Storm synchronous inversion method used in GF-3 typhoon process
By combining the XGBoost machine learning model with ECMWF and WAVEWATCH-III data, the inefficiency and multi-source data dependency issues of traditional wind and wave inversion methods were resolved, enabling efficient and accurate inversion of wind speed and significant wave height during Typhoon GF-3.
Patent Information
- Application Number
- CN202510831056.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-20
- Publication Date
- 2025-09-26
AI Technical Summary
Traditional wind and wave inversion methods require step-by-step inversion or rely on multi-source data, resulting in low efficiency and insufficient accuracy.
The XGBoost machine learning model was combined with ECMWF wind field data and WAVEWATCH-Ⅲ wave model data to establish a synchronous wind and wave inversion method for the GF-3 typhoon process. The inversion algorithm for wind speed and significant wave height was established through the XGBoost model, and the VV and VH polarization NRCS, incident angle, wind direction and other features were used for training and verification.
Efficient and accurate synchronous wind and wave inversion of GF-3 typhoon data has been achieved, which has improved the inversion accuracy of wind speed and significant wave height and reduced the dependence on multi-source data.
Smart Images
Figure CN120706259A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of ocean remote sensing inversion, and in particular to a wind-wave synchronous inversion method used in the GF-3 typhoon process. Background Art
[0002] Over the past decade, synthetic aperture radar (SAR) measurements carried by satellites including RADARSAT-2 (R-2), Sentinel-1 (S-1), and Gaofen-3 (GF-3) have enabled sustained global monitoring of tropical cyclones (TCs). In this context, SAR-derived measurements of sea surface dynamics are a unique source for advancing TC-related research.
[0003] Since the 1980s, the imaging mechanisms of SAR have been extensively studied, particularly after the Seasat mission provided valuable ocean surface datasets. Existing techniques have shown that the ocean surface wind vector is correlated with co-polarized (i.e., vertical-vertical (VV) and horizontal-horizontal (HH)) SAR backscatter, as quantified by the normalized radar cross section (NRCS). Based on this principle, geophysical model functions (GMFs) have been developed for wind retrieval from various SAR systems, including C-band (CMOD series) and X-band (XMOD). Because co-polarized NRCSs experience signal saturation during TCs, vertical-horizontal (VH) NRCSs have been widely used for TC wind speed retrieval. Combining VV and VH polarized NRCS measurements can significantly improve the accuracy of SAR-derived TC winds.
[0004] The principle of SAR wave inversion originates from the mechanism of ocean wave imaging. Fundamentally, the SAR intensity spectrum is generated by modulating the wave spectrum with three mean transfer functions (MTFs). To simplify the inversion process, empirical models such as CWAVE have been developed to derive wave parameters from SAR imaging variables. With advances in artificial intelligence, traditional GMFs for wind inversion and empirical wave models have been enhanced by machine and deep learning methods. In recent work, the extreme gradient boosting (XGBoost) machine learning method was implemented to invert TC wind speed and significant wave height (SWH) from dual-polarization S-1 imagery. Summary of the Invention
[0005] In response to the defects in the existing technology, the purpose of the present invention is to provide a method for synchronous inversion of wind and waves for the GF-3 typhoon process. The method of the present invention provides an effective inversion method for GF-3 typhoon data to address the limitations of traditional wind and wave inversion methods that require gradual inversion or rely on multi-source data.
[0006] In order to solve the above problems, the technical solution of the present invention is:
[0007] A wind-wave synchronous inversion method for the GF-3 typhoon process includes the following steps:
[0008] Obtain ECMWF wind field data, WAVEWATCH-III wave model data and GF-3L2 typhoon data;
[0009] Use the XGBoost machine learning model to establish the GF-3 typhoon wind field inversion algorithm;
[0010] The XGBoost machine learning model was used to establish the GF-3 typhoon significant wave height inversion algorithm;
[0011] Verify the accuracy of the GF-3 typhoon wind and wave inversion algorithm.
[0012] Preferably, the step of obtaining ECMWF wind field data, WAVEWATCH-Ⅲ wave pattern data and GF-3L2 typhoon data specifically includes: collecting GF-3L2 typhoon data in the Chinese waters between 2023 and 2024, collecting ECMWF wind field data and WAVEWATCH-Ⅲ wave pattern data based on the time of GF3 data, and performing time and space matching based on the collected data to establish a training data set for the XGBoost model for model training.
[0013] Preferably, the step of establishing the GF-3 typhoon wind field inversion algorithm using the XGBoost machine learning model specifically includes: using the XGBoost machine learning model to establish the relationship between wind speed and NRCS, the machine learning model includes four input features: VV and VH polarization NRCS, incident angle and wind direction, and the target output is the reconstructed TC wind speed.
[0014] Preferably, the step of establishing the GF-3 typhoon significant wave height inversion algorithm using the XGBoost machine learning model specifically includes: the significant wave height is determined by multiple variables: VV and VH polarization NRCS, total SAR spectral energy, normalized azimuth cutoff wavelength λ and G, image uniformity Cvar and incident angle, and G=S / L, where S is the satellite speed and L is the slant range; for the XGBoost-based significant wave height retrieval, these parameters are used as input features, and the WW3 simulated significant wave height is used as the target output.
