A soil moisture inversion method, apparatus, medium, and device

Abstract

The application discloses a soil humidity inversion method, device, medium and equipment, and relates to the technical field of microwave remote sensing inversion. The application obtains the total backscattering coefficient of electromagnetic waves, the normalized vegetation index, the leaf area index and the average leaf inclination angle of a target region per pixel by combining microwave remote sensing data and atmospheric apparent reflectivity data of the target region; then, the vegetation is spatially clustered according to the normalized vegetation index, the leaf area index and the average leaf inclination angle, and the spatial heterogeneity of crop types is captured; subsequently, based on the same constraints of the surface roughness parameter and the vegetation parameter of the same vegetation type and the same soil humidity in the same surface spatial window, the two-stage iterative inversion of the intra-class scale and the surface spatial window scale is performed, so that the time dynamics is adapted, the parameter calibration function of the in-situ soil humidity data is replaced, and the model parameters and the soil humidity can be determined without the in-situ soil humidity data. The application improves the accuracy and reliability of the soil humidity inversion method.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of microwave remote sensing inversion, and in particular to a soil moisture inversion method, device, medium and equipment. BACKGROUND

[0002] At present, soil moisture, as a key regulating factor of crop root water absorption and transpiration metabolism in an agricultural ecosystem, has important scientific significance for optimizing irrigation management strategies, relieving regional water resource pressure, and even maintaining global food security. Remote sensing technology can obtain continuous surface information in a regional or even global range due to its macro-scale observation capability and periodic revisit characteristics, which brings a revolutionary technological breakthrough for agricultural soil moisture monitoring. Microwave remote sensing, based on the strong coupling relationship between soil dielectric constant and water content and the cloud penetration capability, has shown irreplaceable application potential in the business monitoring of farmland soil conditions.

[0003] In the prior art, in the field of microwave remote sensing soil moisture inversion, the standard process usually includes two key steps: first, a mathematical mapping of radar backscattering and soil dielectric properties is constructed, and this process relies on various scattering models to describe the electromagnetic interaction mechanism; then, a quantitative estimation of soil surface water content is realized by means of a dielectric constant mixing model. Some studies have developed a series of models to characterize the scattering properties of bare soil based on radiation transfer theory and multi-source observation data. Among them, the AIEM model, with its strict electromagnetic scattering theory foundation, combined with the Dobson dielectric model to characterize the coupling of soil moisture, temperature, texture (clay and sand content) and bulk density, effectively solves the parameter adaptability problem of different soil types, and shows excellent inversion performance in complex surface conditions.

[0004] However, although the above-mentioned models have achieved remarkable results in bare soil conditions, their application in vegetation-covered areas still faces the challenge of difficult decoupling of surface scattering characteristics. To solve this problem, some researchers have proposed a water cloud model (Water-Cloud Model, WCM). This model decomposes the total backscattering coefficient of vegetation-covered surface into two parts: vegetation body scattering and soil scattering after vegetation attenuation. Among them, the vegetation body scattering is related to the vegetation water content, and the soil scattering is affected by the vegetation attenuation, and the attenuation degree depends on the vegetation water content and crown type and other factors. In this way, the water cloud model effectively decouples the scattering contribution of vegetation and soil, providing a feasible way for soil moisture inversion in vegetation-covered areas.

[0005] In the actual inversion process, the Advanced Integral Equation Model (AIEM) is the mainstream physical model for simulating the microwave scattering characteristics of bare ground. The AIEM model and the water cloud model face the challenges of parameter dynamic updating and scale matching. For the AIEM model, the surface roughness parameters (root mean square height s , correlation length l ) are the physical basis for describing the interaction between electromagnetic waves and the ground. The numerical value directly determines the simulation accuracy of the backscattering coefficient, but the parameter has significant temporal and spatial heterogeneity. The temporal uncertainty is caused by multi-scale dynamic variation driven by natural and human processes such as precipitation compaction and farming activities. Spatial heterogeneity is determined by the non-uniform distribution of microscale to regional scale caused by terrain slope, soil texture and farming system. The vegetation parameters (volume scattering coefficient A, attenuation coefficient B) in the water cloud model also face the problem of determination. In the spatial dimension, the parameter heterogeneity across vegetation types will significantly differentiate the adaptability of the inversion algorithm. In the time dimension, the dynamic change of crop phenology makes the inversion model based on static parameters have seasonal bias. This temporal and spatial heterogeneity essentially restricts the applicability of model calibration based on limited spatiotemporal prior information. At the same time, the ill-posed nature of the parameter inversion problem makes it almost impossible to achieve accurate calibration of the model at high spatiotemporal resolution.

[0006] Therefore, some studies have introduced temporal and spatial constraints on model parameters to achieve more stable model calibration. Common assumptions include that model parameters remain stable within a time window, and that the same vegetation type shares a uniform set of vegetation parameters. These assumptions are in fact a balance between limited prior information and the contradiction of parameter ill-posed inversion. Based on this, the number of model parameters to be estimated is significantly reduced, and combined with multi-temporal remote sensing data and land cover type data, soil moisture can be inverted at the global or regional scale.

[0007] However, in the farmland area, the crop growth cycle is short, the farming activities are frequent, and the planting structure is complex. The satellite revisit period is difficult to match the change frequency of crop parameters and surface roughness, and the spatial scale and classification standard of land cover type data are relatively rough, which is usually difficult to represent the spatial variation characteristics of crop types. At the same time, the current assumptions still rely on the calibration of in-situ soil moisture data (soil moisture measured in the field, such as the value measured by soil moisture sensors buried in the field soil, or the soil moisture measured by drying method using local soil samples) or existing soil moisture remote sensing data, which is easy to introduce errors in soil moisture data. In summary, the existing inversion method of farmland soil moisture has low accuracy and poor reliability. SUMMARY

[0008] Therefore, it is necessary to provide a soil moisture inversion method, device, medium and equipment to solve the above technical problems.

[0009] The application adopts the technical solutions below:

[0010] The application provides a soil humidity inversion method, which comprises the following steps: determining the total backscattering coefficient of electromagnetic waves of a target region pixel by pixel according to microwave remote sensing data of the target region; determining the normalized vegetation index, leaf area index and average leaf inclination angle of the target region pixel by pixel according to the atmospheric apparent reflectivity data of the target region; clustering the normalized vegetation index, leaf area index and average leaf inclination angle of the target region to determine the vegetation types of the target region and the corresponding regions; coupling the backscattering model of soil, the soil dielectric constant model and the water cloud model to obtain the influence relationship between the surface roughness parameters, vegetation parameters and soil humidity of the vegetation-covered ground and the total backscattering coefficient of electromagnetic waves; initializing the soil humidity pixel by pixel by taking the same surface roughness parameters and vegetation parameters of the same vegetation type and the same soil humidity in the same surface space window as constraints, constructing the influence relationship equation set of the intra-class scale for each vegetation type and solving the equation set to obtain the initial surface roughness parameters and vegetation parameters of each vegetation type; the surface space window corresponds to a plurality of adjacent pixels; and then constructing the influence relationship equation set of the window scale for each surface space window of the target region divided in advance, substituting the initial surface roughness parameters and vegetation parameters of the corresponding vegetation type to solve the equation set to obtain the updated soil humidity of each surface space window, so as to iteratively update and solve the surface roughness parameters and vegetation parameters of each vegetation type and iteratively update and solve the soil humidity of each surface space window.

[0011] The application provides a soil humidity inversion device, which comprises:

[0012] A preliminary determination module is configured to determine the total backscattering coefficient of electromagnetic waves of a target region pixel by pixel according to microwave remote sensing data of the target region, and determine the normalized vegetation index, leaf area index and average leaf inclination angle of the target region pixel by pixel according to the atmospheric apparent reflectivity data of the target region.

[0013] A clustering module is configured to cluster the normalized vegetation index, leaf area index and average leaf inclination angle of the target region to determine the vegetation types of the target region and the corresponding regions.

[0014] A coupling module is configured to couple the backscattering model of soil, the soil dielectric constant model and the water cloud model to obtain the influence relationship between the surface roughness parameters, vegetation parameters and soil humidity of the vegetation-covered ground and the total backscattering coefficient of electromagnetic waves.

