A quantitative evaluation method and system for predicting the content of a surface soil geochemical index and its influencing factors
By constructing a multi-source dataset and combining it with environmental enrichment features, and employing the Pearson correlation and optimal parameter geodetector model, an ensemble learning model was built. This model solved various problems in predicting soil geochemical indicators, achieving accurate prediction and multi-scale application, and improving the model's adaptability and reliability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- LINGNAN NORMAL UNIV
- Filing Date
- 2026-03-03
- Publication Date
- 2026-06-05
Smart Images
Figure CN122157874A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing technology, specifically to a quantitative evaluation method and system for predicting the content of geochemical indicators in surface soil and their influencing factors. Background Technology
[0002] Topsoil geochemical surveys are a crucial foundation for understanding regional soil environmental quality and guiding agricultural production and environmental protection. Traditional soil geochemical surveys rely on intensive sampling and laboratory analysis, which suffers from high costs, long cycles, and insufficient spatial coverage. With the development of geographic information systems, remote sensing technology, and machine learning, it has become possible to predict the content of soil geochemical indicators using multi-source environmental factor data.
[0003] However, existing methods for predicting soil geochemical indicators suffer from the following problems: First, the accuracy of a single model is limited, making it difficult to adapt to the complex spatial differentiation characteristics of multiple geochemical indicators; second, there is a lack of quantitative analysis of the influence mechanisms of environmental factors, making it impossible to clarify the contribution of each factor to the formation of soil geochemical indicators; third, there is a lack of a systematic verification system, making it difficult to guarantee the reliability of the prediction results; fourth, there is a lack of a multi-scale prediction application framework, failing to meet the actual application scenarios with different accuracy requirements; fifth, there is a lack of effective technical correlation between environmental enrichment characteristics and prediction models, resulting in low feature utilization; and sixth, factor interaction analysis is disconnected from model structure design, making it difficult to fully leverage the role of interaction features.
[0004] Therefore, it is necessary to provide a solution to achieve accurate prediction of the content of geochemical indicators in surface soil and quantitative evaluation of its influencing factors, establish a direct technical link between environmental enrichment characteristics and prediction models, effectively integrate the results of factor interaction analysis into model design, and provide a scientific basis for soil resource management and environmental protection. Summary of the Invention
[0005] The main objective of this invention is to propose a quantitative evaluation method and system for predicting the content of geochemical indicators in surface soil and their influencing factors, which can achieve accurate prediction of the content of geochemical indicators in surface soil.
[0006] In a first aspect, embodiments of the present invention provide a quantitative evaluation method for predicting the content of geochemical indicators in surface soil and their influencing factors, the method comprising the following steps:
[0007] Acquire geochemical survey data of surface soil and environmental factor data to construct a multi-source dataset;
[0008] Statistical analysis was performed on the multi-source dataset to obtain the element content distribution pattern and the combination law of geochemical indicators. A comparative analysis of the differences in geochemical indicators between surface soil and deep soil was conducted to obtain environmental enrichment characteristics, which were used as supplementary characteristics of the multi-source dataset.
[0009] Based on the geochemical indicators and continuous environmental factors in the multi-source dataset, the linear relationship between environmental factors and geochemical indicators is determined; based on the optimal parameter geodetector model, single-factor analysis and two-factor interaction analysis are performed on the multi-source dataset to obtain the degree of influence of environmental factors on geochemical indicators and the intensity of interaction, which serve as the basis for the design of model feature interaction.
[0010] Based on the multi-source dataset and the supplementary features, a prediction model is constructed and optimized, and the feature interaction design is incorporated into the model to obtain a trained ensemble learning model.
[0011] Single-factor perturbation experiments and multi-factor perturbation experiments were conducted on the trained ensemble learning model to obtain the independent contribution of environmental factors and the multi-factor synergistic mechanism, and the consistency of the single-factor contribution and interaction degree obtained by the optimal parameter geographic detector model and the ensemble learning model was compared.
[0012] Based on the trained ensemble learning model, a multi-scale prediction application system is constructed to obtain prediction results of geochemical index content at different scales.
[0013] Optionally, the acquisition of surface soil geochemical survey data and environmental factor data to construct a multi-source dataset includes:
[0014] Obtain surface soil geochemical survey data, which includes analytical values of multiple geochemical indicators;
[0015] Environmental factor data are acquired, and spatial registration and resampling techniques are used to unify the resolution of the environmental factor data and standardize the coordinate system.
