A soil organic carbon inversion method based on multi-source remote sensing data and ensemble learning
By combining multi-source remote sensing data with environmental factors and integrating learning models, the high cost and model stability issues of soil organic carbon spatial distribution estimation in plantation areas were resolved, achieving high-precision and stable soil organic carbon inversion.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUILIN UNIVERSITY OF TECHNOLOGY
- Filing Date
- 2026-04-01
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies are costly and difficult to continuously acquire in estimating the spatial distribution of soil organic carbon in plantation areas. Single remote sensing data has limited representation under complex surface conditions, and the redundancy of information from multi-source remote sensing data affects the stability and adaptability of the model.
By jointly processing multi-source remote sensing data and environmental factors, a multi-dimensional feature set is constructed. The optimal feature subset is selected by recursive feature elimination and cross-validation. Combined with an ensemble learning model, multiple prediction results are integrated to form a soil organic carbon inversion system.
It improves the stability and applicability of soil organic carbon inversion results, making it suitable for high-precision estimation under complex surface conditions. It also reduces the redundancy of multi-source high-dimensional features and enhances the reliability and consistency of the model.
Smart Images
Figure CN122432961A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of remote sensing information processing, geographic information systems and artificial intelligence, specifically to a soil organic carbon inversion system and method based on multi-source remote sensing data fusion and integrated learning, applicable to forest ecosystems, especially artificial forest areas for soil organic carbon content estimation and spatial distribution analysis. Background Technology
[0002] Soil organic carbon (SOC) is an important component of soil organic carbon, and its content variations are closely related to regional carbon storage estimation and ecological environment management. In plantation areas, due to differences in stand structure, topography, and climate, soil organic carbon is often spatially unevenly distributed. Traditional methods relying on field sampling and statistical analysis are costly to apply on a large scale and are difficult to obtain continuously.
[0003] With the development of remote sensing data acquisition methods, the analysis of soil-related information using remote sensing imagery has been gradually applied. Existing methods are mostly based on optical or radar remote sensing data for modeling. While this improves data acquisition efficiency, it still has certain limitations under complex surface conditions. A single data source has limited representation of surface information, while using multi-source remote sensing data in combination can easily lead to a large number of features and information redundancy, thus affecting model stability. Furthermore, current technologies generally use a single model for prediction, and the adaptability and consistency of prediction results still have room for improvement when facing multi-source, high-dimensional data.
[0004] To address the above issues, this invention unifies the processing of multi-source remote sensing data and environmental factors. Based on feature construction and screening, it introduces ensemble learning to fuse the prediction results of multiple models, ultimately forming a soil organic carbon inversion system based on multi-source remote sensing data and ensemble learning, thereby improving the stability and applicability of soil organic carbon inversion results. Summary of the Invention
[0005] The purpose of this invention is to provide a soil organic carbon inversion method based on multi-source remote sensing data and ensemble learning. By jointly processing multi-source remote sensing data and environmental factor data, and combining feature construction, feature optimization and ensemble learning, a high-precision inversion method for soil organic carbon content in plantation areas can be achieved, thereby improving the stability and applicability of the inversion results.
[0006] To achieve the above objectives, the technical solution adopted by the present invention includes the following steps: Step S1. Data acquisition and preprocessing: Acquire environmental factor data of soil sampling points in the study area, as well as multi-source remote sensing data corresponding to the spatial location and sampling time of the soil sampling points, and store the data in the database after preprocessing. Step S2. Feature Construction and Feature Optimization: Based on the data obtained in Step S1, construct a multidimensional feature set for soil organic carbon inversion; Step S3. Model Training: Based on the optimal feature subset obtained in step S2, construct multiple prediction models for soil organic carbon inversion, and train the prediction models respectively. Step S4. Integration of Prediction and Output: Based on the multiple prediction models obtained in Step S3, an integrated learning model is constructed to fuse the prediction results of each prediction model and obtain the final inversion result of soil organic carbon.
[0007] The data acquisition and preprocessing in step S1 are as follows: Step S11: Obtain environmental factor data of soil sampling points within the study area, and obtain multi-source remote sensing data corresponding to the soil sampling points through a cloud computing platform. The multi-source remote sensing data includes optical remote sensing data and radar remote sensing data. Step S12: Preprocess the multi-source remote sensing data, including cloud and shadow removal, atmospheric correction and time synthesis processing for optical remote sensing data, and radiometric correction, geometric correction and terrain correction processing for radar remote sensing data. Step S13: Based on the spatial location of the soil sampling points, perform spatial matching and feature extraction on the preprocessed multi-source remote sensing data to obtain remote sensing feature data corresponding to each sampling point, and use it together with environmental factor data as the data basis for subsequent analysis.
