Development of a 3d spatial prediction method for site soil pollutant concentration based on multi-source auxiliary data

Abstract

The present application relates to a method for predicting the concentration of soil pollutants in a three-dimensional space based on multi-source auxiliary data development, which is directed to the multi-source auxiliary data of site pollution such as functional area layout and underground physical properties, uses different derivation methods for further development, obtains more derived auxiliary data representing the prior information of the three-dimensional spatial distribution of pollutant concentration, and then establishes a quantitative relationship model between the derived auxiliary data and the pollutant concentration, so as to realize the three-dimensional spatial prediction of the concentration of soil pollutants in the site. The present application solves the problems of weak correlation between the original auxiliary data and the soil pollutant concentration in the prior art, and the insufficient development and utilization of the site pollution prior information implied in the auxiliary data, and significantly improves the prediction accuracy of the three-dimensional spatial concentration of soil pollutants in the site.

Description

Technical Field

[0001] This invention relates to a three-dimensional spatial prediction method for soil pollutant concentrations in a site based on multi-source auxiliary data, and belongs to the field of soil pollution prediction technology. Background Technology

[0002] Current methods for predicting the three-dimensional spatial distribution of soil pollutant concentrations at sites mainly utilize traditional spatial interpolation methods such as three-dimensional inverse distance weighted interpolation (IDW) and three-dimensional kriging interpolation. However, these methods suffer from significant problems, including a strong smoothing effect (high concentration values ​​are underestimated, and low concentration values ​​are overestimated) and low accuracy. Furthermore, existing spatial interpolation methods for predicting the three-dimensional distribution of soil pollutant concentrations at sites rely solely on borehole soil pollutant concentration data, lacking the integration of auxiliary site pollution data. This further restricts the improvement of prediction accuracy.

[0003] Multi-source auxiliary data, such as site functional zoning layout and subsurface properties, provides a wealth of prior information on the spatial distribution of pollutant concentrations, and these auxiliary data exhibit a certain spatial correlation with site soil pollutant concentrations. For example, functional zoning layout can reflect the impact of site pollution sources (production processes, waste disposal, etc.) on the spatial distribution of pollutant concentrations; electromagnetic data of subsurface media obtained by geophysical exploration methods such as electromagnetic induction (EM) and resistivity tomography (ERT) can reflect the differences in electromagnetic properties between contaminated and uncontaminated soils. Furthermore, compared to soil pollutant concentration data obtained from borehole sampling, these auxiliary data have higher spatial coverage density and completeness, and are acquired at lower cost and faster speed. Therefore, achieving three-dimensional spatial prediction based on the quantitative relationship between auxiliary data and pollutant concentrations is one of the potential technical approaches to overcome the limitations of traditional spatial interpolation methods in predicting the three-dimensional distribution of site soil pollutant concentrations. However, using the quantitative relationship between auxiliary data such as functional zoning layout and subsurface properties and site soil pollutant concentrations to predict the three-dimensional spatial distribution of pollutant concentrations still faces the following challenges:

[0004] (1) The correlation between functional zone layout and raw data of underground physical properties and soil pollutant concentration may be weak, which may lead to a decrease in spatial prediction accuracy, or even lower than the prediction accuracy of traditional interpolation methods. For example, the site space is very complex, and ground debris, soil properties, geological conditions, etc. will affect the EM and ERT detection results, thereby weakening the correlation between raw auxiliary data of underground physical properties such as conductivity and resistivity and soil pollutant concentration, resulting in low spatial prediction accuracy;

[0005] (2) The prior information on site pollution implied in the auxiliary data has not been fully developed and utilized. For example, the functional zone layout is a two-dimensional discrete data, and the current method of simply using dummy variables to identify the quantitative relationship between it and soil pollutant concentration is still difficult to fully reflect the intensity of its influence on the three-dimensional spatial distribution of pollutant concentration. Summary of the Invention

[0006] The technical problem to be solved by this invention is to provide a three-dimensional spatial prediction method for soil pollutant concentration based on multi-source auxiliary data. It adopts a novel design strategy that can effectively improve the prediction efficiency of the three-dimensional spatial distribution of soil pollutant concentration.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention designs a three-dimensional spatial prediction method for soil pollutant concentration based on multi-source auxiliary data. Through steps A to D, a three-dimensional prediction model of soil pollutant concentration in the target area is obtained. Through step i, the three-dimensional prediction model of soil pollutant concentration is applied to realize the prediction of the three-dimensional distribution of soil pollutant concentration in the underground space of the target area.

[0008] Step A. Obtain the measured values ​​of soil pollutant concentrations at preset depths and locations of preset boreholes within the target area, as well as the multi-source auxiliary datasets containing preset data types for the target area, and then proceed to Step B;

[0009] Step B. Based on the pre-defined three-dimensional raster of the underground space of the target area, the multi-source auxiliary dataset corresponding to the target area is preprocessed using a spatial interpolation method, the multi-source auxiliary dataset corresponding to the target area is updated, and then proceed to step C;

[0010] Step C. Based on the preset three-dimensional raster of the underground space of the target area, perform data derivation processing on the multi-source auxiliary dataset corresponding to the target area to obtain the three-dimensional derived auxiliary data of the target area, and then proceed to step D;

[0011] Step D. Based on the three-dimensional derived auxiliary data of the target area, obtain the derived auxiliary data corresponding to each soil sample point. Use the derived auxiliary data corresponding to the soil sample point as the prediction variable and the measured value of the soil pollutant concentration corresponding to the soil sample point as the target variable. Train the preset machine learning model to obtain the three-dimensional prediction model of soil pollutant concentration in the target area.

