Sea surface ocean current inversion method and system integrating ground wave radar and GNSS reflection

Abstract

The invention relates to the technical field of ocean remote sensing and intelligent ocean monitoring, in particular to a sea surface ocean current inversion method and system fusing ground wave radar and GNSS reflection, and the method comprises the steps: obtaining ground wave radar radial flow and GNSS-R original data; aI-driven space-time interpolation is carried out on the ground wave radar data so as to restore missing; constructing a wind-wave-current coupled GNSS-R forward physical model to extract weak ocean current signals; taking a regional ocean model as a framework, dynamically distributing two types of observation weights according to a shore distance, and carrying out partition self-adaptive data assimilation by adopting ensemble Kalman filtering; and finally, a high-resolution seamlessly-connected two-dimensional ocean current field product is output. According to the method, the problems of discontinuous flow field, high data missing rate and difficulty in GNSS-R ocean current signal extraction in a coastal-ocean transition zone of a traditional method are solved, and physically consistent and high-precision business ocean current monitoring from a coastal area to an open sea area is realized.

Description

Technical Field

[0001] This invention relates to the field of marine remote sensing and intelligent marine monitoring technology, specifically to a method and system for inverting sea surface currents by integrating ground wave radar and GNSS reflection. Background Technology

[0002] Currently, remote sensing monitoring of ocean currents mainly relies on technologies such as high-frequency ground-wave radar (HF Radar, HFR) and satellite remote sensing. HF Radar transmits high-frequency radio waves and receives Bragg scattering echoes from the sea surface, enabling it to invert the radial velocity of the sea surface along the line of sight. It boasts a time resolution of minutes and a spatial resolution of 1 to 6 kilometers, and is widely used in monitoring nearshore marine dynamic processes. However, a single-station HF Radar can only acquire the radial velocity component; it requires the cross-coverage of two or more stations to synthesize a complete two-dimensional vector flow field. In actual deployment, due to limitations in coastal topography, construction costs, or the electromagnetic environment, radar stations are often sparse or have poor cross-angles, leading to a significant increase in vector synthesis errors. Furthermore, HF Radar observations are susceptible to radio frequency interference, ionospheric disturbances, and the lack of effective scattered waves from the sea surface under low wind speed conditions, resulting in a high data loss rate. In some sea areas, the daily effective data coverage is less than 70%, severely restricting its reliability in operational marine monitoring.

[0003] On the other hand, ocean remote sensing technology based on Global Navigation Satellite System Reflected Signals (GNSS-R) has developed rapidly in recent years. This technology utilizes low-Earth orbit satellites to receive GNSS navigation signals reflected from the sea surface, and by analyzing parameters such as delayed Doppler images, carrier-to-noise ratio, or interferometric phase, it can retrieve information on sea surface wind speed, significant wave height, and even ocean currents. GNSS-R has the advantages of all-weather, global coverage, and low cost, making it particularly suitable for dynamic monitoring in open sea areas. However, the impact of ocean currents on GNSS-R signals is extremely weak—the resulting Doppler shift is typically less than 0.1 Hz, far lower than the scattering broadening caused by wind and waves (which can reach several Hz), making ocean current signals easily masked by the scattering noise dominated by wind and waves. Simultaneously, spaceborne GNSS-R has low spatial resolution (typical footprint diameter of 5 to 20 kilometers) and a long revisit period, making it difficult to effectively capture key ocean dynamic structures such as mesoscale eddies and fronts. Currently, GNSS-R is still in the exploratory stage in ocean current retrieval, and a stable and reliable operational product has not yet been developed.

