Soil salinization disaster prediction model based on heterogeneous SVM neural network

In the paper，we analyzes the soil salinization in Yinchuan plain of Ningxia, predicts the degree of soil salinization and the potential development trend of soil salinization. Based on the Landsat 8 OLI data and field measurements, we select ground elevation, groundwater burial depth, TDS, vegetation index, salinity index and drought index as indicator and use these indicators and field test sample data to set up the database and establish a disaster model to predict soil salinization which is based on heterogeneous SVM neural network algorithm. The results show that (1) selecting the Radial Basis Function as the kernel function of the early warning model, when c=100 and g=3, can make the accuracy up to 85%. (2) The soil area of mildly salinized soil is about 854.08 km2, the area of moderate salinized soil is about 985.52 km2, and the area of severe salinized soil is about 231.97 km2.They mainly occur in the Xidatan town of Pingluo county, the Luhua area near Yinchuan and the Kushuihe district of Wuzhong.(3) The soil salinization in the northern part of the Yinchuan plain is more serious, and it widely exists in the abandoned areas around the cultivated land and in the shallow areas where groundwater occurs. The soil salinization is serious in the cultivated land of the Yinchuan plain, and attention should be paid to proper irrigation and drainage to increase the sustainable utilization of soil.