WFV-based spatial distribution mapping method for cation exchange capacity of cultivated soil

Abstract

The invention relates to a method for mapping the spatial distribution of cation exchange capacity of cultivated land soil based on WFV. The invention belongs to the field of satellite remote sensing technology and its application, and specifically relates to a method for mapping the spatial distribution of cation exchange capacity of cultivated land soil. The invention is to solve the problem that the existing remote sensing mapping of soil CEC content mostly uses foreign satellites. The method comprises the following steps: obtaining WFV images of the target study area, obtaining surface reflectances of different central wavelengths through data preprocessing; inputting the input variables of the random forest algorithm of the invention (annual average rainfall, annual average temperature, elevation, DI3, DVI, RDVI3, GDVI1), and outputting the cation exchange capacity of cultivated land soil (CEC, cmol / kg), so as to realize the spatial distribution map of the cation exchange capacity of soil in the target study area. The method of the invention can realize the dynamic satellite remote sensing monitoring of soil cation exchange capacity at the regional scale. After verification of the content of soil cation exchange capacity of field samples, the method of the invention has good accuracy and applicability.

Description

Technical Field

[0001] The invention belongs to the field of satellite remote sensing technology and its application, and in particular relates to a method for mapping the spatial distribution of cation exchange capacity of cultivated soil. Background Art

[0002] Soil cation exchange capacity (CEC) is highly correlated with soil solid phase composition and is the sum of various cations adsorbed in the soil. It can reflect the soil buffering cation nutrients (such as NH4 + etc.) and the ability to retain water and fertilizer. Due to the universal exchange and adsorption of soil CEC, CEC is the main storage form of soluble effective cations in the soil. When cationic nutrients are adsorbed on colloids, it is soil fertilizer retention, that is, cationic nutrient retention, and the opposite process is fertilizer supply. Early studies found that CEC has a strong correlation with soil organic matter content and clay content, and increases with the increase of organic matter and clay content, and farmland soil has a stronger ability to retain water and fertilizer. Therefore, soil CEC is used as one of the important evaluation indicators for soil resource evaluation, reasonable and scientific fertilization, soil improvement and soil management. The traditional soil CEC determination method is time-consuming and laborious, requiring field soil sample collection and laboratory determination. With the development of earth observation technology, satellite remote sensing observation methods combined with machine learning algorithms have the advantages of low cost and pollution-free. At present, remote sensing mapping of soil CEC content mostly uses foreign satellites, and this method requires a lot of time and cost through field sampling and indoor experimental analysis, while there is little research on the application of domestic high-resolution series satellites in digital soil mapping. Therefore, the research and development of CEC content prediction based on the reflectance spectrum of domestically produced GF-1 satellite images has important research significance and application value. Summary of the invention

[0003] The purpose of the present invention is to solve the technical problems that the existing soil CEC content remote sensing mapping mostly uses foreign satellites, and the time and cost required for field sampling and indoor experimental analysis are relatively large. A WFV method for mapping the spatial distribution of cultivated soil cation exchange capacity is provided.

[0004] The spatial distribution mapping method of WFV's cultivated soil cation exchange capacity is as follows:

[0005] 1. Obtain the WFV sensor image data of the Gaofen-1 satellite (GF-1);

[0006] Second, the acquired WFV sensor image data of the Gaofen-1 satellite is preprocessed, including atmospheric correction, orthorectification, and radiometric calibration, and finally the surface reflectance of the central wavelength of different bands of the WFV sensor is obtained;

[0007] 3. Obtain geographic resource and environmental data, including annual average temperature with a spatial resolution of 1000m, annual average precipitation with a spatial resolution of 1000m, and DEM elevation data with a spatial resolution of 30m. The annual average temperature and annual average precipitation data with a spatial resolution of 1000m are downloaded from the National Meteorological Center (http: / / data.cma.cn / ), and the DEM elevation data with a spatial resolution of 30m is obtained from USGS (https: / / gdex.cr.usgs.gov / gdex / ). The spatial resolution of the above raster data is different from that of the 16mWFV sensor image data and needs to be resampled to 16m.

