Spatial Prediction And Uncertainty Analysis Of Regional Soil Nutrients

Soil organic matter and total nitrogen are important factors to measure soil health,and can also characterize the quality of the ecological environment to a certain extent.Their precise digital mapping(expanded from limited sampling points to regional area data)is important for realizing soil organic matter and total nitrogen.Regional monitoring and precise fertilization are essential.At present,soil attribute spatial prediction methods mainly include geostatistics,machine learning methods,remote sensing inversion,etc.The prediction accuracy and uncertainty of different models under different number of points,topographic changes and different auxiliary variable combinations need to be further studied.Taking soil organic matter and total nitrogen as the object,using RBF(radial basis function)neural network,partial least squares,and random forest models,selecting elements with greater correlation with organic matter and total nitrogen as auxiliary variables,and carrying out different prediction models and samples Forecast accuracy and uncertainty analysis under the number of points and auxiliary variables.The following main conclusions are obtained:(1)The content of soil organic matter in the study area is extremely significantly correlated with normalized vegetation index,elevation and effective soil thickness indicators,with the largest correlation coefficient.The above combination is used to construct auxiliary variables for soil organic matter spatial prediction;different spatial predictions are combined through different auxiliary variables The comparative analysis of prediction accuracy of the methods shows that no matter which combination of auxiliary variables,the prediction accuracy of random forest is the highest.For the soil organic matter in this study area,the random forest prediction accuracy assisted by the combination of effective soil layer thickness and elevation is the best,with root mean square error(RMSE),average absolute error(MAE),average relative error(MRE)and consistency The index(d)reached 3.20,2.41,18.15 and 0.77 respectively.(2)The total nitrogen content of the soil in the study area is extremely significantly correlated with slope,elevation,cultivated layer thickness,and normalized vegetation index indicators,with the largest correlation coefficient;regardless of the number of sampling points,under specific auxiliary variables,random forest The prediction accuracy of soil total nitrogen in the study area is the highest,and the model fitting effect is the best.The prediction accuracy of different models has different response mechanisms to different number of sampling points.When the number of sampling points is 150-200,the three prediction models have the highest prediction accuracy.When the number of sampling points is 200,the random forest prediction accuracy Optimal,RMSE,MAE,MRE reach 0.15,0.10,and 0.10,respectively.When the number of sampling points is greater than 200,the prediction accuracy will not increase,but will decrease.(3)From the analysis of the uncertainty of soil organic matter,the uncertainty is the smallest when the prediction model is random forest,the auxiliary variables are effective soil thickness and elevation,the standard deviation range is 0.002～4.07g/kg,and the trend analysis chart is The curve trend is the closest to the curve trend of the original data trend analysis graph;from the uncertainty analysis of soil total nitrogen,the prediction model is RBF neural network and the number of sampling points is 200,the curve trend of the trend analysis graph is the trend of the original data The curve trend of the analysis chart is the closest,the standard deviation range is 0.001～0.04 g/kg,and the uncertainty is the lowest.Figure 29 table 6 reference 7...