Degenerated cultivated land soil organic matter three-dimensional mapping method fusing remote sensing, near-earth sensing and historical data

Abstract

The invention discloses a degraded cultivated land soil organic matter three-dimensional mapping method fusing remote sensing, near-earth sensing and historical data, and relates to the technical field of digital soil mapping, the method is based on an INLA-SPDE three-dimensional model framework, remote sensing, near-earth sensing and historical data are fused, and high-precision three-dimensional mapping is achieved. The method comprises the following steps: collecting organic matter data and near-sensing visible near-infrared reflection spectrum data of different depths of a soil profile at each sampling point in a research area, then screening visible near-infrared reflection spectrum characteristic wavebands significantly related to organic matters, collecting and integrating historical data, and completing standardization processing on remote sensing and near-earth sensing data; dividing a modeling set and a verification set, calculating a range value based on a semi-variable function, determining triangular mesh parameters, and constructing an SPDE model; the near sensing data and the remote sensing data serve as covariables and are fused into the INLA-SPDE three-dimensional model, and a soil organic matter three-dimensional space distribution diagram is generated after model fitting and uncertainty is quantified. According to the method, the problem that the vertical change of the soil attribute cannot be accurately simulated by a traditional method is solved, and high-precision and high-efficiency three-dimensional mapping of the regional scale soil organic matter in the horizontal and vertical dimensions is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of digital soil mapping, in particular to a three-dimensional mapping method for degraded farmland soil organic matter by fusing remote sensing, near-ground sensing and historical data. BACKGROUND

[0002] Black soil resources are globally scarce, accounting for only 7% of the global arable land area, but carrying nearly 25% of the food production capacity. The thickness and organic matter content of the black soil layer directly determine the level of soil health. However, in the past 50 years, continuous high-intensity cultivation has led to an average annual loss of 0.3-1 cm of the black soil layer, and in some areas, the thickness is already less than 20 cm. At the same time, the surface SOM content has decreased sharply from 8-10% at the beginning of reclamation to the current 2-3%, and the deep layer organic matter has also decreased synchronously, seriously threatening the strategic role of the "ballast" of national food security. Soil organic matter (SOM) as one of the key properties of soil plays a key role in global carbon cycle, and its content and spatial distribution deeply affect the cation exchange capacity, water holding capacity, formation and stability of soil aggregates, and the structure and activity of soil microbial community. Therefore, accurately quantifying the three-dimensional spatial distribution characteristics of SOM is the scientific basis for evaluating regional soil carbon storage and understanding the mechanism of carbon cycle.

[0003] With the development of technology, digital soil mapping (DSM) has become a high-precision and high-quality means of acquiring soil spatial information, and has been widely applied in practical research. However, existing DSM methods mostly focus on surface or shallow soil, and lack systematic characterization and high-precision modeling of continuous variation on the vertical profile scale of soil. On the other hand, traditional two-dimensional spatial models face challenges in extending to three dimensions, and it is difficult to effectively capture the complex spatial dependence structure of deep soil. The existing INLA-SPDE three-dimensional modeling method still has fundamental limitations, usually using a unified set of environmental covariates to model SOM at all depth levels, without fully considering the differences in environmental driving mechanisms corresponding to different depth layers of soil, limiting the prediction accuracy and adaptability of the model in the vertical dimension.

[0004] Therefore, how to break through the bottleneck of existing methods in simulating the vertical variation of soil organic matter and realize high-precision and high-efficiency three-dimensional mapping of regional-scale SOM has important theoretical significance and practical value for improving the scientificity of soil carbon storage evaluation and strengthening the cognition of carbon sink function in black soil areas. SUMMARY

[0005] The purpose of the present application is to provide a three-dimensional mapping method for degraded farmland soil organic matter by fusing remote sensing, near-ground sensing and historical data, which can realize high-precision and high-efficiency three-dimensional mapping of regional-scale soil organic matter.

[0006] To achieve the above object, the application provides the following scheme:

[0007] A three-dimensional mapping method for degraded farmland soil organic matter by fusing remote sensing, near-ground sensing and historical data, comprising the following steps:

[0008] Collect soil profile samples at a plurality of sampling points in the study area, layer according to the sampling depth of the soil profile sample and determine the soil organic matter content and near-infrared reflectance spectrum data of each layer of each soil profile sample, and obtain a newly collected soil sample data set.

[0009] Pretreat the soil organic matter content and near-infrared reflectance spectrum data in the newly collected soil sample data set and perform correlation analysis, and select the near-infrared reflectance spectrum characteristic band with the highest correlation with the soil organic matter content.

[0010] Obtain a historical soil sample data set, and combine the converted historical soil sample data set with the newly collected soil sample data set to obtain a total soil sample data set; the historical soil sample data set includes soil organic matter content collected at a plurality of historical sampling points in the study area, and the conversion of the historical soil sample data set includes using an equal-area spline function to integrate the historical soil organic matter content and the newly collected soil organic matter content into a unified depth layer.

[0011] Collect topographic data, soil type data, vegetation data and remote sensing meteorological data in the study area, and standardize and pretreat the topographic data, soil type data, vegetation data and remote sensing meteorological data.

[0012] According to the soil organic matter content of the surface layer, each soil profile sample of the newly collected soil sample data set is divided into a fixed profile according to a predetermined proportion, and a modeling set and a verification set are established; the data in the historical soil sample data set are divided into the modeling set; the modeling set includes soil organic matter content collected at a plurality of sampling points and historical sampling points, the selected near-infrared reflectance spectrum characteristic band, and multi-source covariate data; the multi-source covariate data includes pretreated topographic data, soil type data, vegetation data and remote sensing meteorological data; the near-infrared reflectance spectrum characteristic band of the historical sampling point is determined in the near-infrared reflectance spectrum characteristic band distribution map of the study area according to the coordinates of the historical sampling point.

