Soil organic matter digital mapping method and device based on crop rotation historical change

Abstract

The invention discloses a soil organic matter digital mapping method and device based on crop rotation historical change, and the method comprises the steps: bringing crop rotation interannual change and crop rotation change frequency into an environment variable set, and carrying out the screening and combination with other environment covariables; the historical crop rotation change and the long-term frequency thereof are brought into the environment variable set, and the environment variable set is enriched, so that the influence of human activities on soil spatial-temporal heterogeneity is considered in soil organic matter prediction, and the precision of model prediction is improved. By utilizing the method, the influence of historical agricultural practice can be reflected, and the spatial heterogeneity of the farmland soil organic matters can be better described, so that the performance of a soil organic matter spatial prediction model is improved.

Description

Technical Field

[0001] This invention pertains to methods for predicting soil organic matter content, and particularly relates to a method and apparatus for digital mapping of soil organic matter based on crop rotation history. Background Technology

[0002] Arable land, as the foundation of food production, is crucial for food security, ecological balance, and economic development. Soil, as a key participant in the carbon and other element cycles of the global ecosystem, directly determines the productivity of arable land through its fertility and structure. Soil organic matter stores abundant organic carbon, which can characterize the differences in fertility among different arable lands and the availability of soil nutrients. Accurate soil organic matter mapping is of great significance for providing spatiotemporal pattern information on farmland soil quality, optimizing soil fertility management, and improving agricultural production efficiency. Traditional soil monitoring mainly relies on field surveys and laboratory analysis, followed by soil mapping based on expert experience—a time-consuming, labor-intensive, and costly process. With the development of remote sensing technology, geographic information systems (GIS), geostatistics, and computer modeling techniques, large-scale soil information mapping is becoming increasingly efficient and economical. Digital soil mapping based on soil-landscape models has gained widespread acceptance, and numerous scholars have explored its application in soil organic matter mapping.

[0003] Patent application CN116206011A discloses a digital soil mapping method and system based on multi-source data. The method includes acquiring soil use types and satellite remote sensing image raster data of the area to be measured; dividing the area into blocks to obtain multiple soil regions, with each soil use type including at least one soil region; identifying all soil regions to obtain the topographic features of each soil region and the distribution characteristics of soil regions of the same soil use type based on the identification results; determining whether soil regions of the same soil use type are adjacent; if not, acquiring sample points for each soil unit and obtaining sample data for each sample point to perform soil mapping. This patent application makes the acquired sample data more representative and improves the prediction accuracy of the prediction model by dividing the area to be measured into multiple soil regions and then determining the sample points and corresponding sample data for each soil region. However, since this patent application identifies the topographic features and distribution characteristics of soil regions, the obtained sample data is not comprehensive; therefore, the prediction accuracy of the prediction model needs further improvement.

[0004] Patent application CN120147468A discloses a method, apparatus, device, and medium for predicting and mapping soil organic matter based on deep learning, relating to the field of digital soil mapping. The method includes acquiring remote sensing images of soil in a target area and setting a preset number of soil sampling points; comprehensively acquiring environmental data indicators such as topographic factors, vegetation factors, climate factors, and land use factors; constructing three-dimensional environmental tensor data within a preset range for each sampling point; inputting the three-dimensional environmental tensor data into a soil organic matter content prediction model to obtain a set of predicted organic matter content values ​​for soil sampling points within the target area; and combining the predicted value set with the spatial coordinates of the soil sampling points to generate a spatial distribution prediction map of soil organic matter content in the target area. This application improves the accuracy and mapping efficiency of soil organic matter prediction, providing technical support for soil resource management and sustainable agricultural development. However, the sample data in this patent application only comes from environmental data indicators, which is incomplete and affects the accuracy of organic matter content prediction.

[0005] Therefore, there is an urgent need to develop a new digital soil mapping method to better reflect the impact of human activities on the spatiotemporal heterogeneity of soil and improve the prediction accuracy of the model. Summary of the Invention

[0006] This invention provides a method for digital mapping of soil organic matter based on the historical changes in crop rotation. This method can reflect the impact of historical agricultural practices, better characterize the spatial heterogeneity of farmland soil organic matter, and thus improve the performance of soil organic matter spatial prediction models.

