Soil water and salt dynamic monitoring and quality evaluation method based on multispectral remote sensing

Abstract

The invention discloses a soil water and salt dynamic monitoring and quality evaluation method based on multispectral remote sensing, and relates to the field of geological monitoring, and the method comprises the steps: carrying out the scattering deviation correction and roughness correction of an observation spectrum through a satellite remote sensing image, an unmanned aerial vehicle multispectral image, a ground sample and meteorological driving data; the soil intrinsic reflectivity, the vegetation coverage component and the salt crusting component are obtained; a water-salt inversion model is established by combining thermal infrared and red edge wave band information, and multi-temporal soil volumetric moisture content and surface salinity index are inverted; by introducing meteorological conditions and vegetation dynamic characteristics, a water-salt coupling partial differential equation and a graph space-time constraint model are constructed, and continuous space-time distribution of soil water content and salinity is obtained; extracting salinity, moisture, vegetation response and soil health indexes, establishing a soil quality comprehensive evaluation model, and outputting high-risk plaques and a treatment priority sequence. The method realizes dynamic monitoring and quality grading evaluation of soil water and salt, and is suitable for saline-alkali soil treatment and ecological restoration.

Description

TECHNICAL FIELD

[0001] The present specification relates to the field of geological monitoring, and more particularly, the present application relates to a soil water and salt dynamic monitoring and quality evaluation method based on multi-spectral remote sensing. BACKGROUND

[0002] The existing soil water and salt monitoring and quality evaluation technology mainly relies on ground sampling and experimental analysis means, such as soil conductivity measurement, volume water content detection and chemical analysis. Although this kind of method has high accuracy, it has typical "point limitation" and "time lag" problems: first, the number of sampling points is limited, which is difficult to reflect the spatial heterogeneity of water and salt distribution at regional scale; second, the experimental detection period is long, which cannot meet the continuous monitoring demand of soil water and salt dynamic change.

[0003] In addition, although the traditional remote sensing inversion method can realize large-scale estimation, it is usually based on single time image or single spectral index, ignoring the influence of time series information and meteorological driving factors, resulting in poor model stability and insufficient accuracy. At the same time, the existing method often does not consider the spectral interference factors such as surface roughness and incident angle difference, so that the inversion of soil reflectivity has systematic deviation, which further affects the accuracy of water and salt estimation. Further, the existing soil quality evaluation method is mostly based on single index (such as salt concentration or water content) for classification and discrimination, which lacks comprehensive consideration of multi-dimensional ecological factors, and it is difficult to reveal the spatial distribution and evolution trend of soil degradation risk.

[0004] Therefore, it is necessary to propose a soil water and salt dynamic monitoring and quality evaluation method based on multi-spectral remote sensing to solve the above problems. SUMMARY

[0005] A series of simplified concepts are introduced in the summary part, which will be further described in detail in the specific embodiment part. The summary part of the present application does not mean to try to limit the key features and necessary technical features of the claimed technical solution, and does not mean to try to determine the protection scope of the claimed technical solution.

[0006] In the first aspect, the present application proposes a soil water and salt dynamic monitoring and quality evaluation method based on multi-spectral remote sensing, comprising: obtaining multi-source multi-temporal multi-spectral data, ground sample data and meteorological driving data; wherein the multi-source multi-temporal multi-spectral data includes satellite remote sensing image and unmanned aerial vehicle multi-spectral camera image, the ground sample data includes volume water content, surface and profile conductivity and conductivity ratio, soil texture and bulk density, and the meteorological driving data includes precipitation and evapotranspiration; The multi-source multi-temporal multi-spectral data, the ground sample data and the meteorological driving data are assimilated and registered in the time and space scale; Based on the assimilated multi-temporal multi-spectral data and the ground sample data, a soil, vegetation and salt crust three-domain spectral component unmixing model is constructed, anisotropic scattering correction and roughness correction are performed on the observed spectrum in the above multi-temporal multi-spectral data, and soil intrinsic reflectivity, vegetation coverage component and salt crust component are obtained; Based on the above soil intrinsic reflectivity, the above vegetation coverage component and the above salt crust component, and combined with the thermal infrared and red edge band information in the above satellite remote sensing data, a water-salt inversion model is established, and multi-temporal soil water-salt inversion results are obtained; The above multi-temporal water-salt inversion results are taken as observation input, combined with the time series change of the above vegetation coverage component and the assimilated meteorological driving data, a water-salt coupled partial differential equation and a graph space-time regularization constraint model are constructed, and through joint optimization of the physical domain and the data domain, continuous spatio-temporal distribution fields of soil water content and salt content are obtained; Based on the above continuous spatio-temporal distribution field, a salinity characteristic index and a water content characteristic index are extracted, a vegetation response index is extracted based on the time series change of the above vegetation coverage component, and a soil health index is determined based on the above soil intrinsic reflectivity; Based on the above salinity characteristic index, the above water content characteristic index, the above vegetation response index and the above soil health index, a soil quality comprehensive evaluation model is constructed, the soil quality of different regions is quantitatively scored and classified, and high-risk patches, hot migration paths and treatment priority orders are output.

[0007] In a feasible implementation, the above based on the assimilated multi-temporal multi-spectral data and the ground sample data, the soil, vegetation and salt crust three-domain spectral component unmixing model is constructed, the anisotropic scattering correction and the roughness correction are performed on the observed spectrum in the above multi-temporal multi-spectral data, and the soil intrinsic reflectivity, the vegetation coverage component and the salt crust component are obtained, including: Based on the above assimilated multi-temporal multi-spectral data, the observed spectral reflectivity information of each observation pixel is extracted, and combined with the typical soil, vegetation and salt crust sample spectrum obtained from the above ground sample data, the standard spectral characteristics of the three types of end members are determined; The above standard spectral characteristics are taken as input, a mixed spectrum expression composed of soil end members, vegetation end members and salt crust end members is established to represent the composition relationship of each pixel spectrum; Through decomposition and solving of the above mixed spectrum expression, the component weight of the above soil end member, the above vegetation end member and the above salt crust end member in each pixel is obtained, so that the preliminary spectral component proportion is obtained; In view of the reflection difference of the above observed spectrum under different incident angles and observation angles, anisotropic scattering correction operation is performed to correct the reflection error caused by the solar elevation angle and the sensor viewing angle; The spectrum data after anisotropic scattering correction is corrected for surface roughness to eliminate the influence of surface micro-morphology difference on the spectrum, to obtain the soil intrinsic reflectance, the vegetation coverage component and the salt crust component.

[0008] In a feasible implementation, the component weight of the soil endmember, the vegetation endmember and the salt crust endmember in each pixel is obtained by decomposing and solving the mixed spectrum expression, to obtain the preliminary spectrum component ratio, including: The mixed spectrum expression is converted into a matrix equation in matrix form, wherein the matrix equation takes the observed spectrum as an observation vector, and takes a matrix composed of endmember spectrum basis vectors as a coefficient matrix; The least square method is used to solve the algorithm, and the matrix equation is inverted to obtain the component weight initial value of each endmember; The component weight initial value is processed by non-negative constraint and normalization, so that the sum of the component weights of the soil endmember, the vegetation endmember and the salt crust endmember is equal to one; The component weight result after constraint optimization is output as the preliminary spectrum component ratio.

[0009] In a feasible implementation, the anisotropic scattering correction operation is performed on the reflection difference of the observed spectrum under different incident angles and observation angles, to correct the reflection error caused by the solar elevation angle and the sensor viewing angle, including: Based on the imaging geometric information of the multi-temporal multi-spectral data, geometric parameters corresponding to the moment are obtained, wherein the geometric parameters include the solar incident angle, the azimuth angle and the sensor observation angle parameter; According to the geometric parameters, the relative observation geometric relationship of each pixel is calculated, and a reflection anisotropic response function is established; the original reflectance of the observed spectrum is angle normalized by using the anisotropic response function, to obtain the corrected reflectance under the standard incident angle and observation angle conditions; The corrected reflectance is output as the input data for subsequent surface roughness correction, to correct the reflection error caused by the solar elevation angle and the sensor viewing angle.

[0010] In a feasible implementation, the spectrum data after anisotropic scattering correction is corrected for surface roughness to eliminate the influence of surface micro-morphology difference on the spectrum, to obtain the soil intrinsic reflectance, the vegetation coverage component and the salt crust component, including: Based on the ground sample data and the digital elevation model data, regional surface roughness feature parameters are extracted, wherein the surface roughness parameters include the slope, the slope direction and the surface microstructure scale; Based on the surface roughness feature parameters, a roughness influence function and a roughness correction coefficient are constructed; The spectral data after the anisotropic scattering correction and the roughness influence function are combined to correct the spectral reflectance, and the spectral scattering distortion caused by the surface micro-morphology is compensated. The spectral data after the roughness correction is output to obtain the soil intrinsic reflectance, the vegetation cover component and the salt crust component.

