A swot significant wave height inversion method fusing scattering and interference characteristics

Abstract

This invention discloses a SWOT effective wave height inversion method that integrates scattering and interferometric features. The method includes determining model input parameters for effective wave height inversion, including backscattering correlation parameters and interferometric correlation parameters; acquiring SWOT KaRIn L1B level single-look complex data; dividing the observation area into multiple sub-image regions based on satellite orbit information and imaging geometry parameters; calculating the backscattering correlation parameters corresponding to each sub-image region and extracting the interferometric correlation parameters at the corresponding spatial locations; constructing complete model input feature parameters; performing quality control and screening on the model input feature parameters; matching the screened feature parameters with effective wave height data obtained from other satellites in time and space to construct a deep learning training sample set; training an LSTM-based effective wave height inversion model; and using the trained effective wave height inversion model for inversion. This invention can more fully adapt to the observation characteristics of SWOT KaRIn data, and the inversion results are stable and reliable.

Description

Technical Field

[0001] This invention belongs to the field of satellite remote sensing, ocean observation and data inversion, and specifically relates to a SWOT effective wave height inversion method that integrates scattering and interferometric features. Background Technology

[0002] Significant wave height is one of the important parameters characterizing sea surface wave features, and it is widely used in marine engineering design, climate change research, marine disaster early warning, and maritime traffic safety. Significant wave height is usually defined as the average of the highest one-third of the recorded wave heights under specific time and location conditions, and it is a key physical quantity reflecting the strength of sea state.

[0003] The Ka-band radar interferometer aboard the SWOT satellite possesses high spatial resolution and wide swath observation capabilities, and its L2-level product already provides effective wave height parameters. Existing SWOT KaRIn effective wave height products are primarily estimated based on the volume coherence coefficient of interferometry. They obtain effective wave height parameters through a least-squares method combined with an inversion relationship highly sensitive to interferometric phase pairs, and are released as L2-level products of KaRIn. This method represents a breakthrough in obtaining effective wave height from an interferometric perspective, demonstrating a certain degree of innovation. However, existing effective wave height inversion methods mainly rely on interferometric parameters and do not fully utilize the backscattering information simultaneously acquired by KaRIn. In fact, KaRIn's L1B-level data still contains backscattering-related parameters such as normalized radar cross section, incident angle, and image spectrum, which have been proven to have significant correlations with wave characteristics in traditional SAR wave inversion. Due to the limited availability of interferometric imaging radar data for previous research, the potential value of these backscattering-related parameters in KaRIn effective wave height inversion has not yet been fully explored. SWOT analysis shows that data quality is relatively high near the nadir point in KaRIn along-orbit observations, but accuracy is prone to degradation in areas far from the nadir point due to variations in geometry and signal-to-noise ratio. Summary of the Invention

[0004] Existing methods for retrieving effective wave height mainly rely on single radar backscattering characteristics or empirical statistical models, failing to fully utilize the backscattering and interferometric information simultaneously contained in SWOT KaRIn data. Furthermore, under complex sea conditions, a single information source has limited ability to characterize wave scale, wave spatial inhomogeneity, and volume scattering effects, resulting in insufficient accuracy and poor stability in effective wave height retrieval, making it difficult to meet the requirements of high-resolution ocean observation and refined applications. This invention proposes a SWOT effective wave height retrieval method that integrates scattering and interferometric features. The specific technical solution is as follows: A SWOT effective wave height inversion method that integrates scattering and interference features includes the following steps: S1: Determine the model input parameters for effective wave height inversion, including backscattering correlation parameters and interferometric correlation parameters; acquire SWOT KaRIn L1B level single-look complex data, divide the observation area into multiple sub-map regions according to satellite orbit information and imaging geometry parameters, calculate the backscattering correlation parameters corresponding to each sub-map region, extract the interferometric correlation parameters corresponding to the spatial location, and construct complete model input feature parameters; S2: Perform quality control and screening on the input feature parameters of the model; S3: Match the KaRIn feature parameters filtered in S2 with the effective wave height data obtained from other satellites in time and space to construct a training sample set for deep learning; S4: Construct an effective wave height inversion model based on an LSTM neural network. Use the backscattering features obtained in S3 and the interference correlation parameters of the corresponding spatial locations as inputs to the effective wave height inversion model, and use the matched effective wave height as output. Train the effective wave height inversion model using the backpropagation algorithm. S5: Use the effective wave height inversion model trained in S4 for effective wave height inversion.

