A Method for Estimating Global Lake Wind Speed Using Reflected Signals of BDS / GPS / GLONASS / Galileo on Board Tianmu-1 Satellite in China

Abstract

The present invention discloses a method for estimating the global lake wind speed using reflected signals of BDS / GPS / GLONASS / Galileo on the Chinese Tianmu-1 satellite, which includes three major modules, namely, a global lake data extraction module based on the reflected measurements of the four systems (BDS / GPS / GLONASS / Galileo) of the Tianmu-1 constellation, a module for screening important characteristic variables related to the inversion of the global lake wind speed by machine learning algorithms and calculating the importance values of input features by SHAP, and a module for constructing an inversion model of the global lake surface wind speed. The present invention has for the first time achieved quantitative results of inverting the global lake wind speed using the reflected signals of BDS / GPS / GLONASS / Galileo on the Tianmu-1 constellation, and adopted machine learning methods to improve the accuracy of lake wind speed inversion, which strongly promotes the great potential of spaceborne GNSS-R in the wind speed monitoring of inland waters.

Description

Technical Field

[0001] The present invention belongs to the technical field of cross - research on retrieving lake wind speed parameters from GNSS reflected signals and machine learning. Specifically, it relates to a method for estimating the global lake wind speed from the reflected signals of the Tianmu - 1 satellite - borne BDS / GPS / GLONASS / Galileo in China. Background Art

[0002] Lakes, as important fresh water resources, the accurate measurement of the water - surface wind speed is of great significance for key fields such as meteorological forecasting, environmental monitoring, and water resources management. It helps to improve the accuracy of climate and weather prediction, monitor the dynamics of the lake ecosystem, and evaluate the wind energy potential, etc. Currently, the methods for measuring the water - surface wind speed of inland waters such as lakes include using anemometers and wind vanes on meteorological stations and buoys to obtain real - time data, etc. Radar (especially synthetic aperture radar, SAR) can collect data through ground or space platforms (such as satellites and aircraft). Therefore, radar technology can be used for both satellite remote sensing and ground or aerial remote sensing. It has the advantages of real - time monitoring, wide coverage, and strong adaptability in monitoring the lake water - surface wind speed, but also has disadvantages such as high cost, limited accuracy, and electromagnetic interference. In addition, unmanned aerial vehicles (UAVs) carrying sensors flying over lakes to collect wind speed data are also a flexible monitoring means. GNSS - R technology captures the signals transmitted by global navigation satellite systems and reflected by the lake water surface, processes and analyzes the delay and Doppler frequency of these signals, generates a delay - Doppler map (DDM) to extract features related to the wind speed. Utilizing the correlation between the wind speed and the water - surface roughness, and combining physical or empirical models, GNSS - R can convert these features into an estimated value of the lake water - surface wind speed. Although GNSS - R technology has been relatively mature in retrieving sea - surface wind speed, most current studies are limited to using single - system, dual - system, or triple - system reflected - signal observation data and only using a limited number of space - borne GNSS - R satellites to retrieve sea - surface wind speed. Currently, there are no quantitative experimental results on retrieving lake wind speed using GNSS - R, especially space - borne multi - GNSS - R satellites, at home and abroad.

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

[0004] To achieve the quantitative results of global lake wind speed inversion using the reflected signals of the four spaceborne GNSS-R systems (BDS / GPS / GLONASS / Galileo) for the first time, the present invention first proposes a method for estimating the global lake wind speed using the reflected signals of BDS / GPS / GLONASS / Galileo carried by the Tianmu-1 constellation of 22 low-orbit small satellites. The present invention provides a method for estimating the global lake wind speed using the reflected signals of BDS / GPS / GLONASS / Galileo on Tianmu-1. This method mainly includes three major modules, namely, a module for extracting global lake data based on the reflected measurements of the four systems (BDS / GPS / GLONASS / Galileo) of the Tianmu-1 constellation, a module for screening important characteristic variables related to global lake wind speed inversion using machine learning algorithms and calculating the importance values of input features by SHAP, and a module for constructing a global lake surface wind speed inversion model. Among them, in the module for screening important characteristic variables related to global lake wind speed inversion using machine learning algorithms and calculating the importance values of input features by SHAP, the bagged tree model is used to screen the characteristic variables for constructing the lake surface wind speed model, and SHAP is used to calculate the importance values of input features, so as to better understand the relationship between the characteristic variables and the lake wind speed inversion model, and then improve the accuracy of lake wind speed inversion. The present invention has realized for the first time the quantitative results of global lake wind speed inversion using the reflected signals of BDS / GPS / GLONASS / Galileo on the Tianmu-1 constellation, and uses machine learning methods to improve the accuracy of lake wind speed inversion, which strongly promotes the great potential of spaceborne GNSS-R in inland water body wind speed monitoring.

