A multi-source multi-scale data fusion method for three-dimensional distribution inversion of regional soil salinization

Abstract

The present application relates to a kind of regional soil salinization three-dimensional distribution inversion multi-source multi-scale data fusion method, based on the grid of target research area, with each grid soil apparent conductivity, in combination with spatial interpolation, the soil salt content of each soil sampling point corresponding each depth position is measured, then in combination with Landsat image dataset contact salinization characteristic spectral index, construct obtain the quantitative interpretation model M1 between soil salt content measured value and soil apparent conductivity, and the quantitative interpretation model M2 between soil apparent conductivity spatial distribution and each salinization characteristic spectral index, and then obtain the three-dimensional spatial distribution of soil salt content in target research area;Design application is centered with near-ground sensing data, near-ground sensing and remote sensing complementary advantages, "point-block-area" multi-scale linkage and nesting mechanism is clear, while obtaining higher precision regional soil salt content three-dimensional spatial distribution, with the advantages of needing less measured sample points, low cost, high precision, simple operation etc..

Description

Technical Field

[0001] This invention relates to a multi-source, multi-scale data fusion method for inverting the three-dimensional distribution of regional soil salinization, belonging to the field of regional soil salinization monitoring technology. Background Technology

[0002] Saline-alkali land is a global agricultural and ecological problem, and has become one of the main constraints on the efficient utilization of water and soil resources, sustainable agricultural development, and ecological environment improvement in arid and semi-arid regions. With the global water shortage in recent years, water-saving technologies such as saline water and brackish water irrigation, and drip irrigation under mulch have been widely promoted and popularized, causing drastic changes in micro-regional and regional water-salt dynamics. The problem of soil salt accumulation has become prominent and is becoming increasingly serious, with some areas already experiencing severe secondary salinization and degradation. This problem exhibits strong spatial variability and a trend of expansion. Therefore, accurately characterizing the three-dimensional spatiotemporal distribution and dynamic evolution of soil salinization, especially in sparsely sampled areas, unsampled areas, or deep soil layers, is of great significance for the precise analysis and inversion of the salinization process and the water-saving and efficient transformation and utilization of saline-alkali land.

[0003] Currently, research on regional soil salinity dynamic monitoring, prediction, and salinization risk early warning has become a hot research area both domestically and internationally. Countries such as the United States, Israel, Australia, and China have achieved a series of research results on soil water and salt prediction, salinization monitoring, and assessment. However, in this field, there is still a lack of comprehensive, multi-scale, non-contact water and salt monitoring technologies for precise spatiotemporal dynamic monitoring of regional soil salinity. Furthermore, due to significant differences in the detection of soil salinization depth, extent, and spatiotemporal resolution by different monitoring methods, such as remote sensing and near-ground sensing, there is a lack of methods for scale linkage between multi-source, multi-scale soil water and salt information, and the related scale nesting and linkage mechanisms are unclear. This, to some extent, restricts the accuracy and reliability of regional soil water and salt movement prediction and salinization risk early warning.

[0004] Currently, there is considerable interest in monitoring and inverting the three-dimensional spatiotemporal distribution process of soil salinization at the regional scale. However, existing methods lack reliable means to integrate and fuse multi-source and multi-scale data, and still rely primarily on processing based on a single source (such as electromagnetic induction data or Landsat imagery). The current coupling methods for multi-scale data are still based on traditional approaches, namely, undifferentiated soil sampling covering the entire region, electromagnetic induction instrument surveys, and remote sensing image interpretation to construct linear or nonlinear coupling models to achieve regional scale expansion, which has low efficiency in practical applications. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a multi-source, multi-scale data fusion method for inverting the three-dimensional distribution of regional soil salinization. Centered on near-ground sensing data, the method complements the advantages of near-ground sensing and remote sensing, and has a clear multi-scale connection and nesting mechanism of "point-block-region". It has the advantages of requiring fewer measured sample points, low cost, high accuracy and simple operation.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention designs a multi-source, multi-scale data fusion method for inverting the three-dimensional distribution of regional soil salinization, and performs the following steps to obtain the three-dimensional spatial distribution of soil salinity in the target study area;

[0007] Step A. Based on the preset gridding of the target study area and the preset measurement point location in each grid, obtain the soil apparent electrical conductivity in the horizontal mode and the soil apparent electrical conductivity in the vertical mode corresponding to each measurement point location, and then proceed to Step B;

[0008] Step B. Obtain the spatial distribution of apparent electrical conductivity of soil in the horizontal mode and apparent electrical conductivity of soil in the vertical mode in the target study area through spatial interpolation, and determine each soil sampling point in the target study area. Then obtain the measured values ​​of soil salinity at each preset depth position corresponding to each soil sampling point, and then proceed to step C.

[0009] Step C. Based on the execution time points of Steps A to B, obtain the Landsat image dataset corresponding to the execution time point of the target study area, calculate the spectral indices of each salinization feature corresponding to the target study area, and then proceed to Step D.

[0010] Step D. Based on each soil sampling point, construct a quantitative interpretation model M1 between the measured soil salinity at each soil depth and the corresponding apparent electrical conductivity of the soil in the horizontal mode and the apparent electrical conductivity of the soil in the vertical mode, and construct a quantitative interpretation model M2 between the spatial distribution of the apparent electrical conductivity of the soil in the horizontal mode and the apparent electrical conductivity of the soil in the vertical mode in the target study area and the spectral indices of each salinization characteristic, and then proceed to step E.

[0011] Step E. Based on the quantitative interpretation model M2 and the spectral indices of salinization characteristics corresponding to the target study area, calculate and update the spatial distribution of soil apparent electrical conductivity in the target study area under the horizontal mode and the vertical mode. Then, combine it with the quantitative interpretation model M1 to calculate the predicted value of soil salinity spatial distribution in the target study area, thus forming the three-dimensional spatial distribution of soil salinity in the target study area.

[0012] As a preferred technical solution of the present invention, it further includes step F as follows: step F is performed after step E is completed;

[0013] Step F. Add the Kriging interpolation values ​​of the three-dimensional spatial distribution of soil salinity in the target study area to the residuals of the measured soil salinity values ​​at each depth of each soil sampling point to obtain the inverted value of the three-dimensional spatial distribution of soil salinity in the target study area.

