A small-scale drift buoy observation data calibration method based on ground wave radar data

Abstract

The application provides a small-scale drift buoy observation data calibration method based on ground wave radar data, relates to the field of marine environment observation, and comprises the following steps: synchronously receiving wind, wave and flow parameters inverted by a small-scale drift buoy and gridded ocean dynamic field data of a ground wave radar network on the shore; performing space-time matching on the two, extracting radar observation values at the position of the buoy as a reference benchmark; constructing a bias fitting model based on historical matching data; performing batch calibration on the to-be-calibrated buoy data according to a preset time window, subtracting the model predicted bias from the original inverted value to obtain the calibrated value; regularly updating the fitting model and fine-tuning the buoy deep learning inversion model by using accumulated data. The application changes the ground wave radar from a verified object to an active calibration benchmark source, constructs a closed-loop self-evolution system of 'observation-calibration-feedback-optimization', effectively eliminates the cumulative systematic bias of the buoy inversion data, improves the long-term reliability of the data, and reduces the operation and maintenance cost.

Description

Technical Field

[0001] This application relates to the field of marine environmental observation, and in particular to a method for calibrating observation data of a small drifting buoy based on ground wave radar data. Background Technology

[0002] Currently, the following two methods are commonly used in the field of marine environmental observation to obtain surface current, wind field and wave data.

[0003] One approach is the miniaturized drifting buoy independent observation system. These buoys are equipped with MEMS motion sensors and GPS modules. By collecting triaxial acceleration and angular velocity data, they calculate the wave direction spectrum and then use a pre-built inversion algorithm based on the wave direction spectrum and a deep learning model to output key marine environmental parameters such as wind, waves, and currents. This approach, with its advantages of low cost and mass deployment, has become an important supplement to traditional large buoy observation networks. However, its inherent drawbacks include insufficient generalization ability of the deep learning model, limited by the training dataset. It is prone to predictive system biases in complex sea conditions or unfamiliar waters. Furthermore, it lacks the ability to access external independent reference benchmarks during drifting periods lasting several months, making it difficult to consistently guarantee the long-term reliability of the data.

[0004] Secondly, there is the conventional approach of cross-validating onshore ground-wave radar networks with drifting buoys. This approach treats the trajectory of a buoy that accidentally drifts into the radar coverage area as a moving "true reference," and extracts velocity data through spatiotemporal matching to evaluate the a posteriori accuracy of the radar flow field products. This method has significant limitations: it unidirectionally serves radar product evaluation, failing to systematically improve the quality of buoy data by feeding back valuable radar observation information; and the comparison results usually stop at the analysis report, failing to form an optimized closed loop that provides real-time feedback to the buoy data processing chain. Furthermore, the scheme of using satellite remote sensing data for large-scale comparison also suffers from low spatiotemporal resolution and high uncertainty in the inverted products, making it difficult to meet the practical needs of high-frequency, fine-grained calibration near the coast. Summary of the Invention

[0005] The purpose of this invention is to address the lack of a complete technical solution in the existing technology that can actively and continuously utilize a high-precision shore-based remote sensing network to perform real-time calibration of multidimensional inversion data of small drifting buoys, thereby achieving closed-loop optimization of data quality at low cost. This invention provides a method for calibrating observation data of small drifting buoys based on ground wave radar data.

[0006] The above-mentioned objective of this application is achieved through the following technical solution: S1: Acquire gridded ocean dynamic field data from radar observation networks and buoy data from small drifting buoys; S2: Perform spatiotemporal matching between gridded ocean dynamic field data and buoy data to obtain buoy-radar data and store it in the historical matching database; S3: Calculate the deviation value according to the parameter type of buoy-radar data; collect the deviation value within a certain period, combine it with the environmental parameters at the corresponding time, construct the fitting model and train it; S4: Perform batch calibration on newly received buoy data according to a preset time window to obtain calibration values; generate standard format data product files based on calibration values.