[0005] Furthermore, the backscattering related parameters include the normalized radar cross section, radar incident angle, normalized image variance, and image spectral parameters.

[0006] Furthermore, the interference-related parameters include the volume scattering coherence parameters obtained from KaRIn interferometry.

[0007] Furthermore, in S1, the normalized radar cross section is calculated based on radar scattering theory, and the corresponding radar incident angle is extracted according to the imaging geometric parameters.

[0008] Further, step S2 includes the following sub-steps: S2.1: Combine geographic masking to remove data from land and nearshore areas; S2.2: Use rainfall indicator information to remove samples that are significantly affected by rainfall; S2.3: Set a physically reasonable range for each feature parameter and remove outlier samples.

[0009] Furthermore, the lower threshold of the normalized radar cross section is 10 dB, the range of the radar incident angle is set to 0.5° to 4°, the upper threshold of the normalized image variance is 2, and the lower threshold of the volume scattering coherence parameter is 0.3.

[0010] Furthermore, the effective wave height inversion model based on LSTM neural network includes an input layer, a hidden layer, and an output layer. The hidden layer includes four sub-layers. The first and second sub-layers are both long short-term memory neural network structures, used to extract temporal feature information from the input time series. The third and fourth sub-layers are fully connected layers, used to perform further nonlinear mapping on the features extracted by the first and second sub-layers.

[0011] A SWOT effective wave height inversion device that integrates scattering and interference features includes one or more processors for implementing the SWOT effective wave height inversion method that integrates scattering and interference features.

[0012] An electronic device, comprising: One or more processors; A storage device for storing one or more programs that, when executed by the electronic device, enable the electronic device to implement a SWOT effective wave height inversion method that integrates scattering and interference characteristics.

[0013] A computer-readable storage medium having a program stored thereon that, when executed by a processor, implements a SWOT effective wave height inversion method that integrates scattering and interference features.

[0014] The beneficial effects of this invention are as follows: 1. This invention enhances the comprehensive characterization of ocean wave scattering and spatial structure features by fusing SWOT KaRIn backscattering characteristics with volume scattering coherence information obtained from interferometry, thereby improving the physical consistency and comprehensiveness of significant wave height inversion. Compared to traditional altimeter or interferometric radar inversion methods based on a single observation, this invention realizes a comprehensive inversion framework under the synergistic constraints of multiple observation mechanisms, which can more fully adapt to the observation characteristics of SWOT KaRIn data and exhibits better stability and adaptability under complex sea conditions.

[0015] 2. This invention employs a feature extraction and quality control method based on subgraph scale, which effectively reduces the impact of noise, rainfall, and abnormal samples on the inversion results, thereby improving the stability and reliability of the inversion results.

[0016] 3. By introducing the LSTM deep learning model, the nonlinear mapping relationship between multiple parameters is fully explored, improving the accuracy of effective wave height inversion under complex sea conditions. Transforming two-dimensional high-resolution KaRIn spatial structure data into a time-series sample sequence allows LSTM to not only model temporal variations but also learn the implicit correlations between spatial structures; this is a "space-time reconstruction input design."

[0017] 4. The method of this invention is applicable to large-scale, high-resolution SWOT KaRIn data processing, and has good versatility and engineering application value. This invention achieves adaptive compensation for cross-track observation differences in SWOT KaRIn by constructing a multi-source information fusion and LSTM spatiotemporal modeling mechanism, improving the stability of data inversion in regions far from the nadir point and enhancing its large-scale application capability under wide-swath observation conditions. Attached Figure Description

[0018] Figure 1 This is a flowchart of a SWOT effective wave height inversion method based on the fusion of scattering and interference features, according to one embodiment of the present invention.