[0012] Step i: For each grid cell in the underground space of the target area, obtain the derived auxiliary data corresponding to the grid cell, apply the three-dimensional prediction model of soil pollutant concentration to obtain the predicted value of soil pollutant concentration for each grid cell, and thus realize the three-dimensional prediction of soil pollutant concentration in the underground space of the target area.

[0013] As a preferred technical solution of the present invention, it further includes step E as follows: after step D is completed, step E is performed;

[0014] Step E. Apply the independent validation method to calculate the root mean square error between the predicted soil pollutant concentration values ​​of soil samples obtained from the three-dimensional prediction model of soil pollutant concentration and their corresponding measured soil pollutant concentration values, thus obtaining the prediction accuracy of the three-dimensional prediction model of soil pollutant concentration.

[0015] As a preferred technical solution of the present invention: the target area in step A corresponds to a multi-source auxiliary dataset containing preset data types, including the area range of preset functional areas in the target area, the two-dimensional conductivity measurement point data of preset electromagnetic detection frequencies in the target area, and the two-dimensional cross-sectional inversion resistivity data of preset measurement line positions in the target area.

[0016] As a preferred technical solution of the present invention: the underground conductivity of the target area is measured by electromagnetic induction (EM) to obtain two-dimensional measurement data of conductivity at preset electromagnetic detection frequencies in the target area.

[0017] As a preferred technical solution of the present invention: the underground resistivity of the target area is measured by resistivity tomography (ERT), and two-dimensional cross-sectional inversion resistivity data corresponding to each preset survey line position in the target area are obtained.

[0018] As a preferred technical solution of the present invention: In step B, based on the two-dimensional measurement data of conductivity at preset electromagnetic detection frequencies corresponding to the target area in the multi-source auxiliary data, the two-dimensional grid in the horizontal direction of the three-dimensional grid of the underground space of the target area is interpolated by the spatial interpolation method to obtain the two-dimensional spatial distribution of soil conductivity at each electromagnetic detection frequency corresponding to the target area.

[0019] Based on the two-dimensional cross-sectional resistivity data of the target area corresponding to each preset survey line location in the multi-source auxiliary dataset, the three-dimensional spatial distribution of resistivity in the target area is obtained by interpolating the three-dimensional raster of the underground space through spatial interpolation method.

[0020] As a preferred embodiment of the present invention, the spatial interpolation method in step B is the IDW interpolation method or the Kriging interpolation method.

[0021] As a preferred technical solution of the present invention: step C includes the following steps C1 to C4;

[0022] Step C1. Based on the preset area range of various functional areas in the target area, for each grid in the target area, if the grid is located within the preset functional area, assign a value of 1 to the grid; if the grid is not located within the preset functional area, assign a value of 0 to the grid. This completes the assignment of values ​​to each grid in the target area, thus forming the three-dimensional derived auxiliary data of the functional area type corresponding to the target area. At the same time, for each grid in the target area, calculate the nearest Euclidean distance from the grid to the edge of each functional area, thus obtaining the nearest Euclidean distance from each grid in the target area to the edge of each functional area, thus forming the three-dimensional derived auxiliary data of the distance between the functional areas corresponding to the target area. Then proceed to step C2.

[0023] Step C2. Based on the two-dimensional spatial distribution of conductivity at various electromagnetic detection frequencies in the target area, obtain the minimum, maximum, median, mean, and variance of conductivity between each grid in the horizontal direction of the target area at each electromagnetic detection frequency, thus forming two-dimensional derived auxiliary data of conductivity statistics for the target area; simultaneously, based on the sliding of two-dimensional sliding windows of preset sizes in the horizontal direction of the target area, obtain the minimum, maximum, median, mean, variance, and gradient in the two-dimensional direction of conductivity between each grid in each two-dimensional sliding window at each electromagnetic detection frequency, thus forming two-dimensional derived auxiliary data of conductivity statistics for the sliding window corresponding to the target area; then proceed to step C3;

[0024] Step C3. Based on the three-dimensional spatial distribution of resistivity corresponding to the target area, obtain the minimum, maximum, median, mean, and variance of resistivity between each depth of each grid in the horizontal direction of the target area, forming two-dimensional derived auxiliary data of resistivity statistics between each depth of the target area; simultaneously, based on the sliding of three-dimensional sliding windows of preset sizes on the three-dimensional grid of the underground space of the target area, obtain the minimum, maximum, median, mean, variance, and gradient in the three-dimensional direction of resistivity between each grid in each three-dimensional sliding window, forming three-dimensional derived auxiliary data of resistivity statistics of the sliding window corresponding to the target area; then proceed to step C4;

[0025] Step C4. For the two-dimensional derived auxiliary data of conductivity statistics, sliding window conductivity statistics, and resistivity statistics between depths corresponding to the target area, copy them along the depth direction of the three-dimensional grid to obtain the three-dimensional derived auxiliary data of conductivity statistics between depths, sliding window conductivity statistics, and resistivity corresponding to the target area. Combine these with the three-dimensional derived auxiliary data of functional area type, functional area distance, and sliding window resistivity statistics corresponding to the target area to form the three-dimensional derived auxiliary data of the target area. Then proceed to step D.