[0004] To address the aforementioned issues, existing technologies attempt to integrate multiple observation methods. For example, some studies stitch together nearshore current fields retrieved from ground-wave radar with current fields derived from geostrophic currents or inferred from sea surface temperature gradients via satellite altimetry to expand coverage. However, such methods typically employ simple stitching or interpolation between nearshore and open ocean regions, lacking unified physical constraints and dynamic coordination mechanisms. This leads to abrupt changes in current direction and discontinuous velocities in the transition zone (typically 50 to 150 kilometers from the shore), and even the appearance of non-physical circulation structures. Furthermore, most existing fusion systems fail to effectively compensate for missing data from ground-wave radar or fully exploit the weak ocean current information in GNSS-R, making it difficult to achieve truly high-precision, seamless sea surface current field reconstruction. Summary of the Invention

[0005] This invention addresses the technical problems existing in the prior art by providing a method and system for inverting sea surface ocean currents by fusing ground wave radar and GNSS reflections. It solves the problems that existing fusion technologies often use simple splicing or interpolation, lack unified physical constraints, resulting in discontinuities and abrupt changes in the flow field in the transition area between nearshore and open ocean, and even the appearance of non-physical structures. Furthermore, it fails to effectively compensate for missing HFR data or fully mine the weak ocean current information in GNSS-R.

[0006] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: A method for inverting ocean surface currents by integrating ground-wave radar and GNSS reflections includes: S1. Acquire radial velocity observation data from ground wave radar and raw GNSS reflection signal data; S2. For the spatiotemporal regions where the radial velocity observation data is missing, a pre-trained spatiotemporal neural network model is used for interpolation to reconstruct the complete radial velocity field. S3. Based on the dual-scale sea surface scattering theory, a wind-wave-current coupled forward model is constructed to characterize the GNSS reflected signal jointly modulated by wind speed, wave spectrum and ocean current velocity. S4. Using the background current field provided by the regional ocean numerical model as the state space, and based on the distance from the grid point to the shore in the regional ocean numerical model, dynamically allocate the weights of the radial current observations from the ground wave radar and the ocean current information obtained from the GNSS-R signal based on the forward model, and use the ensemble Kalman filter algorithm to assimilate and fuse the data. S5. Output the assimilated surface ocean current vector field and its uncertainty field.

[0007] The beneficial effects of this invention are: through the three-in-one architecture of "AI interpolation + physical model + adaptive assimilation", the fusion of HFR and GNSS-R under a unified dynamic framework is realized for the first time, which solves the problem of flow field discontinuity caused by traditional splicing, and the measured flow field direction error in the transition zone is reduced by more than 60%.

[0008] Furthermore, in step S1, the raw data of the GNSS reflected signal includes complex signals of the L1, L2 and / or L5 frequency bands, satellite orbit parameters and receiving platform location information.

[0009] Furthermore, in step S2, the spatiotemporal neural network model is a U-Net++ network, the input of which includes a spatial neighborhood matrix centered on the missing point, the velocity sequence of multiple time steps before and after, and the missing mask, and the output is the reconstructed radial velocity.

[0010] Furthermore, in step S3, the wind-wave-current coupled forward model specifically refers to: Introducing ocean current velocity into sea surface wave dynamics allows the phase velocity of sea surface waves to be corrected to the sum of the phase velocity when there is no current and the projection of the ocean current in the wave propagation direction. The theoretical Doppler shift Δf_model consists of three parts: platform motion geometry term Δf_plat, wave trajectory velocity term Δf_wave, and ocean current advection term Δf_current: Δf_model = Δf_plat + Δf_wave(v, U10) + Δf_current(u_c), where v is the wave spectrum parameter, U10 is the wind speed, and u_c is the ocean current velocity; The additional interference phase φ_current caused by ocean currents is expressed as: φ_current = (4π / λ) * (u_c ·r ⊥ ) * Δt, where λ is the signal wavelength, r ⊥ Let Δt be the horizontal displacement vector perpendicular to the line of sight, and let Δt be the coherent integration time.

[0011] Furthermore, in step S4, the strategy for dynamically allocating weights is as follows: When d < 50 km, only radial current observations from ground wave radar are assimilated; when 50 ≤ d ≤ 150 km, the weight of ground wave radar observations w_HFR = 1 - (d - 50) / 100, and the weight of GNSS-R inversion information w_GNSS = 1 - w_HFR; when d > 150 km, only ocean current information retrieved from GNSS-R is assimilated. d represents the distance from the grid point in the model to the shore.