[0008] The feature construction and feature optimization steps in step S2 are as follows: Step S21: Based on the optical remote sensing data and radar remote sensing data obtained in step S1, construct spectral band features, vegetation index features, texture features and polarization combination features, and fuse them with environmental factor data to form a multidimensional feature set. Step S22: Use a recursive feature elimination method to optimize the multidimensional feature set. During the recursive feature elimination process, the importance of each input feature is evaluated by a learning model, and features with lower contribution are gradually eliminated according to the feature importance ranking results.
[0009] The importance of the feature can be expressed as: in, This represents the importance score of the i-th feature. represents the corresponding feature input, and f(⋅) represents the feature evaluation function implemented internally by the learning model.
[0010] Step S23: Evaluate the prediction performance of different feature subsets through cross-validation, and select the feature subset with the best performance as the input features for subsequent model training.
[0011] The model training in step S3 includes the following steps: Step S31: Based on the optimal feature subset obtained in step S2, construct multiple prediction models for soil organic carbon inversion; Step S32: Divide the sample data into training and test sets, and optimize the parameters of each prediction model; Step S33: Train each prediction model based on the training set to obtain multiple trained prediction models; Step S34: Use the test set to make predictions for each prediction model and obtain the corresponding prediction results.
[0012] Step S4, integrating prediction and output, includes the following steps: Step S41: Use the prediction results of each prediction model in step S3 as input to construct an ensemble learning model; Step S42: The meta-learner fuses the prediction results of each base learner, and the fusion relationship can be expressed as follows: .
[0013] in, Let represent the prediction result of the i-th base learner, g(⋅) represent the fusion function, and represent the final soil organic carbon inversion result. In a preferred embodiment, the fusion function is a weighted fusion form, and its calculation relationship is as follows: .
[0014] in, These are the weight coefficients of the basis learner, which are learned through training sample data to minimize the error between the predicted result and the measured soil organic carbon value.
[0015] Step S43: Output and store the soil organic carbon inversion results for subsequent analysis and application.
[0016] The present invention has the following beneficial effects and advantages: (1) By combining multi-source remote sensing data with environmental factor data, the adaptability of soil organic carbon inversion to complex surface conditions has been improved. (2) By combining feature construction and feature optimization, the redundancy of multi-source high-dimensional features is reduced and the stability of the model is improved; (3) By integrating the results of multiple prediction models through ensemble learning, the reliability and consistency of soil organic carbon inversion results are improved; (4) The method of the present invention is applicable to the regional scale soil organic carbon inversion and has good engineering application value. Attached Figure Description
[0017] Figure 1 This diagram illustrates the steps of a soil organic carbon inversion method based on multi-source remote sensing data and ensemble learning, as described in this invention.
[0018] Figure 2 This is a flowchart illustrating the implementation of the soil organic carbon inversion method based on ensemble learning in this invention.
[0019] Marked in the image: S1 - Data acquisition and preprocessing, S2 - Feature construction and feature optimization, S3 - Model training, S4 - Ensemble prediction and result output. Detailed Implementation
[0020] Example: like Figure 1 As shown, the technical solution of the present invention includes four steps: data acquisition and preprocessing, feature construction and feature optimization, model training, ensemble prediction and result output.
[0021] Step S1 Data Acquisition and Preprocessing: Acquire environmental factor data of soil sampling points in the study area, and acquire multi-source remote sensing data corresponding to the soil sampling points in terms of spatial location and sampling time. After preprocessing the multi-source remote sensing data, store it in the database. The multi-source remote sensing data includes optical remote sensing data and radar remote sensing data. Step S2 Feature Construction and Feature Optimization: Based on the data preprocessed in step S1, a multidimensional feature set for soil organic carbon inversion is constructed, and the feature set is optimized to obtain the optimal feature subset for modeling. The model training in step S3 involves inputting the optimal feature subset obtained in step S2 into multiple prediction models for soil organic carbon inversion for training, resulting in multiple trained prediction models. Step S4, Integrated Prediction and Result Output: The prediction results of each prediction model in step S3 are input into the integrated learning model for fusion to obtain the final inversion result of soil organic carbon, and the inversion result is output.
[0022] like Figure 2 As shown in the flowchart, the implementation of the ensemble learning method for soil organic carbon inversion consists of the following steps: First, environmental factor data and multi-source remote sensing data within the study area are acquired. In step S1, the environmental factor acquisition and preprocessing method performs unified data processing and preprocessing on the data, including soil physicochemical properties, topographic factors, and climate factors, as well as the multi-source remote sensing data, including optical remote sensing data and radar remote sensing data, to provide basic data support for subsequent feature construction.
[0023] Step S2, Feature Construction and Optimization Method, constructs spectral band features, vegetation index features, texture features, and polarization combination features based on the preprocessed multi-source remote sensing data, and integrates them with environmental factor data to form a multidimensional feature set for soil organic carbon inversion.
