Satellite-borne GNSS-R soil humidity inversion method and system considering earth surface influence factors

Abstract

The invention relates to a satellite-borne GNSS-R soil humidity inversion method and system considering earth surface influence factors, and the method comprises the steps: obtaining CYGNSS L1 reflection data, SMAP soil moisture data and multi-source static earth surface factor data, and carrying out the data preprocessing; the correlation between the preprocessed CYGNSS L1 reflection data and the SMAP soil moisture data is calculated, and the space-time correlation is obtained; evaluating, analyzing and checking the multi-source static surface factor data by adopting a Geodejector method, and screening out an optimal influence factor combination in combination with spatial-temporal correlation; and training a soil moisture inversion model according to the optimal influence factor combination, and obtaining the trained soil moisture inversion model to invert the soil moisture. According to the method, the interpretation force and the interaction effect of the earth surface factors are quantitatively screened, the inversion model which is simple in input, high in precision and high in adaptability is constructed, and high-resolution inversion of the soil moisture is achieved.

Description

Technical Field

[0001] This invention relates to the field of soil moisture inversion technology, and in particular to a spaceborne GNSS-R soil moisture inversion method and system that takes into account surface influence factors. Background Technology

[0002] Soil moisture is a key parameter in the terrestrial hydrological cycle, significantly impacting agricultural production management, drought monitoring, ecosystem dynamics assessment, and climate model simulation. Currently, soil moisture is primarily obtained through three methods: ground-based measurements, optical remote sensing, and microwave remote sensing.

[0003] Ground-based measurement methods rely on sampling points or sensor networks. While they offer the advantage of high precision, they are limited by factors such as high cost, limited spatial coverage, and insufficient deployment density, making it difficult to achieve continuous monitoring of soil moisture at the regional or global scale.

[0004] Optical remote sensing methods typically use vegetation indices and surface reflectance to indirectly estimate soil moisture, but they are highly sensitive to cloud cover, solar altitude angle and surface vegetation cover, limiting their application in areas with high vegetation cover or long-term cloud and fog conditions, and their inversion stability is relatively weak.

[0005] Microwave active or passive remote sensing technologies offer advantages such as cloud penetration and day-night observability, with satellite data like SMAP being a prime example, widely used for global soil moisture monitoring. However, their spatial resolution is typically on the order of 36 km, which is insufficient to meet the demands of agricultural management, hydrological models, and refined surface studies requiring medium- to high-resolution data. Furthermore, the long repetition cycles make it difficult to capture rapid dynamic changes in soil moisture.

[0006] The GNSS-R (Global Navigation Satellite System Reflected Signal) technology, which has developed in recent years, utilizes the passive reception characteristics of navigation satellite signals after scattering on the Earth's surface, providing higher spatiotemporal resolution and being less affected by weather conditions. The CYGNSS satellite system, composed of eight microwave receiving satellites, can achieve multiple daily repeated observations, demonstrating great potential in the field of soil moisture retrieval. However, existing CYGNSS-based soil moisture retrieval methods still have the following problems:

[0007] (1) It usually relies on a lot of external auxiliary data, such as vegetation parameters, soil type, and surface structure, which leads to complex models and redundant parameters;

[0008] (2) There is a lack of systematic quantitative analysis on how surface factors in different regions affect the relationship between CYGNSS reflectance signals and soil moisture;

[0009] (3) Whether there are interactive enhancement or nonlinear coupling effects among the factors has not been fully studied, which affects the interpretability of the model;

[0010] (4) In areas with strong surface heterogeneity (such as hilly areas, farmland and forest areas), the inversion accuracy is unstable and the reliability is insufficient.