[0012] Furthermore, step S5 also includes comparing the output ocean current product with the drifting buoy trajectory to generate a quality assessment report.

[0013] Furthermore, step S2 also includes quality control of the radial velocity observation data: removing data points with an absolute velocity value greater than 2.5 m / s or a signal-to-noise ratio less than 3 dB; removing 3σ outliers within a ±30 minute time window; verifying vector consistency in the multi-station coverage area and removing observations with directional deviations exceeding 45°.

[0014] Furthermore, the wind-wave-current coupled forward model described in step S3 represents the sea surface as a two-scale structure: long waves are described by the improved Pierson-Moskowitz spectrum and driven by a 10-meter-high wind speed, while short waves use the Elfouhaily directional spectrum and are modulated by the long wave slope; ocean currents are superimposed on the sea surface wave propagation background, and the corrected sea surface wave phase velocity is the sum of the original phase velocity and the projection of the ocean current in the wave propagation direction.

[0015] Furthermore, the ensemble Kalman filter algorithm adopts Local Ensemble Transform Kalman Filter (LETKF), the localization radii of HFR and GNSS-R are set to 20 km and 50 km respectively, the assimilation period is 30 minutes, the horizontal resolution of the ocean numerical model is 1 km, and the vertical layering is no less than 50 layers.

[0016] The present invention also provides a sea surface current inversion system for implementing the method, comprising: The data receiving unit is used to acquire radial velocity data from ground wave radar and raw data from GNSS reflection signals; A preprocessing and AI interpolation module, connected to the data receiving unit, is used for quality control of ground wave radar data and data missing interpolation based on neural networks; The physical inversion engine, connected to the data receiving unit, is used to store and run the wind-wave-current coupled forward model to extract ocean current information from GNSS-R signals; An adaptive assimilation computation unit is connected to the preprocessing and AI interpolation module and the physical inversion engine, respectively, and is used to run the regional ocean numerical model and the ensemble Kalman filter algorithm to realize dynamic weight allocation and fusion of observation data. The product service and interface unit is connected to the adaptive assimilation calculation unit and is used to output fused ocean current products and quality assessment results. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the framework of a sea surface current inversion system that integrates ground wave radar and GNSS reflection according to the present invention; Figure 2 A schematic diagram of the U-Net++ neural network provided by this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0020] In the description of this application, it should also be noted that, unless otherwise expressly specified and limited, the terms "set up," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this technology based on the specific circumstances.

[0021] In the description of this application, spatial relation terms such as "below," "under," "below," "below," "above," "over," etc., are used herein to describe the relationship between one element or feature shown in the figures and other elements or features. It should be understood that, in addition to the orientation shown in the figures, spatial relation terms also include different orientations of the device in use and operation. For example, if the device in the figures is flipped, an element or feature described as "below" or "under" or "below" of other elements or features will be oriented "above" other elements or features. Therefore, the exemplary terms "below" and "under" can include both upper and lower orientations. Furthermore, the device may also include other orientations (e.g., rotated 90 degrees or other orientations), and the spatial descriptive terms used herein are interpreted accordingly.

[0022] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0023] like Figure 1 and Figure 2 As shown, this embodiment of the invention provides a method for inverting sea surface currents by fusing ground wave radar and GNSS reflections, including the following steps: S1. Acquisition and Preprocessing of Multi-Source Observation Data: Acquiring radial current velocity observation data from shore-based high-frequency ground wave radar and raw GNSS reflection signal data from low-Earth orbit satellite platforms. Two types of data are acquired from local receiving stations or public data platforms (such as the NOAA HFR network and NASA PO.DAAC). One type is radial current velocity provided by the shore-based HFR network, with a time resolution of 10 minutes and spatial coverage of 10-200 km from the shore. The other type is raw GNSS-R data transmitted from low-Earth orbit satellites (such as CYGNSS), including complex signals in the L1 / L2 / L5 bands, satellite ephemeris, and platform position. This achieves standardized access to multi-source heterogeneous observation data, providing a data foundation for subsequent fusion.

