Multi-source environment covariable driven soil organic matter remote sensing inversion method and device

Abstract

The invention discloses a multi-source environment covariable-driven soil organic matter remote sensing inversion method and device, and belongs to the technical field of agricultural remote sensing and soil monitoring. A multi-spectral remote sensing image is obtained, topographic data and climate data are combined, and after atmospheric correction, cloud shadow removal and resampling preprocessing, the multi-spectral remote sensing image is obtained; calculating a normalized difference index, a ratio index and wave band transformation characteristics; screening multi-source environment covariables significantly related to organic matters, constructing an inversion model based on an XGBoost algorithm, optimizing hyper-parameters through grid search, and realizing high-precision prediction of the soil organic matter content; according to the method, the problems that a traditional method is low in efficiency, high in cost and insufficient in model generalization ability are solved, and technical support is provided for cultivated land quality evaluation and precise agricultural management.

Description

Technical Field

[0001] The present invention belongs to the interdisciplinary field of agricultural remote sensing technology and soil science, and specifically relates to a method and device for rapid inversion of soil organic matter content based on multi-source environmental covariates (including satellite remote sensing images, terrain, climate data, etc.) and machine learning algorithms. The method is suitable for dynamic monitoring of cultivated land quality, precision agriculture management, and sustainable utilization of land resources. Background Art

[0002] Existing soil organic matter determination relies on field sampling and laboratory analysis (such as the potassium dichromate oxidation method), which suffers from low efficiency (a single sample test cycle of 3-5 days), high cost (a single sample costs approximately US$50), and insufficient spatial coverage. The shortcomings of existing remote sensing inversion technology are as follows:

[0003] Data uniformity: Most studies rely on a single remote sensing data source (such as Sentinel-2A or Landsat-8) without integrating environmental covariates such as topography and climate, resulting in insufficient model explanatory power (R 2 Generally below 0.7).

[0004] Algorithm limitations: Traditional algorithms (such as random forest and partial least squares) have limited performance in modeling complex spectrum-environment relationships and lack efficient feature selection mechanisms.

[0005] Poor timeliness: Localized data processing tools (such as ENVI) take weeks and are difficult to support high-frequency dynamic monitoring.

[0006] How to achieve low-cost, large-scale, real-time soil organic matter monitoring while ensuring accuracy through multi-source data fusion and advanced algorithm design is a difficult problem that urgently needs to be broken through in this field. Summary of the Invention

[0007] The purpose of the present invention is to provide a soil organic matter remote sensing inversion method and device driven by multi-source environmental covariates, so as to solve the problems of insufficient data fusion, weak model generalization ability and low mapping efficiency in the existing technology.

[0008] The purpose of the present invention is achieved through the following technical solutions:

[0009] A soil organic matter remote sensing inversion method driven by multi-source environmental covariates includes the following steps:

[0010] (1) Obtain Landsat-8 satellite multi-band remote sensing image data of the target area through the Google Earth Engine platform;

[0011] (2) preprocessing the remote sensing image data, including atmospheric correction, radiometric calibration, orthorectification, and cloud shadow removal;

[0012] (3) Obtain environmental covariate data of the target area, including 30 m resolution digital elevation model (DEM), slope and aspect data, and annual average temperature and annual average rainfall climate data, and resample all environmental covariate data to 30 m resolution;

[0013] (4) Collect surface soil samples during the bare soil period in the target area, record the latitude and longitude coordinates of the sampling points, and measure the soil organic matter content;

[0014] (5) Based on the preprocessed remote sensing image data, the spectral indices of normalized difference index (NDI), ratio index (RI), and difference index (DI) are calculated;

[0015] (6) Inputting the spectral index of step (5), the environmental covariate data of step (3), and the soil organic matter content data of step (4) into the XGBoost algorithm to construct a soil organic matter inversion model;

[0016] (7) The trained XGBoost model is used to invert the soil organic matter content in the target area and generate a 30m resolution soil organic matter spatial distribution map.

[0017] As a more optimal technical solution of the present invention, the spatial resolution of the climate data in step (3) is 1 km, and it is resampled to 30 m resolution by bilinear interpolation.

[0018] As a more optimal technical solution of the present invention, the spectral index in step (5) also includes a derivative feature generated by performing an inverse transformation or an exponential transformation on the original band reflectivity.