[0011] In a feasible implementation, based on the soil intrinsic reflectance, the vegetation cover component and the salt crust component, and combined with the thermal infrared and red edge band information in the satellite remote sensing data, a water-salt inversion model is established to obtain the multi-time-phase soil water-salt inversion result, including: Based on the soil intrinsic reflectance, spectral feature parameters in the near-infrared and short-wave infrared bands are extracted; The spectral feature parameters and the thermal infrared band information in the satellite remote sensing data are jointly analyzed to calculate the surface temperature feature quantity; The salt-sensitive feature index is extracted using the red edge band information, and the salt crust component is introduced as a constraint factor to correct the spectral aliasing error in the salt inversion; The spectral feature parameters, the surface temperature feature quantity and the salt-sensitive feature index are jointly input into a multi-index collaborative inversion model to calculate the soil volume water content and the surface salt content by statistical regression to generate a preliminary inversion result; The preliminary inversion result is corrected for time sequence consistency to maintain the physical continuity of the water-salt change trend between adjacent time phases, thereby obtaining the multi-time-phase soil water-salt inversion result.

[0012] In a feasible implementation, the multi-time-phase water-salt inversion result is input as an observation, combined with the time sequence change of the vegetation cover component and the assimilated and registered meteorological driving data, to construct a water-salt coupled partial differential equation and graph spatiotemporal regularization constraint model, and through joint optimization of the physical domain and the data domain, the continuous spatiotemporal distribution field of the soil water content and the salt content is obtained, including: The multi-time-phase water-salt inversion result is input as an observation, combined with the time sequence change of the vegetation cover component and the assimilated and registered meteorological driving data, the precipitation is taken as the infiltration and water replenishment boundary condition, and the evapotranspiration is taken as the surface water loss term to construct a water-salt coupled partial differential equation and graph spatiotemporal regularization constraint model; Through joint optimization of the physical domain and the data domain, the continuous spatiotemporal distribution field of the soil water content and the salt content under the driving of the meteorological condition is obtained.

[0013] In a feasible implementation, the soil quality comprehensive evaluation model is constructed based on the salinity characteristic index, the moisture characteristic index, the vegetation response index, and the soil health index, the soil quality of different regions is quantitatively scored and classified, the high-risk patches, the hot migration path, and the treatment priority order are output, including: The salinity characteristic index and the moisture characteristic index are associated with the vegetation response index obtained from the time sequence change of the vegetation coverage component The salinity characteristic index, the moisture characteristic index, and the vegetation response index are normalized in combination with the soil health index, and a soil quality evaluation feature matrix is established; Based on the soil quality evaluation feature matrix, a weighted comprehensive evaluation algorithm is used to construct the soil quality comprehensive evaluation model, and the soil quality comprehensive score of each region is calculated; According to the soil quality comprehensive score, grade division and spatial clustering analysis are performed, the high-risk patches are output, the hot migration path is identified, and the treatment priority order is generated.

[0014] In a feasible implementation, the determination of the high-risk patches includes: Based on the soil quality comprehensive score, the score distribution characteristics of each monitoring region are counted, and the statistical results of the score are calculated, wherein the statistical results include the mean, the standard deviation, and the extreme value range; According to the statistical results, the grade threshold interval is set, the soil quality comprehensive score is divided into multiple grades, and the regions in the low score interval are identified as potential high-risk areas; Each grade region is spatially encoded and vectorized to generate a soil quality grade distribution layer; The soil quality grade distribution layer is input into a spatial clustering analysis module, and adjacent high-risk pixel groups are identified based on a density clustering algorithm to form the risk patches.

[0015] In a feasible implementation, the determination of the treatment priority order includes: According to the change trend of the high-risk patches in the time sequence, the hot migration path is extracted; In combination with the spatial clustering intensity of the high-risk patches, the migration rate of the hot migration path, and the soil health score, the comprehensive risk weight of each region is calculated, and the treatment priority order is determined according to the comprehensive risk weight.

[0016] In summary, by simultaneously acquiring satellite remote sensing images, unmanned aerial vehicle multispectral data, ground sample data, and meteorological driving data, macro-micro, multi-temporal-multi-scale fusion monitoring is realized. The method unifies data of different sources and different resolutions to the same spatial coordinates and time scale through spatio-temporal assimilation registration technology, thereby eliminating the offset error between multi-source data and significantly improving the comparability and continuity of the monitoring results. The method proposes a "soil-vegetation-salt crust" three-domain spectral component unmixing model, which effectively compensates for the reflection error caused by the incident angle and the terrain microstructure through anisotropic scattering correction and surface roughness correction, to obtain soil intrinsic reflectance closer to the true physical state. Compared with the traditional linear mixing model, the method can more accurately distinguish the spectral characteristics of soil, vegetation, and salt crust, providing high-quality input for water-salt inversion. The method combines thermal infrared band and red edge band information to establish a water-salt collaborative inversion model, and by introducing the surface temperature feature quantity and the salt sensitive index (SSI), the multi-parameter joint inversion of soil volumetric water content and surface salt is realized. Compared with the traditional model based on single band or single index, the present invention can significantly improve the robustness and temporal consistency of soil water-salt inversion. The present invention builds a water-salt coupled partial differential equation and introduces a graph space regularization constraint model, so that the physical mechanism (precipitation, evapotranspiration, and infiltration) and remote sensing observation data jointly act on the optimization process to obtain the continuous spatio-temporal distribution field of soil water content and salt. The fusion method breaks through the limitation of traditional data-driven models lacking physical constraints, and realizes the dynamic tracking of water-salt temporal evolution. In the soil quality evaluation stage, the method considers saltiness characteristics, water features, vegetation response, and soil health as four types of indicators, calculates the soil quality score by using weighted synthesis or principal component analysis method, and identifies high-risk patches and hot migration paths by combining spatial clustering and temporal variation. By calculating the comprehensive risk weight, the method can output the priority order of treatment, providing quantitative decision-making basis for saline-alkali land treatment and ecological restoration. The method has both the explainability of physical models and the generalization ability of machine learning, and is suitable for long-term water-salt dynamic monitoring and soil degradation evaluation in arid regions, irrigation areas, and coastal salinization areas, and can be widely applied to agricultural irrigation management, saline-alkali land improvement, ecological restoration, and regional sustainable land use planning. In summary, the method fuses multi-source and multi-temporal remote sensing data, builds a joint model of physical constraints and data-driven, and introduces a multi-index comprehensive evaluation system, which not only solves the shortcomings of traditional technology in data continuity, inversion accuracy, and comprehensive evaluation, but also realizes the leap from "static monitoring" to "dynamic prediction".

[0017] Other advantages, objects, and features of the present application will be apparent to those skilled in the art from the following specification. BRIEF DESCRIPTION OF DRAWINGS

[0018] Various other advantages and benefits will become apparent to those of ordinary skill in the art, upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of preferred embodiments, and are not meant to limit the present description. Moreover, the same reference numerals in different figures represent the same or similar elements. In the drawings:

[0019] Figure 1 A flowchart of a soil water and salt dynamic monitoring and quality evaluation method based on multispectral remote sensing provided by an embodiment of the present application is shown in FIG. 1. Figure 2 A flowchart of an anisotropic scattering correction and roughness correction method provided by an embodiment of the present application is shown in FIG. 2. Figure 3 A flowchart of a method for obtaining preliminary spectral component proportions provided by an embodiment of the present application is shown in FIG. 3. Figure 4 A flowchart of an anisotropic scattering correction operation method provided by an embodiment of the present application is shown in FIG. 4. Figure 5 A flowchart of a method for eliminating the influence of surface micro-morphological differences on spectra provided by an embodiment of the present application is shown in FIG. 5. Figure 6 A flowchart of a method for obtaining multi-temporal soil water and salt inversion results provided by an embodiment of the present application is shown in FIG. 6. Figure 7 A flowchart of a method for obtaining continuous spatiotemporal distribution fields of soil water content and salt content provided by an embodiment of the present application is shown in FIG. 7. Figure 8 A flowchart of a method for quantitatively scoring and grading the soil quality of different regions provided by an embodiment of the present application is shown in FIG. 8. DETAILED DESCRIPTION

[0020] The terms "first", "second", "third", "fourth" and the like in the description and in the claims of the present application, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the use of these terms herein is to be construed to cover a general order and / or structure unless otherwise indicated by the context thereof. Furthermore, the term "comprising" and variations thereof, as used in describing and claiming the embodiments of the present application, are intended to be open and permissive, i.e., meaning the processes, methods, articles, or apparatuses encompassed can include more steps, elements, features, parts, or components than those explicitly stated or implied by reference to the disclosure herein. The following description of the embodiments of the present application is provided, together with the accompanying drawings, to show and describe illustrative embodiments of the present application. It is to be understood that the following description is only exemplary and explanatory of the present application and is not intended to be restrictive of the present application, as the present application is defined by the appended claims.

