A method for filling gaps in spaceborne polarimetric SMAP-R data based on multipolarimetric SAR data

Abstract

This invention discloses a method for filling gaps in spaceborne SMAP-R polarimetric data based on multi-polarization SAR data, comprising the following steps: acquiring SMAP-R, SMAP, GF-3, Sentinel-1, MODIS, and SR TM 30m DEM data; preprocessing the above data to obtain total intensity reflectivity, normalized Stokes parameters, and backscattering coefficients; unifying the spatial resolution of the data and dividing the dataset; constructing a data supplementation model for SMAP-R polarization parameters based on multi-polarization SAR backscattering modeling using ensemble learning; verifying the parameter inversion performance using an ANN model; and finally filling in the missing grid data. This method models SMAP-R polarimetric data using SAR backscattering data and verifies it through secondary inversion, effectively filling the gaps in SMAP-R spatial data and improving the spatial resolution of its inverted surface parameters.

Description

Technical Field

[0001] This invention belongs to the field of space downscaling of spaceborne polarized GNSS-R technology. Specifically, it involves using multi-source data fusion to supplement and enhance SMAP-R data, and designing different model schemes for different secondary inversion requirements. Background Technology

[0002] The SMAP mission provides complementary L-band measurements using radiometers and radar, achieving geophysical measurements with fine spatial resolution. However, in 2015, the radar transmitter ceased operation due to a malfunction. By switching the bandpass center frequency of its radar receiver to the GPS L2C band, and after more than seven years of continuous measurements, SMAP-R data has become the largest polarimetric GNSS-R dataset to date. SMAP can receive reflected signals from GNSS at the Earth's surface via V and H channels and is the only dataset to provide polarimetric measurements. A unique feature of SMAP-R compared to other GNSS-R missions is its high-gain antenna and its hybrid compact polarization (HCP) capability. SMAP-R data is collected as in-phase orthogonal (I / Q) samples, processed to extract forward scattering measurements from GPS reflections, and provided in the form of four Stokes parameters (S0, S1, S2, S3) and total intensity reflectivity (Γ0) calibrated by S0. The SMAP-R dataset has shown great potential in surface parameter (vegetation, soil moisture, polar) inversion studies. While SMAP-R boasts high temporal resolution, its weak reflection signals, uneven distribution of reflection points, and incoherent accumulation times, coupled with the sparse composition of observation points, result in numerous spatial gaps or incomplete regions. The average spatial resolution of SMAP-R is 2.8 km for wetlands, 5.3 km for arid lands, 12.1 km for low to medium vegetation, and 26.6 km for very dense vegetation. Therefore, to obtain higher spatial resolution surface parameter inversion results, it is necessary to supplement the SMAP-R data with other data. SAR data can provide high-resolution images, and a certain relationship can be established between L-band radiometers and SAR through brightness temperature. To fully utilize SMAP-R's ability to invert Earth parameters while maintaining its spatial resolution, it is necessary to supplement its sparse data to achieve high spatiotemporal resolution inversion of soil moisture, vegetation optical thickness, vegetation water content, polar parameters, etc. Therefore, the technical solution proposed in this invention is to supplement SMAP-R data by modeling forward scattering (SMAP-R) using backscattering (SAR) to solve the problem of data scarcity of polarized SMAP-R at medium (or small) scales.

[0003] In view of this, the present invention is proposed. Summary of the Invention

[0004] This invention aims to propose a method for filling gaps in spaceborne polarimetric SMAP-R data based on multipolar SAR data. This method uses SAR backscattering data to model SMAP-R polarimetric data, and then verifies the performance of the data through secondary inversion, thereby supplementing the missing data in SMAP-R space and improving the spatial resolution of SMAP-R inverted surface parameters.

[0005] This invention provides the following technical solution:

[0006] A method for filling gaps in spaceborne polarimetric SMAP-R data based on multipolarimetric SAR data includes the following steps:

[0007] Step S1: Obtain SMAP-R, SMAP, GF-3, Sentinel-1, MODIS, and SRTM 30m DEM data.

[0008] Step S2 involves preprocessing the SMAP-R, SAR, and auxiliary data to obtain the total intensity reflectivity, normalized Stokes parameters, and dual-polarization (Sentinel-1) and full-polarization (GF-3) backscattering coefficients.