[0014] As a preferred technical solution of the present invention: in step A, based on the preset gridding of the target research area, the following steps A1 to A3 are performed for each grid of the target research area respectively;

[0015] Step A1. Based on a preset measurement point location in the grid, divide the area in the grid with the measurement point location as the center and a preset radius by applying two mutually orthogonal straight lines whose intersection point is located at the center of the circle to form four quadrants, and then proceed to step A2;

[0016] Step A2. Randomly place a sample point in each of the four quadrants. Combine the position of the measurement point to form five sample points corresponding to the position of the measurement point in the grid. Obtain the apparent electrical conductivity of the soil in the horizontal mode and the apparent electrical conductivity of the soil in the vertical mode corresponding to the five sample points respectively. Then proceed to step A3.

[0017] Step A3. Obtain the apparent average electrical conductivity of the soil in the horizontal mode and the apparent average electrical conductivity of the soil in the vertical mode among the five sample points, and construct the apparent electrical conductivity of the soil in the horizontal mode and the apparent electrical conductivity of the soil in the vertical mode corresponding to the measurement point position in the grid.

[0018] As a preferred technical solution of the present invention: step A further includes steps A2A3-1 to A2A3-2, after step A2 is completed, step A2A3-1 is entered;

[0019] Step A2A3-1. For each of the five sample points, obtain the soil temperature at each preset depth vertically downward from the sample point, and obtain the corresponding average soil temperature to form the soil temperature corresponding to the sample point. That is, obtain the soil temperature corresponding to each sample point, and then proceed to step A2A3-2.

[0020] Step A2A3-2. For each of the five sample points, apply the following formula:

[0021]

[0022] The apparent electrical conductivity EC of the soil under the horizontal mode corresponding to the sampling points ah_r Soil apparent electrical conductivity EC under vertical mode av_r Corrected to the target temperature t respectively 目 Under the horizontal mode, the apparent corrected electrical conductivity EC of the soil ah Soil apparent corrected electrical conductivity EC in vertical mode avThe apparent electrical conductivity of the soil in the horizontal mode and the apparent electrical conductivity of the soil in the vertical mode corresponding to the sample point are updated. That is, the apparent electrical conductivity of the soil in the horizontal mode and the apparent electrical conductivity of the soil in the vertical mode corresponding to each sample point are updated respectively, where t represents the soil temperature corresponding to the sample point, and then proceed to step A3.

[0023] As a preferred technical solution of the present invention: step B includes the following steps B1 to B3;

[0024] Step B1. Based on the apparent electrical conductivity of soil in the horizontal mode and the apparent electrical conductivity of soil in the vertical mode corresponding to the location of each measurement point in the target study area, the apparent electrical conductivity of soil in the horizontal mode and the apparent electrical conductivity of soil in the vertical mode in the target study area are spatially interpolated using the ordinary Kriging method to form the spatial distribution of the apparent electrical conductivity of soil in the horizontal mode and the apparent electrical conductivity of soil in the vertical mode in the target study area, and then proceed to step B2.

[0025] Step B2. Based on the spatial distribution of soil apparent electrical conductivity in the target study area under the horizontal pattern, the variance quadtree method (VQT) is used to determine the locations of each measurement point used for soil sampling in the target study area with a preset maximum variance threshold, thus forming the soil sampling points in the target study area. Then proceed to step B3.

[0026] Step B3. Based on the five sample points corresponding to the measurement points in each grid of the target study area, for each soil sampling point in the target study area, obtain equal mass soil samples at each preset depth position vertically downward from the five sample points corresponding to the soil sampling points, and obtain mixed soil samples from the five sample points at each depth position. Collect the measured soil salinity values ​​of each mixed soil to form the measured soil salinity values ​​at each depth position corresponding to the soil sampling points, that is, obtain the measured soil salinity values ​​at each depth position corresponding to each soil sampling point.

[0027] As a preferred technical solution of the present invention: in step D, the quantitative interpretation model M1 is obtained by following steps D1-1 to D1-2;

[0028] Step D1-1. Construct the measured values ​​of soil salinity at various depths in the target study area and the corresponding apparent electrical conductivity EC under the soil horizontal model. ah apparent electrical conductivity EC of soil under vertical mode av The first model to be trained is as follows:

[0029] S t =a·EC ah +b·EC av +c

[0030] Among them, St The measured values ​​of soil salinity at various soil depths are represented, a and b represent the multiple regression coefficients in the first model to be trained, and c represents the intercept of the first model to be trained. Then proceed to step D1-2.

[0031] Step D1-2. Based on the measured values ​​of soil salinity at each depth corresponding to each soil sampling point and the apparent electrical conductivity of soil in the horizontal mode and the apparent electrical conductivity of soil in the vertical mode, train the first model to be trained to obtain the quantitative interpretation model M1.

[0032] As a preferred technical solution of the present invention: the various salinization characteristic spectral indices include Enhanced Vegetation Index (EVI), Greenness Ratio Vegetation Index (GRVI), Normalized Difference Salinity Index (NDSI), and Salinity Vegetation Index (SRSI).

[0033] As a preferred technical solution of the present invention: in step D, the quantitative interpretation model M2 is obtained by following steps D2-1 to D2-2;

[0034] Step D2-1. Construct the apparent electrical conductivity EC of soil under a horizontal model for the target study area. ah apparent electrical conductivity EC of soil under vertical mode av The second training model relating the spatial distribution of the spectral indices of each salinization characteristic to the spatial distribution of the salinization feature is as follows:

[0035] EC ah =d·EVI+e·GRVI+f·NDSI+g·SRSI+h

[0036] EC av =j·EVI+k·GRVI+l·NDSI+m·SRSI+n

[0037] Where d, e, f, g, j, k, l, and m represent the multiple regression coefficients in the second model to be trained, and h and n represent the intercepts in the second model to be trained. Then proceed to step D2-2.

[0038] Step D2-2. Based on the spatial distribution of soil apparent electrical conductivity in the horizontal and vertical modes in the target study area, and the corresponding salinization characteristic spectral indices in the target study area, train the second training model to obtain the quantitative interpretation model M2.

[0039] As a preferred technical solution of the present invention: In step C, for the Landsat image dataset of the target study area, geometric correction, strip restoration, and atmospheric correction are performed sequentially, and then the result is obtained according to the following formula:

[0040]

[0041]

[0042]

[0043]

[0044] The enhanced vegetation index (EVI), greenness ratio vegetation index (GRVI), normalized difference salinity index (NDSI), and salinity vegetation index (SRSI) corresponding to the target study area were calculated. NIR represents the near-infrared band value of the target study area, and R, G, and B are the red band value, green band value, and blue band value of the target study area, respectively.

