Prediction Of Soil Selenium Spatial Distribution Based On Neural Network Model Combined With Geostatistics

In order to realize the optimization management of "digital soil",it is important to predict the spatial distribution of soil properties according to the limited sampling point data,and how to improve its prediction accuracy is one of the hot issues in soil science research.Based on the domestic and foreign research on the spatial distribution of soil properties,the characteristics of soil selenium distribution and the influencing factors,this paper takes the total selenium content of topsoil in Lishui District of Nanjing as an example,with the support of ArcGIS and MATLAB,Based on the data of total selenium content,terrain vegetation data and soil type data,the distribution characteristics of surface selenium content and the influence of environmental variables on the surface soil were studied by using neural network model and geostatistics.The quantitative analysis of the spatial distribution of the total selenium content of the topsoil was carried out by using the RBF neural network combined with the statistical analysis.The results were based on the common Kriging method in the statistical analysis of the ground.Comparative evaluation.So as to provide reference for regional soil resources optimization and "precision agriculture".The main research results are as follows:(1)Descriptive statistical analysis of surface selenium content in the study area was carried out,and the spatial distribution characteristics were studied by means of geostatistical analysis and Moran's I index.The results showed that the content of total selenium in the study area was 0.037?0.914mg/kg,the mean value was 0.238mg/kg,the degree of variation was strong and the spatial autocorrelation was weak.(2)The effect of quantitative environmental variables on the total selenium content of topsoil was studied by rank correlation analysis.The results showed that the higher the slope,the higher the slope,the lower the soil water content,the smaller the vegetation coverage(P<0.01).The content of total selenium in the soil was significantly different(P<0.01),and the soil content was higher than that of the other soil types(P<0.01).The results showed that the content of total selenium in different soil types was significantly different Has an important influence that can not be ignored.(3)Based on the spatial autocorrelation and heterogeneity of the total selenium content in the study area,the quantitative environment variables and qualitative auxiliary variables were fused.RBF neural network model was used as the tool,and RBF neural network was used to analyze the total selenium content Spatial distribution prediction;and based on the general Kriging space prediction method based on geostatistical analysis as a reference method.The results show that the spatial distribution of topsoil and the effect of environmental variables on the total selenium content of topsoil in the study area are in agreement with each other.(4)The accuracy of the two spatial distribution prediction methods is evaluated.It is found that the prediction method based on RBF neural network combined with geostatistical analysis is more suitable than the prediction method based on geostatistics.Which indicates that it has better ability of prediction and can reduce the smoothing effect of Kriging method.The prediction results can reflect the details of the variation of total selenium content with terrain and so on,and it is closer to the complicated actual situation.