[0015] The intra-class scale solving module is configured to initialize the per-pixel soil moisture under the constraint of the same surface roughness parameter and vegetation parameter of the same vegetation type and the same soil moisture in the same surface space window, construct an influence relationship simultaneous equation set for the intra-class scale of each vegetation type, and solve the influence relationship simultaneous equation set to obtain the initial surface roughness parameter and vegetation parameter of each vegetation type.

[0016] The window scale solving module is configured to construct an influence relationship simultaneous equation set for the window scale of each surface space window of the target region divided in advance, and substitute the initial surface roughness parameter and vegetation parameter of the corresponding vegetation type to solve the influence relationship simultaneous equation set to obtain the updated soil moisture of each surface space window, so as to iteratively update and solve the surface roughness parameter and vegetation parameter of each vegetation type and iteratively update and solve the soil moisture of each surface space window.

[0017] The application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the soil moisture inversion method.

[0018] The application provides a computer device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the soil moisture inversion method when executing the program.

[0019] The above at least one technical solution adopted by the application can achieve the following beneficial effects:

[0020] The application combines microwave remote sensing data and atmospheric apparent reflectivity data of a target region to obtain a total backscattering coefficient of electromagnetic waves, a normalized vegetation index, a leaf area index, and an average leaf inclination angle of each pixel of the target region, and then performs spatial clustering on the vegetation according to the normalized vegetation index, the leaf area index, and the average leaf inclination angle to capture spatial heterogeneity of crop types, and subsequently, based on the constraint of the same surface roughness parameter and vegetation parameter of the same vegetation type and the same soil moisture in the same surface space window, the two-stage iterative inversion of the intra-class scale and the surface space window scale is performed to adapt to time dynamics and replace the parameter calibration function of in-situ soil moisture data, so that the model parameters and the soil moisture can be determined without in-situ soil moisture data. The application accurately depicts the spatiotemporal variation characteristics of the parameters, avoids the introduction error of in-situ soil moisture data, and improves the accuracy and reliability of the soil moisture inversion method. BRIEF DESCRIPTION OF DRAWINGS

[0021] The accompanying drawings, which are included to provide a further understanding of the application and constitute a part of this application, illustrate certain illustrative embodiments of the application and together with the description serve to explain the application. In the drawings:

[0022] Figure 1 This is a schematic diagram of a soil moisture inversion method provided by the present invention;

[0023] Figure 2 A schematic diagram showing the distribution of the research area and soil moisture observation points provided by this invention;

[0024] Figure 3 A schematic diagram of a specific data processing flow for soil moisture inversion provided by the present invention;

[0025] Figure 4 A schematic diagram of experimental results provided by the present invention;

[0026] Figure 5 This is a schematic diagram of the intra-class parameter inversion result provided by the present invention;

[0027] Figure 6 A schematic diagram of soil moisture inversion results for a research area provided by this invention;

[0028] Figure 7 A schematic diagram illustrating the influence of the k-value setting in cluster analysis provided by this invention on the soil moisture inversion method proposed in this study;

[0029] Figure 8 This invention provides a schematic diagram illustrating the influence of window scale variation on the evolution of model parameters.

[0030] Figure 9 This is a schematic diagram of a soil moisture inversion device provided by the present invention. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0032] Currently, researchers have developed a series of models to characterize the scattering properties of bare soil based on radiative transfer theory and multi-source observation data: including SPM, IEM, AIEM and other models that describe the relationship between surface parameters and radar backscatter through electromagnetic scattering theory, as well as Dubois model, Oh model, Shi model and other empirical / semi-empirical models relying on multi-band, multi-polarization, multi-incidence angle observation data. These models are committed to establishing a quantitative relationship between surface roughness, soil moisture and backscatter coefficient. As the mainstream model for soil moisture inversion in bare soil areas, AIEM effectively solves the parameter adaptability problem of different soil types by coupling soil water content, temperature, texture (clay and sand content) and bulk density parameters based on the strict electromagnetic scattering theory foundation of Dobson dielectric model, and exhibits excellent inversion performance under complex surface conditions.

[0033] The water cloud model decomposes the total backscatter coefficient of the vegetation-covered surface into two parts: vegetation scattering and soil scattering after vegetation attenuation. In actual inversion process, AIEM model and water cloud model face the challenges of parameter dynamic updating and scale matching. Researchers achieve more stable model calibration by introducing spatial and temporal constraints to model parameters. However, the crop growth cycle is short, the farming activities are frequent, and the planting structure is complex in farmland areas, and the satellite revisit period is difficult to match the change frequency of crop parameters and surface roughness. The spatial scale and classification standard of land cover type data are relatively rough and usually difficult to represent the spatial variation characteristics of crop types. In addition, although these assumptions have reduced the number of estimated parameters as much as possible, they still rely on in-situ soil moisture data or existing soil moisture remote sensing products for calibration. This limits the use of inversion models in data-scarce areas and introduces errors in soil moisture products.

[0034] To solve the above problems, the present application constructs a parameter multi-scale constraint iterative inversion method that integrates multiple remote sensing radiative transfer models and high-resolution radar-optical remote sensing images. This method uses high spatial resolution information provided by optical remote sensing data to depict the spatial heterogeneity of crops, and implements joint constraint and inversion of model parameters and soil moisture at three scales: within vegetation, local window and global range, aiming to realize stable inversion of model parameters and soil moisture at farmland scale without relying on in-situ soil moisture data.

[0035] The technical solutions provided by the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0036] Figure 1 The soil moisture inversion method flowchart in the present application specifically includes the following steps:

[0037] S101: Determine the normalized vegetation index, leaf area index, average leaf inclination angle and total backscattering coefficient of electromagnetic waves of the target region pixel by pixel according to the atmospheric apparent reflectivity data of the target region.

[0038] S102: Cluster the normalized vegetation index, leaf area index and average leaf inclination angle of the target region to determine the vegetation types of the target region and the corresponding regions.

[0039] S103: Coupling the backscattering model of soil, the dielectric constant model of soil and the water cloud model, obtain the influence relationship of the surface roughness parameters, vegetation parameters and soil moisture of the vegetation covered ground surface and the total backscattering coefficient of electromagnetic waves.

[0040] S104: With the same surface roughness parameters and vegetation parameters of the same vegetation type and the same soil moisture in the same ground surface space window as the constraints, initialize the soil moisture pixel by pixel, construct the intra-class scale influence relationship equation set for each vegetation type and solve it to obtain the initial surface roughness parameters and vegetation parameters of each vegetation type.

[0041] S105: Construct the window scale influence relationship equation set for each ground surface space window of the target region divided in advance, and substitute the initial surface roughness parameters and vegetation parameters of the corresponding vegetation type to solve it to obtain the updated soil moisture of each ground surface space window, to iteratively update and solve the surface roughness parameters and vegetation parameters of each vegetation type, and iteratively update and solve the soil moisture of each ground surface space window; coupling the leaf biochemical parameters, the ground surface space window corresponds to multiple adjacent pixels.

[0042] For convenience of description, only the server is described below as the execution subject. The server mentioned in the present application can be a server arranged in a business platform, or a device such as a desktop computer, a notebook computer, etc. capable of executing the scheme of the present application.

[0043] The present application takes a certain research area as an example for illustration, and the research area is also taken as a test area for subsequent illustration of the effect of the present application. The research area has an elevation of 1032 m to 1050 m, covers an area of about 2.157*10<5> hectares, contains 1.543*10<5> hectares of arable land, and the soil is mainly silty loam, loam and clay loam. The research area is located in the arid, semi-arid and semi-desert grassland zone, has sufficient sunlight, and the annual average sunlight duration is about 3180 hours. The average frost-free period in many years is 130 days to 168 days. The climate of the research area is mainly temperate continental climate, with cold winter and little snow, and hot and dry summer. The annual average temperature is about 6.3℃ to 7.7℃, the average precipitation is 138.2 mm, the average evaporation is 2096.4 mm, and the agricultural production of the research area is highly dependent on irrigation. The irrigation of the irrigation area is mainly based on the Yellow River water irrigation, and the annual water diversion amount is 1 billion to 1.2 billion cubic meters.