[0007] This invention provides a method for digital mapping of soil organic matter based on crop rotation history, comprising: Based on the annual crop rotation spatial distribution map of the selected region, the interannual variation and frequency of crop rotation in consecutive years within a set time period are extracted, and an environmental variable set is constructed based on the interannual variation, frequency of crop rotation and environmental covariates. The feature selection algorithm is used to select multiple environmental covariates from the set of environmental variables that result in the highest accuracy of the random forest model. Various machine learning models are used to model soil organic matter based on the selected environmental covariates. The best soil organic matter prediction model is determined according to the model accuracy. The selected environmental covariates are then input into the best soil organic matter prediction model to obtain the spatial distribution of soil organic matter content.

[0008] Preferably, the method for obtaining the interannual variation of crop rotation in consecutive years within a set time period includes: superimposing the spatial distribution map of crop rotation for each year in chronological order, and extracting the spatial distribution variation of crop rotation patterns in consecutive years within the set time period pixel by pixel, thereby obtaining the interannual variation of crop rotation in consecutive years within the set time period.

[0009] Preferably, the method for obtaining the frequency of crop rotation changes within a set time period includes: overlaying the spatial distribution map of crop rotation for each year in chronological order, calculating the frequency of crop rotation changes in the selected area within a set time period pixel by pixel, thereby obtaining the frequency of crop rotation changes.

[0010] Preferably, the method for obtaining a spatial distribution map of crop rotation for each year in a selected area includes: Based on the crop rotation system of the selected area, the main crop types are obtained. Cloudless remote sensing images of the selected area in mid-March of each year are obtained. The differences between different crops on real color images and standard color composite images are compared. The random forest algorithm is used to classify crops and obtain the spatial distribution map of crop rotation for each year.

[0011] Preferably, a feature selection algorithm is used to select the top K most important environmental covariates from the set of environmental variables, including: S1. Fit a random forest model to the set of environmental variables, and evaluate the importance of each environmental variable by increasing the mean squared error by a percentage; select the most important environmental variable to fit the initial model, and use k-fold cross-validation to calculate the model's prediction accuracy; S2. Combine the most important environmental variables and the remaining environmental variables selected in the previous step in order, fit the model and calculate the model accuracy under different combinations with the remaining environmental variables. S3. Update the most important combination of environmental variables using the environmental variables from the model with the best accuracy selected in the previous step. S4. Repeat steps S2 and S3 until the model accuracy reaches the set threshold. The obtained environmental variables are the filtered environmental variables.

[0012] Preferably, the various machine learning models include random forest, Cubist, XGBoost, and support vector machine.

[0013] Preferably, the environmental covariates include Sentinel-1 synthetic aperture radar remote sensing variables, Sentinel-2 multispectral remote sensing variables, soil indices, topographic variables, location variables, and in-situ spectral data.

[0014] Preferably, the Sentinel-2 multispectral remote sensing variables include the normalized vegetation index, the enhanced vegetation index, and the transformed vegetation index; The Sentinel-1 synthetic aperture radar remote sensing variables include radar indices calculated based on dual-polarization data of SAR images. The soil indices include clay content, silt content, sand content, brightness index, carbonate content index, gypsum index, color index, iron-containing minerals, hue index, saturation index, pressure-related index, reflectance absorption index, and redness index. The topographic variables include elevation, slope, and aspect; The location variables include the nearest distance to the river and the grid size within 30. ◦ , 60 ◦ , 120 ◦ 150 ◦ oblique geographic coordinates; The in-situ spectral data includes the first 10 principal components of the soil spectrum extracted using principal component analysis.

[0015] Preferably, the model accuracy is determined by evaluation metrics for machine learning accuracy, wherein the evaluation metrics for machine learning accuracy are the root mean square error (RMSE) and the coefficient of determination (R²). 2 .

