A Method and System for Correcting Sea Surface Wind Speed ​​Deviation Based on Statistical Learning and Physical Constraints

Abstract

This application proposes a method and system for correcting sea surface wind speed deviation based on statistical learning and physical constraints, belonging to the field of satellite ocean remote sensing and meteorological data reanalysis. This application addresses the technical challenges of existing multi-source satellite wind speed data, which suffer from statistical instability, physical inconsistencies, and the inability to generate accurate and reliable correction results across the entire wind speed range (especially under high wind speed conditions) due to inherent sensor system errors and sparse samples in high-wind-speed intervals. Specifically, it acquires multi-source satellite wind speed data and high-precision reference wind field data for quality control and spatiotemporal matching. An initial deviation correction lookup table is constructed using a dual-reference partitioning strategy based on satellite-observed wind speed and reference wind field wind speed. A hierarchical Bayesian contraction estimator is used to statistically stabilize the lookup table. An adaptive LOESS smoothing algorithm is employed to improve the physical continuity of the wind speed-deviation curve, and an exponential decay function based on turbulent energy spectrum theory is introduced.

Description

Technical Field

[0001] This application relates to the field of satellite ocean remote sensing and meteorological data reanalysis, specifically proposing a multi-source satellite sea surface wind speed deviation correction method and system based on statistical learning and physical constraints. Background Technology

[0002] Sea surface wind speed is a key geophysical parameter for studying air-sea interaction, ocean dynamics, and climate change. Spaceborne microwave scatterometers and radiometers, capable of global sea surface wind speed observations, have become the primary means of acquiring this parameter. To obtain more complete spatiotemporal coverage, fusing multi-source satellite wind field data from different platforms and sensors to generate consistent data products is currently the mainstream direction for ocean remote sensing data applications. However, due to inherent differences in detection principles, calibration methods, and inversion algorithms among different satellite sensors such as scatterometers and radiometers, significant systematic biases exist between the wind speed data retrieved from them. Sensor-related systematic errors are directly transmitted to the fused product, reducing its accuracy and consistency and introducing uncertainty into data use.

[0003] Currently, bias correction for multi-source satellite wind speed data typically relies on traditional post-processing techniques such as simple linear regression, binning, or direct comparison with reference wind field data. These existing methods are applicable when wind speeds are moderate and the observation sample is sufficient, but they exhibit significant limitations in extreme wind speed ranges, especially under sparse high-wind-speed conditions such as typhoons and strong cyclones. These limitations are mainly reflected in the following aspects: First, the sparse sample size in high-wind-speed ranges leads to highly unstable statistical estimations, resulting in violent oscillations in the wind speed bias curve and producing non-physical correction results. Second, traditional binning methods cannot effectively handle the uneven observation density across different wind speed ranges, leading to poor smoothness and insufficient physical continuity of the bias curve. Finally, because the reference wind field data itself may systematically underestimate or overestimate under extreme conditions, simple bias adjustments can easily lead to overcorrection, i.e., introducing new errors to match the biased reference wind field, causing the corrected wind speed to lose its physical accuracy under strong wind conditions. Therefore, the existing multi-source satellite wind speed data suffers from statistical instability and poor physical consistency due to inherent sensor system errors and sparse samples in high-wind-speed ranges. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a sea surface wind speed deviation correction method and system based on statistical learning and physical constraints, in order to solve the technical problems of statistical instability, physical unreasonableness and inability to generate accurate and reliable correction results across the entire wind speed range (especially under high wind speed conditions) caused by the inherent systematic errors of sensors and the sparse sample size in the high wind speed range of existing multi-source satellite wind speed data.

[0005] To achieve the above objectives, the sea surface wind speed deviation correction method based on statistical learning and physical constraints acquires multi-source satellite wind speed data and high-precision reference wind field data for quality control and spatiotemporal matching. An initial deviation correction lookup table is constructed using a dual-reference partitioning strategy based on satellite-observed wind speed and reference wind field wind speed. A hierarchical Bayesian contraction estimator is used to statistically stabilize the lookup table. An adaptive LOESS smoothing algorithm is employed to improve the physical continuity of the wind speed-deviation curve, and an exponential decay function based on turbulent energy spectrum theory is introduced to constrain the correction amplitude in high-wind-speed regions while preventing over-correction due to systemic deviations in the reference wind field.