[0009] Step S3: Unify the spatial resolution of all data and divide the data.

[0010] Step S4: Model construction based on the data supplementation method for SMAP-R polarization parameters in multi-polarization SAR backscattering modeling using ensemble learning.

[0011] Step S5: Performance verification of parameter inversion based on the ANN model.

[0012] Step S6: Fill in the missing grid data.

[0013] Preferably, the data preparation in step S1 includes the following:

[0014] Download SMAP-R, SMAP, GF-3, Sentinel-1, MODIS, and SRT M 30m DEM data from official websites and institutions.

[0015] Preferably, step S2 involves preprocessing the SMAP-R, SAR, and auxiliary data to obtain the total intensity reflectivity, normalized Stokes parameters, and dual-polarization (Sentinel-1) and full-polarization (GF-3) backscattering coefficients. This includes the following sub-steps:

[0016] Step S2.1: First, QGIS and SNAP / PolSARpro are used in conjunction with DEM data to perform radiometric correction, filtering, geometric correction, and georegistration on SAR data (GF-3, Sentinel-1).

[0017] Step S2.2: Perform Freeman-Durden and Cloude-Pottier decomposition on the scattering matrix and eigenvalues ​​of the GF-3 fully polarized data.

[0018] The Freeman-Durden method is based on the incoherent decomposition of the model. It models the polarization coherence matrix as contributions from three scattering mechanisms: surface scattering, double bounce scattering, and volumetric scattering. In SAR images, each pixel is represented by a 3x3 coherence matrix T.

[0019]

[0020] The superscript * indicates conjugate.

[0021] Freeman represents the coherence matrix as:

[0022] T = P s T sueface +P d T double +P v T volume (2)

[0023] Where P s P d and P v T corresponds to the power of each scattering component. sueface T double T volume These are surface scattering, double bounce scattering, and volume scattering, respectively, and their sum equals Span:

[0024] Span = P s +P d +P v =T 11 +T 22 +T 33 (3)

[0025] Surface scattering is modeled by a first-order Bragg surface scatterer as follows:

[0026]

[0027] In the formula, β is the ratio of horizontally polarized waves to vertically polarized waves in the polarization scattering matrix components under surface scattering.

[0028] The double bounce scattering model is as follows:

[0029]

[0030] In the formula, χ is the ratio of the horizontally polarized wave to the vertically polarized wave in the polarization scattering matrix components under double reflection scattering.

[0031] Volume scattering is modeled as follows:

[0032]

[0033] Cloude–Pottier is based on eigenvalue analysis of the polarization coherence matrix, and the decomposition results are angle (α), entropy (H), and anisotropy (A).

[0034] α=p1a1+p2a2+p3a3 (7)

[0035]

[0036] In the formula, a i and p i (i = 1, 2, 3) represent the types and probabilities of the three scattering mechanisms. α, H, and A can characterize the complexity and dominance of the scattering process.

[0037] Step S2.3: Obtain the total intensity reflectance (Γ0) and the normalized first... second third Stokes parameters are calculated using the following formula:

[0038]

[0039] In the formula, σ(τ, f) is the bistatic radar cross section (BRCS) calibrated by the first Stokes parameter, and R T R is the distance from the transmitter to the point of reflection on the mirror. R τ is the distance from the specular reflection point to the receiver; (τ, f) represents the delayed Doppler frequency, and the formula above calculates the integral within 5 delays and ±1.8kHz frames.

[0040] Step S2.4: Extract other auxiliary data parameters: extract soil moisture (SM), surface roughness (Roughness), vegetation water content (VWC), vegetation optical thickness (VOD), and surface temperature (RC) from SMAP data; and enhanced vegetation index (EVI) and normalized difference vegetation index (NDVI) from MODIS data.

[0041] Preferably, step S3, which unifies the spatial resolution of all data and performs data partitioning, includes the following:

[0042] The selected region's SMAP-R, SMAP, MODIS, GF-3, and Sentinel-1 data were resampled using bilinear interpolation and gridded into an EASE Grid 2.09km x 9km grid, generating a dataset with the same number of grid points as the target region. A grid search method was used to average the SMAP-R reflection points falling within the same grid, and grids without stored SMAP-R reflection points were labeled. The dataset was randomly divided: 60% for training, 25% for testing, and 15% for validation.