[0005] To solve the above technical problems, the basic concept of the technical solution adopted by the present invention is:

[0006] A method for estimating the global lake wind speed using the reflected signals of BDS / GPS / GLONASS / Galileo on Tianmu-1 includes the following steps:

[0007] Step S1, obtain the reflected signal data of the four systems (BDS, GPS, GLONASS, and Galileo) of the Tianmu-1 constellation (provided free of charge by Aerospace Tianmu (Chongqing) Satellite Technology Co., Ltd. and Aerospace Science and Industry (Beijing) Space Information Application Co., Ltd.), ERA5 lake surface wind speed, temperature, water depth, lake ice thickness, and rainfall data, and CCMP lake wind speed data;

[0008] Step S2, extract the observed quantities of the Tianmu-1 reflected data, select the auxiliary data variables of ERA5 lake surface temperature, water depth, lake ice thickness, and rainfall, and perform spatio-temporal matching on all data sets;

[0009] Step S3: Extract global lake data using the global lake shp file, and filter out the data with lake ice thickness ≤ 0.05m according to the auxiliary parameters of lake ice thickness in the dataset;

[0010] Step S4: Conduct data preprocessing and data quality control filtering, and divide the training dataset, validation dataset, and test dataset, accounting for 60%, 30%, and 10% of the total dataset respectively;

[0011] Step S5: Use machine learning algorithms to screen the characteristic variables for constructing the lake water surface wind speed model, and use the SHAP (SHapley Additive exPlanations) interpretability framework to calculate the characteristic importance values (SHAP values) of each model input parameter.

[0012] Step S6: Construct a global lake wind speed machine learning inversion model based on the reflected signals of BDS / GPS / GLONASS / Galileo, and use the training dataset and test dataset to train the lake wind speed inversion models of different systems and test the inversion wind speed performance of different systems.

[0013] Furthermore, the spaceborne GNSS-R data uses the reflected signal data of the Tianmu-1 constellation's four systems (BDS, GPS, GLONASS, and Galileo) provided free of charge by Aerospace Tianmu (Chongqing) Satellite Technology Co., Ltd. and Aerospace Science and Industry (Beijing) Space Information Application Co., Ltd. The reference wind speed products use the ERA5 lake water surface wind speed product and the CCMP lake wind speed product. The external auxiliary data uses the ERA5 lake temperature, water depth, lake ice thickness, and rainfall data.

[0014] Furthermore, merge the observed quantities of the Tianmu-1 reflected data and the auxiliary data variables. To avoid the influence of wind speed inversion errors caused by spatio-temporal differences on different datasets, match all auxiliary datasets (ERA5 and CCMP) with the Tianmu-1 data spatio-temporally, unify the spatio-temporal resolution, and obtain the matched dataset. In terms of time, align both ERA5 and CCMP to Coordinated Universal Time (UTC). In terms of space, perform two-dimensional linear interpolation for the Tianmu-1 data and the two auxiliary datasets (ERA5 and CCMP). In addition, for the reference wind speed used, the wind speeds extracted from CCMP and ERA5 are both the u-wind component and the v-wind component. Therefore, the optimal wind speed calculation needs to be carried out first to obtain the true wind speed, and its calculation formula is:

[0015]

[0016] In the formula, WS is the true wind speed, and u and v are the eastward component and northward component of the wind speed respectively.

[0017] Furthermore, global lake data is extracted using the global lake shp file. Lake ice thickness can change the scattering characteristics of signals on the water surface, leading to signal attenuation. The extracted lake data is screened according to the auxiliary parameter of lake ice thickness in the dataset to select data with a lake ice thickness ≤ 0.05m.

