Soil profile conductivity prediction method and device based on resistivity tomography

Abstract

The invention discloses a soil profile conductivity prediction method and device based on resistivity tomography, and relates to the field of soil physics and geophysical exploration, and the method comprises the steps: obtaining real resistivity spatial distribution data; collecting an undisturbed soil profile sample, and measuring the actually measured conductivity, water content and volume weight of the sample; performing space alignment on the actually measured data and the real resistivity space distribution data, and constructing a plurality of machine learning regression models; performing stratified sampling on each machine learning regression model to divide a training set and a verification set, verifying the training set and the verification set, and selecting a model with optimal comprehensive performance as a final prediction model; outputting the predicted conductivity of each grid point based on the optimal prediction model; and performing spatial interpolation on the predicted conductivity points to generate a continuous soil profile conductivity distribution diagram. The problems that a traditional method is low in efficiency, and an electromagnetic induction method is poor in vertical resolution and limited in detection depth are solved, and the method has the remarkable advantages of being non-invasive, high in precision, high in resolution, high in deep detection capacity and the like.

Description

Technical Field

[0001] This application relates to the field of soil physics and geophysical exploration technology, and in particular to a method and apparatus for predicting the electrical conductivity of soil profiles based on resistivity tomography. Background Technology

[0002] Soil electrical conductivity is a key indicator reflecting soil salinity and has a significant impact on agricultural production and the ecological environment. Rapid in-situ acquisition of soil profile electrical conductivity in the field faces severe challenges. Due to the strong spatial heterogeneity of soil salinity distribution, traditional point-scale methods combining soil profile excavation or borehole sampling with indoor testing are insufficient for rapidly acquiring the spatial distribution characteristics of two-dimensional soil profile electrical conductivity, thus hindering the rapid monitoring of soil salinization information. Currently, an electromagnetic induction-based method can be used for rapid in-situ acquisition of soil profile electrical conductivity in the field. This method emits a primary magnetic field from the surface down into the soil and collects primary and secondary magnetic field signals. Using the conversion formula between magnetic field strength and conductivity, the apparent electrical conductivity of the soil profile is calculated. Combined with algorithms such as linear regression, a predictive model between apparent and measured electrical conductivity is established, thereby achieving rapid acquisition of soil profile electrical conductivity. Although this method utilizes electromagnetic induction technology to achieve rapid in-situ acquisition in the field, it has low vertical resolution and limited detection depth (usually within 1.5 meters). The apparent conductivity is a weighted average from the surface to a certain depth underground rather than a value at a specific depth, making it difficult to achieve rapid acquisition of soil profile conductivity that meets both high vertical resolution and large detection depth requirements.

[0003] Therefore, there is an urgent need for a method to obtain soil profile electrical conductivity that combines high resolution, large detection depth, non-invasiveness, and high prediction accuracy. Summary of the Invention

[0004] The purpose of this application is to provide a method and apparatus for predicting the electrical conductivity of soil profiles based on resistivity tomography, so as to overcome the problems of low vertical resolution, limited detection depth and low efficiency in the prior art.

[0005] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a method for predicting the electrical conductivity of soil profiles based on resistivity tomography, including: Obtain true spatial distribution data of resistivity; Uncirculated soil profile samples were collected, and the uncirculated soil profile samples were layered according to depth to determine the measured electrical conductivity, water content and bulk density of each layer; The measured electrical conductivity, water content, and bulk density are spatially aligned with the actual resistivity spatial distribution data based on the three-dimensional spatial coordinates of their sampling points, forming a matching dataset containing coordinates, resistivity, and the corresponding measured soil electrical conductivity, water content, and bulk density. Based on the matching dataset, multiple machine learning regression models are constructed, each with different feature combinations as input variables. The feature combinations include: resistivity + depth, resistivity + depth + water content, resistivity + depth + bulk density, and resistivity + depth + water content + bulk density, with conductivity as the target variable. The training set and validation set of each machine learning regression model are divided by stratified sampling, and the performance of each model is evaluated and the model with the best overall performance is selected as the final prediction model. The optimal prediction model is applied to the real resistivity and depth data obtained by resistivity tomography in the new monitoring area, and the predicted conductivity of each grid point is output. Spatial interpolation is performed on the predicted conductivity points to generate a continuous soil profile conductivity distribution map.

