Using monitoring data of surface soil to predict whole crop-root zone soil water content with PSO-LSSVM, GRNN and WNN

Abstract

Drought is a significant disaster in Beijing and it is important to find a method to assess the drought condition. First, this paper collected data of 85 soil monitoring stations in Beijing, such as soil dry bulk densities, saturated water contents, field capacities. Then, spatial variability characteristics of soil physics parameters were investigated by GIS and other three factors, 10 cm soil moisture content, organic matter and saturated water content which notably influenced soil moisture were extracted by Principal Component Analysis (PCA). Furthermore, four different nonlinear methods were put forward to predict crop-root zone soil water. 15555 single daily data from 2011 were used in parameters determination, while 15470 double daily data were used to test. The result showed that the Least Square Support Vector Machine coupling Particle Swarm Optimization Algorithm (PSO-LSSVM) (R ² = 0. 875) did better than BP Neural Network (R ² = 0. 840), Generalized Regression Neural Network (GRNN) (R ² = 0. 850) and Wavelet Neural Network (WNN) (R ² = 0. 853). As so the POS-LSSVM method was used to evaluate the drought conditions from October 2010 to March 2011 of Beijing, and the result showed that from October 2010 to January 2011, the drought conditions were getting increasingly worse while later relieved from January 2011 to March 2011.