[0018] Furthermore, for the dataset, values with a variable value of -9999.9 are assigned as NaN and removed, and then the dataset is normalized. After data filtering, the dataset is divided into a training set, a test set, and a validation set in a ratio of 6:3:1.

[0019] Furthermore, in step S5, an important feature variable for constructing a lake water surface wind speed model is screened using a Bagging regression model based on XGBoost (the feature variables for screening are shown in Table 1). Among them, the learning rate of the model is set to 0.02; the type of base learner is selected as a tree-based model; the maximum number of leaf nodes is set to 127; to prevent overfitting, the feature ratio of random sampling for each tree is set to 0.6. The SHAP (SHapley Additive exPlanations) interpretability framework is used to calculate the feature importance values (SHAP values) of the top ten important feature variables (Sp_lon, Sp_alt, Sp_lat, Rx_lat, temperature, Ddm_sp_reflectivity, Sample_num, precipitation, Direct_signal_noise, Sp_fresnel_coeff_square) screened out. To perform model interpretation in SHAP, an explainer needs to be created first. The explainer of SHAP used is Kernel. The specific calculation formula of the SHAP value is as follows:

[0020]

[0021] In the formula, φ i is the contribution of feature i to the model output (i.e., the SHAP value), N is the set of all features, S is a subset of features that does not contain feature i, f(S) is the model prediction value on the feature set S, f(S∪{i}) is the model prediction value on features S and feature i, |S|! and (|N|-|S|-1)! are the factorial weights of feature permutations and combinations.

[0022] Table 1 Machine learning screening feature variables

[0023]

[0024]

[0025]

[0026] Further, in step S6: construct a Bagging Tree (BT) machine learning model for global lake surface wind speed inversion, input the training set into the model for training, use the validation set to supervise the model training to prevent overfitting, and use the test set to test the model to obtain the inverted wind speed. Four accuracy evaluation indicators (Root Mean Square Error (RMSE), Bias, Mean Absolute Error (MAE), and Correlation Coefficient (CC)) are used to comprehensively evaluate the predicted wind speed of the lake surface wind speed inversion model constructed from the observed data of the four system reflection signals.

[0027] After adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art.

[0028] The present invention first uses the reflected signal data of spaceborne GNSS-R four systems (BDS / GPS / GLONASS / Galileo) for global lake wind speed inversion and obtains reliable accuracy. And the present invention uses a machine learning model to screen the characteristic variables for constructing the lake surface wind speed model, which effectively improves the lake wind speed inversion accuracy compared with manually extracting the characteristic variables.

[0029] The following further describes in detail the specific embodiments of the present invention with reference to the accompanying drawings. Description of the Drawings

[0030] The accompanying drawings, as part of this application, are used to provide a further understanding of the present invention. The schematic embodiments and descriptions thereof of the present invention are used to explain the present invention, but do not constitute an improper limitation of the present invention. Obviously, the accompanying drawings in the following description are only some embodiments, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts. In the drawings:

[0031] Figure 1 It is a schematic flowchart of the method in the embodiment of the present invention.

[0032] Figure 2 It is a diagram of the importance order of the characteristic variables related to lake wind speed inversion in the embodiment of the present invention.

[0033] Figure 3 It is a scatter density diagram of the global lake wind speed of different systems and the ERA5 reference wind speed inverted by manually screening the characteristics and inputting them into the model in the embodiment of the present invention.

[0034] Figure 4 It is a scatter density diagram of the global lake wind speed of different systems and the CCMP reference wind speed inverted by manually screening the characteristics and inputting them into the model in the embodiment of the present invention.

[0035] Figure 5This is a scatter density plot of the global lake wind speed and ERA5 reference wind speed of different systems inverted by the machine learning screening feature input model in the embodiments of the present invention.

[0036] Figure 6 This is a scatter density plot of the global lake wind speed and CCMP reference wind speed of different systems inverted by the machine learning screening feature input model in the embodiments of the present invention.

[0037] Figure 7 This is a comparison chart of the importance of global lake wind speed inversion features in the embodiments of the present invention and in the implementation cases of the present invention.