[0006] Optionally, obtaining the actual spatial distribution data of resistivity specifically includes the following steps: An electrode array was deployed on the surface of the target area, and a low-frequency stable current was injected using a resistivity tomography system to measure the potential difference between each electrode pair. The apparent resistivity is calculated based on the potential difference, and the actual resistivity spatial distribution data is generated by the least squares inversion algorithm.

[0007] Optionally, the apparent resistivity is calculated based on the potential difference, and the actual resistivity spatial distribution data is generated using the least squares inversion algorithm, specifically using the following formula: ; in, represents the objective function; m represents the logarithm of the true resistivity; Represents the data weight matrix; This represents the measured apparent resistivity; This represents the theoretical apparent resistivity calculated by numerically solving the electric field equations. Represents the regularization factor; This represents the smoothing constraint matrix of the model.

[0008] Optionally, the spatial alignment is achieved through a unified three-dimensional coordinate system, ensuring that the positional error between each measured sample point and the center point of the resistivity grid in the X, Y, and Z directions is less than 15 cm.

[0009] Optionally, the plurality of machine learning regression models include support vector machines (SVM), random forests (RF), extreme gradient boosting trees (XGBoost), and linear regression (LM).

[0010] Optionally, the hierarchical sampling method for dividing the training set and validation set for each of the machine learning regression models specifically includes: applying Gaussian noise perturbation to the numerical features of the training set, with the noise amplitude being 5% of the standard deviation of the feature, in order to enhance the generalization ability of the model.

[0011] Optionally, the verification and evaluation of the performance of each model, and the selection of the model with the best overall performance as the final prediction model, is specifically based on the coefficient of determination R. 2 The root mean square error (RMSE), mean absolute error (MAE), and relative prediction deviation (RPD) are used to select the model with the best overall performance as the final prediction model.

[0012] Optionally, the optimal prediction model is a support vector machine model, which uses a radial basis kernel function, a regularization parameter C=1, a kernel function parameter gamma=0.1, and an epsilon tolerance band of 0.1.

[0013] Optionally, the soil profile conductivity distribution map is presented in two-dimensional or three-dimensional form, using a uniform color scale, and only displays data located within the inverted trapezoidal effective detection area.

[0014] Secondly, this application provides a soil profile conductivity prediction device based on resistivity tomography, comprising: The module for acquiring real resistivity spatial distribution data is used to acquire real resistivity spatial distribution data. The measured electrical conductivity, water content and bulk density determination module is used to collect undisturbed soil profile samples, and then measure the measured electrical conductivity, water content and bulk density of each layer after the undisturbed soil profile samples are layered according to depth. The matching dataset determination module is used to spatially align the measured electrical conductivity, water content, and bulk density with the real resistivity spatial distribution data based on the three-dimensional spatial coordinates of their sampling points, forming a matching dataset containing coordinates, resistivity, and the corresponding measured soil electrical conductivity, water content, and bulk density. The machine learning regression model building module is used to build multiple machine learning regression models based on the matching dataset. Each machine learning regression model uses different feature combinations as input variables. The feature combinations include: resistivity + depth, resistivity + depth + water content, resistivity + depth + bulk density, and resistivity + depth + water content + bulk density. The target variable is conductivity. The model performance verification module is used to divide the training set and the validation set of each machine learning regression model by stratified sampling, and to verify and evaluate the performance of each model, and select the model with the best overall performance as the final prediction model. The predicted conductivity determination module is used to apply the optimal prediction model to the real resistivity and depth data obtained by resistivity tomography in the new monitoring area, and output the predicted conductivity of each grid point. The soil profile electrical conductivity distribution map generation module is used to perform spatial interpolation on the predicted electrical conductivity points to generate a continuous soil profile electrical conductivity distribution map.

[0015] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a method and apparatus for predicting the electrical conductivity of soil profiles based on resistivity tomography, which has the following significant advantages: High vertical resolution and large detection depth: Compared with the electromagnetic induction method, this application uses resistivity tomography to achieve a vertical resolution of less than 10 cm and a detection depth of more than 5 m, which meets the needs of deep soil salinization monitoring.

