A method and system for sea surface wind speed prediction
By combining data fusion and fitting models from GNSS-R and spaceborne microwave scatterometers, the weather dependence and observation range limitations of sea surface wind speed detection have been resolved, enabling high-precision wind speed prediction that is available 24/7, in all weather conditions, and at low cost.
Patent Information
- Application Number
- CN202310405985.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-04-17
- Publication Date
- 2026-02-10
- Estimated Expiration
- 2043-04-17
AI Technical Summary
Existing technologies for sea surface wind speed detection suffer from problems such as significant influence from weather conditions, limited observation range, and high cost, making it difficult to achieve high-precision wind speed prediction.
A method combining GNSS-R and spaceborne microwave scatterometers is adopted, and machine learning regression models are used for data fusion and fitting. Low wind speeds are retrieved by GNSS-R and high wind speeds are retrieved by microwave scatterometers. The results are then combined with the fusion and fitting model to improve prediction accuracy.
It achieves all-day, all-weather, and low-cost sea surface wind speed prediction, with significantly improved accuracy, especially in the medium-to-high wind speed range, and overall prediction accuracy is better than that of a single model.
Smart Images
Figure CN116430480B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of marine data measurement technology, and specifically relates to a method and system for predicting sea surface wind speed. Background Technology
[0002] Sea surface winds are the primary source of power for upper ocean motion, directly related to almost all seawater movement and serving as a crucial parameter in the interaction between the atmosphere and the ocean. Furthermore, near-surface winds are major contributors to and indicators of momentum and energy at the ocean-atmosphere interface. Accurate detection of sea winds not only helps predict trends in atmospheric circulation over the sea surface but also provides feasibility forecasts for human maritime operations, ensuring their smooth and safe operation. Summary of the Invention
[0003] One embodiment of this application provides a method for predicting sea surface wind speed, comprising the following steps:
[0004] Receive GNSS-R data of the ocean surface, and designate it as the first data;
[0005] Receive measurement data of the ocean surface from the spaceborne microwave scatterometer, and designate it as the second data;
[0006] The first and second data are input into the sea surface wind speed prediction model to obtain the predicted sea surface wind speed value.
[0007] The beneficial effect of this embodiment is that by fusing and inverting the observation data of GNSS-R and scatterometers, the wind speed can be accurately predicted. Attached Figure Description
[0008] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of the invention are illustrated in the drawings by way of example and not limitation, wherein:
[0009] Figure 1 A schematic diagram of the construction process of a sea surface wind speed prediction model according to one embodiment of the present invention.
[0010] Figure 2 A schematic diagram of the CNN regression model structure according to one embodiment of the present invention.
[0011] Figure 3 A schematic diagram of the BP regression model structure according to one embodiment of the present invention.
[0012] Figure 4 A schematic diagram of a fusion model structure according to one embodiment of the present invention.
[0013] Figure 5A model fitting diagram of the prediction model results according to one embodiment of the present invention.
[0014] Figure 6 A training structure diagram of a prediction model according to one embodiment of the present invention.
[0015] Figure 7 A schematic diagram of the composition of a sea surface wind speed prediction system according to one embodiment of the present invention. Detailed Implementation
[0016] To address the challenge of sea surface wind speed detection, this disclosure utilizes Global Navigation Satellite System Reflectometry (GNSS-R) technology to invert and calculate sea surface wind speed. This method effectively leverages the advantages of GNSS-R, including:
[0017] (1) GNSS-R is not affected by weather (clouds, fog, etc.);
[0018] (2) It can receive signals from multiple satellites simultaneously, providing a wide observation range;
[0019] (3) It can achieve low-cost loads.
[0020] When retrieving sea surface wind speed using GNSS-R, the normalized bistatic radar cross section (NBRCS) is used as the main characteristic parameter. The NBRCS is primarily related to sea surface roughness. At low wind speeds When the sea surface roughness is low, a strong sea surface reflection signal can be obtained, and the sea surface wind speed can be retrieved with good accuracy using the normalized bistatic radar cross section. However, at medium to high wind speeds, the sea surface roughness increases, resulting in poorer sea surface wind speed retrieval accuracy.