[0016] The present invention also provides a soil organic matter digital mapping device based on crop rotation history changes, characterized in that it includes: a memory and one or more processors, wherein the memory stores executable code, and the one or more processors execute the executable code to implement the soil organic matter digital mapping method based on crop rotation history changes.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention incorporates the interannual variation and frequency of crop rotation into the environmental variable set, and filters and combines them with other environmental covariates. By including historical crop rotation variations and their frequencies into the environmental variable set, the set of environmental variables is enriched. This allows the impact of human activities on soil spatiotemporal heterogeneity to be considered in the prediction of soil organic matter, thereby improving the accuracy of model predictions. Attached Figure Description

[0018] Figure 1 A flowchart of a soil organic matter digital mapping method based on crop rotation history changes is provided in an embodiment of the present invention; Figure 2 The scatter plot of measured and predicted values ​​of soil organic matter content in the validation set provided in the embodiments of the present invention represents the accuracy of the spatial prediction model of soil organic matter estimated by the present invention. Figure 3 This is a distribution map of surface soil organic matter content in Jiashan area obtained from an embodiment of the present invention. Detailed Implementation

[0019] The present invention will be further described and illustrated below with reference to the accompanying drawings and specific embodiments.

[0020] The type and timing of crop planting significantly influence the spatiotemporal dynamics of soil organic matter in farmland. Digital soil mapping using crop rotation as an environmental variable characterizing human activities can better reflect the impact of human activities on soil spatiotemporal heterogeneity, thus improving prediction accuracy. The mapping method provided in this application incorporates interannual variations and frequencies of crop rotation into a set of environmental variables, enabling the capture of the impact of crop rotation dynamics on soil organic matter. It can identify and analyze the cumulative effects and long-term trends of crop rotation, thereby improving the prediction accuracy of soil organic matter content.

[0021] The soil organic matter digital mapping method based on crop rotation history changes provided by this invention, such as... Figure 1 As shown, it includes: Step S1: In a specific embodiment of the present invention, based on the annual crop rotation spatial distribution map of the selected region, the interannual variation of crop rotation and the frequency of crop rotation variation in consecutive years within a set time period are extracted, and an environmental variable set is constructed based on the interannual variation of crop rotation, the frequency of crop rotation variation, and environmental covariates.

[0022] This invention incorporates the interannual variation and frequency of crop rotation over a set period of consecutive years into an environmental variable set. This means that the long-term trend and frequency of crop rotation are included in the environmental variable set. Since planting different crops and how crops are rotated have a significant impact on soil organic matter content and the soil environment, including the long-term trend and frequency of crop rotation in the environmental variable set improves the completeness of the variables included in the set. This facilitates the more accurate selection of environmental variables that contribute more to soil organic matter content, thereby improving the prediction accuracy of soil organic matter.

[0023] The specific steps of step S1 provided in the specific embodiment of the present invention are as follows: S11. Collect surface soil samples and determine soil organic matter content and in-situ spectra: Topsoil samples (0-20 cm) from fallow land in the testing area were collected. Geographic coordinates were recorded using a handheld GPS device, and soil samples were collected using a soil auger. Topsoil from three sub-samples were randomly collected within a 10 m radius of each sample point, thoroughly mixed, and sealed in sample collection bags, each weighing approximately 1 kg. In-situ spectra were measured using an ASD FieldSpec 4 visible-near-infrared spectrometer. Reflectance spectra were acquired at three different locations for each sub-sample, and the average of the nine soil spectra from the three sub-samples was calculated to represent the in-situ spectrum of each soil sampling point. All soil samples were air-dried and ground, and passed through a 2 mm sieve. Soil organic matter content was determined using the dichromate oxidation-external heating method.

[0024] S12. Obtain historical changes in crop rotation, including: In specific embodiments of the present invention, key information such as the main crop types, crop growth and development stages, and dominant crop rotation systems in the selected region are collected. It is understood that each region will formulate its main crop planting and crop rotation systems. In one embodiment, the crops in this embodiment are the crops in the crop rotation system of the selected region.