[0007] Optionally, step S1 includes: Real-time buoy data is collected by small drifting buoys deployed in the target sea area; The small drifting buoy is equipped with a MEMS motion sensor and a GPS module. It calculates the wave direction spectrum by collecting triaxial acceleration and angular velocity data, and outputs the wind, wave and current parameters of the small drifting buoy's location through a pre-built deep learning inversion model. Buoy data includes: buoy identification, latitude and longitude coordinates, timestamp, and inverted wind, wave, and current parameters; The gridded ocean dynamic field data covering the target sea area is continuously generated through the onshore base wave radar observation network. The gridded ocean dynamic field data includes: surface current vector field, sea surface wind field and significant wave height field.

[0008] Optionally, step S2 includes: Based on the latitude and longitude coordinates and timestamps in the buoy data, a nearest neighbor search and interpolation are performed in the three dimensions of longitude, latitude and time in the gridded ocean dynamic field data to extract the radar observation value of the buoy's location at that moment. The radar observations include: radar current vector, radar wind direction, radar wind speed, and radar significant wave height; successfully matched buoy-radar data pairs are marked and stored in the historical matching database; Built-in fault tolerance mechanism during spatiotemporal matching: In terms of spatial dimension, if the buoy's position falls into the radar coverage blind zone, it will search for the nearest valid grid point data within the preset maximum search radius; if there is no valid data within the radius, the matching is deemed to have failed. In terms of time dimension, if the buoy timestamp is not perfectly aligned with the radar data time layer, the system will prioritize the radar data layer that is the nearest in time; data that fails to match will skip the calibration process, retain the original inversion value and mark it as uncalibrated.

[0009] Optionally, step S3 includes: Identify valid matching data pairs of buoy-radar data in the historical matching database; The system has built-in model selection logic that automatically selects the best fit model based on the current amount of data accumulated. When the number of valid matching data pairs is lower than the preset threshold, a multidimensional lookup table method is used to classify and statistically average the deviation values ​​according to the environmental parameter range. When the number of valid matching data pairs exceeds the threshold, the system automatically switches to parametric fitting, using the least squares method to determine the functional relationship between the deviation value and the key environmental variables. In the parametric fitting method, the input features of the fitting model include: radar observations, significant wave height, geographical zoning identifiers of the buoy's location, and radar data quality identifiers; after the fitting model is trained, it is saved to the data center for online calibration.

[0010] Optionally, step S4 includes: For each buoy data point to be calibrated, spatiotemporal matching is performed. If the matching is successful, the pre-trained fitting model is called to predict the expected deviation value corresponding to that buoy data point. Based on the expected deviation value, the buoy data to be calibrated is corrected to obtain the calibration value; The data product file includes: raw inversion values, radar reference values, calibration values, and a quality identification field; The quality identification field is used to indicate the confidence level of the calibration value; Radar reference values ​​are obtained from gridded ocean dynamic field data; The original inversion values ​​were obtained from buoy data.

[0011] Optionally, the method further includes: Periodically retrain the fitted model using newly accumulated effective matching data, and replace the old fitted model with the newly trained fitted model; The buoy deep learning inversion model is fine-tuned periodically using newly accumulated effective matching data, and the calibration feedback signal is applied to the parameters of the deep learning inversion model.

[0012] An electronic device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to enable the electronic device to perform a calibration method for small drifting buoy observation data based on ground wave radar data.

[0013] A computer-readable storage medium storing instructions that, when executed, perform a method for calibrating observation data of a small drifting buoy based on ground wave radar data.

[0014] The beneficial effects of the technical solution provided in this application are: 1. Effectively improves the accuracy of buoy inversion data: This invention utilizes independent, continuous, and high spatiotemporal resolution ocean dynamic parameters provided by a ground-wave radar network as a reference benchmark to systematically correct biases in buoy inversion data. The calibrated current velocity, wind, and wave parameters can effectively eliminate regional and sea state-dependent systematic errors caused by insufficient model generalization ability, thus significantly improving the data quality of low-cost drifting buoys.