[0019] As a more optimal technical solution of the present invention, the soil sample collection in step (4) adopts a five-point sampling method, and each sampling point is composed of a mixture of sub-samples at the four corners and the center of the sample plot.

[0020] As a more optimal technical solution of the present invention, the hyperparameters of the XGBoost algorithm in step (6) are optimized by a grid search method, including the learning rate (η), the maximum depth of the tree (max_depth) and the regularization coefficient (λ).

[0021] As a more optimal technical solution of the present invention, the spatial distribution map of soil organic matter generated in step (7) is visualized and dynamically updated in the cloud through the Google Earth Engine platform.

[0022] As a more preferred technical solution of the present invention, the method is suitable for inversion of soil organic matter at a depth of 0-20 cm in the surface layer of cultivated land. As a more preferred technical solution of the present invention, the diameter a of the riser is 15 mm-110 mm.

[0023] The soil organic matter remote sensing inversion device driven by multi-source environmental covariates provided by the present invention comprises:

[0024] Data acquisition module, including a satellite remote sensing data interface that supports access to multiple sources of data such as Landsat-8 and Sentinel-2, and an environmental covariate database that integrates DEM, climate, land use type and other data;

[0025] Data processing module, including a spectral index calculation unit with built-in 20 index algorithms such as NDI, RI, and DI, and a feature selection engine that uses the Pearson analysis algorithm to achieve automated feature screening;

[0026] Modeling and inversion modules, including a machine learning module based on the XGBoost framework that supports model training, validation, and optimization, and a real-time mapping unit that leverages GEE cloud computing capabilities to achieve minute-level regional mapping;

[0027] The output and interaction module includes a visual interface that provides organic matter spatial distribution maps, statistical reports, and agricultural management suggestions, as well as an API interface that supports data exchange with agricultural Internet of Things systems (such as smart fertilization equipment).

[0028] As a more optimal technical solution of the present invention, the feature selection engine screens at least 26 key variables through the Pearson analysis algorithm, including DEM, slope, aspect, average annual temperature, average annual precipitation and derived spectral characteristics.

[0029] As a more optimal technical solution of the present invention, the real-time mapping unit can generate a 30m resolution soil organic matter spatial distribution map of the target area within 24 hours, and is linked to the intelligent fertilization equipment through the API interface.

[0030] The beneficial effects are as follows:

[0031] This paper proposes a remote sensing inversion method that can efficiently and accurately invert soil organic matter content by integrating Landsat-8 satellite remote sensing images, multi-source environmental covariate data, and the XGBoost machine learning algorithm. This method not only improves the efficiency of obtaining information on the spatial distribution of soil organic matter, but also provides large-scale, dynamic soil quality monitoring support for agricultural production, helping agricultural managers to scientifically formulate fertilization and tillage strategies. Through this innovative remote sensing and algorithm fusion technology, monitoring costs can be significantly reduced, prediction accuracy can be improved, and the precise management and sustainable utilization of cultivated land resources can be promoted. Test set R 2 It reached 0.80, which is significantly higher than the random forest (R 2 =0.72) and partial least squares method (R 2=0.74). Mapping cycles have been shortened from several months to within 24 hours, supporting dynamic monitoring. Regional mapping costs have been reduced to 1 / 20 of traditional laboratory analysis. Compatible with satellite data such as Sentinel-2 and GF-5, it is applicable to different soil types. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is a flow chart of the inversion method of the present invention;

[0033] Figure 2 shows the prediction results of soil organic matter content by RF, SVM, and XG-Boost machine learning algorithms; RF is shown in Figure 2(a), SVM is shown in Figure 2(b), and XG-Boost is shown in Figure 2(c);

[0034] Figure 3 It is the spatial distribution map of soil organic matter content. DETAILED DESCRIPTION

[0035] The present invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0036] like Figure 1 As shown, the present invention proposes a rapid inversion method for soil organic matter content based on Landsat-8 satellite remote sensing imagery and the XGBoost algorithm, aiming to solve the problems of over-exploitation of soil resources and decline in soil fertility. This method makes full use of the Google Earth Engine platform to obtain multi-temporal remote sensing data, and combines pre-processing methods such as atmospheric correction and cloud shadow removal to extract a variety of spectral indices such as normalized difference index and ratio index, while introducing environmental covariate data such as terrain and climate data. Organic matter content data is obtained based on the collected soil sample information, and a high-precision training sample library is constructed in combination with the sampling point information. A satellite remote sensing inversion model is established through the XGBoost algorithm to achieve large-scale rapid mapping of soil organic matter content. Compared with the traditional method that relies on field sampling and experimental analysis, this method has higher spatial coverage capability and prediction accuracy, providing important technical support for cultivated land quality assessment, soil health monitoring and sustainable management of land resources.