[0019] Figure 2 This is a case study of SWOT satellite KaRIn subplot data provided in this embodiment. Figure 2 In the diagram, (a) represents the normalized radar cross section data. Figure 2 (b) in the image is the spectral data extracted from (a).

[0020] Figure 3 The SWOT analysis results provided in this embodiment are the effective wave height of the product with high precision. Figure 3 (a) in the diagram is a schematic diagram of the effective wave height of the SWOT product. Figure 3 (b) in the figure shows the comparison between the inversion results and the ECMWF model data.

[0021] Figure 4 The SWOT analysis results provided in this embodiment are the effective wave height of the product with high precision. Figure 4 (a) is a schematic diagram of the effective wave height of the SWOT product, and (b) is a comparison diagram of the inversion results and HY-2C satellite data.

[0022] Figure 5 Spatial comparison results of the effective wave height inversion results of the SWOT product provided in this embodiment and the HY-2C satellite data.

[0023] Figure 6 The effective wave height inversion results of the SWOT product provided in this embodiment are compared with the changes in COR and RMSE of the KaRIn L2 product under different incident angles and two polarization modes. Detailed Implementation

[0024] The present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. The purpose and effects of the present invention will become clearer. It should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0025] Explanation of technical terms SWOT: Surface Water and Ocean Topography (SMS) NRCS: Normalized Radar Cross Section; SLC: Single Look Complex; SWH: Significant Wave Height, effective wave height; ECMWF, European Centre for Medium-Range Weather Forecasts; LSTM, Long Short-Term Memory.

[0026] like Figure 1 As shown, the SWOT effective wave height inversion method that combines scattering and interference features in this embodiment includes the following steps: Step 1: Determine the model input parameters for effective wave height inversion, including backscattering correlation parameters and interferometric correlation parameters; acquire SWOT KaRIn L1B level single-look complex data; divide the observation area into multiple sub-image regions based on satellite orbit information and imaging geometry parameters; calculate the backscattering correlation parameters corresponding to each sub-image region; extract the interferometric correlation parameters corresponding to the spatial location; and construct complete model input feature parameters.

[0027] Among them, the backscattering related parameters include the normalized radar cross section (NRCS), radar incident angle, normalized image variance, and image spectrum parameters; the interferometric related parameters include the volume scattering coherence parameters obtained by KaRIn interferometry.

[0028] In this embodiment, SWOT KaRIn L1B level Single Look Complex (SLC) data is first acquired. Based on satellite orbit information and imaging geometric parameters, the observation area is divided into several sub-map regions with a spatial scale of 5 km × 5 km, as shown in Figure 2, to ensure that each sub-map region has a sufficient number of statistical samples to meet the requirements of subsequent feature extraction and inversion model training.

[0029] For each sub-image region, this invention extracts model input parameters from KaRIn L1B SLC data, including normalized radar cross section (NRCS), incident angle, normalized image variance, and image spectral features. Except for the L2 volume scattering coherence parameter, all other parameters are derived from L1B data.

[0030] In this embodiment, within each sub-map region, the normalized radar cross section is calculated based on radar scattering theory, and the corresponding radar incident angle is extracted according to the imaging geometry parameters as a key input feature parameter for SWH inversion.

[0031] Numerous studies have demonstrated a significant correlation between NRCS and sea surface significant wave height. Furthermore, NRCS is an effective indicator for describing the reflectivity of extended targets or the sea surface, reflecting sea surface fluctuations and roughness characteristics. Meanwhile, numerous experimental studies show that radar backscattered energy varies proportionally with the incident angle. According to the quasi-specular reflection model, NRCS is modulated by the incident angle; that is, under the same sea conditions, changes in the incident angle will cause corresponding changes in the observed NRCS value. Therefore, in this embodiment, NRCS and the radar incident angle are used as key input parameters to construct the SWH inversion model. By incorporating incident angle information, the influence of radar imaging geometry on scattered energy can be effectively reflected, thereby enhancing the model's adaptability to sea surface wave height under different imaging conditions and its inversion accuracy.