[0026] As a preferred technical solution of the present invention: step D includes the following steps D1 to D3;

[0027] Step D1. Standardize the three-dimensional derived auxiliary data of the target area, and use principal component analysis to obtain the principal component data in the derived auxiliary data, and then proceed to step D2;

[0028] Step D2. Based on the three-dimensional derived auxiliary data of the target area, obtain the data of each principal component data type corresponding to each soil sample point, and construct the derived auxiliary data corresponding to each soil sample point, and then proceed to step D3;

[0029] Step D3. Based on each soil sampling point, using the derived auxiliary data corresponding to the soil sampling point as the prediction variable and the measured value of the soil pollutant concentration corresponding to the soil sampling point as the target variable, train the support vector machine regression model to obtain a three-dimensional prediction model of soil pollutant concentration in the target area.

[0030] The three-dimensional spatial prediction method for site soil pollutant concentration based on multi-source auxiliary data described in this invention has the following technical advantages compared with existing technologies:

[0031] The present invention presents a three-dimensional spatial prediction method for soil pollutant concentration based on multi-source auxiliary data. This method utilizes different derivation methods to further develop multi-source auxiliary data on site pollution, such as functional area layout and underground properties, to obtain more derived auxiliary data characterizing the three-dimensional spatial distribution of soil pollutant concentration. Then, machine learning is used to establish a quantitative relationship model between the derived auxiliary data and pollutant concentration, achieving high-precision three-dimensional spatial prediction of site soil pollutant concentration. This solves the problems of weak correlation between original auxiliary data and soil pollutant concentration in existing technologies, and the insufficient development and utilization of implicit site pollution prior information in the auxiliary data. It improves the accuracy of three-dimensional spatial prediction of pollutant concentration and overcomes the problems of strong smoothing effect and low prediction accuracy in traditional interpolation methods. Attached Figure Description

[0032] Figure 1 This is a flowchart illustrating the implementation of the three-dimensional spatial prediction method for site soil pollutant concentration developed based on multi-source auxiliary data according to the present invention.

[0033] Figure 2 This is a layout diagram of the functional areas of the site and the distribution of borehole sampling points according to an embodiment of the present invention;

[0034] Figure 3 This is a three-dimensional spatial interpolation map of the ERT resistivity of the site according to an embodiment of the present invention;

[0035] Figure 4Examples of three-dimensional derived auxiliary data for functional area layout (left) and resistivity three-dimensional derived auxiliary data (right);

[0036] Figure 5 This is a comparison of the three-dimensional spatial distribution of Zn concentration in the soil of an embodiment predicted by the present invention (left) and by traditional IDW interpolation (right). Detailed Implementation

[0037] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0038] This invention designs a three-dimensional spatial prediction method for site soil pollutant concentration based on multi-source auxiliary data, such as... Figure 1 As shown, in practical applications, steps A to D are specifically executed to obtain a three-dimensional prediction model of soil pollutant concentration in the target area.

[0039] Step A. Obtain the measured values ​​of soil pollutant concentrations at preset depths and locations of preset boreholes within the target area, as well as the multi-source auxiliary datasets containing preset data types for the target area, and then proceed to Step B.

[0040] In the application, the specific design of the preset data types contained in the multi-source auxiliary dataset includes the area range of preset functional areas in the target area, the two-dimensional conductivity measurement point data of the target area at preset electromagnetic detection frequencies, and the two-dimensional cross-sectional inversion resistivity data of the target area at preset measurement line positions.

[0041] In practical applications, high-resolution remote sensing imagery, combined with site reconnaissance, is used to interpret the types and boundaries of various functional zones within the site, obtaining the pre-defined area ranges for each functional zone. Electromagnetic induction (EM) is used to measure the underground conductivity of the target area, obtaining two-dimensional conductivity measurement data at pre-defined electromagnetic detection frequencies. EM generates a primary magnetic field on the surface through an electric current. This primary magnetic field induces a secondary magnetic field in the heterogeneous soil of the site. By receiving this secondary induced magnetic field at the surface, the conductivity information of the underground medium can be obtained. The conductivity information at different detection frequencies reflects the conductivity of the underground medium at different detection depths; the lower the detection frequency, the deeper the detection depth. EM is widely used in contaminated site investigations, and its detection cost is extremely low. Therefore, the two-dimensional measurement points for EM should be set up to cover the target site as evenly as possible.

