Soil water content inversion method based on improved combination roughness

Abstract

The invention discloses a soil water content inversion method based on improved combination roughness, and relates to the field of remote sensing, and the method comprises the steps: synchronously obtaining a Sentinel-1 radar image and a Sentinel-2 optical image, and extracting different polarization backscattering coefficients, local incident angles and normalized water body indexes through preprocessing; the method comprises the following steps: generating a bare soil simulation backscattering coefficient data set, removing vegetation scattering contribution by using a water cloud model, obtaining a real bare soil backscattering coefficient, constructing a training set and a verification set containing actually measured soil water content, constructing a lookup table based on double models, and calculating the water content of the bare soil by minimizing a root-mean-square error between simulation and the real backscattering coefficient. Global search is carried out in a preset parameter space to determine an optimal earth surface root mean square height and a correlation length, a novel polynomial combination roughness is constructed, a physical correlation between the roughness and a backscattering coefficient is established, a dual-polarization empirical equation set is constructed, and simultaneous solution is carried out after parameters are optimized according to a criterion; according to the method, a more reasonable inversion result of the soil water content in a large range can be obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing image data processing, and particularly relates to a soil water content inversion method based on improved combined roughness. BACKGROUND

[0002] Soil moisture is crucial in ecological water cycle, and regulates key processes such as surface runoff, infiltration and evaporation. In agricultural production, insufficient soil moisture can easily cause soil cracking and agricultural drought, leading to large-scale crop yield reduction and threatening food security. Therefore, real-time and accurate acquisition of large-scale farmland soil moisture information has important guiding value for crop yield estimation, drought and flood warning and agricultural management.

[0003] Traditional soil moisture measurement relies on field sampling, which is inefficient and can only obtain sporadic point data. The rise of remote sensing technology provides an effective means for real-time and accurate monitoring of large-scale soil moisture. Microwave remote sensing has strong penetration capability and can achieve all-weather observation. Compared with optical remote sensing, the surface microwave scattering radiation has a significant response to changes in soil dielectric properties, and there is a deterministic physical correlation between the backscattering coefficient and soil moisture, making it a core technology for soil moisture remote sensing monitoring. At present, based on active microwave remote sensing, especially SAR soil moisture inversion model has been relatively mature and has achieved good results. However, in the SAR soil moisture inversion of bare soil area, the surface roughness has become the main source of uncertainty, and the technical means of field measurement of roughness also has many shortcomings. The traditional needle profile measurement is limited by the sampling interval, resulting in insufficient profile length, and the accuracy of the measured roughness is greatly affected. Laser profile measurement is difficult to deploy on a large scale due to the heavy scanning equipment and poor field applicability. Ground-based microwave scatterometer is equipped with accurate roughness measurement means, but its operation is complex, scale matching is difficult, and practical application is limited. In addition, scale mismatch is also a big problem in field measurement of roughness. Existing technologies (contact / non-contact) can only obtain centimeter-level local roughness parameters (s, l), while the SAR backscattering coefficient reflects the integrated surface response of the pixel (≥10m), and there is a lack of effective conversion model between microscopic measurement values and pixel scale scattering mechanism.

[0004] Field measurement of roughness is difficult to accurately measure or determine, and existing solutions including combined roughness, roughness calibration and other methods also have certain shortcomings. The traditional combined roughness form (such as Zs=s 2 / l) cannot fully characterize the multi-scale effect, and the multi-angle inversion in roughness calibration requires multiple transit data, which is time-consuming. The measurement error of the surface roughness parameter and the scale effect have become the core bottleneck restricting the accuracy of SAR soil moisture inversion. SUMMARY

[0005] The present application is just for the deficiencies of the prior art, provide a kind of soil moisture content inversion method based on improved combination roughness, by establishing the roughness model of combination of clear physical mechanism, using multi-order term coupling effective roughness parameter (s, l), realize the roughness characterization of pixel scale. Linear and nonlinear model is used to accurately characterize the coupling mechanism of soil dielectric properties, moisture content and roughness effect, and a set of dual-polarization empirical equations is established for joint inversion, which improves the inversion accuracy and has good inversion effect.