[0011] Therefore, current technologies lack a CYGNSS soil moisture retrieval method that is effective in screening surface influencing factors, has a simple structure, and is highly accurate. There is an urgent need to construct an retrieval model based on key factors that improves retrieval accuracy and spatial adaptability while reducing model complexity, in order to meet the needs of refined hydrological and agricultural monitoring. Summary of the Invention

[0012] The purpose of this invention is to provide a spaceborne GNSS-R soil moisture inversion method and system that considers surface influence factors. It aims to overcome the problems of excessive auxiliary factors, high model complexity, unclear surface factor influence mechanisms, and unstable inversion accuracy in heterogeneous surface regions in the existing soil moisture inversion process based on CYGNSS satellite reflection signals. This invention quantitatively screens the explanatory power and interaction effects of surface factors to construct an inversion model with simple input, high accuracy, and strong adaptability, thereby achieving high-resolution inversion and verification of soil moisture.

[0013] To achieve the above objectives, the present invention provides the following solution:

[0014] A spaceborne GNSS-R soil moisture inversion method considering surface influence factors includes:

[0015] Acquire CYGNSS L1 reflectance data, SMAP soil moisture data, and multi-source static surface factor data, and perform data preprocessing;

[0016] The correlation between the preprocessed CYGNSS L1 reflectance data and SMAP soil moisture data is calculated to obtain the spatiotemporal correlation.

[0017] The Geodetector method was used to evaluate, analyze, and verify the multi-source static surface factor data, and the optimal combination of influencing factors was selected by combining the spatiotemporal correlation.

[0018] A soil moisture inversion model is trained based on the optimal combination of influencing factors, and the trained soil moisture inversion model is used to invert soil moisture. The soil moisture inversion model is constructed using the XGBoost machine learning model, and the optimal combination of influencing factors is land use, reflectance, and vegetation biomass.

[0019] Optionally, data preprocessing may include:

[0020] The CYGNSS L1 reflection data were quality-screened based on signal-to-noise ratio, antenna gain, and incident angle, and the surface reflectivity was calculated based on the bistatic radar cross-section model.

[0021] The multi-source static surface factor data are unified at the pixel scale according to the unified projection, spatial resolution and geographical range;

[0022] Based on time series statistical characteristics, the surface reflectance and SMAP soil moisture data are subjected to long-term normalization processing, and logarithmic transformation and bias enhancement are introduced into the surface reflectance.

[0023] Optionally, calculating the correlation between the preprocessed CYGNSS L1 reflectance data and SMAP soil moisture data includes:

[0024] Calculate the pixel-by-pixel Pearson correlation coefficient between the CYGNSS L1 reflectance data and SMAP soil moisture data monthly to obtain the correlation distribution map;

[0025] Based on the correlation distribution map, the effectiveness of the reflected signal in different regions and seasons is determined, and the spatiotemporal correlation is obtained.

[0026] Optionally, the Geodetector method may be used to evaluate, analyze, and verify the multi-source static surface factor data, including:

[0027] A quantitative assessment of the explanatory power of each factor in the multi-source static surface factor data was performed.

[0028] Two-factor interaction effect analysis was performed on the factors in the multi-source static surface factor data to determine whether there are enhancing or nonlinear interactions between different factors and to obtain the significance test results.

[0029] Based on the explanatory power and the significance test results, the optimal combination of influence factors is selected.

[0030] This invention also provides a spaceborne GNSS-R soil moisture retrieval system that considers surface influence factors, comprising:

[0031] The data acquisition module is used to acquire CYGNSS L1 reflectance data, SMAP soil moisture data, and multi-source static surface factor data.

[0032] The preprocessing module is used to preprocess the collected data.

[0033] The factor screening module is used to calculate the correlation between the preprocessed CYGNSS L1 reflectance data and SMAP soil moisture data, obtain the spatiotemporal correlation, evaluate, analyze and verify the multi-source static surface factor data using the Geodetector method, and screen out the optimal combination of influencing factors based on the spatiotemporal correlation.