[0015] 2. Reduced lifecycle maintenance costs: Replacing traditional on-site comparative voyage verification with operational online calibration significantly reduces the manpower and ship time costs associated with verifying data accuracy. The calibration process is automated and requires no manual intervention, making long-term, large-scale data quality control economically feasible and further solidifying the core advantage of low cost for small drifting buoys.

[0016] 3. Enhancing the completeness and data reliability of the observation system: This invention organically combines single-point mobile observation with a shore-based surface remote sensing network to form a "point-surface fusion" integrated observation-calibration system. The calibrated data products are accompanied by quality labels, and the data production chain is transparent and traceable, improving the authority and acceptance of the data in fields such as scientific research assimilation, marine forecasting, and engineering applications.

[0017] 4. Promoting the continuous evolution of the inversion model: The closed-loop feedback mechanism enables the buoy inversion model to be fine-tuned periodically using the massive amount of deviation samples accumulated during the calibration process. After multiple iterations, the model's generalization ability and inversion accuracy gradually improve, forming a virtuous cycle of "the more data, the more accurate the model, and the lighter the calibration," laying the technological foundation for building a new generation of self-learning marine intelligent sensors.

[0018] 5. Qualitative Comparison with Uncalibrated Data: Compared to the raw buoy inversion data not calibrated using this method, the data product calibrated using this method exhibits significant advantages in the following aspects: When the buoy traverses different sea state areas, the consistency between the calibrated data and the radar reference standard is significantly higher, the transition between regions is smoother, and regional jumps caused by insufficient model generalization are eliminated. During the long-term drift of the buoy, the time series stability of the calibrated data is significantly improved, effectively suppressing the growth trend of cumulative systematic bias. This improvement makes the data product of low-cost drifting buoys closer in quality to the standard of large fixed buoys, significantly expanding its applicable scenarios in downstream applications.

[0019] The various components of this system work together to form an enhanced closed loop of observation, transmission, fusion, service, and optimization. The buoy and radar network conduct independent and synchronous observations in the same sea area, with data from both sides transmitted in real-time to the data center via a communication link. Within the data center, the calibration method of this invention is triggered and executed, using the high-precision surface data of the radar network as a benchmark to correct the buoy point data, producing a fused data product. This product is provided to users through the platform, and the deviation analysis and model optimization results generated during the calibration process can be fed back to improve the next-generation buoy inversion algorithm or guide the optimization of observation tasks. Attached Figure Description

[0020] The present application will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a step diagram of an embodiment of this application; Figure 2 This is a comparison diagram of the buoy flow velocity deviation distribution before and after calibration in the embodiments of this application; Figure 3 This is a closed-loop self-optimization learning curve diagram in the embodiments of this application; Figure 4 This is a comparison chart of the effects of different calibration methods in the embodiments of this application; Figure 5 This is a schematic diagram of the electronic device structure in the embodiments of this application. Detailed Implementation

[0021] To provide a clearer understanding of the technical features, objectives, and effects of this application, the specific embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0022] The embodiments of this application provide a method for calibrating observation data of a small drifting buoy based on ground wave radar data.

[0023] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating the steps of a method for calibrating observation data of a small drifting buoy based on ground wave radar data, as described in an embodiment of this application, including: S1: Acquire gridded ocean dynamic field data from radar observation networks and buoy data from small drifting buoys; S2: Perform spatiotemporal matching between gridded ocean dynamic field data and buoy data to obtain buoy-radar data and store it in the historical matching database; S3: Calculate the deviation value according to the parameter type of buoy-radar data; collect the deviation value within a certain period, combine it with the environmental parameters at the corresponding time, construct the fitting model and train it; S4: Perform batch calibration on newly received buoy data according to a preset time window to obtain calibration values; generate standard format data product files based on calibration values.