[0044] The research area is located in the southern part of the North China Plain and the eastern foot of the Taihang Mountains, belongs to the temperate continental monsoon climate zone, and is located in the semi-humid climate zone. The soil in the region is mainly loam and sandy loam, with local distribution of clay loam, and the soil layer is deep, which is suitable for agricultural production. The climate is distinct in four seasons, with cold and dry winter, hot and rainy summer, short spring and autumn, and climate change. The annual average sunlight duration is about 2300 hours, and the average frost-free period in many years is 190 days to 220 days, providing light and heat conditions for crop growth. The annual average temperature is about 13.0℃, the annual average precipitation is 600 mm to 700 mm, 60% to 70% of which is concentrated in summer, and the annual average evaporation is 1500 mm to 2000 mm, which is slightly higher than the precipitation. The irrigation system for agricultural production in the research area is relatively complete, and the water source includes groundwater, reservoir storage and Yellow River water diversion. Through water transfer by water conservancy projects such as the Hongqi Canal, the annual water diversion amount reaches hundreds of millions of cubic meters, and the Yellow River water irrigation plays a role in some areas. Figure 2 Fig. 1 is a distribution diagram of a research area and soil moisture observation points in the present application, Figure 2 Fig. 2 shows two research areas and their respective observation point distributions.

[0045] The present application also takes four types of data sets as examples for soil moisture inversion, which are: (1) Sentinel-1 GRD IW mode active microwave remote sensing data; (2) Sentinel-2 L1C atmospheric apparent reflectance data; (3) SoilGrids global gridded soil information; (4) in-situ observed surface soil moisture data. As shown in Table 1.

[0046] Table 1 Data sets for soil moisture inversion

[0047]

[0048] Sentinel-1 GRD IW mode active microwave remote sensing data: Sentinel-1 is composed of two twin satellites, Sentinel-1A and Sentinel-1B. The revisit period of a single satellite is 12 days, and the double satellites can be reduced to 6 days, and 5m~40m resolution satellite images can be obtained under all-weather conditions. Sentinel-1 satellite carries C-band synthetic aperture radar (SAR) sensor, its working frequency is 5.4GHz, has multi-polarization imaging capability and four imaging modes. The Sentinel-1 image used in the present application is the descending data of the interferometric wide swath mode, the corresponding ground resolution in this mode is 5m x 20m, the data level is Level-1, and the product data is in ground range detected (GRD) format. The time resolution in the study area is 12 days. Dual polarization is cross-polarization VH (vertical horizontal) and vertical polarization VV (vertical vertical). Considering that VV polarization is more suitable for being used to retrieve bare soil moisture, and AIEM can well simulate the backscattering of VV polarization, the present application only selects VV polarization as the total backscattering coefficient of electromagnetic wave in the target area. The data is obtained from the earth engine platform and is processed by Refined Lee filtering, and is exported at a spatial resolution of 20m.

[0049] Sentinel-2 L1C atmospheric apparent reflectance data: Like Sentinel-1, Sentinel-2 satellites are also composed of two twin satellites, Sentinel-2A and Sentinel-2B. The revisit period of a single satellite is 10 days, and the complement of the two satellites can reduce it to 5 days, with a revisit period of 2-3 days in the study area. Sentinel-2 satellites carry a multi-spectral imager with 13 spectral bands, a width of 290 km, and a ground resolution ranging from 10 m to 60 m depending on the band. Sentinel-2 L1C (S2L1C) products were selected and atmospheric correction was performed according to existing research results to obtain S2L2A products. To match the Sentinel-1 data, the Sentinel-2 date closest to the Sentinel-1 overpass date was selected. Time series linear interpolation and S-G filtering were used to fill in the local blanks after cloud removal using the QA60 band. The present example selects 10 m and 20 m spatial resolution 10 bands (B02-B8A, B11 and B12) covering the spectral region from visible light to short-wave infrared, and all are resampled to 20 m spatial resolution. At the same time, the normalized difference vegetation index (NDVI) grid data is calculated and output.

[0050] SoilGrids global gridded soil information: The relationship between soil dielectric constant and soil moisture depends on properties such as clay content, sand content, and soil bulk density. For regional scales, the spatial heterogeneity of soil cannot be ignored. SoilGrids is a global digital soil mapping system that uses state-of-the-art machine learning methods to map the spatial distribution of global soil properties. The prediction model is fitted using more than 230,000 soil profile observations from the WoSIS database and a series of environmental covariates. Covariates are selected from more than 400 environmental layers derived from Earth observation products and other environmental information, including climate, land cover, and topography. The output of SoilGrids is a global soil property map with a spatial resolution of 250 meters for six standard depth intervals (0-5 cm, 5-15 cm, 15-30 cm, 30-60 cm, 60-100 cm, and 100-200 cm). This helps to more accurately characterize the spatial variation of soil properties. The present example selects the clay content (clay_mean), sand content (sand_mean), and soil bulk density (bdod_mean) of the 0-5 cm depth in SoilGrids, resampled to 20 m, and used as covariates to convert soil moisture and dielectric constant.

[0051] In-situ observed surface soil moisture data: Soil volumetric water content measurements were conducted at multiple observation points in the study area before. For example, soil volumetric water content measurements were conducted twice at 100 observation points in study area 2 in Figure 2

[0052] Soil moisture inversion in farmland area can be refined into four problems: 1) parameterization of crop canopy; 2) extraction of soil backscattering component; 3) calculation of soil dielectric constant; 4) conversion of soil moisture. The present application uses PROSAIL model, water cloud model, AIEM model and Dobson model respectively to solve the above four problems.

[0053] PROSAIL model, as the current international widely used vegetation canopy radiation transfer model, is developed by combining leaf optical property model PROSPECT and canopy radiation transfer model SAIL (Verhoef & Baret, 2009). The model can accurately simulate canopy directional reflectance in the wavelength range of 400 nm-2500 nm by coupling leaf biochemical parameters (such as chlorophyll content, leaf structure parameters) and canopy three-dimensional geometric structure (such as leaf area index, leaf angle distribution). In the present application, Sentinel-2 image is taken as observation data, and leaf area index (LAI) and average leaf angle (ALA) of the target area are taken as target inversion parameters to construct the search population of target inversion parameters and establish the iterative inversion process including model forward simulation and optimization algorithm. The canopy directional reflectance can be simulated by coupling leaf biochemical parameters and canopy three-dimensional geometric structure through PROSAIL model, and the fitness value of each individual in the search population is determined according to the observation data, so as to minimize the spectral residual between the canopy directional reflectance simulated based on the search population and the observation data as the optimization objective, and the target inversion parameters are optimized. Specifically, particle swarm optimization algorithm (PSO) can be used to minimize the spectral residual between the model simulation value and the image observation value, search for the optimal parameter combination in the PROSAIL model parameter space, and finally obtain the LAI and ALA thematic raster data of the study area. Further, canopy water content (CWC) is defined as the product of leaf area index (LAI) and leaf water content (LWC). Therefore, the canopy water content of the study area can be further obtained through the parameter search result.

[0054] ​A water cloud model was proposed to estimate soil moisture in agricultural areas by studying the backscattering properties of crop-covered surfaces. The model considers that the total backscattering of a vegetation-covered surface is composed of two parts: the volume scattering term directly reflected by the vegetation canopy and the ground backscattering term after double attenuation by the vegetation canopy. The water cloud model is based on the following assumptions: 1) the vegetation layer is assumed to be a homogeneous cloud; 2) only single scattering is considered, and multiple scattering between the vegetation layer and the ground is ignored; and 3) only the variables of vegetation water content, soil moisture, and incident angle need to be considered in the model.

[0055] The expression of the water cloud model is as follows:

[0056] ;

[0057] ;

[0058] .