[0032] This invention extracts NRCS and radar incident angle parameters, which not only makes full use of radar echo intensity information, but also comprehensively considers the modulation effect of imaging geometry on the observation data, providing a reliable feature basis for subsequent multi-source data fusion and deep learning, thereby achieving high-precision and high-robust effective wave height inversion.

[0033] The formula for calculating NRCS is as follows:

[0034] Where σ0 represents the normalized radar cross section, θ Indicates the radar incident angle. R (0) is the Fresnel reflection coefficient under normal incidence conditions, and its square value is R (0) 2 =0.631; It represents the root mean square slope (or mean square slope) of the sea surface.

[0035] Image variance is a statistical measure used to describe the degree of variation in image pixel values. Ocean waves can alter the variation characteristics of radar images through various modulation mechanisms, including tilt modulation, hydrodynamic modulation, and velocity beamforming. Therefore, there is a correlation between image variance and ocean wave parameters. By analyzing the spatial variation characteristics of image variance, ocean wave parameters can be retrieved or inferred. For the above reasons, normalized image variance (nv) is selected as the model input parameter, and its calculation formula is as follows:

[0036] in, I Indicates image intensity, The mean value of the image intensity. This indicates variance calculation.

[0037] Theoretically, radar image spectra can be viewed as modulations of ocean wave spectra, thus containing rich ocean wave information. However, directly introducing the complete image spectrum into the inversion model significantly increases computational complexity, hindering efficient model implementation. Therefore, this embodiment employs an orthogonal function-based method to extract feature parameters from the image spectrum, replacing direct use of the image spectrum. The extracted spectral parameters can characterize most of the spectral information and are computationally more efficient than directly using the complete spectrum. This embodiment performs frequency domain transformation on the sub-image data and uses orthogonal function expansion to extract image spectral parameters to characterize wave scale distribution features.

[0038] like Figure 2 As shown in (a), sea surface ripples can be clearly observed in each 5 km × 5 km sub-map region, indicating that L1B SLC data can reflect the spatial structure characteristics of sea surface fluctuations.

[0039] The image spectrum was obtained by performing a Fourier transform on the sub-image region, as shown in Figure 2(b). Although there is a slight truncation in the azimuth direction, similar to traditional synthetic aperture radar (SAR) imaging, the overall spectrum remains intact without any abrupt changes at the spectrum edges. This indicates that the KaRIn L1B SLC data can still preserve sea surface wave information well in the spatial frequency domain and has the potential to be used for effective wave height (SWH) observation.

[0040] Ocean waves form a volume scattering layer within the angular range corresponding to the range resolution unit, thereby causing volume scattering-related phenomena. This phenomenon arises because the irregular undulations of the ocean surface cause electromagnetic signals to be scattered at different incident and scattering angles within the resolution unit, leading to reduced coherence as the signal interacts with the wave structure. Based on this, this invention extracts volume scattering coherence parameters corresponding to the spatial location of the target submap region from the SWOT KaRIn second-order interferometry product during the interferometric feature extraction process. These parameters are used to characterize the volume scattering characteristics of ocean waves and their vertical structure information.

[0041] Step 2: Perform quality control and screening on the model input feature parameters.

[0042] To ensure the robustness and accuracy of the inversion model, a three-level quality control process is employed to rigorously control and screen the initial sample set before inputting the extracted multidimensional feature vectors into the deep learning model. Therefore, step two specifically includes the following sub-steps: S2.1; Combine geographic masking to remove data from land and nearshore areas.

[0043] In this embodiment, high-precision seabed topographic data (such as ETOPO1) is used to identify and remove observation samples located on land and in nearshore areas with water depths shallower than -50 meters. This step aims to avoid interference from complex nearshore dynamic processes (such as wave refraction, diffraction, and bottom friction effects) and land contamination of radar signals on sea surface wave parameter inversion, ensuring that the training data represent the sea conditions of open waters.

[0044] S2.2: Use rainfall indicator information to remove samples that are significantly affected by rainfall.

[0045] Radar signals, especially Ka-band signals, are significantly affected by atmospheric precipitation attenuation during propagation. Therefore, this embodiment identifies and removes observation samples with precipitation rates exceeding a set threshold by matching rainfall rate products from contemporaneous European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis data. This eliminates or reduces the impact of precipitation attenuation on non-wave factors affecting radar backscattering and interferometric coherence characteristics, thereby improving data quality.

