A method for retrieving SAR range velocity from SAR ocean product data

Abstract

The application discloses a method for inverting SAR distance-direction flow velocity from SAR marine product data, which comprises the following steps: acquiring marine product data and HYCOM data of a synthetic aperture radar; determining a mapping relationship between DCA Doppler frequency shift and Stokes drift, wind speed and flow velocity in the SAR distance direction by using the marine product data and the HYCOM data; constructing an inversion model for inverting the SAR distance-direction flow velocity from the marine product data based on deep learning according to the mapping relationship; verifying the accuracy of the inversion model; and predicting the SAR distance-direction flow velocity inverted from the marine product data of the synthetic aperture radar by using the model. The method is used for exploring the relationship between the DCA Doppler frequency shift and the Stokes drift, the wind speed and the flow velocity in the SAR distance direction by using OCN and HYCOM data, and solves the problem that the flow velocity in the SAR distance direction is difficult to invert.

Description

Technical Field

[0001] This invention relates to the field of synthetic aperture radar technology, specifically to a method for constructing a model for inverting range-direction current velocity in marine products using synthetic aperture radar. Background Technology

[0002] In fact, sea surface currents are an important aspect of upper ocean dynamics, and ocean circulation plays a crucial role in climate change. Although ocean currents are typically observed using shipboard instruments, global ocean current monitoring remains a significant challenge. Synthetic Aperture Radar (SAR) is well-known for its ability to provide high spatial resolution sea surface monitoring, specifically 8-megapixel images acquired by China's Gaofen-3 (GF-3) satellite in four-polarization mode. Over the past few decades, algorithms for extracting upper ocean dynamics and detecting targets—wind, waves, sea ice, and oil spills—from SAR imagery have been significantly developed, particularly for extreme ocean conditions and torrential rains during tropical cyclones (TCs).

[0003] In principle, ocean current-induced shearing and convergence affect sea state, leading to changes in SAR roughness, a behavior clearly observable in SAR images of frontal dominance areas. Assume two methods are applicable to current velocity inversion from SAR images along the radar line of sight. One technique, denoted as along-orbit INSAR (ATI), relies on phase information between two receiving antennas. For single-beam SAR, the Doppler shift frequency associated with the Doppler Centroid Anomaly (DCA) helps explain the relationship between backscattering roughness patterns and upper ocean dynamics, as both short-period waves and surface currents influence the DCA. In the literature, current velocities in the radar observation direction can be inverted from the DCA because the wave contribution is known at that time. Since 2016, Sentinel-1 (S-1), operating in the C-band as the successor to Envisat-ASAR, has officially provided DCA, wind, and wave inversion information stored in Oceanographic Nomenclature (OCN) products. This provides an opportunity to quantitatively analyze the dependence of upper ocean dynamics on DCA. Summary of the Invention

[0004] Based on previous research, in the context of the big data era, Sentinel-1 (S-1), operating in the C-band, provides Oceanographic Content Array (OCN) product data. This invention utilizes DCA, wind, and wave inversion information from the OCN product data to explore the dependencies between these information and sea surface currents. Based on this, a practical SAR range-up current velocity inversion algorithm is proposed. This algorithm was applied to all images and validated with HYCOM data. The technical solution is as follows:

[0005] A method for retrieving SAR range current velocity from SAR marine product data, comprising:

[0006] Acquire marine product data and HYCOM data from synthetic aperture radar;

[0007] Using marine product data and HYCOM data, the mapping relationship between DCA Doppler frequency shift and Stokes drift, wind speed and current velocity in the SAR range direction was determined;

[0008] Based on the mapping relationship, a deep learning-based inversion model for SAR range velocity inversion from marine product data is constructed.

[0009] Verify the accuracy of the inversion model;

[0010] The model is used to predict the SAR range current velocity based on marine product data from synthetic aperture radar.

[0011] A further improvement of this invention lies in acquiring marine product data and HYCOM data from synthetic aperture radar, specifically including:

[0012] Sentinel-1 OCN data was collected as marine product data; the OCN data included radar echo frequency shift at the beam center (called measured Doppler shift), predicted Doppler shift related to satellite velocity and Earth rotation, wind speed and wave parameters retrieved by SAR; HYCOM data consisted of sea surface current fields simulated using the open data source HYCOM, with a spatial resolution of ~0.08° and an interval of 3 hours.