[0013] For each sampling point, calculate the semi-variogram according to the latitude and longitude of the sampling point and the soil organic matter content of each layer to obtain the range value of the soil organic matter content of each layer, and set the MESH grid parameters according to the maximum range.

[0014] The pretreated terrain data, soil type data, vegetation data, remote sensing meteorological data and near-sensing visible near-infrared reflectance characteristic wave band are introduced into the INLA model under the Bayesian inference framework as fixed effect variables, and are used to explain the systematic spatial variation of soil organic matter.

[0015] A SPDE Gaussian random field model is constructed, a covariance function is introduced to describe the three-dimensional spatial correlation, and the finite element method is used to discretize the random field process on the MESH grid. The INLA-SPDE three-dimensional model integrating remote sensing, near-surface sensing and historical data is constructed by integrating the fixed effect variable, the spatial random effect term based on the SPDE Gaussian random field model and the depth hierarchical model. The SPDE Gaussian random field model is used to simulate the spatial random effect term, the depth variation of soil is analogized as a time dynamic sequence, and the depth hierarchical model is used to capture the depth variation law of soil organic matter.

[0016] The spatial coordinates of the sample points of the overall soil sample data set and the pretreated terrain data, soil type data, vegetation data, remote sensing meteorological data and screened near-sensing visible near-infrared reflectance characteristic wave band are used as input variables, and the INLA-SPDE three-dimensional model is used for three-dimensional mapping of soil organic matter.

[0017] Optionally, soil profile samples are collected at a plurality of sampling points in the study area, and the soil profile samples are layered according to the sampling depth of the soil profile samples and the soil organic matter content and near-sensing visible near-infrared reflectance spectrum data of each layer of each soil profile sample are measured to obtain a newly collected soil sample data set, which specifically includes the following steps:

[0018] At each sampling point in the study area, a soil profile sampler is used to collect a soil profile of 0-100 cm, which is placed in a PVC pipe and sealed.

[0019] The PVC pipe is cut longitudinally in half and evenly divided into 10 equal parts along the depth direction, and three positions are randomly selected for each layer of soil profile to be scanned. After measuring 10 times at each point, the average spectrum is taken as the near-sensing visible near-infrared reflectance spectrum data.

[0020] The other half of each layer of soil profile is divided and layered, and the potassium dichromate volumetric method-external heating method is used to measure the soil organic matter content. The soil organic matter content and near-sensing visible near-infrared reflectance spectrum data of each layer of each soil profile sample constitute the newly collected soil sample data set.

[0021] Optionally, a historical soil sample data set is obtained, and the historical soil sample data set is converted and combined with the newly collected soil sample data set to obtain an overall soil sample data set, which specifically includes the following steps:

[0022] The historical soil sample data set is obtained.

[0023] The historical collected soil organic matter content and the newly collected soil organic matter content are integrated into a unified depth layer by using an equal-area spline function.

[0024] The data in the converted historical collected soil sample dataset and the data in the newly collected soil sample dataset are merged to obtain an overall soil sample dataset.

[0025] Optionally, the soil organic matter content and the near-sensing visible near-infrared reflectance spectrum data in the newly collected soil sample dataset are preprocessed and correlation analysis is performed to screen out the near-sensing visible near-infrared reflectance spectrum characteristic bands with the highest correlation with the soil organic matter content, including the following steps:

[0026] The near-sensing visible near-infrared reflectance spectrum data is resampled to 1 nm resolution, and absorbance conversion and Savitzky-Golay smoothing and denoising processing are performed.

[0027] The soil organic matter content is subjected to log transformation processing to make it normally distributed.

[0028] The wave band with the highest correlation with the soil organic matter content is screened out by Pearson correlation analysis and is used as the near-sensing visible near-infrared reflectance spectrum characteristic band; the near-sensing visible near-infrared reflectance spectrum characteristic band also includes the bow curve difference at 600 nm.

[0029] Optionally, the standardization preprocessing of the terrain data, the soil type data, the vegetation data and the remote sensing meteorological data respectively includes spatial resampling, spatial cropping according to the boundary of the study area, numerical standardization and collinearity check; and interpolation is performed according to the near-sensing visible near-infrared reflectance spectrum characteristic bands in the newly collected soil sample dataset to obtain a near-sensing visible near-infrared reflectance spectrum characteristic band distribution map of the study area.

[0030] Optionally, according to the soil organic matter content of the surface layer, each soil profile sample in the newly collected soil sample dataset is divided into a fixed profile according to a preset ratio to establish a modeling set and a verification set, specifically: according to the soil organic matter content of the surface layer, each soil profile sample in the newly collected soil sample dataset is divided into a fixed profile according to a preset ratio of 4:1, and the classification of all profile layer sample data is completed according to the number to establish a modeling set and a verification set; the historical collected soil sample dataset is added to the modeling set as a whole; the near-sensing visible near-infrared reflectance spectrum characteristic bands of the historical sampling points are determined in the near-sensing visible near-infrared reflectance spectrum characteristic band distribution map of the study area according to the coordinates of the historical sampling points.

[0031] Optionally, the MESH grid parameters include an offset, a cutoff distance, and a maximum triangle edge length max.edge; the MESH grid parameters are set according to the maximum range, including: the offset is set according to 1 / 3-3 / 5 of the maximum range; the maximum triangle edge length is set according to 1 / 10-3 / 20 of the maximum range; and the cutoff distance is set according to 1 / 100-1 / 600 of the maximum range.