[0024] S2. AI-driven spatiotemporal interpolation of ground wave radar data: The radial velocity observation data is subjected to quality control, and for spatiotemporal regions with missing data after quality control, a pre-trained spatiotemporal neural network model is used for interpolation to reconstruct the complete radial velocity field.

[0025] Multi-stage quality control was implemented for the HFR radial flow. First, invalid points with an absolute velocity greater than 2.5 m / s or a signal-to-noise ratio less than 3 dB were removed. Second, outliers deviating from the mean by more than three standard deviations were removed from the time series data within a ±30 minute window. For grid points still missing after quality control, a pre-trained U-Net++ neural network was used for imputation. Figure 2As shown, the network employs an encoder-decoder structure with dense skip connections. The input consists of a 7×7 spatial neighborhood centered on the missing point, the radial flow field across three time steps, and the corresponding missing mask matrix. The network is trained on a historical high-quality HFR dataset to learn the spatiotemporal evolution of the flow field. The interpolation process is executed on a GPU server, with a single processing time of less than 2 minutes.

[0026] AI-driven spatiotemporal interpolation effectively repaired data gaps caused by interference or low sea state, increasing the effective coverage of HFR data from approximately 68% to over 95%, significantly enhancing the integrity of nearshore flow field information and laying the foundation for high-precision vector synthesis.

[0027] S3. Constructing a GNSS-R Wind-Wave-Current Coupled Forward Model: Based on the dual-scale sea surface scattering theory, a wind-wave-current coupled forward model is constructed to characterize the GNSS reflected signal modulated by wind speed, wave spectrum, and ocean current velocity. Specifically, this includes: S31, Sea Surface Modeling: A dual-scale sea surface model is employed. The long-wave spectrum uses a modified Pierson-Moskowitz spectrum driven by wind speed U10; the short-wave spectrum uses the Elfouhaily directional spectrum and is modulated by the long-wave slope. The key innovation lies in incorporating ocean current velocity into sea surface wave dynamics, where the phase velocity of the sea surface wave is corrected to the sum of the phase velocity in the absence of current and the projection of the ocean current onto the wave propagation direction.

[0028] S32, Scattering Calculation: Based on the geometric relationship between the satellite and the receiving platform, calculate the incident and reflection directions of the GNSS signal, and correct the local incident angle and azimuth angle according to the local slope of the sea surface, and perform scattering calculation in the real local coordinate system.

[0029] S33, Doppler and Phase Modeling: The theoretical Doppler frequency shift Δf_model consists of three parts: the platform motion geometry term Δf_plat, the wave trajectory velocity term Δf_wave, and the ocean current advection term Δf_current. That is: Δf_model = Δf_plat + Δf_wave(v, U10) + Δf_current(u_c). Among them, the ocean current term Δf_current is linearly related to the projection of the ocean current velocity u_c in the wavenumber direction.

[0030] S34. Simultaneously, utilizing the phase information of the GNSS-R signal, the additional interference phase φ_current caused by ocean currents can be expressed as: φ_current = (4π / λ) * (u_c · r ⊥ ) * Δt, where λ is the signal wavelength, r ⊥ Let Δt be the displacement vector perpendicular to the line of sight, and let Δt be the coherent integration time.

[0031] S35 encapsulates the above process into a forward function F(U10, u_c, Geometry), which takes wind speed, ocean current, and observation geometry as input and outputs theoretical delay Doppler plot, Doppler frequency shift, or interferometric phase. This function supports automatic differentiation.

[0032] By constructing a dedicated physical model, the modulation mechanism of ocean currents on sea surface waves and electromagnetic scattering is systematically incorporated into the GNSS-R forward simulation for the first time, making it possible to separate weak ocean current signals from the observed signals and providing a reliable physical basis for subsequent high-precision inversion.