[0016] Non-invasive and highly efficient: Only electrodes need to be laid on the ground, eliminating the need for large-scale drilling and significantly improving the efficiency of field operations.

[0017] High prediction accuracy and strong generalization ability: Through multi-feature combination and machine learning model optimization, especially the SVM model, the performance in validation is stable (R 2 (≈0.63, RPD≈1.71), which is better than the traditional linear method.

[0018] Flexible adaptability: It can choose whether to introduce auxiliary variables such as water content and bulk density according to actual conditions, so as to achieve a balance between speed and accuracy.

[0019] Highly professional visualization: Reliable and intuitive conductivity profiles are generated through Kriging interpolation and "inverted trapezoidal" effective region constraints, facilitating decision support. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application 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 this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is an application environment diagram of a soil profile conductivity prediction method based on resistivity tomography in one embodiment of this application. Figure 2 A schematic diagram showing the ranking of the relative importance of variables in a soil profile electrical conductivity prediction model according to an embodiment of this application; Figure 3 A schematic diagram showing the comparison between predicted and measured soil EC values ​​based on a support vector machine model (variable combination being soil resistivity, depth, and water content) in an embodiment of this application. Figure 4A comparison chart of predicted and measured soil EC values ​​at borehole points based on a support vector machine model, provided as an embodiment of this application (modeling variables include soil resistivity, depth, and water content). Figure 5 This application provides a soil profile electrical conductivity distribution map based on SVM model prediction for one embodiment of the present application. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] In one exemplary embodiment, such as Figure 1 As shown, a method for predicting soil profile conductivity based on resistivity tomography is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method includes the following steps: Step 101: Obtain the actual spatial distribution data of resistivity.

[0025] Specifically, an electrode array is deployed on the surface of the target area using a resistivity tomography measurement system. The system injects a low-frequency stable current (I) into the ground and simultaneously measures the potential difference (ΔV) between each electrode pair. The apparent resistivity is calculated based on Ohm's law, using the following formula: Where K is a device coefficient related to the geometry of the electrode arrangement. Subsequently, the data processing equipment uses inversion algorithms such as the least squares method to input the original apparent resistivity dataset into the Res2Dinv software. The model resistivity values ​​are updated by minimizing the objective function (such as a least squares objective function with smoothing constraints). The process stops when the RMSE error no longer decreases significantly or when the preset number of iterations is reached. The resistivity output in the last iteration represents the model parameters at that point, generating true resistivity distribution data that reflects the actual underground electrical structure. The equations used by the Res2Dinv software are shown below: ; in, represents the objective function; m represents the logarithm of the true resistivity; Represents the data weight matrix; This represents the measured apparent resistivity; This represents the theoretical apparent resistivity calculated by numerically solving the electric field equations. Represents the regularization factor; This represents the smoothing constraint matrix of the model.

[0026] Step 102: Collect undisturbed soil profile samples, and then measure the electrical conductivity, water content and bulk density of each layer after dividing the undisturbed soil profile samples into layers according to depth.

[0027] Specifically, undisturbed soil profile samples are collected using a drilling rig at pre-defined spatial locations along the resistivity tomography system's measurement line. In the laboratory, the soil profile samples are layered at preset depth intervals, and the soil conductivity, soil moisture content, and soil bulk density of each layer are measured.

[0028] Step 103: Spatially align the measured electrical conductivity, water content, and bulk density with the actual resistivity spatial distribution data based on the three-dimensional spatial coordinates of their sampling points to form a matching dataset containing coordinates, resistivity, and the corresponding measured soil electrical conductivity, water content, and bulk density.

[0029] The data processing equipment precisely matches and aligns the laboratory-measured soil physicochemical property data (electrical conductivity, water content, and bulk density) with the resistivity distribution data generated in step 101, based on the three-dimensional spatial coordinates (X, Y, Z) of the sampling points. This creates a series of spatially corresponding data pairs, each containing coordinate information, resistivity (r), and the corresponding measured soil electrical conductivity, water content, and bulk density. The purpose of this step is to establish the spatial correspondence between soil electrical conductivity and soil resistivity, providing a high-quality dataset for subsequent modeling.