[0034] The inversion model module is used to train a soil moisture inversion model based on the optimal combination of influencing factors, and to obtain the trained soil moisture inversion model to invert soil moisture. The soil moisture inversion model is constructed using the XGBoost machine learning model, and the optimal combination of influencing factors is land use, reflectance, and vegetation biomass.

[0035] Optionally, data preprocessing may include:

[0036] The CYGNSS L1 reflection data were quality-screened based on signal-to-noise ratio, antenna gain, and incident angle, and the surface reflectivity was calculated based on the bistatic radar cross-section model.

[0037] The multi-source static surface factor data are unified at the pixel scale according to the unified projection, spatial resolution and geographical range;

[0038] Based on time series statistical characteristics, the surface reflectance and SMAP soil moisture data are subjected to long-term normalization processing, and logarithmic transformation and bias enhancement are introduced into the surface reflectance.

[0039] Optionally, calculating the correlation between the preprocessed CYGNSS L1 reflectance data and SMAP soil moisture data includes:

[0040] Calculate the pixel-by-pixel Pearson correlation coefficient between the CYGNSS L1 reflectance data and SMAP soil moisture data monthly to obtain the correlation distribution map;

[0041] Based on the correlation distribution map, the effectiveness of the reflected signal in different regions and seasons is determined, and the spatiotemporal correlation is obtained.

[0042] Optionally, the Geodetector method may be used to evaluate, analyze, and verify the multi-source static surface factor data, including:

[0043] A quantitative assessment of the explanatory power of each factor in the multi-source static surface factor data was performed.

[0044] Two-factor interaction effect analysis was performed on the factors in the multi-source static surface factor data to determine whether there are enhancing or nonlinear interactions between different factors and to obtain the significance test results.

[0045] Based on the explanatory power and the significance test results, the optimal combination of influence factors is selected.

[0046] The beneficial effects of this invention are as follows:

[0047] 1. A quantitative screening mechanism for surface factors is proposed:

[0048] In existing technologies, the selection of auxiliary parameters often relies on empirical judgment or simple correlation analysis, lacking a systematic study on the quantitative mechanisms of different surface parameters in the reflection process and their time-varying characteristics. This invention uses Geodetector to analyze the explanatory power and interaction effects of surface factors, clarifying the mechanisms of action of key and non-key factors, making the model structure more interpretable.

[0049] 2. The model structure is significantly simplified while maintaining high accuracy:

[0050] Using only three key factors—reflectance, biomass, and land use—it can achieve an inversion accuracy similar to that of a full-factor model (RMSE < 0.06 cm³ / cm³), reducing model complexity and improving computational efficiency.

[0051] 3. Strong inversion stability and wide adaptability:

[0052] It maintains good inversion results in plains, hills, farmland, and areas with complex vegetation distribution, and has strong robustness and regional adaptability.

[0053] 4. Data is easy to obtain and can be promoted globally:

[0054] CYGNSS and related surface factors are globally available and suitable for global soil moisture monitoring, providing high spatiotemporal resolution data support for agricultural, meteorological, hydrological and ecological monitoring. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 This is a flowchart of a spaceborne GNSS-R soil moisture inversion method considering surface influence factors according to an embodiment of the present invention;

[0057] Figure 2 This is a graph showing the analysis of influence factor results under different discretization methods in this invention embodiment;

[0058] Figure 3This is a schematic diagram of factor interaction effects in an embodiment of the present invention;

[0059] Figure 4 Scatter plots showing model training and testing under different feature combination schemes in embodiments of the present invention;

[0060] Figure 5 This is a histogram of the accuracy results in an embodiment of the present invention;

[0061] Figure 6 The following are spatial RMSE distribution diagrams of the inversion results in embodiments of the present invention, wherein (a) is the spatial RMSE distribution diagram of the inversion results of the complete feature set in Scheme 1, (b) is the spatial RMSE distribution diagram of the inversion results of the first three features in Scheme 2, and (c) is the spatial RMSE distribution diagram of the inversion results of the last four features in Scheme 3. Detailed Implementation

[0062] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0064] like Figure 1 As shown, this embodiment proposes a spaceborne GNSS-R soil moisture inversion method that considers surface influence factors, including:

[0065] Acquire CYGNSS L1 reflectance data, SMAP soil moisture data, and multi-source static surface factor data, and perform data preprocessing;

[0066] The correlation between preprocessed CYGNSS L1 reflectance data and SMAP soil moisture data was calculated to obtain the spatiotemporal correlation.