[0043] Preferably, the model construction of the data supplementation method for SMAP-R polarization parameters based on ensemble learning for multi-polarization SAR backscattering modeling in step S4 includes the following:

[0044] The gradient-boosting ensemble machine learning algorithm CatBoost uses full polarization (GF-3) and dual polarization (Sentinel-1) backscattering coefficients and auxiliary data to evaluate the SMAP-R polarization parameters (total intensity reflectivity Γ0 and the three normalized Stokes parameters). Modeling is performed using the following steps: All filtered and normalized data, including (GF-3 and Sentinel-1 backscattering coefficients; SMAP SM, RC, VWC, VOD, Roughness; MODIS NDVI, EVI) as input, are used with SMAP-R polarization parameters. As output, different input features are combined according to different secondary inversion requirements. After training, SMAP-R parameters with different secondary inversion capabilities are obtained from the SAR backscattering coefficients.

[0045] Preferably, the parameter inversion performance verification based on the ANN model in step S5 includes the following:

[0046] The parameters with different second-order inversion capabilities obtained from step S4 are... The corresponding auxiliary parameters are input into the ANN model for training. The results are compared and validated with SMAP data. The selected accuracy validation metrics are correlation coefficient (CC) and root mean square error (RMSE).

[0047] Preferably, the missing grid data filling in step S6 includes the following: after modeling is completed, the missing grids in the selected area are filled with the inverted data based on the previously marked missing grids. Attached Figure Description

[0048] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the implementation cases will be briefly introduced below.

[0049] Figure 1 This is a flowchart illustrating an embodiment of the present invention.

[0050] Figure 2 This is a schematic diagram of the structure of the ANN model in an embodiment of the present invention.

[0051] Figure 3 The accompanying figure shows the comparison results of the vegetation optical depth inversion solution proposed in this embodiment of the invention with reference data. The figure is in color so that the comparison results can be more intuitively presented.

[0052] Figure 4 The accompanying figure shows the comparison results of vegetation water content retrieved by the technical solution in this embodiment of the invention with reference data. The figure is in color so that the comparison results can be more intuitively presented.

[0053] Figure 5 The accompanying figure shows the comparison results of soil moisture inversion in the technical solution proposed in this embodiment of the invention with reference data. The figure is in color so that the comparison results can be more intuitively presented. Detailed Implementation

[0054] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0055] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0056] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments. The present invention provides the following embodiments.

[0057] Example 1

[0058] To verify the feasibility and reliability of the SMAP-R data retrieved in this invention, the invention first reconstructs SMAP-R parameters from SAR data through ensemble learning. Then, it uses the reconstructed data to retrieve vegetation optical depth, vegetation water content, and soil moisture. Specific implementation examples are as follows: SMAP-R data (SM, Roughness, RC, VOD, VWC) and MODIS data (MOD13C2 NDVI, EVI) were downloaded from relevant public websites, covering the period from January 1, 2017 to December 31, 2020. SMAP data was used as reference data. GF-3 fully polarized surface scattering, double bounce scattering, volume scattering coefficients, angles, entropy, and anisotropy were used; Sentinel-1 vertical polarization and cross-polarization backscattering coefficients were used; SMAP SM, Roughness, VWC, RC, and VOD were used; MODIS EVI and NDVI data were used as input data for model inversion; and SRTM 30mDEM was used for geometric correction of SAR data processing. The temporal and spatial resolution information of the six types of data used in this invention, along with the parameters used, are shown in Table 1:

[0059] Table 1. Temporal and spatial resolution information of the six types of data used in this invention.

[0060]

[0061] A method for filling gaps in spaceborne polarimetric SMAP-R data based on multipolarimetric SAR data includes the following steps:

[0062] Step S1: Obtain SMAP-R, SMAP, GF-3, Sentinel-1, MODIS, and SRTM 30m DEM data.

[0063] Step S2 involves preprocessing the SMAP-R, SAR, and auxiliary data to obtain the total intensity reflectivity, normalized Stokes parameters, and dual-polarization (Sentinel-1) and full-polarization (GF-3) backscattering coefficients.

[0064] Step S3: Unify the spatial resolution of all data and divide the data.