[0021] As an important active microwave remote sensor, the spaceborne microwave scatterometer can rapidly acquire global sea surface wind field information around the clock and in all weather conditions by observing the backscattering coefficient of the sea surface from multiple azimuth angles. It is also a primary means of satellite remote sensing of sea surface wind fields. The backscattering cross section is affected by Bragg scattering and specular reflection. Bragg scattering is mainly influenced by gravitational capillary waves on the sea surface, and the generation of gravitational capillary waves is caused by changes in the wind field on the sea surface. Some of the wind energy is absorbed by the ocean and transferred from gravitational capillary waves to gravitational waves and long waves in time and space. With increased wind speed, even longer wavelength waves may be formed. This energy absorbed by the ocean exists in various forms of long and short waves with almost no energy loss. Therefore, the microwave scatterometer is less affected by sea surface roughness. The wind speed is relatively high.At wind speeds above 10 m / s, GNSS-R has better inversion results than GNSS-R; however, at lower wind speeds (below 10 m / s), GNSS-R is less affected by sea surface roughness and has better accuracy in sea surface wind speed inversion.
[0022] Therefore, based on GNSS-R forward scattering and microwave scattering backscattering, this disclosure proposes a new method for fusing GNSS-R and scattering to retrieve sea surface wind speed.
[0023] According to one or more embodiments, a method for predicting sea surface wind speed includes the following steps:
[0024] First, sea surface wind speed was retrieved using GNSS-R and microwave scatterometer respectively, employing machine learning (ML) regression models.
[0025] Then, taking full account of the advantages of both, the wind speed inversion results of the regression model are fused at the result level.
[0026] Finally, the wind speed inversion results are fitted with the actual wind speed to achieve the optimal accuracy result. Figure 1 This diagram illustrates the process of building a sea surface wind speed prediction model. The models in steps 1, 2, and 3 are trained using three different batches of data. In step 1, training data 1 is input into the ML regression model to obtain output result 1; in step 2, training data 2 is input into the ML regression model to obtain output result 2; and in step 3, training data 3 is input into the ML regression model to obtain output result 3. Output results 1, 2, and 3 are then sequentially input into a result fusion model to obtain wind speed inversion result 1. Further, real wind speed test data is input into the ML regression model to obtain output result 4. This output result 4, along with wind speed inversion result 1, is then input again into the result fusion model and / or result fitting model to obtain the final inverted wind speed result.
[0027] Furthermore, the establishment of ML regression models includes:
[0028] Sea surface wind speed inversion was performed using CNN regression and BP regression models, respectively, with GNSS-R and microwave scatterometer. Specific model results are shown below. Figure 2 , Figure 3 As shown. Figure 2 It includes multiple convolutional layers, as well as unfolded layers and fully connected layers.
[0029] After extensive data validation, the final determination for the CNN regression model was as follows: 3 convolutional layers, no pooling layers, a dropout rate of 0.3, 32 kernels in each convolutional layer, a batch size of 1000 parameters, and 2000 training epochs. The BP regression model had 2 hidden layers with 20 nodes in each, and tanh was chosen as the activation function.
[0030] To further utilize the BP regression model results from the microwave scatterometer to improve the accuracy of GNSS-R global sea surface wind speed retrieval, a result fusion model integrating the regression model results from both is proposed. This model primarily concatenates the BP regression model results from the microwave scatterometer with the CNN regression model results from GNSS-R, while adding a flag to distinguish the two data sources. This data, along with the ground truth wind speed data, serves as the input to the result fusion model.
[0031] Figure 4 The fusion model structure used in this embodiment is as follows. After verification with a large amount of data, the final determination is that the number of CNN convolutional layers is 4, no pooling layers are set, the dropout rate of hidden layer nodes is 0.3, the number of convolutional kernels is 3, the number of parameters in a single training session (batch-size) is 1000, and the number of training epochs is 1000.
[0032] After the ML regression model and the result fusion model, some discrete points may still exist that deviate from the center. To correct these potential discrete points, after completing the regression model and result fusion model inversion module, the final results of the two models are further fitted with the actual wind speed value to achieve the purpose of correcting the discrete points. The result fitting model diagram is shown below. Figure 5 As shown in the figure, the fusion results output by the result fusion model, the machine learning regression results output by the ML regression model, and / or the true wind speed can be input into the result fitting model to obtain the final wind speed prediction result.
[0033] Based on the distribution of the training data and the performance of different fitting models after training, the final function form used for fitting is determined as follows:
[0034] U ECMWF =AU Regression +BU Fusion +C (1)
[0035] In equation (1), A, B, and C are the fitting coefficients to be determined, and U ECMWF U represents the actual wind speed at sea surface. Regression For the wind speed inversion results of the ML regression model, U FusionThis is the wind speed inversion result from the fusion model. After using the fitted model, the discrete points in the inversion result will be corrected as much as possible to improve the final inversion result.
