CN115266612A - A method for mapping soil available phosphorus in cultivated land in southern hilly areas based on high-resolution environmental variables - Google Patents

Abstract

The invention relates to the technical field of soil available phosphorus research, in particular to a method for mapping soil available phosphorus in cultivated land in southern hilly areas based on high-resolution environmental variables. , S3: Data preprocessing, S4: Extraction and screening of environmental variables, S5: Model construction and accuracy evaluation, and S6: Predicting available phosphorus in the surface soil (0-15cm) of all cultivated land in the test area at the same time as the sampling time of soil sample points Content: The method for mapping soil available phosphorus in cultivated land in southern hilly areas based on high-resolution environmental variables provided by the present invention can significantly improve the mapping accuracy of soil available phosphorus in cultivated land in southern hilly areas, and quickly realize the improvement of the content of available phosphorus in large-area cultivated land in southern hilly areas. The prediction and its spatial distribution are suitable for further promotion and application.

Description

A method of soil available phosphorus in cultivated land in southern hilly areas based on high-resolution environmental variables Mapping method

technical field

The invention relates to the technical field of soil available phosphorus research, in particular to a method for mapping soil available phosphorus in cultivated land in southern hilly areas based on high-resolution environmental variables.

Background technique

Traditional analysis and mapping of soil available phosphorus content requires the collection of a large number of soil samples, which is often time-consuming, laborious, and costly. It also increases ecological and environmental risks, and the accuracy of mapping for larger areas is low. Satellite remote sensing images can reflect soil surface information and are used in digital soil mapping research. Many satellite remote sensing images can be obtained for free, and the spatial coverage of remote sensing data is continuous. As a secondary data source, it can improve the mapping efficiency of sparse sample soil properties. The use of satellite remote sensing images combined with other environmental variables has become an important means of digital soil mapping of large areas with sparse soil samples. However, due to the low spatial resolution of the previously available free remote sensing satellite data, the accuracy of the final digital soil mapping is not high enough. At present, with the emergence of new remote sensing data with higher spatial resolution and easy access, such as Sentinel-2 satellites, compared with commonly used Landsat data, Sentinel-2 data has higher spatial resolution and 3 red-edge bands, and The red-edge band has a clear advantage in predicting soil properties such as organic matter and total nitrogen. Relevant studies have confirmed that adding remote sensing data can effectively improve the prediction accuracy of soil phosphorus. Therefore, the introduction of Sentinel-2 data can improve the accuracy of soil phosphorus mapping.

In addition, the previous digital soil mapping was mainly for the cultivated land in the northern flat area. The single field area is large and large-scale, and the soil spectral characteristics and topography of the cultivated land are relatively uniform. Therefore, the current digital soil available phosphorus mapping method is better. However, most of the cultivated land in the south is hilly area, with large terrain undulations, and the cultivated land is relatively fragmented. The area of a single field is small, and the accuracy of traditional digital mapping is extremely poor. Therefore, the currently commonly used digital soil mapping methods are not suitable for the southern hilly areas. Many scholars have significantly improved the accuracy of digital soil carbon and nitrogen mapping by using remote sensing variables or topographic variables with higher precision, that is, higher spatial resolution.

However, it is still blank to use higher spatial resolution remote sensing variables or terrain variables to improve the accuracy of digital soil phosphorus mapping. Therefore, this patent uses higher spatial resolution remote sensing variables and terrain variables to improve the accuracy of digital soil available phosphorus mapping.

Contents of the invention

In order to solve the above problems, the present invention provides a method for mapping soil available phosphorus in cultivated land in southern hilly areas based on high-resolution environmental variables to improve the accuracy of mapping soil available phosphorus in the surface layer (0-15cm) of cultivated land in southern hilly areas.

In order to solve the problems of the technologies described above, the technical solution adopted in the present invention is:

A method for mapping soil available phosphorus in cultivated land in southern hilly areas based on high-resolution environmental variables, including the following steps:

S1. Acquisition of soil sample point data and building a database: obtain the physical and chemical attribute data of the surface soil sample points of the cultivated land in the area to be tested, and establish a soil sample point database in ArcGIS10.2 software;

S2. Acquisition of environmental data: Acquire the environmental data of the area at the same time as the soil sampling point, including medium and high resolution remote sensing variables, medium and high resolution terrain data and meteorological data;

S3. Data preprocessing: project the remote sensing data, terrain data and meteorological data to the Xi’an 1980 coordinate system, and check the consistency of the spatial matching between the environmental data and the soil sample point attributes, and connect the spatial attributes after correctness. Administrative division map, use the vector clipping function in ArcGIS10.2 software to get the cultivated land surface soil in the area to be measured;

S4. Extraction and screening of environmental variables: The selected predictive variables include remote sensing variables, topographical variables, meteorological variables and soil pH value; in order to simplify the model input, the auxiliary predictors are optimized before modeling, and the relationship between soil and soil is retained through Pearson correlation analysis. For the variables with significant correlation with available phosphorus, the auxiliary variables that finally participate in the modeling are retained through the backward elimination method;