[0042] Regarding the two-dimensional cross-sectional resistivity data, the specific application design utilizes resistivity tomography (ERT) to measure the subsurface resistivity of the target area, obtaining two-dimensional cross-sectional resistivity data for each preset survey line location within the target area. Changes in soil pollutant concentration often alter resistivity. ERT involves passing direct current underground and observing the potential difference between the supplied current intensity and the measuring electrodes to calculate the apparent resistivity. During measurement, the geographical coordinates (x, y), electrode spacing, and electrode setup information of the electrodes are recorded, thereby obtaining two-dimensional cross-sectional resistivity data for different ERT survey line locations through inversion. The two-dimensional cross-sectional resistivity data obtained after inversion from different ERT survey lines need to be correlated with the geographical coordinates (x, y) of each electrode and the inversion depth z to obtain three-dimensional scatter plot resistivity data. ERT detection is also relatively inexpensive; therefore, the survey line setup should cover the target site as uniformly as possible.

[0043] Step B. Based on the pre-defined three-dimensional raster of the underground space of the target area, the multi-source auxiliary dataset corresponding to the target area is preprocessed using a spatial interpolation method, the multi-source auxiliary dataset corresponding to the target area is updated, and then proceed to step C.

[0044] In the specific implementation of step B above, based on the two-dimensional measurement data of conductivity at preset electromagnetic detection frequencies corresponding to the target area in the multi-source auxiliary data center, spatial interpolation methods such as IDW interpolation or Kriging interpolation are used to interpolate the two-dimensional grid in the horizontal direction of the three-dimensional grid of the underground space of the target area to obtain the two-dimensional spatial distribution of soil conductivity at each electromagnetic detection frequency corresponding to the target area.

[0045] Based on the two-dimensional cross-sectional resistivity data of the target area corresponding to each preset survey line location in the multi-source auxiliary dataset, the three-dimensional spatial distribution of resistivity in the target area is obtained by interpolating the three-dimensional raster of the underground space through spatial interpolation methods such as IDW interpolation or Kriging interpolation.

[0046] Step C. Based on the preset three-dimensional raster of the underground space of the target area, perform data derivation processing on the multi-source auxiliary dataset corresponding to the target area to obtain the three-dimensional derived auxiliary data of the target area, and then proceed to step D.

[0047] In practical applications, step C above includes steps C1 to C4.

[0048] Step C1. Based on the preset area range of various functional zones in the target area, for each grid in the target area, if the grid is located within the preset functional zone, assign a value of 1 to the grid; if the grid is not located within the preset functional zone, assign a value of 0 to the grid. This completes the assignment of values ​​to each grid in the target area, thus forming three-dimensional derived auxiliary data of the functional zone type corresponding to the target area. At the same time, for each grid in the target area, calculate the nearest Euclidean distance from the grid to the edge of each functional zone, thus obtaining the nearest Euclidean distance from each grid in the target area to the edge of each functional zone, thus forming three-dimensional derived auxiliary data of the distance to the functional zone corresponding to the target area. Different functional zone types and distances to the edges of functional zones reflect the intensity of the impact of potential pollution sources on the soil pollutant concentration. For example, the soil characteristic pollutant concentrations in waste dumping areas and production workshops are often high, and the closer the distance, the higher the pollutant concentration usually is. Then proceed to step C2.

[0049] Step C2. Based on the two-dimensional spatial distribution of conductivity at various electromagnetic detection frequencies in the target area, obtain the minimum, maximum, median, mean, and variance of conductivity between each grid in the horizontal direction of the target area at each electromagnetic detection frequency, thus forming two-dimensional derived auxiliary data of conductivity statistics for the target area; simultaneously, based on the sliding of two-dimensional sliding windows of preset sizes in the horizontal direction of the target area, obtain the minimum, maximum, median, mean, variance, and gradient in the two-dimensional direction of conductivity between each grid in each two-dimensional sliding window at each electromagnetic detection frequency, thus forming two-dimensional derived auxiliary data of conductivity statistics for the sliding window corresponding to the target area; then proceed to step C3.

[0050] Step C3. Based on the three-dimensional spatial distribution of resistivity corresponding to the target area, obtain the minimum, maximum, median, mean, and variance of resistivity between each depth of each grid in the horizontal direction of the target area, forming two-dimensional derived auxiliary data of resistivity statistics between each depth of the target area; simultaneously, based on the sliding of three-dimensional sliding windows of preset sizes on the three-dimensional grid of the underground space of the target area, obtain the minimum, maximum, median, mean, variance, and gradient in the three-dimensional direction of resistivity between each grid in each three-dimensional sliding window, forming three-dimensional derived auxiliary data of resistivity statistics of the sliding window corresponding to the target area; then proceed to step C4.

[0051] Step C4. For the two-dimensional derived auxiliary data of conductivity statistics, sliding window conductivity statistics, and resistivity statistics between depths corresponding to the target area, copy them along the depth direction of the three-dimensional grid to obtain the three-dimensional derived auxiliary data of conductivity statistics between depths, sliding window conductivity statistics, and resistivity corresponding to the target area. Combine these with the three-dimensional derived auxiliary data of functional area type, functional area distance, and sliding window resistivity statistics corresponding to the target area to form the three-dimensional derived auxiliary data of the target area. Then proceed to step D.

[0052] The purpose of step C is to further mine the information hidden in the two-dimensional and three-dimensional auxiliary data that is more strongly correlated with the spatial distribution of pollutant concentration, so as to use it to build a spatial prediction model and improve the accuracy of pollutant concentration prediction.