[0013] A further improvement of this invention lies in determining the relationship between DCA Doppler frequency shift and Stokes drift velocity, wind speed, and current velocity in the SAR range direction using marine product data and HYCOM data, specifically including:

[0014] The acquired marine product data and HYCOM data are used to calculate the corresponding DCA Doppler frequency shift and Stokes drift velocity, wind speed, and current velocity based on the variables in the marine product data and HYCOM data. These are all decomposed into the SAR range direction. Then, the marine product data and HYCOM data are fully co-located to obtain a dataset. Two-thirds of the dataset is used to train a deep learning-based range-direction current velocity inversion model, and the remainder is used for validation.

[0015] A further improvement of the present invention is that the inversion model is a neural network model, represented as a multilayer perceptron; the inversion model of SAR range velocity based on deep learning is constructed according to the following steps;

[0016] (1) Calculate the DCA Doppler shift, Stokes shift, and wind speed component in the SAR range direction, as well as the radar incident angle, for each piece of marine product data;

[0017] (2) Decompose the flow velocity in the HYCOM data to the corresponding SAR range direction;

[0018] (3) Co-locate marine product data with the HYCOM dataset to obtain the training dataset for the model;

[0019] (4) Use two-thirds of the data in the dataset for model training and one-third for model validation;

[0020] (5) The inversion model is used to predict the SAR range velocity of a single SAR image.

[0021] A further improvement of the present invention is that, in the process of verifying the accuracy of the constructed inversion model, the input of the inversion model includes the radar readjustment Doppler frequency f. DR Predicting Doppler frequency shift f DP Wind speed U r Stokes Drift S r And the radar incident angle θ, the flow velocity V in HYCOM data r This serves as the output of the inversion model.

[0022] The beneficial effects of the technical solution provided by this invention are:

[0023] This invention discloses a method for inverting SAR range-direction current velocity from SAR marine product data. The method obtains radar incident angle information from OCN data; explores the relationship between DCA Doppler shift, Stokes drift, wind speed, and current velocity in the SAR range direction using OCN and HYCOM data; and constructs a range-direction current velocity inversion algorithm based on the explored relationship between current velocity and DCA Doppler shift, Stokes drift, and wind speed, and verifies its accuracy. Attached Figure Description

[0024] To more clearly illustrate the development technical solution of this invention, the accompanying drawings required in the algorithm development description will be briefly introduced below.

[0025] Figure 1 This is a flowchart of a method for retrieving SAR range current velocity from SAR marine product data according to the present invention.

[0026] Figure 2 The geographical locations of S-1OCN data from 2020 to 2022 are shown in Figure (a) for the western Pacific Ocean and Figure (b) for the western Atlantic Ocean.

[0027] Figure 3 The following are ocean surface current field diagrams for HYCOM. Figure (a) shows the location in the western Pacific Ocean, and Figure (b) shows the location in the western Atlantic Ocean.

[0028] Figure 4 For DCA f DC The relationship between the wind speed U and the upper ocean variables observed by radar is shown in Figure (a). r Figure (b) shows the HYCOM flow velocity V. r Figure (c) shows the Stokes drift S derived from the surge spectrum in the OCN products. r .

[0029] Figure 5 This is the flowchart of the MLP algorithm.

[0030] Figure 6 This refers to the performance during the training process.

[0031] Figure 7 The velocity is the inverse velocity derived from the distance upwards.

[0032] Figure 8 This is a comparison between the distance-inverted velocity and the distance-inverted component of the HYCOM velocity. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0034] Figure 1 This is a flowchart of a method for retrieving SAR range current velocity from SAR marine product data according to the present invention. The method includes:

[0035] 101: Obtain raw Sentinel-1 marine product data (OCN data) and HYCOM data, including sea surface flow field data.

[0036] Specifically, S-1OCN data (ocean product data) stored as a secondary product between 2020 and 2022 was collected from the western boundary current region of the ocean, acquired using the IW (Interferometric Wide Swath) model. Specifically, the OCN product provides the radar echo frequency shift at the beam center (called the measured Doppler shift), the predicted Doppler shift related to satellite velocity and Earth rotation, and wind speed and wave parameters retrieved from SAR. The geographical location of the OCN data is as follows: Figure 2 As shown, Figure (a) is in the western Pacific Ocean and Figure (b) is in the western Atlantic Ocean. Furthermore, sea surface flow fields simulated using the open data source HYCOM were directly employed, with a spatial resolution of ~0.08° and an interval of 3 hours. Figure 3 The HYCOM flow field map showing the location of the OCN data is displayed. It is necessary to calculate that the time difference and distance difference between the HYCOM flow field and the OCN product are within 1 hour and 2 kilometers, respectively. Two-thirds of the dataset will be used to develop a distance-based velocity inversion algorithm, and the remainder will be used for validation.