[0006] The present application is just for the deficiencies of the prior art, provide a kind of soil moisture content inversion method based on improved combination roughness, by establishing the roughness model of combination of clear physical mechanism, using multi-order term coupling effective roughness parameter (s, l), realize the roughness characterization of pixel scale. Linear and nonlinear model is used to accurately characterize the coupling mechanism of soil dielectric properties, moisture content and roughness effect, and a set of dual-polarization empirical equations is established for joint inversion, which improves the inversion accuracy and has good inversion effect.

[0007] According to the soil moisture content inversion method based on improved combination roughness provided by the present application, the following steps are included:

[0008] Step one, data acquisition and image preprocessing;

[0009] First, obtain Sentinel-1 radar image and Sentinel-2 optical image; second, perform preprocessing operations including radiation calibration, image filtering, geographic coding, etc. on Sentinel-1 data, and perform preprocessing operations such as radiation calibration, atmospheric correction, orthorectification, etc. on Sentinel-2 optical image; finally, register and resample the two image data, and normalize the incident angle of the backscattering coefficient extracted from the Sentinel-1 radar image, output the backscattering coefficient under two polarization modes Local incident angle θ and normalized water index NDWI and other characteristic parameters;

[0010] Step two, data set construction;

[0011] First, based on AIEM and OH model, simulate the influence law of soil moisture content on co-polarization (VV) and cross-polarization (VH) backscattering coefficient Under different ground roughness conditions, generate bare soil simulated backscattering coefficient, and construct simulated data set of soil moisture content-backscattering response relationship, for establishing different polarization backscattering coefficient lookup table (LUT); second, using water cloud model, quantitatively separate and remove the scattering contribution of vegetation layer to radar signal from the collected radar backscattering observation data in the study area, to obtain the real value of bare soil backscattering coefficient representing the real physical characteristics of the ground surface Finally, divide the measured sample data into training set and validation set, construct real data set containing backscattering coefficient real value and measured soil moisture content Mv, for inverting effective roughness of each sample point;

[0012] Step three, inversion of effective roughness;

[0013] First, two lookup tables (LUTs) are constructed based on the AIEM and Oh models. Within each LUT, roughness parameters within a preset step size and the measured soil moisture content Mv of the training set samples are input to obtain the simulated backscattering coefficients of bare soil with different polarizations. Roughness parameters include root mean square height *s* and correlation length *l*. Secondly, using a minimum cost function strategy, the simulated backscattering coefficients of bare soil for each training sample are obtained based on different values ​​of *s*, *l*, and *Mv*. With the corresponding backscattering coefficient of real bare soil The root mean square error (RMSE) between the simulated and true values ​​is calculated, and this RMSE value is used as an evaluation index of the accuracy of the effective roughness estimation of the training set samples. Finally, a global traversal search is performed within the range of roughness parameter values ​​within the preset step size, and the root mean square height s and related length l corresponding to the minimum calculated RMSE value are selected and established as the effective roughness representing each sample point in the entire study area.

[0014] Step 4: Construct the combined roughness;

[0015] First, based on surface scattering theory, the backscattering coefficient is calculated using the perturbation method (SPM). Expanding into a series form, the higher-order physical mechanism of surface scattering is condensed into a polynomial combination of measurable parameters (s,l). Secondly, based on existing combined roughness forms, a novel combined roughness Zs composed of polynomials is constructed, with the dominant terms of each order solution of the series serving as subterms of Zs, each subterm of Zs contributing to the normalized scattering cross section. Finally, the backscattering coefficient is established. The functional relationship between the combined roughness Zs and the soil moisture content inversion model is used to construct the soil moisture content inversion model.

[0016] Step 5: Construct a soil moisture content inversion model;

[0017] First, based on the simulation dataset generated in step two, the backscattering coefficients are analyzed. The response relationship between combined roughness Zs and soil moisture content Mv was investigated by constructing a logarithmic model and determining the relationship between combined roughness Zs and soil moisture content Mv through rigorous statistical tests. Based on the correlation essence, candidate models for soil moisture content inversion are constructed. Secondly, least squares fitting is performed on each model using training set data to determine the undetermined coefficients. The residual sum of squares (RSS) and Akaike information (AIC) of the candidate models are calculated to determine the optimal fitting parameters and establish a dual-polarization empirical equation system. Finally, the simultaneous equation system is solved for the test set samples to invert the soil moisture content of each test set sample. The inversion results are compared and analyzed with the measured soil moisture content (Mv) of the test set, and the error index is calculated to complete the accuracy verification.