[0053] Step D. Based on the three-dimensional derived auxiliary data of the target area, obtain the derived auxiliary data corresponding to each soil sample point. Use the derived auxiliary data corresponding to the soil sample point as the prediction variable and the measured value of the soil pollutant concentration corresponding to the soil sample point as the target variable. Train the preset machine learning model to obtain the three-dimensional prediction model of soil pollutant concentration in the target area, and then proceed to step E.

[0054] In practical applications, step D above is specifically executed as steps D1 to D3.

[0055] Step D1. Since machine learning prediction models are easily affected by the order of magnitude of the predicted variables, it is necessary to standardize the derived auxiliary data before constructing the three-dimensional prediction model of pollutant concentration in order to eliminate the order of magnitude differences between different derived auxiliary data. Therefore, the three-dimensional derived auxiliary data of the target area is standardized, and principal component analysis is used to obtain the principal component data in the derived auxiliary data, and then proceed to step D2.

[0056] Step D2. Based on the three-dimensional derived auxiliary data of the target area, obtain the data of each principal component data type corresponding to each soil sample point, and construct the derived auxiliary data corresponding to each soil sample point, and then proceed to step D3.

[0057] Step D3. Based on each soil sampling point, using the derived auxiliary data corresponding to the soil sampling point as the prediction variable and the measured value of the soil pollutant concentration corresponding to the soil sampling point as the target variable, train the support vector machine regression model to obtain a three-dimensional prediction model of soil pollutant concentration in the target area.

[0058] Support vector machine regression can map low-dimensional data to a high-dimensional feature space, and then make predictions by finding a hyperplane that satisfies the regression. When building the model, cross-validation of the training set data is used to determine the optimal values ​​of the model parameters.

[0059] Step E. Apply the independent validation method to calculate the root mean square error between the predicted soil pollutant concentration values ​​of soil samples obtained from the three-dimensional prediction model of soil pollutant concentration and their corresponding measured soil pollutant concentration values, thus obtaining the prediction accuracy of the three-dimensional prediction model of soil pollutant concentration.

[0060] Based on the acquisition of the three-dimensional prediction model of soil pollutant concentration in the target area, step i is further used to apply the three-dimensional prediction model of soil pollutant concentration to achieve the prediction of the three-dimensional distribution of soil pollutant concentration in the underground space of the target area.

[0061] Step i: For each grid cell in the underground space of the target area, obtain the derived auxiliary data corresponding to the grid cell, apply the three-dimensional prediction model of soil pollutant concentration to obtain the predicted value of soil pollutant concentration for each grid cell, and thus realize the three-dimensional prediction of soil pollutant concentration in the underground space of the target area.

[0062] In practical application, taking a rubber factory site as an example, the above design method is used to predict the three-dimensional spatial distribution of soil Zn concentration. The implementation process is as follows: Figure 1 As shown, the specific implementation method is described in detail below:

[0063] Step A. Obtain the measured values ​​of soil pollutant concentrations at preset depths and locations of preset boreholes within the target area, as well as the multi-source auxiliary datasets containing preset data types for the target area, and then proceed to Step B.

[0064] (1.1) Drilling was carried out at the example site, and a total of 282 soil samples were collected at different depths. The three-dimensional coordinates (x, y, z) of the soil sampling points were recorded and laboratory analysis was conducted to determine the soil Zn concentration.

[0065] (1.2) Using Gaofen-2 remote sensing imagery within the example site area, and combined with on-site reconnaissance, the categories and boundaries of each functional zone were interpreted and identified to obtain a site functional zone layout map, as shown below. Figure 2 As shown;

[0066] (1.3) The two-dimensional measurement data of the underground conductivity of the site were collected using a multi-frequency electromagnetic detector. Five detection frequencies were used in the EM measurement: 474Hz, 1625Hz, 5475Hz, 18575Hz and 63025Hz. The two-dimensional measurement data of the EM conductivity at the five detection frequencies were obtained respectively.

[0067] (1.4) The apparent resistivity of the underground surface was measured using a high-density resistivity meter. A total of 14 ERT lines were laid out on the site, with the shortest line being 130m and the longest being 150m. The electrode spacing was 2m. The apparent resistivity was measured using a Wenner-Schlumberger electrode device, and the geographic coordinates (x, y) of the electrode positions were recorded. The apparent resistivity data of each ERT line were inverted using AGI EarthImager software to obtain the resistivity data of 14 two-dimensional sections.

[0068] Step B. Based on the pre-defined three-dimensional raster of the underground space of the target area, the multi-source auxiliary dataset corresponding to the target area is preprocessed using a spatial interpolation method, the multi-source auxiliary dataset corresponding to the target area is updated, and then proceed to step C.

[0069] (2.1) Based on the horizontal and vertical range of the site, the grid resolution is set to 2m×2m×1m (length×width×depth), and the underground space of the site is discretized into a three-dimensional grid. Subsequent auxiliary data preprocessing, derivative development and three-dimensional spatial prediction of soil Zn concentration are all based on this three-dimensional grid.

[0070] (2.2) Using the two-dimensional grid in the horizontal direction of the three-dimensional grid as the grid point positions to be interpolated, IDW interpolation is used to interpolate the conductivity two-dimensional measurement point data of the five detection frequencies into two-dimensional grid data, resulting in five conductivity two-dimensional grid data, which serve as two-dimensional auxiliary data; using the three-dimensional grid as the grid point positions to be interpolated, three-dimensional kriging interpolation is used to interpolate the resistivity data of 14 two-dimensional sections into one three-dimensional grid data, such as... Figure 3 As shown, this serves as three-dimensional auxiliary data.