[0032] Optionally, the INLA-SPDE three-dimensional model constructed by integrating the fixed effect variable, the spatial random effect term based on the SPDE Gaussian random field model, and the deep hierarchical model includes a fixed effect, a random effect, white noise, and an intercept, as shown in the following formula:

[0033] Y(s i ,d)=β0+z1(s i ,d)β1+z2(s i ,d)β2+…+z h (s i ,d)β n +ξ(s i ,d)+ε(s i ,d)。

[0034] wherein y(s i ,d) is the organic matter content at a geographic location s i and a depth d, (i=1, 2, 3, …, n), β0 is an intercept, β1, …, β h are corresponding environmental covariate coefficients, z h (s i ,d) is the value of the hth environmental covariate at a geographic location s i and a depth d, i.e., a fixed effect, ξ(s i ,d) is a spatial random effect, wherein d is the soil profile depth, and the deep hierarchical model is calculated in the spatial random effect, i.e., a depth effect, ε(s i ,d) represents a measurement error with a mean of 0 and a variance of , ε is white noise unrelated to the spatial position.

[0035] Optionally, the spatial coordinates of the total soil sample data set sample points and the preprocessed terrain data, soil type data, vegetation data, remote sensing meteorological data, and screened near-sensing visible near-infrared reflectance spectral characteristic bands are used as input variables, and the INLA-SPDE three-dimensional model is applied to soil organic matter three-dimensional mapping, including the following steps:

[0036] The spatial coordinates of the total soil sample data set sample points, and the preprocessed terrain data, soil type data, vegetation data, remote sensing meteorological data and screened near-sensing visible near-infrared reflection spectrum characteristic bands are taken as input variables, the INLA-SPDE three-dimensional model is applied to three-dimensional mapping of soil organic matter, and the posterior distribution estimation value of the soil organic matter is obtained; the posterior distribution estimation value includes the posterior distribution mean value and the posterior distribution standard deviation value, which are used to reflect the spatial distribution characteristics of the predicted value and the uncertainty thereof.

[0037] Optionally, the SPDE Gaussian random field model is constructed based on a Matern covariance function, the SPDE Gaussian random field model is solved by a finite element method to obtain a discretized Gaussian Markov random field, and the INLA method is used for Bayesian inference to obtain the posterior distribution estimation value of the soil organic matter.

[0038] According to the specific embodiments provided in the application, the following technical effects are disclosed:

[0039] The application provides a three-dimensional mapping method for degraded farmland soil organic matter by fusing remote sensing, near-ground sensing and historical data, which significantly improves the efficiency and accuracy of deep soil organic matter mapping in black soil areas. First, traditional regional-scale deep soil organic matter three-dimensional mapping requires a large amount of soil profile excavation, and obtaining samples at different depths consumes a lot of manpower and resources, especially in large-scale black soil areas, which is extremely costly and difficult to apply on a large scale. The method proposes a modeling strategy that fuses “limited new sampling points” and “wide-range shallow historical data”, which greatly simplifies manpower and resources. The historical data provide a broad background of organic matter distribution, and the new sampling points provide key profile deep information. This design realizes high-precision three-dimensional prediction with minimal new sampling investment, significantly reducing data collection costs. Second, the application realizes a double breakthrough in accuracy and efficiency. In the INLA-SPDE three-dimensional model method, the soil profile depth is introduced into the SPDE model, which is analogous to the construction of a Gaussian random field for time series, effectively expressing the spatial heterogeneity of soil organic matter in the vertical dimension. At the same time, the near-ground visible near-infrared spectrum data are fused as key covariates to capture the vertical changes of organic matter, which significantly improves the response ability of the model to changes in deep soil properties. By integrating historical data and new sampling points for modeling, more extensive spatial coverage and background information are provided, which enhances the accurate expression of the model to regional spatial variation and reduces the prediction random error. The method can also accurately quantify the uncertainty of the prediction results, providing reliable scientific support for regional carbon storage assessment and other practical applications. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below only illustrate some of the embodiments of the present application, and for those skilled in the art, other drawings can be obtained from these drawings without creative labor.

[0041] Figure 1 A flow chart of a three-dimensional mapping method of degraded farmland soil organic matter fusing remote sensing, near-ground sensing and historical data provided by an embodiment of the present application.

[0042] Figure 2 A flow chart of step A1 in a three-dimensional mapping method of degraded farmland soil organic matter fusing remote sensing, near-ground sensing and historical data provided by an embodiment of the present application.

[0043] Figure 3 A flow chart of step A2 in a three-dimensional mapping method of degraded farmland soil organic matter fusing remote sensing, near-ground sensing and historical data provided by an embodiment of the present application.

[0044] Figure 4 A flow chart of step A3 in a three-dimensional mapping method of degraded farmland soil organic matter fusing remote sensing, near-ground sensing and historical data provided by an embodiment of the present application.

[0045] Figure 5 A predicted distribution diagram of soil organic matter in a three-dimensional spatial mapping method of soil organic matter fusing multi-source sensing data and historical data provided by an embodiment of the present application.

[0046] Figure 6 A predicted standard deviation distribution diagram in a three-dimensional spatial mapping method of soil organic matter fusing multi-source sensing data and historical data provided by an embodiment of the present application. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0048] In order to make the above purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0049] The embodiment of the present application provides a three-dimensional mapping method for degraded farmland soil organic matter by fusing remote sensing, near-ground sensing and historical data. Figure 1 As shown in the figure,

[0050] A1, collecting soil profile samples at a plurality of sampling points in a study area, layering according to the sampling depth of the soil profile samples and measuring the soil organic matter content and near-sensing visible near-infrared reflectance spectrum data of each layer of each soil profile sample to obtain a newly collected soil sample data set. In the embodiment, as shown in the figure, Figure 2 The step A1 specifically includes the following steps:

[0051] A11, at each sampling point in the study area, a soil profile of 0-100cm is collected using a soil profile sampler and is closed in a PVC pipe. The PVC pipe is closed with a plastic film outside to prevent water loss and pollution. As for the layout of the sampling points, environmental factors need to be considered comprehensively in the study area, and the conditional Latin hypercube method is used. In an exemplary embodiment, the setting of the sampling points comprehensively considers soil types, soil texture, topography, land use, rainfall and temperature and the like in the whole region of Lishu County, and the conditional Latin hypercube method is used to set 73 sampling points in Lishu County by referring to the sampling positions of the second national soil survey.

