Providing Land Use and Land Cover Maps Using Remote Sensing Data and Artificial Neural Network

Land use and land cover (LULC) maps are necessary for many management activities, hydrology and erosion analysis. Remote sensing data have a high potential for providing up-to-date LULC maps. The objective of this study was to provide and evaluate the LULC maps of Gavshan dam watershed in west of Iran using Landsat 8 satellite images and artificial neural network. Hence, 1320 ground control points or reference points were used to train and test the ANN model for providing LULC maps. Land use classification at each point was specified with a land survey or using Google Earth images. The identified LULC classes in this basin included agriculture, buffer forests (riverside trees), orchards, bushes, forage grasslands, residential areas, roads and water. The ANN used in this study was a feed-forward perceptron that was trained using a coupled conjugate gradient backpropagation algorithm. The input variables for the ANN model were the revised spectral reflectance of bands 1 to 7 of Landsat 8 satellite images. The evaluation of the ANN model made by the ground control data showed a high accuracy for the used method with a general accuracy of 78.5% and kappa coefficient of 68.5%. The results of this study indicated that the utilization of the ANN and Landsat 8 satellite images provides an opportunity to produce LULC maps with high accuracy.