CN-122024901-A - Resistivity tomography-based soil profile conductivity prediction method and device

CN122024901ACN 122024901 ACN122024901 ACN 122024901ACN-122024901-A

Abstract

The application discloses a soil profile conductivity prediction method and device based on resistivity tomography, and relates to the field of soil physics and geophysical exploration, wherein the method comprises the steps of obtaining real resistivity spatial distribution data; collecting undisturbed soil profile samples, measuring the actual measured conductivity, water content and volume weight, carrying out spatial alignment on the measured data and the real resistivity spatial distribution data to construct a plurality of machine learning regression models, dividing a training set and a verification set by adopting layered sampling for each machine learning regression model, verifying, 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 carrying out spatial interpolation on the predicted conductivity points to generate a continuous soil profile conductivity distribution map. The application solves the problems of low efficiency, poor vertical resolution and limited detection depth of the traditional method by using the electromagnetic induction method, and has the remarkable advantages of non-invasiveness, high precision, high resolution, deep detection capability and the like.

Inventors

WU HUAYONG
YIN XIANYUAN
LIU XUELIAN
LU YING
YANG FEI
ZHAO YUGUO
ZHANG GANLIN

Assignees

中国科学院南京土壤研究所
衢州市美丽乡村建设中心

Dates

Publication Date: 20260512
Application Date: 20260129

Claims (10)

1. The soil profile conductivity prediction method based on resistivity tomography is characterized by comprising the following steps of: acquiring real resistivity space distribution data; collecting an undisturbed soil profile sample, layering the undisturbed soil profile sample according to depth, and measuring the actual measured conductivity, water content and volume weight of each layer; Carrying out space alignment on the actually measured conductivity, the water content and the volume weight according to the three-dimensional space coordinates of the sampling points and the real resistivity space distribution data to form a matching data set containing coordinates, resistivity and corresponding actually measured soil conductivity, water content and volume weight; Based on the matching data set, constructing a plurality of machine learning regression models, wherein each machine learning regression model takes different characteristic combinations as input variables, and the characteristic combinations comprise resistivity, depth, volume weight, resistivity, depth, water content and volume weight, and the target variables are electric conductivity; dividing a training set and a verification set by adopting hierarchical sampling for each machine learning regression model, verifying and evaluating the performance of each model, and selecting a model with optimal comprehensive performance as a final prediction model; Applying the optimal prediction model to real resistivity and depth data obtained by resistivity tomography of a new monitoring area, and outputting the predicted conductivity of each grid point; and performing spatial interpolation on the predicted conductivity points to generate a continuous soil profile conductivity distribution map.
2. The method for predicting the conductivity of a soil profile based on resistivity tomography as claimed in claim 1, wherein the step of obtaining the real resistivity spatial distribution data comprises the steps of: Arranging an electrode array on the surface of a target area, injecting low-frequency stable current by using a resistivity tomography system, and measuring potential difference between each electrode pair; And calculating apparent resistivity based on the potential difference, and generating real resistivity spatial distribution data through a least square inversion algorithm.
3. The resistivity tomography-based soil profile conductivity prediction method of claim 2, wherein the apparent resistivity is calculated based on the potential difference and the true resistivity spatial distribution data is generated by a least squares inversion algorithm using the following formula: ; Wherein, the M represents the logarithmic value of the true resistivity; representing a data weight matrix; Representing the measured apparent resistivity; representing the theoretical apparent resistivity calculated by numerically solving the electric field equation; Representing a regularization factor; Representing a model smoothness constraint matrix.
4. The resistivity tomography-based soil profile conductivity prediction method of claim 1, wherein the spatial alignment is achieved by a unified three-dimensional coordinate system, ensuring that each measured sample point has a position error of less than 15 cm from a resistivity grid center point in a X, Y, Z direction.
5. The resistivity tomography based soil profile conductivity prediction method of claim 1, wherein the plurality of machine learning regression models includes a support vector machine SVM, a random forest RF, an extreme gradient lift tree XGBoost, and a linear regression LM.
6. The resistivity tomography-based soil profile conductivity prediction method as claimed in claim 1, wherein the employing hierarchical sampling to divide the training set and the validation set for each machine learning regression model specifically comprises applying gaussian noise disturbance to a numerical feature of the training set with a noise amplitude of 5% of the standard deviation of the feature to enhance model generalization ability.
7. The method for predicting the conductivity of a soil profile based on resistivity tomography according to claim 1, wherein the verification evaluates the performance of each model, and selects a model with optimal comprehensive performance as a final prediction model, specifically, selects a model with optimal comprehensive performance as a final prediction model according to a determination coefficient R 2 , a root mean square error RMSE, an average absolute error MAE and a relative prediction deviation RPD index.
8. The method for predicting the conductivity of a soil profile based on resistivity tomography according to claim 1, wherein the optimal prediction model is a support vector machine model, a radial basis function is adopted, a regularization parameter c=1, a kernel parameter gamma=0.1, and an epsilon tolerance band is 0.1.
9. The resistivity tomography-based soil profile conductivity prediction method of claim 1, wherein the soil profile conductivity profile is presented in two or three dimensions and is uniformly color coded and only displays data within an inverted trapezoid effective detection area.
10. The device for predicting the conductivity of the soil profile based on the resistivity tomography is characterized by comprising the following components The real resistivity spatial distribution data acquisition module is used for acquiring real resistivity spatial distribution data; The measured conductivity, water content and volume weight determining module is used for collecting an undisturbed soil profile sample, layering the undisturbed soil profile sample according to depth, and measuring the measured conductivity, water content and volume weight of each layer; The matching data set determining module is used for carrying out space alignment on the actually measured conductivity, the water content and the volume weight according to the three-dimensional space coordinates of the sampling points and the real resistivity space distribution data to form a matching data set containing coordinates, resistivity and corresponding actually measured soil conductivity, water content and volume weight; The machine learning regression model construction module is used for constructing a plurality of machine learning regression models based on the matching data set, wherein each machine learning regression model takes different characteristic combinations as input variables, and the characteristic combinations comprise resistivity + depth, resistivity + depth + water content, resistivity + depth + volume weight and resistivity + depth + water content + volume weight, and the target variables are electric conductivity; the model performance verification module is used for dividing a training set and a verification set by adopting hierarchical sampling for each machine learning regression model, verifying and evaluating the performance of each model, and selecting a model with optimal comprehensive performance as a final prediction model; The predicted conductivity determining module is used for applying the optimal prediction model to real resistivity and depth data acquired by resistivity tomography of a new monitoring area and outputting the predicted conductivity of each grid point; And the soil profile conductivity distribution map generation module is used for carrying out spatial interpolation on the predicted conductivity points to generate a continuous soil profile conductivity distribution map.