[0045] As a preferred technical solution of the present invention: using an EM38 or EM38-MK2 magnetic induction earth conductivity meter as a near-ground sensor to measure the apparent conductivity of soil in horizontal mode and the apparent conductivity of soil in vertical mode.

[0046] The multi-source, multi-scale data fusion method for inverting the three-dimensional distribution of regional soil salinization described in this invention has the following technical advantages compared with existing technologies:

[0047] (1) The multi-source, multi-scale data fusion method for inverting the three-dimensional distribution of regional soil salinization designed in this invention is based on the gridding of the target study area. The apparent electrical conductivity of the soil in each grid is combined with spatial interpolation to perform soil salinity measurements at each depth location corresponding to each soil sampling point. Then, the Landsat image dataset is combined with the salinization characteristic spectral index to construct a quantitative interpretation model M1 between the measured soil salinity and the apparent electrical conductivity of the soil, and a quantitative interpretation model M2 between the spatial distribution of the apparent electrical conductivity of the soil and each salinization characteristic spectral index. The two quantitative interpretation models are then applied sequentially to obtain the three-dimensional spatial distribution of soil salinity in the target study area. The design and application are centered on near-ground sensing data, with the advantages of near-ground sensing and remote sensing complementing each other. The multi-scale connection and nesting mechanism of "point-block-area" is clear. While obtaining a high-precision three-dimensional spatial distribution of regional soil salinity, it has the advantages of requiring fewer measured sample points, low cost, high accuracy, and simple operation.

[0048] (2) The multi-source, multi-scale data fusion method for three-dimensional distribution inversion of regional soil salinization designed in this invention has four advantages in practical application: 1) High sampling efficiency: Utilizing the high correlation between near-ground sensor data and soil salinity, and based on the spatial distribution of apparent conductivity and the VQT method, the representativeness of sampling points is enhanced, effectively reducing the number of soil samples required and improving sampling efficiency by more than 30%; 2) High inversion accuracy: Four improved salinization characteristic spectral indices—Enhanced Vegetation Index (EVI), Greenness Ratio Vegetation Index (GRVI), Normalized Difference Salinity Index (NDSI), and Salinity Vegetation Index (SRSI)—are selected, and soil salinity is inverted based on the comprehensive information of salinity index and vegetation index. More efficient and accurate; 3) High computational efficiency: A multivariate regression method based on partial least squares regression (PLSR) is adopted to construct an inversion model centered on near-ground sensing data and connecting soil sampling "point" data with Landsat improved salinization characteristic spectral index "regional" data, reducing the computational load of nonlinear methods and improving computational efficiency; 4) Excellent synergistic effect: Through the improved salinization characteristic spectral index and multivariate regression model, the soil profile information obtained by near-ground sensing is extended to the regional scale. At the same time, combined with the regression kriging method, the measured sampling information of field soil salinity is integrated into the inversion of regional soil salinization to the greatest extent, and the accuracy is improved by more than 15% compared with existing methods. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating the multi-source, multi-scale data fusion method for inverting the three-dimensional distribution of soil salinization in the design area according to the present invention.

[0050] Figure 2 This is a schematic diagram of the grid division of the study area, near-ground sensor measurements, and soil sampling points in the design of this invention.

[0051] Figure 3 Spatial distribution of apparent electrical conductivity ECah in soil of the study area (a) and its regional divisions (b)

[0052] Figure 4 Soil sampling patterns based on different variance thresholds using the VQT algorithm. (a) Variance threshold of 1% of maximum variance; (b) Variance threshold of 2.5% of maximum variance; (c) Variance threshold of 5% of maximum variance; (d) Variance threshold of 10% of maximum variance;

[0053] Figure 5 Spatial distribution of soil salinity at all sampling points under the VQT sampling strategy. (a) Number of sampling points: 295; (b) Number of sampling points: 141

[0054] Figure 6 Relationship between maximum intra-layer variance and number of splits (a) and maximum variance threshold and number of layers (b)

[0055] Figure 7 Inverted spatial distribution map of soil salinity in the regional profile. (a) Based on Landsat 8OLI imagery; (b) Based on multivariate data fusion of Landsat 8OLI, near-ground sensor data, and measured soil salinity residuals. Detailed Implementation

[0056] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0057] The multi-source, multi-scale data fusion method for inverting the three-dimensional distribution of regional soil salinization designed in this invention can be applied in practical situations, such as... Figure 1 As shown, the following steps A to F are performed to obtain the three-dimensional spatial distribution of soil salinity in the target study area.

[0058] Step A. Figure 2 As shown, based on the preset gridding of the target study area and the preset measurement point position in each grid, the apparent electrical conductivity of the soil in the horizontal mode and the apparent electrical conductivity of the soil in the vertical mode corresponding to each measurement point position are obtained, and then the process proceeds to step B.

[0059] Regarding the pre-defined gridding of the target research area, in practical applications, for research areas at the provincial scale, the grid center point spacing should be 4000-20000m; for research areas at the county scale, the grid size should be 2000-4000m; for research areas at the landscape scale, the grid size should be 200-2000m; and for research areas at the plot scale, the grid size should be 10-200m.

[0060] In practical applications, in step A above, based on the preset gridding of the target study area, steps A1 to A3 are performed for each grid of the target study area.

[0061] Step A1. As Figure 2 As shown, based on a preset measurement point location in the grid, the area in the grid with the measurement point location as the center and a preset radius of 15m is divided into four quadrants by two mutually orthogonal straight lines whose intersection point is located at the center of the circle, and then proceeds to step A2.

[0062] Step A2. Randomly place a sample point in each of the four quadrants. Combine the location of the measurement point to form five sample points corresponding to the measurement point location in the grid. Use an EM38 or EM38-MK2 magnetic induction earth conductivity meter as a near-ground sensor to measure the apparent soil conductivity in the horizontal mode and the apparent soil conductivity in the vertical mode at the five sample point locations. Then proceed to step A2A3-1.

[0063] Step A2A3-1. For each of the five sample points, use an insertion thermometer to measure the soil temperature at preset depths such as 5cm, 10cm, 20cm, 40cm, and 60cm vertically downwards from the sample point, and obtain the corresponding average soil temperature to form the soil temperature corresponding to the sample point. That is, obtain the soil temperature corresponding to each sample point, and then proceed to step A2A3-2.