[0016] The attributes of the environmental factor data are assigned to soil sampling points using spatial connectivity tools to construct the multi-source dataset.
[0017] The multi-source dataset is cleaned to obtain a preprocessed multi-source dataset.
[0018] Optionally, the statistical analysis of the multi-source dataset to obtain the elemental content distribution pattern and geochemical index combination law, and the comparative analysis of the differences in geochemical indices between surface soil and deep soil to obtain environmental enrichment characteristics, which serve as supplementary characteristics to the multi-source dataset, include:
[0019] The multi-source dataset was grouped and statistically analyzed to obtain the element content distribution pattern;
[0020] Correlation analysis, cluster analysis, and factor analysis were performed on the multi-source dataset to obtain the combination patterns of the geochemical indicators;
[0021] The environmental enrichment characteristics were obtained by comparing and analyzing the differences in environmental enrichment coefficients, inheritance, and coefficients of variation of geochemical indicators between surface soil and deep soil.
[0022] The environmental enrichment features are added to the multi-source dataset as supplementary features for the model input.
[0023] Optionally, the process involves determining the linear relationship between environmental factors and geochemical indicators based on geochemical indicators and continuous environmental factors in the multi-source dataset; and performing single-factor analysis and two-factor interaction analysis on the multi-source dataset based on the optimal parameter geodetector model to obtain the degree of influence of environmental factors on geochemical indicators and the strength of their interaction, which serves as the basis for the design of model feature interactions, including:
[0024] Pearson correlation analysis was used to perform correlation analysis and linear fitting on the geochemical indicators and continuous environmental factors in the multi-source dataset to obtain the linear relationship between environmental factors and geochemical indicators.
[0025] The multi-source dataset is coupled with the optimal parameter geodetector model to quantitatively analyze the influence of environmental factors on geochemical indicators and the contribution of individual factors.
[0026] The environmental factors are classified into multiple categories of influencing factors, and a two-factor interaction analysis is performed based on the optimal parameter geographic detector model to obtain the interaction intensity.
[0027] Based on the degree of influence and the strength of interaction, the feature interaction terms that need to be focused on being modeled in the ensemble learning model are determined.
[0028] Optionally, the step of constructing and optimizing the prediction model based on the multi-source dataset and the supplementary features, and incorporating the feature interaction design into the model to obtain the trained ensemble learning model, includes:
[0029] The multi-source dataset and the supplementary features are preprocessed and divided into training set, validation set and test set;
[0030] Based on the training set, a spatial interpolation model and a machine learning benchmark model are trained respectively to obtain a benchmark model set;
[0031] An ensemble learning model is constructed based on the feature interaction design, using gradient boosting decision tree model and random forest model as base learners and deep neural network model as secondary learners.
[0032] The ensemble learning model is optimized and evaluated using the validation set and the test set to obtain the trained ensemble learning model.
[0033] Optionally, the step of conducting single-factor perturbation experiments and multi-factor perturbation experiments on the trained ensemble learning model to obtain the independent contribution of environmental factors and the multi-factor synergistic mechanism, and comparing the consistency of the single-factor contribution and interaction degree obtained by the optimal parameter geographic detector model and the ensemble learning model, includes:
[0034] The independent contribution of the trained ensemble learning model was analyzed by single-factor perturbation experiments. The values of individual environmental factors were randomly shuffled using a permutation function, and the changes in model performance were observed to obtain the independent contribution of the environmental factors.
[0035] The interaction analysis of the trained ensemble learning model was conducted through multi-factor perturbation experiments. The values of two or three environmental factors were randomly shuffled using a permutation function, and the changes in model performance were observed to obtain the multi-factor synergistic mechanism.
[0036] The single-factor contribution level and interaction strength obtained from the analysis of the optimal parameter geographic detector model are compared with the single-factor independent contribution level and multi-factor synergistic mechanism obtained by the integrated learning model through perturbation experiments, and the rationality is verified by combining the literature knowledge base.
[0037] Optionally, the multi-scale prediction application system constructed based on the trained ensemble learning model obtains prediction results of geochemical index content at different scales, including:
[0038] Based on the trained ensemble learning model, the content of geochemical indicators at virtual sampling points is predicted to obtain grid-scale prediction results.
[0039] Based on the trained ensemble learning model, the content of geochemical indicators in unsampled areas is predicted to obtain regional-scale prediction results.
[0040] Based on the trained ensemble learning model, the content of geochemical indicators in sparse sampling scenarios is predicted, and the optimized prediction results of sampling density are obtained.
