Predicting soil erosion in sloping Hilly Areas of Central Sichuan based on SVM model

Based on Support Vector Machine (SVM) model, observational data from 1991 to 2000 of the six sloping runoff plots at Suining Soil and Water Conservation Experiment Station which located in Hilly Areas of Central Sichuan, China, were used for modeling and predicting the soil erosion. For modeling, five factors (rainfall, rainfall duration, rainfall intensity, vegetation coverage, slope), nine factors (five factors plus early rainfall, early rainfall duration, pre-rainfall intensity and time interval before and after the rain), ten factors (nine factors added to soil and water conservation measures) were as inputs to SVM model, respectively, and the erosion was as output. The results indicated that, the coefficients of efficiency of 5 factors and 9 factors on the soil erosion were 0.52 and 0.55, while that of 10 factors including soil and water conservation measures factor was 0.90. Compared with 5 and 9 factors, the model of 10 factors as input achieved a more satisfied prediction result and could be used for business forecasting. For a particular farming method, the values of soil and water conservation measures factor varied with the slopes, and the values of 15° were 1.46-2.03 times of those of 10°. For a certain farming method, the model with terrain, rainfall, rainfall and the vegetation coverage factors as input factors, could achieve a good effect.