[0064] Step A2A3-2. For each of the five sample points, apply the following formula:

[0065]

[0066] The apparent electrical conductivity EC of the soil under the horizontal mode corresponding to the sampling points ah_r Soil apparent electrical conductivity EC under vertical mode av_r Corrected to the target temperature t respectively 目 For example, the apparent corrected electrical conductivity EC of soil under horizontal mode at 25°C ah Soil apparent corrected electrical conductivity EC in vertical mode av The apparent electrical conductivity of the soil in the horizontal mode and the apparent electrical conductivity of the soil in the vertical mode corresponding to the sample point are updated. That is, the apparent electrical conductivity of the soil in the horizontal mode and the apparent electrical conductivity of the soil in the vertical mode corresponding to each sample point are updated respectively, where t represents the soil temperature corresponding to the sample point, and then proceed to step A3.

[0067] Step A3. Obtain the apparent average electrical conductivity of the soil in the horizontal mode and the apparent average electrical conductivity of the soil in the vertical mode among the five sample points, and construct the apparent electrical conductivity of the soil in the horizontal mode and the apparent electrical conductivity of the soil in the vertical mode corresponding to the measurement point position in the grid.

[0068] Step B. Obtain the spatial distribution of soil apparent electrical conductivity in the target study area under horizontal and vertical modes through spatial interpolation, determine each soil sampling point in the target study area, and then obtain the measured soil salinity values ​​at each preset depth location corresponding to each soil sampling point, and then proceed to step C.

[0069] In practical applications, step B above includes steps B1 to B3.

[0070] Step B1. Based on the apparent electrical conductivity of soil in the horizontal mode and the apparent electrical conductivity of soil in the vertical mode corresponding to the location of each measurement point in the target study area, the apparent electrical conductivity of soil in the horizontal mode and the apparent electrical conductivity of soil in the vertical mode in the target study area are spatially interpolated using the ordinary Kriging method to form the spatial distribution of the apparent electrical conductivity of soil in the horizontal mode and the apparent electrical conductivity of soil in the vertical mode in the target study area, and then proceed to step B2.

[0071] Step B2. Based on the spatial distribution of soil apparent electrical conductivity in the target study area under the horizontal pattern, the variance quad-tree (VQT) method is used to determine the locations of each measurement point used for soil sampling in the target study area with a preset maximum variance threshold, thus forming the soil sampling points in the target study area. Then proceed to step B3.

[0072] Specifically based on EC ah The spatial distribution map of EC in the target study area was analyzed using the variance quad-tree (VQT) method. ah Areas with high variance are repeatedly divided into four equal parts until the maximum variance is less than a specified threshold, i.e., 10% of the maximum variance in the target study area. Generally speaking, the greater the variability of spatial distribution, the more segmentation iterations are required, and the more layers (i.e., the number of soil sampling points) are generated. Finally, the optimal number and location of soil sampling points required for the target study area are determined.

[0073] Step B3. Based on the five sample points corresponding to the measurement points in each grid of the target study area, for each soil sampling point in the target study area, obtain equal mass soil samples at preset depths such as 0-10cm, 10-20cm, 20-40cm, 40-60cm, 60-80cm, and 80-100cm vertically downwards from the five sample points. Then, obtain mixed soil samples from the five sample points at each depth. After air-drying, grinding, and passing through a 20-mesh sieve, determine the soil salinity of each mixed soil sample using the residue drying method. This constitutes the measured soil salinity values ​​for each depth corresponding to the soil sampling point, i.e., obtaining the measured soil salinity values ​​for each depth corresponding to each soil sampling point.

[0074] Step C. Based on the execution time points of Steps A to B, obtain the Landsat image dataset for the target study area corresponding to the execution time points, provided there was no significant rainfall during that period. Then, perform geometric correction, strip restoration, and atmospheric correction on the Landsat image dataset sequentially, and apply the following formula:

[0075]

[0076]

[0077]

[0078]

[0079] The enhanced vegetation index (WVI), greenness ratio vegetation index (GRVI), normalized difference salinity index (NDSI), and salinity vegetation index (SRSI) corresponding to the target study area are calculated. Here, NIR represents the near-infrared band value of the target study area, and R, G, and B are the red band value, green band value, and blue band value of the target study area, respectively. Then, proceed to step D.

[0080] Step D. Based on each soil sampling point, construct a quantitative interpretation model M1 between the measured soil salinity at each soil depth and the corresponding apparent electrical conductivity under horizontal and vertical modes, and construct a quantitative interpretation model M2 between the spatial distribution of apparent electrical conductivity under horizontal and vertical modes in the target study area and the spectral indices of various salinization characteristics, and then proceed to step E.

[0081] In practical applications, in step D above, the quantitative interpretation model M1 is obtained by following steps D1-1 to D1-2.

[0082] Step D1-1. Construct the measured values ​​of soil salinity at various depths in the target study area and the corresponding apparent electrical conductivity EC under the soil horizontal model. ah apparent electrical conductivity EC of soil under vertical mode av The first model to be trained is as follows:

[0083] S t =a·EC ah +b·EC av +c

[0084] Among them, S t The values ​​represent the measured soil salinity at various soil depths, a and b represent the multiple regression coefficients in the first model to be trained, and c represents the intercept in the first model to be trained. Then proceed to step D1-2.

[0085] Step D1-2. Based on the measured values ​​of soil salinity at each depth corresponding to each soil sampling point and the apparent electrical conductivity of soil in the horizontal mode and the apparent electrical conductivity of soil in the vertical mode, train the first model to be trained to obtain the quantitative interpretation model M1.

[0086] Furthermore, based on various salinization characteristic spectral indices including Enhanced Vegetation Index (EVI), Greenness Ratio Vegetation Index (GRVI), Normalized Difference Salinity Index (NDSI), and Salinity Vegetation Index (SRSI), in practical applications, in step D above, the quantitative interpretation model M2 is obtained by following steps D2-1 to D2-2.

[0087] Step D2-1. Construct the apparent electrical conductivity EC of soil under a horizontal model for the target study area. ah apparent electrical conductivity EC of soil under vertical modeav The second training model relating the spatial distribution of the spectral indices of each salinization characteristic to the spatial distribution of the salinization feature is as follows:

[0088] EC ah =d·EVI+e·GRVI+f·NDSI+g·SRSI+h

[0089] EC av =j·EVI+k·GRVI+l·NDSI+m·SRSI+n

[0090] Where d, e, f, g, j, k, l, and m represent the multiple regression coefficients in the second model to be trained, and h and n represent the intercepts in the second model to be trained. Then proceed to step D2-2.

