Ptf-based method for predicting target soil property and content

A PTF-based method for predicting a target soil property and content, which is used to predict the target soil property in a target region and comprises following steps:

• Step A, selecting, based on existing soil data, sampling sites of which data values of corresponding preset soil physicochemical properties are non-null from the target region, and partitioning a first-level region with a smallest bounding polygon covering all the selected sampling sites, wherein the sampling sites are used as sampling sites corresponding to the first-level region, the preset soil physicochemical properties comprise a target soil property that is referred to as a soil dependent variable and is defined as the target soil property, and rest of the soil physicochemical properties that are used to form a soil independent variable set;
• and then, proceeding to Step B;
• Step B, acquiring layers, covering the first-level region, of specified environmental variables related to the soil dependent variable;
• deriving, for the sampling sites corresponding to the first-level region, values from the specified environmental variables at the sampling sites, and adding the derived values to the soil independent variable set, to update the soil independent variable set;
• and then, proceeding to Step C;
• Step C, deleting independent variables causing multicollinearity and independent variables of which the correlation with the soil dependent variable is less than a preset significant difference threshold, from the soil independent variable set, to update the soil independent variable set;
• and then, proceeding to Step D;
• Step D, training, for the sampling sites corresponding to the first-level region, a linear relationship between the soil dependent variables and the soil independent variables in the soil independent variable set by using a stepwise multiple linear regression model based on a preset number of iterations;
• separately acquiring a temporary optimal independent variable set during each training iteration, and recording a selection number of each of temporary optimal independent variable sets;
• adopting, after completion of training for the preset number of iterations, a temporary optimal independent variable set which is most frequently screened as the optimal independent variable set corresponding to the first-level region;
• and then, proceeding to Step E;
• Step E, acquiring layers of a soil region with preset properties that cover the first-level region, and extracting, for the sampling sites corresponding to the first-level region, soil zones with the preset properties where the sampling sites are located;
• analyzing and obtaining, based on the data values of the soil dependent variable at the sampling sites, difference results in the soil dependent variable between different soil zones;
• if none of the difference results are greater than the preset significant difference threshold, proceeding to Step G;
• merging, if there is a difference result greater than the preset significant difference threshold in the obtained difference results, the soil zones with the difference results not greater than the preset significant difference threshold, and partitioning second-level regions in combination with the soil zones that are not merged;
• acquiring, based on the sampling sites corresponding to the first-level region, sampling sites corresponding to the second-level regions;
• and then, proceeding to Step F;
• Step F, acquiring, for each of the second-level regions, an optimal independent variable set corresponding to each of the second-level regions by using the method described in Step D, and then proceeding to Step G;
• Step G, training, for the sampling sites corresponding to the first-level region, a linear regression model and a nonlinear regression model between the data values of the soil dependent variables and the data values of independent variables in the corresponding optimal independent variable set;
• and acquiring a coefficient of determination of the linear regression model and that of the nonlinear regression model, wherein the coefficient of determination of the linear regression model corresponding to the first-level region is R_OLS and the coefficient of determination of the nonlinear regression model corresponding to the first-level region is R_NLS;
• further, if there is no second-level region, then, proceeding to Step H;
• or if there are second-level regions, for the second-level regions, separately training a linear regression model and a nonlinear regression model between the data values of the soil dependent variables at the sampling sites and the data values of the independent variables in the corresponding optimal independent variable set, and acquiring the coefficient of determination of the linear regression model and that of the nonlinear regression model, thus acquiring coefficients of determination of linear regression models and nonlinear regression models corresponding to all the second-level regions, and further acquiring mean values of these coefficients of determination of the linear regression models and nonlinear regression models corresponding to all the second-level regions, which are termed R_OLS mean and R_NLS mean, respectively;
• and then, proceeding to Step H;
• Step H, if there is no second-level region, proceeding to Step I;
• if there are second-level regions, when R_OLS is greater than both R_OLS mean and R_NLS mean or R_NLS is greater than both R_OLS mean and R_NLS mean, proceeding to Step I;
• and
• when R_OLS mean is greater than both R_OLS and R_NLS or R_NLS mean is greater than both R_OLS and R NLS, proceeding to Step M;
• Step I, based on the sampling sites corresponding to the first-level region and the corresponding optimal independent variable set, and according to data values of soil physicochemical properties in the optimal independent variable set corresponding to the sampling sites, acquiring a prediction model based on all the specified environmental variables in Step B for the soil physicochemical properties in the optimal independent variable set;
• acquiring spatial distribution prediction layers of the soil physicochemical properties in the optimal independent variable set corresponding to the first-level region according to the layers of the specified environmental variables in Step B;
• and then, proceeding to Step J;
• Step J, merging the spatial distribution prediction layers of the soil physicochemical properties and the environmental variables in the optimal independent variable set corresponding to the first-level region, to form an optimal independent variable layer set corresponding to the first-level region;
• and then, proceeding to Step K;
• Step K, if R_OLS≥R_NLS, for the sampling sites corresponding to the first-level region, deriving data values of the independent variables from the optimal independent variable layer set, and training a linear regression model between the data values of the independent variables in the optimal independent variable set and the data values of the soil dependent variables, to form a first-level region prediction model;
• and then, proceeding to Step L;
• if R_NLS>R_OLS, for the sampling sites corresponding to the first-level region, deriving data values of independent variables from the optimal independent variable layer set, and training a nonlinear regression model between the data values of the independent variables in the optimal independent variable set and the data values of the soil dependent variables, to form a first-level region prediction model;
• and then, proceeding to Step L;
• Step L, generating, according to the optimal independent variable layer set corresponding to the first-level region, a spatial distribution map of the soil dependent variable, namely, a spatial distribution map of the target soil property in the target region, by applying the first-level region prediction model, to realize prediction of the target soil property in the target region;
• Step M, for each second-level region, acquiring an optimal independent variable layer set corresponding to the each second-level region by using the method descried in Steps I to J;
• and then, proceeding to Step N;
• Step N, if R_OLS mean≥R_NLS mean, for each second-level region and for the sampling sites corresponding to the second-level region, deriving data values of independent variables from the optimal independent variable set corresponding to the second-level region, and training a linear regression model between the data values of the independent variables and the data values of the soil dependent variables, to form a second-level region prediction model, thus obtaining the second-level region prediction model for each of the second-level regions;
• and then, proceeding to Step O;
• if R_NLS mean>R_OLS mean, for each second-level region and for the sampling sites corresponding to the second-level region, deriving data values of independent variables from the optimal independent variable set corresponding to the second-level region, and training a nonlinear regression model between the data values of the independent variables and the data values of the soil dependent variables, to form a second-level region prediction model, thus obtaining second-level region prediction model for each of the second-level regions;
• and then, proceeding to Step O;
• and
• Step O, generating, for each second-level region, according to the optimal independent variable layer set corresponding to the second-level region, a spatial distribution map of the soil dependent variable in the second-level region by applying the second-level region prediction model, thus obtaining a spatial distribution map of the soil dependent variable in each of the second-level regions;
• and obtaining a spatial distribution map of the target soil property in the target region after combination, to predict the target soil property and content in the target region.