[0038] Figure 8 This is a global deviation distribution map of the global lake wind speed and ERA5 reference wind speed of different systems inverted by the machine learning screening feature input model in the embodiments of the present invention.

[0039] Figure 9 This is a global deviation distribution map of the global lake wind speed and CCMP reference wind speed of different systems inverted by the machine learning screening feature input model in the embodiments of the present invention.

[0040] It should be noted that these drawings and textual descriptions are not intended to limit the scope of the concept of the present invention in any way, but to illustrate the concept of the present invention to those skilled in the art by referring to specific embodiments. Detailed Embodiments

[0041] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. The following embodiments are used to illustrate the present invention but are not used to limit the scope of the present invention.

[0042] Embodiment 1

[0043] To verify the effectiveness of the method proposed in the present invention, GNSS-R observation data of four systems of the Tianmu-1 constellation, ERA5 lake surface wind speed, temperature, water depth, lake ice thickness, and rainfall data for a total of 20 days from March 11, 2024, to March 31, 2024, and CCMP lake wind speed data were obtained for experiments.

[0044] Combined with the attached Figures 1-9 , the method for estimating the global lake wind speed using reflected signals of BDS / GPS / GLONASS / Galileo on the Chinese Tianmu-1 satellite in this embodiment includes:

[0045] Step S1, obtain the reflected signal data of the Tianmu-1 constellation's four systems (BDS, GPS, GLONASS, and Galileo) (provided free of charge by Aerospace Tianmu (Chongqing) Satellite Technology Co., Ltd. and Aerospace Science and Industry (Beijing) Space Information Application Co., Ltd.), ERA5 lake surface wind speed, temperature, water depth, lake ice thickness, and rainfall data, and CCMP lake wind speed data;

[0046] Step S2, extract the observed variables of the Tianmu-1 reflected data, select the ERA5 lake surface temperature, water depth, lake ice thickness, and rainfall auxiliary data variables, and perform spatio-temporal matching on all datasets;

[0047] Step S3, extract global lake data using the global lake shp file, and filter out the data with lake ice thickness ≤ 0.05 m according to the lake ice thickness auxiliary parameter in the dataset;

[0048] Step S4, perform data preprocessing and data quality control filtering, and divide the training dataset, validation dataset, and test dataset, accounting for 60%, 30%, and 10% of the total dataset respectively;

[0049] Step S5, use machine learning algorithms to screen the characteristic variables for constructing the lake surface wind speed model, and use the SHAP (SHapley Additive exPlanations) interpretability framework to calculate the characteristic importance values (SHAP values) of each model input parameter.

[0050] Step S6, construct a global lake wind speed machine learning inversion model based on the BDS / GPS / GLONASS / Galileo reflected signals, and use the training dataset and test dataset to train the lake wind speed inversion models of different systems and test the inversion wind speed performance of different systems.

[0051] As an implementation manner of this embodiment, in Step S1, the spaceborne GNSS-R data uses the reflected signal data of the Tianmu-1 constellation's four systems (BDS, GPS, GLONASS, and Galileo) provided free of charge by Aerospace Tianmu (Chongqing) Satellite Technology Co., Ltd. and Aerospace Science and Industry (Beijing) Space Information Application Co., Ltd., the reference wind speed product uses the ERA5 lake surface wind speed product and the CCMP lake wind speed product. The external auxiliary data uses the ERA5 lake temperature, water depth, lake ice thickness, and rainfall data.

[0052] As an implementation manner of this embodiment, in step S2, first, the observation data of Tianmu-1 reflectance and the auxiliary data variables are merged. To avoid the influence of wind speed inversion error caused by spatio-temporal differences on different data sets, all auxiliary data sets (ERA5 and CCMP) are spatio-temporally matched with Tianmu-1 data to unify the spatio-temporal resolution, and the matched data set is obtained. The specific implementation method is as follows: in terms of time, both ERA5 and CCMP are aligned to Coordinated Universal Time (UTC); in terms of space, two-dimensional linear interpolation is used for spatial interpolation of Tianmu-1 data and the two auxiliary data (ERA5 and CCMP) sets. In addition, for the reference wind speed used in the embodiment, the wind speeds extracted from CCMP and ERA5 are both the u-wind component and the v-wind component. Therefore, the optimal wind speed needs to be calculated first to obtain the true wind speed, and its calculation formula is:

[0053]

[0054] In the formula, WS is the true wind speed, and u and v are the eastward component and northward component of the wind speed respectively.