[0059] In the formula, τ is the total backscattering coefficient of electromagnetic waves; τv is the direct vegetation backscattering coefficient; τs is the soil backscattering coefficient; w is the vegetation water content; τv is the vegetation double-layer attenuation factor (transmittance); θ is the incident angle of electromagnetic waves; and the value of depends on the vegetation type and the frequency of incident electromagnetic waves. ,

[0060] The Advanced Integral Equation Model (AIEM) is a mainstream physical model for simulating the microwave scattering properties of bare surfaces. It is improved on the basis of the integral equation model and can adapt to different roughness conditions of the ground surface by improving the surface roughness power spectrum and the Fresnel reflection coefficient, thus more accurately reproducing the backscattering of real surfaces. It is widely used in the simulation and analysis of microwave surface radiation and scattering.

[0061] The backscattering form of AIEM can be generalized as follows:

[0062] .

[0063] In the formula, φ is the scattering azimuth angle (the backscattering is fixed at 180°); L is the correlation length; h is the root mean square height; and ε is the soil complex permittivity. ​​​​​​​​​​​The electromagnetic wave frequency is a constant value (5.4 GHz in the present application).

[0064] For the correlation length, the ground surface is not completely random ups and downs, and the height of adjacent points often has certain correlation (for example, near the top of a small soil slope, the adjacent points are probably in the same inclined section of the slope body, rather than suddenly becoming a depression). The correlation length is the maximum distance at which such correlation "exists effectively" calculated by mathematical methods (such as "autocorrelation function").

[0065] The root mean square height (RMSH) ) and the correlation length (CL) ) together describe the roughness of the ground surface and have certain effects on the backscattering. A combined roughness is proposed to more simply describe the roughness of the ground surface and has achieved good results. The AIEM model can be expressed as:

[0066] .

[0067] The Dobson model is a classic soil dielectric constant model, which aims to quantitatively describe the coupling relationship between the radar incident wave frequency, the soil physical properties and the soil moisture content, and is an important basis for soil moisture inversion in the field of microwave remote sensing. The commonly used soil dielectric model is an improved model, which further expands the application range of the model, and can effectively depict the variation of the soil complex dielectric constant with the soil volume moisture content, the electromagnetic wave frequency, the soil texture (sand / clay content), the soil temperature and the bulk density.

[0068] The Dobson model can be generalized as:

[0069] .

[0070] In the formula, is the soil temperature (the effect on the simulation results is not significant, and the present application is fixed at 25℃); is the soil moisture; is the clay content; is the sand content; is the soil bulk density.

[0071] When the soil moisture is inverted in combination with the above-mentioned model, the coupling of the present application model is divided into four steps, 1) coupling the AIEM and the Dobson model; 2) simplifying the AIEM_Dobson model; 3) connecting the water cloud model; and 4) using the canopy water content to proxy the vegetation water content.

[0072] Firstly, AIEM model is coupled with Dobson model, which can convert soil moisture into soil complex permittivity under certain soil properties, providing input parameters for AIEM model, and then the theoretical relationship between soil moisture and soil backscattering coefficient is built.

[0073] At this time, the soil backscattering coefficient can be expressed as:

[0074] .

[0075] AIEM_Dobson model simplification. The high nonlinearity of AIEM model and Dobson model will bring challenges to parameter inversion, so it is necessary to simplify AIEM_Dobson model. Existing studies have shown that when , , and parameters are constant, the above model can be simplified as:

[0076] .

[0077] In the formula, , , and are empirical parameters, which are estimated by AIEM_Dobson model. The specific steps are as follows: fix the parameters of AIEM_Dobson model , , and , and traverse and to generate a simulated data set, and finally use the least square method to calculate the coefficients. The parameters , , and in the examples of the present application are provided by the angle properties of Sentinel-1 and SoilGrids data, and then the grid data of , , and of the study area are generated pixel by pixel. For the determination of , , and , the electromagnetic wave incidence angle can be determined by microwave remote sensing data of the target area; the clay content , the sand content and the soil bulk density It can be determined by using partial data of the target area corresponding to the global gridded soil information.

[0078] Then, the water cloud model is connected in series, and the water cloud model contains... The parameters are simulated by the AIEM_Dobson model, and the above formulas can be rearranged to obtain:

[0079] .

[0080] This formula describes the relationship between soil moisture and backscattering on vegetated surfaces.

[0081] Next, canopy water content was used to proxy vegetation water content. Sentinel-2 imagery was used as observation data, and the canopy water content (CWC) raster data of the study area was obtained through inversion using the PROSAIL model. Here, it is assumed that the vegetation water content (VWC) of the same crop is a linear function of the canopy water content (CWC).

[0082] .

[0083] Substituting into the formula, we get:

[0084] ,

[0085] , .

[0086] Right now, .

[0087] and The specific values ​​still depend on the vegetation type and the frequency of the incident electromagnetic waves; therefore, further research is needed. and These are considered as the transformed vegetation parameters.

[0088] Finally, parameter inversion is performed for a specific pixel. of All are known, and the parameters to be inverted are: For the study area There are a total of 100 pixels. There are equations, but there are Solving for unknown parameters is difficult. Therefore, it is necessary to add constraints to achieve parameter inversion.

[0089] Considering parameters and Depending on the vegetation type, it can be hypothesized that there are [various vegetation types] within the study area. Crops of different types, and each crop type shares the same set of parameters. and Meanwhile, each crop type shares the same parameter This is because the surface roughness of farmland area is mainly affected by tillage measures, and the same crop type has similar tillage measures. Due to the surface runoff of water and the large-scale characteristics of the water event, it is assumed that The pixels in the window have the same .

[0090] The present application considers that the influence of vegetation types on backscattering is mainly derived from the difference in the crown structure of different vegetation, and the crown structure can be simply described by the leaf area index and the average leaf angle. In addition, the NDVI is used to describe the difference in vegetation coverage. Therefore, the vegetation types in the research area are divided into types by clustering the NDVI calculated by Sentinel-2 and the LAI and ALA inverted by PROSAIL. Based on the above assumption, the number of unknown parameters is reduced to If and are valued so that , the condition for preliminary parameter inversion is met. In the embodiment of the present application in the research area, the value of is set to 5, and the value of is set to 3.

[0091] The parameter inversion can be divided into two parts: the inversion of the intra-class parameter and the inversion of the window parameter . Since the two parts of parameters are not in the same spatial scale, and the inversion results depend on each other, the present application proposes an iterative trial inversion method. The specific steps are as follows: first, assuming that the soil moisture of the research area is , the initial window parameter is obtained, and it is substituted into the equation; second, the intra-class parameter is solved by simultaneously solving the equation set in the same class (intra-class scale), and the same operation is performed for each class; third, the intra-class parameter is substituted back into the equation, and the new window parameter is solved by simultaneously solving the equation set in the window (window scale), and each window is traversed; fourth, it is judged whether the maximum value of the difference between all windows and is less than an acceptable threshold. If the judgment condition is not met, the value of is updated, and the iteration is continued. Otherwise, the iteration is stopped, the value of at this time is saved, and the backscattering coefficient is reconstructed based on this at the global scale (global scale), and the reconstruction accuracy is calculated. Finally, the value of is traversed, the optimal backscattering coefficient reconstruction accuracy is screened, and the corresponding at this time is output., parameter inversion is completed. A special case is that the canopy water content of a certain class is close to 0, and the class is considered as bare soil, and the in-class parameters are solved only by , and the remaining steps remain unchanged. Figure 3 It is a specific data processing flowchart of soil moisture inversion in the application.

[0092] Taking the above research area as an example, the model is simplified pixel by pixel on the research area by fusing the soil properties of SoilGrids and the incident angle of Sentinel-1 data, and satisfactory results are obtained. The determination coefficient of the simulation results of the simplified empirical model and the AIEM_Dobson model is almost above 0.8, as shown in Figure 4 , Figure 4 is a schematic diagram of an experimental result in the application, Figure 4 (a) is the known input parameter of the AIEM_Dobson model, (b) is the coefficient of the simplified semi-empirical model, (c) is the simplification effect of the semi-empirical model on the AIEM_Dobson model in the research area, and (d) is the correlation between the input parameters and the semi-empirical coefficients.