[0037] Compared with traditional soil organic matter content monitoring methods, the use of satellite remote sensing technology breaks through the limitations of traditional methods, estimates and inverts the organic matter content of cultivated land in large-scale areas, and realizes dynamic monitoring of the organic matter content of cultivated land surface soil, which is helpful for soil health monitoring, land resource protection and sustainable utilization management.

[0038] Example 1

[0039] The rapid inversion method for soil organic matter content constructed in this invention utilizes a training model based on measured organic matter content values ​​from 104 soil samples in western Jilin Province. This method then screens for significantly correlated band reflectance values, determines the algorithm's input variables, and employs a suitable machine learning algorithm for remote sensing inversion of soil organic matter content. The algorithm's predicted soil organic matter content has good applicability, avoiding or reducing traditional soil sample collection and laboratory soil organic matter content determination methods, reducing the time and cost of data acquisition and improving the efficiency of soil health assessment and sustainable development.

[0040] (1) Obtaining raw data;

[0041] Remote sensing image data of various bands of the Landsat-8 satellite within the study area were obtained and downloaded on the Google Earth Engine geospatial data processing platform (https: / / earthengine.google.com / ).

[0042] (2) Remote sensing image data preprocessing;

[0043] The original data is preprocessed, including atmospheric correction, radiation correction, orthorectification, cloud and shadow removal, etc., to obtain the reflectance data of each band of the remote sensing image.

[0044] (3) Acquisition of environmental covariate data;

[0045] Obtain environmental data within the study area, including topographic data and climate data.

[0046] The terrain data consisted of 30m-resolution digital elevation model (DEM) data, as well as slope and aspect data. The 30m-resolution DEM data was obtained from the Geospatial Data Cloud Platform (https: / / www.gscloud.cn / search). The slope and aspect data were calculated using the 30m-resolution DEM data in ArcGIS.

[0047] Climate data include annual average temperature data with a resolution of 1 km and annual average rainfall data with a resolution of 1 km, both of which were obtained from the National Earth System Science Data Center (https: / / www.geodata.cn / data / ). All data were resampled to a spatial resolution of 30 m to match the spatial resolution of Landsat-8 satellite remote sensing images.

[0048] (4) Soil sample collection;

[0049] During the bare soil period (late April) in western Jilin Province, a total of 104 soil samples with a surface depth of 0-20 cm were collected. A five-point sampling method was used to collect 500 g of soil with a surface thickness of 0-20 cm. A handheld GPS locator was used to record the latitude and longitude coordinates of the sampling points, and a field soil sample point collection record sheet was filled in to record the collection time and site description of the sampling points.

[0050] All collected soil samples were numbered and bagged, placed in an oven for drying, and residual impurities in the soil were removed. The samples were ground and sieved, and the soil organic carbon was oxidized using potassium dichromate-sulfuric acid solution to calculate the organic matter content.

[0051] (5) Spectral index calculation and correlation analysis;

[0052] Because multispectral sensors have limited bands, predicting organic matter from multispectral images requires calculating spectral indices for each band to obtain more information. This paper calculates multiple spectral indices and basic band functions based on the raw spectral data of remote sensing images, including the normalized difference index (NDI), ratio index (RI), difference index (DI), reciprocal (lgB), and exponential (expB).

[0053] NDI=(Band i-Band j) / (Band i+Band j)

[0054] RI=Band i / Band j

[0055] DI=Band i-Band j

[0056] Band i and Band j both represent the remote sensing reflectance of the first 7 bands of the Landsat-8 satellite OLI sensor, Band i represents the i-th band, and Band j represents the j-th band. A total of 91 spectral indices are obtained.

[0057] Pearson correlation analysis was used to determine the strength of association between soil organic matter content and various spectral indices, climate data, and topographic data.

[0058] Pearson correlation coefficient expression:

[0059]

[0060] Among them, r is the Pearson correlation coefficient, xi is the value of the i-th variable x, and yi is the value of the i-th variable y. and are the average values ​​of two variables x and y respectively, and n is the number of samples.