[0018] As a further optimization scheme of the soil moisture content inversion method based on the improved combined roughness, the normalization processing of the incidence angle in step one is performed according to the following steps:

[0019] According to the latitude and longitude information of the sampling point, the backscattering coefficients of the two polarization modes of the sampling point are extracted from the preprocessed Sentinel-1 image, the two polarization modes include VV and VH, and the extracted backscattering coefficients are normalized by the incidence angle, and the calculation method is as follows:

[0020]

[0021] Wherein, θ is the local incidence angle; θ ref is the reference incidence angle;

[0022] As a further optimization scheme of the soil moisture content inversion method based on the improved combined roughness, the calculation method of the normalized water body index in step one is as follows:

[0023]

[0024] Wherein, R NIR is the near-infrared band reflectivity; R SWIR is the short-wave infrared band reflectivity;

[0025] As a further optimization scheme of the soil moisture content inversion method based on the improved combined roughness, the AIEM and Oh surface scattering model in step two is established according to the following formula:

[0026]

[0027] Wherein, ε is the soil dielectric constant; s is the root mean square height of the surface; l is the correlation length; ACF is the autocorrelation function; k is the free space wave number; f is the frequency of Sentinel-1 satellite; θ is the incidence angle of Sentinel-1 satellite; Mv is the soil moisture content; pp is the polarization mode (VV, VH) ;

[0028] As a further optimization scheme of the soil moisture content inversion method based on the improved combined roughness, the water cloud model removes the influence of vegetation from the radar backscattering coefficient according to the following steps:

[0029] The normalized water body index NDWI of the sampling point is extracted, and the vegetation water content VWC is calculated by using the empirical formula established by Jackson et al. according to the measured crop water content, and the empirical formula is:

[0030] VWC = 1.44NDWI 2+1.36NDWI+0.34 (5),

[0031] Then, the vegetation influence is removed from the radar backscatter coefficient using a water cloud model, the water cloud model expression is as follows:

[0032]

[0033] τ 2 (θ)=exp(-2B×VWCsecθ) (8),

[0034] Where, is the total radar backscatter coefficient; is the vegetation layer backscatter coefficient; is the bare soil backscatter coefficient; τ 2 (θ) is the two-layer attenuation factor of the vegetation layer; θ is the incident angle; A and B are parameters depending on the vegetation type;

[0035] As a further optimization scheme of the soil moisture inversion method based on the improved combined roughness, the minimum cost function inversion strategy in step three is calculated according to the following formula:

[0036]

[0037] As a further optimization scheme of the soil moisture inversion method based on the improved combined roughness, the combined roughness Zs in step four is constructed according to the following formula:

[0038] Zs=as 3 / l 2 +bs 3 / l+cs 2 / l+ds / l+e (10),

[0039] Wherein, s is the root mean square height of the ground; l is the correlation length; a, b, c, d and e are all undetermined coefficients; the least square multiple regression is used to calculate the undetermined coefficients of the empirical equation by using the data of the training set;

[0040] As a further optimization scheme of the soil moisture inversion method based on the improved combined roughness, the calculation formula of the residual sum of squares RSS and the Akaike information AIC in step five is as follows:

[0041]

[0042] Wherein, is the true value; is the fitted curve value; N is the sample size; k is the number of model parameters; the smaller the AIC value, the better the model;

[0043] As a further optimization scheme of the soil moisture content inversion method based on the improved combined roughness of the application, the empirical equation in step five is established according to the following formula:

[0044]

[0045] wherein, represents the backscattering coefficient of different polarized bare soil surface; Mv is the soil moisture content; Zs is the combined roughness; α, β, α', β', p, q, l, m, n are undetermined coefficients; the least square multiple regression is performed on the data of the training set to obtain the undetermined coefficients of the empirical equation;

[0046] Compared with the prior art, the application has the beneficial effects that: the soil moisture content inversion method based on the improved combined roughness provided by the application quantitatively eliminates the contribution of vegetation scattering from radar observation data using a water cloud model to obtain the real bare soil backscattering coefficient, which is simple and efficient in calculation. The AIEM and Oh model are combined to construct a simulation data set, and the multi-polarization information is fully utilized. The representative effective roughness parameter is inverted from the lookup table (LUT) through the least cost function strategy and global traversal search, avoiding the difficulty of traditional field measurement. The combined roughness model with clear physical mechanism is established, the effective roughness parameter (s, l) is coupled by using multi-order terms, and the pixel scale roughness representation is realized. The linear and nonlinear models are adopted to accurately represent the coupling mechanism of soil dielectric properties, moisture content and roughness effect, and the dual-polarization empirical equation set is established for joint inversion, which improves the inversion accuracy of soil moisture to a certain extent. The quantitative accuracy evaluation is realized through the independent test set, and an efficient and reliable technical scheme is provided for regional soil moisture monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 is a flowchart of the method of the application;