[0093] In Figure 4 part (d), it can be found that the coefficient a of the semi-empirical model is mainly affected by parameters such as , and , and all are negatively correlated. Coefficient b is mainly affected by parameters and , which are positively correlated with and negatively correlated with . Coefficient d also shows a similar response law as coefficient b, but coefficient b is more dependent on , and coefficient d is more dependent on . Coefficient c is mainly affected by parameters , and , which are positively correlated with and negatively correlated with and . On the regional scale, the input parameters of the model all have spatial heterogeneity, which makes the semi-empirical model coefficients after simplification also have certain spatial heterogeneity. Although this feature is not obvious, the pixel-by-pixel operation also avoids the negative effects of using fixed model coefficients when performing soil moisture inversion on the regional scale.

[0094] Figure 5 is a schematic diagram of in-class parameter inversion results in the application. Figure 5 ​The clustering results of study area (1) on July 9, 2018, August 3, 2018, September 3, 2019 and study area (2) on August 29, 2024 and the inversion results of each category A', B' and Z are shown in (a-d) respectively. (e) shows the relationships between parameters A' and Z and NDVI, LAI and ALA. The clustering results are only used as labels in a specific spatiotemporal range and are not universal in different regions or different periods.

[0095] In the examples of the present application, the vegetation in a specific time and a specific study area is divided into 5 categories by using clustering analysis. It can be found that the distribution of each category of vegetation in the study area presents a phenomenon of overall aggregation, i.e. a certain category of vegetation is mainly concentrated in a specific range. This is generally consistent with the characteristics of land types and the planting preferences of farmers in the study area.

[0096] On July 10, 2018, the parameters A' of each category of vegetation in study area (1) were -20.764, -19.126, -18.995 and -14.825 respectively. The parameters B' were 0.105, 0.210, 0.138 and 0.137 respectively. The parameters Z were 0.112, 0.066, 0.063, 0.033 and 0.207 respectively. Among them, category 5 is considered to be no vegetation coverage because the overall canopy water content is low, and the parameters A' and B' are directly set to 0. On August 3, 2018, the parameters A' of each category of vegetation in study area (1) were -23.746, -21.034, -20.712 and -15.255 respectively. The parameters B' were 0.099, 0.109, 0.101 and 0.074 respectively. The parameters Z were 0.505, 0.822, 0.366, 0.735 and 0.828 respectively. Like on July 9, 2018, the parameters A' and B' of category 5 were directly set to 0. On September 3, 2019, the parameters A' of each category of vegetation in study area (1) were -21.309, -19.282, -16.289 and -15.502 respectively. The parameters B' were 0.031, 0.154, 0.075 and 0.176 respectively. The parameters Z were 0.027, 0.007, 0.009, 0.005 and 0.023 respectively. The parameters A' and B' of category 5 were directly set to 0. On August 29, 2024, the parameters A' of each category of vegetation in study area (2) were -18.188, -15.679, -14.944, -12.615 and -11.274 respectively. The parameters B' were 0.110, 0.130, 0.136, 0.112 and 0.091 respectively. The parameters Z were 0.538, 0.299, 0.197, 0.180 and 0.332 respectively.

[0097] From study area (1) on July 10, 2018 (a), August 3, 2018 (b), September 3, 2019 (c) and study area (2) on August 29, 2024 (d), the relationships between parameters A' and Z and NDVI, LAI and ALA are shown in (e). Figure 5(a) , August 3, 2018 ( Figure 5 (b) and September 3, 2019 ( Figure 4 From the clustering results and parameter inversion of (c) in the study area, the distribution of each category and the parameter values ​​differ between months and years. This reflects that the change in vegetation structure and coverage over time, coupled with agricultural activities, affects the parameters of the water cloud model, demonstrating the dynamic response of the model parameters to the temporal changes in farmland surface characteristics. The parameter inversion results of the study area (2) are as follows: Figure 4 The results in the middle (d) and the study area (1) at different time phases are significantly different. This is due to the differences in vegetation type and structure, soil properties, and farming patterns between the two areas, indicating that the parameters of the water cloud model are regionally specific and constrained by the comprehensive characteristics of the underlying surface. In summary, the same parameter varies at different times, locations, and vegetation types. Ignoring the spatiotemporal heterogeneity of the parameters may have a negative effect on the inversion of soil moisture.

[0098] This invention also calculates the mean of three canopy features (NDVI, LAI, and ALA) from the cluster analysis, attempting to explore their impact on parameters A' and B'. Figure 5 As shown in equation (d), parameter A' generally increases with increasing NDVI and LAI, but its relationship with ALA is not significant. Parameter B' generally increases with increasing NDVI and LAI and then tends to plateau. When LAI is the same, the smaller ALA is, the larger parameter B' is. However, when NDVI is approximately greater than 0.5, the effect of ALA on parameter B' is no longer significant. According to the equation, parameter A' is related to vegetation backscattering. NDVI and LAI represent vegetation density; the increase of parameter A' along with NDVI and LAI means that the greater the vegetation density, the greater the contribution of vegetation to radar wave backscattering. Parameter B' mainly reflects the attenuation of radar waves propagating in the vegetation layer. The larger the value of B', the more severe the attenuation of radar waves propagating in the vegetation layer, that is, the stronger the shielding effect of vegetation on radar waves. B' initially increases and then stabilizes with increasing NDVI and LAI, essentially due to the combined effect of the vegetation canopy's microwave attenuation mechanism and the canopy structure's saturation effect: When LAI is small, the canopy leaves are sparse, and microwave attenuation is mainly due to single scattering. As LAI increases, the absorption and scattering paths of microwaves by the leaves lengthen, leading to an increase in the B' value. When LAI exceeds a certain threshold, the overlapping canopy leaves form a closed layer, and microwaves are largely absorbed or scattered at the top of the canopy. The excess leaves only increase internal multiple scattering, weakening the overall attenuation efficiency improvement, and the B' value tends to stabilize. Parameter Z is mainly affected by cultivation practices.

[0099] Soil moisture inversion results: In order to verify the reliability of the method, the in-situ observation data of research area (1) and research area (2) are used to verify the inversion results. At the same time, the method of the application is compared with the model calibrated by part of the data of research area (2) across regions.

[0100] Figure 6 The soil moisture inversion results of a research area in the application are shown in the figure. (a) is the spatial distribution of soil moisture at different times and places based on the method. (b) is the relationship between the vegetation backscattering coefficient and the actual scattering coefficient based on the soil moisture inversion results. (c) is the soil moisture inversion accuracy based on multi-scale parameter optimization and unified parameter calibration. For the method of unified parameter calibration, 50 measured data of research area (2) are used to calibrate the vegetation parameters, and the remaining 50 of research area (2) and all in-situ soil moisture data of research area (1) are used to verify the inversion accuracy of the two methods.

[0101] The surface soil moisture value range of the research area at different times and places is stably distributed in the interval of 0 to 0.4. Affected by different crop irrigation systems, the irrigation time shows significant spatial heterogeneity, resulting in a strong correlation between soil moisture distribution pattern and vegetation category. Taking research area (1) on July 9, 2018 as an example, the soil moisture in the northeast region is relatively higher than that in other regions, forming a certain soil wetting concentration area.

[0102] Based on the soil moisture and intra-class parameter inversion results, the model is constructed to simulate the soil backscattering coefficient. The results show that under the global spatio-temporal scale, the model determination coefficient R² has no big difference, and all remain above 0.88, indicating that the model can well describe the soil backscattering characteristics when facing the spatial heterogeneity of different vegetation structures, and effectively reflects the influence mechanism of soil moisture and vegetation canopy characteristics on backscattering coefficient.