[0008] 4. Based on the random forest algorithm, input variables: average annual rainfall, average annual temperature, elevation, DI3, DVI, RDVI3, GDVI1;

[0009] Random forest algorithm output variables: soil cation exchange capacity (CEC), unit: cmol / kg;

[0010] The spectral index

[0011] DI3=Band4-Band1

[0012] DVI=Band4-Band3

[0013]

[0014]

[0015] Band1 represents the surface reflectance of band 1 of the WFV sensor image;

[0016] Band2 represents the surface reflectance of band 2 of the WFV sensor image;

[0017] Band3 represents the surface reflectance of band 3 of the WFV sensor image;

[0018] Band4 represents the surface reflectance of band 4 of the WFV sensor image;

[0019] 5. The method of mapping the spatial distribution of soil cation exchange capacity is to substitute the prediction model into the WFV sensor image of the Gaofen-1 satellite to realize the spatial distribution mapping of regional soil cation exchange capacity.

[0020] The input variables of the random forest algorithm described in step 4 are model building based on the measured soil cation exchange capacity, and the construction process is as follows:

[0021] a. During the bare soil period, a total of 81-164 soil samples from the 0-20 cm surface layer of cultivated land in the main grain-producing areas of the Songnen Plain were collected, and the GPS locations of the soil samples were recorded. The collected soil samples were air-dried, ground and sieved in the laboratory, and the soil cation exchange capacity was obtained using the sodium acetate-ammonium acetate exchange-ICP method, in cmol / kg;

[0022] b. After obtaining the WFV sensor image of the Gaofen-1 satellite, extract the corresponding remote sensing reflectivity according to the GPS position of the collected soil samples;

[0023] c. Obtain environmental covariate data: average annual rainfall, average annual temperature, elevation, and resample to 16m;

[0024] d. Based on the measured soil cation exchange capacity data and remote sensing reflectance, a random forest inversion model of soil cation exchange capacity was established:

[0025] The measured cation exchange capacity of 81-164 soil samples and the corresponding remote sensing reflectance data were selected as the calibration data set of the model. The linear correlation between the remote sensing reflectance, spectral index, environmental covariate data of Band1-Band4 and the measured soil cation exchange capacity content data was analyzed. Finally, the input variables were selected as the average annual rainfall, average annual temperature, elevation, DI3, DVI, RDVI3, and GDVI1. The selected variables were used as the input variables of the random forest algorithm to estimate the soil cation exchange capacity, and a linear fit was performed on the measured cation exchange capacity of 81-164 soil samples to establish a soil cation exchange capacity model.

[0026] The linear correlation analysis was implemented in SPSS 16.0 software platform, and the Pearson correlation coefficient r was expressed as follows:

[0027]

[0028] Among them, x i is the measured value of soil cation exchange capacity, y i is the remote sensing reflectance of the WFV sensor band that matches the measured value of soil cation exchange capacity, and N is the number of samples;

[0029] e. The random forest algorithm requires linear fitting error analysis of the measured values ​​and predicted values, and the expression is as follows:

[0030]

[0031] Where RMSE is the root mean square error, N is the number of samples, and y i is the measured value of soil cation exchange capacity, y i ' is the estimated soil cation exchange capacity;

[0032] f. For the calibration data set, the algorithm linearly fits R 2 =0.71-0.79, RMSE is 3.98 cmol / kg-5.55 cmol / kg.

[0033] The random forest inversion algorithm described in step 4 uses the RandomForest package, the number of decision trees ntree is 500, and the number of random variables of the split nodes mtry is 1 / 3 of the input. At this time, the value is small and the error in the model is basically stable.

[0034] Based on the spatial distribution mapping method of cultivated soil cation exchange capacity of WFV of the present invention, the estimation of cultivated soil cation exchange capacity content and spatial distribution mapping can be carried out. The present invention constructs a random forest algorithm for cultivated soil cation exchange capacity, which is based on field sampling and indoor experimental analysis data of soil samples in Songnen Plain. The field collection samples cover a large range and can cover different soil types, agronomic measures and different geographical environment and climate backgrounds in Songnen Plain. The algorithm has good applicability and can achieve high-precision estimation and mapping of soil cation exchange capacity, avoiding the time and cost required by traditional methods through field sampling and indoor experimental analysis, obtaining spatial distribution maps of large-scale areas and different time periods, and ensuring effective soil fertility monitoring and cultivated land quality assessment.

[0035] The method of the present invention can realize dynamic satellite remote sensing monitoring of soil cation exchange capacity at a regional scale. Verification of the content of soil cation exchange capacity by field samples shows that the method of the present invention has good accuracy and applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is the distribution map of sampling points of measured soil samples in the algorithm of the present invention;

[0037] Figure 2 It is the calibration graph of the random forest algorithm model of soil cation exchange capacity;

[0038] Figure 3 This is the validation diagram of the random forest algorithm model for soil cation exchange capacity;

[0039] Figure 4 It is a schematic diagram of the steps of the method for mapping the spatial distribution of cation exchange capacity of cultivated soil of WFV of the present invention. DETAILED DESCRIPTION

[0040] The technical solution of the present invention is not limited to the specific implementation modes listed below, but also includes any combination of the specific implementation modes.