[0025] Specifically, a method for obtaining an annual crop rotation spatial distribution map of a selected region according to a specific embodiment of the present invention includes: Based on field surveys of selected areas, the main crop types were identified. Cloud-free remote sensing images of the selected areas were obtained each year. The differences between different crops on real color images and standard color composite images were compared. The random forest algorithm was used to classify crops and obtain the spatial distribution map of crop rotation each year.

[0026] Specifically, a method for obtaining the interannual variation of crop rotation in consecutive years within a set time period according to a specific embodiment of the present invention includes: superimposing the spatial distribution map of crop rotation for each year in chronological order, and extracting the spatial distribution variation of crop rotation patterns in consecutive years within a set time period pixel by pixel, thereby obtaining the interannual variation of crop rotation in consecutive years within a set time period.

[0027] Specifically, a method for obtaining the frequency of crop rotation changes within a set time period according to a specific embodiment of the present invention includes: superimposing the spatial distribution map of crop rotation for each year in chronological order, calculating the frequency of crop rotation changes within a set time period for a selected area pixel by pixel, thereby obtaining the frequency of crop rotation changes.

[0028] In one embodiment, this embodiment combines auxiliary imagery data from Google Earth to form prior knowledge of crop rotation and builds sample labels for crops. Based on cloudless remote sensing images from late March each year, the differences between different crops on true-color images and standard false-color composite images are compared, and a random forest algorithm is used for classification to obtain the annual crop rotation distribution. The spatial distribution maps of crop rotation for each year are superimposed sequentially, and the interannual variations of crop rotation over three, four, and five years are extracted pixel by pixel. The frequency of crop rotation changes over three, four, and five years is calculated to represent the intensity of human agricultural activities. The interannual variation of crop rotation and the frequency of crop rotation changes constitute historical crop rotation changes. The interannual variation of crop rotation refers to the spatial distribution changes of rotation patterns over different time periods, such as the change in the rotation system of a certain pixel in consecutive years. The frequency of crop rotation changes refers to the frequency of these changes, which is a quantification of the intensity of human agricultural activities.

[0029] S13. The specific embodiments of the present invention provide for obtaining other environmental covariates, including Sentinel-1 synthetic aperture radar remote sensing variables, Sentinel-2 multispectral remote sensing variables, soil indices, topographic variables, location variables, and in-situ spectral data. Sentinel-1, Sentinel-2, and the Space Shuttle Radar Topography Mission (SRTM) products are all from the Google Earth Engine platform.

[0030] Specifically, the Sentinel-1 synthetic aperture radar remote sensing variables provided in this embodiment include radar indices calculated based on SAR image dual-polarization data; the soil indices include clay content, silt content, sand content, brightness index, carbonate content index, gypsum index, color index, iron-bearing minerals, hue index, saturation index, pressure-related index, reflectance absorption index, and redness index; the topographic variables include elevation, slope, and aspect; and the location variables include the nearest distance to the river and the grid size within 30. ◦ , 60 ◦ , 120 ◦ 150 ◦ The oblique geographic coordinates; the in-situ spectral data includes the first 10 principal components of the soil spectrum extracted using principal component analysis.

[0031] S14. The environmental coordination provided in the specific embodiments of the present invention is unified into the CGCS WGS 1984 geographic coordinate system, and resampled to a set spatial resolution using the bilinear method. In one specific embodiment, the set spatial resolution is 10 m.

[0032] Step S2: In a specific embodiment of the present invention, a feature selection algorithm is used to select multiple environmental variables from the set of environmental variables that result in the highest model accuracy. The feature selection algorithm is used to reduce the redundancy of the prediction model and improve the computational efficiency of spatial prediction of soil organic matter. The specific steps are as follows: S21. Fit a random forest model to the set of environmental variables, and evaluate the importance of each environmental variable by the percentage increase in mean squared error (%IncMSE); select the most important environmental variable to fit the initial model, and use k-fold cross-validation to calculate the model's prediction accuracy. S22. Combine the most important environmental variables and the remaining environmental variables selected in the previous step in order, fit the model and calculate the model accuracy under different combinations with the remaining environmental variables. S23. Use the environmental variables from the model with the best accuracy selected in the previous step to update the most important combination of environmental variables, and increase the number of environmental variables in the fitted model from 3 to n. S24. Repeat steps S22 and S23 until the model accuracy no longer improves. The obtained environmental variables are the filtered environmental variables.