[0052] A12, the PVC pipe is cut longitudinally in half and evenly divided into 10 equal parts (0-10cm, 10-20cm, 20-30cm, …, 90-100cm) along the depth direction, and three positions are randomly selected for each layer of soil profile to be scanned, and the average spectrum is taken after 10 measurements at each point as the near-sensing visible near-infrared reflectance spectrum data. For the collection of visible near-infrared reflectance spectrum data, a strict calibration and standardization process needs to be followed, and for the 73 one-meter soil profile samples, the PVC pipe is first cut longitudinally in half in the embodiment, and the profile is evenly divided into 10 equal parts of 10cm long along the side of the PVC pipe using a marker pen, and then the visible near-infrared reflectance spectrum measurement is completed using the complete profile. The standard reference white plate is used for spectrometer calibration before and after each measurement, and the optical fiber probe is ensured to be in close contact with the surface of the soil sample during measurement. The average spectrum is taken as the final spectrum of the sample after 10 measurements at each point, so as to ensure the consistency and accuracy of the data.

[0053] Before performing visible-near-infrared reflectance spectroscopy measurements, the soil profile must be ensured to be flat and clean. Plant roots and other foreign objects should be removed, and a location with no gaps on the profile surface should be selected for collecting visible-near-infrared (Vis-NIR) spectra. A Quality Spec Trek portable spectrometer was used to measure the Vis-NIR spectrum, with a spectral range of 350-2500 nm.

[0054] A13. After dividing the other half of each soil profile, the soil organic matter content was determined by the potassium dichromate titration method-external heating method. The soil organic matter content and near-sensing visible and near-infrared reflectance spectral data of each layer of each soil profile sample constituted a newly collected soil sample dataset. In an exemplary embodiment, the data of the other half of 73 1-meter soil profiles were divided into layers to obtain soil samples at different soil depths (0-10, 10-20, 20-30, 30-40, 40-50, 50-60, 60-70, 70-80, 80-90, 90-100 cm). The samples were air-dried, ground, and sieved through a 2 mm sieve. The soil organic matter content was determined by the potassium dichromate titration method-external heating method, and then multiplied by the Van Bemmelen factor (1.724) to obtain the soil organic matter content.

[0055] A2. The soil organic matter content and near-infrared reflectance spectral data from the newly collected soil sample dataset are preprocessed and correlation analysis is performed to screen out the near-infrared reflectance spectral characteristic bands that have the highest correlation with soil organic matter content. In this embodiment, as... Figure 3 As shown, step A2 includes the following steps:

[0056] A21. Resample the near-sensing visible and near-infrared reflectance spectral data to 1nm resolution, and perform absorbance conversion and Savitzky-Golay smoothing and noise reduction processing.

[0057] A22. The soil organic matter content was subjected to log transformation to make it normally distributed.

[0058] A23. The bands with the highest correlation to soil organic matter content were selected by Pearson correlation analysis as the characteristic bands of near-visible and near-infrared reflectance spectra. The characteristic bands of near-visible and near-infrared reflectance spectra also include the bowing difference at 600 nm.

[0059] A3. Obtain the historical soil sample dataset, and merge the converted historical soil sample dataset with the newly collected soil sample dataset to obtain the overall soil sample dataset.

[0060] The historical collected soil sample data set includes soil organic matter content collected at a plurality of historical sampling points in the study area. The conversion of the historical collected soil sample data set includes integrating the historical collected soil organic matter content and the newly collected soil organic matter content into unified depth layers using an equal-area spline function.

[0061] In this embodiment, as shown in Figure 4 Step A3 specifically includes the following steps:

[0062] A31, obtaining a historical collected soil sample data set. Specifically, in this embodiment, the sampling depth of the historical soil sample data set in the degraded black soil area of Lishu County is 0-20 cm. Before integration, the historical soil organic matter data and the newly collected soil organic matter data need to be integrated into unified depth layers using an equal-area spline function.

[0063] A32, integrating the historical collected soil organic matter content and the newly collected soil organic matter content into unified depth layers using an equal-area spline function. Specifically, first, the newly collected 10 cm depth layer organic matter data is converted into unified 5 cm depth layer data. Based on the conversion result, a quantitative regression model of 0-20 cm depth layer organic matter content and original 0-10 cm and 10-20 cm depth layer content is established, and model verification is performed using significance analysis. After meeting the prediction accuracy requirement (P<0.001), the regression model is used to complete the work of predicting the corresponding 0-10 cm depth layer organic matter content using the historical data 0-20 cm soil organic matter content. After conversion, a total of 319 historical soil organic matter data of 0-10 cm layer is collected.

[0064] A33, merging the data in the converted historical collected soil sample data set and the data in the newly collected soil sample data set to obtain an overall soil sample data set.

[0065] A4, collecting topographic data, soil type data, vegetation data and remote sensing meteorological data in the study area, and standardizing and preprocessing the topographic data, soil type data, vegetation data and remote sensing meteorological data. Specifically, the standardization preprocessing of the topographic data (elevation, slope direction), soil type data (soil clay content), vegetation data (NDVI) and remote sensing meteorological data (annual average rainfall, temperature) respectively includes spatial resampling, spatial cropping according to the boundary of the study area, numerical standardization and collinearity check; and interpolation is performed according to the near-infrared visible reflectance spectral characteristic bands in the newly collected soil sample data set to obtain a near-infrared visible reflectance spectral characteristic band distribution map of the study area. Specifically, the near-infrared visible reflectance spectral data obtained from the 73 newly collected sample points are interpolated using IDW to obtain a distribution map of the visible near-infrared reflectance spectrum in Lishu County.