[0065] Step S4: Model construction based on the data supplementation method for SMAP-R polarization parameters in multi-polarization SAR backscattering modeling using ensemble learning.

[0066] Step S5: Fill in the missing grid data.

[0067] Step S6: Performance verification of parameter inversion based on the ANN model.

[0068] As one implementation of this embodiment, step S2 includes the following sub-steps:

[0069] Step S2.1, SAR data (GF-3, Sentinel-1) preprocessing. Import Sentinel-1 data into SNAP, use Calibration to select the backscattering coefficients of the VV and VH channels, then use the Refined Lee filter to remove speckle and filter the data, selecting a 5x5 window. Then input the stitched DEM for geometric correction, selecting WGS84 as the coordinate system. Import GF-3 data into PolSARpro, select Calibration for polarization channel amplitude correction, and select normalized output. Then use the Refined Lee filter to filter the covariance matrix. Input the filtered data into SNAP, use the DEM for geometric correction, and select WGS84 as the output coordinate system.

[0070] Step S2.2: Perform Freeman-Durden and Cloude-Pottier decomposition on the scattering matrix and eigenvalues ​​of the GF-3 fully polarized data.

[0071] The Freeman-Durden method is based on the incoherent decomposition of the model. It models the polarization coherence matrix as contributions from three scattering mechanisms: surface scattering, double bounce scattering, and volumetric scattering. In SAR images, each pixel is represented by a 3x3 coherence matrix T.

[0072]

[0073] The superscript * indicates conjugate.

[0074] Freeman represents the coherence matrix as:

[0075] T = P s T sueface +P d T double +P v T volume (2)

[0076] Where P s P d and P v T corresponds to the power of each scattering component. sueface T double T volume These are surface scattering, double bounce scattering, and volume scattering, respectively, and their sum equals Span:

[0077] Span = P s +P d +P v =T11 +T 22 +T 33 (3)

[0078] Surface scattering is modeled by a first-order Bragg surface scatterer as follows:

[0079]

[0080] In the formula, β is the ratio of horizontally polarized waves to vertically polarized waves in the polarization scattering matrix components under surface scattering.

[0081] The double bounce scattering model is as follows:

[0082]

[0083] In the formula, χ is the ratio of the horizontally polarized wave to the vertically polarized wave in the polarization scattering matrix components under double reflection scattering.

[0084] Volume scattering is modeled as follows:

[0085]

[0086] Cloude–Pottier is based on eigenvalue analysis of the polarization coherence matrix, and the decomposition results are angle (α), entropy (H), and anisotropy (A).

[0087] α=p1a1+p2a2+p3a3 (7)

[0088]

[0089] In the formula, a i and p i (i = 1, 2, 3) represent the types and probabilities of the three scattering mechanisms. α, H, and A can characterize the complexity and dominance of the scattering process.

[0090] Step S2.3: Obtain the total intensity reflectance (Γ0) and the normalized first... second third Stokes parameters are calculated using the following formula:

[0091]

[0092] In the formula, σ(τ, f) is the bistatic radar cross section (BRCS) calibrated by the first Stokes parameter, and R T R is the distance from the transmitter to the point of reflection on the mirror. R τ is the distance from the specular reflection point to the receiver; (τ, f) represents the delayed Doppler frequency, and the formula above calculates the integral within 5 delays and ±1.8kHz frames.

[0093] Step S2.4, Preprocessing of other auxiliary data parameters: Extract SM, Roughness, VWC, VOD, and RC from SMAP data; and EVI and NDVI from MODIS data. Perform quality control on the above data: assign NAN to SM values ​​less than 0.02 and equal to -9999; assign NAN to VWC, VOD, Roughness, and RC values ​​equal to -9999; and assign NAN to quality indicators equal to 0 and 8 using the quality indicators of SMAP data itself. Assign NAN to NDVI values ​​less than -2000 or greater than 10000; and filter out data with non-zero bits 0-1 using the quality indicators of MODIS data itself.

[0094] As one implementation of this embodiment, step S3 includes the following:

[0095] The SMAP-R, SMAP, MODIS, GF-3, and Sentinel-1 data for the selected region were resampled using bilinear interpolation and gridded into an EASE Grid 2.09km x 9km grid, generating a dataset with the same number of grid points as the target region. A grid search method was used to average the SMAP-R reflection points falling within the same grid, and grids without stored SMAP-R reflection points were labeled. The dataset was randomly divided: 60% for training, 25% for testing, and 15% for validation.