[0033] Step S3: Using multiple machine learning models, soil organic matter is modeled based on multiple screened environmental variables to obtain multiple soil organic matter prediction models. Based on the model accuracy, the best soil organic matter prediction model is determined from the multiple soil organic matter prediction models.

[0034] Specifically, the machine learning models provided in this embodiment employ random forest, Cubist, XGBoost, and support vector machine.

[0035] Specifically, model accuracy is determined using machine learning accuracy evaluation metrics, namely the root mean square error (RMSE) and the coefficient of determination (R²). 2 The details are as follows: in, n The number of sample data. and For the first i Observed and predicted values ​​for each sample. It is the average of all observed values.

[0036] The present invention also provides a soil organic matter digital mapping device based on crop rotation history changes, comprising: a memory and one or more processors, wherein the memory stores executable code, and the one or more processors execute the executable code to implement the soil organic matter digital mapping method based on crop rotation history changes.

[0037] This invention analyzes the role of historical crop rotation changes in soil organic matter digital mapping, identifying the environmental covariates that contribute most to soil organic matter modeling. This reflects the impact of human activities on soil spatiotemporal heterogeneity while improving the accuracy of model predictions. It helps to understand and utilize past agricultural management practices, enabling us to better study farmland soil quality, thereby improving agricultural productivity and supporting sustainable development.

[0038] Example 1 In this embodiment, Jiashan County, Jiaxing City, in northeastern Zhejiang Province, my country, was selected as the study area. Using data from 202 surface soil samples collected in 2023, and employing a soil organic matter digital mapping method based on crop rotation history, the spatial distribution of surface soil organic matter content with a spatial resolution of 10 m was finally obtained. The basic steps of this method are as described in steps (1) to (5) of the aforementioned embodiment, and will not be repeated in detail. The following mainly presents the specific data and implementation details: Step (1): Collect topsoil samples from fallow land after wheat or rapeseed harvest and before the next planting.

[0039] Step (2): While conducting field sampling, key information such as the main crop types, crop growth and development stages, and dominant crop rotation systems in the study area over the past five years was systematically collected. The main crop rotation systems in Xitang Town are wheat-rice and rapeseed-rice. Rice is planted from late May to early June each year, and wheat or rapeseed is planted after the rice is harvested in November, and harvested the following May. Therefore, the focus is mainly on these two crop rotation patterns, and other non-dominant crop rotation methods are classified as "other crop rotation systems". Analyzing the growth cycle and phenological characteristics of wheat and rapeseed in the study area, it was found that rapeseed is in the flowering stage from March to early April each year, while wheat is in the jointing stage where leaf area is rapidly increasing. The two crops show obvious differences in remote sensing images. In true-color remote sensing images, rapeseed flowers appear yellowish-green, wheat appears dark green, and other farmland appears pinkish-purple. In standard false-color composite images with enhanced vegetation features, rapeseed appears light pink, wheat appears red, and other farmland appears grayish-green. Based on cloudless remote sensing images from late March each year, the differences between different crops on true-color images and standard false-color composite images are compared. Random forest algorithm is used for classification to obtain the crop rotation distribution for each year. The interannual variation of crop rotation and the frequency of crop rotation variation are further extracted to form historical crop rotation variation.

[0040] Step (3): The system collected other environmental covariates, including Sentinel-1 synthetic aperture radar remote sensing variables, Sentinel-2 multispectral remote sensing variables, topographic variables, location variables, and in-situ spectral data. Sentinel-1, Sentinel-2, and the Space Shuttle Radar Topography Mission (SRTM) products were all from the Google Earth Engine platform (https: / / earthengine.google.com / ).