[0006] The sea surface wind speed deviation correction method includes the following steps:

[0007] Step S101: Data preparation and preprocessing;

[0008] Acquire multi-source satellite wind speed data and corresponding spatiotemporal reference wind field data, and match and bin them according to sensor type and wind speed value; perform a strict quality control process on all acquired data, including but not limited to removing data points marked as invalid or of poor quality, removing outliers that are obviously outside the reasonable physical range, and filtering out observations near land to avoid land pollution effects; for spatiotemporal matching, find the closest reference field data in spatiotemporal space for each satellite observation grid point; classify and store all matched data according to satellite sensor ID, in preparation for subsequently building a high-precision bias correction lookup table independently for each sensor;

[0009] Step S102: Construction of deviation correction lookup table and system stabilization;

[0010] A dual-partitioning method is adopted, based on a dual-reference wind speed partitioning strategy of satellite and reference wind field. The average deviation between satellite wind speed and reference wind field wind speed in each wind speed box is calculated, and an initial deviation correction lookup table is constructed. For each wind speed box, a hierarchical Bayesian estimator is used to perform weighted fusion of the local average deviation of the wind speed box and the global average prior deviation of the same sensor to obtain a statistically stable deviation estimate. The matching dataset generated by sensor classification in step S101 above is subjected to box-based statistical analysis to divide the wind speed range into multiple continuous wind speed boxes.

[0011] Step S103: Physical smoothing and constraints of the lookup table;

[0012] An adaptive local weighted scatter plot smoothing process is applied to the stabilized deviation correction lookup table, with the smoothing bandwidth dynamically adjusted according to the wind speed range. A larger bandwidth is used in the high wind speed range where the sample is sparse to ensure curve smoothness, while a smaller bandwidth is used in the low wind speed range where the sample is abundant to preserve details, thereby suppressing high-frequency fluctuations in the wind speed deviation relationship curve and enhancing its physical continuity. An exponential function-based decay constraint is applied to the correction amount in the high wind speed range to prevent overcorrection. Thus, the stabilized deviation correction lookup table is physically optimized.

[0013] Step S104: Satellite wind speed data correction;

[0014] For the satellite wind speed data to be corrected, the final deviation correction lookup table after processing is queried according to its sensor type and wind speed value to obtain the corresponding deviation correction amount, and the original wind speed is corrected and output according to the preset correction rules.

[0015] The preset correction rule is that when the wind speed is lower than or equal to the wind speed threshold, the deviation correction amount is applied in full; when the wind speed is higher than the wind speed threshold, the deviation correction amount needs to be multiplied by a decay coefficient that decreases exponentially with the increase of wind speed.

[0016] The wind speed threshold and the empirical attenuation rate parameter are set based on the turbulent energy spectrum scale law to optimize the correction behavior under high wind speeds in accordance with the laws of fluid dynamics.

[0017] In step S101, the multi-source satellite data refers to wind speed data acquired by satellite remote sensing, which is satellite scatterometer data, satellite radiometer data, or satellite altimeter data.

[0018] In step S101, the reference wind field data is model data, reanalysis data, multi-source fusion data, or measured data, and it is within the same spatiotemporal range as satellite observations.

[0019] Step S102 employs a dual-reference wind speed zoning strategy; using the initial satellite wind speed respectively... and reference wind field wind speed As a reference benchmark, a separate binning statistical analysis is performed to calculate two preliminary local average deviation sequences. These two deviation sequences are then merged by averaging or weighted averaging to form the final deviation correction lookup table.

[0020] In step S102, a hierarchical Bayesian contraction estimator is used to perform the following statistical stabilization process; for the nth anemometer, its statistically stable deviation estimate is... The calculation is as follows:

[0021] (1);

[0022] (2);

[0023] (3);

[0024] in, Indicates the first The effective sample size in each anemometer box; It is the local average deviation between the wind speeds of different satellites and the reference wind field within the container; This represents the global average deviation for the same sensor ID, which is used as the prior mean in the hierarchical Bayesian shrinkage estimator. It is the total sample size of all anemometers; parameters This represents the prior weights, which are interpreted as pseudo-sample sizes and used to quantify the global average bias. With local average The relative confidence level.

[0025] The statistically stable deviation estimate is calculated by multiplying the local average deviation by the effective sample size of the anemometer as a first weight, multiplying the global average prior deviation by a preset prior pseudo-sample size as a second weight, and dividing the sum of the two weighted results by the sum of the effective sample size and the prior pseudo-sample size; wherein, the prior pseudo-sample size is determined based on the total sample size of the sensor data and is set as the square root of the total sample size of the sensor data.