S5. Model construction and accuracy evaluation: the soil available phosphorus content of the preset number of cultivated land surfaces in the area to be measured is used as the dependent variable, and the medium and high resolution remote sensing variables, topographic variables, meteorological variables and soil pH value screened above are used as independent variables On the Python platform, the random forest model calculation is performed based on the medium and high resolution environmental variables, and the random forest model with the best prediction accuracy and its variable importance analysis are automatically selected, and the sampling time of the soil sample points based on the medium and high resolution environmental variables are respectively obtained The predictive model of the available phosphorus content in the surface soil of the cultivated land in the area to be tested in the same period;

S6. Predict the available phosphorus content of the surface soil (0-15cm) of all cultivated land in the area to be tested at the same time as the sampling time of the soil sample points: the best model constructed according to the high-resolution (10m) environmental variables and input high-precision (10m) m) environment variable, execute the random forest algorithm in Python software, predict the available phosphorus content of all cultivated land surface soil (0-15cm) in the same area to be measured, and obtain all the cultivated land surface soil (0-15cm) in the area to be measured in this period Spatial distribution map of available phosphorus.

Further, S1 specifically includes the following steps:

S11: Study the physical and chemical attribute data of the preset number of cultivated land surface soil samples in the area to be measured, which comes from the data of the cultivated land quality inspection and evaluation sample points of the Ministry of Agriculture and Rural Affairs, including soil available phosphorus, soil pH value, geographical coordinates, and elevation;

S12: According to the geographical coordinates and physical and chemical attribute data of the soil samples, the soil sample points database was established in ArcGIS10.2 software, and the spatial distribution map of the soil samples under the projection of Xi'an 1980 coordinate system was obtained.

Further, the pH value of the soil in S11 was measured by the acidity meter method, and the available phosphorus in the soil was measured by the sodium bicarbonate leaching-molybdenum antimony anti-colorimetric method.

Further, the topographic data described in S2 includes a high-resolution (12.5 meters) DEM and a DEM with a spatial resolution of 30 meters commonly used in digital soil mapping;

The remote sensing data includes 4 high-resolution (10 meters) Sentinel-2A remote sensing images and Landsat-8OLI images with a spatial resolution of 30 meters commonly used in digital soil mapping;

The meteorological data are 1km×1km grid data of historical monthly average precipitation and monthly average temperature in the area to be measured.

Further, S3 specifically includes the following steps:

S31. Remote sensing data preprocessing: remote sensing data preprocessing includes radiometric calibration, atmospheric correction, geometric fine correction, and image mosaic; for the Sentinel-2A remote sensing images of the 4 scenes in the study area, the SNAP software performs radiometric calibration for each scene separately. Atmospheric correction, the DN value of the remote sensing image is converted into the surface reflectance, and then the image is mosaiced and stitched, and then the finely corrected SPOT5 reference image is used in the ENVI5.3 software to perform geometric fine correction, and according to the administrative division of the area to be measured The image is cropped; the Landsat-8OLI remote sensing image is a geometrically corrected image, and the atmospheric correction is performed in the ENVI5.3 software. The geometrically corrected error of all remote sensing images is controlled within 1 pixel, and they are uniformly projected into the Xi'an 1980 coordinate system;

S32. Preprocessing of terrain variables: firstly, after checking that the DEM data with spatial resolutions of 12.5m and 30m are correct, in ArcGIS10.2 software, they are uniformly converted to the Xi’an 1980 coordinate system through projection transformation, and the 12.5 The DEM of m is produced as DEM data with a spatial resolution of 10m;

S33. Preprocessing of meteorological variables: using arcgis10.2 software to project the meteorological data into the Xi'an 1980 coordinate system;

S34. In the ArcGIS10.2 software, check the consistency of the attribute spatial matching of the above remote sensing data, terrain data, meteorological data and soil sample point data, and establish a geographical connection between the spatial data and the soil attribute after the inspection is correct.

Further, the backward culling method in S4 is specifically:

The variables were screened according to the increase or decrease of the root mean square error after each variable was excluded from the model in turn during the modeling. If the RMSE increased, the variables were retained, and vice versa.