[0041] Optical imagery was acquired using the Sentinel-2 Level-2A surface reflectance (SR) product (COPERNICUS / S2_SR), which underwent atmospheric and orthorectification preprocessing. Sentinel-2 SR images with cloud cover less than 20% were selected, and monthly or quarterly median composites were performed based on the data acquisition time. Vegetation indices, including Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), and Transformed Vegetation Index (TVI), were calculated using the median composite Sentinel-2 SR images as optical predictors. Bare soil pixels from all available Sentinel-2 SR images were extracted using the Two-Dimensional Bare Soil Separation (TDBSS) framework. Ten soil indices, including Brightness Index (BI), Hue Index (HI), and Saturation Index (SI), were then calculated using the median composite images of all bare soil pixels (Table 1). Some existing digital soil product results, such as clay content, silt content, and sand content, have also been added to the soil factors.

[0042] SAR images were obtained from C-band Sentinel-1 Level-1 Ground Range Detected (GRD) products (COPERNICUS / S1_GRD). The data was processed using dual-polarization data (Vertical transmit Vertical receiving (VV) and Vertical transmit Horizontal receiving (VH)), and thermal noise cancellation, radiometric calibration, and terrain correction were performed. Multi-year median composites were created using Sentinel-1 GRD data from May of each year. Based on the median data, 17 radar indices were calculated: 5 predictive features including VV / VH, VH / VV, VH–VV, (VH–VV) / (VH+VV), and (VH+VV) / 2. Additionally, 12 features were calculated using the Gray-Level Co-occurrence Matrix (GLMC).

[0043] Elevation data was obtained from STRM, and slope and aspect were calculated using SAGAGIS software as topographic factors. The nearest distance to the river was calculated pixel-by-pixel, and the grid was then analyzed using latitude and longitude coordinates within a 30-kilometer radius. ◦ , 60 ◦ , 120 ◦ 150 ◦ The oblique geographic coordinates.

[0044] Step (4): The importance of environmental covariates is evaluated by %IncMSE in the random forest model. %IncMSE is calculated by the percentage increase in the mean squared error (MSE) of the model when the value of a certain feature is randomly shuffled (i.e., permuted). The higher the %IncMSE, the greater the contribution of the variable to the model's prediction.

[0045] Step (5): Divide the sample dataset into a 7:3 ratio and train each machine learning model using 10-fold cross-validation to ensure the robustness and reliability of the model. Based on the coefficient of determination (R²), 2 The results of feature selection and root mean square error (RMSE) analysis showed that the model based on the Cubist model achieved the highest accuracy. 2 The value was 0.85, and the RMSE was 4.23 g kg⁻¹. The Cubist model is implemented in Python using the Cubist function from the cubist library. Step (6): The distribution map of soil organic matter content is drawn in Python using the rasterio library, and the final TIFF format file is output through the rasterio.open function.

[0046] After variable selection in this implementation, the number of environmental covariates used for modeling is nine, reducing model complexity, avoiding interference from irrelevant variables on model predictive performance, enabling the model to run more efficiently, and reducing computation time and memory usage. The evaluation accuracy of the optimal soil organic matter digital mapping model based on crop rotation history changes in this embodiment is as follows: Figure 2 As shown, R 2 The value was 0.85 and the RMSE was 4.23 g kg⁻¹, indicating good predictive performance. Jiashan County, Zhejiang Province, my country, was selected as the study area. The spatial distribution of topsoil organic matter content predicted in this example is as follows: Figure 3 As shown, the spatial resolution is 10 m.

[0047] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained through equivalent substitution or transformation fall within the protection scope of the present invention.

Claims

1. A method for digital mapping of soil organic matter based on crop rotation history, characterized in that, include: Based on the annual crop rotation spatial distribution map of the selected region, the interannual variation and frequency of crop rotation in consecutive years within a set time period are extracted, and an environmental variable set is constructed based on the interannual variation, frequency of crop rotation and environmental covariates. The feature selection algorithm is used to select multiple environmental covariates from the set of environmental variables that make the accuracy of the random forest model higher than a set threshold. Multiple machine learning models are used to model soil organic matter based on the selected environmental covariates. The best soil organic matter prediction model is determined according to the model accuracy. The selected environmental covariates are input into the best soil organic matter prediction model to obtain the spatial distribution of soil organic matter content.