[0021] Referring to Figure 1 A flowchart of a soil water and salt dynamic monitoring and quality evaluation method based on multispectral remote sensing provided for an embodiment of the present application can specifically include: In a first aspect, the present application provides a soil water and salt dynamic monitoring and quality evaluation method based on multispectral remote sensing, comprising: S110, acquiring multi-source multi-temporal multispectral data, ground sample data and meteorological driving data; wherein the multi-source multi-temporal multispectral data includes satellite remote sensing images and unmanned aerial vehicle multispectral camera images, the ground sample data includes volumetric water content, surface and profile conductivity and conductivity ratio, soil texture and bulk density, and the meteorological driving data includes precipitation and evapotranspiration.

[0022] S120, assimilating and registering the multi-source multi-temporal multispectral data, the ground sample data and the meteorological driving data in the time and space scales.

[0023] S130, based on the assimilated and registered multi-temporal multispectral data and ground sample data, constructing a soil, vegetation and salt crust three-domain spectral component unmixing model, performing anisotropic scattering correction and roughness correction on the observed spectrum in the multi-temporal multispectral data to obtain soil intrinsic reflectance, vegetation cover component and salt crust component.

[0024] S140, based on the soil intrinsic reflectance, the vegetation cover component and the salt crust component, and combining the thermal infrared and red edge band information in the satellite remote sensing data, establishing a water and salt inversion model to obtain multi-temporal soil water and salt inversion results.

[0025] S150, taking the multi-temporal water and salt inversion results as observation input, combining the temporal variation of the vegetation cover component and the assimilated meteorological driving data, constructing a water and salt coupled partial differential equation and graph time and space regularization constraint model, and obtaining continuous spatiotemporal distribution fields of soil water content and salt content through joint optimization of the physical domain and the data domain.

[0026] S160, based on the continuous spatiotemporal distribution fields, extracting salt concentration characteristic indexes and water content characteristic indexes, extracting vegetation response indexes based on the temporal variation of the vegetation cover component, and determining soil health indexes based on the soil intrinsic reflectance.

[0027] S170, based on the salt concentration characteristic indexes, the water content characteristic indexes, the vegetation response indexes and the soil health indexes, constructing a soil quality comprehensive evaluation model, quantitatively scoring and grading the soil quality of different regions, and outputting high-risk patches, hot migration paths and treatment priority orders.

[0028] Exemplarily, in step S110, multi-source, multi-temporal and multi-spectral data, ground sample data and meteorological driving data are acquired to provide multi-dimensional information support for subsequent analysis. The multi-spectral data cover satellite remote sensing images and unmanned aerial vehicle multi-spectral camera images, which can simultaneously capture macro features and local high-resolution details; the ground sample data provide real physical properties such as volumetric water content, conductivity ratio and soil texture parameters; and the precipitation and evapotranspiration in the meteorological driving data reflect the external water cycle environment, providing boundary conditions for dynamic modeling.

[0029] In step S120, the data consistency and comparability are achieved by aligning different time, spatial resolution and source data to a unified coordinate and time reference through spatio-temporal assimilation registration of the above-mentioned multi-source heterogeneous data. The process makes remote sensing observation results and ground measured data cooperate in the spatio-temporal domain through the fusion of geographic reference points and time series interpolation.

[0030] In step S130, a soil-vegetation-salt crust three-domain spectral component unmixing model is constructed based on the multi-temporal and multi-spectral data and the ground sample data after assimilation registration. Through anisotropic scattering correction and roughness correction of the observed spectrum, the interference of different incident angles and surface micro-morphology on spectral reflection is eliminated, so as to extract soil intrinsic reflectance, vegetation coverage component and salt crust component, which provide independent physical characteristics for subsequent water and salt inversion.

[0031] In step S140, a multi-index water and salt collaborative inversion model is established by combining the extracted spectral features and the thermal infrared and red edge band information in the satellite remote sensing data. The model analyzes the change of the surface temperature in the thermal infrared band and the sensitive response of the red edge band to salt, and inverses the soil volumetric water content and the surface salt index, so as to realize the quantitative estimation of the soil water and salt distribution at different time nodes.

[0032] In step S150, the multi-temporal water and salt inversion results, the time series change of the vegetation coverage component and the meteorological driving data are jointly input into a water and salt coupling partial differential equation model. By introducing the precipitation and evapotranspiration boundary conditions in the physical domain and combining the spatio-temporal regularization constraints, joint optimization is performed in the data domain, so as to obtain the soil water content and salt distribution field in the continuous spatio-temporal scale, and reveal the time series evolution law.

[0033] In step S160, the salinity characteristic index and the water characteristic index are extracted based on the above-mentioned continuous spatio-temporal distribution field, so as to quantitatively characterize the soil salt accumulation and water change characteristics; at the same time, the vegetation response index is extracted according to the time series change of the vegetation coverage component, so as to reflect the adaptation or stress effect of the vegetation growth to the water and salt change; and the soil health index is determined based on the soil intrinsic reflectance inversion result, which is used to evaluate the long-term stable properties such as soil structure, texture and organic matter content.

[0034] In step S170, the soil quality comprehensive evaluation model is constructed by synthesizing the aforementioned multiple indicators. The model calculates the soil quality comprehensive score of each region through a weighted synthesis algorithm or principal component analysis method, and classifies it according to the classification threshold interval to output the high-risk patch region. Combined with the time series change trend, the hot migration path is further extracted to reveal the spatiotemporal migration law of the high-salt and high-water risk area. Finally, according to the comprehensive risk weight and spatial clustering characteristics, the management priority order is generated to provide scientific decision-making basis for soil improvement, water resource regulation and ecological restoration.

[0035] In summary, the method realizes macro-micro, multi-temporal-multi-scale fusion monitoring by simultaneously acquiring satellite remote sensing images, unmanned aerial vehicle multi-spectral data, ground sample data and meteorological driving data. The method unifies data of different sources and different resolutions to the same spatial coordinates and time scale through space-time assimilation registration technology, thereby eliminating the offset error between multi-source data and significantly improving the comparability and continuity of the monitoring results. The method proposes a "soil-vegetation-salt crust" three-domain spectral component unmixing model, which effectively compensates for the reflection error caused by the incident angle and the terrain microstructure through anisotropic scattering correction and surface roughness correction, to obtain soil intrinsic reflectivity closer to the true physical state. Compared with the traditional linear mixing model, the method can more accurately distinguish the spectral characteristics of soil, vegetation and salt crust, providing high-quality input for water-salt inversion. The method combines thermal infrared band and red edge band information to establish a water-salt collaborative inversion model, and realizes the multi-parameter joint inversion of soil volumetric water content and surface salt by introducing the surface temperature characteristic quantity and the salt sensitive index (SSI). Compared with the traditional model based on single band or single index, the present invention can significantly improve the robustness and time sequence consistency of soil water-salt inversion. The present invention builds a water-salt coupled partial differential equation and introduces a graph space regularization constraint model, so that the physical mechanism (precipitation, evapotranspiration, infiltration) and remote sensing observation data jointly act on the optimization process, thereby obtaining the continuous spatio-temporal distribution field of soil water content and salt. The fusion method breaks through the limitation of traditional data-driven models lacking physical constraints, and realizes the dynamic tracking of water-salt time series evolution. In the soil quality evaluation stage, the method considers saltiness characteristics, water characteristics, vegetation response and soil health four indexes, and calculates the soil quality score by using weighted comprehensive or principal component analysis method, and identifies high-risk patches and hot migration paths by combining spatial clustering and time sequence change. By calculating the comprehensive risk weight, the method can output the priority of treatment, providing quantitative decision basis for saline-alkali land treatment and ecological restoration. The method has both the explainability of physical models and the generalization ability of machine learning, and is suitable for long-term water-salt dynamic monitoring and soil degradation evaluation in arid regions, irrigation areas and coastal salinization areas, and can be widely used in agricultural irrigation management, saline-alkali land improvement, ecological restoration and regional sustainable land use planning scenes. In summary, the method fuses multi-source multi-temporal remote sensing data, builds a joint model of physical constraints and data-driven, and introduces a multi-index comprehensive evaluation system, which not only solves the shortcomings of traditional technology in data continuity, inversion accuracy and comprehensive evaluation, but also realizes the leap from "static monitoring" to "dynamic prediction".