[0033] The model first describes the sea surface state, which is characterized as a two-scale structure consisting of long waves and short waves. The long waves are represented by a modified Pierson-Moskowitz spectrum, with wind speeds at a height of 10 meters as the threshold. The propagation of the sea surface wave is driven by the Elfouhaily directional spectrum, which is modulated by the slope of the long wave. The key point is that the propagation of the sea surface wave does not occur on a static background, but is superimposed on the ocean current. The phase velocity of the sea surface wave is corrected to the sum of the original phase velocity and the projection of the ocean current in the direction of wave propagation.

[0034] The physical inversion engine involves a "wind-wave-current coupled scattering forward model" that calculates the incident and reflection directions of the GNSS signal based on the real-time positions of the satellite and the receiving platform, and corrects the local incident angle and azimuth angle by incorporating the local sea surface slope. This process ensures that the scattering calculations are performed in a realistic local sea surface coordinate system, improving the accuracy of the forward simulation.

[0035] To connect ocean currents with GNSS-R observations, a Doppler shift was incorporated. The theoretical Doppler shift consists of three parts: geometric Doppler caused by the relative motion of the platform, orbital velocity Doppler caused by the vertical motion of sea surface waves, and advection Doppler caused by the horizontal motion of ocean currents. The theoretical Doppler shift is the sum of the platform term, wave term, and ocean current term, allowing the observation residuals to directly reflect ocean current information.

[0036] S4. Zonal Adaptive Data Assimilation: Using the background current field provided by the regional ocean numerical model as the state space, the weights of ground-wave radar radial current observations and ocean current information retrieved from GNSS-R signals based on the distance of model grid points from the shore are dynamically allocated. An ensemble Kalman filter algorithm is then used for data assimilation and fusion. The background field of the regional ocean numerical model (e.g., ROMS, horizontal resolution 1 km) is used as the state space. The Local Ensemble Transform Kalman Filter (LETKF) algorithm is employed, with an assimilation period of 30 minutes.

[0037] The dynamic weight allocation strategy involves dynamically allocating observation weights w based on the distance d (km) from the model grid point to the shore. When d < 50 km: Assimilate only HFR radial flow observations (w_HFR=1, w_GNSS=0).

[0038] When 50 ≤ d ≤ 150 km: The ocean current information retrieved from HFR and GNSS-R is assimilated together with weights. The weight calculation formulas are: w_HFR = 1 - (d - 50) / 100, w_GNSS = 1 - w_HFR. At this time, the GNSS-R information is used to obtain the initial ocean current estimate by embedding the forward model from step S3 into the 4D-Var framework.

[0039] When d>150 km: only the ocean current vector information retrieved by GNSS-R is assimilated (w_HFR=0, w_GNSS=1).

[0040] To suppress spurious correlations, the localization radius of HFR observations was set to 20 km, and the localization radius of GNSS-R observations was set to 50 km.

[0041] By employing a dynamic weighting mechanism, high-precision HFR observations are fully relied upon nearshore, while the global coverage capability of GNSS-R is primarily used in the open ocean. A smooth transition and optimal fusion of these two methods are achieved in the transition zone. This not only solves the discontinuity problem of traditional splicing but also dynamically constrains observational information within a physically consistent dynamic framework through data assimilation, significantly improving the physical rationality and accuracy of the flow field throughout the study area.

[0042] S5. Fusion Product Generation and Output: Outputs the assimilated surface ocean current vector field and its uncertainty field. After assimilation, the system outputs a high-resolution two-dimensional ocean current field (u, v) for the 0-1 meter surface layer in NetCDF format. Simultaneously, based on the ensemble variance output by the LETKF algorithm, an uncertainty distribution map of the current field is generated. The system automatically compares the fused product with global drifting buoy trajectories daily, generating a quality assessment report. The product is published via API or FTP service, with configurable update frequency. Standardized, high-quality ocean current products are generated that can directly serve marine forecasting, emergency response, and other operational needs, and objective quality assessments are provided, ensuring the reliability and practicality of the method.