[0067] The Geodetector method was used to evaluate, analyze and verify multi-source static surface factor data, and the optimal combination of influencing factors was selected by combining spatiotemporal correlation.

[0068] The soil moisture inversion model was trained based on the optimal combination of influencing factors, and the soil moisture inversion model was obtained after training to invert soil moisture. The soil moisture inversion model was constructed using the XGBoost machine learning model, and the optimal combination of influencing factors was land use, reflectance and vegetation biomass.

[0069] Further data preprocessing includes:

[0070] The quality of CYGNSS L1 reflection data was screened based on signal-to-noise ratio, antenna gain, and incident angle, and the surface reflectivity was calculated based on the bistatic radar cross-section model.

[0071] The multi-source static surface factor data are unified at the pixel scale based on unified projection, spatial resolution and geographical range;

[0072] Based on time series statistical characteristics, long-term normalization processing was performed on surface reflectance and SMAP soil moisture data, and logarithmic transformation and bias enhancement were introduced into the surface reflectance.

[0073] Specifically, the enhanced preprocessing of CYGNSS reflection data includes: acquiring CYGNSS L1 reflection data, performing quality screening based on signal-to-noise ratio, antenna gain, and incident angle, and calculating surface reflectivity based on a bistatic cross-sectional area model.

[0074] Spatial isomorphic processing of multi-source surface factors includes: acquiring SMAP soil moisture data and multi-source static surface factor data such as digital elevation model, slope, land use type, vegetation biomass, NDVI, and tree height, and forming a one-to-one isomorphic dataset at the pixel scale by unifying projection, spatial resolution (0.025°), and geographical extent, providing a foundation for subsequent pixel-by-pixel analysis and machine learning modeling.

[0075] Time series enhancement methods based on long-term statistical characteristics include: addressing the inconsistency between CYGNSS reflectance calculated based on the bi-base cross-sectional area model and soil moisture on an absolute numerical scale, instead of using single-time normalization, long-term normalization is performed based on multi-year time series statistical characteristics. Logarithmic transformation and bias enhancement are introduced into the reflectance calculated based on the bi-base cross-sectional area model to highlight its relative variation characteristics, making the reflectance signal more suitable for cross-regional and cross-seasonal modeling.

[0076] Furthermore, the correlation between the preprocessed CYGNSS L1 reflectance data and SMAP soil moisture data was calculated, including:

[0077] Calculate the pixel-by-pixel Pearson correlation coefficient of CYGNSS L1 reflectance data and SMAP soil moisture data monthly to obtain the correlation distribution map;

[0078] Based on the correlation distribution map, the effectiveness of the reflected signal in different regions and seasons is determined, and the spatiotemporal correlation is obtained.

[0079] Specifically, the feature pre-analysis mechanism that introduces correlation structure constraints includes: before factor screening, constructing the spatiotemporal correlation structure between CYGNSS reflectance and soil moisture through pixel-by-pixel Pearson correlation analysis, which is used to determine the effectiveness of the reflectance signal in different regions and seasons, to further verify the temporal characteristics of the correlation in subsequent experiments, and to analyze the temporal synergistic trend between reflectance and soil moisture.