[0055] As an implementation manner of this embodiment, in step S3, the global lake data is extracted using the global lake shp file. The lake ice thickness will change the scattering characteristics of the signal on the water surface and cause signal attenuation. The extracted lake data is screened according to the lake ice thickness auxiliary parameter in the data set to select the data with lake ice thickness ≤ 0.05m.

[0056] As an implementation manner of this embodiment, in step S4, for the data set, the value of -9999.9 is assigned as NaN and removed, and then the data set is normalized. After data filtering, the data set is divided into a training set, a test set, and a validation set according to the ratio of 6:3:1.

[0057] As an implementation manner of this embodiment, in step S5, a Bagging regression model based on XGBoost is used to screen important feature variables for constructing the lake surface wind speed model (the feature variables for screening are shown in Table 1). Among them, the learning rate of the model is set to 0.02; the type of the base learner is selected as a tree-based model; the maximum number of leaf nodes is set to 127; to prevent overfitting, the proportion of randomly sampled features for each tree is set to 0.6. The SHAP (SHapley Additive exPlanations) interpretability framework is used to calculate the feature importance values (SHAP values) of the top ten important feature variables (Sp_lon, Sp_alt, Sp_lat, Rx_lat, temperature, Ddm_sp_reflectivity, Sample_num, precipitation, Direct_signal_noise, Sp_fresnel_coeff_square) screened out. In SHAP, to perform model interpretation, an explainer needs to be created first. The explainer of SHAP used is Kernel. The specific calculation formula of the SHAP value is as follows:

[0058]

[0059] In the formula, φ i is the contribution of feature i to the model output (i.e., the SHAP value), N is the set of all features, S is a subset of features that does not contain feature i, f(S) is the model prediction value on the feature set S, f(S∪{i}) is the model prediction value on features S and feature i, |S|! and (|N|-|S|-1)! are the factorial weights of feature permutations and combinations.

[0060] Table 1 Machine learning screening feature variables

[0061]

[0062]

[0063]

[0064] Appendix Figure 7 is a comparison chart of the feature importance of global lake wind speed inversion. It can be analyzed that:

[0065] From Appendix Figure 7From the upper left figure, it can be seen that features such as Sp_lon, Sp_alt, and Sp_lat have relatively large positive SHAP values, which contribute greatly to the positive prediction of the model; temperature and Ddm_sp_reflectivity have both positive and negative contributions, indicating that high and low values of these features will have different impacts on the model in different directions; Direct_signal_noise and Sp_fresnel_coeff_square are more concentrated in the negative contribution area, meaning that high values of these features may reduce the prediction results of the model. From the attached Figure 7 From the middle figure, it can be seen that Sp_lon is the most important feature, having the greatest impact on model prediction. Followed by Sp_alt and Sp_lat, these features also contribute relatively greatly to the model; temperature, Ddm_sp_reflectivity, and Sample_num are also relatively important features in model prediction. From the attached Figure 7 From the upper right figure of the attached, it can be seen that Sp_lon has the greatest positive contribution to this sample, pushing up the predicted value by 0.29; Sp_alt also has a relatively large positive contribution, increasing by 0.26; Rx_lat brings a negative contribution, decreasing by 0.11. From the attached Figure 7 From the lower figure in the attached, it can be seen that Sp_alt and Sp_lon are the two main positive contribution features, pushing up the predicted values by 0.26 and 0.29 respectively. Rx_lat has a negative impact on the predicted value, decreasing by 0.11; other features such as Sp_lat, Sp_fresnel_coeff_square, etc. also make corresponding contributions to the prediction results.