[0071] Step C. Based on the preset three-dimensional raster of the underground space of the target area, perform data derivation processing on the multi-source auxiliary dataset corresponding to the target area to obtain the three-dimensional derived auxiliary data of the target area, and then proceed to step D.

[0072] (3.1) For the site functional area layout map, firstly, the 10 functional area types are converted into 0-1 binary discrete data. Then, the grids in the 3D mesh are assigned values. That is, for a certain functional area type, when the grid point is within the underground range of that functional area, it is assigned a value of 1; otherwise, it is assigned a value of 0. This process is repeated for all 10 functional area types, resulting in 10 sets of 3D derived auxiliary data for functional area types. Secondly, the nearest Euclidean distance from the grid point in the 3D mesh to the edge of the polygon of each of the 10 functional areas is calculated, resulting in 10 sets of 3D derived auxiliary data for functional area distances, such as... Figure 4 (Left) shows the three-dimensional distribution of the grid and the Euclidean distance between the grid and the gelation workshop in the three-dimensional mesh;

[0073] (3.2) For the two-dimensional spatial distribution data of EM conductivity at 5 detection frequencies, firstly, the minimum, maximum, median, mean and variance of conductivity at different detection frequencies were calculated, resulting in 5 two-dimensional auxiliary data of conductivity; secondly, 3 two-dimensional sliding windows of sizes 3×3, 9×9 and 11×11 were set, and for the two-dimensional spatial distribution data of EM conductivity at each detection frequency, the minimum, maximum, median, mean, variance and gradients in the x and y directions of conductivity data within different sliding windows were calculated, resulting in 105 two-dimensional auxiliary data of EM conductivity.

[0074] (3.3) For the three-dimensional spatial distribution data of ERT resistivity, firstly, the two-dimensional distribution data of resistivity at different depths were extracted, and the minimum, maximum, median, mean, and variance of resistivity at different depths were calculated, resulting in 5 two-dimensional derivative auxiliary data of resistivity. Secondly, three three-dimensional sliding windows with sizes of 3×3×3, 9×9×3, and 11×11×3 were set, and then the minimum, maximum, median, mean, variance, and gradients in the x, y, and z directions of resistivity data within the sliding window were calculated under different sliding windows, resulting in 24 three-dimensional derivative auxiliary data of resistivity sliding window statistics, such as... Figure 4 (Right) shows the three-dimensional spatial distribution of the maximum resistivity within a sliding window of 11×11×3;

[0075] (3.4) For the 110 EM conductivity two-dimensional derivative auxiliary data and the 5 resistivity two-dimensional derivative auxiliary data calculated by depth, they were copied sequentially according to the depth layer of the three-dimensional grid and extended to the three-dimensional space respectively, so as to be consistent with the defined underground three-dimensional grid.

[0076] After the above steps, a total of 159 three-dimensional derived auxiliary data (i.e., 159 three-dimensional raster data) were obtained in this embodiment, including 20 functional area layout derived auxiliary data, 110 EM conductivity derived auxiliary data, and 29 ERT resistivity derived auxiliary data.

[0077] Step D. Based on the three-dimensional derived auxiliary data of the target area, obtain the derived auxiliary data corresponding to each soil sample point. Using the derived auxiliary data corresponding to the soil sample point as the prediction variable and the measured value of the soil pollutant concentration corresponding to the soil sample point as the target variable, train the preset machine learning model to obtain the three-dimensional prediction model of soil pollutant concentration in the target area.

[0078] (4.1) First, the derived auxiliary data (i.e., 159 three-dimensional raster data) were standardized to eliminate the difference in magnitude between different variables. Then, the dimensionality of the derived auxiliary data was reduced by principal component analysis, and a total of 14 principal components with eigenvalues ​​≥1 were obtained, with a cumulative variance contribution rate of 96%.

[0079] (4.2) Extract the principal component scores of the derived auxiliary data corresponding to the locations of 282 soil sampling points, construct a soil sampling point dataset with Zn concentration data and 14 principal components, and randomly divide the soil sampling point dataset into training set and validation set in an 8:2 ratio;

[0080] (4.3) Based on the soil sample training set, the principal component scores of the derived auxiliary data are used as the prediction variable and the soil Zn concentration is used as the target variable. A three-dimensional spatial prediction model of soil Zn concentration is established by using support vector machine regression. During the model construction process, its optimal parameters are set through the 5-fold cross-validation results of the training set data, as follows: the kernel function is set to Gaussian radial basis function, the penalty factor C is set to 100, and the insensitive loss function gamma is set to 0.1.

[0081] Step E. Apply the independent validation method to calculate the root mean square error between the predicted soil pollutant concentration values ​​of soil samples obtained from the three-dimensional prediction model of soil pollutant concentration and their corresponding measured soil pollutant concentration values, thus obtaining the prediction accuracy of the three-dimensional prediction model of soil pollutant concentration.