[0080] Furthermore, the Geodetector method is used to evaluate, analyze, and verify multi-source static surface factor data, including:

[0081] Quantitative assessment of the explanatory power of each factor in multi-source static surface factor data;

[0082] Two-factor interaction effect analysis was performed on factors in multi-source static surface factor data to determine whether there are enhancing or nonlinear interactions between different factors and to obtain significance test results.

[0083] Based on the explanatory power and significance test results, the optimal combination of influence factors was selected.

[0084] Specifically, such as Figure 3 As shown, the Geodetector-based joint screening mechanism for factor explanatory power and interaction effects addresses the issue that existing methods only focus on the importance of a single factor. This embodiment introduces the Geodetector method to simultaneously screen multiple surface factors.

[0085] (1) Quantitative assessment of the explanatory power of single factors;

[0086] (2) Two-factor interaction enhancement effect analysis;

[0087] (3) Statistical significance test;

[0088] This study identified key factors that play a dominant role in the spatial differentiation of soil moisture and exhibit a nonlinear reinforcing relationship, and screened out three core factors: land use, reflectance, and vegetation biomass.

[0089] Furthermore, this embodiment also provides a method for constructing a machine learning inversion model based on comparative feature combinations:

[0090] In the inversion model construction phase, instead of building a single model, three feature combination schemes are designed:

[0091] (1) Full factorial combination model;

[0092] (2) Key factor combination model based on Geodetector screening;

[0093] (3) Non-critical factor combination model;

[0094] The XGBoost architecture was used for training. By comparing the model performance differences under different feature structures, the effectiveness of key factor selection was verified at the model level. Scatter plots of model training and testing under different feature combination schemes are shown below. Figure 4 As shown, Figure 5 The results shown validate the proposed scheme of the geographic detector. The schemes using all features and those using the main features exhibit higher overall inversion accuracy. The spatial RMSE distribution of the inversion results for the three feature combination schemes is shown in the figure below. Figure 6 As shown in (a)-(c).

[0095] Furthermore, the method in this embodiment also provides a multi-level independent verification and stability evaluation mechanism:

[0096] The inversion results were independently validated in both spatial and temporal dimensions using test set error indices (RMSE, MAE, R²) and SMCI1.0 in-situ observation station data. The inversion stability of different land surface types was also evaluated, and finally, a high-resolution soil moisture product was generated.

[0097] This embodiment also provides a spaceborne GNSS-R soil moisture inversion system that considers surface influence factors, including:

[0098] The data acquisition module is used to acquire CYGNSS L1 reflectance data, SMAP soil moisture data, and multi-source static surface factor data. The data acquisition module includes a CYGNSS L1 data interface, an SMAP soil moisture interface, a vegetation biomass database interface, a land use database interface, and an NDVI data interface.

[0099] The preprocessing module is used to preprocess the collected data.

[0100] The factor screening module is used to calculate the correlation between preprocessed CYGNSS L1 reflectance data and SMAP soil moisture data to obtain spatiotemporal correlation. The Geodetector method is used to evaluate, analyze and verify multi-source static surface factor data, and the optimal combination of influencing factors is screened by combining spatiotemporal correlation. Among them, based on the explanatory power q value and the intensity of two-factor interaction of the factors output by the Geodetector model, reflectance, biomass and land use are identified as the main features according to the q value ranking.

[0101] The inversion model module is used to train a soil moisture inversion model based on the optimal combination of influencing factors, and to retrieve soil moisture from the trained soil moisture inversion model. The soil moisture inversion model is constructed using the XGBoost machine learning model, and the optimal combination of influencing factors is land use, reflectance and vegetation biomass. The soil moisture inversion model can generate soil moisture raster data at a resolution of 0.025° or higher.

[0102] Further data preprocessing includes:

[0103] The quality of CYGNSS L1 reflection data was screened based on signal-to-noise ratio, antenna gain, and incident angle, and the surface reflectivity was calculated based on the bistatic radar cross-section model.