[0091] Step D2-2. Based on the spatial distribution of soil apparent electrical conductivity in the horizontal and vertical modes in the target study area, and the corresponding salinization characteristic spectral indices in the target study area, train the second training model to obtain the quantitative interpretation model M2.

[0092] Step E. Based on the quantitative interpretation model M2 and the spectral indices of salinization characteristics corresponding to the target study area, calculate and update the spatial distribution of soil apparent electrical conductivity in the target study area under the horizontal mode and the vertical mode. Then, combine the quantitative interpretation model M1 to calculate the predicted value of soil salinity spatial distribution in the target study area, thus forming the three-dimensional spatial distribution of soil salinity in the target study area. Then proceed to step F.

[0093] Step F. Add the Kriging interpolation values ​​of the three-dimensional spatial distribution of soil salinity in the target study area to the residuals of the measured soil salinity values ​​at each depth of each soil sampling point to obtain the inverted value of the three-dimensional spatial distribution of soil salinity in the target study area.

[0094] The above technical solution designs a multi-source, multi-scale data fusion method for inverting the three-dimensional distribution of regional soil salinization. Based on the gridding of the target study area, it uses the apparent electrical conductivity of the soil in each grid and combines spatial interpolation to perform on-site measurements of soil salinity at each depth location corresponding to each soil sampling point. Then, it combines Landsat image datasets with salinization characteristic spectral indices to construct a quantitative interpretation model M1 between the measured soil salinity and the apparent electrical conductivity of the soil, and a quantitative interpretation model M2 between the spatial distribution of the apparent electrical conductivity of the soil and each salinization characteristic spectral indice. The two quantitative interpretation models are then applied sequentially to obtain the three-dimensional spatial distribution of soil salinity in the target study area. The design and application are centered on near-ground sensing data, with the advantages of near-ground sensing and remote sensing complementing each other. The "point-block-region" multi-scale connection and nesting mechanism is clear. While obtaining a high-precision three-dimensional spatial distribution of regional soil salinity, it has the advantages of requiring fewer on-site sampling points, low cost, high accuracy, and simple operation.

[0095] Applying this invention to practical applications, Example 1 examines a study area located in Jinhai Farm, Dafeng District, Yancheng City, Jiangsu Province, approximately 4 km east of the Yellow Sea coastline. This area is situated in the northern subtropical monsoon climate zone, exhibiting distinct transitional, maritime, and monsoon characteristics. The average annual precipitation is 1058.4 mm, primarily concentrated in the rainy season from June to August. The soils in this area are mainly composed of marine sediments, with sandy loam as the dominant texture. Salinization is a major obstacle to agricultural development in this region. Despite continuous cultivation in recent years, the degree of variation in soil salinity remains significant, leading to considerable differences in crop growth in different areas.

[0096] Applying the design method of this invention to this embodiment 1, in the actual application process in this region, specifically includes the following steps:

[0097] Step A: Based on the 45m×45m grid of the target study area, the EM38 near-ground sensor—electromagnetic induction instrument—was used to measure the apparent electrical conductivity of the soil in the horizontal and vertical modes corresponding to the measurement points in each grid. Specifically, for each measurement point, based on the five sample points corresponding to the measurement point, the apparent electrical conductivity of the soil at each sample point was corrected by referring to the soil temperature at depths of 5cm, 10cm, 20cm, 40cm, and 60cm vertically downward from each sample point measured by an insertion digital thermometer (Beken ST digital thermometer). Then, the average value method was used to obtain the apparent electrical conductivity of the soil in the horizontal and vertical modes corresponding to the measurement points in each grid. In Example 1, a total of 285 measurement points were measured for the apparent electrical conductivity of the soil.

[0098] Step B involves using the Kriging method to spatially interpolate and map the soil apparent electrical conductivity data at 285 measurement points, obtaining their spatial distribution as shown below. Figure 3 As shown in (a), the apparent electrical conductivity of this plot shows a trend of being higher in the southeast and lower in the northwest. Overall, the eastern part is higher than the western part, and the southern part is higher than the northern part. Moreover, its spatial variability is extremely significant, and the apparent electrical conductivity of the soil varies greatly in some local areas.

[0099] Given the significant spatial variability in the apparent electrical conductivity distribution of the soil in the study area, and based on the requirements of the variance quadtree method (VQT), the study area was divided into three equal-area regions, A, B, and C, from north to south. Figure 3 As shown in (b), region C exhibits the greatest variation intensity, while region A shows the least. This facilitates a quantitative comparison of the differences between the VQT algorithm's calculation results for different spatial distribution patterns. Overall, the maximum variance of each region decreases rapidly with the increase of the number of segmentation iterations (i.e., the increase of the number of strata), then the rate of decrease begins to slow down and eventually stabilizes. Under the same variance threshold, the stronger the variability of the spatial distribution, the more segmentation iterations are required, and the more strata are generated. Furthermore, within the same region, the smaller the variance threshold, the more segmentation iterations are required and the more strata are generated.

[0100] like Figure 4 The sampling design diagrams shown in (a), (b), (c), and (d) divide the target study area of ​​Example 1 into 420, 276, 195, and 141 layers, respectively. It can be seen that the layer density is higher in areas with strong variability in soil apparent conductivity, while the layer density is sparse in areas with strong homogeneity. That is, the variance quadtree method (VQT) can identify the areas with the greatest variation in soil apparent conductivity over a small range. During field sampling, the sampling density can be increased in these areas and sparsely sampled in areas with good continuity. This can save sampling and analysis costs and improve sampling efficiency.

[0101] Therefore, the variance quadtree method (VQT) is used to determine the locations of various measurement points for soil sampling in the target study area by setting a maximum variance threshold. These points constitute the soil sampling points in the target study area. Furthermore, for each soil sampling point, five sample points are selected, and equal mass soil samples are taken at each depth vertically downward from each sample point. Mixed soil samples of equal mass from the five sample points are obtained at each depth. After air drying, grinding, and passing through a 20-mesh sieve, the soil salinity of the mixed soil is determined by the residue drying method. Thus, the measured soil salinity values ​​at each depth corresponding to each soil sampling point are obtained.

[0102] Further execute steps C to F to obtain the inversion values ​​of the three-dimensional spatial distribution of soil salinity in the target study area.