[0082] (5.1) Using the three-dimensional prediction model of soil Zn concentration constructed in step four, the grid positions in the defined three-dimensional grid are predicted one by one to obtain the three-dimensional spatial distribution map of soil Zn concentration in the site, as shown in the figure. Figure 5 As shown on the left;

[0083] (5.2) Using the three-dimensional prediction model of soil Zn concentration constructed in step 4, the Zn concentration at the location of soil sampling points in the validation set is predicted, and the root mean square error (RMSE) between the predicted and measured Zn concentration at the location of soil sampling points in the validation set is calculated to evaluate the spatial prediction accuracy.

[0084] In this embodiment, the original data of the three-dimensional resistivity distribution ( Figure 3 The correlation coefficient between resistivity and soil Zn concentration was only 0.29, while the auxiliary data derived from the maximum value of the 11×11×3 sliding window of resistivity obtained after auxiliary data development ( Figure 4 The correlation coefficients between the derived auxiliary data (right) and the 3×3×3 sliding window variance-derived auxiliary data and soil Zn concentration increased to 0.5 and 0.51, respectively; in addition, the correlation coefficient between the derived auxiliary data of resistivity z-direction (i.e., vertical direction) gradient and Zn concentration increased to 0.31. Compared with the original auxiliary data, the correlation between the derived auxiliary data and soil pollution concentration was significantly improved.

[0085] In this embodiment, the RMSE of the three-dimensional prediction of soil Zn concentration independently validated is 175.01 mg / kg, while the RMSE of the traditional IDW spatial interpolation method using the same independent validation dataset is as high as 242.33 mg / kg. Compared with the traditional IDW interpolation method, the prediction error of the method of the present invention is reduced by approximately 28%. Furthermore, the method of the present invention ( Figure 5 (Left) and traditional IDW interpolation method ( Figure 5 The predicted maximum Zn concentrations (right) are 841.00 mg / kg and 861.00 mg / kg, respectively, while the minimum values ​​are 3.15 mg / kg and 42.6 mg / kg, respectively. This demonstrates that the present invention can significantly reduce the "underestimation" effect of pollutant concentration, thereby reducing the smoothing effect and better characterizing the detailed features of the three-dimensional spatial distribution of pollutant concentration.

[0086] This embodiment demonstrates that by further developing site pollution auxiliary data such as functional area layout and underground physical properties, its correlation with pollutant concentration can be improved, and the prior information of site pollution implied in the auxiliary data can be fully utilized, thereby improving the three-dimensional spatial prediction accuracy of site soil pollutant concentration and reducing the smoothing effect of prediction results.

[0087] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A three-dimensional spatial prediction method for site soil pollutant concentration based on multi-source auxiliary data, characterized by: Through steps A to D, a three-dimensional prediction model of soil pollutant concentration in the target area is obtained. Through step i, the three-dimensional prediction model of soil pollutant concentration is applied to realize the prediction of the three-dimensional distribution of soil pollutant concentration in the underground space of the target area. Step A. Obtain the measured values ​​of soil pollutant concentrations at each preset depth and at each preset borehole location within the target area, as well as the multi-source auxiliary dataset containing preset data types for the target area, and then proceed to Step B; Step B. Based on the pre-defined three-dimensional raster of the underground space of the target area, the multi-source auxiliary dataset corresponding to the target area is preprocessed using a spatial interpolation method, the multi-source auxiliary dataset corresponding to the target area is updated, and then proceed to step C; Step C. Based on the preset three-dimensional raster of the underground space of the target area, perform data derivation processing on the multi-source auxiliary dataset corresponding to the target area to obtain three-dimensional derived auxiliary data of the target area, and then proceed to step D; Step C above includes steps C1 to C4 as follows; Step C1. Based on the preset area range of various functional areas in the target area, for each grid in the target area, if the grid is located within the preset functional area, assign a value of 1 to the grid; if the grid is not located within the preset functional area, assign a value of 0 to the grid. This completes the assignment of values ​​to each grid in the target area, thus forming the three-dimensional derived auxiliary data of the functional area type corresponding to the target area. At the same time, for each grid in the target area, calculate the nearest Euclidean distance from the grid to the edge of each functional area, thus obtaining the nearest Euclidean distance from each grid in the target area to the edge of each functional area, thus forming the three-dimensional derived auxiliary data of the distance between the functional areas corresponding to the target area. Then proceed to step C2. Step C2. Based on the two-dimensional spatial distribution of conductivity at various electromagnetic detection frequencies in the target area, obtain the minimum, maximum, median, mean, and variance of conductivity between each grid in the horizontal direction of the target area at each electromagnetic detection frequency, thus forming two-dimensional derived auxiliary data of conductivity statistics for the target area; simultaneously, based on the sliding of two-dimensional sliding windows of preset sizes in the horizontal direction of the target area, obtain the minimum, maximum, median, mean, variance, and gradient in the two-dimensional direction of conductivity between each grid in each two-dimensional sliding window at each electromagnetic detection frequency, thus forming two-dimensional derived auxiliary data of conductivity statistics for the sliding window corresponding to the target area; then proceed to step C3; Step C3. Based on the three-dimensional spatial distribution of resistivity corresponding to the target area, obtain the minimum, maximum, median, mean, and variance of resistivity between each depth of each grid in the horizontal direction of the target area, forming two-dimensional derived auxiliary data of resistivity statistics between each depth of the target area; simultaneously, based on the sliding of three-dimensional sliding windows of preset sizes on the three-dimensional grid of the underground space of the target area, obtain the minimum, maximum, median, mean, variance, and gradient in the three-dimensional direction of resistivity between each grid in each three-dimensional sliding window, forming three-dimensional derived auxiliary data of resistivity statistics of the sliding window corresponding to the target area; then proceed to step C4; Step C4. For the two-dimensional derived auxiliary data of conductivity statistics, sliding window conductivity statistics, and resistivity statistics between depths corresponding to the target area, copy them along the depth direction of the three-dimensional grid to obtain the three-dimensional derived auxiliary data of conductivity statistics between depths, sliding window conductivity statistics, and resistivity corresponding to the target area. Combine these with the three-dimensional derived auxiliary data of functional area type, functional area distance, and sliding window resistivity statistics corresponding to the target area to form the three-dimensional derived auxiliary data of the target area. Then proceed to step D. Step D. Based on the three-dimensional derived auxiliary data of the target area, obtain the derived auxiliary data corresponding to each soil sample point. Use the derived auxiliary data corresponding to the soil sample point as the prediction variable and the measured value of the soil pollutant concentration corresponding to the soil sample point as the target variable. Train the preset machine learning model to obtain the three-dimensional prediction model of soil pollutant concentration in the target area. Step i: For each grid cell in the underground space of the target area, obtain the derived auxiliary data corresponding to the grid cell, apply the three-dimensional prediction model of soil pollutant concentration to obtain the predicted value of soil pollutant concentration for each grid cell, and thus realize the three-dimensional prediction of soil pollutant concentration in the underground space of the target area.