[0026] To implement the aforementioned sea surface wind speed deviation correction method based on statistical learning and physical constraints, this application also proposes the following sea surface wind speed deviation correction system based on statistical learning and physical constraints: including,

[0027] 1) Data preparation module;

[0028] This module is used to acquire multi-source satellite wind speed data and corresponding reference wind field data, and to perform quality control, spatiotemporal matching, and classification of the acquired data according to sensor type. It is also responsible for performing box-based statistics according to preset wind speed intervals, calculating the local average deviation between the satellite wind speed and the reference wind field wind speed in each wind speed box, and the effective sample size, so as to provide structured basic statistical data for subsequent processing.

[0029] 2) Lookup table construction module;

[0030] This module is used to construct an initial bias correction lookup table using a dual-reference wind speed partitioning strategy. The initial satellite wind speed and the reference wind field wind speed are used as independent references for binning statistics, and the two results are fused by weighted averaging to reduce selection bias. The module also uses a hierarchical Bayesian contraction estimator to perform statistical stabilization processing on the initial lookup table by weighted fusion of the local average bias of each wind speed bin with the global average prior bias of the sensor.

[0031] 3) Physical constraint module;

[0032] This module is used to perform adaptive local weighted scatter plot smoothing on the stabilized deviation correction lookup table. Its smoothing bandwidth is dynamically adjusted with the sample density in the wind speed range to filter out high-frequency noise while maintaining the continuity of the curve. The module also introduces an exponential decay function based on the turbulent energy spectrum scale law to impose physical constraints on the correction amount in the high wind speed range to prevent overcorrection caused by the deviation of the reference wind field system.

[0033] 4) Calibration application module;

[0034] This module is used to correct and output satellite wind speed data based on the final optimized deviation correction lookup table. It calls the corresponding lookup table according to the sensor type, queries each satellite wind speed observation value to obtain the corresponding deviation correction amount and attenuation coefficient, and calculates the final wind speed product with effective suppression of system deviation through the correction formula.

[0035] This application proposes a novel computer device comprising one or more processors and a storage device, wherein the storage device stores one or more programs, and when the one or more programs are executed by the one or more processors, the aforementioned sea surface wind speed deviation correction method based on statistical learning and physical constraints is implemented.

[0036] This application proposes a novel computer-readable storage medium storing one or more executable programs. When the programs are executed by the processor of a computer device, the aforementioned sea surface wind speed deviation correction method based on statistical learning and physical constraints is implemented.

[0037] In summary, the beneficial effects and advantages of this application compared with the prior art include:

[0038] 1. This application can effectively solve the limitations of the existing bias correction technology. It not only has strong statistical robustness to deal with the problem of sample sparsity, but also can integrate physical constraints such as fluid dynamics into the correction process. Thus, it can produce physically reasonable and statistically reliable correction results in the full wind speed range, especially under high wind speed conditions, providing a reliable data foundation for high-precision multi-source satellite wind field fusion, data assimilation and extreme marine meteorological research.

[0039] 2. This application can significantly improve statistical stability. By introducing a hierarchical Bayesian shrinkage estimator, local sample information and global prior information are optimally weighted and fused, thereby effectively overcoming the problem of drastic fluctuations in bias estimation caused by sparse samples in high wind speed ranges, and significantly improving the robustness and statistical stability of the correction algorithm.

[0040] 3. This application has effective guarantee performance of physical rationality. Through adaptive local weighted scatter plot smoothing, the physical continuity of the wind speed and deviation relationship curve is ensured. Furthermore, by combining the exponential decay constraint based on turbulence theory, the correction range under high wind speed is scientifically limited, effectively preventing physically unreasonable overcorrection behavior, and making the correction results more in line with the dynamic laws of the ocean-atmospheric boundary layer.

[0041] 4. This application has a high-precision correction effect across the entire wind speed range. It creatively integrates statistical learning technology with physical constraint mechanisms, systematically solving the problem of full-range deviation correction from conventional wind speed to extreme wind speed. The resulting wind speed product has higher accuracy, consistency and reliability. Attached Figure Description

[0042] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. Some specific embodiments of this application will be described in detail below with reference to the accompanying drawings in an exemplary and non-limiting manner. The same reference numerals in the drawings designate the same or similar parts or components. It should be understood by those skilled in the art that these drawings are not necessarily drawn to scale.

