Establish Soil Erosion Model Based On The Machine Learning Theory

Soil erosion is one kind of widespread geophysical dynamics phenomena under the synthetic effect of internal-force and external-force. Soil erosion is the main reason of desertification and environmental degradation. It leads to decline in fertility, physical and chemical properties bad and reduction of soil utilization rate. It has bad influence on both environmental protection and guarantee of agricultural land. The prediction of soil erosion can help the government to make the policy and measure to defend soil erosion. After many experts’ improvements, the common soil erosion models have became more and more accurate. But these models also have some shortcomings such as low applicability, small prediction scale, complicated input data, etc.This research attempts to establish a flexible soil erosion model which uses SVM as the kernel to contact the factors with soil erosion in Zhuji Zhejiang. Using GIS to extract the input factors such as precipitation, soil property, gradient, slope length, land use pattern. We got the daily precipitation data of Zhuji in2000to2011from Zhuji Puyang river hydrometric station; got the soil property data from Nanjing soil research of Chinese academy of sciences; downloaded the DEM of China in2009and remote diagram of Zhuji in2005from the database of Chinese Academy of Sciences. By calculating and sifting the precipitation data and soil property data, using GIS to calculate the gradient and slope length, using GIS and ERDAS to statistics the land use pattern data, we got the input data and output data of the prediction model.Randomly select the input data and output data to put into SVM and proceed the parameter and model optimizing. After step length selection, input data adjustment, testing of the data volume to put into the model, we got the optimum parameters which can make the prediction model accurately, steadily and rathe. So we got a prediction model which can predict the amount of soil erosion by precipitation, soil property, gradient, slope length and land use pattern. The predictive accuracy of the model can reach75%.At last, we made a discussion and analysis for the model. The model has some advantage such as getting input data easily, data modification expediently, autoregulation and strong applicability. It also has some imperfection to improve such as data collection and pretreatment, perfection of influence factor, selection of the optimum parameters and practical application.This research establish a soil erosion prediction model by using SVM as the kernel to contact the factors with soil erosion in Zhuji Zhejiang. After data filtration. preferences and model optimizing, the predictive accuracy of the model reach75%.