Estimation of Soil Erosion by Satellite Images Integration and SVM Method (Case Study: South Khorasan Province)

Landsat 8, Modis and Support Vector Machine (SVM) classification algorithm are used to map soil erosion in South Khorasan province in the present study. For this purpose, three information layers of land use map, slope and vegetation are used. In the proposed method to prepare the land use map, four supervised classification algorithms of maximum likelihood, Mahalobani distance, minimum distance and artificial neural network with thermal infrared band are used. Six integration algorithms of NNDiffuse, HPF, Brovey, Gram-Schmidt, PC and CN are used to prepare the vegetation map. The ASTER satellite DEM map is used to prepare the slope map of the area. The results of the experiments show that the maximum likelihood algorithm with thermal band is the most accurate in land use mapping. The NNDiffuse algorithm is also more accurate for integrating the red and infrared bands near Landsat 8 and Modis. After preparing the land use, slope and three vegetation maps derived from Landsat 8, Modis and Landsat 8 and Modis integrated images, three erosion maps are prepared using SVM algorithm. The results show that the highest value of Kappa coefficient (69.1) is related to the erosion map obtained from Landsat 8 and the lowest value (57.1) is related to Modis image. The integration of Landsat 8 and Modis increases the kappa coefficient to 67.3 %.