[0036] In one possible implementation, as Figure 2As shown in the above step S130, based on the assimilated multi-temporal multi-spectral data and the ground sample data, a soil, vegetation and salt crust three-domain spectral component unmixing model is constructed, anisotropic scattering correction and roughness correction are performed on the observed spectrum in the above multi-temporal multi-spectral data to obtain soil intrinsic reflectivity, vegetation coverage component and salt crust component, including: S1301, based on the above assimilated multi-temporal multi-spectral data, the observed spectral reflectivity information of each observed pixel is extracted, and combined with the typical soil, vegetation and salt crust sample spectrum obtained from the above ground sample data, the standard spectral characteristics of the three types of end members are determined; S1302, the above standard spectral characteristics are input to establish a mixed spectrum expression composed of soil end members, vegetation end members and salt crust end members to represent the composition relationship of each pixel spectrum; S1303, by decomposing and solving the above mixed spectrum expression, the component weight of the above soil end member, the above vegetation end member and the above salt crust end member in each pixel is obtained, so as to obtain the preliminary spectral component proportion; S1304, according to the reflection difference of the above observed spectrum under different incident angles and observation angles, anisotropic scattering correction operation is performed to correct the reflection error caused by the solar elevation angle and the sensor viewing angle; S1305, the surface roughness correction is performed on the spectrum data after anisotropic scattering correction to eliminate the influence of the micro-morphological difference of the ground surface on the spectrum, so as to obtain the above soil intrinsic reflectivity, the above vegetation coverage component and the above salt crust component.

[0037] For example, the embodiment describes in detail a soil water and salt dynamic monitoring and quality evaluation method based on multi-spectral remote sensing. By combining multi-temporal multi-spectral data with ground sample data, using unmixing technology, physical model and data-driven analysis method, the method can accurately evaluate the soil water and salt condition and monitor the soil quality change.

[0038] Using the assimilated multi-temporal multi-spectral data and the ground sample data, a soil, vegetation and salt crust three-domain spectral component unmixing model is constructed. This process accurately extracts the contribution proportion of soil, vegetation and salt crust components through multi-step unmixing calculation.

[0039] Based on the assimilated multi-temporal multi-spectral data, the spectral reflectivity information of each observed pixel is extracted. The reflectivity of each pixel is obtained from the remote sensing image and can reflect the spectral characteristics of the position. By combining the typical soil, vegetation and salt crust sample spectrum in the ground sample data, the standard spectral characteristics of the three types of end members (soil, vegetation and salt crust) can be determined.

[0040]

[0041] Among them: is the observed spectral reflectance, representing the spectral characteristics of each pixel in the remote sensing image; is the standard spectral basis vector of each endmember, obtained from ground sample data; is the spectral component weight of each endmember, representing the contribution of different components in the observed spectrum; is the error term, representing noise and other factors not considered.

[0042] The spectral characteristics of soil, vegetation, and salt crust are combined into the total observed spectrum through weighted summation. The weight of each endmember reflects the contribution of the endmember to each pixel spectrum. The error term is used to describe the noise and errors not considered.

[0043] Based on the extracted endmember standard spectrum, a mixed spectral expression is established to represent the composition relationship of each pixel spectrum:

[0044] where: is the observed spectrum; is the standard spectrum of each endmember; is the spectral component proportion of the endmember. This formula indicates that the standard spectrum of each endmember (such as soil, vegetation, and salt crust) is weighted proportionally to obtain the observed spectrum. The weight of each endmember indicates the contribution proportion of the endmember in the overall spectrum.

[0045] By decomposing the above mixed spectral expression, the component weight of soil, vegetation, and salt crust in each pixel is solved, and the contribution proportion of each endmember in each pixel is obtained:

[0046] where: is the spectral component proportion of each endmember; is the observed spectrum; is the standard spectrum of each endmember. This formula indicates that by minimizing the difference between the observed spectrum and the predicted spectrum, the component weight of each endmember in each pixel is inverted . Through least squares method or other optimization algorithm, the contribution proportion of each endmember to the observed spectrum is calculated, and the preliminary spectral component proportion is obtained.

[0047] The observed spectrum of remote sensing data will be affected by different incident angles and observation angles. In order to eliminate this effect, we perform anisotropic scattering correction operation in this step to correct the reflection error caused by the solar elevation angle and the sensor viewing angle:

[0048]

[0049] wherein: is the corrected spectral reflectance; is the observed spectrum; is the anisotropic response function, which depends on the solar incidence angle and the sensor viewing angle .

[0050] The observed spectrum is corrected by the anisotropic response function to eliminate the reflection error caused by the difference in solar elevation angle and sensor viewing angle. This ensures the comparability of spectral data at different observation angles, making the spectral information of remote sensing data more accurate.

[0051] The differences in surface micro-morphology (such as slope, cracks, etc.) will affect the spectrum, and roughness correction needs to be performed on the spectral data to eliminate the effects of these differences:

[0052] wherein: is the final corrected spectrum; is the spectrum after scattering correction; is the roughness correction coefficient; roughness is the surface roughness. This formula indicates that the spectral data is corrected by the roughness correction coefficient to compensate for the spectral differences caused by surface micro-morphology (such as slope, cracks, etc.). The roughness parameter (roughness) is calculated by a digital elevation model, reflecting the morphological characteristics of the surface.

[0053] This embodiment combines multi-temporal and multi-spectral remote sensing data with ground sample data, uses advanced spectral unmixing models, scattering correction, and roughness correction techniques to accurately extract soil, vegetation, and salt crust spectral components. Each step processes the observed spectrum through mathematical models and optimization algorithms to ensure the accuracy of spectral component decomposition and provide high-precision data support for subsequent water-salt inversion and soil quality assessment.

[0054] In one possible implementation, as shown in Figure 3 the above step S1303 is performed by decomposing and solving the above mixed spectral expression to obtain the component weight of the above soil endmember, the above vegetation endmember, and the above salt crust endmember in each pixel, thereby obtaining the preliminary spectral component ratio, including: S13031, convert the above mixed spectral expression into a matrix equation in matrix form, wherein the matrix equation takes the above observed spectrum as the observation vector and the matrix composed of each endmember spectral basis vector as the coefficient matrix; S13032, solve the matrix equation using the least squares method, and obtain the component weight initial value of each end member; S13033, non-negative constraint and normalization processing on the component weight initial value, so that the sum of the component weights of the soil end member, the vegetation end member and the salt crust end member is equal to one; S13034, output the component weight result after constraint optimization as the preliminary spectral component ratio.

[0055] Exemplarily, the embodiment realizes the extraction of the spectral components of the soil, vegetation and salt crust three types of end members by using the unmixing technology through the combination of multi-temporal multi-spectral data and ground sample data, and then performs the monitoring and quality assessment of the soil water and salt state.

[0056] By decomposing the mixed spectrum expression, the component weight of each end member in each pixel is obtained, and thus the preliminary spectral component ratio is obtained. This is one of the core steps of the remote sensing data unmixing. The mixed spectrum expression is converted into a matrix equation in the form of a matrix. The goal of this equation is to establish a mathematical relationship between the observed spectrum and the end member spectrum through the matrix form, which facilitates subsequent calculations:

[0057]

[0058] Wherein: is the observed spectrum reflectance, which represents the spectral information of each pixel obtained through the remote sensing image; A is a matrix composed of end member spectral basis vectors (such as soil, vegetation, salt crust), and the list of the matrix represents the spectrum of different end members; is the component weight vector of the end member, which represents the proportion of each end member in the observed spectrum; is the error term, which considers measurement errors, noise or factors not considered. is the observed spectrum reflectance, which is obtained from the remote sensing image and represents the reflectance of the pixel at different wavebands. is the end member spectrum matrix, which contains the standard spectrum of different end members (such as soil, vegetation, salt crust) at each waveband. is the weight coefficient of each end member, which represents the contribution proportion of the component to the overall observed spectrum. is the error term, which is used to represent the part that cannot be explained by the known end members, such as noise, etc. This equation converts the mixed spectrum model into a linear equation, which represents the relationship between the observed spectrum and the end member spectrum through the matrix. The goal is to solve the weight of each end member in the observed spectrum.