[0046] S2.3: Set a physically reasonable range for each feature parameter and remove outlier samples.

[0047] This embodiment sets reasonable thresholds for key input parameters based on the physical mechanisms of radar ocean remote sensing and statistical analysis of historical observation data. For example: The lower threshold of the normalized radar cross section (NRCS) is set to 10 dB to eliminate data samples that may indicate invalid observations or special sea conditions due to abnormally low scattering intensity. This threshold can effectively exclude abnormal measurements caused by weak radar signals or extreme sea surface conditions, ensuring that the KaRIn data used has a reliable signal strength basis in the effective wave height inversion process, thereby improving inversion accuracy and model stability.

[0048] By setting the radar incident angle range to 0.5° to 4°, observation data caused by excessively small or large incident angles are discarded. This measure avoids imaging distortions that may occur under low incident angle conditions, as well as the impact of radar signal attenuation and scattering geometry effects on the inversion results under high incident angle conditions, thereby ensuring the consistency of the data used for inversion in terms of spatial geometry and signal strength.

[0049] Setting an upper threshold of 2 for the normalized image variance is used to remove samples whose excessively high image variance may be caused by data anomalies, strong ocean phenomena (such as storms and swells), or residual system noise. This threshold filtering ensures the stability of the spatial statistical properties of the data used, reduces the interference of outlier samples on the training and prediction results of the inversion model, and improves the reliability and repeatability of the inversion results.

[0050] The lower threshold for KaRIn L2 volume correlation is set to 0.3 to exclude observational data with minimal volume scattering contribution or insufficient interferometric coherence. This threshold ensures that the data used has sufficient interferometric coherence within the radar resolution cell, resulting in a high signal-to-noise ratio and consistency when retrieving significant wave height. By excluding low-correlation data, the stability of model training can be significantly improved, error accumulation reduced, and the inversion results maintained at a high accuracy under different sea states and incident angles.

[0051] By setting physically reasonable ranges for each feature parameter, outliers caused by instrument noise, processing artifacts, or extremely rare sea conditions can be effectively filtered out, making the distribution of the training dataset more concentrated and reasonable.

[0052] After completing the three-level quality control in step two, a high-quality, clean matching dataset is obtained. Subsequently, the remaining samples (200,000 groups) are divided into training and validation sets (e.g., in an 8:2 ratio) according to time sequence to ensure the independence of the model training and evaluation processes. This quality control process is a crucial prerequisite for building a high-precision inversion model, laying a reliable data foundation for the effective learning of subsequent deep learning models.

[0053] Step 3: Match the KaRIn feature parameters selected in Step 2 with the effective wave height data obtained from other satellites in time and space to construct a training sample set for deep learning.

[0054] In this embodiment, the KaRIn feature parameters selected in step two are matched temporally and spatially with the significant wave height (SWH) in the ECMWF reanalysis data and the significant wave height data observed by the HY-2C satellite altimeter. Specifically, the temporal difference between the KaRIn data and the HY-2C altimeter data is required to be no more than 0.5 hours, and the spatial difference is required to be no more than 5 km, thereby ensuring the spatiotemporal consistency and reliability of the matched samples. Simultaneously, the ECMWF reanalysis data is used to provide reference environmental conditions and sea surface state information, further enhancing the completeness and representativeness of the training samples.

[0055] This multi-source data matching method effectively integrates the advantages of satellite observation data and reanalysis data, providing high-quality input features and corresponding real labels for deep learning models, improving the SWH inversion accuracy and generalization ability of KaRIn data, and thus ensuring that the inversion method described in this invention has stable and reliable performance under different sea states, different polarization modes and different incident angles.

[0056] Step 4: Construct an effective wave height inversion model based on an LSTM neural network. Use the backscattering features obtained in Step 3 and the corresponding spatial interference correlation parameters as inputs to the effective wave height inversion model, and use the matched effective wave height as the output. Train the effective wave height inversion model using the backpropagation algorithm.

