Study On Interpolation Method Of Soil Spatial Information Based On RBF Neural Networks

Affected by the nature and human factors, soil property is non- homogeneous in the spatial distribution, that is spatial variability of soil property. The impact of human activity on soil Spatial variability becomes stronger.So, applicable math method must be recurred to in order to apprehend evolvement trend of soil Spatial variability quickly and exactly. The spatial interpolation method is the basic means to the research of soil Spatial variability.By improving the spatial interpolation method,the law of soil Spatial variability can be opaned out with lesser samples,which becomes a hotspot in the research field.TaiHe town and Shangyi towon, MeiShan county was selected for the study.Two samples were collected in different scales, one was collected in large scale and included 80 soil points, between which there was a interval which was 700m,the other one was collected in small scale and included 30 soil points, between which there was a interval which was 50m.Soil SOM.total K and availabe Mn were selected as the object for research.The samples were divided into training and validation datum sets. In order to research the performance of RBF Neural Networks for soil spatial information interpolation, 4 sampling distribution were designed based on the two training sample distribution, and the performance of RBF Neural Networks was compared with that of Kriging method which is widely applied.RBF Neural Networks(Radias Basis Function Neural Networks),which has a strong nonlinear computing competence,is an effective tool for the nonlinear system problem.In this paper, two methods of input of the Neural Networks were designed for training and simulating of the net, one was that: the coordinate of the soil point was designed as the input of net, the value of the soil property in corresponding soil point was the output of the net. The relation was only founded between the coordinate and the value of the soil property,which was named CRBFNN. That was there were 2 nodes(nerve cell) in the input layer and 1 node in the output layer. The other one was that: the coordinate of the soil point and the values of 5 soil points which were close to the point which needed interpolating was designed as the input of net, the...