[0036] According to one or more embodiments, a method for retrieving sea surface wind speed based on GNSS-R and scatterometer fusion includes the following steps:
[0037] S101, wind speed inversion is performed using CNN model and BP model respectively for GNSS-R data and microwave scatterometer data;
[0038] S102, the inversion results of the CNN model and the inversion results of the BP model are put into the result fusion model. The flag bit of the GNSS-R inverted wind speed data is set to "1" and the flag bit of the microwave scatterometer inverted wind speed data is set to "0". The two data are merged together and the fusion model is trained.
[0039] S103, fit the wind speed results of the regression model and the wind speed results of the result fusion model with the actual wind speed, and further correct the abnormal wind speed results after fusion;
[0040] S104 analyzes the wind speed results from the regression model, the fusion model, and the fitted correction model to verify the feasibility of using the fusion model and the fitted model to retrieve sea surface wind speed. The specific process and test set results flow are as follows... Figure 6 As shown.
[0041] According to one or more embodiments, a sea surface wind speed prediction system includes a GNSS-R low-Earth orbit satellite, a scatterometer satellite, a ground control station, and a ground server. The GNSS-R low-Earth orbit satellite transmits GNSS-R data through a GNSS-R payload and an air-to-ground communication link. The scatterometer satellite transmits scatterometer data through a scatterometer payload and an air-to-ground communication link. The ground control station receives the GNSS-R data and the scatterometer data, transmits them to the server, and after processing by the aforementioned regression model, fusion model, and fitting model, obtains the sea surface wind speed prediction result data.
[0042] The beneficial effects of this disclosure include:
[0043] (1) After the result fusion model, the accuracy of sea surface wind speed above 10m / s using GNSS-R will be improved, which is more conducive to wind speed inversion in the medium and high wind speed range.
[0044] (2) The overall fusion model inversion results are further improved on the basis of the single ML model, which will make the sea surface wind speed inversion model more accurate;
[0045] (3) The inversion accuracy of the result fitting model is the best at all wind speeds, and the overall inversion accuracy is significantly improved compared with the result fusion model.
[0046] If implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0047] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A method for predicting sea surface wind speed, characterized in that, Including the following steps: Receive GNSS-R data of the ocean surface, and designate it as the first data; Receive measurement data of the ocean surface from the spaceborne microwave scatterometer, and designate it as the second data; Input the first data and the second data into the sea surface wind speed prediction model to obtain the predicted sea surface wind speed value; The sea surface wind speed prediction model includes a machine learning (ML) regression model. The first data and the second data are inverted and fused through the ML regression model to obtain the predicted sea surface wind speed value. The ML regression model includes a CNN regression model and a BP regression model. The first data is input into the CNN regression model, and the second data is input into the BP regression model. The sea surface wind speed prediction model includes a fusion model, which fuses the outputs of the CNN regression model and the BP regression model, and further fits and corrects the predicted sea surface wind speed value output by the machine learning ML regression model with the real sea surface wind speed data, which is also used as the input of the fusion model, to obtain the corrected sea surface wind speed prediction result.
2. The sea surface wind speed prediction method according to claim 1, characterized in that, The marine meteorological wind speed data In equation (1), A, B, and C are the fitting coefficients to be determined. This represents the actual wind speed at sea. The wind speed inversion results are from the ML regression model. The wind speed inversion result is given by the fusion model.
3. A sea surface wind speed prediction system, characterized in that, The system includes GNSS-R low-Earth orbit satellites, scatterometer satellites, ground tracking and control stations, and ground servers. GNSS-R low-Earth orbit satellites transmit GNSS-R data via GNSS-R payloads and air-to-ground communication links, while scatterometer satellites transmit scatterometer data via scatterometer payloads and air-to-ground communication links. Ground control stations receive the GNSS-R data and scatterometer data and transmit them to ground servers. The ground server includes a memory; and A processor coupled to the memory, the processor being configured to execute instructions stored in the memory to implement the method as described in any one of claims 1 to 2.
4. The sea surface wind speed prediction system according to claim 3, characterized in that, GNSS-R data includes normalized bistatic radar cross section parameters obtained through bistatic radar, and scatterometer data includes sea surface backscattering coefficients obtained through spaceborne microwave scatterometers.
5. A storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 2.