[0066] A5、According to the soil organic matter content of the surface layer, the soil profile samples in the newly collected soil sample data set are divided into fixed profiles according to a preset ratio, and a modeling set and a verification set are established; the data in the historically collected soil sample data set are divided into the modeling set; the modeling set includes soil organic matter content collected at a plurality of sampling points and historical sampling points, selected near-sensing visible near-infrared reflectance spectral characteristic bands, and multi-source covariate data; the multi-source covariate data includes pre-processed terrain data, soil type data, vegetation data, and remote sensing meteorological data; the near-sensing visible near-infrared reflectance spectral characteristic bands of the historical sampling points are determined in a near-sensing visible near-infrared reflectance spectral characteristic band distribution map of the research area according to the coordinates of the historical sampling points.

[0067] In an exemplary embodiment, according to the soil organic matter content of the surface layer (0-10 cm), the fixed profile is divided according to a preset ratio of 4:1, and the modeling set and the verification set are established according to the numbering to complete the classification of all profile layer sample data. In this embodiment, the collected soil profile samples are numbered, and are numbered 1-73, respectively. Each soil profile sample is divided into 10 layers, and the numbers of the 10 layers are the numbers of the profile samples, for example, the soil profile sample numbered 1 can be divided into 10 layers, and the numbers of the 10 layer samples are all 1. According to the calibrated modeling set and verification set numbers, the classification of all different layer sample data is completed. In this way, the data set division of all soil samples according to the fixed profile is completed.

[0068] A6、For each sampling point, the semi-variogram function is calculated according to the latitude and longitude of the sampling point and the soil organic matter content of each layer, the range value of the soil organic matter content of each layer is obtained, and the MESH grid parameters are set according to the maximum range. Specifically, the MESH grid parameters include offset, cutoff distance, and maximum triangle edge length. In step A6, "setting the MESH grid parameters according to the maximum range" specifically includes: setting the offset according to 1 / 3, 2 / 5, and 3 / 5 of the maximum range; setting the maximum triangle edge length according to 1 / 10, 1 / 5, and 3 / 20 of the maximum range; and setting the cutoff distance according to 1 / 100, 1 / 200, 1 / 300, and 1 / 600 of the maximum range.

[0069] After adjusting the three parameters to set the MESH grid, the triangular grid parameters are screened according to the deviance information criterion (DIC) and the root mean square error (RMSE). The smaller the DIC and RMSE values, the higher the fitting degree of the model, and the optimal model triangular grid parameters. The specific formulas of DIC and RMSE are as follows:

[0070] DIC = D bar + pD.

[0071] where D bar is the goodness of fit at the posterior mean, and pD is the effective number of parameters of the model.

[0072]

[0073] where n is the total number of samples, is the mean of the observed values, Y a is the observed values.

[0074] The RMSE and DIC of the triangular grid parameters under various combinations are compared, as shown in Table 1.

[0075] Table 1 Triangular grid parameters and their DIC and RMSE values under different combinations

[0076]

[0077]

[0078] A7 introduces the pre-processed terrain data, soil type data, vegetation data, remote sensing meteorological data and near-sensing visible and near-infrared reflectance spectral feature bands as fixed effect variables into the INLA model under the Bayesian inference framework, for explaining the systematic spatial variation of soil organic matter. In an exemplary embodiment, the INLA method is used for efficient Bayesian inference to obtain the posterior distribution estimate of soil organic matter in the spatial dimension.

[0079] A8, constructs a SPDE Gaussian random field model, introduces a covariance function to describe the three-dimensional spatial correlation, and discretizes the random field process on the MESH grid by means of the finite element method, integrates the fixed effect variable, the spatial random effect term based on the SPDE Gaussian random field model and the depth hierarchical model to construct an INLA-SPDE three-dimensional model integrating remote sensing, near-ground sensing and historical data; the SPDE Gaussian random field model is used to simulate the spatial random effect term, the depth variation of soil is analogous to a time dynamic sequence, and the depth hierarchical model is used to capture the depth variation law of soil organic matter.

[0080] The SPDE Gaussian random field model is constructed based on the Matérn covariance function, the SPDE Gaussian random field model is solved by the finite element method to obtain the discretized Gaussian Markov random field, the INLA method is used for Bayesian inference to obtain the posterior distribution estimate of soil organic matter. Specifically, the INLA-SPDE three-dimensional model constructed by integrating the fixed effect variable, the spatial random effect term based on the SPDE Gaussian random field model and the depth hierarchical model includes fixed effects, random effects, white noise and intercept, as shown in the following formula:

[0081] y(s i ,d)=β0+z1(s i ,d)β1+z2(s i ,d)β2+…+z h (s i ,d)β n +ξ(s i ,d)+ε(s i ,d)。

[0082] where y(s i ,d) is the soil organic matter content at geographic location s i and depth d, (i = 1, 2, 3, …, n), β0is the intercept, β1,…, β h are the corresponding environmental covariate coefficients, z h (s i ,d) is the hth environmental covariate value at geographic location s i and depth d, i.e., the fixed effect, ξ(s i ,d) is the spatial random effect, where d is the soil profile depth, the spatial random effect is calculated by the depth-level model, i.e., the depth effect, ε(s i ,d) represents the measurement error with mean 0 and variance , and ε is the white noise independent of the spatial location.