Further, S4 specifically includes the following steps:

S41. Extraction of remote sensing variables: extracting remote sensing variables from Sentinel-2A and Landsat-8OLI in ArcGIS10.2 software to calculate 46 spectral indices for digital soil mapping;

S42. Extraction of topographic variables: 8 topographic variables commonly used in digital soil mapping are extracted from the resampled 10-meter resolution DEM and 30-meter DEM in ArcGIS10.2 software;

S43. Extraction of meteorological variables: Calculate the average monthly precipitation and average temperature data for many years in ArcGIS10.2 software, and then use the grid calculator to calculate the average value of the average monthly precipitation and average temperature for many years in the study area to obtain the study area Years of average annual temperature and average annual precipitation data, and finally use the nearest neighbor method to resample the annual average precipitation and annual average temperature data to raster data with a spatial resolution of 10 meters and 30 meters;

S44. Acquisition of the spatial distribution map of soil pH: use the Kriging interpolation method in the ArcGIS10.2 software to obtain the soil pH in the study area for interpolation to obtain its spatial distribution map;

S45, in the ArcGIS10.2 software, the soil sample points are geographically correlated with the above remote sensing variables, terrain variables, meteorological variables and soil pH data, and all environmental variables corresponding to the soil sample points in the study area are obtained;

S46. Screening of environmental variables involved in modeling: Perform Pearson correlation analysis on the Python platform to screen variables that have a significant correlation with soil available phosphorus. Through the backward elimination method, each variable is excluded from the model in turn according to the mean square of the model. The increase or decrease of the root error (RMSE) screens the variables, the RMSE increases and the variables are retained, otherwise, the auxiliary variables that are finally involved in the modeling are retained, and finally the sample set that participates in the modeling is obtained.

Further, S5 specifically includes the following steps:

S51. Constructing a training sample set and a verification sample set: randomly divide the above sample set into a training set and a verification set in proportion, wherein the samples of the training set are used for modeling, and the samples of the verification set are used for testing the prediction accuracy of the model;

S52. Determine the dependent variable and independent variable of the model: use the soil available phosphorus content of the cultivated land surface of the soil sample point in the area to be measured as the dependent variable, and use the screened significant correlation high-resolution (10 meters) remote sensing variables, high-resolution The topographic variables, meteorological variables and soil pH value of the rate (10 meters) are used as independent variables, and the random forest algorithm is executed on the Python platform to automatically screen the effective soil surface soil of the cultivated land in the area to be measured based on the high-precision environmental variables at the same time as the sampling time of the soil samples. Phosphorus content prediction model;

S53, model prediction accuracy evaluation: filter the best prediction model by R ² (coefficient of determination), MAE (mean absolute error) and RMSE (root mean square error), the closer R ² is to 1, the smaller the MAE and RMSE are, indicating that the prediction The higher the accuracy of the model; the accuracy evaluation formula (1-3) is:

为实测值平均值；In the formula, n is the number of sampling points, O _i and P _i are the measured and predicted values of sampling point i,

is the average value of the measured value;

S54. Taking the soil available phosphorus content in the surface layer (0-15cm) of the cultivated land of the soil sample point as the dependent variable, and using the remote sensing variable with a moderate resolution (30 meters) and the terrain variable with a medium resolution (30 meters) that are significantly related to the screening , Meteorological variables and soil pH value as independent variables, repeat the same steps as S51, and obtain the best available phosphorus content of cultivated land surface soil (0-15cm) based on the medium-resolution environmental variables and the sampling time of the soil sample point at the same time. random forest prediction model;

S55. Obtain the relative importance score of variables in the process of predicting soil available phosphorus content by taking the average value of multiple iterations of the RF model;

The relative importance of variables based on the modeling of environmental variables with a resolution of 10m shows that the relative importance scores of meteorological variables, topographical variables, remote sensing variables, and soil pH are 30.64%, 30.38%, 22.87%, and 16.11%; The order of importance of individual variables from large to small is annual average temperature, pH value, topographic humidity index, DEM value, annual average rainfall, enhanced vegetation index, first principal component and red edge band6;

The relative importance of variables based on environmental variables with a spatial resolution of 30 meters shows that the relative importance scores of meteorological variables, topographical variables, remote sensing variables and soil pH are 25.86%, 32.44%, 21.31% and 20.39% respectively; The order of importance from large to small is elevation (DEM value), pH value, average annual rainfall, enhanced vegetation index, B5 and the first principal component.

The beneficial effects of the present invention are:

The method for mapping available soil phosphorus in cultivated land soil in southern hilly areas based on high-resolution environmental variables provided by the present invention can significantly improve the accuracy of mapping available phosphorus in cultivated land soil in southern hilly areas, and quickly realize the prediction and analysis of available phosphorus content in large-scale cultivated land soil in southern hilly areas. Its spatial distribution is suitable for further popularization and application.