[0057] The effective wave height inversion model consists of three parts: an input layer, a hidden layer, and an output layer. The input layer receives the model's input feature parameters, the hidden layer includes multiple sub-layers, and is used to extract temporal features and nonlinear mapping relationships from the input features. The output layer includes one neuron, which outputs the effective wave height inversion result for the corresponding time position.

[0058] In this embodiment, the input layer comprises 24 neurons, corresponding to 24 model input parameters. These parameters include the normalized radar cross section, radar incident angle, normalized image variance, image spectral parameters, and volume scattering coherence. The input parameters are arranged in the order of continuous observations along the KaRIn orbit to form a time-series input effective wave height inversion model based on an LSTM neural network.

[0059] In this embodiment, the hidden layer is divided into four sub-layers, namely: a first LSTM sub-layer and a second LSTM sub-layer, both of which are long short-term memory neural network structures, each containing 50 neurons, used to extract temporal feature information from the input time series; the third and fourth sub-layers are fully connected layers, containing 20 and 10 neurons respectively, used to further nonlinearly map the features extracted by the LSTM sub-layers. Both the LSTM sub-layers and the fully connected sub-layers use the Rectified Linear Unit (ReLU) as the activation function to enhance the nonlinear expressive power of the model and improve training stability.

[0060] In this embodiment, the initial learning rate is set to 0.01 during the model training phase. To improve the stability and convergence of the model training, a dynamic learning rate decay strategy is adopted during training. That is, after each training epoch, the current learning rate is adjusted to 95% of the previous epoch's learning rate. By gradually reducing the learning rate, the magnitude of model parameter updates gradually decreases during the training process, thereby improving the model's convergence performance. In this embodiment, the model training is performed for a total of 30 training epochs.

[0061] To reduce hardware computing resource consumption and shorten model training time, the model training adopts a batch training method, dividing the training samples into 64 batches, which are then input into the model sequentially for parameter updates.

[0062] By introducing the LSTM deep learning model, the nonlinear mapping relationship between multiple parameters can be fully explored, improving the accuracy of effective wave height inversion under complex sea conditions. At the same time, the two-dimensional high-resolution KaRIn spatial structure data is transformed into a time-series sample sequence, enabling the LSTM to not only model time variations but also learn the implicit correlations between spatial structures. This is a "space-time reconstruction input design".

[0063] Step 5: Use the effective wave height inversion model trained in Step 4 for effective wave height inversion.

[0064] Corresponding to the previous embodiments of the SWOT effective wave height inversion method that combines scattering and interference features, the present invention also provides an embodiment of a SWOT effective wave height inversion device that combines scattering and interference features.

[0065] The SWOT effective wave height inversion device for fusion scattering and interference features provided in this embodiment includes one or more processors for implementing the SWOT effective wave height inversion method for fusion scattering and interference features in the above embodiment.

[0066] The embodiments of the SWOT effective wave height inversion device integrating scattering and interference characteristics of the present invention can be applied to any device with data processing capabilities, such as a computer. The device embodiments can be implemented through software, hardware, or a combination of both. Taking software implementation as an example, as a logical device, it is formed by the processor of any data processing device loading the corresponding computer program instructions from non-volatile memory into memory for execution. From a hardware perspective, in addition to the processor, memory, network interface, and non-volatile memory, the data processing device in the embodiments typically includes other hardware depending on its actual functions; these will not be elaborated further.

[0067] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0068] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the present invention according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0069] This invention also provides a computer-readable storage medium storing a program that, when executed by a processor, implements the SWOT effective wave height inversion method based on the fusion of scattering and interference features described in the above embodiments.

[0070] In this embodiment, after training the LSTM-based inversion model, the constructed model is applied to independent KaRIn data for SWH inversion. Specifically, the feature parameters extracted from the KaRIn data, including normalized radar cross section (NRCS), incident angle, normalized image variance, image spectral parameters, and KaRIn L2 volume correlation, are input into the trained LSTM model to obtain the corresponding SWH inversion results.