[0083] Specifically, the SPDE method is used to model the random effect by simulating the spatial and depth-dependent variability of soil organic matter. This method constructs a spatial Gaussian random field based on the Matérn covariance function, and ξ(s i ,d) represents the part of the variogram function in depth simulated by an autoregressive model based on the spatial variability model, which is expressed as follows:

[0084]

[0085] where ξ(s i ,d-1) is the soil organic matter process at layer d-1, a is the autocorrelation coefficient; ω(s i ,d) is the measurement error with mean 0 and variance , and satisfies the characteristics of the spatiotemporal covariance function. Assuming that the depth and space are stationary and the spatial process follows autoregressive dynamics in time, the covariance function can be expressed as the following function:

[0086]

[0087] where C(h) is determined by the Euclidean distance between two points, h is the Euclidean distance, x(i, j) is a random variable at the coordinates and depth, and d represents the depth dimension.

[0088] For the above prediction, the root mean square error (RMSE), the determination coefficient (R 2 ) and the relative analysis error (RPD) are selected as indicators to evaluate the prediction accuracy of the model, wherein the greater the values of RPD and R 2 , the better the prediction effect of the model, and the smaller the RMSE, the better the prediction effect of the model and the stronger the prediction ability.

[0089] The application discloses a three-dimensional mapping method for degraded farmland soil organic matter by fusing remote sensing, near-ground sensing and historical data, which shows the influence of different variable combinations on the prediction accuracy of the soil organic matter model, and eight variable combination schemes are used, including no variable, only near-ground Vis-NIR spectrum, only remote sensing, remote sensing combined with near-ground Vis-NIR spectrum and adding historical data, and the results are shown in Table 2.

[0090] Table 2 INLA-SPDE three-dimensional modeling performance of soil organic matter in Lishu County under different variable combinations

[0091]

[0092] The application evaluates the prediction performance of the INLA-SPDE three-dimensional model under different variable combinations, and clearly reveals the optimization mechanism of multi-source data cooperation for three-dimensional mapping of soil organic matter. As shown in Table 2, when no environmental variable is introduced, the model RMSE reaches 5.37 g / kg, R 2 is only 0.37, and RPD is 1.26; when only historical data is added and no environmental variable is introduced, the error is reduced to 4.56 g / kg, R 2 is increased to 0.5, and RPD is increased to 0.5 to reach the prediction level. When only remote sensing data is integrated, the RMSE is 5.09 g / kg, and after adding historical data, the RMSE result is increased to 4.52 g / kg, and only using ground Vis-NIR spectrum data can significantly improve the prediction accuracy (RMSE = 4.36 g / kg, R 2 = 0.58, RPD = 1.55), which proves the core value of capturing the inherent properties of soil. By fusing remote sensing and spectrum data, the model performance is further improved (RMSE = 4.16 g / kg), which shows that the combination of near sensing and remote sensing can effectively supplement the horizontal spatial heterogeneity information. Finally, under the full-factor scheme of integrating remote sensing, ground spectrum and historical data, the RMSE is further reduced to 4.03 g / kg, which is 25% lower than the prediction error of the initial model, R 2The result is 0.72, and the RPD is 1.89, which has reached the level of an excellent model, and the R 2 The increase of 95%, the increase of RPD of 50%, and the synchronous increase of the three indicators prove that Vis-NIR spectrum is the key driver for deep soil modeling, remote sensing data enhances spatial representation capability, and historical sampling points reduce model random error through longitudinal calibration. This method not only guarantees accuracy, but also proves the effectiveness of proximal sensing data, remote sensing data, and historical data, providing technical support for efficient monitoring of carbon storage in large-scale black soil areas.

[0093] A9, the spatial coordinates of the overall soil sample data set points, and the pre-processed terrain data, soil type data, vegetation data, remote sensing meteorological data and selected proximal visible near-infrared reflectance spectral characteristic bands are used as input variables, and the INLA-SPDE three-dimensional model is used for three-dimensional mapping of soil organic matter. In this embodiment, step A9 is specifically: the spatial coordinates of the overall soil sample data set points, and the pre-processed terrain data, soil type data, vegetation data, remote sensing meteorological data and selected proximal visible near-infrared reflectance spectral characteristic bands are used as input variables, and the INLA-SPDE three-dimensional model is used for three-dimensional mapping of soil organic matter, and the posterior distribution estimate value of soil organic matter is obtained; the posterior distribution estimate value includes the posterior distribution mean and the posterior distribution standard deviation value, which is used to reflect the spatial distribution characteristics of the predicted value and its uncertainty.

[0094] In an exemplary embodiment, the global three-dimensional coordinate grid and the multi-source variable matrix with annual precipitation (AMP), annual mean temperature (AMT), digital elevation model (DEM), slope aspect (Aspect), soil clay content (Clay), normalized vegetation index (NDVI), ground Vis-NIR spectrum at 600 nm, and ground Vis-NIR spectrum at 790 nm as covariates are input into the trained INLA-SPDE three-dimensional model for running, the model outputs the mean and standard deviation of the predicted soil organic matter content of each grid point, and stores it as a three-dimensional mapping result in the form of a grid, and finally completes the organic matter digital soil mapping and uncertainty spatial distribution map of 10 different depths (0-10, 10-20, 20-30, 30-40, 40-50, 50-60, 60-70, 70-80, 80-90, 90-100 cm) by stacking, as shown in Figure 5 and Figure 6 .

[0095] The method significantly optimizes the three-dimensional mapping efficiency of deep soil organic matter in black soil areas. First, traditional regional-scale deep soil organic matter three-dimensional mapping requires a large number of soil profile excavations, and obtaining samples at different depths consumes a lot of manpower and resources, especially in large-scale black soil areas, which is extremely costly and difficult to implement on a large scale. The method innovatively integrates "limited new sampling points" and "wide coverage of shallow historical data" to significantly reduce manpower and resources. The historical data serves as a spatial pattern constraint, and the spatial correlation modeling is used to assist in improving the regional background consistency. The new sampling points provide deep modeling anchors by collecting profile information. This design supports high-precision three-dimensional prediction with minimal new sampling investment.