[0041] Secondly, embodiments of the present invention provide a quantitative evaluation system for predicting the content of geochemical indicators in surface soil and their influencing factors, the system comprising:
[0042] At least one processor;
[0043] At least one memory for storing at least one program;
[0044] When the at least one program is executed by the at least one processor, the at least one processor performs the method as described in any of the preceding statements.
[0045] The beneficial effects of this invention are as follows: By constructing a multi-source dataset and comprehensively utilizing multivariate statistical analysis and machine learning methods, this invention achieves accurate prediction of the content of geochemical indicators in surface soil; by using environmental enrichment features as supplementary features input into the prediction model, it improves the model's ability to identify polluted areas; by exploring the linear relationship between environmental factors and geochemical indicators through Pearson correlation analysis and linear fitting, it quantitatively analyzes the influence degree, single-factor contribution, and interaction of environmental factors through an optimal parameter geospatial detector model, and uses the influence degree and interaction strength as the basis for model feature interaction design, achieving an organic integration of factor analysis and model structure; by constructing an ensemble learning model, it improves prediction accuracy and generalization ability; by designing factor perturbation experiments and comparing the consistency of single-factor contribution and interaction degree obtained by OPGD and machine learning, two different principle methods, it establishes a complete verification system; by constructing a multi-scale prediction application system, it meets the needs of different application scenarios. This invention forms a complete technical closed loop of feature analysis, model construction, and mechanism verification, eliminating isolated technical features and providing scientific and technical support for soil resource management, agricultural production, and environmental protection. Attached Figure Description
[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0047] Figure 1 This is a flowchart illustrating the quantitative evaluation method for predicting the content of geochemical indicators in surface soil and their influencing factors in an embodiment of the present invention.
[0048] Figure 2 This is a technical roadmap diagram in an embodiment of the present invention.
[0049] Figure 3 This is a schematic diagram of the structure of a quantitative evaluation system for predicting the content of geochemical indicators in surface soil and their influencing factors in an embodiment of the present invention. Detailed Implementation
[0050] The following will clearly and completely describe the concept, specific structure, and technical effects of the present invention in conjunction with embodiments and accompanying drawings, so as to fully understand the purpose, solution, and effects of the present invention. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present invention can be combined with each other.
[0051] See Figure 1 and Figure 2 This invention provides a quantitative evaluation method for predicting the content of geochemical indicators in surface soil and their influencing factors. The method includes the following steps:
[0052] S100: Obtain surface soil geochemical survey data and environmental factor data to construct a multi-source dataset.
[0053] In some embodiments, the surface soil analysis data obtained from a 1:50,000 soil geochemical survey are first systematically organized, covering the analytical values of 21 geochemical indicators (nitrogen, phosphorus, potassium, organic matter, arsenic, copper, chromium, cadmium, mercury, nickel, lead, zinc, vanadium, manganese, fluorine, iodine, boron, cobalt, germanium, molybdenum, and selenium). Three-level quality control is used to ensure data reliability. A literature review is conducted using a knowledge database to summarize the main influencing factors of spatial differentiation for each indicator. Environmental factor data is collected through a geographic and geological data sharing platform. Spatial registration and resampling techniques are used to unify the data resolution to 500m, standardize it to the WGS84 coordinate system, and crop it to the study area boundary. Using the spatial connection tool in ArcGIS software, environmental factor attribute data are accurately assigned to soil sampling points to construct a multi-source dataset. Finally, data cleaning (including missing value imputation, outlier removal, and data standardization) is performed to ensure data integrity.
[0054] S200, Statistical analysis is performed on the multi-source dataset to obtain the element content distribution pattern and geochemical index combination law. The differences in geochemical indices between surface soil and deep soil are compared and analyzed to obtain environmental enrichment characteristics, which are used as supplementary characteristics of the multi-source dataset.
[0055] In some embodiments, grouped statistics were used to reveal the distribution patterns of elemental content, and Pearson correlation, K-means clustering, and factor analysis were employed to analyze the combination patterns of 21 indicators. A time-series comparative analysis (2010 vs. 2019) was used to assess the stability of the surface soil environment in the study area. By comparing the differences in environmental enrichment coefficients, inheritance, and coefficients of variation of geochemical indicators between surface and deep soils, the environmental enrichment characteristics of surface soil geochemical indicators were explored.