[0096] As one implementation of this embodiment, step S4 includes the following:

[0097] The SMAP-R polarization parameters (total intensity reflectivity and three normalized Stokes parameters) were modeled using backscattering coefficients of full polarization (GF-3) and dual polarization (Sentinel-1) and auxiliary data, respectively. The chosen ensemble learning method was CatBoost. The specific training process was as follows: All filtered and normalized data, including (GF-3 and Sentinel-1 backscattering coefficients; SMAP SM, Roughness, VWC, VOD, RC; MODIS NDVI, EVI) as input, were used to model the SMAP-R polarization parameters (total intensity reflectivity and three normalized Stokes parameters). As output.

[0098] Based on the different requirements of the second inversion of the filled data, and to prevent overfitting, the input scheme is designed as follows:

[0099]

[0100] As one implementation of this embodiment, step S5 includes the following sub-steps:

[0101] Step S5.1 involves taking the SMAP-R parameters with different second-order inversion capabilities obtained from the SAR backscattering coefficients after training in step S4. The input is fed into an ANN network for a second inversion, with the target parameters being VWC, VOD, and SM.

[0102] Step S5.2, for the ANN model, as shown in the appendix. Figure 2 As shown, the model consists of 8 fully connected layers, with the number of neurons in each layer being 32, 64, 64, 32, 64, 128, 128, and 64 respectively. The ReLU activation function is used uniformly across layers 1-6, while layers 7-8 do not use an activation function. A Dropout mechanism is added in between to suppress overfitting, with a Dropout ratio of 0.1. The output layer uses a Linear function to enhance output stability. The Adam optimizer is used as the optimization algorithm, with an initial learning rate of 0.001. The mean squared error (MSE) loss function is used, defined as follows: where...

[0103]

[0104] In the formula, y i This represents the observation value of the i-th sample. This represents the model's predicted value, where n is the total number of samples.

[0105] Step S5.3 involves comparing and validating the results with the SMAP data. During training, if the validation set error does not significantly decrease over 10 consecutive iterations, training is terminated early. The overall model, after constructing training samples for the complete region, is applied to predictively fill in missing grid lines, achieving a complete reconstruction of the spatial distribution of the SMAP-R inversion parameters.

[0106] Step S5.4: The selected accuracy indicators are the correlation coefficient (CC) and the root mean square error (RMSE).

[0107]

[0108] In the formula, n is the number of data samples. The true values ​​of the four parameters of SMAP-R are y i The predicted value of the model. They are respectively and y i The average value.

[0109] As one implementation of this embodiment, step S6 includes the following: after completing modeling and verification, fill the missing grids in the selected area with the inverted data according to the previously marked missing grids.

[0110] The final result of the second inversion using ANN is as follows: Figures 3-5 As shown, scatter plots of vegetation optical thickness, vegetation water content, and soil moisture retrieved from SMAP-R data, along with correlation coefficients and root mean square errors, are presented.

[0111] As can be seen from the figure, there is a strong correlation between the three inversion results and the validation data (inverted vegetation optical thickness CC = 0.89, vegetation water content CC = 0.88, and soil moisture CC = 0.83), with the inverted soil moisture showing the lowest RMSE of 0.054 cm. 3 / cm 3 The RMSE of vegetation optical thickness was 0.076, and the RMSE of vegetation water content was 0.645 kg / m³. 2 The above results demonstrate that SMAP-R data modeled from SAR backscatter data can successfully invert surface parameters with good accuracy. It can serve as a substitute for missing data, thereby filling the spatial gaps in SMAP-R inversion parameters and improving spatial resolution.

[0112] The above description, in conjunction with specific embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, several simple deductions or substitutions can be made without departing from the concept of the present invention, and all such deductions or substitutions should be considered to fall within the scope of protection defined by the claims submitted herein.