2. The method for digital mapping of soil organic matter based on crop rotation history changes according to claim 1, characterized in that, A method for obtaining the interannual variation of crop rotation in consecutive years within a set time period includes: overlaying the spatial distribution map of crop rotation for each year in chronological order, extracting the spatial distribution variation of crop rotation patterns in consecutive years within the set time period pixel by pixel, thereby obtaining the interannual variation of crop rotation in consecutive years within the set time period.

3. The method for digital mapping of soil organic matter based on crop rotation history changes according to claim 1, characterized in that, A method for obtaining the frequency of crop rotation changes within a set time period includes: overlaying the spatial distribution map of crop rotation for each year in chronological order, calculating the frequency of crop rotation changes in the selected area pixel by pixel within the set time period, thereby obtaining the frequency of crop rotation changes.

4. The method for digital mapping of soil organic matter based on crop rotation history changes according to claim 1, characterized in that, Methods for obtaining spatial distribution maps of crop rotation in selected regions each year include: Based on the crop rotation system of the selected area, the main crop types are obtained. Cloudless remote sensing images of the selected area in mid-March of each year are obtained. The differences between different crops on real color images and standard color composite images are compared. The random forest algorithm is used to classify crops and obtain the spatial distribution map of crop rotation for each year.

5. The method for digital mapping of soil organic matter based on crop rotation history changes according to claim 1, characterized in that, Feature selection algorithms were used to select the top K most important environmental covariates from the set of environmental variables, including: S1. Fit a random forest model to the set of environmental variables, and evaluate the importance of each environmental variable by increasing the mean squared error by a percentage; select the most important environmental variable to fit the initial model, and use k-fold cross-validation to calculate the model's prediction accuracy; S2. Combine the most important environmental variables and the remaining environmental variables selected in the previous step in order, fit the model and calculate the model accuracy under different combinations with the remaining environmental variables. S3. Update the most important combination of environmental variables using the environmental variables from the model with the best accuracy selected in the previous step. S4. Repeat steps S2 and S3 until the model accuracy is higher than the set accuracy threshold. The obtained environmental variables are the filtered environmental variables.

6. The method for digital mapping of soil organic matter based on crop rotation history changes according to claim 1, characterized in that, The various machine learning models mentioned include random forest, Cubist, XGBoost, and support vector machine.

7. The method for digital mapping of soil organic matter based on crop rotation history changes according to claim 1, characterized in that, The environmental covariates include Sentinel-1 synthetic aperture radar remote sensing variables, Sentinel-2 multispectral remote sensing variables, soil indices, topographic variables, location variables, and in-situ spectral data.

8. The method for digital mapping of soil organic matter based on crop rotation history changes according to claim 7, characterized in that, The Sentinel-2 multispectral remote sensing variables include the normalized vegetation index, enhanced vegetation index, and transformed vegetation index. The Sentinel-1 synthetic aperture radar remote sensing variables include radar indices calculated based on dual-polarization data of SAR images. The soil indices include clay content, silt content, sand content, brightness index, carbonate content index, gypsum index, color index, iron-containing minerals, hue index, saturation index, pressure-related index, reflectance absorption index, and redness index. The topographic variables include elevation, slope, and aspect; The location variables include the nearest distance to the river and the grid size within 30. ◦ , 60 ◦ , 120 ◦ 150 ◦ oblique geographic coordinates; The in-situ spectral data includes the first 10 principal components of the soil spectrum extracted using principal component analysis.

9. The method for digital mapping of soil organic matter based on crop rotation history changes according to claim 1, characterized in that, Model accuracy is determined using machine learning accuracy evaluation metrics, namely the root mean square error (RMSE) and the coefficient of determination (R²). 2 .

10. A digital mapping device for soil organic matter based on crop rotation history changes, characterized in that, include: The system includes a memory and one or more processors, wherein the memory stores executable code, and the one or more processors execute the executable code to implement the soil organic matter digital mapping method based on crop rotation history changes as described in any one of claims 1-9.

Images