[0043] Compared with the prior art, this invention has the following core innovations and direct beneficial effects: (1) A pioneering adaptive fusion mechanism for ground wave radar and GNSS-R under a unified dynamic framework was developed. By using the ocean numerical model as a bridge, two types of observation weights were dynamically allocated according to the distance from the shore, achieving seamless connection of the ocean current field from nearshore (<50 km) to the open ocean (>150 km), thus solving the problem of discontinuity and abrupt change in direction of the current field in the transition zone in the traditional splicing method.

[0044] (2) A method of "AI interpolation + physical prior" is proposed to repair missing ground wave radar data. Using spatiotemporal neural networks such as U-Net++, combined with historical flow field and model background field, high-fidelity reconstruction is performed on HFR data missing due to interference or low sea state, which significantly improves the integrity rate of nearshore flow field (from 68% to over 95% in actual measurements).

[0045] (3) Construct a wind-wave-current coupled GNSS-R forward model to effectively extract weak ocean current signals. Utilize multi-frequency interferometric phase and decouple the effects of wind and waves through 4D-Var inversion. For the first time, reliable inversion of ocean currents by GNSS-R is achieved within an operationally feasible framework, reducing ocean current errors by more than 50%.

[0046] (4) Establish an end-to-end ocean current inversion system that can be operated commercially. The entire method is based on existing hardware and open-source models (such as ROMS and LETKF), requires no new infrastructure, and the output products are compatible with mainstream ocean forecasting and emergency platforms, and have strong engineering applicability.

[0047] like Figure 1 As shown, based on the same inventive concept, this embodiment of the invention also provides a sea surface current inversion system for implementing the method, comprising: The data receiving unit 101 is used to acquire ground wave radar radial flow velocity data and raw GNSS reflection signal data; it is responsible for pulling HFR radial flow files and GNSS-R Level-1 data from various data sources in real time or at regular intervals.

[0048] The preprocessing and AI interpolation module 102 is connected to the data receiving unit and is used for quality control and neural network-based data missing interpolation of the ground wave radar data; it includes a quality control submodule and a U-Net++ neural network interpolation submodule, and is used to perform step S2 in embodiment 1.

[0049] The physical inversion engine 103, connected to the data receiving unit, is used to store and run the wind-wave-current coupled forward model to extract ocean current information from GNSS-R signals. It incorporates the wind-wave-current coupled forward model and the 4D-Var inversion algorithm to initially invert ocean current information from GNSS-R data (performing the core calculation of step S3 in Example 1).

[0050] The adaptive assimilation computation unit 104 is connected to the preprocessing and AI interpolation module and the physical inversion engine, respectively, and is used to run the regional ocean numerical model and the ensemble Kalman filter algorithm to realize dynamic weight allocation and fusion of observation data. As the core computation unit, it is equipped with a regional ocean numerical model (such as ROMS) and the LETKF assimilation program. Based on the configured dynamic weight strategy, this unit calls the observation data processed by modules 102 and 103, executes step S4 in Example 1, and completes the fusion inversion.

[0051] Product service and interface unit 105, connected to the adaptive assimilation calculation unit, is used to output fused ocean current products and quality assessment results. It is responsible for formatting and encapsulating the raw results output by unit 104 to generate the final product (ocean current field, uncertainty field), and publishing it externally via a network interface, while also managing the quality assessment process. After assimilation, the system outputs a surface (0–1 m) ocean current vector field (u, v) in NetCDF format, and simultaneously generates an uncertainty field and a quality mask 106. The product is automatically published by product service and interface unit 105 via API or FTP, with an update frequency configurable from 5–60 minutes. The system also automatically generates a quality assessment report daily, comparing the fused product with global drifting buoy trajectories.

[0052] During implementation, data receiving unit 101 first acquires two types of observation data: one is radial current velocity observations from shore-based high-frequency ground wave radar (HFR), with a typical time resolution of 10 minutes and a spatial coverage range of 10–200 km from the shore; the other is raw GNSS-R reflected signal data from low-Earth orbit satellites (such as CYGNSS), including complex signals in the L1 / L2 / L5 bands, satellite orbit parameters, and the location of the receiving platform. Data can be acquired through local receiving stations or downloaded from public platforms such as the NOAA HFR network or NASA PO.DAAC.