[0024] Based on this, a recursive feature elimination process is performed on the multidimensional feature set. Features with lower contribution are gradually eliminated through feature importance evaluation, and the predictive performance of the feature subset is judged. When the performance of the feature subset is not optimal, the feature elimination process continues until the optimal feature subset is obtained. The feature importance satisfies the following relationship: in, Indicates the first The importance score of each feature This represents the corresponding feature input. This represents the feature evaluation function implemented internally by the learning model; and the predictive performance of different feature subsets is evaluated through cross-validation, and the feature subset with the best predictive performance is selected as the input features for subsequent model training.
[0025] Next, a prediction model is constructed based on the optimal feature subset. In step S3, the model training method divides the sample data into a training set and a test set. The training set is used to train multiple base prediction models, and the test set is used to obtain the prediction results corresponding to each base model.
[0026] Step S4, the ensemble prediction and result method, uses the prediction results of multiple base prediction models as input to construct an ensemble learning model, and fuses the prediction results through a meta-learner. The fusion relationship satisfies the following equation: .
[0027] in, Indicates the first The prediction results of each base learner The fusion function represents the final soil organic carbon inversion result. In a preferred embodiment, the fusion function is a weighted fusion form, and its calculation relationship is as follows: .
[0028] in, These are the weight coefficients for the basis learner, which are learned through training sample data to minimize the error between the predicted results and the measured soil organic carbon values. Finally, the soil organic carbon inversion results are output, realizing the quantitative inversion of soil organic carbon content in the study area, and the process ends.
[0029] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.
Claims
1. A method for retrieving soil organic carbon based on multi-source remote sensing data and ensemble learning, characterized in that, Includes the following steps: Step S1. Data acquisition and preprocessing: Acquire environmental factor data of soil sampling points in the study area, as well as multi-source remote sensing data corresponding to the spatial location and sampling time of the soil sampling points, and store the data in the database after preprocessing. Step S2. Feature Construction and Feature Optimization: Based on the data obtained in Step S1, construct a multidimensional feature set for soil organic carbon inversion; Step S3. Model Training: Based on the optimal feature subset obtained in step S2, construct multiple prediction models for soil organic carbon inversion, and train the prediction models respectively. Step S4. Integration of Prediction and Output: Based on the multiple prediction models obtained in Step S3, an integrated learning model is constructed to fuse the prediction results of each prediction model and obtain the final inversion result of soil organic carbon.
2. The soil organic carbon inversion method based on multi-source remote sensing data and ensemble learning according to claim 1 is characterized in that, The data acquisition and preprocessing method is used to acquire multi-source remote sensing data corresponding to the spatial location and sampling time of soil sampling points through a cloud computing platform, and to preprocess the multi-source remote sensing data, including cloud and shadow removal, atmospheric correction and time synthesis processing for optical remote sensing data, and radiometric correction, geometric correction and topographic correction processing for radar remote sensing data; and based on the spatial location of soil sampling points, to perform spatial matching and feature extraction on the preprocessed multi-source remote sensing data to obtain remote sensing feature data corresponding to each sampling point, which, together with environmental factor data, serves as the input data basis for soil organic carbon inversion.
3. The soil organic carbon inversion method based on multi-source remote sensing data and ensemble learning according to claim 1 is characterized in that, The feature construction and optimization steps are used to construct spectral band features, vegetation index features, texture features, and polarization combination features based on the optical and radar remote sensing data obtained in step S1, and fuse them with environmental factor data to form a multidimensional feature set. Further, a recursive feature elimination method is used to optimize the multidimensional feature set, evaluating feature importance through a learning model and gradually eliminating features with low contribution, wherein the feature importance satisfies the following relationship: in, Indicates the first The importance score of each feature This represents the corresponding feature input. This represents the feature evaluation function implemented internally by the learning model; and the predictive performance of different feature subsets is evaluated through cross-validation, and the feature subset with the best predictive performance is selected as the input features for subsequent model training.
4. The soil organic carbon inversion method based on multi-source remote sensing data and ensemble learning according to claim 1 is characterized in that, The model training method is used to construct multiple prediction models for soil organic carbon inversion based on the optimal feature subset obtained in step S2, and to divide the sample data into training set and test set, optimize and train the parameters of each prediction model, and obtain multiple trained prediction models and their corresponding prediction results.
5. The soil organic carbon inversion method based on multi-source remote sensing data and ensemble learning according to claim 1 is characterized in that, The ensemble prediction and result output method is used to construct an ensemble learning model by taking the prediction results of multiple prediction models as input, and to fuse the prediction results through a meta-learner. The fusion relationship satisfies the following equation: in, Indicates the first The prediction results of each base learner The fusion function represents the final soil organic carbon inversion result. In a preferred embodiment, the fusion function is a weighted fusion form, and its calculation relationship is as follows: . in, These are the weight coefficients of the basis learner, which are learned through training sample data to minimize the error between the predicted result and the measured soil organic carbon value. The soil organic carbon inversion result is then output and stored for subsequent analysis and application.