[0061] The spectral data and environmental covariate data were imported into SPSS16.0 software, and correlation analysis was performed with the measured values ​​of soil organic matter content at the sample points. Twenty-six variables with significant correlation were screened out as characteristic parameters of soil organic matter content, which served as the basic data for subsequent model establishment, namely: DEM, slope, aspect, average annual temperature, average annual precipitation, B1, B2, B3, B4, B5, B6, B7, D21, D61, D76, R74, NDI32, NDI72, 1 / B3, lgB1, lgB2, lgB3, lgB4, lgB6, lgB7, expB1, expB2, expB3, expB4, expB6, expB7.

[0062] (6) Estimation model establishment and algorithm inversion;

[0063] The 26 selected spectral characteristic parameters, along with DEM digital elevation data, slope, aspect, average annual temperature, and average annual precipitation, were used as the model input data, and the measured values ​​of soil organic matter content were used as the model output data. To evaluate the accuracy and stability of different models in predicting soil organic matter content, the coefficient of determination (R2) and root mean square error (RMSE) were used to evaluate the predictive ability of each model, as shown in the following formula:

[0064]

[0065] Several studies have shown that using different machine learning modeling methods to establish soil organic matter content inversion models will lead to differences in their prediction results and accuracy.

[0066] Therefore, in order to explore the effects of different modeling methods on the inversion results of soil organic matter content, the present invention uses three machine learning algorithms, namely random forest model (RF), support vector machine model (SVM), and extreme gradient boosting model (XG-Boost), to analyze the prediction results of soil organic matter content, as shown in Figure 2. 2 It reaches 0.80, which is more than 30% higher than random forest (0.72) and partial least squares (0.74); the mapping cost is reduced to 1 / 20 of traditional laboratory analysis, and the spatial resolution reaches 30m, supporting accurate inversion of 0-20cm depth of cultivated land surface.

[0067] (1) The model results are as follows:

[0068] The screened Landsat8 satellite remote sensing band data and environmental covariate data were used as input data, and the sample data of soil organic matter content obtained through field sampling was used as output data to construct an XGBoost regression model. Through training and verification of the model, it was determined that it had high accuracy and stability. The trained model was then applied to the entire study area to achieve the drawing of the spatial distribution map of soil organic matter content at a spatial resolution of 30 meters. Finally, a spatial distribution map of soil organic matter content covering the western part of Jilin Province was generated. Figure 3 As shown in the figure, the inversion results of soil organic matter content in the study area were obtained. The mapping cycle was shortened from several months to real-time dynamic updates, meeting the needs of high-frequency monitoring.

[0069] The multi-source environmental covariate-driven soil organic matter remote sensing inversion device provided by the present invention includes: a data acquisition module, including a satellite remote sensing data interface that supports access to multi-source data such as Landsat-8 and Sentinel-2, and an environmental covariate database that integrates data such as DEM, climate, and land use types; a data processing module, including a spectral index calculation unit with built-in 20 index algorithms such as NDI, RI, and DI, and a feature selection engine that uses the Pearson analysis algorithm to achieve automated feature screening; a modeling and inversion module, including a machine learning module based on the XGBoost framework that supports model training, verification, and optimization, and a real-time mapping unit that relies on GEE cloud computing capabilities to achieve minute-level regional mapping; an output and interaction module, including a visual interface that provides organic matter spatial distribution maps, statistical reports, and agricultural management suggestions, and an API interface that supports data intercommunication with agricultural Internet of Things systems (such as smart fertilization equipment). It supports multi-source satellite data such as Landsat-8, Sentinel-2, and GF-5, and is adapted to different soil types and regional conditions. Through the API interface and linkage with smart agricultural equipment, it provides direct technical support for precision fertilization and soil remediation, and promotes sustainable agricultural management.

[0070] This invention deeply integrates it with the soil organic matter inversion process to realize the cloud-based operation of the entire chain of "preprocessing-modeling-visualization", solving the problem that traditional local tools such as ENVI take several weeks.

[0071] This invention systematically integrates Landsat-8 spectral, topographic, and climate data for the first time, overcoming the limitations of a single data source through XGBoost modeling. The combination of Pearson analysis feature selection and XGBoost hyperparameter optimization significantly improves model accuracy and stability. Full cloud-based process integration: GEE-based real-time processing and dynamic updates overcome traditional timeliness bottlenecks.

