defined and training samples<br />obtained via field visits. The obtained results show that, over time period of 1987-2000, the<br />extent of low-density rangeland and farmland in study region had been increased by 89.09<br />and 321.08 ha, respectively, while good rangeland and fair rangeland faced a declining<br />trend of 358.29 ha and 48.89 ha. Also, during time period of 2000-2006, the extent of poor<br />rangeland and farmland within study region has increased by 64.98 and 727.12 ha,<br />respectively, while good rangeland and fair rangeland faced a declining trend of 144.01 ha<br />and 648.1 ha. Accuracy of vegetation maps resulting from satellite data classification using<br />algorithm of artificial fuzzy ARTMAP neural network was 90.97% and 94% for TM<br />(1987) images and ETM+ (2000,2006) respectively which indicates high accuracy of<br />ARTMAP algorithms for classifying satellite. Therefore, this study proves high efficiency<br />and potential of artificial fuzzy ARTMAP neural network for classification of remote<br />sensing images.