Description of drawings

Fig. 1 is the sample point distribution map of cultivated land surface soil (0-15cm);

Figure 2 is based on the spatial distribution of soil available phosphorus content in the study area;

Fig. 3 is a schematic flow chart of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are part of the present invention Examples, not all examples. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

Referring to the accompanying drawing 3, the present embodiment takes the cultivated land in Jian'ou City, Fujian Province as an object, and carries out mapping of available phosphorus in cultivated land soil. Specific steps are as follows:

S1, Soil sample point data acquisition and database building. The physical and chemical property data of the cultivated land surface (0-15cm) soil sample points in Jian'ou City, Fujian Province in 2017 were obtained, and the soil sample point database was established in ArcGIS10.2 software. Specifically include the following steps:

S11, the physical and chemical attribute data of 97 cultivated land surface soil (0-20cm) samples in the study area are derived from the data of the cultivated land quality inspection and evaluation sample points at the end of 2017 by the Ministry of Agriculture and Rural Affairs of the People’s Republic of China, including soil available phosphorus, soil pH value, geographical coordinates , elevation, etc. Among them, the pH value of the soil was measured by the acidity meter method, and the available phosphorus in the soil was measured by the sodium bicarbonate leaching-molybdenum antimony anti-colorimetric method. A total of 96 soil sampling points were used, excluding one sampling point affected by clouds on the satellite remote sensing image in the same period.

S12, according to the geographical coordinates, available phosphorus content, pH value and other physical and chemical properties of 96 soil sample points, the soil sample point database was established in ArcGIS10.2 software, and the spatial distribution map of soil sample points under the projection of Xi’an 1980 coordinate system was obtained, see figure 1.

S2, acquisition of environmental data. Obtain the environmental data of the area at the same time as the soil sampling point, including medium and high resolution remote sensing variables, medium and high resolution terrain data and meteorological data. Among them, topographic data include the high-resolution (12.5m) DEM of Jian'ou City, Fujian Province in 2017 and the DEM with a spatial resolution of 30m commonly used in digital soil mapping. The remote sensing data are 4-scene Sentinel-2A remote sensing images with high resolution (10m) at the same time as the sampling time of soil samples and Landsat-8OLI images with a spatial resolution of 30m commonly used in digital soil mapping. Meteorological data are 1km×1km raster data of monthly average precipitation (January-December) and monthly average temperature (January-December) from 1970 to 2000.

S3, data preprocessing. Project the remote sensing data, terrain data and meteorological data to the Xi'an 1980 coordinate system, and check the consistency of the spatial matching between all the environmental data and the soil sample point attributes, and connect the spatial attributes after correctness. Through the administrative division map of Jian'ou City, Using the vector clipping function in ArcGIS10.2 software, the surface (0-15cm) soil of cultivated land in Jian'ou City was obtained. Specifically include the following steps:

S31, remote sensing data preprocessing. It mainly includes radiometric calibration, atmospheric correction, geometric precision correction, and image mosaic. For the Sentinel-2A remote sensing images of the four scenes in the study area, the SNAP software was used to perform radiometric calibration on each scene and then perform atmospheric correction. The DN value of the remote sensing image was converted into the surface reflectance, and then the images were mosaiced and stitched, and then in ENVI5.3 The software uses the finely corrected SPOT5 reference image to perform geometric fine correction, and cuts it according to the administrative division map of Jian'ou City, Fujian Province. The Landsat-8OLI remote sensing image is a geometrically corrected image, and the atmospheric correction is carried out in the ENVI5.3 software. The error of geometric fine correction of all remote sensing images is controlled within 1 pixel, and they are uniformly projected into the Xi'an 1980 coordinate system.

S32, preprocessing of terrain variables. First, after checking that the DEM data with a spatial resolution of 12.5 meters and 30 meters are correct, in the ArcGIS10.2 software, they are uniformly converted to the Xi’an 1980 coordinate system through projection transformation, and the 12.5-meter DEM is produced as a spatial resolution using the nearest neighbor method. DEM data with a rate of 10 m.

S33, preprocessing of meteorological variables: use arcgis10.2 software to uniformly project the meteorological data to the Xi'an 1980 coordinate system.

S34. In the ArcGIS10.2 software, check the consistency of the attribute spatial matching of the above remote sensing data, terrain data, meteorological data and soil sample point data, and establish a geographical connection between the spatial data and the soil attribute after the check is correct.

S4, extraction and screening of environment variables. The predictor variables selected in this study include remote sensing variables, topographical variables, meteorological variables and soil pH. In order to further simplify the model input, the auxiliary predictors need to be optimized before modeling. The variables with significant correlation with soil available phosphorus were retained by Pearson correlation analysis, and the variables were screened by the backward elimination method according to the increase or decrease of the root mean square error (RMSE) after each variable was excluded from the model in turn during modeling. If the RMSE increases, the variables will be retained; otherwise, the auxiliary variables that will eventually participate in the modeling will be retained. Specifically include the following steps:

S41, extraction of remote sensing variables. In ArcGIS10.2 software, remote sensing variables were extracted from Sentinel-2A and Landsat-8OLI to calculate 46 spectral indices for digital soil mapping.

S42, extraction of terrain variables. Eight topographic variables commonly used in digital soil mapping were extracted from the resampled 10-meter resolution DEM and 30-meter DEM in ArcGIS10.2 software.