[0037] 102: Exploring the relationship between DCA Doppler frequency shift and Stokes drift, wind speed, and current velocity in the SAR range direction using OCN and HYCOM data.

[0038] Specifically, all acquired OCN and HYCOM data are used to calculate the corresponding DCA Doppler shift (SAR measured Doppler shift - predicted Doppler shift), Stokes drift, wind speed, and current velocity based on the variables in the OCN and HYCOM data. By fully co-locating the OCN and HYCOM data, nearly 3 million statistical samples are obtained, which is sufficient to develop a practical algorithm for SAR range-to-current velocity inversion based on deep learning. The specific parameter calculations are as follows:

[0039] Regarding single-antenna SAR, DCAf in the range direction DC Radar-remodulated Doppler frequency f DR And predicting Doppler frequency shift f DP The difference between them is represented as:

[0040] f DC =f DR -f DP +Δ (1)

[0041] In the formula, Δ is the change in azimuth angle of the radar backscattered signal represented by the normalized radar cross section (NRCS), caused by the change in the azimuth angle. DR and f DP The estimation error. In the OCN data, f DR and f DP These values ​​are provided after azimuth offset correction, therefore, Δ is assumed to be zero.

[0042] According to wave theory, Stokes drift S r Defined as the average Lagrangian velocity of water particles along the direction of wave propagation, it plays an important role in ocean circulation, and its calculation formula is as follows:

[0043]

[0044] In the formula, H represents the velocity of surface Stokes drift. s It is the significant wave height (SWH), T m It is the Mean Wave Period (MWP). This is the direction of wave propagation. The above parameters can be calculated from the surge spectrum in the OCN product. Although the amplitude of surface Stokes drift is relatively small, it has an impact on DCA. Based on the principle of geometric projection, the velocity vector of surface Stokes drift is converted into the component of the radar observation direction using formula (3):

[0045] x r =ucosγ+vsinγ (3)

[0046] In the formula, x r γ represents the ranging velocity; u and v represent the meridional and zonal components of either wind or ocean current; γ is the angle from the geographic location, rotating counterclockwise eastward to the radar line of sight.

[0047] Figure 4 DCA f was displayed DC The relationship between the upper ocean variables in the direction of radar observation, i.e., the OCN wind speed U in Figure (a). r Figure (b) shows the current speed V of HYCOM. r Figure (c) shows the Stokes drift S derived from the surge spectrum in the OCN product. r The black line represents linear regression. DCA f can be clearly observed. DC With Stokes Drift S r They have a linear relationship.

[0048] 103: Based on the research results, construct a deep learning-based OCN product inversion model for velocity in the distance direction.

[0049] This invention employs a neural network model, represented as a Multi-Layer Perceptron (MLP), which is an adaptive nonlinear system. The basic principles of MLP are as follows: Figure 5As shown, this is a three-layer model; in practical applications, deeper neural network models with more layers can be used. Follow these steps to construct a deep learning-based SAR range-direction current velocity inversion model;

[0050] (1) Calculate the DCA Doppler shift, Stokes drift, and wind speed components in the SAR range direction, as well as the radar incident angle, for each data point in the obtained OCN data (marine product data);

[0051] (2) Decompose the flow velocity in the HYCOM data to the corresponding SAR range direction;

[0052] (3) Co-localize the OCN dataset and the HYCOM dataset to obtain the training dataset for the model;

[0053] (4) Use two-thirds of the data in the dataset for model training and one-third for model validation;

[0054] (5) Finally, the trained model is used for prediction of a single image;

[0055] 104: Verify the accuracy of the inversion algorithm.

[0056] Specifically, in this invention, the input includes the radar readjustment Doppler frequency f. DR Predicting Doppler frequency shift f DP OCN wind speed U r Stokes Drift r And radar incident angle θ, HYCOM ocean current velocity V r This is the output. The performance of two-thirds of the training process on the dataset (approximately 2.7 million samples) is as follows: Figure 6 As shown, this indicates that the model gradually converges as the number of training iterations increases. The remaining portion of the dataset (300,000 samples) was used to validate the MLP model. As an example, Figure 7 A diagram showing the inversion of velocity along the distance is presented. Statistical analysis is as follows: Figure 8 As shown, the result is 0.19ms. -1 The RMSE velocity has a correlation coefficient (r) of 0.61. Overall, MLP is a deep learning-based algorithm that considers Stokes drift caused by swells and is suitable for extracting distance-oriented flow velocities from S-1OCN products.