[0030] In summary, steps 101-103 are completed collaboratively by the apparent resistivity measured by the resistivity tomography measurement equipment, the water content, bulk density, and conductivity obtained from borehole sampling, and the data processing equipment. The purpose is to obtain a standardized, spatially matched benchmark dataset required for training and validating the prediction model.

[0031] Step 104: Based on the matching dataset, construct multiple machine learning regression models, each with different feature combinations as input variables; the feature combinations include: resistivity + depth, resistivity + depth + water content, resistivity + depth + bulk density, and resistivity + depth + water content + bulk density, with conductivity as the target variable.

[0032] Step 105: Divide the training set and validation set of each machine learning regression model using stratified sampling, evaluate the performance of each model, and select the model with the best overall performance as the final prediction model.

[0033] Specifically, this application constructs a systematic machine learning workflow for predictive modeling and feature engineering evaluation of soil electrical conductivity. The entire process follows rigorous experimental design principles to ensure the reproducibility and statistical significance of the results, as follows: Experiment initialization and traceability management: The system generates unique run identifiers using timestamps, establishing a complete experiment log system. Each run creates a CSV log file containing metadata such as the run identifier, timestamp, and status, enabling full tracking of the experiment process. The output directory uses a hierarchical structure of "D: / GIS / " to ensure systematic archiving and storage of experimental results.

[0034] Data Preprocessing and Quality Control: The workflow loads raw observation data (depth, resistivity, bulk density, water content, and conductivity) from an Excel file and performs systematic data cleaning. For missing resistivity values, multiple interpolation is used to fill in the gaps. Distribution checks, sensitivity analysis, and uncertainty reports are employed to replace traditional model precision calculations. Results show good overall performance, strong stability of the interpolated data, and high consistency with the original data distribution, ensuring data integrity. For the target variable, soil EC, the 3-interquartile range criterion based on Tukey's method is applied to identify and handle extreme outliers, eliminating their potential interference with model training. Simultaneously, the overall standard deviation of soil EC is calculated to establish a benchmark for subsequent standardized evaluation of model performance.

[0035] Data augmentation and regularization strategies: To improve the model's generalization ability, controlled data perturbation is applied to the numerical features of the training set. A Gaussian noise injection mechanism is employed, with the noise amplitude set to 5% of the feature standard deviation. This increases the diversity of the training data through local perturbation of the feature space. This strategy strictly protects the target variable from modification, ensuring the effectiveness of supervised learning, while introducing appropriate uncertainty to prevent model overfitting.

[0036] Multi-factor experimental design architecture: A complete 4×4 factorial experimental design was constructed to systematically evaluate the interaction between different feature combinations and algorithms. A progressive strategy was adopted for feature dimension design, gradually expanding from basic electrical-geometric feature combinations to a multi-dimensional feature space including soil physical properties. Four machine learning algorithms were selected: Random Forest (RF), Extreme Gradient Boosting Tree (XGBoost), Support Vector Machine (SVM), and Linear Regression (LM). Comparative analysis of 16 independent modeling scenarios comprehensively evaluated the applicability of each method.

[0037] Training and validation framework and statistical significance assurance: A stratified sampling method based on the quantiles of the target variable is employed, dividing the dataset into an 80% training set and a 20% validation set to ensure consistency between the two subsets in the target distribution. To reduce the impact of randomness in a single random partition, 10 repeated experiments are conducted, each time using a different random seed to re-split the data and train the model. Model hyperparameters are adaptively configured according to algorithm characteristics and feature dimensions; for example, the `mtry` parameter in random forests is dynamically adjusted as the number of features changes.

[0038] Model-independent feature importance quantification: A permutation-based feature importance calculation method is adopted to establish a cross-algorithm comparable feature contribution evaluation system. Each feature is independently and randomly permuted, and its importance is quantified by calculating the relative degradation of model performance, expressed as a percentage increase in mean squared error (IncMSE). This method is independent of specific algorithm implementations, providing a unified evaluation standard for feature selection under different modeling techniques.