[0048] Figure 2 is a schematic diagram of the research area in the specific embodiment of the application, and the selected research area is a planting area in Jiangning District of Nanjing;

[0049] Figure 3 , Figure 4 is the response relationship between the improved combined roughness Zs and the backscattering coefficient in the specific embodiment of the application;

[0050] Figure 5 is the regional mapping of the soil moisture content of the planting area obtained in the specific embodiment of the application;

[0051] Figure 6 is the estimation accuracy of the empirical equation based on the improved roughness on the test set in the specific embodiment of the application. DETAILED DESCRIPTION

[0052] The technical solutions of the present application will be described in detail below with reference to the drawings.

[0053] The present application designs a soil moisture content inversion method based on improved combined roughness, as shown in Figure 1 , comprising the following steps:

[0054] Step one, data acquisition and image preprocessing;

[0055] Step 1-1, data synchronous acquisition. Obtain the Sentinel-1 SAR image (dual polarization mode) and Sentinel-2 optical image (spatial resolution 10m) passing through the same date, and highlight the planting area, extract the planting area vector boundary based on crop classification, perform spatial superposition display, see Figure 2 ;

[0056] Step 1-2, Sentinel-1 SAR image processing. Use ESASnap tool to perform absolute radiometric calibration:

[0057]

[0058] Where K is the calibration constant; α loc is the local incidence angle;

[0059] The terrain correction uses 3 arc second SRTM3 DEM (vertical accuracy ± 5m), the adaptive filter optimization uses 7x7 window size Refined Lee filter, morphological edge detection is used to avoid farmland boundary blur, geographic coding uses coordinate system WGS84 UTM Zone 50N, and the resampling method uses bicubic convolution interpolation (maintains the statistical characteristics of backscattering);

[0060] Step 1-3, Sentinel-2 optical image processing. Use Sen2Cor 2.11 processor to generate L2A level data for Sentinel-2 optical image, and calculate normalized water index NDWI;

[0061] Step 1-4, multi-source data collaborative registration. Use Sentinel-2 (10m resolution) as reference data, use SIFT feature matching algorithm to automatically extract road intersections, RMS error is controlled within 0.5 pixels, register and resample two images to the unified coordinate system, normalize the extracted backscattering coefficient according to the incidence angle, and the calculation method is as follows:

[0062]

[0063] Where, θ is the local incidence angle; θ refFor reference angle of incidence;

[0064] Obtaining the backscattering coefficient Normalized water index (NDWI) and measured soil moisture content (Mv) at the sampling points;

[0065] Step 2: Constructing the dataset

[0066] Step 2-1: Based on the advanced integral equation models AIEM and Oh, simulate the effect of soil water content on the backscattering coefficients of symmetric (VV) and cross-polarization (VH) under different surface roughness conditions. The influence of backscattering on soil was investigated, and a simulated backscattering coefficient for bare soil was generated. A simulated dataset of soil moisture content-backscattering response relationship was constructed to establish a lookup table (LUT) for backscattering coefficients of different polarizations. The AIEM model constructed a functional relationship between the same polarization backscattering coefficient and surface roughness parameters (root mean square height, correlation length), incident angle, incident wavelength, and soil dielectric constant, which can be expressed as the following formula:

[0067]

[0068] Where PQ represents the polarization mode (VV, VH), and W... n Surface roughness spectrum; R pq ε is the Fresnel reflection coefficient; r is the soil dielectric constant; s and l represent the root mean square height and correlation length, respectively; k is the wavenumber; θ is the radar incident angle;

[0069] The Oh model establishes a functional relationship between the cross-polarization backscattering coefficient and the soil dielectric constant, surface roughness, and dielectric constant, which can be expressed as the following formula:

[0070]

[0071] Where p and q are defined as the same polarization ratio and cross polarization ratio, respectively, θ is the incident angle, k is the free space wavenumber, and s and l represent the root mean square height and correlation length, respectively.