[0072] The present invention addresses the limitations of traditional soil organic matter monitoring methods. Existing soil organic matter determination methods rely on field sampling and laboratory analysis. Although they are highly accurate, they are time-consuming, costly, and have limited spatial coverage, making it difficult to meet the needs of large-scale, real-time dynamic monitoring. The present invention aims to use remote sensing technology to break through the limitations of traditional methods and achieve rapid, low-cost, large-scale soil organic matter monitoring. A soil organic matter inversion model based on remote sensing images and environmental variables is constructed. The present invention uses Landsat-8 satellite remote sensing images, combined with environmental covariates such as digital elevation models and climate data, to extract characteristic information related to soil organic matter by calculating spectral indices and band transformations. A prediction model is established using an appropriate inversion algorithm to improve the estimation accuracy of soil organic matter content. Rapid spatial distribution mapping of soil organic matter is achieved to support agricultural and ecological management. By constructing an efficient soil organic matter content inversion model, the present invention can generate soil organic matter distribution maps on a large scale, providing a scientific basis for precision agriculture, soil resource management, and ecological environmental protection, and facilitating soil quality monitoring and sustainable utilization.

[0073] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A soil organic matter remote sensing inversion method driven by multi-source environmental covariates, characterized in that: The following steps are involved: (1) Obtain Landsat-8 satellite multi-band remote sensing image data of the target area through the Google Earth Engine platform; (2) Preprocessing the remote sensing image data, including atmospheric correction, radiometric calibration, orthorectification, and cloud shadow removal; (3) Obtain environmental covariate data of the target area, including 30m resolution digital elevation model (DEM), slope and aspect data, as well as annual average temperature and annual average rainfall climate data, and resample all environmental covariate data to 30m resolution; (4) Collect surface soil samples during the bare soil period in the target area, record the longitude and latitude coordinates of the sampling points, and measure the soil organic matter content; (5) Based on the preprocessed remote sensing image data, calculate the spectral indices of normalized difference index (NDI), ratio index (RI), and difference index (DI); (6) Inputting the spectral index of step (5), the environmental covariate data of step (3), and the soil organic matter content data of step (4) into the XGBoost algorithm to construct a soil organic matter inversion model; (7) The trained XGBoost model is used to invert the soil organic matter content in the target area and generate a spatial distribution map of soil organic matter.

2. The method according to claim 1, characterized in that The original spatial resolution of the climate data in step (3) is 1 km, which is resampled to 30 m resolution by bilinear interpolation.

3. The method according to claim 1, characterized in that The spectral index in step (5) also includes a derivative feature generated by performing an inverse transformation or an exponential transformation on the original band reflectance.

4. The method according to claim 1, wherein The soil samples in step (4) are collected using a five-point sampling method, where each sampling point is composed of a mixture of sub-samples at the four corners and the center of the sample plot.

5. The method according to claim 1, wherein The hyperparameters of the XGBoost algorithm in step (6) are optimized by a grid search method, including the learning rate (η), the maximum depth of the tree (max_depth) and the regularization coefficient (λ).

6. The method according to claim 1, characterized in that The soil organic matter spatial distribution map generated in step (7) is visualized and dynamically updated in the cloud through the Google Earth Engine platform.

7. The method according to claim 1, characterized in that The method is suitable for inversion of soil organic matter at a depth of 0-20 cm in the surface layer of cultivated land.

8. A soil organic matter remote sensing inversion device driven by multi-source environmental covariates, characterized in that: include: The data acquisition module includes a satellite remote sensing data interface that supports access to multi-source data from Landsat-8 and Sentinel-2, as well as an environmental covariate database that integrates DEM, climate, and land use type data. The data processing module includes a built-in NDI, RI, and DI spectral index calculation unit, as well as a feature selection engine that uses the Pearson analysis algorithm for automated feature screening. Modeling and inversion modules, including a machine learning unit based on the XGBoost framework and a real-time mapping unit based on the cloud computing capabilities of Google Earth Engine; The output and interaction module includes a visual interface that provides a spatial distribution map of organic matter, and an API interface that supports data exchange with the agricultural Internet of Things system.

9. The device according to claim 8, characterized in that The feature selection engine screens at least 26 key variables using a Pearson analysis algorithm, including DEM, slope, aspect, average annual temperature, average annual precipitation, and derived spectral features.

10. The device according to claim 8, characterized in that The real-time mapping unit can dynamically adjust the mapping time according to the area of ​​the target area to generate a soil organic matter spatial distribution map, and is linked to the intelligent fertilization equipment through the API interface.

Images