Claims

1. A method for filling gaps in spaceborne polarimetric SMAP-R data based on multi-polarimetric SAR data, characterized in that, Includes the following steps: Step S1: Acquire SMAP-R data, SMAP data, GF-3 fully polarimetric SAR data, Sentinel-1 dual polarimetric SAR data, MODIS vegetation index data, and SRTM 30m DEM topographic data; Step S2: Preprocess the SMAP-R data, SMAP data, GF-3 fully polarimetric SAR data, Sentinel-1 dual polarimetric SAR data, MODIS vegetation index data, and SRTM 30m DEM topographic data. Extract the total intensity reflectance and normalized Stokes parameters of SMAP-R. Obtain the surface scattering, double bounce scattering, volume scattering components and entropy, anisotropy, and angle of GF-3 fully polarimetric data. Extract the vertical polarization backscattering coefficient and cross polarization backscattering coefficient of Sentinel-1. Integrate the soil moisture (SM), surface roughness (Roughness), vegetation water content (VWC), vegetation optical thickness (VOD), and surface temperature (RC) of SMAP, and the enhanced vegetation index (EVI) and normalized vegetation index (NDVI) of MODIS. Step S3: Resample the SMAP-R, SMAP, MODIS, GF-3, and Sentinel-1 data to the EASE Grid 2.0 9 km × 9 km standard grid, and divide them into training, test, and validation sets; Step S4: Using the gradient boosting-based ensemble machine learning CatBoost algorithm, the backscattering coefficients of fully polarized GF-3 and dual-polarized Sentinel-1, along with auxiliary data, are used to evaluate the total intensity reflectivity of the SMAP-R polarization parameters. and the three Stokes parameters normalized The modeling process involves the following steps: All filtered and normalized data, including the backscattering coefficients of fully polarized GF-3 and bipolarized Sentinel-1, soil moisture (SM) and surface temperature (RC) from SMAP, vegetation water content (VWC), vegetation optical thickness (VOD), surface roughness, and enhanced vegetation index (EVI) and normalized vegetation index (NDVI) from MODIS, are used as model inputs. The SMAP-R polarization parameters are also considered. As output, different input features are combined according to different secondary inversion requirements to construct an inversion model of SMAP-R polarization parameters; Step S5: Verify the validity of the data by using an artificial neural network (ANN) to perform a second inversion of vegetation parameters and soil moisture. Step S6: Fill in the spatial missing grid in the SMAP-R data based on the model output. Specifically, this involves: inverting the model... , , , The parameters are mapped to the EASE Grid, and the missing grid cells are filled by interpolation to generate a spatially continuous SMAP-R polarization dataset.

2. The method for filling gaps in spaceborne polarimetric SMAP-R data based on multi-polarimetric SAR data according to claim 1, characterized in that, The preprocessing in step S2 includes: Freeman-Durden decomposition was performed on GF-3 data to obtain surface scattering, double bounce scattering, and volume scattering components; entropy H, anisotropy A, and scattering angle α were extracted by Cloude-Pottier decomposition; normalized Stokes parameters were calculated on SMAP-R data; and vertical polarization and cross-polarization backscattering coefficients were extracted from Sentinel-1 data after radiometric correction and geometric registration.

3. The method for filling gaps in spaceborne polarimetric SMAP-R data based on multi-polarimetric SAR data according to claim 1, characterized in that, The data partitioning ratio in step S3 is as follows: The training set comprises 60%, the test set 25%, and the validation set 15%; grids with missing SMAP-R data are marked as null values.

4. The method for filling gaps in spaceborne polarimetric SMAP-R data based on multi-polarimetric SAR data according to claim 1, characterized in that, The ensemble learning model in step S4 is the CatBoost algorithm, and the input features include: Fully polarized mode: , , , , , , , , , , , , ; Dual polarization mode: , , , , , , , , .

5. The method for filling gaps in spaceborne polarimetric SMAP-R data based on multi-polarimetric SAR data according to claim 1, characterized in that, The ANN model in step S5 consists of an 8-layer fully connected network, with ReLU as the activation function, mean squared error (MSE) as the loss function, Adam optimizer as the optimization algorithm, and VOD, VWC, and SM as the output layer inversion parameters.

6. A method for filling gaps in spaceborne polarimetric SMAP-R data based on multi-polarimetric SAR data according to any one of claims 1-5, characterized in that, The method described herein is applicable to improving the spatial resolution of SMAP-R data in vegetation monitoring, soil moisture retrieval, and polar parameter extraction, and filling the spatial gaps caused by the sparse distribution of the original data.

Images