[0104] The multi-source static surface factor data are unified at the pixel scale based on unified projection, spatial resolution and geographical range;

[0105] Based on the time series statistical characteristics, long-term normalization processing was performed on surface reflectance and SMAP soil moisture data, and logarithmic transformation and bias enhancement were introduced for surface reflectance.

[0106] Furthermore, the correlation between the preprocessed CYGNSS L1 reflectance data and SMAP soil moisture data was calculated, including:

[0107] Calculate the pixel-by-pixel Pearson correlation coefficient of CYGNSS L1 reflectance data and SMAP soil moisture data monthly to obtain the correlation distribution map;

[0108] Based on the correlation distribution map, the effectiveness of the reflected signal in different regions and seasons is determined, and the spatiotemporal correlation is obtained.

[0109] Furthermore, the Geodetector method is used to evaluate, analyze, and verify multi-source static surface factor data, including:

[0110] Quantitative assessment of the explanatory power of each factor in multi-source static surface factor data;

[0111] Two-factor interaction effect analysis was performed on factors in multi-source static surface factor data to determine whether there are enhancing or nonlinear interactions between different factors and to obtain significance test results.

[0112] Based on the explanatory power and significance test results, the optimal combination of influence factors was selected.

[0113] The present invention will be further described below with reference to embodiments:

[0114] A spaceborne GNSS-R soil moisture inversion method considering surface influence factors includes the following steps:

[0115] CYGNSS Data Processing:

[0116] (1) Obtain CYGNSS L1 satellite scattering and reflection data from 2019 to 2022, including signal-to-noise ratio (SNR), antenna gain, BRCS, and geometric parameters of the transmitter and receiver.

[0117] (2) Set data filtering conditions:

[0118] a. Signal-to-noise ratio > 2dB;

[0119] b. Antenna gain range: 0–13 dB;

[0120] c. Angle of incidence < 65° to exclude noisy data far from the specular reflection area.

[0121] (3) The surface reflectance is calculated using the CYGNSS bistatic scattering model, and the formula is as follows:

[0122] ;

[0123] Among them, R t and R r Let be the distances from the launching and receiving satellites to the mirror point, respectively, and σ be the bistatic cross-sectional area.

[0124] (4) Spatial aggregation of reflectance data according to geographic grid, and uniformity to 0.025° resolution.

[0125] Multi-source surface factor data preprocessing:

[0126] (1) Download the SMAP soil moisture product that is in the same period as CYGNSS.

[0127] (2) Obtain digital elevation model (DEM), slope, land use, biomass, tree height and NDVI data, and unify them to 0.025° resolution based on bilinear interpolation or nearest neighbor algorithm.

[0128] (3) Use WGS84 projection to convert all data to a unified coordinate system.

[0129] (4) Crop the data according to the study area to obtain a multi-source dataset with consistent size and cell alignment.

[0130] Time series normalization and bias enhancement:

[0131] (1) Long-term normalization of CYGNSS reflectance and soil moisture was performed. The normalization range was calculated based on the minimum and maximum values ​​over many years to enhance the temporal variation characteristics of reflectance.

[0132] (2) Based on the statistical differences between SMAP and CYGNSS over many years, the reflectance deviation term is calculated to improve the sensitivity of reflectance to changes in soil moisture.

[0133] (3) Perform a logarithmic transformation on the reflectivity to stretch the distribution of low reflectivity areas and improve the distinguishability of abnormal areas.

[0134] Correlation analysis between CYGNSS and soil moisture:

[0135] (1) Calculate the Pearson correlation coefficient between CYGNSS reflectance and SMAP soil moisture pixel by pixel on a monthly basis to obtain the correlation distribution map.

[0136] (2) Further analyze its seasonal variation pattern based on the correlation distribution map calculated on a monthly basis, and identify the difference in sensitivity of reflectance to wet and dry seasons.

[0137] (3) The correlation analysis between CYGNSS and soil moisture in this region is used as a priori guarantee to ensure the feasibility and effectiveness of subsequent experiments.