[0039] A multi-dimensional performance evaluation index system is established, comprising a three-tiered system of performance indicators including goodness of fit, error measurement, and standardized evaluation. The coefficient of determination measures the proportion of the variance of the target variable explained by the model, while the root mean square error and mean absolute error assess prediction accuracy from the perspectives of squared error and absolute error, respectively. Relative prediction bias is specifically introduced as a core evaluation index, which provides a quality grading standard directly related to the accuracy of measurement techniques by using the ratio of the original standard deviation of soil EC to the model prediction error.

[0040] The training accuracy and independent prediction accuracy of the soil profile electrical conductivity prediction model are detailed in Tables 1 and 2: Table 1 Training accuracy of soil profile electrical conductivity prediction model

[0041] Table 2. Independent validation accuracy of the soil profile electrical conductivity prediction model

[0042] Comprehensive Results Integration and Knowledge Discovery: Statistical aggregation of the results from 10 repeated experiments was performed to calculate the average value of each performance indicator and the average contribution of feature importance. A feature importance matrix was constructed, and feature contribution patterns under different modeling scenarios were visualized using heatmaps. Based on the validation set performance, 16 modeling schemes were ranked, and the optimal feature-algorithm combination was identified, providing clear modeling suggestions for practical applications.

[0043] See Figure 2 , Figure 2The results show that, among the four models (a) random forest model, b) extreme gradient boosting tree model, c) support vector machine model, and d) linear regression model, soil resistivity (r) is the most important predictor variable, followed by depth, while water content (wc) and bulk density (bd) are relatively less important.

[0044] Figure 3 This image shows a comparison between predicted and measured soil EC values ​​based on a Support Vector Machine (SVM) model (with soil resistivity, depth, and water content as variables). The predicted and measured values ​​were analyzed using the sum of 10 experiments conducted on the validation set. The analysis revealed that among the four models, the SVM model had the highest prediction accuracy. 2 The optimal value is 0.61.

[0045] Figure 4 The image shows a comparison between the predicted and measured values ​​of soil EC at borehole points based on the support vector machine model (the modeling variables are soil resistivity, depth, and water content). It was found that the predicted soil EC values ​​have a high degree of overlap with the actual values ​​and can reflect the actual situation well, which further verifies the reliability of the SVM model.

[0046] Persistent Output and Robustness: Structured output files are generated, including detailed performance summary tables, feature importance ranking tables, and diagnostic charts. All charts are saved in high-resolution PNG format (600 DPI) to ensure publication-quality visualization. Complete analysis results are serialized and stored in RDS format, supporting subsequent in-depth analysis and extended research. The entire workflow incorporates multi-layered anomaly handling mechanisms, including file existence checks, data integrity verification, and training process monitoring, ensuring robust execution of the analysis process.

[0047] The framework described above automates the entire process from data loading to model evaluation. Through systematic experimental design, rigorous statistical validation, and standardized output formats, it establishes a standardized analytical paradigm for soil electrical conductivity prediction modeling. Its modular architecture supports flexible expansion of feature sets and rapid iteration of modeling algorithms, providing a reliable methodological foundation for high-precision prediction and feature engineering optimization of soil electrical conductivity.

[0048] Step 106: Apply the optimal prediction model to the real resistivity and depth data obtained by resistivity tomography in the new monitoring area, and output the predicted conductivity of each grid point.

[0049] Step 105 yielded the optimal model and its parameters. The optimal model is the SVM model, which employs a radial basis function kernel for soil EC prediction. The core parameter settings reflect a balance between model complexity and prediction accuracy. Specifically, the regularization parameter C is set to 1, using moderate regularization to prevent overfitting; the kernel parameter gamma is set to 0.1, ensuring the model remains moderately sensitive to local patterns; and the epsilon parameter is set to 0.1, defining the tolerance range for regression prediction, allowing a prediction error of ±0.1 to be excluded from the loss. This parameter combination constitutes a support vector regression configuration with moderate complexity and moderate accuracy. It considers the potential nonlinear variations in soil electrical conductivity while avoiding overfitting risks on small sample datasets through appropriate regularization, providing a relatively robust modeling foundation for soil EC prediction.