[0103] The soil moisture inversion method constructed by the application shows robust applicability under the global scale, and the determination coefficients (R2) are 0.265, 0.363, 0.367 and 0.379 respectively under four scenarios. The root mean square error (RMSE) is 0.083, 0.097, 0.087 and 0.127 respectively. Compared with this, the unified parameter inversion model calibrated based on the in-situ data of the research area (2) achieves higher precision locally (R2=0.548, RMSE=0.096). But its spatial generalization ability is significantly insufficient, and the determination coefficients under the three scenarios of the research area (1) are 0.105, 0.202 and 0.197 respectively. The root mean square error is 0.120, 0.138 and 0.113 respectively, and the inversion accuracy is not as good as the method proposed in the application. This shows that the temporal and spatial migration of the single parameter calibrated based on the data of a certain space-time has obvious limitations. The above phenomenon reveals the limitation of calibrating the parameter depending on the data of a certain region: the unified parameter system has strong dependence on the observation data of the research area (2), which makes it difficult to directly adapt to other geographical environments. Considering that large-scale in-situ data acquisition faces high human and economic costs, and it is difficult to implement in areas where ground observation data is scarce, the application scenario of such calibration method is significantly restricted. Compared with this, the inversion method proposed in the present application breaks through the data dependence bottleneck and still maintains stable performance under the condition of limited data acquisition, providing a more universal technical path for soil moisture inversion.

[0104] In the farmland area, the crop phenology process is fast, the canopy structure changes greatly with the growth stage, and the differences in cultivation system cause the canopy characteristics to be significantly differentiated among fields, so that the model parameters representing the scattering and attenuation characteristics of the vegetation body need to be dynamically adjusted according to the canopy characteristics; at the same time, human activities such as cultivation, irrigation and harvesting frequently change the surface micro-topography, which superimposes natural processes such as precipitation compaction, causing the surface roughness to fluctuate in a short time, and the differences in field-scale management measures further exacerbate the spatial heterogeneity. These factors jointly cause the microwave scattering model parameters in the farmland area to present strong temporal and spatial variation. The parameters calibrated depending on the data of a certain region are difficult to adapt to the dynamic changes under different crop types, growth stages and management measures, resulting in a significant decrease in the inversion accuracy when the model is applied across regions or over a long time. The application couples multiple remote sensing radiation transfer models, proposes a multi-scale constraint of parameters based on the assumption of parameter sharing within the vegetation type and the consistency of window-scale soil moisture, and reduces the dimension of unknown parameters. By finely representing the spatial heterogeneity of crops through high-resolution remote sensing data, and combining the iterative inversion mechanism to realize the coupled solution of the model parameters and soil moisture, the limitation of calibrating the model parameters based on in-situ observation data is avoided, and the applicability of the model in the farmland area and the stability of the parameter inversion are improved.

[0105] To make the system error and the scope of application of the proposed method more clear, taking the irrigation area (1) on July 10, 2018 as an example, the following three aspects are discussed: first, the sensitivity of the number of clusters (k value) to the accuracy of soil moisture inversion is analyzed, and the internal relationship between the classification granularity of vegetation and the adaptability of the model is revealed; second, the influence of window scale change on the evolution law of model parameters is discussed, and the coupling response mechanism between spatial heterogeneity assumption and scattering process modeling is revealed.

[0106] Influence of cluster number: the algorithm proposed in the present application classifies LAI, ALA and NDVI to describe the differentiation of crop types in the study area. However, the setting of k value is subjective, and the influence of k value on the inversion of soil moisture needs to be clarified. Figure 7 The influence of the k value of a cluster analysis in the present application on the soil moisture inversion method proposed in the present application is shown in Figure (a), which is the variation coefficient of soil moisture inversion based on different k values in the study area. (b) is the soil moisture inversion accuracy based on different k values. (c) is the model parameter when k is 10. (d) is the variation coefficient with the change of leaf area index.

[0107] Figure 7 (a) shows the variation coefficient of soil moisture inversion based on different k values in the study area. The variation coefficient is distributed between 0 and 0.04, and the relatively high value is mainly concentrated in the northeast corner. In order to further illustrate this phenomenon, Figure 7 (d) shows the variation coefficient with the change of LAI. As LAI gradually increases, the variation coefficient increases. This is because when LAI tends to 0, the radar backscattering of the ground surface is mainly the direct backscattering of bare soil, and the weight of vegetation participation is small, so the classification granularity of vegetation will not affect the applicability of the model. As LAI gradually increases, the weight of vegetation participating in backscattering becomes larger, and accurately describing the heterogeneity of vegetation parameters (A', B') in the model helps the model to reasonably reproduce the radar backscattering process. Therefore, the variation coefficient of soil moisture inversion based on different k values increases with the increase of LAI.

[0108] Figure 7The middle (b) shows the accuracy of the model for soil moisture changes with k value. It can be found that with the increase of k value, the inversion accuracy of soil moisture presents the state of first increasing and then stable (R2 first increases and then tends to be stable, RMSE first decreases and then tends to be stable). This means that the model has stable performance when facing the uncertainty of k value selection, and does not need to distinguish the vegetation in the study area too much. This is mainly because when the k value is less than a certain threshold, the difference of the canopy characteristics between different types of vegetation is still obvious. At this time, increasing the k value helps the algorithm to analyze the spatial heterogeneity of the model parameters. But when the k value is greater than a certain threshold, the difference of the canopy characteristics between some types of vegetation is no longer obvious, and the heterogeneity of the corresponding model parameters is also not significant. For example, Figure 7 As shown in the middle (c), when the k value is set to 10, the model parameters of category 3 and category 4 are very similar, and the same is true for category 5 and category 6. At this time, the increase of k value gradually reduces the help to the model accuracy, and increases the computational resource overhead. In this study, Figure 7 The soil moisture variation shown in the middle (a) is also mainly due to the change of k value between 1 and 5. In practical application, a larger k value can be selected first to avoid insufficient description of parameter heterogeneity, and then the size of k value can be adjusted according to the planting structure in the region, so as to balance the accuracy and computational resource overhead of the model.

[0109] The influence of window size: the algorithm proposed in the present application assumes that the soil moisture in the same window is consistent, the number of model parameters is reduced, the solution of model equation is more stable, the parameters are more convergent, and the mathematical identifiability of inversion is improved. But with the increase of window size, this assumption gradually deviates from the actual spatial distribution characteristics of soil moisture. The actual surface soil moisture has spatial heterogeneity, which cannot be accurately expressed by fixed SM parameters in a larger window range, which limits the explanation ability of the model to backscattering. Therefore, it is necessary to explore the influence of window size on the inversion of model parameters. The present application sets the window size from 2 to 7 with a step of 1 to inverse the parameters. The results show that: in order to reduce the structural error caused by the deviation of the above assumption from the fact, the model parameters will show a kind of "statistical compensation mechanism" in the process of parameter inversion.

[0110] Figure 8 A schematic diagram of the influence of window size on the evolution of model parameters in the present application, (a) is the coefficient A' of vegetation body scattering term based on different window sizes. (b) is the coefficient B' of vegetation attenuation term based on different window sizes. (c) is the surface roughness Z based on different window sizes. (d) is the accuracy of the model reconstructed backscattering coefficient with the change of window size.

[0111] Specifically, the water cloud model parameters A' (vegetation scattering coefficient) and B' (vegetation attenuation coefficient) exhibit a monotonically decreasing and stabilizing characteristic as the window size increases. Figure 8 (a, b)). This trend reflects that the model reduces the weight of the vegetation term in the total scattering, thereby decreasing its sensitivity to spatial variations and avoiding the erroneous attribution of spatial heterogeneity of soil moisture to vegetation factors. This does not mean that the canopy physical processes themselves are weakened, but rather that in statistical fitting, the model chooses to suppress the canopy parameter as a compensatory strategy to fit the simplified SM assumptions. As the window continues to increase, the model gradually reaches the optimal balance between the contribution of the vegetation term and error control, thus both A' and B' exhibit convergence characteristics.

[0112] Meanwhile, the surface roughness parameter Z is also affected by the accuracy of the SM assumption, and its value shows an increasing trend in the early stage of window expansion for most surface types. Figure 8 (c) Since Z and SM are modeled together as interaction terms, this phenomenon can be understood as the model enhancing its ability to explain the spatial variation of SM by improving Z, thereby compensating to some extent for the model residuals caused by the decrease in soil moisture accuracy. However, as the window continues to expand, the actual expressive power of SM-related terms gradually decreases, the "compensation space" of Z tends to saturate, the parameter estimation enters a stationary state, and thus tends to stabilize at large scales.