[0072] Step 2-2: Vegetation Scattering Contribution Separation. The Normalized Difference Water Index (NDWI) of the sampling points is extracted. The vegetation water content (VWC) is calculated using the empirical formula established by Jackson et al. based on measured crop water content. The empirical formula is:

[0073] VWC = 1.44NDWI 2 +1.36NDWI+0.34 (5),

[0074] Then, the vegetation influence is removed from the radar backscattering coefficient using a water cloud model. The expression for the water cloud model is as follows: An improved water cloud model is used for vegetation layer scattering separation. The expression for the water cloud model is as follows:

[0075]

[0076] τ 2 (θ) = exp(-2B x VWC sec θ) (8),

[0077] where, is the total radar backscatter coefficient; is the vegetation layer backscatter coefficient; is the bare soil backscatter coefficient; τ 2 (θ) is the two-layer attenuation factor of the vegetation layer; θ is the incidence angle 39.2°; A and B are parameters depending on the vegetation type; here A and B take the crop vegetation type A = 0.12, B = 0.18;

[0078] Step 2-3, bare soil backscatter coefficient extraction. Calculate the vegetation water content through the vegetation water content conversion model, separate the bare soil signal by using the water cloud model, and obtain the real bare soil backscatter coefficient The measured data is divided into training set and validation set, wherein the number of training set samples is 189, and the number of validation set samples is 43;

[0079] Step three, inversion of effective roughness

[0080] Step 3-1, double model lookup table (LUT) construction. The advanced integral equation model AIEM and Oh are used to construct a simulated data set of soil water content-backscatter response relationship, which is used to establish a lookup table (LUT) of different polarization backscatter coefficients. The surface roughness parameter range is set as follows: the root mean square height s is 0.1 cm-3 cm (step length is 0.1 cm); the correlation length l is 5 cm-70 cm (step length is 1 cm); the soil water content Mv gradient is set to 0%-50% (volume water content, step length is 1%); the incidence angle is set to the standard imaging angle θ = 39.2° of the Sentinel-1 satellite data in the study area; the frequency f = 5.405 GHz;

[0081] Step 3-2, generate lookup table (LUT). A three-dimensional lookup table is constructed, the main input parameters are the root mean square height s; the correlation length l; the soil water content Mv; the input fixed imaging angle θ = 39.2°; the frequency f = 5.405 GHz; and the output target is the different polarization backscatter coefficient values corresponding to the training set sample points

[0082] Step 3-3, polarization scattering response simulation. Based on the AIEM model and the Oh model, the soil water content Mv and the backscatter coefficient The response relationship between the simulated backscattering coefficient and the real backscattering coefficient of bare soil is established by calculating the volume scattering coefficient of bare soil

[0083] Step 3-4, strategy of minimum cost function, according to different s, l and Mv values, the simulated backscattering coefficient of bare soil of each training sample is obtained and the corresponding real backscattering coefficient of bare soil The root mean square error (RMSE) between the simulated value and the real value is calculated, and the RMSE value is used as an evaluation index of the effective roughness estimation accuracy of the training set samples, and the formula is as follows:

[0084]

[0085] Step 3-5, finally, the global traversal search is performed in the roughness parameter value range within the preset step range, and the root mean square height s and the correlation length l corresponding to the minimum RMSE value calculated are selected and established as the effective roughness representing each sample point in the whole study area;

[0086] Step four, construct the combined roughness:

[0087] Step 4-1, scattering solution is expanded into series. Based on the surface scattering theory, the backscattering coefficient is expanded into series form by using the perturbation method (SPM), and the high-order physical mechanism of surface scattering is condensed into the polynomial combination of measurable parameters (s, l), and the series form is as follows:

[0088]

[0089] The expressions of each order solution are given below (taking scalar approximation as an example, and ignoring constant factors):

[0090] The first order solution (n=1) corresponds to single scattering (Bragg scattering):

[0091]

[0092] Where, p, q represent polarization (VV, VH); k is the wave number; θ is the incidence angle; |f pq | 2 is the polarization coupling coefficient; W(2ksinθ, 0) is the surface roughness spectrum (i.e. the value of the power spectral density function of surface height in the scattering vector direction), and the simplified form of the dominant term is as follows:

[0093]

[0094] The second order solution (n=2) corresponds to double scattering (including cross polarization):