[0024] Step S1 includes: Real-time buoy data is collected by small drifting buoys deployed in the target sea area; The small drifting buoy is equipped with a MEMS motion sensor and a GPS module. It calculates the wave direction spectrum by collecting triaxial acceleration and angular velocity data, and outputs the wind, wave and current parameters of the small drifting buoy's location through a pre-built deep learning inversion model. Buoy data includes: buoy identification, latitude and longitude coordinates, timestamp, and inverted wind, wave, and current parameters; The gridded ocean dynamic field data covering the target sea area is continuously generated through the onshore base wave radar observation network. The gridded ocean dynamic field data includes: surface current vector field, sea surface wind field and significant wave height field.

[0025] As one embodiment, the ground wave radar observation network consists of at least two ground wave radars, with a spatial resolution of 1 to 3 kilometers and a temporal resolution of 10 to 60 minutes for the radar data products; small drifting buoys periodically upload data packets via satellite communication links, the data packets containing buoy identification, latitude and longitude coordinates, timestamps, and inverted wind, wave, and current parameters.

[0026] Step S2 includes: Based on the latitude and longitude coordinates and timestamps in the buoy data, a nearest neighbor search and interpolation are performed in the three dimensions of longitude, latitude and time in the gridded ocean dynamic field data to extract the radar observation value of the buoy's location at that moment. The radar observations include: radar current vector, radar wind direction, radar wind speed, and radar significant wave height; successfully matched buoy-radar data pairs are marked and stored in the historical matching database; Built-in fault tolerance mechanism during spatiotemporal matching: In terms of spatial dimension, if the buoy's position falls into the radar coverage blind zone, it will search for the nearest valid grid point data within the preset maximum search radius; if there is no valid data within the radius, the matching is deemed to have failed. In terms of time dimension, if the buoy timestamp is not perfectly aligned with the radar data time layer, the system will prioritize the radar data layer that is the nearest in time; data that fails to match will skip the calibration process, retain the original inversion value and mark it as uncalibrated.

[0027] The data processing center performs spatiotemporal matching between the received buoy data and radar grid data. Specifically, based on the latitude and longitude coordinates and timestamp in the buoy data packet, it performs nearest neighbor search and bilinear interpolation along the longitude, latitude, and time dimensions of the radar data cube to extract the radar observation value of the buoy's location at that moment. The extracted parameters include radar current vector, radar wind direction, radar wind speed, and radar significant wave height.

[0028] As one implementation, the system incorporates a fault-tolerant mechanism during the spatiotemporal matching process. Spatially, if the buoy's position happens to fall within a radar coverage blind spot (such as island obstruction or near-shore edge), the system will search for the nearest valid grid point data within a preset maximum search radius; if no valid data is found within the radius, the matching is considered a failure. Temporally, if the buoy's timestamp and the radar data time layer are not perfectly aligned, the system prioritizes the radar data layer that is the temporally nearest neighbor. This fault-tolerant mechanism ensures that as many valid matching pairs as possible can be obtained even under boundary conditions. Successfully matched buoy-radar data pairs are marked and stored in the historical matching database. Data that fails to match skips the calibration process, retains the original inversion value, and is marked as "uncalibrated."

[0029] Step S3 includes: Identify valid matching data pairs of buoy-radar data in the historical matching database; The system has built-in model selection logic that automatically selects the best fit model based on the current amount of data accumulated. When the number of valid matching data pairs is lower than the preset threshold, a multidimensional lookup table method is used to classify and statistically average the deviation values ​​according to the environmental parameter range. Once the number of valid matching data pairs exceeds the threshold, the system automatically switches to parametric fitting, using the least squares method to determine the functional relationship between the deviation value and key environmental variables.

[0030] In the parametric fitting method, the input features of the fitting model include: radar observations, significant wave height, geographical zoning identifiers of the buoy's location, and radar data quality identifiers; after the fitting model is trained, it is saved to the data center for online calibration.