[0043] Figure 1 This is a flowchart of the sea surface wind speed deviation correction method based on statistical learning and physical constraints described in this application;

[0044] Figure 2 This is an architecture diagram of the sea surface wind speed deviation correction system based on statistical learning and physical constraints described in this application. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0046] The following embodiments of this application provide a multi-source satellite sea surface wind speed deviation correction method based on statistical learning and physical constraints. This method systematically solves the systematic bias of multi-source satellite wind speed data and the estimation instability caused by the sparse high wind speed samples by integrating hierarchical Bayesian statistical learning and physical constraint smoothing technology. Finally, it generates a high-precision correction product that is physically reasonable and statistically stable across the entire wind speed range.

[0047] like Figure 1As shown, the sea surface wind speed deviation correction method proposed in this application, based on statistical learning and physical constraints, performs quality control and spatiotemporal matching by acquiring multi-source satellite wind speed data and high-precision reference wind field data. It constructs an initial deviation correction lookup table using a dual-reference partitioning strategy based on satellite-observed wind speed and reference wind field wind speed. The lookup table is statistically stabilized using a hierarchical Bayesian contraction estimator. The physical continuity of the wind speed-deviation curve is improved by using an adaptive LOESS smoothing algorithm, and an exponential decay function based on turbulent energy spectrum theory is introduced to constrain the correction amplitude in high wind speed areas and prevent overcorrection caused by systematic deviations in the reference wind field.

[0048] The overall process of this method can be found in [reference needed]. Figure 1 As shown, it includes the following steps:

[0049] Step S101: Data preparation and preprocessing;

[0050] Acquire multi-source satellite wind speed data and corresponding spatiotemporal reference wind field data, and match and bin them according to sensor type and wind speed value;

[0051] A rigorous quality control process is implemented for all acquired data, including but not limited to removing data points marked as invalid or of poor quality, removing outliers that are clearly outside the reasonable physical range, and filtering out observations of nearby land to avoid land pollution effects;

[0052] For spatiotemporal matching, the goal is to find the closest reference field data in spatiotemporal space for each satellite observation grid point. In this embodiment, the set time matching window is 0.5 h before and after the satellite observation time point, and the spatial matching window is 25 km.

[0053] Finally, all the matched data are classified and stored according to satellite sensor ID (such as "ASCAT-A", "AMSR2", etc.) to prepare for the subsequent construction of a high-precision deviation correction lookup table for each sensor independently.

[0054] Specifically, the data bins are spaced less than 25 km apart and timed less than 0.5 h apart, aiming to provide a high-quality, spatiotemporally consistent input data source for subsequent bias correction;

[0055] The multi-source satellite data mentioned above are wind speed data acquired by satellite remote sensing, such as satellite scatterometer data, satellite radiometer data, or satellite altimeter data. Multi-source satellite sea surface wind speed data for the target area and specific time period can be obtained from relevant data archiving centers, such as the European Space Agency, the U.S. National Environmental Prediction Center, and the China National Satellite Ocean Application Center.

[0056] The multi-source satellite data includes, but is not limited to, scatterometer wind field data from HY-2B / C / D series satellites (HSCAT scatterometer), CFOSAT satellites (CSCAT scatterometer), and MetOp-A / B / C series satellites (ASCAT scatterometer); as well as radiometer wind speed data from DMSPF16 / 17 / 18 satellites (SSMIS radiometer), GCOM-W1 satellite (AMSR2 radiometer), GPM satellite (GMI radiometer), SMOS satellite (MIRAS radiometer), and SMAP satellite (L-band radiometer).

[0057] The reference wind field data can be model data, reanalysis data, multi-source fusion data or measured data, etc., within the same spatiotemporal range as satellite observations.

[0058] Step S102: Construction of deviation correction lookup table and system stabilization;

[0059] The dual partitioning method is adopted, based on the dual reference wind speed partitioning strategy of satellite and reference wind field, to calculate the average deviation between the satellite wind speed and the reference wind field wind speed in each wind speed box, and to construct an initial deviation correction lookup table.

[0060] For each wind speed box, a hierarchical Bayesian estimator is used to weight and fuse the local average deviation of the wind speed box with the global average prior deviation of the same sensor to obtain a statistically stable deviation estimate.

[0061] Perform binning statistical analysis on the matching dataset generated by sensor classification in step S101 above. For example, divide the wind speed range (e.g., 0 m / s to 70 m / s) into multiple consecutive wind speed bins at intervals of 0.5 m / s.

[0062] To mitigate the selectivity bias that may be introduced by using only a single data source such as satellite or reference wind field for binning, this application adopts a dual reference wind speed zoning strategy.

[0063] Specifically, based on the initial wind speed of the satellite and reference wind field wind speed As a reference benchmark, an independent binning statistical analysis was conducted to calculate two preliminary local average deviation sequences.