[0053] Subsequently, the preprocessing and AI interpolation module 102 processes the two types of data separately. For HFR data, the system performs multi-level quality control: first, data points with an absolute velocity greater than 2.5 m / s or a signal-to-noise ratio less than 3 dB are removed; second, 3σ outliers are removed in a ±30 minute window over time; if multi-station coverage exists, vector consistency is verified and observations with directional deviations exceeding 45° are removed. For regions still missing after quality control, a U-Net++ neural network is used for interpolation. This network contains an encoder-decoder structure with dense skip connections. The input is a 7×7 spatial neighborhood and the flow field and missing mask for the preceding and following three time steps, and the output is the reconstructed radial velocity. The network is pre-trained on historical high-quality HFR data, and the interpolation process is completed on a GPU server, with a single processing time of less than 2 minutes.

[0054] For GNSS-R data, the Physics Inversion Engine 103 is a wind-wave-current coupled scattering forward model specifically built for ocean current inversion based on classical electromagnetic scattering theory. The model first describes the sea surface state, characterized as a two-scale structure consisting of long waves and short waves. The long waves utilize a modified Pierson-Moskowitz spectrum, with wind speeds at 10 meters above the ground. The propagation of the sea surface wave is driven by the Elfouhaily directional spectrum, which is modulated by the slope of the long wave. The key point is that the propagation of the sea surface wave does not occur on a static background, but is superimposed on the ocean current. The phase velocity of the sea surface wave is corrected to the sum of the original phase velocity and the projection of the ocean current in the direction of wave propagation.

[0055] The system design is entirely based on open-source tools (ROMS, PyTorch) and existing observation networks, requiring no new infrastructure. It has already achieved operational trial run in the East China Sea demonstration area, with product updates occurring every 10 minutes, meeting emergency needs such as maritime search and rescue. All modules can be implemented on a general-purpose computing platform. Ground-wave radar and GNSS-R are existing observation methods, while ocean models (such as ROMS) and AI frameworks (such as PyTorch) are open-source tools. The entire process can be automated and is suitable for operational scenarios such as marine environmental monitoring, maritime search and rescue, and oil spill response.

[0056] While embodiments or examples of this disclosure have been described with reference to the accompanying drawings, it should be understood that the above embodiments are merely exemplary embodiments or examples, and the scope of the invention is not limited by these embodiments or examples, but only by the granted claims and their equivalents. Various elements in the embodiments or examples may be omitted or replaced by their equivalents. Furthermore, the steps may be performed in a different order than that described in this disclosure. Further, various elements in the embodiments or examples may be combined in various ways. Importantly, as the technology evolves, many elements described herein can be replaced by equivalents that appear after this disclosure.

Claims

1. A method for inverting ocean surface currents by integrating ground-wave radar and GNSS reflection, characterized in that, include: S1. Acquire radial velocity observation data from ground wave radar and raw GNSS reflection signal data; S2. For the spatiotemporal regions where the radial velocity observation data is missing, a pre-trained spatiotemporal neural network model is used for interpolation to reconstruct the complete radial velocity field. S3. Based on the dual-scale sea surface scattering theory, a wind-wave-current coupled forward model is constructed to characterize the GNSS reflected signal jointly modulated by wind speed, wave spectrum and ocean current velocity. S4. Using the background current field provided by the regional ocean numerical model as the state space, and based on the distance from the grid point to the shore in the regional ocean numerical model, dynamically allocate the weights of the radial current observations from the ground wave radar and the ocean current information obtained from the GNSS-R signal based on the forward model, and use the ensemble Kalman filter algorithm to assimilate and fuse the data. S5. Output the assimilated surface ocean current vector field and its uncertainty field.

2. The method according to claim 1, characterized in that, In step S1, the raw data of the GNSS reflected signal includes complex signals of the L1, L2 and / or L5 frequency bands, satellite orbit parameters and receiving platform location information.