[0031] As one embodiment, the fitting model is used to describe the statistical regularity of the deviation between the buoy inversion value (buoy data) and the radar reference value. This step is performed offline or quasi-online, aiming to establish the statistical regularity of the deviation between the buoy inversion value and the radar reference value.

[0032] As one implementation, for valid matching data pairs accumulated in the historical matching database, deviation values ​​are calculated separately according to parameter type. Taking flow velocity as an example, the deviation is defined as the difference between the buoy-inverted flow velocity and the radar-observed flow velocity. Deviation data is collected over a certain period and combined with environmental parameters at the corresponding time to construct a lightweight fitting model. The system has built-in model selection logic, which automatically selects a suitable model based on the current amount of accumulated data: when the number of valid matching data pairs is lower than a preset threshold, a multidimensional lookup table method is used to classify and statistically average the deviation values ​​according to the environmental parameter intervals; when the amount of data exceeds the threshold, the system automatically switches to the parametric fitting method, using the least squares method to determine the functional relationship between the deviation and key environmental variables. The input features of the fitting model include radar observations, significant wave height, the geographical zoning identifier of the buoy's location, and radar data quality identifiers. After the fitting model is trained, it is saved to the data center for online calibration. More importantly, the accumulated deviation data will be used periodically to fine-tune the buoy's original deep learning inversion model. The fine-tuning process directly applies the calibration feedback signal to the model parameters, optimizing the accuracy of the inversion algorithm from the root. This is the core component of the closed-loop self-optimization mechanism of this invention.

[0033] Step S4 includes: For each buoy data point to be calibrated, spatiotemporal matching is performed. If the matching is successful, the pre-trained fitting model is called to predict the expected deviation value corresponding to that buoy data point. Based on the expected deviation value, the buoy data to be calibrated is corrected to obtain the calibration value; The data product file includes: raw inversion values, radar reference values, calibration values, and a quality identification field; The quality identification field is used to indicate the confidence level of the calibration value; Radar reference values ​​are obtained from gridded ocean dynamic field data; The original inversion values ​​were obtained from buoy data.

[0034] As one implementation, the data center performs batch calibration on newly received buoy data within a preset time window. The batch window size is set to match the buoy data upload frequency and radar data update cycle, typically set to 6 hours or 24 hours. If a piece of data in a batch cannot be calibrated due to a matching failure, the system will skip that piece of data without affecting the normal processing of other data in the same batch. In addition, the system supports an event-triggered priority processing mode: when a buoy is detected to enter the radar coverage area for the first time, a priority calibration of the latest data for that buoy can be triggered immediately to quickly obtain high-quality status information.

[0035] As one example, the batch window size is set to match the buoy data upload frequency and the radar data update cycle; the system supports an event-triggered priority processing mode: when a buoy is detected to enter the radar coverage area for the first time, a priority calibration of the latest data for that buoy is triggered.

[0036] The method further includes: Periodically retrain the fitted model using newly accumulated effective matching data, and replace the old fitted model with the newly trained fitted model; The buoy deep learning inversion model is fine-tuned periodically using newly accumulated effective matching data, and the calibration feedback signal is applied to the parameters of the deep learning inversion model.

[0037] As one implementation, the system periodically executes a model update process. Newly accumulated valid matching data pairs are incorporated into the training dataset during the period, the fitted model is retrained, and the old model is replaced with the new model. Before replacing the old model, the system evaluates the new model using a reserved validation dataset. The replacement operation is only performed if the new model's calibration performance on the validation set is no worse than the old model. This anti-degradation mechanism ensures that the calibrated model continues to optimize in actual operation without performance regression. Simultaneously, the buoy deep learning inversion model is periodically fine-tuned using accumulated data. The fine-tuning frequency of the deep learning model is typically lower than the update frequency of the fitted model; both can be configured independently based on the data accumulation rate. This closed-loop calibration-feedback-optimization mechanism enables the system to continuously self-evolve.