[0064] By averaging or weighted averaging, the two sets of deviation sequences are merged to form the final deviation correction lookup table.

[0065] For example, to address the problem of severe oscillations in the local mean bias estimate caused by sparse samples in high wind speed ranges (such as wind speeds greater than 15 m / s), this application introduces a hierarchical Bayesian contraction estimator for statistical stabilization.

[0066] For the nth anemometer, its statistically stable bias estimate is The calculation is as follows:

[0067] (1);

[0068] (2);

[0069] (3);

[0070] in, Indicates the first The effective sample size in each anemometer box; It is the local average deviation between the wind speeds of different satellites and the reference wind field within the container; This represents the global average deviation for the same sensor ID, which is used as the prior mean in the hierarchical Bayesian shrinkage estimator. It is the total sample size of all anemometers; parameters This represents the prior weights, which are interpreted as pseudo-sample sizes and used to quantify the global average bias. With local average The relative confidence level.

[0071] Using the estimator described above, the bias estimate of the wind speed box with a small sample size will "shrink" towards the global prior, thereby effectively suppressing small sample noise and obtaining a physically smoother and statistically more robust bias estimate sequence, which constitutes a stabilized bias correction lookup table.

[0072] The statistically stable deviation estimate is calculated by multiplying the local average deviation by the effective sample size of the anemometer as a first weight, multiplying the global average prior deviation by a preset prior pseudo-sample size as a second weight, and dividing the sum of the two weighted results by the sum of the effective sample size and the prior pseudo-sample size; wherein, the prior pseudo-sample size is determined based on the total sample size of the sensor data and is set as the square root of the total sample size of the sensor data.

[0073] Step S103: Physical smoothing and constraints of the lookup table;

[0074] The stabilized deviation correction lookup table is subjected to adaptive local weighted scatter plot smoothing, and the smoothing bandwidth is dynamically adjusted according to the wind speed range. A larger bandwidth is used in the high wind speed range where the sample is sparse to ensure curve smoothness, and a smaller bandwidth is used in the low wind speed range where the sample is abundant to retain details, so as to suppress high-frequency fluctuations of the wind speed deviation relationship curve and enhance its physical continuity.

[0075] Apply an exponential function-based attenuation constraint to the correction amount in the high wind speed range to prevent overcorrection;

[0076] Therefore, the deviation correction lookup table after stabilization is physically optimized to ensure that its output is physically continuous and reasonable.

[0077] Specifically, a variable bandwidth locally weighted scatter plot smoothing algorithm is applied to the smoothing of the scatter plot. The data sequence is smoothed, and the smoothing process can be represented as follows:

[0078] (4);

[0079] Among them, bandwidth The system dynamically adjusts with the wind speed interval to maintain the continuity of the wind speed gradient while filtering out high-frequency observation noise. The adjustment strategy is as follows: in the high wind speed range where the sample is sparse, a larger bandwidth is automatically used, such as covering a range of 5 m / s, to force the smoothness of the curve; in the low wind speed range where the sample is abundant, a smaller bandwidth is used, such as covering a range of 1.5 m / s, to retain more realistic details of local deviation changes. This can effectively suppress residual high-frequency fluctuations and form a physically continuous and smooth wind speed-deviation relationship curve.

[0080] To prevent overcorrection of satellite data under high wind speed conditions, and considering that the reference wind field itself may underestimate extreme wind speeds, an exponential decay function is introduced. Constraints are imposed on the correction amount; for example, setting the attenuation threshold to 20 m / s. When the wind speed is below 20 m / s, the full correction amount from the lookup table is applied; when the wind speed is above 20 m / s, the applied correction amount will decrease exponentially with increasing wind speed, ensuring that the correction behavior under extreme wind speeds conforms to the following turbulent energy spectrum scale law formula:

[0081] (5);

[0082] in, is the decay initiation threshold, which is set to 20 m / s in this embodiment based on the typical typhoon intensity wind speed and the reliability of the reference wind field; is the scale parameter controlling the decay rate, which is optimized in this embodiment based on the energy decay characteristics characterized by the turbulent energy spectrum scaling law; this function ensures that when the wind speed is lower than... At that time, the correction amount was applied in full. When the wind speed is higher than When the applied correction decreases exponentially with increasing wind speed, the physical rationality of the correction behavior is maintained under extreme weather conditions.