[0096] Secondly, after deeply integrating the INLA-SPDE framework, the method uses ground Vis-NIR spectra as a key driver to capture the vertical variation of organic matter, and then uses spatial correlation modeling of historical data and new sampling points to assist in improving the regional background accuracy, effectively reducing the random error of prediction. Specifically, the historical data plays a spatial correlation role in the model, while the ground Vis-NIR spectrum is used as a deep specific environmental factor to fully capture the trend of the inherent property changes in deep soil. The fusion of the two types of data combined with the enhanced horizontal spatial heterogeneity representation of remote sensing data improves the accuracy of three-dimensional modeling. At the same time, the triangular mesh MESH improves the three-dimensional spatial correlation modeling capability, ensuring the continuity of soil types in the horizontal and vertical dimensions. The efficient Bayesian inference of the INLA algorithm significantly improves the calculation efficiency, supports fast complex model fitting and multivariate combination optimization screening, and accurately quantifies the uncertainty of the prediction results, providing reliable scientific support for regional carbon storage assessment and other applications.

[0097] The core advantage of the method is the integration of historical data and ground visible near-infrared (Vis-NIR) spectral data, which significantly improves the accuracy, efficiency and reliability of regional-scale soil organic matter three-dimensional mapping, and innovatively solves the key problem of vertical dimension modeling. The final output of the three-dimensional prediction map and the uncertainty evaluation map provides a scientific and reliable tool for organic matter dynamic monitoring and carbon storage assessment in black soil areas. The method realizes the unification of low-cost sampling and high-precision three-dimensional output through precise sampling strategy and deep coupling of multi-source data.

[0098] The technical features of the above embodiments can be combined in any way. To make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combinations of the technical features do not exist, they should be considered as the scope of the present disclosure.

[0099] The principles and implementation manners of the present application are described herein by using specific examples, and the above examples are only used to help understand the method of the present application and its core idea; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation manners and application ranges will have changes. In conclusion, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A method of mapping soil organic matter in degraded cropland in three dimensions by fusing remote sensing, proximal sensing and historical data, characterized in that, Comprise: Collect soil profile samples at several sampling points in the study area, and divide them into layers according to the sampling depth of the soil profile samples and measure the soil organic matter content and near-infrared reflectance spectrum data of each layer of the soil profile samples to obtain a newly collected soil sample dataset; Pretreat the soil organic matter content and near-infrared reflectance spectrum data in the newly collected soil sample dataset and perform correlation analysis to select the near-infrared reflectance spectrum characteristic band with the highest correlation with the soil organic matter content; Obtain a historical soil sample dataset and combine it with the newly collected soil sample dataset after conversion to obtain a total soil sample dataset; the historical soil sample dataset includes soil organic matter content collected at several historical sampling points in the study area, and the conversion of the historical soil sample dataset includes using an equal-area spline function to integrate the historical soil organic matter content and the newly collected soil organic matter content into a unified depth layer; Collect topographic data, soil type data, vegetation data, and remote sensing meteorological data in the study area, and standardize and pretreat the topographic data, soil type data, vegetation data, and remote sensing meteorological data; According to the soil organic matter content of the surface layer, divide each soil profile sample in the newly collected soil sample dataset into a fixed profile according to a predetermined proportion, and establish a modeling set and a validation set; the data in the historical soil sample dataset are divided into the modeling set; the modeling set includes soil organic matter content collected at several sampling points and historical sampling points, selected near-infrared reflectance spectrum characteristic bands, and multi-source covariate data; the multi-source covariate data includes pretreated topographic data, soil type data, vegetation data, and remote sensing meteorological data; the near-infrared reflectance spectrum characteristic bands of the historical sampling points are determined according to the coordinates of the historical sampling points in the near-infrared reflectance spectrum characteristic band distribution map of the study area; For each sampling point, calculate the semi-variogram according to the latitude and longitude of the sampling point and the soil organic matter content of each layer to obtain the range value of the soil organic matter content of each layer, and set the MESH grid parameters according to the maximum range; Introduce the pretreated topographic data, soil type data, vegetation data, remote sensing meteorological data, and near-infrared reflectance spectrum characteristic bands as fixed effect variables into the INLA model under the Bayesian inference framework to explain the systematic spatial variation of soil organic matter. The SPDE Gaussian random field model is constructed, a covariance function is introduced to describe the three-dimensional spatial correlation, and the finite element method is used to discretize the random field process on the MESH grid, and the INLA-SPDE three-dimensional model integrating remote sensing, near-ground sensing and historical data is constructed by integrating fixed effect variables, spatial random effect items based on the SPDE Gaussian random field model and depth hierarchical model; the SPDE Gaussian random field model is used to simulate the spatial random effect item, the depth variation of the soil layer is analogized as a time dynamic sequence, and the depth hierarchical model is used to capture the depth variation law of the soil organic matter; The spatial coordinates of the total soil sample data set sample points, the pretreated terrain data, soil type data, vegetation data, remote sensing meteorological data and screened near-sensing visible near-infrared reflectance spectral characteristic bands are used as input variables, and the INLA-SPDE three-dimensional model is used for soil organic matter three-dimensional mapping.