Description

Resistivity tomography-based soil profile conductivity prediction method and device Technical Field The application relates to the technical field of soil physics and geophysical exploration, in particular to a soil profile conductivity prediction method and device based on resistivity tomography. Background The soil conductivity is a key index reflecting the salt content of the soil, and has important influence on agricultural production and ecological environment. The field in-situ rapid acquisition of the conductivity of the soil profile faces a serious challenge, and the rapid acquisition of the spatial distribution characteristics of the conductivity of the two-dimensional profile of the soil is difficult to realize by the traditional method of the soil profile mining or drilling sampling combined with the indoor detection of the point position scale due to the strong heterogeneity of the spatial distribution of the salinity of the soil, so that the rapid monitoring of the salinization information of the soil profile is restricted. At present, the method based on electromagnetic induction can be used for rapidly acquiring the conductivity of a soil section in situ in the open air, and the method is used for rapidly acquiring the conductivity of the soil section by transmitting a primary magnetic field from the ground surface to the soil at the lower part and acquiring a primary magnetic field signal and a secondary magnetic field signal, calculating the apparent conductivity of the soil section by utilizing a conversion formula between the magnetic field intensity and the conductivity, and establishing a prediction model between the apparent conductivity and the actually measured conductivity by combining algorithms such as linear regression. Although the method utilizes electromagnetic induction technology to realize field in-situ rapid acquisition, the method has low vertical resolution and limited detection depth (usually within 1.5 meters), the apparent conductivity is a weighted average value of a certain depth from the earth surface to the underground rather than a numerical value of a specific depth, and the method has difficulty in realizing rapid acquisition of the soil profile conductivity which meets the requirements of high vertical resolution and large detection depth. Therefore, there is a need for a method of acquiring conductivity of a soil profile that combines high resolution, large detection depth, non-invasiveness and high prediction accuracy. Disclosure of Invention The application aims to provide a soil profile conductivity prediction method and device based on resistivity tomography, which are used for solving the problems of low vertical resolution, limited detection depth, low efficiency and the like in the prior art. In order to achieve the above object, the present application provides the following solutions: In a first aspect, the application provides a soil profile conductivity prediction method based on resistivity tomography, comprising: acquiring real resistivity space distribution data; collecting an undisturbed soil profile sample, layering the undisturbed soil profile sample according to depth, and measuring the actual measured conductivity, water content and volume weight of each layer; Carrying out space alignment on the actually measured conductivity, the water content and the volume weight according to the three-dimensional space coordinates of the sampling points and the real resistivity space distribution data to form a matching data set containing coordinates, resistivity and corresponding actually measured soil conductivity, water content and volume weight; Based on the matching data set, constructing a plurality of machine learning regression models, wherein each machine learning regression model takes different characteristic combinations as input variables, and the characteristic combinations comprise resistivity, depth, volume weight, resistivity, depth, water content and volume weight, and the target variables are electric conductivity; dividing a training set and a verification set by adopting hierarchical sampling for each machine learning regression model, verifying and evaluating the performance of each model, and selecting a model with optimal comprehensive performance as a final prediction model; Applying the optimal prediction model to real resistivity and depth data obtained by resistivity tomography of a new monitoring area, and outputting the predicted conductivity of each grid point; and performing spatial interpolation on the predicted conductivity points to generate a continuous soil profile conductivity distribution map. Optionally, the acquiring real resistivity spatial distribution data specifically includes the following steps: Arranging an electrode array on the surface of a target area, injecting low-frequency stable current by using a resistivity tomography system, and measuring potential difference between each electrode pair; And ca