[0083] Step S104: Satellite wind speed data correction;

[0084] For the satellite wind speed data to be corrected, the final deviation correction lookup table after processing is queried according to its sensor type and wind speed value to obtain the corresponding deviation correction amount, and the original wind speed is corrected and output according to the preset correction rules.

[0085] The preset correction rule is that when the wind speed is lower than or equal to the wind speed threshold, the deviation correction amount is applied in full; when the wind speed is higher than the wind speed threshold, the deviation correction amount needs to be multiplied by a decay coefficient that decreases exponentially with the increase of wind speed.

[0086] The wind speed threshold and the empirical attenuation rate parameter are set based on the turbulent energy spectrum scale law to optimize the correction behavior under high wind speeds in accordance with the laws of fluid dynamics.

[0087] Specifically, for each satellite wind speed observation value Based on its value, find the record row in the lookup table that is closest to the midpoint of the wind speed range, and read the corresponding core deviation correction amount (denoted as ). and ) and attenuation coefficient (denoted as and The physically constrained bias correction is subtracted from the raw satellite wind speeds. This process is performed point-by-point on all satellite data to be corrected, ultimately outputting a high-quality multi-source satellite sea surface wind speed product with effectively suppressed systematic biases and high physical consistency across the entire wind speed range.

[0088] The final correction formula calculates the corrected wind speed value. for:

[0089] (6);

[0090] (7);

[0091] This yields the final deviation correction lookup table, which is a structured data file or database table. Its core content and format include the following fields:

[0092] Sensor ID (Sensor_ID), a string type, is used to uniquely identify the data source, such as "CSCAT";

[0093] Wind_Speed_Bin_Center, a floating-point number in m / s, represents the center value of a wind speed range, such as 10.5 m / s.

[0094] Core Bias Correction (Core_Bias_Correction), a floating-point number in m / s, is the value after LOESS smoothing. value;

[0095] Attenuation Factor, a floating-point number, dimensionless, which corresponds to the midpoint of the interval. value;

[0096] Valid Sample Count (Valid_Sample_Count), an integer, represents the number of samples in the anemometer. ;

[0097] Local Mean Bias (Local_Mean_Bias), a floating-point number in m / s, is the value before smoothing. .

[0098] To implement the aforementioned multi-source satellite sea surface wind speed deviation correction method based on statistical learning and physical constraints, this application also proposes the following multi-source satellite sea surface wind speed deviation correction system based on statistical learning and physical constraints:

[0099] 1) Data preparation module;

[0100] This module is used to acquire multi-source satellite wind speed data and corresponding reference wind field data, and to perform quality control, spatiotemporal matching, and classification of the acquired data according to sensor type. This module is also responsible for performing box-based statistics according to preset wind speed intervals, calculating the local average deviation between the satellite wind speed and the reference wind field wind speed in each wind speed box, and the effective sample size, so as to provide structured basic statistical data for subsequent processing.

[0101] 2) Lookup table construction module;

[0102] This module is used to construct an initial bias correction lookup table using a dual-reference wind speed partitioning strategy. It performs binning statistics using satellite initial wind speed and reference wind field wind speed as independent references, and uses a weighted average to fuse the two results to reduce selection bias. The module also uses a hierarchical Bayesian shrinkage estimator to perform statistical stabilization processing on the initial lookup table. By weighting and fusing the local average bias of each wind speed bin with the global average prior bias of the sensor, it effectively suppresses estimation noise caused by sample sparsity in high wind speed ranges.

[0103] 3) Physical constraint module;

[0104] This module is used to perform adaptive local weighted scatter plot smoothing on the stabilized deviation correction lookup table. Its smoothing bandwidth is dynamically adjusted with the sample density of the wind speed range to filter out high-frequency noise while maintaining the continuity of the curve. The module also introduces an exponential decay function based on the turbulent energy spectrum scale law to impose physical constraints on the correction amount in the high wind speed range, preventing overcorrection caused by the deviation of the reference wind field system.

[0105] 4) Calibration application module;

[0106] This module is used to correct and output satellite wind speed data based on the final optimized deviation correction lookup table. The module calls the corresponding lookup table according to the sensor type, queries the corresponding deviation correction amount and attenuation coefficient for each satellite wind speed observation value, and calculates the final wind speed product with effective suppression of system deviation through the correction formula.

[0107] In summary, the multi-source satellite sea surface wind speed deviation correction method based on statistical learning and physical constraints proposed in this application significantly improves statistical stability compared to traditional binning statistical methods and demonstrates marked improvement in physical rationality compared to single statistical learning methods. Although this embodiment describes the deviation correction of sea surface wind speed in detail, the core idea of ​​this application method—integrating statistical learning techniques with physical constraint mechanisms—is also applicable to the systematic deviation correction problems of other marine remote sensing parameters, such as significant wave height and sea surface temperature.