[0103] To verify the sampling layout obtained by the Variance Quadtree Method (VQT), ordinary Kriging interpolation was performed on the measured soil salinity values ​​of all 285 sampling points in the target study area of ​​Example 1 and 141 soil sampling points obtained using the VQT algorithm (with a variance threshold of 10% of the maximum variance). The spatial distribution of soil salinity was obtained as follows: Figure 5 As shown in the figure, the spatial distribution map of soil salinity formed by the 141 sampling points is as follows: Figure 5 The distribution map obtained by comparing (b) with all sampling points is shown below. Figure 5 (a) has an extremely similar spatial pattern, although some details of local variations are smoothed out, Figure 5 b still retains the regions with greater variation intensity very well. Although the number of samples differs by nearly half, the analysis results show that their spatial distribution similarity is 90%.

[0104] The analysis results of this embodiment show that, based on near-ground sensing data and the variance quadtree method, the field sample requirement for soil salinization can be significantly reduced, with a maximum reduction of 52% in sampling volume, thus significantly improving sample representativeness and sampling efficiency.

[0105] Example 2: The Qingtongxia Irrigation District is located in the northern part of Ningxia Hui Autonomous Region, on the alluvial plain at the eastern foot of the Helan Mountains, in the lower reaches of the Yellow River (within Ningxia). Situated in the arid and semi-arid region of Northwest my country, it has a temperate continental climate. The 200mm isohyet passes through the irrigation district, which is divided into the East and West Irrigation Districts by the Yellow River flowing from northwest to southeast. This includes eight cities and counties: Qingtongxia, Wuzhong, Lingwu, Yongning, Yinchuan, Helan, Pingluo, and Shizuishan. The Qingtongxia Irrigation District has fertile soil due to the Yellow River's alluvial deposits. The soil on the left side of the Yellow River is heavy, meadow soda saline soil or marsh soda saline-alkali soil. The soil on the right side of the Yellow River is mainly chloride-sulfate residual saline soil and subsurface saline light gray-calcic soil, forming soils in the irrigated area between the old terraces on the left and right sides of the Yellow River. This often results in small patches of alluvial saline soil or saline-irrigated alluvial soil.

[0106] The design method of the present invention is applied to the target research area of ​​the above embodiment 2, and the following steps are specifically performed:

[0107] Step A: Divide the entire Qingtongxia Irrigation District into regular grid units, considering that the irrigated area of ​​the Qingtongxia Irrigation District is 366,700 hectares. 2 The grid size was set to 4000m × 4000m, dividing the study area into 210 grids. Soil apparent conductivity was corrected for by considering the temperature at each measurement point within the grid, resulting in the soil apparent conductivity in both horizontal and vertical modes for each measurement point.

[0108] Step B: The Kriging method was used to analyze the soil apparent electrical conductivity data (EC) from 210 sampling points. ah and EC av Spatial interpolation and mapping are performed to obtain its spatial distribution.

[0109] Given the apparent electrical conductivity EC of the soil in some plots of the irrigation area ah The spatial distribution exhibits strong variability. Based on the requirements of the Variance Quadritree Method (VQT), the study area is divided into two regions, A and B, from north to south; for example... Figure 6 Figure 'a' shows the variance values ​​of the layers with the maximum prediction variance found in regions A and B as the number of iterations of the VQT (Variance Quadtree Method) increases. It can be seen that the maximum variance of each region decreases rapidly with the increase of the number of segmentation iterations (i.e., the number of layers), then the rate of decrease begins to slow down, and eventually stabilizes. The maximum intra-layer variance in region A begins to stabilize after 21 segmentations, while in region B it is after 16. For the same variance threshold (based on the 5% maximum variance in region A), region A begins to stabilize after 21 segmentations, while region B requires 25. According to... Figure 6 A can be used to calculate a reasonable number of strata to ensure that the maximum variance within each stratum is within the ideal range. When the variance threshold is 10% of the maximum variance, 100 soil sampling points in the Qingtongxia irrigation area are sufficient to meet the requirements.

[0110] Soil samples were then collected from the 100 soil sampling points identified above, with sampling depths set at 0–10, 10–20, 20–40, 40–60, and 60–100 cm. Five soil samples were collected from each of the five corresponding sampling points and mixed to form the test sample. The soil sampling points coincided with the measurement points of the EM38-MK2. The collected mixed soil samples were brought back to the laboratory for natural air drying, ground, and sieved through a 2 mm sieve for later use. A 1:5 soil-to-water ratio extract was prepared from all soil samples, shaken, allowed to stand, and impurities were filtered out. A certain amount of the soil extract was placed in a porcelain evaporating dish, evaporated to dryness in a water bath, oxidized with hydrogen peroxide, and then dried to constant weight in an oven at 105–110 °C. The measured soil salinity was then calculated.

[0111] Step C: Considering the low rainfall in the Qingtongxia irrigation area during spring, the Landsat imagery data from the period closest to the near-ground sensing measurements and soil sample collection time was selected. Landsat 8OLI has a spectral resolution of 0.43–2.3 μm and 11 bands; bands 1 to 7 have a spatial resolution of 30 m, and bands 8 to 11 have a spatial resolution of 15 m. In addition, Landsat 8OLI has a panchromatic band with a spatial resolution of 15 m. Landsat 8OLI satellite payload parameters are shown in Table 1.

[0112] Table 1

[0113]

[0114] After geometric correction, strip restoration, and atmospheric correction of the Landsat 8OLI imagery, four improved salinization characteristic spectral indices—Enhanced Vegetation Index (EVI), Greenness Ratio Vegetation Index (GRVI), Normalized Difference Salinity Index (NDSI), and Salinity Vegetation Index (SRSI)—were calculated using the following formulas:

[0115]

[0116]

[0117]

[0118]

[0119] The enhanced vegetation index (EVI), greenness ratio vegetation index (GRVI), normalized difference salinity index (NDSI), and salinity vegetation index (SRSI) corresponding to the target study area were calculated. NIR represents the near-infrared band value of the target study area, and R, G, and B are the red band value, green band value, and blue band value of the target study area, respectively.

[0120] Step D: Based on each soil sampling point, a quantitative interpretation model M1 is constructed to obtain the measured soil salinity values ​​at each soil depth and the corresponding apparent electrical conductivity under horizontal and vertical modes. Specifically, based on the partial least squares regression (PLSR) multiple regression method, the measured soil salinity values ​​S at different depths of the profile are constructed. t EC with near-ground sensor measurement data ah EC av The multiple linear regression model between them is shown in Table 2.