[0015] Preferably, the step of verifying the accuracy of the GF-3 typhoon wind and wave inversion algorithm specifically includes: based on the established XGBoost algorithm, using WAVEWATCH-III wave pattern data and SMAP typhoon observation data to verify the SAR inversion wind and wave results, and verifying the accuracy of the inversion algorithm.
[0016] Compared to existing technologies, this invention uses ECMWF wind data and WW3 ocean wave pattern data as training data to construct a synchronous wind and wave inversion algorithm for GF-3 typhoon processes. The accuracy of the results is verified using WW3 ocean wave pattern data and SMAP typhoon observation data. This invention provides an effective inversion method for GF-3 typhoon data, overcoming the limitations of traditional wind and wave inversion methods, which require step-by-step inversion or rely on multiple data sources. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments with reference to the following drawings:
[0018] Figure 1 This is a flowchart of the wind-wave synchronous inversion method used in the GF-3 typhoon process of the present invention;
[0019] Figure 2 To show the geographical distribution of Gaofen-3 images during tropical cyclones, the wide-scan SAR mode was used to obtain the best track data map from the Japan Meteorological Agency (JMA);
[0020] Figure 3a This is the ERA-5 wind map at 10:00 UTC on August 4, 2023 during TC Khanun;
[0021] Figure 3b SWH diagram simulated for WW3.
[0022] Figure 4a Active and passive wind speed observation maps of soil moisture at 21:00 UTC on November 14, 2024, during TC Manyi;
[0023] Figure 4b This is a graph of significant wave height measurements from the Ocean 2 (HY-2) altimeter on August 28, 2024;
[0024] Figure 5 Performance graph of an XGBoost model trained for SAR wind field inversion;
[0025] Figure 6 Performance plot of training an XGBoost model for SAR SWH inversion;
[0026] Figure 7a Retrieve the wind speed map from the GF-3SAR image during TC Khanun on August 4, 2023 at 09:52UTC;
[0027] Figure 7b A scatter plot comparing SAR-derived wind speeds with SMAP observations;
[0028] Figure 8a The SWH map was retrieved from the GF-3SAR imagery at 09:52 UTC on August 4, 2023, during TC Khanun;
[0029] Figure 8b Comparison of SWH derived from SAR with HY-2 altimeter measurements and WW3 simulations. DETAILED DESCRIPTION
[0030] The present invention will be described in detail below with reference to specific embodiments. The following examples will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those skilled in the art, several changes and improvements can be made without departing from the scope of the present invention. These all fall within the scope of protection of the present invention.
[0031] Specifically, the present invention provides a method for synchronous inversion of wind and waves used in GF-3 typhoon process, such as Figure 1 As shown, the method includes the following steps:
[0032] S1: Obtain ECMWF wind field data, WAVEWATCH-III wave model data and GF-3L2 typhoon data;
[0033] GF-3L2 typhoon data were collected in the Chinese waters between 2023 and 2024. ECMWF wind field data and WAVEWATCH-III (WW3) wave model data were collected based on the time of GF3 data. Based on the collected data, time and space matching was performed, and a training dataset for the XGBoost model was established for model training.
[0034] Specifically, a total of 28 GF-3 images acquired in wide-scan SAR (WSC) mode were collected during a tropical cyclone (TC) over the Northwestern Pacific (NWP). For wind wave retrieval, NRCS with VV and VH polarizations were used. Figure 2 All geographic coordinates are shown for the VV-polarized GF-3 WSC imagery, which is overlaid with Japan Meteorological Agency (JMA) best track data and the corresponding quick view imagery for TC Khanun on August 4, 2023, at 10:00 UTC. The GF-3 imagery is juxtaposed with ERA-5 data (0.25° wind and 0.5° SWH grids, 1-hour temporal resolution). Note that the TC wind field is reconstructed using a combination of ERA-5 wind and JMA best track data to address the known ERA-5 TC wind underestimation. Furthermore, the SWH hindcast is performed using the third-generation numerical model WW3 at a spatial resolution of 0.25° and a temporal resolution of 0.5 hours to develop the SAR wave inversion algorithm. Figure 3ashows the ERA-5 wind field at 10:00 UTC on August 4, 2023 during TCKhanun, while Figure 3b The corresponding SWH diagram of the WW3 simulation is also shown. Figure 4a The Soil Moisture Active Passive (SMAP) wind speed observations at 21:00 UTC on November 14, 2024, during TC Manyi. Figure 4b The significant wave height (SWH) measured by the Ocean-2 (HY-2) altimeter on August 28, 2024.