[0050] In the actual processing, the input feature variables (known quantities) include resistivity (r) and depth (Z) obtained from step 101. To improve model accuracy and physical interpretability, the preferred combination of input feature variables also includes soil moisture content and soil bulk density. The target variable (the unknown quantity to be predicted) is soil electrical conductivity.

[0051] The data processing device randomly divides the matching dataset into a training set and independent test sets. Then, one or more machine learning algorithms are selected and trained. Preferably, a support vector machine (SVM) algorithm is used. Taking one feature combination as an example, the SVM algorithm aims to find an optimal hyperplane or nonlinear function f that minimizes the prediction error. Its core optimization problem can be expressed as minimizing the following objective function: ; The constraints are: ; in, The input feature vector represents soil resistivity, depth, water content, and bulk density. Here, w and b are the corresponding measured conductivity, and w and b are the model parameters. A kernel function (such as a radial basis function RBF) maps features to a high-dimensional space, where C is the penalty parameter. These are slack variables. Data processing equipment optimizes model hyperparameters such as kernel function parameters and penalty parameters using techniques such as cross-validation.

[0052] The performance of the trained model is evaluated using an independent test set, including the coefficient of determination (R²). 2The key aspect of this application is the parallel construction of multiple prediction models based on different feature combinations and the evaluation of model performance. These feature combinations include resistivity and depth, resistivity, depth and water content, resistivity, depth and bulk density, and resistivity, depth, water content and bulk density. Based on the evaluation results, the data processing equipment automatically selects the model with the highest prediction accuracy and best generalization ability, determining it as the final soil profile conductivity prediction model. The purpose of this step is to determine the optimal model scheme by comparing and validating the models, balancing verification accuracy with the feasibility of rapid measurement.

[0053] Step 107: Perform spatial interpolation on the predicted conductivity points to generate a continuous soil profile conductivity distribution map.

[0054] This step is performed by the data processing equipment. Its function is to apply the optimal model to a new area that only requires resistivity tomography measurements, thereby enabling rapid acquisition and visualization of the soil profile conductivity distribution in that area. The specific steps are as follows: 1. New Area Data Input. For new target monitoring areas, obtain the resistivity distribution data and spatial coordinate information of that area. If an optimized model including soil moisture content or bulk density characteristics is used, this data should be obtained simultaneously.

[0055] 2. Soil Profile Electrical Conductivity Prediction. The data processing equipment inputs characteristic data such as resistivity and depth of the new area into the optimal soil electrical conductivity prediction model. The model automatically outputs a predicted soil electrical conductivity for each two-dimensional or three-dimensional grid point. This process achieves batch, rapid, and quantitative conversion from soil resistivity to soil electrical conductivity.

[0056] 3. Spatial Interpolation and Distribution Map Generation. The data processing equipment uses Kriging interpolation and other methods to perform spatial interpolation on discrete prediction points, generating a continuous and smooth spatial distribution surface of soil electrical conductivity. This surface is then rendered into a two-dimensional profile or a three-dimensional image. During mapping, a uniform color scale is used, and the effective data range is constrained to the "inverted trapezoidal" region (i.e., a region wider at the top and narrower at the bottom) where resistivity tomography detection has the highest theoretical accuracy. This shields low-precision data at the edges, improving the reliability and professionalism of the resulting map. The final generated soil profile electrical conductivity distribution map clearly and intuitively displays the characteristics of soil profile electrical conductivity distribution. See details... Figure 5 .

[0057] Through the above technical solutions, this application systematically combines the high-resolution imaging capability of resistivity tomography with the intelligent decoupling capability of machine learning for the first time, and constructs a complete and operational method for rapid quantitative prediction of soil profile conductivity. This effectively solves the industry problems of low efficiency of traditional methods, low vertical resolution and limited detection depth of existing electromagnetic induction methods.