[0041] Specific implementation method 1: The spatial distribution mapping method of cultivated soil cation exchange capacity of WFV in this implementation method is as follows:

[0042] 1. Obtain the WFV sensor image data of the Gaofen-1 satellite (GF-1);

[0043] Second, the acquired WFV sensor image data of the Gaofen-1 satellite is preprocessed, including atmospheric correction, orthorectification, and radiometric calibration, and finally the surface reflectance of the central wavelength of different bands of the WFV sensor is obtained;

[0044] 3. Obtain geographic resource and environmental data, including annual average temperature with a spatial resolution of 1000m, annual average precipitation with a spatial resolution of 1000m, and DEM elevation data with a spatial resolution of 30m. The annual average temperature and annual average precipitation data with a spatial resolution of 1000m are downloaded from the National Meteorological Center (http: / / data.cma.cn / ), and the DEM elevation data with a spatial resolution of 30m is obtained from USGS (https: / / gdex.cr.usgs.gov / gdex / ). The spatial resolution of the above raster data is different from that of the 16mWFV sensor image data and needs to be resampled to 16m.

[0045] 4. Based on the random forest algorithm, input variables: average annual rainfall, average annual temperature, elevation, DI3, DVI, RDVI3, GDVI1;

[0046] Random forest algorithm output variables: soil cation exchange capacity (CEC), unit: cmol / kg;

[0047] The spectral index

[0048] DI3=Band4-Band1

[0049] DVI=Band4-Band3

[0050]

[0051]

[0052] Band1 represents the surface reflectance of band 1 of the WFV sensor image;

[0053] Band2 represents the surface reflectance of band 2 of the WFV sensor image;

[0054] Band3 represents the surface reflectance of band 3 of the WFV sensor image;

[0055] Band4 represents the surface reflectance of band 4 of the WFV sensor image;

[0056] 5. The method of mapping the spatial distribution of soil cation exchange capacity is to substitute the prediction model into the WFV sensor image of the Gaofen-1 satellite to realize the spatial distribution mapping of regional soil cation exchange capacity.

[0057] Specific implementation method 2: This implementation method is different from the specific implementation method 1 in that the input variable of the random forest algorithm described in step 4 is a model built based on the measured soil cation exchange capacity and the construction process is as follows:

[0058] a. During the bare soil period, a total of 81-164 soil samples from the 0-20 cm surface layer of cultivated land in the main grain-producing areas of the Songnen Plain were collected, and the GPS locations of the soil samples were recorded. The collected soil samples were air-dried, ground and sieved in the laboratory, and the soil cation exchange capacity was obtained using the sodium acetate-ammonium acetate exchange-ICP method, in cmol / kg;

[0059] b. After obtaining the WFV sensor image of the Gaofen-1 satellite, extract the corresponding remote sensing reflectivity according to the GPS position of the collected soil samples;

[0060] c. Obtain environmental covariate data: average annual rainfall, average annual temperature, elevation, and resample to 16m;

[0061] d. Based on the measured soil cation exchange capacity data and remote sensing reflectance, a random forest inversion model of soil cation exchange capacity was established:

[0062] The measured cation exchange capacity of 81-164 soil samples and the corresponding remote sensing reflectance data were selected as the calibration data set of the model. The linear correlation between the remote sensing reflectance, spectral index, environmental covariate data of Band1-Band4 and the measured soil cation exchange capacity content data was analyzed. Finally, the input variables were selected as the average annual rainfall, average annual temperature, elevation, DI3, DVI, RDVI3, and GDVI1. The selected variables were used as the input variables of the random forest algorithm to estimate the soil cation exchange capacity, and a linear fit was performed on the measured cation exchange capacity of 81-164 soil samples to establish a soil cation exchange capacity model.

[0063] The linear correlation analysis was implemented in SPSS 16.0 software platform, and the Pearson correlation coefficient r was expressed as follows:

[0064]

[0065] Among them, x i is the measured value of soil cation exchange capacity, y i is the remote sensing reflectance of the WFV sensor band that matches the measured value of soil cation exchange capacity, and N is the number of samples;

[0066] e. The random forest algorithm requires linear fitting error analysis of the measured values ​​and predicted values, and the expression is as follows:

[0067]

[0068] Where RMSE is the root mean square error, N is the number of samples, and y i is the measured value of soil cation exchange capacity, y i ' is the estimated soil cation exchange capacity;

[0069] f. For the calibration data set, the algorithm linearly fits R 2 =0.71-0.79, RMSE is 3.98 cmol / kg-5.55 cmol / kg.