3. The method according to claim 1, characterized in that, In step S2, the spatiotemporal neural network model is a U-Net++ network. The input includes a spatial neighborhood matrix centered on the missing point, the velocity sequence of multiple time steps before and after the missing point, and the missing mask. The output is the reconstructed radial velocity.

4. The method according to claim 1, characterized in that, In step S3, the wind-wave-current coupled forward model is specifically as follows: Introducing ocean current velocity into sea surface wave dynamics allows the phase velocity of sea surface waves to be corrected to the sum of the phase velocity when there is no current and the projection of the ocean current in the wave propagation direction. The theoretical Doppler shift Δf_model consists of three parts: platform motion geometry term Δf_plat, wave trajectory velocity term Δf_wave, and ocean current advection term Δf_current: Δf_model = Δf_plat + Δf_wave(v, U10) + Δf_current(u_c), where v is the wave spectrum parameter, U10 is the wind speed, and u_c is the ocean current velocity; The additional interference phase φ_current caused by ocean currents is expressed as: φ_current = (4π / λ) * (u_c · r ⊥ )* Δt, where λ is the signal wavelength, r ⊥ Let Δt be the horizontal displacement vector perpendicular to the line of sight, and let Δt be the coherent integration time.

5. The method according to claim 1, characterized in that, In step S4, the strategy for dynamically allocating weights is as follows: When d < 50 km, only radial flow observations from ground wave radar are assimilated; When 50 ≤ d ≤ 150 km, the ground wave radar observation weight w_HFR = 1 - (d - 50) / 100, and the GNSS-R inversion information weight w_GNSS = 1 - w_HFR; When d > 150 km, only ocean current information retrieved from GNSS-R is assimilated; d represents the distance from the grid point in the model to the shore.

6. The method according to claim 1, characterized in that, Step S5 also includes comparing the output ocean current product with the drifting buoy trajectory to generate a quality assessment report.

7. The method according to claim 1, characterized in that, Step S2 also includes quality control of the radial velocity observation data: removing data points with an absolute velocity value greater than 2.5 m / s or a signal-to-noise ratio less than 3 dB; removing 3σ outliers within a ±30 minute time window; verifying vector consistency in the multi-station coverage area and removing observations with directional deviations exceeding 45°.

8. The method according to claim 1, characterized in that, The wind-wave-current coupled forward model described in step S3 represents the sea surface as a two-scale structure: long waves are described by the improved Pierson-Moskowitz spectrum and driven by a 10-meter-high wind speed, while short waves use the Elfouhaily directional spectrum and are modulated by the long wave slope; ocean currents are superimposed on the sea surface wave propagation background, and the corrected sea surface wave phase velocity is the sum of the original phase velocity and the projection of the ocean current in the wave propagation direction.

9. The method according to claim 1, characterized in that, The ensemble Kalman filter algorithm adopts local ensemble transformation Kalman filtering. The localization radii of the ground wave radar and GNSS-R are set to 20 km and 50 km respectively, the assimilation period is 30 minutes, the horizontal resolution of the ocean numerical model is 1 km, and the vertical layering is no less than 50 layers.

10. A sea surface current inversion system for implementing the method according to any one of claims 1-9, characterized in that, include: The data receiving unit is used to acquire radial velocity data from ground wave radar and raw data from GNSS reflection signals; A preprocessing and AI interpolation module, connected to the data receiving unit, is used for quality control of ground wave radar data and data missing interpolation based on neural networks; The physical inversion engine, connected to the data receiving unit, is used to store and run the wind-wave-current coupled forward model to extract ocean current information from GNSS-R signals; An adaptive assimilation computation unit is connected to the preprocessing and AI interpolation module and the physical inversion engine, respectively, and is used to run the regional ocean numerical model and the ensemble Kalman filter algorithm to realize dynamic weight allocation and fusion of observation data. The product service and interface unit is connected to the adaptive assimilation calculation unit and is used to output fused ocean current products and quality assessment results.

Images