2. The three-dimensional spatial prediction method for site soil pollutant concentration based on multi-source auxiliary data as described in claim 1, characterized in that: It also includes step E as follows: after step D is completed, proceed to step E; Step E. Apply the independent validation method to calculate the root mean square error between the predicted soil pollutant concentration values ​​of soil samples obtained from the three-dimensional prediction model of soil pollutant concentration and their corresponding measured soil pollutant concentration values, thus obtaining the prediction accuracy of the three-dimensional prediction model of soil pollutant concentration.

3. The three-dimensional spatial prediction method for site soil pollutant concentration based on multi-source auxiliary data as described in claim 1, characterized in that: In step A, the target area corresponds to a multi-source auxiliary dataset containing preset data types, including the area range of preset functional areas in the target area, the two-dimensional conductivity measurement point data of preset electromagnetic detection frequencies in the target area, and the two-dimensional cross-sectional inversion resistivity data of preset measurement line positions in the target area.

4. The three-dimensional spatial prediction method for site soil pollutant concentration based on multi-source auxiliary data as described in claim 3, characterized in that: The underground conductivity of the target area is measured by electromagnetic induction (EM) to obtain two-dimensional measurement data of conductivity at preset electromagnetic detection frequencies in the target area.

5. The three-dimensional spatial prediction method for site soil pollutant concentration based on multi-source auxiliary data as described in claim 3, characterized in that: The resistivity of the subsurface in the target area is measured by resistivity tomography (ERT), and two-dimensional cross-sectional inversion resistivity data corresponding to each preset survey line location in the target area are obtained.

6. The three-dimensional spatial prediction method for site soil pollutant concentration based on multi-source auxiliary data as described in claim 1, characterized in that: In step B, based on the two-dimensional conductivity measurement point data of the target area corresponding to each preset electromagnetic detection frequency in the multi-source auxiliary data, the two-dimensional grid in the horizontal direction of the three-dimensional grid of the underground space of the target area is interpolated by the spatial interpolation method to obtain the two-dimensional spatial distribution of soil conductivity of the target area corresponding to each electromagnetic detection frequency. Based on the two-dimensional cross-sectional resistivity data of the target area corresponding to each preset survey line location in the multi-source auxiliary dataset, the three-dimensional spatial distribution of resistivity in the target area is obtained by interpolating the three-dimensional raster of the underground space through spatial interpolation method.

7. The three-dimensional spatial prediction method for site soil pollutant concentration based on multi-source auxiliary data as described in claim 6, characterized in that: The spatial interpolation method in step B is either the IDW interpolation method or the Kriging interpolation method.

8. The three-dimensional spatial prediction method for site soil pollutant concentration based on multi-source auxiliary data as described in claim 1, characterized in that: Step D includes the following steps D1 to D3; Step D1. Standardize the three-dimensional derived auxiliary data of the target area, and use principal component analysis to obtain the principal component data in the derived auxiliary data, and then proceed to step D2; Step D2. Based on the three-dimensional derived auxiliary data of the target area, obtain the data of each principal component data type corresponding to each soil sample point, and construct the derived auxiliary data corresponding to each soil sample point, and then proceed to step D3; Step D3. Based on each soil sampling point, using the derived auxiliary data corresponding to the soil sampling point as the prediction variable and the measured value of the soil pollutant concentration corresponding to the soil sampling point as the target variable, train the support vector machine regression model to obtain a three-dimensional prediction model of soil pollutant concentration in the target area.

Images