Study on Spatial Variability of Cultivated Soil Available Phosphorus with Garbf Neural Network

The spatial variability of cultivated soil available phosphorus(P) contents is the work foundation of adjusting measures for cultivated management, rationally applying P fertilizer, minimizing P loss, decreasing non-point source pollution of water. A case of Gaozhou City, we collected 664 soil samples in cultivate horizon, then the characteristics of spatial variability, spatial distribution pattern and is cause of county cultivated soil available phosphorus, used Radial Basis Function Network optimized by Genetic Algorithm(GARBF for short) and Geo-statistical methods and so on. The results showed that, while in the whole region of Gaozhou City, the semi-variance structure was existed for spatial variability of surface soil of cultivated land, and models best fit exponential or spherical, there was structural spatial correlation in surface cultivated soil available phosphorus in 5 sampling scales, which indicated that a poor spatial correlation in content developed in large-block scale. The spatial interpolation ability of GARBF neural network method was better than RBF neural network prediction model based on several closest neighbors and Ordinary Kriging method. The P application to cultivated soils in cropping systems in Gaozhou City exceeded P offtake by crops. This surplus could lead to P accumulation in cultivated soils, making them long-term diffuse sources of P loss water, which formed a serious threat to the regional water environment.