[0059] The least squares method is used to solve the above matrix equation, and the component weight initial value of each end member is obtained. The least squares method minimizes the error between the observed spectrum and the calculated spectrum to obtain the optimal component weight:

[0060]

[0061] wherein: is the endmember component weight obtained by least square method; is the observed spectrum; is the endmember spectrum matrix, is the coefficient matrix containing each endmember spectrum, representing the contribution of endmember spectrum to the observed spectrum. is the component weight vector of endmember, representing the proportion of each endmember. The purpose of least square method is to minimize the difference between observed spectrum and the spectrum calculated by known endmember spectrum and component weight. In this way, we can deduce the weight of each endmember in the pixel, that is, its contribution in the overall spectrum.

[0062] Since the endmember component weight must be non-negative and its sum equal to 1 (i.e. the spectral composition of each pixel is positive, and the sum is , the initial value is subjected to non-negative constraint and normalization processing to ensure physical rationality:

[0063] wherein: represents the spectral composition proportion of each endmember. The constraint ensures that the proportion of each endmember is non-negative, which is in line with the physical meaning. The normalization constraint ensures that the sum of the proportions of all endmembers is 1. This step ensures that the contribution of each endmember is reasonable by imposing non-negative constraint and normalization condition.

[0064] The endmember component weight after non-negative constraint and normalization optimization is output as the preliminary spectral composition proportion:

[0065] wherein: is the optimized endmember component weight, representing the contribution proportion of each endmember. is the optimized endmember weight, which meets the non-negative and normalization constraints. This step ensures that the output endmember weight meets the physical conditions and serves as the input for subsequent analysis. These results can be used for subsequent water-salt inversion, quality assessment and governance priority ranking.

[0066] In this embodiment, the spectral composition proportions of soil, vegetation and salt crust are accurately extracted from multi-temporal and multi-spectral data, and the scientificity and physical rationality of spectral unmixing are ensured through least square method, non-negative constraint and normalization processing.

[0067] In a feasible implementation manner, as Figure 4As shown, the step S1304 performs anisotropic scattering correction operation on the reflection difference of the observation spectrum at different incident angles and observation angles to correct the reflection error caused by the solar elevation angle and the sensor viewing angle, including: S13041, based on the imaging geometric information of the multi-temporal multi-spectral data, obtaining the geometric parameters at the corresponding time, wherein the geometric parameters include the solar incident angle, the azimuth angle and the sensor observation angle parameters; S13042, calculating the relative observation geometric relationship of each pixel according to the geometric parameters, establishing a reflection anisotropic response function; using the anisotropic response function to perform angle normalization processing on the original reflectivity of the observation spectrum to obtain the corrected reflectivity under the standard incident angle and observation angle conditions; S13043, outputting the corrected reflectivity as the input data for subsequent surface roughness correction to correct the reflection error caused by the solar elevation angle and the sensor viewing angle.

[0068] For example, in remote sensing data processing, the reflectivity of the observation spectrum is affected by the incident angle and the observation angle. Therefore, in this step, anisotropic scattering correction operation is performed to correct the reflection error caused by the solar elevation angle and the sensor viewing angle. Specifically, the following steps perform angle normalization correction on the spectral data by mathematical method to ensure the comparability of the data under different viewing angles.

[0069] First, through the imaging geometric information of the multi-temporal multi-spectral data, the geometric parameters at the corresponding time are obtained. These geometric parameters include the solar incident angle, the azimuth angle and the sensor observation angle, which have a direct impact on the reflected spectrum wherein: is the solar incident angle, indicating the angle of solar radiation to the ground surface; is the solar azimuth angle, indicating the angle between the sun's rays and the horizontal plane of the ground; is the sensor observation angle, indicating the angle at which the remote sensing sensor receives the reflected light. These geometric parameters reflect the angle relationship between solar radiation and sensor receiving reflected light, and are the basis for performing anisotropic scattering correction operation. By obtaining these geometric parameters.

[0070] Based on the geometric parameters, we calculate the relative observation geometric relationship of each pixel, and further establish a reflection anisotropic response function. This response function describes how the spectral reflectivity changes with the change of angle. Using this response function to perform angle normalization processing on the original reflected spectrum, the corrected reflectivity under the standard incident angle and observation angle conditions is obtained:

[0071]

[0072] wherein: the corrected spectral reflectance; the observed spectral reflectance; the anisotropic response function, which describes how the spectral reflectance changes with the change of angle.

[0073] The corrected spectral data after anisotropic scattering correction is output and used as input data for subsequent surface roughness correction to further eliminate the influence of surface micro-morphology on the spectrum:

[0074] wherein, the final corrected spectral reflectance; the spectral reflectance after scattering correction; the roughness correction coefficient; roughness is the surface roughness parameter. The spectral data after scattering correction is further corrected for roughness to compensate for the influence of surface micro-morphology (such as slope, crack, etc.) on spectral reflectance. The roughness parameter is calculated based on the ground sample data and digital elevation model (DEM), which can effectively eliminate the error caused by surface micro-morphology.

[0075] The present embodiment corrects the reflection error caused by different incident angles and observation angles, and combines the surface roughness information to obtain more standardized and accurate spectral data. These corrected spectral data provide accurate basic data for dynamic monitoring and quality assessment of soil water and salt.

[0076] In a feasible implementation manner, as shown in Figure 5 The above step S1305 corrects the spectral data after anisotropic scattering correction for surface roughness to eliminate the influence of surface micro-morphology on the spectrum, and obtains the above soil intrinsic reflectance, the above vegetation cover component, and the above salt crust component, including: S13051, based on the ground sample data and digital elevation model data, extracting regional surface roughness feature parameters, wherein the surface roughness parameters include slope, slope direction, and surface microstructure scale; S13052, constructing a roughness influence function and a roughness correction coefficient based on the surface roughness feature parameters; S13053, combining the spectral data after the above anisotropic scattering correction and the above roughness influence function to correct the spectral reflectance for roughness and compensate for the spectral scattering distortion caused by surface micro-morphology; S13054, outputting the spectral data after roughness correction to obtain the above soil intrinsic reflectance, the above vegetation cover component, and the above salt crust component.

[0077] Exemplarily, the embodiment details the dynamic monitoring of soil, vegetation, and salt crust by combining multi-temporal and multi-spectral data with ground sample data, using advanced physical and mathematical methods to process the spectral data. The spectral data that has been corrected for anisotropic scattering is corrected for surface roughness. The roughness correction aims to eliminate the effects of surface micro-topography differences (such as slope, cracks, etc.) on the spectral data, thereby improving the accuracy of the spectral data.

[0078] Based on the ground sample data and digital elevation model (DEM) data, we extract the surface roughness characteristic parameters of each region. The surface roughness parameters usually include slope, aspect, and surface micro structure scale. Through the calculation of these parameters, we can accurately describe the micro-topographic characteristics of the surface:

[0079] roughness Slope, Aspect, Surface micro structure scale

[0080] wherein roughness is the surface roughness parameter, representing the micro-topographic characteristics of the surface, such as slope, aspect, and surface micro structure scale; Slope is the slope, representing the inclination angle of the surface; Aspect is the aspect, representing the angle of the slope facing; Surface micro structure scale is the surface micro structure scale, representing the details of the surface texture and morphology. Slope and Aspect are extracted from the ground data through digital elevation model (DEM). Surface micro structure scale is obtained through ground sample data, usually by analyzing the high-frequency changes related to surface roughness.

[0081] Based on the above extracted surface roughness characteristic parameters, the roughness influence function (RIF) and the roughness correction coefficient (RCC) are constructed. These functions and coefficients will be used to describe and compensate for the effects of surface micro-topographic differences on the spectrum:

[0082]

[0083] wherein: RIF (roughness) is the roughness influence function, representing the influence of surface roughness on spectral reflection; RCC is the roughness correction coefficient, used to adjust the influence of roughness on the spectrum; roughness is the surface roughness parameter, representing the micro-topographic characteristics of the surface. Through the roughness correction coefficient and the roughness influence function, the spectral data after scattering correction can be further corrected.