[0095]

[0096] where p, q represent polarization modes (VV, VH); k is the wave number; denotes the scattering physical mechanism, S (1) denotes double surface reflection, S (2) denotes surface-volume interaction, S (3) denotes cross-polarization coupling, S (4) denotes edge diffraction, S (5) denotes cavity scattering, S (6) denotes angular reflection; W(k1)W(k2) is the surface roughness spectrum (i.e. the value of the power spectral density function of the surface height in the direction of the scattering vector), the dominant term is simplified as follows:

[0097]

[0098] is actually simplified as to reduce the dimension

[0099] The third-order solution (n = 3) corresponds to triple scattering, and the expression is more complex, usually containing multiple terms, here a simplified form is given:

[0100]

[0101] Step 4-2, polynomial structure design. Based on the existing combined roughness form, a new combined roughness Zs composed of polynomials is constructed, and the dominant term of each order solution of the series is taken as a sub-term of Zs, each sub-term of Zs is a normalized scattering cross-section contribution, and the form is as follows:

[0102] Zs = as 3 / l 2 + bs 3 / l + cs 2 / l + ds / l + e (16),

[0103] where s is the root mean square height of the ground; l is the correlation length; a, b, c, d, e are all fitting coefficients;

[0104] Step five, building soil moisture inversion model

[0105] Step 5-1, correlation type decision. Based on the advanced integral equation model AIEM and Oh, a simulation data set of soil moisture-backscattering response relationship is constructed, and the response relationship between backscattering coefficient and soil moisture Mv is analyzed, a logarithmic model is constructed, and the general relationship between backscattering coefficient and soil moisture Mv can be expressed as: The backscattering coefficient The response relationship between the combination roughness Zs and the correlation between Zs and The correlation between Zs and the correlation between Zs and

[0106] Log-linear model:

[0107] Log-linear model:

[0108] Where PQ represents the polarization mode (VV, VH); α, β, α', β', p, q, l, m, n are undetermined coefficients;

[0109] Step 5-2, coefficient fitting.

[0110] First, use the 189 training set data to perform least squares fitting on each model, and the input data includes the real bare soil backscattering coefficient The measured soil moisture Mv, the effective roughness s and l obtained by inversion in step three, and the calculated Zs are normalized to limit Zs ∈ [0.20, 1.00]; the incident angle θ = 39.2°; the frequency f = 5.4GHz, and the undetermined coefficients are as follows:

[0111] Log-linear model (VV polarization): α = -3.46, β = 1.28, p = 10.85, q

[0112] = -7.92, a = 0.147, b = -0.082, c = 1.236, d = 0.574, e

[0113] = 0.021

[0114] Log-linear model (VV polarization): α = -3.46, β = 1.28, p = 10.85, q

[0115] = 1.12, n = -5.34, a = 0.028, b = -0.193, c = 0.421, d

[0116] = 1.087, e = 0.073

[0117] Log-linear model (VV polarization): α = -3.46, β = 1.28, p = 10.85, q

[0118] = -12.25, a = 0.281, b = -0.112, c = 0.849, d = 0.757, e

[0119] = 0.018

[0120] Log-linear model (VV polarization): a = 0.025, b = -0.176, c = 0.398, d = 1.124, e = 0.068

[0121] Log-linear model (VV polarization): a = 0.025, b = -0.176, c = 0.398, d = 1.124, e = 0.068

[0122] Log-linear model (VV polarization): a = 0.025, b = -0.176, c = 0.398, d = 1.124, e = 0.068

[0123] Secondly, the residual sum of squares RSS and Akaike information AIC of the four models are calculated, which are as follows:

[0124]

[0125] Where, is the true value; is the fitted curve value; N is the sample size; k is the number of model parameters (k = 9 for linear model, k = 10 for power-log model), the smaller the AIC value, the better the model;

[0126] Finally, the RSS and AIC of the log-linear model under VV polarization are calculated as 36.85 dB 2 and -276.3, respectively, and the RSS and AIC of the log-power-log model are 38.72 dB 2 and -271.6, respectively; the RSS and AIC of the log-linear model under VH polarization are 48.63 dB 2 and -227.8, respectively, and the RSS and AIC of the log-power-log model are 44.17 dB 2 and -241.5, respectively; therefore, the log-linear model is used under VV polarization, and the log-power-log model is used under VH polarization, and the specific form of the empirical equation set is as follows:

[0127]

[0128] Figure 3 , Figure 4 The fitting results of part of the sample points (20) under VV polarization and VH polarization with fixed soil water content (Mv = 25%) are shown in the following table:

[0129] Step 5-3, soil water content inversion and accuracy verification;

[0130] The optimal parameters determined in step 5-2 are substituted into the empirical inversion equation set to construct a simultaneous equation set for the test set samples, and the Newton-Raphson iteration method is used for numerical solution. The inversion soil water content value is output for each test set sample, and the soil water content regional mapping of the study area is completed, as shown in the following table: Figure 5; The inversion result is compared with the measured soil water content Mv of the test subset, error indexes are calculated, RMSE and R 2 , etc. are calculated by comparison with the measured values to complete precision verification, see Figure 6 , and quantitative verification of the model inversion precision is completed.

[0131] It should be noted that the above content only illustrates the technical idea of the present application and cannot limit the protection scope of the present application. For ordinary skilled persons in the art, some improvements and refinements can be made without departing from the principles of the present application, and these improvements and refinements fall within the protection scope of the claims of the present application.

Claims

1. A soil moisture content inversion method based on improved combined roughness, characterized by, Comprise the following steps: Step one, data acquisition and image preprocessing; Firstly, Sentinel-1 radar image and Sentinel-2 optical image are acquired, secondly, the Sentinel-1 data and the Sentinel-2 optical image are preprocessed, and finally, the two image data are registered and resampled, and the backscattering coefficient extracted from the Sentinel-1 radar image is normalized by incident angle to output the cross-polarization backscattering coefficient Cross-polarization backscattering coefficient Local incident angle θ and normalized water index NDWI characteristic parameters Step two, building dataset; Firstly, based on the advanced integral equation model AIEM and Oh model, the influence of soil moisture on the monostatic backscattering coefficient and the cross-polarized backscattering coefficient under different surface roughness conditions is simulated to generate the simulated backscattering coefficient of bare soil, and the simulated dataset of the relationship between soil moisture and backscattering response is constructed to establish the lookup table LUT of different polarized backscattering coefficients. Secondly, the water cloud model is used to quantitatively separate and remove the scattering contribution of the vegetation layer to the radar signal from the radar backscatter observation data collected in the study area, to obtain the real value of the co-polarized bare soil backscatter coefficient representing the real physical characteristics of the ground surface and the real value of the cross-polarized bare soil backscatter coefficient Finally, the measured sample data is divided into a training set and a validation set, and a real data set containing the real value of the backscatter coefficient and the measured soil moisture Mv is constructed to invert the effective roughness of each sample point. Step three, inversion of effective roughness; Firstly, two look-up tables (LUTs) are constructed based on the AIEM model and Oh model. In the LUTs, the input is the roughness parameter in the preset step range and the measured soil moisture Mv of the training set samples, and the output is the co-polarized bare soil simulated backscattering coefficient and the cross-polarized bare soil simulated backscattering coefficient The roughness parameter includes the root mean square height s and the correlation length l. Secondly, the strategy of the minimum cost function is adopted. According to the values of s, l and Mv, the co-polarized bare soil simulated backscattering coefficient and the cross-polarized bare soil simulated backscattering coefficient of each training sample are obtained. The co-polarized real bare soil backscattering coefficient and the cross-polarized real bare soil backscattering coefficient are obtained. The root mean square error (RMSE) between the simulated value and the real value is calculated, and the RMSE value is used as an evaluation index of the effective roughness estimation accuracy of the training set samples. Finally, the roughness parameter is globally searched in the preset step range, and the root mean square height s and the correlation length l corresponding to the minimum RMSE value are selected and established as the effective roughness representing each sample point in the whole study area. Step four, building combined roughness; Firstly, based on the surface scattering theory, the monostatic backscattering coefficient and cross-polarization are expanded into series form, and the high-order physical mechanism of surface scattering is condensed into the polynomial combination of measurable parameters (s, l); Secondly, based on the existing combined roughness form, a new combined roughness Zs composed of polynomials is constructed, and the dominant term of each order solution of the series is taken as a sub-term of Zs. Each sub-term of Zs is contributed by the normalized scattering cross-section; Finally, the functional relationship between the monostatic backscattering coefficient and cross-polarization and the combined roughness Zs is established, which is used to construct the soil moisture inversion model; Step five, building soil moisture inversion model; Firstly, based on the simulated data set generated in step two, the response relationship between the co-polarized backscatter coefficient and cross-polarized backscatter coefficient and soil moisture content Mv is analyzed, a logarithmic model is constructed, and the correlation between the combined roughness Zs and the co-polarized backscatter coefficient and cross-polarized backscatter coefficient is determined by strict statistical test to construct the soil moisture content inversion candidate model; secondly, the least square fitting is performed on each model using the training set data to obtain the undetermined coefficients, the residual sum of squares RSS and Akaike information AIC of the inversion candidate model are calculated, the optimal fitting parameters are determined, and the dual-polarization empirical equation set is established; finally, the soil moisture content of each test set sample is obtained by solving the equation set, the inversion results are compared with the measured soil moisture content Mv of the test set, the error index is calculated, and the precision verification is completed.