[0066] As an implementation manner of this embodiment, for the step S6 of constructing a global lake wind speed machine learning inversion model based on BDS / GPS / GLONASS / Galileo reflected signals, a Bagging Tree machine learning model is used to construct a global lake surface wind speed inversion model. The training set is input into the model for training, the validation set is used to supervise the model training to prevent overfitting, and the test set is used to test the model to obtain the inverted wind speed. Four accuracy evaluation indicators (Root Mean Square Error (RMSE), Bias, Mean Absolute Error (MAE), Correlation Coefficient (CC)) are used to comprehensively evaluate the predicted wind speed of the lake surface wind speed inversion model constructed from the observed data of the reflected signals of the four systems.

[0067] The following are the comprehensive evaluation results of the four systems for predicting lake wind speed: Attached Figure 3Shows the scatter density plots of the global lake wind speeds of different systems inversed by the input model with manually screened features (60, as shown in Table 1) and the ERA5 reference wind speeds. The evaluation results of the accuracy indicators of its four systems (BDS, GPS, GAL, GLO) are RMSE: 1.506 m / s, 1.877 m / s, 1.662 m / s, 1.612 m / s, CC: 0.74, 0.65, 0.64, 0.7, MAE: 0.19 m / s, 0.556 m / s, 0.19 m / s, 0.357 m / s. Attached Figure 5 Shows the scatter density plots of the global lake wind speeds of different systems inversed by the input model with machine learning screened features (10). The evaluation results of the accuracy indicators of its four systems (BDS, GPS, GAL, GLO) are RMSE: 1.237 m / s, 1.569 m / s, 1.488 m / s, 1.527 m / s, CC: 0.84, 0.77, 0.72, 0.73, MAE: 0.099 m / s, 0.482 m / s, 0.073 m / s, 0.213 m / s. By comparison, Attached Figure 3 and Attached Figure 5 It can be seen that by using machine learning to screen the features required for the inversion of the global lake wind speed, the inversion wind speeds of the four systems have been improved in terms of RMSE, MAE and CC. In addition, Attached Figure 4 and Attached Figure 6 Correspond to the scatter density plots of the global lake wind speeds of different systems inversed by the input model with manually screened features and machine learning screened features and the CCMP reference wind speeds. Attached Figure 8 and Attached Figure 9 Are respectively the global deviation distribution plots of the global lake wind speeds of different systems inversed by the input model with machine learning screened features and the ERA5 and CCMP reference wind speeds. From Attached Figure 8 and Attached Figure 9 It can be seen that the deviation of the wind speed inversion results of the four systems of Tianmu-1 from the ERA5 reference wind speed is generally better than that from the CCMP reference wind speed. In summary, by using the machine learning method to screen the feature variables, the accuracy of the wind speed inversion has been significantly improved. Further analysis shows that the comparison accuracy of the wind speed inversion results of the four systems of Tianmu-1 with the ERA5 reference wind speed is generally better than that with the CCMP reference wind speed. In addition, among the four systems of the Tianmu-1 constellation, the overall performance of the wind speed inversion results of the BDS system is better than that of the GPS, GAL and GLO systems.

[0068] The above are only the preferred embodiments of the present invention, and do not impose any form of limitation on the present invention. Although the present invention has been disclosed above with the preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art of this patent, without departing from the scope of the technical solution of the present invention, can make some changes or modifications using the technical content prompted above to obtain equivalent embodiments of equivalent changes. However, as long as it does not depart from the content of the technical solution of the present invention, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention still fall within the scope of the present invention's solution.