[0056] The key point is that this invention not only uses environmental enrichment features as descriptive analysis results, but also incorporates them as supplementary features into multi-source datasets. Specifically, the calculated environmental enrichment coefficient, inheritance coefficient, and coefficient of variation are added as new feature columns to the dataset, enabling the model to directly utilize these features for learning. For example, for heavy metal elements, if their content in topsoil is significantly higher than in deep soil (high enrichment coefficient), the model can learn to identify the association between this enrichment pattern and environmental factors, thereby improving the prediction accuracy of polluted areas.
[0057] S300, Based on the geochemical indicators and continuous environmental factors in the multi-source dataset, determine the linear relationship between environmental factors and geochemical indicators; Based on the optimal parameter geodetector model, perform single-factor analysis and two-factor interaction analysis on the multi-source dataset to obtain the degree of influence of environmental factors on geochemical indicators and the intensity of interaction, and use it as the basis for the design of model feature interaction.
[0058] In some embodiments, Pearson correlation analysis is first used to analyze the relationship between geochemical indicators and continuous environmental factors, followed by linear fitting to explore the linear relationship between environmental factors and geochemical indicators. Secondly, a multi-source dataset is coupled with an Optimal Parameters-based Geographical Detector (OPGD) model to quantitatively analyze the influence of environmental factors on 21 geochemical indicators and their individual contribution. The OPGD model is a novel statistical method for detecting spatial heterogeneity and revealing its underlying driving factors. This method has no linear assumptions, possesses an elegant form, and clear physical meaning. The basic idea is: assuming the study area is divided into several sub-regions, spatial heterogeneity exists if the sum of the variances of the sub-regions is less than the total variance of the region; if the spatial distributions of two variables tend to be consistent, a statistical correlation exists between them. Finally, environmental factors are classified into several categories of influencing factors, and a two-factor interaction analysis is conducted based on OPGD to explore the strength of the interaction between the influencing factors.
[0059] The innovation of this invention lies in directly applying the results of single-factor contribution and interaction strength analysis to the design of model feature interactions. Specifically, based on the influence degree and interaction strength ranking obtained from OPGD analysis, single-factor and two-factor interaction terms that significantly affect geochemical indicators are identified. When constructing the ensemble learning model, these significant features are either input as explicit features into the model or used to guide the model to automatically learn these interaction patterns. For example, if OPGD analysis finds that the interaction strength between soil type and pH value is the highest, this interaction feature is explicitly constructed in the model, or the model structure is adjusted to prioritize learning this interaction relationship.
[0060] S400, construct and optimize the prediction model based on the multi-source dataset and the supplementary features, and incorporate the feature interaction design into the model to obtain the trained ensemble learning model.
[0061] In some embodiments, a two-stage modeling strategy of single-model benchmarking and ensemble model innovation is designed: data preprocessing of multi-source datasets and supplementary features, followed by partitioning into training, validation, and test sets (reservation method), and system training and optimization. Benchmark models include spatial interpolation models (Inverse Distance Weighted Interpolation (IDW) and Ordinary Kriging (OK)) and machine learning models (including Gradient Boosting Decision Tree (LightGBM), Random Forest (RF), and Deep Neural Network (DNN)). Based on this, using LightGBM and RF as base learners and DNN as secondary learners, a new ensemble model (LRD) is constructed based on feature interaction design. Specifically, important single factors and interaction terms identified by OPGD are incorporated into the feature engineering of the base learners, and an interaction feature layer is added to the network structure of the secondary learners. Model performance is evaluated using metrics such as Mean Squared Error (MSE), Mean Absolute Percentage Error (MAPE), and Coefficient of Determination (R²), comparing the performance of each model in terms of prediction accuracy, multi-metric adaptability, and result consistency.
[0062] S500, Single-factor perturbation experiments and multi-factor perturbation experiments are conducted on the trained ensemble learning model to obtain the independent contribution of environmental factors and the multi-factor synergistic mechanism, and the consistency of the single-factor contribution and interaction degree obtained by the optimal parameter geographic detector model and the ensemble learning model is compared.
[0063] In some embodiments, a dual validation system of factor perturbation experiments and model comparison validation is designed: The independent contribution of each environmental factor is quantified through single-factor perturbation experiments. The basic idea is to randomly shuffle the values of individual environmental factors using a permutation function, observe the impact on model performance (MSE, MAPE, R², etc.), and then determine the importance of this feature to the model. This independent contribution corresponds to the strength of the single-factor effect. A high-dimensional interaction network is constructed to reveal the multi-factor synergistic mechanism. The values of two or three environmental factors are randomly shuffled using a permutation function, and the impact on model performance (MSE, MAPE, R², etc.) under the interaction of two or three environmental factors is observed. This determines the importance of the shuffled interaction of the two or three environmental factors to the model, and this multi-factor synergistic mechanism corresponds to the interaction strength.