[0057] In summary, quantitative estimation of ocean currents remains a challenge for SAR inversion, and some experiments have been conducted on simple phenomena under ideal conditions. A major problem in range-direction current velocity inversion is eliminating the influence of wind-generated short-period waves on the DCA (Distribution-Directional Aspect). In particular, the contribution of swell-induced Stokes drift to the DCA warrants further investigation. In our work, a large number of S-1OCN products from 2020 to 2022 were combined with HYCOM current field data. Using a dataset of approximately 3 million matched data points, the dependence of upper ocean dynamics on the DCA of range-direction currents, namely wind, ocean currents, and Stokes drift, was investigated. A range-direction current velocity inversion algorithm based on a deep learning method was then proposed. The conclusions are as follows.

[0058] (1) Apart from sea surface wind and ocean current, the Stokes drift caused by swells is linearly related to DCA.

[0059] (2) When the DCA-based flow velocity inversion algorithm is trained on the dataset to two-thirds of its training using a deep learning method (called MLP), the RMSE of distance to flow velocity converges to 0.15ms. -1 As the number of iterations approaches 60.

[0060] (3) One-third of the dataset was used for MLP validation, and the distance-to-flow velocity generated 0.19ms. -1 RMSE, with 0.61r and 0.89 bias.

[0061] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for retrieving SAR range current velocity from SAR marine product data, characterized in that, include: Acquire marine product data and HYCOM data from synthetic aperture radar; Using marine product data and HYCOM data, the mapping relationship between DCA Doppler frequency shift and Stokes drift, wind speed and current velocity in the SAR range direction was determined; Based on the mapping relationship, a deep learning-based inversion model for SAR range velocity inversion from marine product data is constructed. Verify the accuracy of the inversion model; The model is used to predict the SAR range current velocity based on marine product data from synthetic aperture radar. Using marine product data and HYCOM data, the relationship between DCA Doppler frequency shift and Stokes drift velocity, wind speed, and current velocity in the SAR range direction was determined, specifically including: The acquired marine product data and HYCOM data are used to calculate the corresponding DCA Doppler frequency shift and Stokes drift velocity, wind speed, and current velocity based on the variables in the marine product data and HYCOM data. These are all decomposed into SAR range direction. Then, the marine product data and HYCOM data are fully co-located to obtain a dataset. Two-thirds of the dataset is used to train a deep learning-based range-direction current velocity inversion model, and the remainder is used for validation. The inversion model is a neural network model, represented as a multilayer perceptron; the following steps are followed to construct a deep learning-based SAR range velocity inversion model; (1) Calculate the DCA Doppler shift, Stokes shift, and wind speed components in the SAR range direction, as well as the radar incident angle, for each data point of the obtained marine product data; (2) Decompose the flow velocity in the HYCOM data into the corresponding SAR range direction; (3) Co-locate marine product data with the HYCOM dataset to obtain the training dataset for the model; (4) Use two-thirds of the data in the dataset for model training and one-third for model validation; (5) The inversion model is used to predict the SAR range velocity of a single SAR image.

2. The method for retrieving SAR range current velocity from SAR marine product data according to claim 1, characterized in that, Acquire marine product data and HYCOM data from synthetic aperture radar, specifically including: Sentinel-1 OCN data was collected as marine product data; the OCN data included radar echo frequency shift at the beam center, predicted Doppler frequency shift related to satellite velocity and Earth rotation, wind speed and wave parameters retrieved by SAR; the HYCOM data was the sea surface flow field simulated using the open data source HYCOM, with a spatial resolution of ~0.08° and an interval of 3 hours.

3. The method for retrieving SAR range current velocity from SAR marine product data according to claim 1, characterized in that, In verifying the accuracy of the constructed inversion model, the inputs to the inversion model include the radar readjustment Doppler frequency. f DR Predicting Doppler frequency shift f DP Wind speed U r Stokes Drift S r and radar incident angle Flow rate in HYCOM data V r This serves as the output of the inversion model.

Images