S43, extraction of meteorological variables. Calculate the average monthly precipitation and temperature data for many years in ArcGIS10.2 software, and then use the grid calculator to calculate the average monthly average precipitation and average temperature for many years in the study area, and obtain the annual average temperature and temperature of the study area for many years precipitation data. Finally, the nearest neighbor method was used to resample the annual average precipitation and annual average temperature data to raster data with a spatial resolution of 10 meters and 30 meters.

S44, obtaining the spatial distribution map of soil pH. In the ArcGIS10.2 software, the Kriging interpolation method was used to interpolate the soil pH in the study area to obtain its spatial distribution map.

S45, in ArcGIS10.2 software, geographically correlate the soil sample points with the above remote sensing variables, terrain variables, meteorological variables and soil pH and other spatial data to obtain all the environmental variables corresponding to the soil sample points in the study area.

S46, screening of environmental variables involved in modeling. Perform Pearson correlation analysis on the Python platform to screen the variables that have a significant correlation with soil available phosphorus, and further use the backward elimination method to eliminate the root mean square error (RMSE) of each variable in turn after the modeling. Variables are screened, if the RMSE increases, the variables are retained, otherwise, the auxiliary variables that participate in the modeling are retained, and finally the sample set that participates in the modeling is obtained. The auxiliary variables that participate in the modeling are shown in Table 1.

Table 1 Auxiliary variables for soil available phosphorus modeling

S5, model construction and accuracy evaluation. Taking the soil available phosphorus content of 96 cultivated land surfaces (0-15cm) in Jian'ou City as the dependent variable, and the medium and high resolution remote sensing variables, terrain variables, meteorological variables and soil pH value screened above as the independent variables, the Python platform based on The medium and high resolution environmental variables were respectively executed with the random forest model calculation, and the random forest model with the best prediction accuracy and the importance analysis of the variables were automatically selected, and the surface soil of cultivated land in Jian'ou City in 2017 (0 -15cm) prediction model of available phosphorus content; and evaluate the prediction accuracy of the model. Specifically include the following steps:

S51. Construct a training sample set and a verification sample set. The above sample set is randomly divided into a training set and a verification set according to 9:1, wherein the samples of the training set are used for modeling, and the samples of the verification set are used to test the prediction accuracy of the model.

S52, determining dependent variables and independent variables of the model. Taking the soil available phosphorus content of 96 cultivated land surfaces (0-15cm) in Jian’ou City as the dependent variable, the high-resolution (10m) remote sensing variables and high-resolution (10m) Terrain variables, meteorological variables, and soil pH were used as independent variables, and the random forest algorithm was executed on the Python platform to automatically screen the best random value of the available phosphorus content in the surface soil (0-15cm) of cultivated land in Jian'ou City in 2017 based on high-precision environmental variables. Forest Prediction Model.

S53, evaluation of model prediction accuracy. The best prediction model is screened by R ² (coefficient of determination), MAE (mean absolute error) and RMSE (root mean square error). The closer R ² is to 1, the smaller MAE and RMSE indicate the higher accuracy of the prediction model. The accuracy evaluation formula (1-3) is:

为实测值平均值。In the formula, n is the number of sampling points, O _i and P _i are the measured and predicted values of sampling point i,

is the average value of the measured values.

Based on high-precision (10 meters) environmental variables, the prediction accuracy index R ² of the soil available phosphorus content in the surface layer (0-15cm) of cultivated land in Jian'ou City is 0.59, the MAE is 19.04mg·kg ^-1 , and the RMSE error is 25.26mg·kg ^-1 .

S54, taking the soil available phosphorus content of 96 cultivated land surfaces (0-15cm) in Jian’ou City as the dependent variable, and the remote sensing variables of the moderate resolution (30 meters) and the remote sensing variables of the medium resolution (30 meters) Topographic variables, meteorological variables, and soil pH values are used as independent variables, and the same steps as S51 are repeated to obtain the best available phosphorus content in surface soil (0-15cm) of cultivated land in Jian'ou City in 2017 based on commonly used medium-resolution environmental variables. Random Forest Prediction Model.

Based on the commonly used environmental variables with medium resolution (30 meters), the prediction accuracy index R ² of soil available phosphorus content in the surface layer (0-15cm) of cultivated land in Jian'ou City is 0.42, the MAE error is 21.84 mg·kg ^-1 , and the RMSE error is 30.18 mg·kg ^-1 .

S55, obtaining the relative importance scores of variables in the process of predicting soil available phosphorus content by taking the average value of the RF model for multiple iterations. The relative importance of variables based on environmental variable modeling with a resolution of 10m showed that the relative importance scores of meteorological variables, topographical variables, remote sensing variables, and soil pH were 30.64%, 30.38%, 22.87%, and 16.11%, respectively. According to the importance of individual variables, they are annual average temperature, pH value, topographic humidity index, DEM value, annual average rainfall, enhanced vegetation index, first principal component and red edge band6.