[0064] The key innovation of this invention lies in establishing a comparative verification mechanism between OPGD and machine learning, two methods based on different principles. Specifically, the single-factor contribution levels and interaction strengths obtained from OPGD analysis are ranked and compared with the single-factor independent contribution levels and multi-factor synergistic mechanisms obtained from permutation experiments. If the two are highly consistent, it indicates that the model has successfully learned the real factor influence relationships and interactions. If there are significant differences, further analysis is needed, which may indicate that the model has not fully captured certain interaction patterns or that the OPGD analysis has biases. Through this comparative verification, combined with a literature knowledge base, a rationality verification is conducted to establish a driving mechanism explanation framework that includes direct effects and interaction effects.
[0065] S600, based on the trained ensemble learning model, construct a multi-scale prediction application system to obtain prediction results of geochemical index content at different scales.
[0066] In some embodiments, a progressive hierarchical prediction framework is constructed:
[0067] Prediction of geochemical index content at virtual sampling points: 500m grid prediction for the study area;
[0068] Prediction of geochemical index content in unsampled areas: Prediction is carried out for reserved sampling points (test sets);
[0069] Sparse sampling optimization prediction: Simulating low-to-medium density sampling scenarios, the sampling density threshold effect is explored by predicting the number of sampling points and the content of geochemical indicators.
[0070] and Figure 1 The corresponding method is referenced. Figure 3 This invention provides a quantitative evaluation system for predicting the content of geochemical indicators in surface soil and their influencing factors, comprising:
[0071] At least one processor;
[0072] At least one memory for storing at least one program;
[0073] When the at least one program is executed by the at least one processor, the at least one processor performs the method described above.
[0074] It is evident that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.
[0075] Furthermore, embodiments of the present invention also disclose a computer program product or computer program stored in a computer-readable storage medium. A processor of a computer device can read the computer program from the computer-readable storage medium, and the processor executes the computer program, causing the computer device to perform the described method. Similarly, the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.
[0076] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.
[0077] The above is a detailed description of the preferred embodiments of this disclosure. However, this disclosure is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this disclosure. All such equivalent modifications or substitutions are included within the scope defined by the claims of this disclosure.
Claims
1. A quantitative evaluation method for predicting the content of geochemical indicators in surface soil and their influencing factors, characterized in that, The method includes the following steps: Acquire geochemical survey data of surface soil and environmental factor data to construct a multi-source dataset; Statistical analysis was performed on the multi-source dataset to obtain the element content distribution pattern and the combination law of geochemical indicators. A comparative analysis of the differences in geochemical indicators between surface soil and deep soil was conducted to obtain environmental enrichment characteristics, which were used as supplementary characteristics of the multi-source dataset. Based on the geochemical indicators and continuous environmental factors in the multi-source dataset, the linear relationship between environmental factors and geochemical indicators is determined; based on the optimal parameter geodetector model, single-factor analysis and two-factor interaction analysis are performed on the multi-source dataset to obtain the degree of influence of environmental factors on geochemical indicators and the intensity of interaction, which serve as the basis for the design of model feature interaction. Based on the multi-source dataset and the supplementary features, a prediction model is constructed and optimized, and the feature interaction design is incorporated into the model to obtain a trained ensemble learning model. Single-factor perturbation experiments and multi-factor perturbation experiments were conducted on the trained ensemble learning model to obtain the independent contribution of environmental factors and the multi-factor synergistic mechanism, and the consistency of the single-factor contribution and interaction degree obtained by the optimal parameter geographic detector model and the ensemble learning model was compared. Based on the trained ensemble learning model, a multi-scale prediction application system is constructed to obtain prediction results of geochemical index content at different scales.
2. The method according to claim 1, characterized in that, The acquisition of surface soil geochemical survey data and environmental factor data, and the construction of a multi-source dataset, includes: Obtain surface soil geochemical survey data, which includes analytical values of multiple geochemical indicators; Environmental factor data are acquired, and spatial registration and resampling techniques are used to unify the resolution of the environmental factor data and standardize the coordinate system. The attributes of the environmental factor data are assigned to soil sampling points using spatial connectivity tools to construct the multi-source dataset. The multi-source dataset is cleaned to obtain a preprocessed multi-source dataset.