Claims

1. A method for estimating global lake wind speed from BDS / GPS / GLONASS / Galileo reflected signals onboard China's Tianmu-1 satellite, characterized in that: The following steps are involved: Step S1, obtaining the reflected signal data of the four systems BDS, GPS, GLONASS and Galileo of the Tianmu-1 constellation, the ERA5 lake surface wind speed, temperature, water depth, lake ice thickness and rainfall data, and the CCMP lake wind speed data; Step S2, extract the Tianmu-1 reflection data observations, select the ERA5 lake surface temperature, water depth, lake ice thickness and rainfall auxiliary variable data, first merge the Tianmu-1 reflection data observations and auxiliary data variables, then perform spatiotemporal matching of all auxiliary data sets with the Tianmu-1 data, unify the spatiotemporal resolution, and obtain the matched data set. Specific implementation: In terms of time, ERA5 and CCMP are aligned to Coordinated Universal Time; in terms of space, two-dimensional linear interpolation is used to spatially interpolate the Tianmu-1 data and two auxiliary data sets, the two auxiliary data sets are ERA5 and CCMP; in addition, for the reference wind speed used, the wind speeds extracted from CCMP and ERA5 are both u wind components and v wind components, so the optimal wind speed calculation must be performed first to obtain the true wind speed, and the calculation formula is: Where WS is the true wind speed, u and v are the eastward and northward components of the wind speed respectively; Step S3, using the global lake shp file to extract global lake GNSS-R observation data, and filtering out lake ice thickness ≤ 0.05 m data according to lake ice thickness auxiliary parameters in the data set; Step S4, data preprocessing and data quality control filtering, and dividing the training data set, validation data set and test data set, accounting for 60%, 30% and 10% of the total data set respectively; Step S5, using a machine learning algorithm to screen characteristic variables for constructing a lake surface wind speed model, and using a SHAP interpretability framework to calculate the SHAP value of each model input parameter; Step S6, construct a global lake wind speed machine learning inversion model based on BDS / GPS / GLONASS / Galileo reflection signals, and use the training data set and test data set to train the lake wind speed inversion models of different systems and test the inversion wind speed performance of different systems.

2. A method for estimating global lake wind speed based on BDS / GPS / GLONASS / Galileo reflected signals from China's Tianmu-1 satellite according to claim 1, characterized in that: In step S1, the satellite-borne GNSS-R data adopts the reflected signal data of the four systems BDS, GPS, GLONASS and Galileo of the Tianmu-1 constellation, the reference wind speed product adopts the ERA5 lake surface wind speed product and CCMP lake wind speed product, and the external auxiliary data adopts the ERA5 lake temperature, water depth, lake ice thickness and rainfall data.

3. A method for estimating global lake wind speed based on BDS / GPS / GLONASS / Galileo reflected signals from China's Tianmu-1 satellite according to claim 1, characterized in that: Step S4 is specifically as follows: for the data set, the variable value of -9999.9 is assigned to NaN and removed, and then the data set is normalized. After data filtering, the data set is divided into a training set, a test set, and a validation set in a ratio of 6:3:

1.

4. A method for estimating global lake wind speed based on BDS / GPS / GLONASS / Galileo reflected signals from China's Tianmu-1 satellite according to claim 1, characterized in that: Step S5 is specifically as follows: the bagging regression model of XGBoost is used to select important characteristic variables for constructing the lake surface wind speed model from 60 variables, wherein the learning rate of the model is set to 0.02; the basic learner type selects a tree-based model; the maximum number of leaf nodes is set to 127; in order to prevent overfitting, the feature ratio of each tree randomly sampled is set to 0.6; the SHAP interpretability framework is used to calculate the SHAP values ​​of the top ten important characteristic variables Sp_lon, Sp_alt, Sp_lat, Rx_lat, temperature, Ddm_sp_reflectivity, Sample_num, precipitation, Direct_signal_noise, and Sp_fresnel_coeff_square. To interpret the model in SHAP, an interpreter needs to be created first. The SHAP interpreter used is Kernel, and the specific calculation formula of the SHAP value is as follows: In the formula, φ i is the SHAP value of feature i for the model output, N is the set of all features, S is the feature subset excluding feature i, f(S) is the model prediction value on feature set S, f(S∪{i}) is the model prediction value on feature S and feature i, |S|! and (|N|-|S|-1)! are the factorial weights of the feature permutations.

5. A method for estimating global lake wind speed based on BDS / GPS / GLONASS / Galileo reflected signals from China's Tianmu-1 satellite according to claim 1, characterized in that: Step S6 is specifically as follows: construct a bagged tree machine learning model for global lake surface wind speed inversion, input the training set into the model training, use the validation set to supervise the model training to prevent overfitting, use the test set to test the model, obtain the inverted wind speed, and use the four accuracy evaluation indicators of root mean square error, deviation, mean absolute error, and correlation coefficient to comprehensively evaluate the predicted wind speed of the lake surface wind speed inversion model constructed based on the reflection signal observation data of the four systems.

Images