The relative importance of variables based on environmental variables with a spatial resolution of 30 meters showed that the relative importance scores of meteorological variables, topographical variables, remote sensing variables, and soil pH were 25.86%, 32.44%, 21.31%, and 20.39%, respectively. In descending order of importance, they are elevation (DEM value), pH value, average annual rainfall, enhanced vegetation index, B5 and the first principal component.

S6, method effect evaluation. Based on the environmental variables with a spatial resolution of 30 meters, the modeling of soil available phosphorus in the top layer of cultivated land (0-15cm) in Jian'ou City in 2017 is used as a reference standard, and the three indicators of R ² , MAE and RMSE are used to analyze the proposed method based on this patent. The accuracy of high-resolution (10m) environmental variables to predict the soil available phosphorus content in the surface layer (0-15cm) of cultivated land in Jian'ou City in 2017 was evaluated. The results show that the precision based on 10-meter remote sensing data, 12.5-meter topographic data combined with other environmental variables is significantly improved compared with the commonly used 30-meter remote sensing data, 30-meter topographic data combined with other variables, and the modeling accuracy R2 is increased by 40.5% ( From 0.42 to 0.59), the MAE error decreased by 12.8% (from 21.84mg·kg ^-1 to 19.04mg·kg ^-1 ), the RMSE error decreased by 16.3% (from 30.18mg·kg ^-1 to 25.26mg·kg ^{- 1} ). Specific steps include:

S61, constructing method improvement effect indicators. Based on the environmental variables with a spatial resolution of 30 meters, the modeling of soil available phosphorus in the cultivated land surface (0-15cm) of Jian'ou City in 2017 is used as a reference. From the three indicators of R ² , MAE and RMSE, the high Environmental variables with a resolution (10 meters) can predict the accuracy of soil available phosphorus content in the surface layer (0-15cm) of cultivated land in Jian'ou City in 2017 to evaluate the effect. The formula of method improvement effect P is:

P＝(X _10m- X _30m )/X _30m *100％

Among them, P represents the ratio of R2 improvement or the ratio of MAE and RMSE error reduction, ^X10m represents the accuracy of predicting soil available phosphorus content in the surface layer (0-15cm) of cultivated land in southern hilly areas based on high-resolution (10m) environmental variables, X30 M represents the accuracy of predicting soil available phosphorus content in the surface layer (0-15cm) of cultivated land in southern hilly areas based on the routinely used 30-meter environmental variable.

S62. Calculate the R ² improvement effect. According to the above formula, the modeling accuracy R ² calculated in the excel table is increased by 40.5% (from 0.42 to 0.59).

S63. Calculate the ratio of error reduction. According to the above formula, calculate the reduction ratio of MAE error and RMSE error in the excel table. The MAE error decreased by 12.8% (from 21.84 mg·kg ^-1 to 19.04 mg·kg ^-1 ), and the RMSE error decreased by 16.3% (from 30.18 mg·kg ^-1 to 25.26 mg·kg ^-1 ).

S64, the results are shown in Table 2. In terms of the three evaluation indicators, the combination of 10-meter remote sensing data and 12.5-meter topographic data combined with other environmental variables is better than the commonly used 30-meter remote sensing data and 30-meter topographic data combined with other variables. The prediction accuracy is significantly improved.

Table 2 Improvement evaluation results of this method

S7, predicting the available phosphorus content in the surface soil (0-15cm) of all cultivated land in Jian'ou City in 2017. Based on the best model constructed with high-resolution (10 meters) environmental variables and the input of high-precision (10 meters) environmental variables, the random forest algorithm is executed in Python software to predict the surface soil of all cultivated land in Jian'ou City in 2017 (0- 15cm) available phosphorus content, and the spatial distribution map of available phosphorus in all cultivated land surface soil (0-15cm) in Jian'ou City in 2017 was obtained, as shown in Figure 2.

The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still be described in the foregoing embodiments Modifications are made to the recorded technical solutions, or equivalent replacements are made to some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (8)