[0108] This application proposes a novel computer device comprising one or more processors and a storage device, wherein the storage device stores one or more programs, and when the one or more programs are executed by the one or more processors, the aforementioned multi-source satellite sea surface wind speed deviation correction method based on statistical learning and physical constraints is implemented.

[0109] This application proposes a novel computer-readable storage medium storing one or more executable programs. When the programs are executed by the processor of a computer device, the aforementioned multi-source satellite sea surface wind speed deviation correction method based on statistical learning and physical constraints can be implemented.

[0110] The above description of the embodiments is only for the purpose of helping to understand the method and core ideas of this application; at the same time, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. The content of this specification should not be construed as a limitation of this application.

Claims

1. A method for correcting sea surface wind speed deviation based on statistical learning and physical constraints, characterized in that: Quality control and spatiotemporal matching are performed by acquiring multi-source satellite wind speed data and high-precision reference wind field data. An initial deviation correction lookup table is constructed using a dual-reference partitioning strategy based on satellite-observed wind speed and reference wind field wind speed. The lookup table is statistically stabilized using a hierarchical Bayesian contraction estimator. The physical continuity of the wind speed-deviation curve is improved by using an adaptive LOESS smoothing algorithm, and an exponential decay function based on turbulent energy spectrum theory is introduced to constrain the correction amplitude in high wind speed areas and prevent overcorrection caused by the deviation of the reference wind field system. Includes the following steps, Step S101: Data preparation and preprocessing; Acquire multi-source satellite wind speed data and corresponding spatiotemporal reference wind field data, and match and bin them according to sensor type and wind speed value; A rigorous quality control process is implemented for all acquired data, including removing data points marked as invalid or of poor quality, removing outliers that are clearly outside the reasonable physical range, and filtering out observations of nearby land to avoid land pollution effects; for spatiotemporal matching, the closest reference field data in spatiotemporal space is found for each satellite observation grid point; all matched data are classified and stored according to satellite sensor IDs to prepare for the subsequent construction of high-precision bias correction lookup tables for each sensor independently; Step S102: Construction of deviation correction lookup table and system stabilization; A dual-partitioning method is adopted, based on a dual-reference wind speed partitioning strategy of satellite and reference wind field. The average deviation between satellite wind speed and reference wind field wind speed in each wind speed box is calculated, and an initial deviation correction lookup table is constructed. For each wind speed box, a hierarchical Bayesian estimator is used to perform weighted fusion of the local average deviation of the wind speed box and the global average prior deviation of the same sensor to obtain a statistically stable deviation estimate. The matching dataset generated by sensor classification in step S101 above is subjected to box-based statistical analysis to divide the wind speed range into multiple continuous wind speed boxes. Step S103: Physical smoothing and constraints of the lookup table; An adaptive local weighted scatter plot smoothing process is applied to the stabilized deviation correction lookup table, with the smoothing bandwidth dynamically adjusted according to the wind speed range. A larger bandwidth is used in the high wind speed range where the sample is sparse to ensure curve smoothness, while a smaller bandwidth is used in the low wind speed range where the sample is abundant to preserve details, thereby suppressing high-frequency fluctuations in the wind speed deviation relationship curve and enhancing its physical continuity. An exponential function-based attenuation constraint is applied to the correction amount in the high wind speed range to prevent overcorrection; thereby, the deviation correction lookup table after stabilization is physically optimized. Step S104: Satellite wind speed data correction; For the satellite wind speed data to be corrected, the final deviation correction lookup table after processing is queried according to its sensor type and wind speed value to obtain the corresponding deviation correction amount, and the original wind speed is corrected and output according to the preset correction rules. The preset correction rule is that when the wind speed is lower than or equal to the wind speed threshold, the deviation correction amount is applied in full; when the wind speed is higher than the wind speed threshold, the deviation correction amount needs to be multiplied by a decay coefficient that decreases exponentially with the increase of wind speed. The setting of wind speed threshold and empirical attenuation rate parameters is optimized based on the turbulent energy spectrum scale law, so that the correction behavior under high wind speed conforms to the laws of fluid dynamics.

2. The sea surface wind speed deviation correction method based on statistical learning and physical constraints according to claim 1, characterized in that: In step S101, the multi-source satellite data refers to wind speed data acquired by satellite remote sensing, which includes satellite scatterometer data, satellite radiometer data, or satellite altimeter data.