[0071] To verify the accuracy of the inversion results, the retrieved significant wave height was compared with the spatiotemporally matched external reference significant wave height data. The external reference data included ECMWF reanalysis significant wave height data and HY2C altimeter significant wave height data. The correlation coefficient, root mean square error, and bias index were used to quantitatively evaluate the inversion results and verify the inversion performance of the method in this embodiment.

[0072] Figure 3 (a) shows the comparison results between the significant wave height data published in the SWOT KaRIn L2 product and the significant wave height data from the ECMWF reanalysis. Figure 3Table (b) shows the comparison between the KaRIn significant wave height obtained by the method of this embodiment and the ECMWF reanalysis significant wave height data. The comparison results show that the correlation coefficient between the significant wave height corresponding to the KaRIn L2 product and the ECMWF data is 0.95, the root mean square error is 0.45 m, and the deviation is -0.20 m; the correlation coefficient between the significant wave height obtained by the method of this embodiment and the ECMWF data is 0.96, the root mean square error is 0.22 m, and the deviation is 0.06 m. Statistical analysis with ECMWF model data shows that the inversion results of the method of this invention are significantly better than those of the L2 product in terms of accuracy and consistency.

[0073] Further comparative analysis with the HY2C altimeter is shown in Figure 4. Figure 4 (a) shows the comparison results between the SWH data published in the SWOT KaRIn L2 product and the SWH data observed by the HY2C satellite altimeter. Figure 4 (b) shows the comparison results between the KaRIn SWH and HY2C altimeter observation data obtained by the method of this embodiment. Statistical results show that the correlation coefficient between the SWH data released by the SWOT KaRIn L2 product and the HY2C altimeter observation data is 0.88, the root mean square error is 0.41 m, and the average deviation is -0.11 m; while the correlation coefficient between the KaRIn SWH and HY2C data obtained by the method of this invention is 0.97, the root mean square error is 0.20 m, and the deviation is close to 0 (0.00 m). This further illustrates that the inversion method combining radar backscattering characteristics and interferometric characteristics proposed in this embodiment can significantly improve the accuracy of effective wave height inversion compared to methods using only interferometric information, achieving higher accuracy and reliability.

[0074] This invention, by comprehensively utilizing multi-source information from KaRIn data, can obtain more accurate and stable SWH inversion results under different sea states and observation conditions, providing reliable data support for ocean dynamic monitoring and sea surface wave research.

[0075] In addition to data from the European Centre for Medium-Range Weather Forecasts, SWH data from the HY2C wave altimeter were also used to validate the wave height inversion results from KaRIn.

[0076] Figure 5 shows the spatial distribution of the KaRIn inversion results and the HY2C wave height data from the same period. As shown in the figure, their spatial variations are consistent, both showing a trend of gradually increasing wave height from south to north.

[0077] Through overall analysis, this invention further analyzes the inversion results under different polarization modes and incident angles. First, the observed data are classified according to vertical polarization (VV) and horizontal polarization (HH); then, the data under each polarization mode are segmented according to an incident angle interval of 0.5°. As shown in Figure 6, the figure illustrates the variation of the correlation coefficient (COR) and root mean square error (RMSE) of the inversion results of the KaRIn L2 product and this embodiment under different incident angles. The results show that under all polarization modes and incident angles, the COR is greater than 0.99 and the RMSE is approximately 0.14 m, indicating that both methods have good wave height inversion performance. Specifically, as shown in Figures 6(a) and (b), the two methods exhibit high correlation under all polarization modes and incident angles. However, the correlation coefficient of the KaRIn L2 product shows a significant decreasing trend with the increase of the incident angle, from about 0.98 to 0.90; in contrast, the inversion result proposed in this embodiment maintains a relatively stable correlation of about 0.95 throughout the entire incident angle range.

[0078] As shown in Figures 6(c) and (d), the overall RMSE levels of the two methods are comparable under different polarization modes. However, the RMSE of the KaRIn L2 product increases significantly with increasing incident angle, exceeding 0.5 m in the 3.0°–3.5° range; while the RMSE of the inversion results in this embodiment remains basically constant at approximately 0.25 m under different incident angles. Figure 6 The results demonstrate the adaptability of the method of the present invention to the observation differences at different cross-track positions, thereby mitigating the accuracy degradation phenomenon in areas far from the nadir point and improving the consistency and stability of the inversion results across the entire observation range.