1. A method for mapping soil available phosphorus in cultivated land in southern hilly areas based on high-resolution environmental variables, characterized in that it comprises the following steps: S1. Acquisition of soil sample point data and building a database: obtain the physical and chemical attribute data of the surface soil sample points of the cultivated land in the area to be tested, and establish a soil sample point database in ArcGIS10.2 software; S2. Acquisition of environmental data: Acquire the environmental data of the area at the same time as the soil sampling point, including medium and high resolution remote sensing variables, medium and high resolution terrain data and meteorological data; S3. Data preprocessing: project the remote sensing data, terrain data and meteorological data to the Xi’an 1980 coordinate system, and check the consistency of the spatial matching between the environmental data and the soil sample point attributes, and connect the spatial attributes after correctness. Administrative division map, use the vector clipping function in ArcGIS10.2 software to get the cultivated land surface soil in the area to be measured; S4. Extraction and screening of environmental variables: The selected predictive variables include remote sensing variables, topographical variables, meteorological variables and soil pH value; in order to simplify the model input, the auxiliary predictors are optimized before modeling, and the relationship between soil and soil is retained through Pearson correlation analysis. For the variables with significant correlation with available phosphorus, the auxiliary variables that finally participate in the modeling are retained through the backward elimination method; S5. Model construction and accuracy evaluation: the soil available phosphorus content of the preset number of cultivated land surfaces in the area to be measured is used as the dependent variable, and the medium and high resolution remote sensing variables, topographic variables, meteorological variables and soil pH value screened above are used as independent variables On the Python platform, the random forest model calculation is performed based on the medium and high resolution environmental variables, and the random forest model with the best prediction accuracy and its variable importance analysis are automatically selected, and the sampling time of the soil sample points based on the medium and high resolution environmental variables are respectively obtained The predictive model of the available phosphorus content in the surface soil of the cultivated land in the area to be tested in the same period; S6. Predict the available phosphorus content of the surface soil (0-15cm) of all cultivated land in the area to be tested at the same time as the sampling time of the soil sample points: the best model constructed according to the high-resolution (10m) environmental variables and input high-precision (10m) m) environment variable, execute the random forest algorithm in Python software, predict the available phosphorus content of all cultivated land surface soil (0-15cm) in the same area to be measured, and obtain all the cultivated land surface soil (0-15cm) in the area to be measured in this period Spatial distribution map of available phosphorus. 2. A method for mapping the available phosphorus in cultivated land soil in southern hilly areas based on high-resolution environmental variables according to claim 1, wherein S1 specifically comprises the following steps: S11: Study the physical and chemical attribute data of the preset number of cultivated land surface soil samples in the area to be measured, which comes from the data of the cultivated land quality inspection and evaluation sample points of the Ministry of Agriculture and Rural Affairs, including soil available phosphorus, soil pH value, geographical coordinates, and elevation; S12: According to the geographical coordinates and physical and chemical attribute data of the soil samples, the soil sample points database was established in ArcGIS10.2 software, and the spatial distribution map of the soil samples under the projection of Xi'an 1980 coordinate system was obtained. 3. according to claim 2, a method for mapping soil available phosphorus in arable land in southern hilly areas based on high-resolution environmental variables, is characterized in that, in S11, the soil pH value adopts the acidity meter method to measure, and the soil available phosphorus adopts sodium bicarbonate Extraction-molybdenum antimony anti-colorimetric method. 4. A kind of method for mapping available phosphorus in cultivated land soil in southern hilly areas based on high-resolution environmental variables according to claim 1, characterized in that, the terrain data described in S2 includes high-resolution (12.5 meters) DEM and commonly used DEMs with a spatial resolution of 30 meters for digital soil mapping; The remote sensing data includes 4 high-resolution (10 meters) Sentinel-2A remote sensing images and Landsat-8OLI images with a spatial resolution of 30 meters commonly used in digital soil mapping; The meteorological data are 1km×1km grid data of historical monthly average precipitation and monthly average temperature in the area to be measured. 5. A method for mapping the available phosphorus in cultivated land soil in southern hilly areas based on high-resolution environmental variables according to claim 1, wherein S3 specifically comprises the following steps: S31. Remote sensing data preprocessing: remote sensing data preprocessing includes radiometric calibration, atmospheric correction, geometric fine correction, and image mosaic; for the Sentinel-2A remote sensing images of the 4 scenes in the study area, the SNAP software performs radiometric calibration for each scene separately. Atmospheric correction, the DN value of the remote sensing image is converted into the surface reflectance, and then the image is mosaiced and stitched, and then the finely corrected SPOT5 reference image is used in the ENVI5.3 software to perform geometric fine correction, and according to the administrative division of the area to be measured The image is cropped; the Landsat-8OLI remote sensing image is a geometrically corrected image, and the atmospheric correction is performed in the ENVI5.3 software. The geometrically corrected error of all remote sensing images is controlled within 1 pixel, and they are uniformly projected into the Xi'an 1980 coordinate system; S32. Preprocessing of terrain variables: firstly, after checking that the DEM data with spatial resolutions of 12.5m and 30m are correct, in ArcGIS10.2 software, they are uniformly converted to the Xi’an 1980 coordinate system through projection transformation, and the 12.5 The DEM of m is produced as DEM data with a spatial resolution of 10m; S33. Preprocessing of meteorological variables: using arcgis10.2 software to project the meteorological data into the Xi'an 1980 coordinate system; S34. In the ArcGIS10.2 software, check the consistency of the attribute spatial matching of the above remote sensing data, terrain data, meteorological data and soil sample point data, and establish a geographical connection between the spatial data and the soil attribute after the inspection is correct. 6. A method for mapping the available phosphorus in cultivated land soil in southern hilly areas based on high-resolution environmental variables according to claim 1, characterized in that the backward elimination method in S4 is specifically: The variables were screened according to the increase or decrease of the root mean square error after each variable was excluded from the model in turn during the modeling. If the RMSE increased, the variables were retained, and vice versa. 7. A method for mapping available soil phosphorus in arable land in southern hilly areas based on high-resolution environmental variables according to claim 6, wherein S4 specifically includes the following steps: S41. Extraction of remote sensing variables: extracting remote sensing variables from Sentinel-2A and Landsat-8OLI in ArcGIS10.2 software to calculate 46 spectral indices for digital soil mapping; S42. Extraction of topographic variables: 8 topographic variables commonly used in digital soil mapping are extracted from the resampled 10-meter resolution DEM and 30-meter DEM in ArcGIS10.2 software; S43. Extraction of meteorological variables: Calculate the average monthly precipitation and average temperature data for many years in ArcGIS10.2 software, and then use the grid calculator to calculate the average value of the average monthly precipitation and average temperature for many years in the study area to obtain the study area Years of average annual temperature and average annual precipitation data, and finally use the nearest neighbor method to resample the annual average precipitation and annual average temperature data to raster data with a spatial resolution of 10 meters and 30 meters; S44. Acquisition of the spatial distribution map of soil pH: use the Kriging interpolation method in the ArcGIS10.2 software to obtain the soil pH in the study area for interpolation to obtain its spatial distribution map; S45, in the ArcGIS10.2 software, the soil sample points are geographically correlated with the above remote sensing variables, terrain variables, meteorological variables and soil pH data, and all environmental variables corresponding to the soil sample points in the study area are obtained; S46. Screening of environmental variables involved in modeling: Perform Pearson correlation analysis on the Python platform to screen variables that have a significant correlation with soil available phosphorus. Through the backward elimination method, each variable is excluded from the model in turn according to the mean square of the model. The increase or decrease of the root error (RMSE) screens the variables, the RMSE increases and the variables are retained, otherwise, the auxiliary variables that are finally involved in the modeling are retained, and finally the sample set that participates in the modeling is obtained. 8. A method for mapping available soil phosphorus in arable land in southern hilly areas based on high-resolution environmental variables according to claim 7, wherein S5 specifically comprises the following steps: S51. Constructing a training sample set and a verification sample set: randomly divide the above sample set into a training set and a verification set in proportion, wherein the samples of the training set are used for modeling, and the samples of the verification set are used for testing the prediction accuracy of the model; S52. Determine the dependent variable and independent variable of the model: use the soil available phosphorus content of the cultivated land surface of the soil sample point in the area to be measured as the dependent variable, and use the screened significant correlation high-resolution (10 meters) remote sensing variables, high-resolution The topographic variables, meteorological variables and soil pH value of the rate (10 meters) are used as independent variables, and the random forest algorithm is executed on the Python platform to automatically screen the effective soil surface soil of the cultivated land in the area to be measured based on the high-precision environmental variables at the same time as the sampling time of the soil samples. Phosphorus content prediction model; S53, model prediction accuracy evaluation: filter the best prediction model by R ² (coefficient of determination), MAE (mean absolute error) and RMSE (root mean square error), the closer R ² is to 1, the smaller the MAE and RMSE are, indicating that the prediction The higher the accuracy of the model; the accuracy evaluation formula (1-3) is:

In the formula, n is the number of sampling points, O _i and P _i are the measured and predicted values of sampling point i,

is the average value of the measured value; S54. Taking the soil available phosphorus content in the surface layer (0-15cm) of the cultivated land of the soil sample point as the dependent variable, and using the remote sensing variable with a moderate resolution (30 meters) and the terrain variable with a medium resolution (30 meters) that are significantly related to the screening , Meteorological variables and soil pH value as independent variables, repeat the same steps as S51, and obtain the best available phosphorus content of cultivated land surface soil (0-15cm) based on the medium-resolution environmental variables and the sampling time of the soil sample point at the same time. random forest prediction model; S55. Obtain the relative importance score of variables in the process of predicting soil available phosphorus content by taking the average value of multiple iterations of the RF model; The relative importance of variables based on the modeling of environmental variables with a resolution of 10m shows that the relative importance scores of meteorological variables, topographical variables, remote sensing variables, and soil pH are 30.64%, 30.38%, 22.87%, and 16.11%; The order of importance of individual variables from large to small is annual average temperature, pH value, topographic humidity index, DEM value, annual average rainfall, enhanced vegetation index, first principal component and red edge band6; The relative importance of variables based on environmental variables with a spatial resolution of 30 meters shows that the relative importance scores of meteorological variables, topographical variables, remote sensing variables and soil pH are 25.86%, 32.44%, 21.31% and 20.39% respectively; The order of importance from large to small is elevation (DEM value), pH value, average annual rainfall, enhanced vegetation index, B5 and the first principal component.

Images