3. The sea surface wind speed deviation correction method based on statistical learning and physical constraints according to claim 1, characterized in that: In step S101, the reference wind field data is model data, reanalysis data, multi-source fusion data, or measured data, and it is within the same spatiotemporal range as satellite observations.

4. The sea surface wind speed deviation correction method based on statistical learning and physical constraints according to claim 1, characterized in that: Step S102 employs a dual-reference wind speed zoning strategy; Using the initial wind speed of the satellite respectively and reference wind field wind speed As a reference benchmark, an independent binning statistical analysis was conducted to calculate two preliminary local average deviation sequences. The two sets of deviation sequences are merged by averaging or weighted averaging to form the final deviation correction lookup table.

5. The sea surface wind speed deviation correction method based on statistical learning and physical constraints according to claim 4, characterized in that: Step S102 employs a hierarchical Bayesian shrinkage estimator based on different satellite sensors. initial wind speed of the satellite and reference wind field wind speed The following statistical stabilization process was performed; For the nth anemometer, its statistically stable bias estimate is The calculation is as follows: （1） （2） （3） in, Indicates the first The effective sample size in each anemometer box; It is the local average deviation between the wind speeds of different satellites and the reference wind field within the anemometer. This represents the global average deviation for the same sensor ID, which is used as the prior mean in the hierarchical Bayesian shrinkage estimator. It is the total sample size of all anemometers; parameters This represents the prior weights, which are interpreted as pseudo-sample sizes and used to quantify the global average bias. With local average The relative confidence level.

6. The sea surface wind speed deviation correction method based on statistical learning and physical constraints according to claim 5, characterized in that: The statistically stable deviation estimate is obtained by weighting the local average deviation with the effective sample size of the anemometer as the weight to obtain a first value, weighting the global average prior deviation with the prior pseudo-sample size as the weight to obtain a second value, and then calculating the sum of the first value and the second value by dividing the sum of the effective sample size and the prior pseudo-sample size; wherein, the prior pseudo-sample size is determined based on the total sample size of the satellite sensor data and is set as the square root of the total sample size of the sensor data.

7. A sea surface wind speed deviation correction system based on statistical learning and physical constraints, applying the sea surface wind speed deviation correction method based on statistical learning and physical constraints as described in any one of claims 1 to 6, characterized in that: Including, 1) Data preparation module; This module is used to acquire multi-source satellite wind speed data and corresponding reference wind field data, and to perform quality control, spatiotemporal matching, and classification of the acquired data according to sensor type. It is also responsible for performing boxed statistics according to preset wind speed intervals, calculating the local average deviation between the satellite wind speed and the reference wind field wind speed in each wind speed box and the effective sample size, so as to provide structured basic statistical data for subsequent processing. 2) Lookup table construction module; This module is used to construct an initial bias correction lookup table using a dual-reference wind speed partitioning strategy. The initial satellite wind speed and the reference wind field wind speed are used as independent references for binning statistics, and the two results are fused by weighted average to reduce selection bias. The module also uses a hierarchical Bayesian contraction estimator to perform statistical stabilization processing on the initial lookup table by weighted fusion of the local average bias of each wind speed bin with the global average prior bias of the sensor. 3) Physical constraint module; This module is used to perform adaptive local weighted scatter plot smoothing on the stabilized deviation correction lookup table. Its smoothing bandwidth is dynamically adjusted with the sample density in the wind speed range to filter out high-frequency noise while maintaining the continuity of the curve. The module also introduces an exponential decay function based on the turbulent energy spectrum scale law to impose physical constraints on the correction amount in the high wind speed range to prevent overcorrection caused by the deviation of the reference wind field system. 4) Calibration application module; This module is used to correct and output satellite wind speed data based on the final optimized deviation correction lookup table. It calls the corresponding lookup table according to the sensor type, queries each satellite wind speed observation value to obtain the corresponding deviation correction amount and attenuation coefficient, and calculates the final wind speed product with effective suppression of system deviation through the correction formula.

8. A computer device, characterized in that: The method includes one or more processors and a storage device, wherein the storage device stores one or more programs, and when the one or more programs are executed by the one or more processors, the sea surface wind speed deviation correction method based on statistical learning and physical constraints described in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that: It stores one or more executable programs, and when the programs are executed by the processor of a computer device, the sea surface wind speed deviation correction method based on statistical learning and physical constraints described in any one of claims 1 to 6 is implemented.

Images