[0070] The rest is the same as the first specific implementation mode.

[0071] Specific implementation method three: This implementation method is different from specific implementation method one or two in that the random forest inversion algorithm described in step four uses the RandomForest package, the number of decision trees ntree is 500, and the number of random variables of the split nodes mtry is 1 / 3 of the input amount, at which time the value is small and the error within the model is basically stable. The rest is the same as specific implementation method one or two.

[0072] The following experiments were used to verify the effect of the present invention:

[0073] Experiment 1;

[0074] The spatial distribution mapping method of WFV's cultivated soil cation exchange capacity is as follows:

[0075] 1. Obtain the WFV sensor image data of the Gaofen-1 satellite (GF-1);

[0076] Second, the acquired WFV sensor image data of the Gaofen-1 satellite is preprocessed, including atmospheric correction, orthorectification, and radiometric calibration, and finally the surface reflectance of the central wavelength of different bands of the WFV sensor is obtained;

[0077] 3. Obtain geographic resource and environmental data, including annual average temperature with a spatial resolution of 1000m, annual average precipitation with a spatial resolution of 1000m, and DEM elevation data with a spatial resolution of 30m. The annual average temperature and annual average precipitation data with a spatial resolution of 1000m are downloaded from the National Meteorological Center (http: / / data.cma.cn / ), and the DEM elevation data with a spatial resolution of 30m is obtained from USGS (https: / / gdex.cr.usgs.gov / gdex / ). The spatial resolution of the above raster data is different from that of the 16mWFV sensor image data and needs to be resampled to 16m.

[0078] 4. Based on the random forest algorithm, input variables: average annual rainfall, average annual temperature, elevation, DI3, DVI, RDVI3, GDVI1;

[0079] Random forest algorithm output variables: soil cation exchange capacity (CEC), unit: cmol / kg;

[0080] The spectral index

[0081] DI3=Band4-Band1

[0082] DVI=Band4-Band3

[0083]

[0084]

[0085] Band1 represents the surface reflectance of band 1 of the WFV sensor image;

[0086] Band2 represents the surface reflectance of band 2 of the WFV sensor image;

[0087] Band3 represents the surface reflectance of band 3 of the WFV sensor image;

[0088] Band4 represents the surface reflectance of band 4 of the WFV sensor image;

[0089] 5. The method of mapping the spatial distribution of soil cation exchange capacity is to substitute the prediction model into the WFV sensor image of the Gaofen-1 satellite to realize the spatial distribution mapping of regional soil cation exchange capacity.

[0090] The following experimental steps are used to verify the effect of the present invention:

[0091] a. In the bare soil period of 2021 (end of April), a total of 81 soil samples with a depth of 0-20 cm were collected from the main grain-producing areas of the Songnen Plain, and the GPS locations of the soil samples were recorded ( Figure 1 ), the collected soil samples were air-dried, ground and sieved in the laboratory, and the soil cation exchange capacity was obtained using the sodium acetate-ammonium acetate exchange-ICP method (in cmol / kg);

[0092] b. After obtaining the WFV sensor image of the Gaofen-1 satellite, extract the corresponding remote sensing reflectivity according to the GPS position of the collected soil samples;

[0093] c. Obtain environmental covariate data: average annual rainfall, average annual temperature, elevation, and resample to 16m;

[0094] d. Calculate the spectral indexes DI3, DVI, RDVI3, and GDVI1;

[0095] e. According to the random forest algorithm of soil cation exchange capacity established in this patent, the variables such as average annual rainfall, average annual temperature, elevation, DI3, DVI, RDVI3, GDVI1, etc. are input. The random forest inversion algorithm uses the RandomForest package, the number of decision trees (ntree) is 500, and the number of random variables for splitting nodes (mtry) is 1 / 3 of the input amount;

[0096] f. For the validation data set, the algorithm linearly fits R 2 =0.71, RMSE is 5.55cmol / kg, which meets the internationally accepted quantitative remote sensing inversion accuracy requirements (>45%). Figure 3 .

[0097] The random forest inversion algorithm described in step 4 uses the RandomForest package, the number of decision trees ntree is 500, and the number of random variables of the split nodes mtry is 1 / 3 of the input. At this time, the value is small and the error in the model is basically stable.