[0084] ​The roughness-corrected spectral data is outputted, and the soil intrinsic reflectance, vegetation cover component and salt crust component are obtained. The corrected spectral data will be used as input for subsequent soil water-salt inversion and soil quality assessment:

[0085]

[0086] The final spectral data is the data after anisotropic scattering correction and roughness correction. After eliminating the effects of solar angle, sensor view angle and surface micro-morphology, these data can provide more accurate information of soil, vegetation and salt crust components.

[0087] Through the above steps, in the soil water-salt dynamic monitoring and quality assessment method based on multi-spectral remote sensing, the ground sample data and digital elevation model are used for roughness correction. By accurately adjusting the effect of surface roughness on the spectrum, it is ensured that the final spectral data can accurately reflect the true situation of soil, vegetation and salt crust on the ground.

[0088] In a feasible implementation manner, as shown in Figure 6 S140, based on the soil intrinsic reflectance, the vegetation cover component and the salt crust component, and combined with the thermal infrared and red edge band information in the satellite remote sensing data, a water-salt inversion model is established to obtain the multi-temporal soil water-salt inversion result, including: S1401, extracting spectral feature parameters in near-infrared and short-wave infrared bands based on the soil intrinsic reflectance; S1402, jointly analyzing the spectral feature parameters and the thermal infrared band information in the satellite remote sensing data to calculate the surface temperature feature quantity; S1403, extracting a salt-sensitive feature index using the red edge band information, and introducing the salt crust component as a constraint factor to correct the spectral aliasing error in salt inversion; S1404, inputting the spectral feature parameters, the surface temperature feature quantity and the salt-sensitive feature index into a multi-index collaborative inversion model to calculate the soil volume water content and the surface salt index through statistical regression to generate a preliminary inversion result; S1405, performing time series consistency correction on the preliminary inversion result to maintain the physical continuity of the water-salt change trend between adjacent time phases, thereby obtaining the multi-temporal soil water-salt inversion result.

[0089] For example, the soil intrinsic reflectance, the vegetation cover component and the salt crust component are used to establish a water-salt inversion model in combination with the thermal infrared and red edge band information in the satellite remote sensing data to obtain the multi-temporal inversion result of soil water-salt. This process includes multiple sub-steps, which describe in detail how to accurately obtain the soil water-salt state through the combination of spectral features and models.

[0090] Spectral feature parameters in the near-infrared (NIR) and shortwave infrared (SWIR) bands are extracted from the intrinsic reflectance of the soil. These bands play an important role in soil water-salt inversion because they are sensitive to the water content and salt content of the soil:

[0091]

[0092] where: is the reflectance in the near-infrared band; is the reflectance in the shortwave infrared band; is the solar incidence angle; is the sensor observation angle; are the wavelengths in the near-infrared and shortwave infrared bands, respectively.

[0093] The extracted spectral feature parameters (such as near-infrared and shortwave infrared reflectance) are combined with thermal infrared band information to calculate the surface temperature characteristic quantity. Surface temperature has a significant impact on changes in soil water and salt content, and can effectively assist in water-salt inversion.

[0094]

[0095] where: is the surface temperature; is the reflectance in the thermal infrared band. Surface temperature is one of the key parameters for water-salt inversion. By combining information from the near-infrared, shortwave infrared, and thermal infrared bands, the inversion accuracy can be improved. The thermal infrared band is particularly suitable for monitoring surface temperature and can be combined with other band data for joint analysis, providing more accurate soil moisture information.

[0096] The salt-sensitive feature index (SSI) is extracted through the red edge band information. The red edge band is very sensitive to changes in salt content, especially in areas with soil salt crusts, so this band plays an important role in salt inversion.

[0097]

[0098] where: SSI is the salt-sensitive feature index; is the reflectance in the red edge band; is the reflectance in the shortwave infrared band. There is a significant relationship between the reflectance in the red edge band and the salt content of the soil. By calculating the salt-sensitive feature index (SSI), the salt content can be better identified, especially in areas where salt crusts have formed.

[0099] The extracted spectral feature parameters, the surface temperature feature quantity, and the salt sensitive feature index are input into a multi-index collaborative inversion model, and soil volume water content (VWC) and surface salt index (ECe) are calculated through statistical regression to generate preliminary inversion results.

[0100]

[0101] wherein: is the soil volume water content; is the surface salt index; is the regression coefficient of the soil water inversion model; is the regression coefficient of the salt inversion model; is an error term.

[0102] Through the establishment of a multi-index collaborative inversion model, combined with multiple spectral features and surface temperature information, statistical regression analysis is used to invert the soil volume water content and the surface salt index.

[0103] After obtaining the preliminary inversion results, time series consistency correction is performed on these results to ensure that the water and salt change trends between adjacent time phases remain physically continuous.

[0104]

[0105] wherein: is the corrected soil volume water content change; is the original soil volume water content change; is a consistency correction factor. Time series consistency correction can ensure the consistency of soil water and salt inversion results between different time points, thereby ensuring the physical continuity of the inversion results and avoiding data jumps or unreasonable changes.

[0106] Through the above steps, a water and salt inversion model is established based on multi-spectral remote sensing data and multiple spectral feature parameters (such as near-infrared, short-wave infrared, red edge band, etc.). In this process, sensitive features of soil water and salt are extracted, and through multi-index collaborative inversion and time series consistency correction, the accuracy and physical reasonableness of the water and salt inversion results are ensured.

[0107] In a feasible implementation manner, as shown in FIG. 1 1, Figure 7 the step S150 inputs the above multi-time phase water and salt inversion results as observations, combines the time series changes of the vegetation coverage component and the assimilated registered meteorological driving data, constructs a water and salt coupled partial differential equation and graph space regularization constraint model, and obtains continuous spatio-temporal distribution fields of soil water content and salt content through joint optimization of the physical domain and the data domain, including: S1501, combine the multi-temporal water-salt inversion results with the temporal changes of vegetation cover components and the assimilated meteorological driving data, take the precipitation as the infiltration and recharge boundary condition, and take the evapotranspiration as the surface water loss term to construct a water-salt coupling partial differential equation and graph spatial-temporal regularization constraint model; S1502, obtain the continuous spatial-temporal distribution field of soil water content and salt under the driving of meteorological conditions through joint optimization of the physical domain and the data domain.

[0108] For example, a water-salt coupling partial differential equation (PDE) and graph spatial-temporal regularization constraint model is constructed by combining the multi-temporal water-salt inversion results with the temporal changes of vegetation cover components and the meteorological driving data. Through joint optimization of the physical model and the data domain, the continuous spatial-temporal distribution field of soil water content and salt is finally obtained.

[0109] A water-salt coupling partial differential equation (PDE) and graph spatial-temporal regularization constraint model is constructed, and multiple input data (such as multi-temporal water-salt inversion results, temporal changes of vegetation cover components, meteorological driving data, etc.) are combined into the model. The model needs to consider meteorological conditions such as precipitation and evapotranspiration.

[0110]

[0111] Wherein: VWC is the soil water content, representing the water content in unit volume of soil. ECe is the salt concentration in soil. t is the time variable. D is the water diffusion coefficient, describing the diffusion process of water in soil. D is the salt diffusion coefficient, describing the diffusion process of salt in soil. Laplace operator represents the spatial variation of water and salt. f is the effective rate of soil water evaporation, which depends on soil water content and surface temperature. f is the effective migration rate of salt, which depends on soil water content and salt concentration. P is the precipitation in meteorological driving factors, representing the infiltration and recharge boundary condition. E is the evapotranspiration in meteorological driving factors, representing the surface water loss term. These two equations represent the spatial and temporal evolution process of water and salt in soil. The first equation describes the change process of soil water content, including diffusion, evaporation and precipitation recharge; the second equation describes the diffusion and change of salt in soil, considering the coupling effect of water and salt. By solving these two equations, the spatial-temporal distribution of soil water content and salt can be obtained.

[0112] By joint optimization of physical domain and data domain, the observation results obtained from remote sensing data and ground data are combined with the theoretical model to further optimize the spatial distribution of soil water content and salt. The goal of joint optimization is to minimize the error between the model prediction results and the observation data, ensuring the consistency of the physical model and the actual data.

[0113]

[0114] wherein: L is the loss function, representing the optimization goal of the model. is the predicted soil moisture at the th spatiotemporal position. is the observed soil moisture at the th spatiotemporal position. is the predicted salt at the th spatiotemporal position. is the observed salt at the th spatiotemporal position. is the regularization coefficient, controlling the smoothness of the model. is the smoothing term of the prediction results, ensuring the physical continuity of the results. By minimizing the loss function L, joint optimization between the physical model and the actual observation data is achieved, ensuring that the obtained spatiotemporal distribution fields of soil moisture and salt not only conform to the physical laws, but also are consistent with the actual observation data as much as possible.