[0058] Implementing the method described in this application has the following significant advantages: First, this method is non-invasive and highly efficient, eliminating the need for large-scale soil profile excavation or drilling. Soil resistivity data can be rapidly acquired simply by deploying an electrode array on the surface, significantly improving the efficiency of field monitoring. Second, this technology demonstrates high vertical resolution and deep detection capabilities. Compared to electromagnetic induction methods, resistivity tomography provides a vertical resolution of less than 10 cm and a detection depth exceeding 5 meters. Finally, the constructed machine learning model exhibits strong generalization ability, especially the support vector machine model, which demonstrates stable validation accuracy and high prediction accuracy, outperforming linear regression and some tree models, showcasing its advantage in handling nonlinear relationships. Furthermore, the method offers flexible and adaptable feature combination design, allowing for the selection of different feature combinations based on actual needs. It can utilize soil resistivity and depth for maximum rapid acquisition, or incorporate parameters such as soil moisture content to enhance model interpretability and generalization ability.

[0059] Based on the same inventive concept, this application also provides a soil profile conductivity prediction device based on resistivity tomography for implementing the aforementioned method for predicting soil profile conductivity based on resistivity tomography. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the soil profile conductivity prediction device based on resistivity tomography provided below can be found in the limitations of the soil profile conductivity prediction method based on resistivity tomography described above, and will not be repeated here.

[0060] In one exemplary embodiment, a soil profile conductivity prediction device based on resistivity tomography is provided, comprising: The module for acquiring real resistivity spatial distribution data is used to acquire real resistivity spatial distribution data. The measured electrical conductivity, water content and bulk density determination module is used to collect undisturbed soil profile samples, and then measure the measured electrical conductivity, water content and bulk density of each layer after the undisturbed soil profile samples are layered according to depth. The matching dataset determination module is used to spatially align the measured electrical conductivity, water content, and bulk density with the real resistivity spatial distribution data based on the three-dimensional spatial coordinates of their sampling points, forming a matching dataset containing coordinates, resistivity, and the corresponding measured soil electrical conductivity, water content, and bulk density. The machine learning regression model building module is used to build multiple machine learning regression models based on the matching dataset. Each machine learning regression model uses different feature combinations as input variables. The feature combinations include: resistivity + depth, resistivity + depth + water content, resistivity + depth + bulk density, and resistivity + depth + water content + bulk density. The target variable is conductivity. The model performance verification module is used to divide the training set and the validation set of each machine learning regression model by stratified sampling, and to verify and evaluate the performance of each model, and select the model with the best overall performance as the final prediction model. The predicted conductivity determination module is used to apply the optimal prediction model to the real resistivity and depth data obtained by resistivity tomography in the new monitoring area, and output the predicted conductivity of each grid point. The soil profile electrical conductivity distribution map generation module is used to perform spatial interpolation on the predicted electrical conductivity points to generate a continuous soil profile electrical conductivity distribution map.

[0061] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0062] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting soil profile electrical conductivity based on resistivity tomography, characterized in that, The method for predicting soil profile electrical conductivity based on resistivity tomography includes: Obtain true spatial distribution data of resistivity; Uncirculated soil profile samples were collected, and the uncirculated soil profile samples were layered according to depth to determine the measured electrical conductivity, water content and bulk density of each layer; The measured electrical conductivity, water content, and bulk density are spatially aligned with the actual resistivity spatial distribution data based on the three-dimensional spatial coordinates of their sampling points, forming a matching dataset containing coordinates, resistivity, and the corresponding measured soil electrical conductivity, water content, and bulk density. Based on the matching dataset, multiple machine learning regression models are constructed, each with different feature combinations as input variables. The feature combinations include: resistivity + depth, resistivity + depth + water content, resistivity + depth + bulk density, and resistivity + depth + water content + bulk density, with conductivity as the target variable. The training set and validation set of each machine learning regression model are divided by stratified sampling, and the performance of each model is evaluated and the model with the best overall performance is selected as the final prediction model. The optimal prediction model is applied to the real resistivity and depth data obtained by resistivity tomography in the new monitoring area, and the predicted conductivity of each grid point is output. Spatial interpolation is performed on the predicted conductivity points to generate a continuous soil profile conductivity distribution map.

2. The method for predicting soil profile conductivity based on resistivity tomography according to claim 1, characterized in that, The process of obtaining the actual spatial distribution data of resistivity specifically includes the following steps: An electrode array was deployed on the surface of the target area, and a low-frequency stable current was injected using a resistivity tomography system to measure the potential difference between each electrode pair. The apparent resistivity is calculated based on the potential difference, and the actual resistivity spatial distribution data is generated by the least squares inversion algorithm.