[0121] Table 2

[0122]

[0123] In the table, a and b are multiple regression coefficients, c is the intercept, and r² is the coefficient of determination.

[0124] Furthermore, a quantitative interpretation model M2 was constructed to obtain the spatial distribution of soil apparent electrical conductivity in the target study area under horizontal and vertical modes, and the relationship between these distributions and various salinization characteristic spectral indices. Specifically, this model was based on a partial least squares regression (PLSR) multiple regression method to establish near-ground sensor measurement data EC. ah EC av The multiple linear regression models between the four improved salinization characteristic spectral indices, EVI, GRVI, NDSI and SRSI, are shown in Table 3.

[0125] Table 3

[0126]

[0127] Where d, e, f, g, j, k, l, and m represent the multiple regression coefficients in the second model to be trained, h and n represent the intercepts in the second model to be trained, and r represents the multivariate regression coefficients in the second model to be trained. 2 The coefficient of determination.

[0128] Finally, steps E and F are performed. Based on the Landsat 8OLI imagery of the study area, four improved salinization characteristic spectral indices—Enhanced Vegetation Index (EVI), Greenness Ratio Vegetation Index (GRVI), Normalized Differential Salinity Index (NDSI), and Salinity Vegetation Index (SRSI)—are calculated. The EC is then calculated using the model in Table 4. ah and EC av Spatial distribution; then, using the model in Table 1, the predicted spatial distribution values ​​of soil salinity at different depths are calculated; next, the measured soil salinity values ​​at different depths at the sampling points are subtracted from the predicted salinity values ​​calculated based on the models in Tables 3 and 1, i.e., the residuals, and Kriging interpolation is performed to form a residual spatial distribution map; finally, the predicted spatial distribution values ​​of soil salinity calculated based on the models in Tables 3 and 1 are added to the residual spatial distribution map to obtain the inverted three-dimensional spatial distribution values ​​of soil salinity in the profile, as shown below. Figure 7 As shown. Among them. Figure 7 (a) shows the soil salinity distribution profile obtained directly from Landsat 8OLI imagery. Figure 7 (b) shows the integration of Landsat 8OLI imagery and near-ground sensor data EC. a A spatial distribution map of soil salinity profile obtained by fusing multi-source information from data, measured soil salinity values, and residuals. It can be seen that... Figure 7 (b) Figure 7Image (a) provides a more intuitive and detailed reflection of soil salinity variability in local areas, especially highlighting the continuous patchy high-salinity distribution areas (northwest region) on the profile. Table 4 shows the cross-validation results of soil salinity profiles obtained using different methods. The method based on the fusion of Landsat 8OLI, near-ground sensor data, and residual multivariate data of measured soil salinity values ​​yields a significantly lower root mean square error (RMSE) for soil salinity profiles compared to the method based solely on Landsat 8OLI imagery, indicating higher precision and accuracy.

[0129] Table 4

[0130]

[0131] In practical applications, this invention has four advantages: 1) High sampling efficiency: Utilizing the high correlation between near-ground sensing data and soil salinity, and based on the spatial distribution of apparent conductivity and the VQT method, it enhances the representativeness of sampling points, effectively reducing the number of soil samples required and improving sampling efficiency by more than 30%; 2) High inversion accuracy: By selecting four improved salinization characteristic spectral indices—Enhanced Vegetation Index (EVI), Greenness Ratio Vegetation Index (GRVI), Normalized Difference Salinity Index (NDSI), and Salinity Vegetation Index (SRSI)—the inversion of soil salinity based on the comprehensive information of salinity and vegetation indices is more efficient and accurate; 3) High computational efficiency: Using a multivariate regression method based on partial least squares regression (PLSR), an inversion model is constructed with near-ground sensing data as the center, connecting soil sampling "point" data and Landsat improved salinity characteristic spectral index "regional" data, reducing the computational load of nonlinear methods and improving computational efficiency; 4) Excellent synergistic effect: Through the improved salinity characteristic spectral index and multivariate regression model, the soil profile information obtained by near-ground sensing is extended to the regional scale. At the same time, combined with the regression kriging method, the measured sampling information of field soil salinity is integrated into the inversion of regional soil salinity to the greatest extent, and the accuracy is improved by more than 15% compared with existing methods.