[0115] In this embodiment, remote sensing data, ground sample data and meteorological driving data are combined to establish a water-salt coupling partial differential equation and graph spatiotemporal regularization constraint model. Through this method, the dynamic changes of soil moisture and salt can be accurately simulated, and the spatiotemporal distribution thereof can be obtained, and the joint optimization method is used to ensure the physical reasonableness and data consistency of the results.

[0116] In a feasible implementation manner, as shown in Figure 8 , the step S170 constructs a soil quality comprehensive evaluation model based on the salinity characteristic index, the water content characteristic index, the vegetation response index and the soil health index, quantitatively scores and classifies the soil quality of different regions, and outputs high-risk patches, hot migration paths and treatment priority orders, including: S1701, correlating the salinity characteristic index and the water content characteristic index with the vegetation response index obtained from the time series change of the vegetation coverage component; S1702, combining the soil health index, normalizing the salinity characteristic index, the water content characteristic index and the vegetation response index, and establishing a soil quality evaluation feature matrix; S1703、Based on the above soil quality evaluation characteristic matrix, a weighted comprehensive evaluation algorithm is used to construct the above soil quality comprehensive evaluation model to calculate the soil quality comprehensive score of each region; S1704、According to the above soil quality comprehensive score, grade division and spatial clustering analysis are carried out to output the above high-risk patches, identify the above hot migration path, and generate the above management priority order.

[0117] In a feasible implementation, the determination of the high-risk patch comprises: Based on the soil quality comprehensive score, the score distribution characteristics of each monitoring region are counted, and the statistical results of the score are calculated, wherein the statistical results include the mean, standard deviation and extreme value range; According to the statistical results, set the grading threshold interval, divide the soil quality comprehensive score into multiple grades, and identify the region in the low score interval as a potential high-risk area; Each grade region is spatially encoded and vectorized to generate a soil quality grade distribution layer; The soil quality grade distribution layer is input into the spatial clustering analysis module to identify adjacent high-risk pixel groups using a density-based clustering algorithm to form the above risk patches.

[0118] In a feasible implementation, the determination of the management priority order comprises: According to the change trend of the high-risk patch in the time series, the hot migration path is extracted; Combined with the spatial clustering intensity of the high-risk patch, the migration rate of the hot migration path and the soil health score, the comprehensive risk weight of each region is calculated, and the management priority order is determined according to the comprehensive risk weight.

[0119] For example, according to the salinity characteristic index, the water characteristic index, the vegetation response index and the soil health index, a soil quality comprehensive evaluation model is constructed to score and grade the soil quality of different regions, and output the high-risk patch, the hot migration path and the management priority order. The following is a detailed description of each sub-step.

[0120] The salinity characteristic index, the water characteristic index and the vegetation response index (obtained from the time series change of vegetation coverage component) are analyzed. This analysis helps us understand the correlation between these indicators, which provides a basis for subsequent soil quality evaluation.

[0121]

[0122] Wherein: is the correlation between the salinity characteristic index and the vegetation response index; is the salinity characteristic index; Vegetation response indicator. By calculating the correlation between salinity characteristics and vegetation response, we can understand how vegetation status affects the distribution of salt, and thus provide a more accurate model for soil quality assessment.

[0123] In combination with soil health indicators, normalize the salinity characteristics, water characteristics, and vegetation response indicators. The normalized indicator data will form the soil quality evaluation feature matrix, which will be used for subsequent comprehensive evaluation.

[0124]

[0125] Where: is the normalized feature value; is the original feature value (salinity, water, vegetation response, etc.); is the mean of the feature value; is the standard deviation of the feature value. represents different feature values, which can be salinity, water, vegetation response, or other related indicators. and are the mean and standard deviation of the feature value, respectively, used for standardizing data to ensure that each feature has equal influence in the evaluation model. Normalization helps to eliminate the dimensional differences of different feature values, making each indicator consistent in the comprehensive evaluation, and providing standardized data input for subsequent weighted comprehensive evaluation.

[0126] Based on the normalized soil quality evaluation feature matrix, use weighted comprehensive evaluation algorithm or principal component analysis (PCA) algorithm to construct the soil quality comprehensive evaluation model, and calculate the soil quality comprehensive score of each region.

[0127]

[0128] Where: is the soil quality comprehensive score; is the normalized feature value (salinity characteristics, water characteristics, vegetation response, etc.); is the weight of the corresponding feature; is the number of features. Through weighted comprehensive evaluation or principal component analysis method, we will weight and sum all the normalized features (such as salinity, water, vegetation response) to calculate the soil quality comprehensive score of each region.

[0129] Grade division and spatial clustering analysis, output high-risk patches and hot migration paths Based on the soil quality comprehensive score, we perform grade division and spatial clustering analysis to identify high-risk patches and hot migration paths, and generate the priority order for governance.

[0130]

[0131] where: is the soil quality rating; is the soil quality composite score; cluster() is a spatial clustering algorithm based on density or other clustering methods for hierarchical and hotspot identification. By performing spatial clustering analysis on the soil quality composite score, we divide the region into different risk levels. Low-score areas are identified as potential high-risk zones, and high-risk patches and hotspot migration paths are identified based on soil health conditions.

[0132] The determination of high-risk patches involves statistical analysis of the distribution characteristics of the soil quality composite score and calculation of the mean, standard deviation, and extreme value range.

[0133]

[0134] where: is the mean of the soil quality composite score; is the standard deviation of the soil quality composite score; is the number of regions. By calculating the mean and standard deviation of the score, I can understand the overall distribution of the soil quality score.

[0135] By using a spatial clustering algorithm (such as a density-based clustering method), adjacent high-risk pixel groups are identified, forming high-risk patches.

[0136]

[0137] where: cluster is the th high-risk patch; is the soil quality score of each region. Through clustering analysis, multiple high-risk patches can be identified, and these areas need to be prioritized for treatment.

[0138] Based on the temporal changes of high-risk patches, hotspot migration paths are extracted, and a treatment priority order is generated.

[0139]

[0140] where: path is the hotspot migration path; change in risk is the change in high-risk areas in the time series. By analyzing the change trend of high-risk areas in the time series, hotspot migration paths can be identified.

[0141] Through the above steps, a complete comprehensive soil quality assessment model was constructed, incorporating considerations of salinity, moisture, vegetation response, and soil health. A weighted comprehensive evaluation algorithm was used to calculate the soil quality score, and spatial clustering and temporal series analysis were employed to identify high-risk patches and hotspot migration pathways. This model not only enables quantitative scoring and classification of soil quality but also identifies areas requiring priority remediation, providing a scientific basis for soil management.

[0142] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A soil water and salt dynamic monitoring and quality evaluation method based on multispectral remote sensing, characterized in that, The method comprises the following steps: acquiring multi-source multi-temporal multi-spectral data, ground sample data and meteorological driving data; wherein the multi-source multi-temporal multi-spectral data comprises satellite remote sensing images and unmanned aerial vehicle multi-spectral camera images, the ground sample data comprises volumetric water content, surface and profile electrical conductivity and electrical conductivity ratio, soil texture and bulk density, and the meteorological driving data comprises precipitation and evapotranspiration; assimilating and registering the multi-source multi-temporal multi-spectral data, the ground sample data and the meteorological driving data in the time and space scales; based on the assimilated multi-temporal multi-spectral data and the ground sample data, constructing a soil, vegetation and salt crust three-domain spectral component unmixing model, performing anisotropic scattering correction and roughness correction on the observed spectrum in the multi-temporal multi-spectral data to obtain soil intrinsic reflectance, vegetation cover component and salt crust component; based on the soil intrinsic reflectance, the vegetation cover component and the salt crust component, and combining the thermal infrared and red edge band information in the satellite remote sensing data, establishing a water-salt inversion model to obtain multi-temporal soil water-salt inversion results; taking the multi-temporal water-salt inversion results as observation input, combining the time series variation of the vegetation cover component and the assimilated meteorological driving data, constructing a water-salt coupled partial differential equation and graph time-space regularization constraint model, and obtaining continuous time-space distribution fields of soil water content and salt content through joint optimization of the physical domain and the data domain; based on the continuous time-space distribution fields, extracting salt concentration characteristic indexes and water content characteristic indexes, extracting vegetation response indexes based on the time series variation of the vegetation cover component, and determining soil health indexes based on the soil intrinsic reflectance; based on the salt concentration characteristic indexes, the water content characteristic indexes, the vegetation response indexes and the soil health indexes, constructing a soil quality comprehensive evaluation model to quantitatively score and classify the soil quality in different regions, and outputting high-risk patches, hot migration paths and treatment priorities.