3. The method for predicting soil profile conductivity based on resistivity tomography according to claim 2, characterized in that, The apparent resistivity is calculated based on the potential difference, and the actual spatial distribution data of resistivity is generated using the least squares inversion algorithm. The specific formula used is as follows: ； in, represents the objective function; m represents the logarithm of the true resistivity; Represents the data weight matrix; This represents the measured apparent resistivity; This represents the theoretical apparent resistivity calculated by numerically solving the electric field equations. Represents the regularization factor; This represents the smoothing constraint matrix of the model.

4. The method for predicting soil profile conductivity based on resistivity tomography according to claim 1, characterized in that, The spatial alignment is achieved through a unified three-dimensional coordinate system, ensuring that the positional error between each measured sample point and the center point of the resistivity grid in the X, Y, and Z directions is less than 15 cm.

5. The method for predicting soil profile conductivity based on resistivity tomography according to claim 1, characterized in that, The multiple machine learning regression models include Support Vector Machine (SVM), Random Forest (RF), Extreme Gradient Boosting Tree (XGBoost), and Linear Regression (LM).

6. The method for predicting soil profile conductivity based on resistivity tomography according to claim 1, characterized in that, The hierarchical sampling method for dividing the training and validation sets of each machine learning regression model includes: applying Gaussian noise perturbation to the numerical features of the training set, with the noise amplitude being 5% of the standard deviation of the feature, in order to enhance the generalization ability of the model.

7. The method for predicting soil profile conductivity based on resistivity tomography according to claim 1, characterized in that, The verification and evaluation of each model's performance, and the selection of the model with the best overall performance as the final prediction model, are specifically based on the coefficient of determination R. 2 The root mean square error (RMSE), mean absolute error (MAE), and relative prediction deviation (RPD) are used to select the model with the best overall performance as the final prediction model.

8. The method for predicting soil profile conductivity based on resistivity tomography according to claim 1, characterized in that, The optimal prediction model is a support vector machine model, which uses a radial basis kernel function, a regularization parameter C=1, a kernel function parameter gamma=0.1, and an epsilon tolerance band of 0.

1.

9. The method for predicting soil profile conductivity based on resistivity tomography according to claim 1, characterized in that, The soil profile conductivity distribution map is presented in two-dimensional or three-dimensional form, using a uniform color scale, and only displays data located within the inverted trapezoidal effective detection area.

10. A soil profile conductivity prediction device based on resistivity tomography, characterized in that, The soil profile conductivity prediction device based on resistivity tomography includes: The module for acquiring real resistivity spatial distribution data is used to acquire real resistivity spatial distribution data. The measured electrical conductivity, water content and bulk density determination module is used to collect undisturbed soil profile samples, and then measure the measured electrical conductivity, water content and bulk density of each layer after the undisturbed soil profile samples are layered according to depth. The matching dataset determination module is used to spatially align the measured electrical conductivity, water content, and bulk density with the real resistivity spatial distribution data based on the three-dimensional spatial coordinates of their sampling points, forming a matching dataset containing coordinates, resistivity, and the corresponding measured soil electrical conductivity, water content, and bulk density. The machine learning regression model building module is used to build multiple machine learning regression models based on the matching dataset, with each machine learning regression model using different feature combinations as input variables; The feature combinations include: resistivity + depth, resistivity + depth + water content, resistivity + depth + bulk density, and resistivity + depth + water content + bulk density, with conductivity as the target variable. The model performance verification module is used to divide the training set and the validation set of each machine learning regression model by stratified sampling, and to verify and evaluate the performance of each model, and select the model with the best overall performance as the final prediction model. The predicted conductivity determination module is used to apply the optimal prediction model to the real resistivity and depth data obtained by resistivity tomography in the new monitoring area, and output the predicted conductivity of each grid point. The soil profile electrical conductivity distribution map generation module is used to perform spatial interpolation on the predicted electrical conductivity points to generate a continuous soil profile electrical conductivity distribution map.

Images