3. The method according to claim 1, characterized in that, The statistical analysis of the multi-source dataset yields the elemental content distribution patterns and geochemical index combination rules. A comparative analysis of the differences in geochemical indices between surface and deep soil layers is conducted to obtain environmental enrichment characteristics, which serve as supplementary features to the multi-source dataset, including: The multi-source dataset was grouped and statistically analyzed to obtain the element content distribution pattern; Correlation analysis, cluster analysis, and factor analysis were performed on the multi-source dataset to obtain the combination patterns of the geochemical indicators; The environmental enrichment characteristics were obtained by comparing and analyzing the differences in environmental enrichment coefficients, inheritance, and coefficients of variation of geochemical indicators between surface soil and deep soil. The environmental enrichment features are added to the multi-source dataset as supplementary features for the model input.
4. The method according to claim 1, characterized in that, The process involves determining the linear relationship between environmental factors and geochemical indicators based on geochemical indicators and continuous environmental factors in the multi-source dataset; performing single-factor analysis and two-factor interaction analysis on the multi-source dataset based on the optimal parameter geodetector model to obtain the degree of influence of environmental factors on geochemical indicators and the strength of their interaction, which serves as the basis for the design of model feature interactions, including: Pearson correlation analysis was used to perform correlation analysis and linear fitting on the geochemical indicators and continuous environmental factors in the multi-source dataset to obtain the linear relationship between environmental factors and geochemical indicators. The multi-source dataset is coupled with the optimal parameter geodetector model to quantitatively analyze the influence of environmental factors on geochemical indicators and the contribution of individual factors. The environmental factors are classified into multiple categories of influencing factors, and a two-factor interaction analysis is performed based on the optimal parameter geographic detector model to obtain the intensity of the interaction. Based on the degree of influence and the strength of interaction, the feature interaction terms that need to be focused on being modeled in the ensemble learning model are determined.
5. The method according to claim 1, characterized in that, The process of constructing and optimizing a prediction model based on the multi-source dataset and the supplementary features, and incorporating the feature interaction design into the model to obtain a trained ensemble learning model, includes: The multi-source dataset and the supplementary features are preprocessed and divided into training set, validation set and test set; Based on the training set, a spatial interpolation model and a machine learning benchmark model are trained respectively to obtain a benchmark model set; An ensemble learning model is constructed based on the feature interaction design, using gradient boosting decision tree model and random forest model as base learners and deep neural network model as secondary learners. The ensemble learning model is optimized and evaluated using the validation set and the test set to obtain the trained ensemble learning model.
6. The method according to claim 1, characterized in that, The process involves conducting single-factor and multi-factor perturbation experiments on the trained ensemble learning model to obtain the independent contribution of environmental factors and the multi-factor synergistic mechanism. The consistency of the single-factor contribution and interaction levels obtained by the optimal parameter geographic detector model and the ensemble learning model is then compared, including: The independent contribution of the trained ensemble learning model was analyzed by single-factor perturbation experiments. The values of individual environmental factors were randomly shuffled using a permutation function, and the changes in model performance were observed to obtain the independent contribution of the environmental factors. The interaction analysis of the trained ensemble learning model was conducted through multi-factor perturbation experiments. The values of two or three environmental factors were randomly shuffled using a permutation function, and the changes in model performance were observed to obtain the multi-factor synergistic mechanism. The single-factor contribution level and interaction strength obtained from the analysis of the optimal parameter geographic detector model are compared with the single-factor independent contribution level and multi-factor synergistic mechanism obtained by the integrated learning model through perturbation experiments, and the rationality is verified by combining the literature knowledge base.
7. The method according to claim 1, characterized in that, The multi-scale prediction application system constructed based on the trained ensemble learning model obtains prediction results of geochemical index content at different scales, including: Based on the trained ensemble learning model, the content of geochemical indicators at virtual sampling points is predicted to obtain grid-scale prediction results. Based on the trained ensemble learning model, the content of geochemical indicators in unsampled areas is predicted to obtain regional-scale prediction results. Based on the trained ensemble learning model, the content of geochemical indicators in sparse sampling scenarios is predicted, and the optimized prediction results of sampling density are obtained.
8. A quantitative evaluation system for predicting the content of geochemical indicators in surface soil and their influencing factors, characterized in that, The system includes: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor performs the method as described in any one of claims 1 to 7.