[0113] However, for bare soil, Z shows a decreasing trend as the window size increases, revealing the difference in model error compensation mechanisms across different surface types. In bare soil areas, backscattering is mainly controlled directly by surface roughness and soil moisture, with vegetation having a very weak or even negligible impact. With a larger window, the contribution of the SM assumption error to total scattering becomes more sensitive on this type of surface. However, due to the lack of a vegetation term to adjust for bare soil surfaces, the model must suppress Z to reduce the risk of overfitting the roughness term in the SM error interpretation, thus avoiding an overestimation of scattering changes caused by surface structure. In other words, in the absence of redundant variables for error absorption, the model maintains overall fitting stability by weakening the effect of the roughness scattering term.

[0114] During soil moisture inversion with different window sizes, the accuracy of the model reconstructed backscattering coefficient showed a typical trend of "first increasing and then decreasing". Figure 8(d) This phenomenon intuitively reflects the dynamic trade-off between parameter estimation stability and model structural error. In the early stage of window expansion, parameter estimation gradually stabilizes and the error absorption mechanism is enhanced, leading to an improvement in model fitting accuracy. However, as the window continues to expand, the heterogeneity of the SM space can no longer be offset by the simple "parameter compensation" mechanism, and the accumulation of structural error leads to a decrease in reconstruction accuracy. The window size at which the model accuracy reaches its peak coincides with the inflection point where the changes in parameters (especially A', B', and Z) tend to stabilize, suggesting that in this study region, this scale may be the optimal window size for balancing physical consistency and statistical fit optimality.

[0115] In summary, the impact of window size variation on model parameter estimation reflects the spatial heterogeneity of soil moisture, the compensation effect in the parameter estimation process, and the complex relationship of coupling different scattering mechanisms in statistical modeling. This phenomenon emphasizes the necessity of appropriately selecting the window size in microwave remote sensing soil moisture inversion to improve the physical reliability and robustness of the model.

[0116] based on Figure 1 The soil moisture retrieval method presented in this invention combines microwave remote sensing data and atmospheric apparent reflectance data of the target area to obtain the total backscattering coefficient, normalized vegetation index, leaf area index, and mean leaf tilt angle of electromagnetic waves per pixel in the target area. Then, based on the normalized vegetation index, leaf area index, and mean leaf tilt angle, the vegetation is spatially clustered to capture the spatial heterogeneity of crop types. Subsequently, based on the constraints that the surface roughness parameters and vegetation parameters of the same vegetation type are the same, and the soil moisture is the same within the same surface spatial window, a two-stage iterative retrieval at the intra-class scale and the surface spatial window scale is performed. This method adapts to temporal dynamics and replaces the parameter calibration function of in-situ soil moisture data, allowing the determination of model parameters and soil moisture without the need for in-situ soil moisture data. This invention improves the accuracy and reliability of the soil moisture retrieval method by accurately characterizing the spatiotemporal variation of parameters while avoiding the errors introduced by in-situ soil moisture data.

[0117] Existing technologies for microwave remote sensing inversion of farmland soil moisture face three major technical bottlenecks. This solution addresses these bottlenecks through a layered design of "model coupling - multi-scale constraints - iterative inversion," with the specific solution logic as follows:

[0118] Technical bottleneck 1. Decoupling of scattering signals in vegetation-covered areas requires in-situ data to calibrate model parameters (such as vegetation parameters and surface roughness in water cloud models). However, the prior information of limited in-situ data cannot adapt to the dynamic changes in farmland, leading to the failure of inversion in areas with scarce data.

[0119] The solution logic of the present application is: through the triple scale constraint of "vegetation clustering-local window-global verification", the parameter calibration function of in-situ soil moisture data is replaced, and the model parameters and soil moisture (SM) can be determined without in-situ soil moisture data.

[0120] Technical bottleneck 2. Assuming that the model parameters (vegetation parameters and surface roughness) are spatially and temporally stable (such as uniform vegetation parameters and fixed roughness), but the vegetation parameters in the farmland change dynamically with the crop phenology (NDVI / LAI change), and the surface roughness changes spatially and temporally with the farming activities (such as irrigation and plowing), resulting in deviation of the inversion results from the actual situation.

[0121] The solution logic of the present application is: through "clustering differentiation capturing spatial heterogeneity + iterative inversion adapting temporal dynamics", the spatial and temporal variation characteristics of the parameters are accurately described.

[0122] Technical bottleneck 3. Parameter inversion is ill-posed (unknown quantity > equation number), and the model is difficult to converge.

[0123] The solution logic of the present application is: through "parameter constraint reduces dimension + model simplification reduces nonlinearity", the model is realized to converge stably.

[0124] The present application couples multiple remote sensing radiation transfer models, fuses high-resolution radar and optical remote sensing images, and constructs an iterative inversion framework based on parameter multi-scale constraint. The method uses optical remote sensing data to describe the spatial heterogeneity characteristics of crops at a fine scale, and jointly constrains and solves the model parameters and soil moisture at three scales of vegetation type, local window and global range, realizes the stable inversion of soil moisture and model parameters without relying on in-situ observation.

[0125] The method of the present application performs better in soil moisture inversion in data scarce areas: compared with the uniform parameter model relying on in-situ data calibration, the inversion results (R² is 0.265-0.379, RMSE is 0.083-0.127) in the study area (1) and the study area (2) not only have more stable accuracy, but also show stronger spatial generalization ability.

[0126] The water cloud model parameters (A', B') and the surface roughness parameters (Z) have significant spatial and temporal variations. Parameter A' increases with the increase of NDVI and LAI, and parameter B' increases first and then stabilizes with the increase of NDVI and LAI.

[0127] The number of clusters and the window size have a significant impact on the inversion accuracy: the accuracy first increases and then stabilizes with the increase of k value; there is an optimal window size that balances the soil moisture spatial heterogeneity assumption and the parameter estimation stability, at which the model reconstructs the backscattering coefficient with the highest accuracy.

[0128] The soil moisture inversion method provided by one or more embodiments of the present application is based on the same idea, and the present application further provides a corresponding soil moisture inversion device, such as Figure 9

[0129] Figure 9 The soil moisture inversion device provided by the present application is shown in a schematic diagram, which comprises:

[0130] The preliminary determination module 201 is configured to determine the total backscattering coefficient of electromagnetic waves of the target region pixel by pixel according to the microwave remote sensing data of the target region; and determine the normalized vegetation index, leaf area index and average leaf inclination angle of the target region pixel by pixel according to the atmospheric apparent reflectivity data of the target region.

[0131] The clustering module 202 is configured to cluster the normalized vegetation index, leaf area index and average leaf inclination angle of the target region, and determine each vegetation type and the corresponding region of the target region.

[0132] The coupling module 203 is configured to couple the backscattering model of the soil, the dielectric constant model of the soil and the water cloud model, so as to obtain the influence relationship between the total backscattering coefficient of electromagnetic waves and the surface roughness parameter, vegetation parameter and soil moisture of the vegetation-covered ground.

[0133] The intra-class scale solving module 204 is configured to initialize the soil moisture pixel by pixel by taking the same surface roughness parameter and vegetation parameter of the same vegetation type and the same soil moisture in the same surface space window as constraints, construct an influence relationship simultaneous equation set of intra-class scale for each vegetation type, and solve the equation set to obtain the initial surface roughness parameter and vegetation parameter of each vegetation type. The surface space window corresponds to a plurality of adjacent pixels.

[0134] The window scale solving module 205 is configured to construct an influence relationship simultaneous equation set of window scale for each surface space window of the target region divided in advance, and substitute the initial surface roughness parameter and vegetation parameter of the corresponding vegetation type to solve the equation set, so as to obtain the updated soil moisture of each surface space window, and iteratively update and solve the surface roughness parameter and vegetation parameter of each vegetation type, and iteratively update and solve the soil moisture of each surface space window.