[0038] In one embodiment, the core innovation of this invention lies in designing the calibration process as a self-evolving closed-loop system. After the buoy inversion data is calibrated against a radar benchmark, the calibration results not only directly improve the quality of the data product but also serve as feedback signals used periodically to fine-tune the buoy's deep learning inversion model. This closed-loop mechanism of "observation-calibration-feedback-optimization" allows the accuracy of the buoy inversion model to continuously improve over time, fundamentally reducing the correction magnitude of subsequent calibrations. The longer the system runs and the more data accumulated, the more accurate the inversion model becomes, forming a virtuous cycle. This is the essential difference from existing methods that only perform one-time comparative analysis, and it is also the most important technical innovation of this invention.

[0039] In another embodiment, the present invention employs a background algorithm framework of "spatiotemporal matching combined with statistical fitting," avoiding the high computational overhead and engineering implementation difficulties of complex numerical assimilation models. The fitting model has low computational complexity, can run stably on conventional servers, and facilitates rapid deployment and long-term maintenance in operational marine observation systems. The combination of batch processing mode and event-triggered priority mechanism further optimizes system resource utilization efficiency and ensures scalability in large-scale buoy networking scenarios. This design gives the solution clear engineering practical value.

[0040] In one exemplary embodiment, the present invention breaks with the conventional understanding in the art of "using buoys to verify radar," and for the first time systematically transforms ground wave radar from a "verified object" into an "active calibration reference source." This transformation upgrades buoy data quality control from passive, random post-hoc verification to proactive, continuous operational calibration, solving the industry pain point that drifting buoys cannot obtain quality assurance after leaving the radar coverage area. Figure 2 This is a comparison chart of the buoy velocity deviation distribution before and after calibration. As shown in the figure, after calibration using this method, the mean deviation decreased from 0.25 m / s to 0.15 m / s, and the deviation dispersion was significantly narrowed, indicating that the calibration effectively suppressed random errors. Figure 3 This is a closed-loop self-optimization learning curve. As the system running time increases and the inversion model is periodically fine-tuned, the average residual deviation after calibration gradually decreases from approximately 0.15 m / s initially to approximately 0.06 m / s, verifying the effectiveness of the "observation-calibration-feedback-optimization" closed-loop mechanism. Figure 4 This is a comparison chart of the effects of different calibration methods. Using the root mean square error (RMSE) of flow velocity as the indicator, the RMSE after calibration using this method is 0.12 m / s, which is about 57% more accurate than the uncalibrated data and about 40% more accurate than the traditional linear regression calibration, demonstrating the significant advantages of this approach.

[0041] This application also discloses an electronic device. (See reference...) Figure 5 , Figure 5 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of this application. The electronic device 500 may include: at least one processor 501, at least one network interface 504, a user interface 503, a memory 505, and at least one communication bus 502.

[0042] The communication bus 502 is used to enable communication between these components.

[0043] The user interface 503 may include a display screen, and optionally, the user interface 503 may also include a standard wired interface or a wireless interface.

[0044] The network interface 504 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0045] This application also discloses a computer-readable storage medium storing multiple instructions adapted for loading by a processor to execute the above-described method for calibrating small drifting buoy observation data based on ground wave radar data.

[0046] The above are merely exemplary embodiments of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure.

[0047] This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A method for calibrating observation data from a small drifting buoy based on ground wave radar data, characterized in that, The method includes the following steps: S1: Acquire gridded ocean dynamic field data from radar observation networks and buoy data from small drifting buoys; S2: Perform spatiotemporal matching between gridded ocean dynamic field data and buoy data to obtain buoy-radar data and store it in the historical matching database; S3: Calculate the deviation value according to the parameter type of buoy-radar data; collect the deviation value within a certain period, combine it with the environmental parameters at the corresponding time, construct the fitting model and train it; S4: Perform batch calibration on newly received buoy data according to a preset time window to obtain calibration values; generate standard format data product files based on calibration values.