[0035] S2: Use XGBoost machine learning model to establish GF-3 typhoon wind field inversion algorithm;
[0036] Specifically, an XGBoost machine learning model is used to establish the relationship between wind speed and NRCS. The machine learning model contains four input features: VV and VH polarization NRCS, incident angle and wind direction (derived from SAR wind ripples), and the target output is the reconstructed TC wind speed (derived from ECMWF). The pairing of 20 images is used to train the XGBoost model, such as Figure 5 As shown, the XGBoost model achieved the best performance (RMSE = 1.5 m / s) after 400 training iterations.
[0037] S3: Establishing the GF-3 Typhoon Significant Wave Height Inversion Algorithm Using XGBoost Machine Learning Model;
[0038] Specifically, the significant wave height (SWH) in this algorithm is determined by multiple variables: the VV and VH polarization NRCS, the total SAR spectral energy, the normalized azimuth cutoff wavelength λ and G (G = S / L, where S is the satellite velocity and L is the slant range), image uniformity Cvar, and the angle of incidence. For XGBoost-based SWH retrieval, these parameters serve as input features, and the WW3 simulated SWH serves as the target output. Figure 6 The training performance is demonstrated on samples extracted from 20 images, achieving a stable SWH RMSE of approximately 0.7m after 200 iterations.
[0039] S4: Verify the accuracy of the GF-3 typhoon wind and wave inversion algorithm.
[0040] Based on the XGBoost algorithm established above, the WAVEWATCH-Ⅲ wave pattern data and SMAP typhoon observation data were used to verify the SAR inversion wind and wave results to verify the accuracy of the inversion algorithm.
[0041] Specifically, the trained XGBoost model was applied to eight GF-3 images co-located with the SMAP wind product. Figure 7aThe wind speed map retrieved from GF-3SAR imagery during TC Khanun on August 4, 2023 at 09:52 UTC is shown. Figure 7b The statistical analysis of all validation samples is shown, and the results show that the wind speed RMSE is 3.31m / s, the correlation coefficient (COR) is 0.80, and the scattering index (SI) is 0.29. However, the validation dataset contains limited samples of extreme wind speeds (>40m / s), which may lead to larger retrieval errors in high wind conditions. The trained XGBoost model is used to retrieve SWH from SAR imagery and validated on HY-2 altimeter data. Figure 8a The SWH map retrieved during TC Khanun at 09:52 UTC on 4 August 2023 is shown. Due to the limited collocation of HY-2, the WW3 simulation was additionally included for verification. Figure 8b A comparison between SAR-derived SWH (up to 10 m), HY-2 measurements, and WW3 simulations is shown, showing an RMSE of 0.74 m, a COR of 0.83, and an SI of 0.25.
[0042] The above describes specific embodiments of the present invention. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art may make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. The embodiments of this application and the features in the embodiments may be combined with each other in any manner unless there is a conflict.
Claims
1. A wind-wave synchronous inversion method used in the GF-3 typhoon process, characterized in that: The method comprises the following steps: Obtain ECMWF wind field data, WAVEWATCH-III wave model data and GF-3L2 typhoon data; Use the XGBoost machine learning model to establish the GF-3 typhoon wind field inversion algorithm; The XGBoost machine learning model was used to establish the GF-3 typhoon significant wave height inversion algorithm; Verify the accuracy of the GF-3 typhoon wind and wave inversion algorithm.
2. The synchronous wind-wave inversion method for GF-3 typhoon process according to claim 1 is characterized in that: The steps of obtaining ECMWF wind field data, WAVEWATCH-III wave pattern data and GF-3L2 typhoon data specifically include: collecting GF-3L2 typhoon data in the Chinese waters between 2023 and 2024, collecting ECMWF wind field data and WAVEWATCH-III wave pattern data based on the time of GF3 data, and performing time-space matching based on the collected data to establish a training data set for the XGBoost model for model training.
3. The synchronous wind-wave inversion method for GF-3 typhoon process according to claim 1 is characterized in that: The steps of establishing the GF-3 typhoon wind field inversion algorithm using the XGBoost machine learning model specifically include: using the XGBoost machine learning model to establish the relationship between wind speed and NRCS, the machine learning model includes four input features: VV and VH polarization NRCS, incident angle and wind direction, and the target output is the reconstructed TC wind speed.
4. The synchronous wind-wave inversion method for GF-3 typhoon process according to claim 1 is characterized in that: The steps of establishing the GF-3 typhoon significant wave height inversion algorithm using the XGBoost machine learning model specifically include: the significant wave height is determined by multiple variables: VV and VH polarization NRCS, total SAR spectral energy, normalized azimuth cutoff wavelength λ and G, image uniformity Cvar and incident angle, wherein G=S / L, where S is the satellite speed and L is the slant range; for the XGBoost-based significant wave height retrieval, these parameters are used as input features, and the WW3 simulated significant wave height is used as the target output.
5. The synchronous wind-wave inversion method for GF-3 typhoon process according to claim 1 is characterized in that: The steps of verifying the accuracy of the GF-3 typhoon wind and wave inversion algorithm specifically include: based on the established XGBoost algorithm, using WAVEWATCH-III wave pattern data and SMAP typhoon observation data to verify the SAR inversion wind and wave results, and verifying the accuracy of the inversion algorithm.