[0135] The specific limitations of the soil moisture inversion device can be referred to the limitations of the soil moisture inversion method in the foregoing, which will not be repeated here. Each module in the soil moisture inversion device described above can be realized by software, hardware and their combinations in whole or in part. Each module described above can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to each module.

[0136] ​The application further provides a computer readable storage medium, which stores a computer program. Figure 1 The application further provides a soil moisture inversion method.

[0137] The application further provides a computer device, which comprises a processor, an internal bus, a network interface, a memory and a nonvolatile memory at the hardware level, and can further comprise other hardware required by a business. The processor reads the corresponding computer program from the nonvolatile memory into the memory and then runs to implement the above-mentioned Figure 1 The application further provides a soil moisture inversion method.

[0138] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware, and the computer program can be stored in a nonvolatile computer readable storage medium. When the computer program is executed, the computer program can include the processes of the above-mentioned embodiments. In each embodiment of the application, any reference to a memory, a storage, a database or other medium can include at least one of a nonvolatile memory and a volatile memory. The nonvolatile memory can include a read-only memory (ROM), a magnetic tape, a floppy disk, a flash memory or an optical memory. The volatile memory can include a random access memory (RAM) or an external cache memory.

[0139] The technical features of the above-mentioned embodiments can be combined in any manner. In order to make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described, however, as long as the combinations of the technical features do not exist contradictions, they should be considered as the range disclosed by the application.

Claims

1. A soil moisture inversion method characterized by, The method comprises the following steps: According to the microwave remote sensing data of the target area, the total backscattering coefficient of electromagnetic waves of the target area is determined pixel by pixel; according to the atmospheric apparent reflectivity data of the target area, the normalized vegetation index, the leaf area index and the average leaf inclination angle of the target area are determined pixel by pixel; The normalized vegetation index, the leaf area index and the average leaf inclination angle of the target area are clustered to determine the vegetation types of the target area and the corresponding regions; The backscattering model of soil, the soil dielectric constant model and the water cloud model are coupled to obtain the influence relationship between the total backscattering coefficient of electromagnetic waves and the surface roughness parameter, the vegetation parameter and the soil moisture of the vegetation-covered ground surface; The soil moisture is initialized pixel by pixel by taking the same surface roughness parameter and vegetation parameter of the same vegetation type and the same soil moisture in the same surface space window as constraints, and the influence relationship equation set of the intra-class scale is constructed for each vegetation type and solved to obtain the initial surface roughness parameter and vegetation parameter of each vegetation type; the surface space window corresponds to a plurality of adjacent pixels; The influence relationship equation set of the window scale is constructed for each surface space window of the target area divided in advance, and the initial surface roughness parameter and vegetation parameter of the corresponding vegetation type are substituted to solve, to obtain the updated soil moisture of each surface space window, so as to iteratively update and solve the surface roughness parameter and vegetation parameter of each vegetation type, and iteratively update and solve the soil moisture of each surface space window.

2. The soil moisture inversion method of claim 1, wherein, The leaf area index and the average leaf inclination angle of the target area are determined pixel by pixel according to the atmospheric apparent reflectivity data of the target area, and the method comprises the following steps: The atmospheric apparent reflectivity data of the target area is taken as observation data, and the leaf area index and the average leaf inclination angle of the target area are taken as target inversion parameters to construct a search population of the target inversion parameters; The PROSAIL model is coupled with the leaf biochemical parameters and the three-dimensional geometric structure of the canopy to simulate the canopy directional reflectance, the fitness value of each individual in the search population is determined according to the observation data, the spectral residual error between the canopy directional reflectance simulated based on the search population and the observation data is minimized as an optimization target, and parameter optimization is performed on the target inversion parameters.

3. The soil moisture inversion method of claim 1, wherein, The backscattering model of soil, the soil dielectric constant model and the water cloud model are coupled to obtain the influence relationship between the total backscattering coefficient of electromagnetic waves and the surface roughness parameter, the vegetation parameter and the soil moisture of the vegetation-covered ground surface, and the method comprises the following steps: The backscattering model of soil is constructed based on the advanced integral equation model by the following formula: ； The soil dielectric constant model is constructed by the following formula: ； The water cloud model is constructed by the following formula: ， ， ； When the electromagnetic wave incidence angle , clay ratio , sand ratio , and soil bulk density are determined, the backscattering model of the soil and the soil dielectric constant model are coupled and simplified by the following equation: ， ； The coupled and simplified model is connected with the water cloud model by the following formula to obtain the influence relationship between the total backscattering coefficient of electromagnetic waves and the surface roughness parameter, the vegetation parameter and the soil moisture of the vegetation-covered ground surface: ； wherein, is the soil backscattering coefficient, is the electromagnetic wave incidence angle, is the scattering azimuth angle, is the correlation length, is the root mean square height, is the soil complex permittivity, is the electromagnetic wave frequency, denotes the high-level integral equation model; is the soil temperature, is the soil moisture, is the clay fraction, is the sand fraction, is the soil bulk density, denotes Dobson the model; is the total electromagnetic wave backscattering coefficient, is the direct vegetation backscattering coefficient, is the vegetation water content, is the vegetation two-layer attenuation factor, is the volume scattering coefficient, is the attenuation coefficient; , , and are empirical parameters, is the combined roughness.

4. The soil moisture inversion method of claim 3, wherein, The vegetation water content is represented by the following equation: ; wherein, is the vegetation water content, is the canopy water content, is a linear coefficient; According to the linear relationship between the vegetation water content and the canopy water content, the influence relationship is represented by the following formula: ， ， ； The method further comprises the following steps: The canopy water content of the target area is determined pixel by pixel according to the atmospheric apparent reflectivity data of the target area.

5. The soil moisture inversion method of claim 3, wherein, determining the empirical parameters , , and in particular comprising: The backscattering model of soil and the soil dielectric constant model are coupled by the following formula to obtain a coupled model: ； According to the values of the parameters of the predetermined coupling model of each pixel , , and , the values of the parameters and are obtained to obtain the simulation data set corresponding to and and . The simulation data set is fitted to the simplified coupling model by the least square method to obtain the empirical parameters , , and of each pixel.

6. The soil moisture inversion method of claim 5, wherein, The electromagnetic wave incidence angle determined by microwave remote sensing data of the target area; The clay ratio The sand ratio And the soil bulk density Determined by the global gridded soil information corresponding to the target area part data.

7. A soil moisture inversion apparatus, characterized by, The method comprises the following steps: The preliminary determination module is configured to determine the total backscattering coefficient of electromagnetic waves of the target region pixel by pixel according to microwave remote sensing data of the target region; and determine the normalized vegetation index, leaf area index and average leaf inclination angle of the target region pixel by pixel according to atmospheric apparent reflectivity data of the target region; The clustering module is configured to cluster the normalized vegetation index, leaf area index and average leaf inclination angle of the target region, and determine each vegetation type of the target region and the corresponding region; The coupling module is configured to couple the backscattering model of soil, the dielectric constant model of soil and the water cloud model, and obtain the influence relationship between the total backscattering coefficient of electromagnetic waves and the surface roughness parameter, the vegetation parameter and the soil moisture of the vegetation-covered ground surface; The intra-class scale solving module is configured to initialize the soil moisture pixel by pixel by taking the same surface roughness parameter and vegetation parameter of the same vegetation type and the same soil moisture in the same surface space window as constraints, construct the influence relationship of intra-class scale for each vegetation type, and solve the simultaneous equations to obtain the initial surface roughness parameter and vegetation parameter of each vegetation type; the surface space window corresponds to a plurality of adjacent pixels. The window scale solving module is configured to construct the influence relationship of window scale for each surface space window of the target region which is divided in advance, and substitute the initial surface roughness parameter and vegetation parameter of the corresponding vegetation type to solve, to obtain the updated soil moisture of each surface space window, to iteratively update and solve the surface roughness parameter and vegetation parameter of each vegetation type, and iteratively update and solve the soil moisture of each surface space window.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, and the computer program is executed by the processor to implement the method in any one of claims 1-6.

9. A computer device, comprising: The computer program is stored in the memory and executable on the processor, and the processor implements the method in any one of claims 1-6 when executing the computer program.

Images