[0138] Factor screening based on Geodetector:

[0139] (1) Land use, NDVI, biomass, tree height, and other factors are classified according to the natural breakpoint method, standard deviation stratification method, or quantile method. Discretized data classification facilitates the next step of geographic detector statistical regularity and result derivation, such as... Figure 2 As shown in the figure. Among them, reflectance and tree height were divided into 7 categories using the standard deviation stratification method; NDVI was divided into 7 categories using the equidistant stratification method; DEM, slope, and biomass were stratified using the quantile method; and land use was classified using the natural breakpoint method.

[0140] (2) Calculate the single-factor explanatory power q value for each factor to determine the degree of influence of the factor on the spatial distribution of soil moisture.

[0141] (3) Conduct a two-factor interaction effect analysis to determine whether there is an enhancing or nonlinear interaction between different factors.

[0142] (4) Based on the explanatory power ranking and the results of the two-factor interaction, three key factors were screened: land use, reflectance and biomass.

[0143] The influence of surface factors is ranked as follows: land use > reflectivity > biomass > tree height > NDVI > slope > DEM.

[0144] Construction of XGBoost Soil Moisture Inversion Model:

[0145] (1) Set up the training plan:

[0146] Option 1: Input all factors;

[0147] Option II: Input only key factors (reflectivity, biomass, land use);

[0148] Option III: Input only non-critical factors (tree height, NDVI, slope, DEM).

[0149] (2) Model parameters include:

[0150] a. Tree depth 15;

[0151] b. Learning rate 0.05;

[0152] c. Number of base learners: 1000;

[0153] d. Sampling ratio 0.6;

[0154] e. L1 and L2 regularization penalty terms are used to suppress overfitting.

[0155] (3) The data is randomly divided into a training set and a test set in a ratio of 8:2.

[0156] (4) Use the training set to complete the model training, and use the test set to calculate RMSE, MAE and R².

[0157] Model validation and result output:

[0158] (1) The accuracy of the model was tested using the test set. The RMSE of the key factor scheme (scheme II) was ≈ 0.0586 cm³ / cm³, which was better than the non-key factor scheme and close to the full factor scheme.

[0159] (2) The inversion results were independently verified with SMCI1.0 in-situ soil moisture data, and the consistency of time series, dynamic response and error distribution were compared.

[0160] (3) Generate high-resolution soil moisture spatial products and make spatial RMSE distribution maps and time series verification maps.

[0161] (4) Analyze the abnormal areas to ensure the stability of the model under different land surface types.

[0162] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A spaceborne GNSS-R soil moisture inversion method considering surface influence factors, characterized in that, include: Acquire CYGNSS L1 reflectance data, SMAP soil moisture data, and multi-source static surface factor data, and perform data preprocessing; The correlation between the preprocessed CYGNSS L1 reflectance data and SMAP soil moisture data is calculated to obtain the spatiotemporal correlation. The Geodetector method was used to evaluate, analyze, and verify the multi-source static surface factor data, and the optimal combination of influencing factors was selected by combining the spatiotemporal correlation. A soil moisture inversion model is trained based on the optimal combination of influencing factors, and the trained soil moisture inversion model is used to invert soil moisture. The soil moisture inversion model is constructed using the XGBoost machine learning model, and the optimal combination of influencing factors is land use, reflectance, and vegetation biomass.

2. The spaceborne GNSS-R soil moisture inversion method considering surface influence factors according to claim 1, characterized in that, Data preprocessing includes: The CYGNSS L1 reflection data were quality-screened based on signal-to-noise ratio, antenna gain, and incident angle, and the surface reflectivity was calculated based on the bistatic radar cross-section model. The multi-source static surface factor data are unified at the pixel scale according to the unified projection, spatial resolution and geographical range; Based on time series statistical characteristics, the surface reflectance and SMAP soil moisture data are subjected to long-term normalization processing, and logarithmic transformation and bias enhancement are introduced into the surface reflectance.