2. The soil moisture content inversion method based on improved combined roughness according to claim 1, characterized in that, The normalization processing of incident angle in step one is carried out according to the following steps: According to the latitude and longitude information of the sampling point, the backscattering coefficients of two polarization modes of the sampling point are extracted from the preprocessed Sentinel-1 image, the two polarization modes include co-polarization VV and cross-polarization VH, and the extracted backscattering coefficients are normalized to incident angle, and the calculation method is as follows: where θ is the local angle of incidence; θ ref is the reference angle of incidence.

3. The soil moisture content inversion method based on improved combined roughness according to claim 1, characterized in that, The calculation method of normalized water index NDWI in step one is as follows: Wherein, R NIR is the reflectivity in the near-infrared band; R SWIR is the reflectivity in the short-wave infrared band.

4. The soil moisture content inversion method based on improved combined roughness according to claim 1, characterized in that, In step two, AIEM and Oh surface scattering model are established according to the following formula: Wherein, epsilon is the soil dielectric constant, s is the root mean square height of the ground, l is the correlation length, ACF is the autocorrelation function, k is the free space wave number, f is the frequency of Sentinel-1 satellite, theta is the incident angle of Sentinel-1 satellite, Mv is the soil moisture content, and pp is the polarization mode.

5. The soil moisture content inversion method based on improved combined roughness according to claim 1, characterized in that, In step two, the water cloud model removes the influence of vegetation from the radar backscattering coefficient according to the following steps: The normalized water index NDWI of the sampling point is extracted, and the vegetation water content VWC is calculated by using the empirical formula established according to the measured crop water content, and the empirical formula is as follows: VWC = 1.44NDWI 2 + 1.36NDWI + 0.34 (5), Then, the water cloud model is used to remove the influence of vegetation from the radar backscattering coefficient, and the expression of the water cloud model is as follows: τ 2 (0) = exp(-2B x VWCsec0) (8), wherein, is the total radar backscatter coefficient, is the vegetation layer backscatter coefficient, is the bare soil backscatter coefficient, τ 2 (θ) is the two-layer attenuation factor for the vegetation layer, θ is the incidence angle, and A and B are parameters depending on the vegetation type.

6. The soil moisture content inversion method based on improved combined roughness according to claim 1, characterized in that, In step three, the minimum cost function inversion strategy is calculated according to the following formula:

7. The soil moisture content inversion method based on improved combined roughness according to claim 1, characterized in that, The combined roughness Zs in step four is built according to the following formula: Zs = as 3 / 1 2 + bs 3 / 1 + cs 2 / 1 + ds / 1 + e (10), Wherein, s is the root mean square height of the ground, l is the correlation length, a, b, c, d, e are all undetermined coefficients, and the least square multiple regression is used to calculate the undetermined coefficients of the empirical equation.

8. The soil moisture content inversion method based on improved combined roughness according to claim 1, characterized in that, The calculation formula of residual sum of squares RSS and Akaike information AIC in step five is as follows: wherein, is the true value, is the fitted curve value, N is the sample size, k is the number of model parameters, and the smaller the AlC value, the better the model.

9. The soil moisture content inversion method based on improved combined roughness according to claim 1, characterized in that, The empirical equation in step five is established according to the following formula: wherein, wherein, Mv is the soil moisture content, Zs is the combined roughness, a, b, a', b', p, q, l, m, n are undetermined coefficients, and the undetermined coefficients of the empirical equation can be solved by using the data of the training set for least squares multiple regression.

Images