2. The soil water and salt dynamic monitoring and quality evaluation method based on multispectral remote sensing according to claim 1, characterized in that, The method based on the assimilated multi-temporal multi-spectral data and the ground sample data, constructing a soil, vegetation and salt crust three-domain spectral component unmixing model, performing anisotropic scattering correction and roughness correction on the observed spectrum in the multi-temporal multi-spectral data to obtain soil intrinsic reflectance, vegetation cover component and salt crust component, comprises the following steps: extracting observed spectral reflectance information of each observation pixel based on the assimilated multi-temporal multi-spectral data, and combining the typical soil, vegetation and salt crust sample spectra obtained from the ground sample data to determine the standard spectral characteristics of the three types of end members; taking the standard spectral characteristics as input, establishing a mixed spectral expression composed of soil end members, vegetation end members and salt crust end members to represent the composition relationship of each pixel spectrum; by decomposing and solving the mixed spectral expression, the component weights of the soil end members, the vegetation end members and the salt crust end members in each pixel are obtained, thereby obtaining the preliminary spectral component proportions; performing anisotropic scattering correction operation on the reflection difference of the observed spectrum under different incident angles and observation angles to correct the reflection error caused by the solar elevation angle and the sensor viewing angle; The spectrum data after anisotropic scattering correction is subjected to surface roughness correction to eliminate the influence of surface micro-morphology difference on the spectrum, so as to obtain the soil intrinsic reflectivity, the vegetation coverage component and the salt crust component.

3. The soil water and salt dynamic monitoring and quality evaluation method based on multispectral remote sensing according to claim 2, characterized in that, The mixed spectrum expression is transformed into a matrix equation in a matrix form, wherein the matrix equation takes the observed spectrum as an observation vector and takes a matrix composed of each end-member spectrum basis vector as a coefficient matrix. A least square method is used to solve the matrix equation to obtain the component weight initial value of each end-member. The component weight initial value is subjected to non-negative constraint and normalization processing, so that the sum of the component weights of the soil end-member, the vegetation end-member and the salt crust end-member is equal to one. The component weight result after constraint optimization is output as a preliminary spectrum component ratio. The difference in reflection of the observed spectrum under different incident angles and observation angles is subjected to anisotropic scattering correction operation to correct the reflection error caused by the solar elevation angle and the sensor viewing angle.

4. The soil water and salt dynamic monitoring and quality evaluation method based on multispectral remote sensing according to claim 2, characterized in that, Based on the imaging geometry information of the multi-temporal multi-spectral data, the geometric parameters corresponding to the time are obtained, wherein the geometric parameters include the solar incident angle, the azimuth angle and the sensor observation angle parameters. The relative observation geometry relationship of each pixel is calculated according to the geometric parameters, and a reflection anisotropy response function is established. The original reflectivity of the observed spectrum is subjected to angle normalization processing by using the anisotropy response function to obtain the corrected reflectivity under the standard incident angle and observation angle conditions. The corrected reflectivity is output as the input data for subsequent surface roughness correction to correct the reflection error caused by the solar elevation angle and the sensor viewing angle.

5. The soil water and salt dynamic monitoring and quality evaluation method based on multispectral remote sensing according to claim 2, characterized in that, The spectrum data after anisotropic scattering correction is subjected to surface roughness correction to eliminate the influence of surface micro-morphology difference on the spectrum, so as to obtain the soil intrinsic reflectivity, the vegetation coverage component and the salt crust component. Based on the ground sample data and the digital elevation model data, the regional surface roughness characteristic parameters are extracted, wherein the surface roughness parameters include the slope, the slope direction and the surface microstructure scale. Based on the surface roughness characteristic parameters, the roughness influence function and the roughness correction coefficient are constructed. The spectrum data after the anisotropic scattering correction and the roughness influence function are combined to correct the spectrum reflectivity and compensate the spectrum scattering distortion caused by the surface micro-morphology. The spectrum data after the roughness correction is output to obtain the soil intrinsic reflectivity, the vegetation coverage component and the salt crust component.

6. The soil water and salt dynamic monitoring and quality evaluation method based on multispectral remote sensing according to claim 1, characterized in that, Based on the soil intrinsic reflectivity, the vegetation coverage component and the salt crust component, and combined with the thermal infrared and red edge band information in the satellite remote sensing data, a water-salt inversion model is established to obtain multi-temporal soil water-salt inversion results, including: Based on the soil intrinsic reflectivity, the spectrum characteristic parameters in the near-infrared and short-wave infrared bands are extracted. Jointly analyze the spectral feature parameters and the thermal infrared band information in the satellite remote sensing data to calculate a ground surface temperature characteristic quantity; Extract a salt sensitivity characteristic index using the red edge band information, and introduce the salt crust component as a constraint factor to correct spectral aliasing errors in salt inversion; Input the spectral feature parameters, the ground surface temperature characteristic quantity and the salt sensitivity characteristic index into a multi-index collaborative inversion model to calculate soil volume water content and surface salt content by statistical regression, and generate a preliminary inversion result; Perform time series consistency correction on the preliminary inversion result to maintain physical continuity of water and salt change trends between adjacent time phases, thereby obtaining the multi-time phase soil water and salt inversion result.

7. The soil water and salt dynamic monitoring and quality evaluation method based on multispectral remote sensing according to claim 1, characterized in that, The multi-time phase water and salt inversion result is taken as observation input, combined with time series changes of the vegetation coverage component and assimilated and registered meteorological driving data to construct a water and salt coupled partial differential equation and graph spatiotemporal regularization constraint model, and a continuous spatiotemporal distribution field of soil water content and salt content is obtained through joint optimization of a physical domain and a data domain, including: The multi-time phase water and salt inversion result is taken as observation input, combined with time series changes of the vegetation coverage component and assimilated and registered meteorological driving data, the precipitation is taken as an infiltration and replenishment boundary condition, and the evapotranspiration is taken as a surface water loss term to construct a water and salt coupled partial differential equation and graph spatiotemporal regularization constraint model; A continuous spatiotemporal distribution field of soil water content and salt content under meteorological condition driving is obtained through joint optimization of a physical domain and a data domain.

8. The soil water and salt dynamic monitoring and quality evaluation method based on multispectral remote sensing according to claim 1, characterized in that, Based on the salinity characteristic index, the water content characteristic index, the vegetation response index and the soil health index, a soil quality comprehensive evaluation model is constructed to quantitatively score and classify soil quality in different regions, and a high-risk patch, a hot spot migration path and a treatment priority order are output, including: Correlation analysis is performed on the salinity characteristic index and the water content characteristic index and the vegetation response index obtained from time series changes of the vegetation coverage component The salinity characteristic index, the water content characteristic index and the vegetation response index are normalized in combination with the soil health index, and a soil quality evaluation feature matrix is established; Based on the soil quality evaluation feature matrix, a weighted comprehensive evaluation algorithm is used to construct the soil quality comprehensive evaluation model to calculate soil quality comprehensive scores of regions; According to the soil quality comprehensive scores, grade division and spatial cluster analysis are performed to output the high-risk patch, identify the hot spot migration path and generate the treatment priority order.

9. The soil water and salt dynamic monitoring and quality evaluation method based on multispectral remote sensing according to claim 1, characterized in that, The determination of the high-risk patch includes: Based on the soil quality comprehensive scores, score distribution characteristics of each monitoring region are counted, and statistical results of the scores are calculated, wherein the statistical results include mean value, standard deviation and extreme value range; Grade threshold intervals are set according to the statistical results, the soil quality comprehensive scores are divided into multiple grades, and regions in a low score interval are identified as potential high-risk areas; Each grade region is spatially encoded and vectorized to generate a soil quality grade distribution layer; The soil quality grade distribution layer is input into a spatial clustering analysis module, and a density-based clustering algorithm is used to identify adjacent high-risk pixel groups to form the risk patches.

10. The soil water and salt dynamic monitoring and quality evaluation method based on multispectral remote sensing according to claim 9, characterized in that, The determining of the management priority order comprises: According to the change trend of the high-risk patches in the time sequence, the hotspot migration path is extracted; In combination with the spatial clustering intensity of the high-risk patches, the migration rate of the hotspot migration path and the soil health score, the comprehensive risk weight of each region is calculated, and the management priority order is determined according to the comprehensive risk weight.

Images