ABSTRACT
Huang Fuchuan is one sandy tributary in middle reaches of Yellow River with watershed area of 3240km2, ravine density of 6.7km/km2, channel frequency of 34 /km2, ravine splitting degree of 35.9%, and an annual average sand supply of 50 million tons to Yellow River. The soil erosion situation of so fragile land and so strong erosion is regarded as the "worst situation of the globe". Because of lack of quantitative research on soil erosion in small watersheds, a uniform soil erosion modulus of 18000t/km2-a was used in soil erosion treatment planning and benefit evaluation in the comprehensive soil erosion treatment campaign. Actually the difference in erosion factors can cause the erosion modulus to be several-fold different or even tens of folds different. In the last 30 years, models of soil erosion have been consecutively developed in various countries The existence of fewer process-based models suitable for specific situation in China was caused by the extremely complex physical process of sand production in soil erosion and the difficulty in describing the physical mechanism of soil erosion by mathematical approaches. Support Vector Machines (SVMs) are new types of general - purpose learning algorithms based on Statistical Learning Theory. SVMs have become the hot topics in machine learning researches and have been successfully applied in classification and regression problems. Based on the difference of problems, they are called Support Vector Classification (SVC) and Support Vector Regression (SVR), respectively. This paper tried to apply the theory of Support Vector Regression in predicting sand production in small watersheds.