[0079] The above results show that the LSTM-based inversion model that integrates interferometry and backscattering characteristics in this embodiment can effectively obtain sea surface SWH from KaRIn data under different polarization modes and incident angles, achieving high-precision and robust inversion performance.

[0080] It will be understood by those skilled in the art that the above descriptions are merely preferred examples of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the invention should be included within the scope of protection of the invention.

Claims

1. A SWOT significant wave height inversion method fusing scattering and interference features, characterized in that, Includes the following steps: S1: Determine the model input parameters for effective wave height inversion, including backscattering correlation parameters and interferometric correlation parameters; acquire SWOT KaRIn L1B level single-look complex data, divide the observation area into multiple sub-map regions according to satellite orbit information and imaging geometry parameters, calculate the backscattering correlation parameters corresponding to each sub-map region, extract the interferometric correlation parameters corresponding to the spatial location, and construct complete model input feature parameters; The backscattering related parameters include normalized radar cross section, radar incident angle, normalized image variance, and image spectral parameters; The interference-related parameters include the volume scattering coherence parameters obtained from KaRIn interferometry; S2: Perform quality control and screening on the input feature parameters of the model; S3: Match the KaRIn feature parameters filtered in S2 with the effective wave height data obtained from other satellites in time and space to construct a training sample set for deep learning; S4: Construct an effective wave height inversion model based on an LSTM neural network. Use the backscattering features obtained in S3 and the interference correlation parameters of the corresponding spatial locations as inputs to the effective wave height inversion model, and use the matched effective wave height as output. Train the effective wave height inversion model using the backpropagation algorithm. Transforming two-dimensional high-resolution KaRIn spatial structure data into time-series sample sequences enables LSTM to not only model temporal variations but also learn the implicit correlations between spatial structures. This is a "space-time reconstruction input design". S5: Use the effective wave height inversion model trained in S4 for effective wave height inversion.

2. The SWOT significant wave height inversion method fusing scatter and interference characteristics according to claim 1, characterized in that, In step S1, the normalized radar cross section is calculated based on radar scattering theory, and the corresponding radar incident angle is extracted according to the imaging geometric parameters.

3. The SWOT effective wave height inversion method based on the fusion of scattering and interference characteristics according to claim 1, characterized in that, S2 includes the following sub-steps: S2.1: Combine geographic masking to remove data from land and nearshore areas; S2.2: Use rainfall indicator information to remove samples that are significantly affected by rainfall; S2.3: Set a physically reasonable range for each feature parameter and remove outlier samples.

4. The SWOT effective wave height inversion method based on the fusion of scattering and interference characteristics according to claim 3, characterized in that, The lower threshold of the normalized radar cross section is 10 dB, the range of the radar incident angle is set to 0.5° to 4°, the upper threshold of the normalized image variance is 2, and the lower threshold of the volume scattering coherence parameter is 0.

3.

5. The SWOT effective wave height inversion method based on fused scattering and interference characteristics according to claim 1, characterized in that, The effective wave height inversion model based on LSTM neural network includes an input layer, a hidden layer, and an output layer. The hidden layer includes four sub-layers. The first and second sub-layers are both long short-term memory neural network structures, used to extract temporal feature information from the input time series. The third and fourth sub-layers are fully connected layers, used to perform further nonlinear mapping on the features extracted by the first and second sub-layers.

6. A SWOT effective wave height inversion device that integrates scattering and interference characteristics, characterized in that, It includes one or more processors for implementing the SWOT effective wave height inversion method based on the fusion of scattering and interference features as described in any one of claims 1 to 5.

7. An electronic device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by the electronic device, cause the electronic device to implement the SWOT effective wave height inversion method based on fused scattering and interference features as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, It stores a program that, when executed by a processor, implements the SWOT effective wave height inversion method based on the fusion of scattering and interference features as described in any one of claims 1 to 5.

Images