2. The method for calibrating observation data of a small drifting buoy based on ground wave radar data as described in claim 1, characterized in that, Step S1 includes: Real-time buoy data is collected by small drifting buoys deployed in the target sea area; The small drifting buoy is equipped with a MEMS motion sensor and a GPS module. It calculates the wave direction spectrum by collecting triaxial acceleration and angular velocity data, and outputs the wind, wave and current parameters of the small drifting buoy's location through a pre-built deep learning inversion model. Buoy data includes: buoy identification, latitude and longitude coordinates, timestamp, and inverted wind, wave, and current parameters; The gridded ocean dynamic field data covering the target sea area is continuously generated through the onshore base wave radar observation network. The gridded ocean dynamic field data includes: surface current vector field, sea surface wind field and significant wave height field.

3. The method for calibrating observation data of a small drifting buoy based on ground wave radar data as described in claim 1, characterized in that, Step S2 includes: Based on the latitude and longitude coordinates and timestamps in the buoy data, a nearest neighbor search and interpolation are performed in the three dimensions of longitude, latitude and time in the gridded ocean dynamic field data to extract the radar observation value of the buoy's location at that moment. The radar observations include: radar current vector, radar wind direction, radar wind speed, and radar significant wave height; successfully matched buoy-radar data pairs are marked and stored in the historical matching database; Built-in fault tolerance mechanism during spatiotemporal matching: In terms of spatial dimension, if the buoy's position falls into the radar coverage blind zone, it will search for the nearest valid grid point data within the preset maximum search radius; if there is no valid data within the radius, the matching is deemed to have failed. In terms of time dimension, if the buoy timestamp is not perfectly aligned with the radar data time layer, the system will prioritize the radar data layer that is the nearest in time; data that fails to match will skip the calibration process, retain the original inversion value and mark it as uncalibrated.

4. The method for calibrating observation data of a small drifting buoy based on ground wave radar data as described in claim 2, characterized in that, Step S3 includes: Identify valid matching data pairs of buoy-radar data in the historical matching database; The system has built-in model selection logic that automatically selects the best fit model based on the amount of data currently accumulated. When the number of valid matching data pairs is lower than a preset threshold, a multidimensional lookup table method is used to classify and statistically average the deviation values ​​according to the environmental parameter range. When the number of valid matching data pairs exceeds the threshold, the system automatically switches to parametric fitting, using the least squares method to determine the functional relationship between the deviation value and the key environmental variables. In the parametric fitting method, the input features of the fitting model include: radar observations, significant wave height, geographical zoning identifiers of the buoy's location, and radar data quality identifiers; after the fitting model is trained, it is saved to the data center for online calibration.

5. The method for calibrating observation data of a small drifting buoy based on ground wave radar data as described in claim 1, characterized in that, Step S4 includes: For each buoy data point to be calibrated, spatiotemporal matching is performed. If the matching is successful, the pre-trained fitting model is called to predict the expected deviation value corresponding to that buoy data point. Based on the expected deviation value, the buoy data to be calibrated is corrected to obtain the calibration value; The data product file includes: raw inversion values, radar reference values, calibration values, and a quality identification field; The quality identification field is used to indicate the confidence level of the calibration value; Radar reference values ​​are obtained from gridded ocean dynamic field data; The original inversion values ​​were obtained from buoy data.

6. The method for calibrating observation data of a small drifting buoy based on ground wave radar data as described in claim 4, characterized in that, The method further includes: Periodically retrain the fitted model using newly accumulated effective matching data, and replace the old fitted model with the newly trained fitted model; The buoy deep learning inversion model is fine-tuned periodically using newly accumulated effective matching data, and the calibration feedback signal is applied to the parameters of the deep learning inversion model.

7. An electronic device, characterized in that, The device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to enable the electronic device to perform the small drifting buoy observation data calibration method based on ground wave radar data as described in any one of claims 1-6.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed by a computer, perform the small drifting buoy observation data calibration method based on ground wave radar data as described in any one of claims 1-6.

Images