Claims

1. A method for mapping the spatial distribution of cation exchange capacity of cultivated soil based on WFV, characterized in that The spatial distribution mapping method of the WFV cultivated soil cation exchange capacity is as follows:

1. Obtain the WFV sensor image data of the Gaofen-1 satellite; Second, the acquired WFV sensor image data of the Gaofen-1 satellite is preprocessed, including atmospheric correction, orthorectification, and radiometric calibration, and finally the surface reflectance of the central wavelength of different bands of the WFV sensor is obtained; 3. Obtain geographic resource and environmental data, including annual average temperature with a spatial resolution of 1000m, annual average rainfall with a spatial resolution of 1000m, and DEM elevation data with a spatial resolution of 30m. The annual average temperature and annual average rainfall data with a spatial resolution of 1000m are downloaded from the National Meteorological Center, and the DEM elevation data with a spatial resolution of 30m is obtained from USGS. The spatial resolution of the above raster data is different from that of the 16m WFV sensor image data and needs to be resampled to 16m.

4. Based on the random forest algorithm, input variables: average annual rainfall, average annual temperature, elevation, DI3, DVI, RDVI3, GDVI1; Random forest algorithm output variables: soil cation exchange capacity, unit cmol / kg; Said DI3=Band4-Band1 DVI=Band4-Band3 Band1 represents the surface reflectance of band 1 of the WFV sensor image; Band3 represents the surface reflectance of band 3 of the WFV sensor image; Band4 represents the surface reflectance of band 4 of the WFV sensor image; 5. The method of mapping the spatial distribution of soil cation exchange capacity is to substitute the prediction model into the WFV sensor image of the Gaofen-1 satellite to realize the spatial distribution mapping of regional soil cation exchange capacity.

2. The method for mapping the spatial distribution of cation exchange capacity of cultivated soil by WFV according to claim 1, characterized in that The input variables of the random forest algorithm described in step 4 are model construction trained based on the measured soil cation exchange capacity, and the construction process is as follows: a. During the bare soil period, a total of 81-164 soil samples from the 0-20 cm surface layer of cultivated land in the main grain-producing areas of the Songnen Plain were collected, and the GPS locations of the soil samples were recorded. The collected soil samples were air-dried, ground and sieved in the laboratory, and the soil cation exchange capacity was obtained using the sodium acetate-ammonium acetate exchange-ICP method, in cmol / kg; b. After obtaining the WFV sensor image of the Gaofen-1 satellite, extract the corresponding remote sensing reflectivity according to the GPS position of the collected soil samples; c. Obtain environmental covariate data: average annual rainfall, average annual temperature, elevation, and resample to 16m; d. Based on the measured soil cation exchange capacity data and remote sensing reflectance, a random forest inversion model of soil cation exchange capacity was established: The measured cation exchange capacity of 81-164 soil samples and the corresponding remote sensing reflectance data were selected as the calibration data set of the model. The linear correlation between the remote sensing reflectance, spectral index, environmental covariate data of Band1-Band4 and the measured soil cation exchange capacity content data was analyzed. Finally, the input variables were selected as the average annual rainfall, average annual temperature, elevation, DI3, DVI, RDVI3, and GDVI1. The selected variables were used as the input variables of the random forest algorithm to estimate the soil cation exchange capacity, and a linear fit was performed on the measured cation exchange capacity of 81-164 soil samples to establish a soil cation exchange capacity model. The linear correlation analysis was implemented in SPSS 16.0 software platform, and the Pearson correlation coefficient r was expressed as follows: Among them, x i is the measured value of soil cation exchange capacity, y i is the remote sensing reflectance of the WFV sensor band that matches the measured value of soil cation exchange capacity, and N is the number of samples; e. The random forest algorithm requires linear fitting error analysis of the measured values ​​and predicted values, and the expression is as follows: Where RMSE is the root mean square error, N is the number of samples, and y i is the measured value of soil cation exchange capacity, y i ' is the estimated soil cation exchange capacity; f. For the calibration data set, the algorithm linearly fits R 2 =0.71-0.79, RMSE is 3.98 cmol / kg-5.55 cmol / kg.

3. The method for mapping the spatial distribution of cation exchange capacity of cultivated land soil according to WFV of claim 1, characterized in that The random forest inversion algorithm described in step 4 uses the RandomForest package, the number of decision trees ntree is 500, and the number of random variables of the split nodes mtry is 1 / 3 of the input. At this time, the value is small and the error in the model is basically stable.

Images