[0132] 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 multi-source, multi-scale data fusion method for inverting the three-dimensional distribution of regional soil salinization, characterized in that: Perform the following steps to obtain the three-dimensional spatial distribution of soil salinity in the target study area; Step A. Based on the preset gridding of the target study area and the preset measurement point location in each grid, obtain the soil apparent electrical conductivity in the horizontal mode and the soil apparent electrical conductivity in the vertical mode corresponding to each measurement point location, and then proceed to Step B; In step A above, based on the preset gridding of the target study area, steps A1 to A3 are performed for each grid of the target study area respectively; Step A1. Based on a preset measurement point location in the grid, divide the area of ​​the grid with the measurement point location as the center and a preset radius by applying two mutually orthogonal straight lines whose intersection point is located at the center of the circle to form four quadrants, and then proceed to step A2; Step A2. Randomly place a sample point in each of the four quadrants. Combine the position of the measurement point to form five sample points corresponding to the position of the measurement point in the grid. Obtain the apparent electrical conductivity of the soil in the horizontal mode and the apparent electrical conductivity of the soil in the vertical mode corresponding to the five sample points respectively. Then proceed to step A3. Step A3. Obtain the apparent average electrical conductivity of soil in the horizontal mode and the apparent average electrical conductivity of soil in the vertical mode among the five sample points, and construct the apparent electrical conductivity of soil in the horizontal mode and the apparent electrical conductivity of soil in the vertical mode corresponding to the measurement point position in the grid. Step B. Obtain the spatial distribution of soil apparent electrical conductivity in the target study area under horizontal and vertical modes through spatial interpolation, determine each soil sampling point in the target study area, and then obtain the measured values ​​of soil salinity at each preset depth location corresponding to each soil sampling point, and then proceed to step C. Step C. Based on the execution time points of Steps A to B, obtain the Landsat image dataset corresponding to the execution time points for the target study area, and calculate the spectral indices of salinization characteristics for the target study area. Then proceed to Step D; the spectral indices of salinization characteristics include the vegetation enhancement index. Greenness ratio and vegetation index Normalized Differential Salinity Index Salt vegetation index ; Step D. Based on each soil sampling point, construct a quantitative interpretation model M1 between the measured soil salinity at each soil depth and the corresponding apparent electrical conductivity of the soil in the horizontal mode and the apparent electrical conductivity of the soil in the vertical mode, and construct a quantitative interpretation model M2 between the spatial distribution of the apparent electrical conductivity of the soil in the horizontal mode and the apparent electrical conductivity of the soil in the vertical mode in the target study area and the spectral indices of each salinization characteristic, and then proceed to step E; In step D above, the quantitative interpretation model M1 is obtained by following steps D1-1 to D1-2; Step D1-1. Construct the measured values ​​of soil salinity at various depths in the target study area and the corresponding apparent electrical conductivity of soil under horizontal models. apparent electrical conductivity of soil under vertical mode The first model to be trained is as follows: ； in, This represents the measured values ​​of soil salinity at various depths. , represent the multiple regression coefficients in the first model to be trained. This represents the intercept in the first model to be trained, and then proceeds to step D1-2; Step D1-2. Based on the measured values ​​of soil salinity at each depth corresponding to each soil sampling point and the apparent electrical conductivity of soil in the horizontal mode and the apparent electrical conductivity of soil in the vertical mode, train the first model to be trained to obtain the quantitative interpretation model M1. In step D, follow steps D2-1 to D2-2 as follows to obtain the quantitative interpretation model M2; Step D2-1. Constructing the apparent electrical conductivity of soil under a horizontal model for the target study area. apparent electrical conductivity of soil under vertical mode The second training model relating the spatial distribution of the spectral indices of each salinization characteristic to the spatial distribution of the salinization feature is as follows: ； ； in, , , , , , , , These represent the multiple regression coefficients in the second model to be trained. , These represent the intercepts in the second model to be trained, and then proceed to step D2-2; Step D2-2. Based on the spatial distribution of soil apparent electrical conductivity in the horizontal and vertical modes in the target study area, and the corresponding salinization characteristic spectral indices in the target study area, train the second training model to obtain the quantitative interpretation model M2. Step E. Based on the quantitative interpretation model M2 and the spectral indices of salinization characteristics corresponding to the target study area, calculate and update the spatial distribution of soil apparent electrical conductivity in the target study area under the horizontal mode and the vertical mode. Then, combine it with the quantitative interpretation model M1 to calculate the predicted value of soil salinity spatial distribution in the target study area, thus forming the three-dimensional spatial distribution of soil salinity in the target study area.

2. The multi-source, multi-scale data fusion method for inverting the three-dimensional distribution of regional soil salinization according to claim 1, characterized in that: It also includes step F as follows: after step E is completed, step F is performed. Step F. Add the Kriging interpolation values ​​of the three-dimensional spatial distribution of soil salinity in the target study area to the residuals of the measured soil salinity values ​​at each depth of each soil sampling point to obtain the inverted value of the three-dimensional spatial distribution of soil salinity in the target study area.

3. The multi-source, multi-scale data fusion method for inverting the three-dimensional distribution of regional soil salinization according to claim 1, characterized in that: Step A further includes steps A2A3-1 to A2A3-2. After step A2 is completed, step A2A3-1 is entered. Step A2A3-1. For each of the five sample points, obtain the soil temperature at each preset depth vertically downward from the sample point, and obtain the corresponding average soil temperature to form the soil temperature corresponding to that sample point. That is, obtain the soil temperature corresponding to each sample point, and then proceed to step A2A3-2. Step A2A3-2. For each of the five sample points, apply the following formula: ； ； The apparent electrical conductivity of the soil under the horizontal mode corresponding to the sampling points Soil apparent electrical conductivity under vertical mode Corrected to the target temperature respectively Soil apparent corrected electrical conductivity under horizontal mode Soil apparent corrected electrical conductivity in vertical mode The soil apparent electrical conductivity is updated to the corresponding horizontal and vertical modes for each sampling point. Specifically, the soil apparent electrical conductivity for each sampling point is updated to the corresponding horizontal and vertical modes. This indicates the soil temperature corresponding to the sample point, and then proceeds to step A3.

4. The multi-source, multi-scale data fusion method for inverting the three-dimensional distribution of regional soil salinization according to claim 1, characterized in that: Step B includes the following steps B1 to B3; Step B1. Based on the apparent electrical conductivity of soil in the horizontal mode and the apparent electrical conductivity of soil in the vertical mode corresponding to the location of each measurement point in the target study area, the apparent electrical conductivity of soil in the horizontal mode and the apparent electrical conductivity of soil in the vertical mode in the target study area are spatially interpolated using the ordinary Kriging method to form the spatial distribution of the apparent electrical conductivity of soil in the horizontal mode and the apparent electrical conductivity of soil in the vertical mode in the target study area, and then proceed to step B2. Step B2. Based on the spatial distribution of soil apparent electrical conductivity in the target study area under the horizontal pattern, the variance quadtree method (VQT) is used to determine the locations of each measurement point used for soil sampling in the target study area with a preset maximum variance threshold, thus forming the soil sampling points in the target study area. Then proceed to step B3. Step B3. Based on the five sample points corresponding to the measurement points in each grid of the target study area, for each soil sampling point in the target study area, obtain equal mass soil samples at each preset depth vertically downward from the five sample points corresponding to the soil sampling points, and obtain mixed soil samples from the five sample points at each depth. Collect the measured soil salinity values ​​of each mixed soil to form the measured soil salinity values ​​at each depth corresponding to the soil sampling points, that is, obtain the measured soil salinity values ​​at each depth corresponding to each soil sampling point.

5. The multi-source, multi-scale data fusion method for inverting the three-dimensional distribution of regional soil salinization according to claim 1, characterized in that: In step C, for the Landsat image dataset of the target study area, geometric correction, strip restoration, and atmospheric correction are performed sequentially, followed by the following formula: ； ； ； ； Calculate the enhanced vegetation index corresponding to the target study area Greenness ratio and vegetation index Normalized Differential Salinity Index Salt vegetation index ,in, Indicates the near-infrared band value of the target study area. , , The values ​​are, in order, the red band value, the green band value, and the blue band value of the target study area.

6. The multi-source, multi-scale data fusion method for inverting the three-dimensional distribution of regional soil salinization according to claim 1, characterized in that: The EM38 or EM38-MK2 magnetic induction geoelectric conductivity meter was used as a near-ground sensor to measure the apparent electrical conductivity of soil in both horizontal and vertical modes.

Images