2. The method of claim 1, wherein the method further comprises: Soil profile samples are collected at several sampling points in the study area, and the soil profile samples are layered according to the sampling depth of the soil profile samples, and the soil organic matter content and near-sensing visible near-infrared reflectance spectral data of each layer of each soil profile sample are measured to obtain a newly collected soil sample data set, which specifically includes: At each sampling point in the study area, a soil profile sampler is used to collect a soil profile of 0-100 cm, which is placed in a PVC pipe and sealed; The PVC pipe is cut longitudinally and evenly divided into 10 equal parts along the depth direction, and three positions are randomly selected for each layer of soil profile to be scanned, and the average spectrum is taken after measuring 10 times at each point as the near-sensing visible near-infrared reflectance spectral data; The other half of each layer of soil profile is divided and layered, and the potassium dichromate volumetric method-external heating method is used to measure the soil organic matter content; the soil organic matter content and near-sensing visible near-infrared reflectance spectral data of each layer of each soil profile sample constitute the newly collected soil sample data set.

3. The method of claim 1, wherein the method further comprises: The historical collected soil sample data set is obtained, and the converted historical collected soil sample data set is combined with the newly collected soil sample data set to obtain a total soil sample data set, which specifically includes: The historical collected soil sample data set is obtained; The historical collected soil organic matter content and the newly collected soil organic matter content are integrated into a unified depth layer by using an equal-area spline function; The data in the converted historical collected soil sample data set and the data in the newly collected soil sample data set are combined to obtain a total soil sample data set.

4. The method of claim 1, wherein the method further comprises: The soil organic matter content and near-sensing visible near-infrared reflectance spectral data in the newly collected soil sample data set are pretreated and correlation analysis is performed, and the near-sensing visible near-infrared reflectance spectral characteristic bands with the highest correlation with the soil organic matter content are screened out, including: The near-sensing visible near-infrared reflectance spectral data is resampled to 1 nm resolution, and absorbance conversion and Savitzky-Golay smoothing denoising processing are performed; The soil organic matter content is log transformed to make it normally distributed; The band with the highest correlation with the soil organic matter content is screened through a Pearson correlation analysis, as a near-sensing visible near-infrared reflectance spectral characteristic band; the near-sensing visible near-infrared reflectance spectral characteristic band also includes a bowtie difference at 600 nm.

5. The method of claim 1, wherein the method further comprises: The standardization preprocessing of the terrain data, the soil type data, the vegetation data and the remote sensing meteorological data respectively includes spatial resampling, spatial cropping according to the boundary of the study area, numerical standardization and collinearity checking; and interpolation is performed according to the near-sensing visible near-infrared reflectance spectral characteristic band in the newly collected soil sample data set, to obtain a near-sensing visible near-infrared reflectance spectral characteristic band distribution map of the study area.

6. The method of claim 1, wherein the method further comprises: According to the soil organic matter content of the surface layer, each soil profile sample in the newly collected soil sample data set is divided into fixed profiles according to a preset ratio, to establish a modeling set and a verification set, specifically: according to the soil organic matter content of the surface layer, each soil profile sample in the newly collected soil sample data set is divided into fixed profiles according to a preset ratio of 4:1, and the classification of all profile layer sample data is completed according to the number, to establish a modeling set and a verification set; the entire historical collected soil sample data set is added to the modeling set; the near-sensing visible near-infrared reflectance spectral characteristic band of the historical sampling point is determined in the near-sensing visible near-infrared reflectance spectral characteristic band distribution map of the study area according to the coordinates of the historical sampling point.

7. The method of claim 1, wherein the method further comprises: The MESH grid parameters include an offset, a cutoff distance and a maximum triangle edge length; the MESH grid parameters are set according to the maximum range, including: the offset is set according to 1 / 3-3 / 5 of the maximum range; the maximum triangle edge length is set according to 1 / 10-3 / 20 of the maximum range; and the cutoff distance is set according to 1 / 100-1 / 600 of the maximum range.

8. The method of claim 1, wherein the method further comprises: The INLA-SPDE three-dimensional model integrating a fixed effect variable, a spatial random effect term based on a SPDE Gaussian random field model and a depth hierarchical model includes a fixed effect, a random effect, white noise and an intercept, as shown in the following formula: y(s i ,d) = β0+ z1(s i ,d)β1+ z2(s i ,d)β2+... + z h (s i ,d)β n + ξ(s i ,d) + ε(s i ,d) ; where y(s i ,d) is the organic matter content at geographic location s i and depth d, (i = 1, 2, 3,..., n), β0is the intercept, β1,..., β h are the corresponding environmental covariate coefficients, z h (s i ,d) is the hth environmental covariate value at geographic location s i and depth d, i.e. the fixed effect, ξ(s i ,d) is the spatial random effect, where d is the soil profile depth, and the depth level model is calculated in the spatial random effect, i.e. the depth effect, ε(s i ,d) represents the measurement error with mean 0 and variance , ε is the white noise independent of the spatial location.

9. The method of claim 1, wherein the method further comprises: The spatial coordinates of the sample points of the overall soil sample data set and the preprocessed terrain data, soil type data, vegetation data, remote sensing meteorological data and screened near-sensing visible near-infrared reflectance spectral characteristic bands are used as input variables, and the INLA-SPDE three-dimensional model is applied to perform soil organic matter three-dimensional mapping, including: The spatial coordinates of the sample points of the overall soil sample data set and the preprocessed terrain data, soil type data, vegetation data, remote sensing meteorological data and screened near-sensing visible near-infrared reflectance spectral characteristic bands are used as input variables, and the INLA-SPDE three-dimensional model is applied to perform soil organic matter three-dimensional mapping, and a posterior distribution estimation value of the soil organic matter is obtained; the posterior distribution estimation value includes a posterior distribution mean value and a posterior distribution standard deviation value, for reflecting the spatial distribution characteristics and uncertainty of the predicted value.

10. The method of claim 1, wherein the method further comprises: The SPDE Gaussian random field model is constructed based on a Matern covariance function, is solved by a finite element method to obtain a discretized Gaussian Markov random field, and is subjected to Bayesian inference by an INLA method to obtain an estimated value of a posterior distribution of soil organic matter.

Images