3. The spaceborne GNSS-R soil moisture inversion method considering surface influence factors according to claim 1, characterized in that, The correlation calculation between the preprocessed CYGNSS L1 reflectance data and SMAP soil moisture data includes: Calculate the pixel-by-pixel Pearson correlation coefficient between the CYGNSS L1 reflectance data and SMAP soil moisture data monthly to obtain the correlation distribution map; Based on the correlation distribution map, the effectiveness of the reflected signal in different regions and seasons is determined, and the spatiotemporal correlation is obtained.

4. The spaceborne GNSS-R soil moisture inversion method considering surface influence factors according to claim 1, characterized in that, The Geodetector method was used to evaluate, analyze, and verify the multi-source static surface factor data, including: A quantitative assessment of the explanatory power of each factor in the multi-source static surface factor data was performed. Two-factor interaction effect analysis was performed on the factors in the multi-source static surface factor data to determine whether there are enhancing or nonlinear interactions between different factors and to obtain the significance test results. Based on the explanatory power and the significance test results, the optimal combination of influence factors is selected.

5. A spaceborne GNSS-R soil moisture inversion system considering surface influence factors, used to implement the spaceborne GNSS-R soil moisture inversion method considering surface influence factors as described in any one of claims 1-4, characterized in that, include: The data acquisition module is used to acquire CYGNSS L1 reflectance data, SMAP soil moisture data, and multi-source static surface factor data. The preprocessing module is used to preprocess the collected data. The factor screening module is used to calculate the correlation between the preprocessed CYGNSS L1 reflectance data and SMAP soil moisture data, obtain the spatiotemporal correlation, evaluate, analyze and verify the multi-source static surface factor data using the Geodetector method, and screen out the optimal combination of influencing factors based on the spatiotemporal correlation. The inversion model module is used to train a soil moisture inversion model based on the optimal combination of influencing factors, and to obtain the trained soil moisture inversion model to invert soil moisture. The soil moisture inversion model is constructed using the XGBoost machine learning model, and the optimal combination of influencing factors is land use, reflectance, and vegetation biomass.

6. The spaceborne GNSS-R soil moisture inversion system considering surface influence factors according to claim 5, characterized in that, Data preprocessing includes: The CYGNSS L1 reflection data were quality-screened based on signal-to-noise ratio, antenna gain, and incident angle, and the surface reflectivity was calculated based on the bistatic radar cross-section model. The multi-source static surface factor data are unified at the pixel scale according to the unified projection, spatial resolution and geographical range; Based on time series statistical characteristics, the surface reflectance and SMAP soil moisture data are subjected to long-term normalization processing, and logarithmic transformation and bias enhancement are introduced into the surface reflectance.

7. The spaceborne GNSS-R soil moisture inversion system considering surface influence factors according to claim 5, characterized in that, The correlation calculation between the preprocessed CYGNSS L1 reflectance data and SMAP soil moisture data includes: Calculate the pixel-by-pixel Pearson correlation coefficient between the CYGNSS L1 reflectance data and SMAP soil moisture data monthly to obtain the correlation distribution map; Based on the correlation distribution map, the effectiveness of the reflected signal in different regions and seasons is determined, and the spatiotemporal correlation is obtained.

8. The spaceborne GNSS-R soil moisture inversion system considering surface influence factors according to claim 5, characterized in that, The Geodetector method was used to evaluate, analyze, and verify the multi-source static surface factor data, including: A quantitative assessment of the explanatory power of each factor in the multi-source static surface factor data was performed. Two-factor interaction effect analysis was performed on the factors in the multi-source static surface factor data to determine whether there are enhancing or nonlinear interactions between different factors and to obtain the significance test results. Based on the explanatory power and the significance test results, the optimal combination of influence factors is selected.

Images