Rain-resistant wind speed retrieval method based on GNSS-R and dual-frequency scatterometer of same satellite platform

Abstract

The application discloses a kind of anti-rainfall wind speed inversion methods based on same satellite platform GNSS-R and dual-frequency scatterometer, utilize same platform dual-frequency scatterometer data and synchronous precipitation data, construct dual-band rainfall detection model, identify the rainfall area in the observation width of scatterometer;In the identified rainfall area, the observation point that coincides with same platform GNSS-R space-time is screened, and the anti-rainfall wind speed inverted by GNSS-R is used as reference true value, trains dual-band wind speed inversion model specially used in rainfall condition;The trained model is applied to the entire scatterometer observation width, and the unit determined as rainfall is corrected for wind speed, and the non-rainfall unit is inverted using the standard model, so as to generate high-precision, full-coverage sea surface wind field product under rainfall condition.The application makes full use of the anti-interference characteristics of GNSS-R under rainfall condition, the wide-width detection capability of dual-frequency scatterometer and the advantages of synchronous observation time of same platform, and improves the precision, reliability and spatial integrity of wide-width sea surface wind field product under complex weather.

Description

Technical Field

[0001] This invention relates to the field of satellite ocean remote sensing technology, and in particular to a method for retrieving rainfall-resistant wind speed based on the same satellite platform GNSS-R and a dual-frequency scatterometer. Background Technology

[0002] Spaceborne microwave scatterometers are the primary means of acquiring global sea surface wind fields. Dual-frequency scatterometers (such as C-band and Ku-band) can provide wide-amplitude, high-precision wind field information under clear-sky conditions. However, under rainfall conditions, radar signals suffer significant scattering and absorption attenuation from raindrops, particularly in the shorter wavelength Ku-band, where the impact is more severe. Simultaneously, the "rain noise" generated by raindrops hitting the sea surface masks the sea surface roughness signal caused by wind, leading to a significant decrease in wind speed inversion accuracy or even failure.

[0003] Global Navigation Satellite System Reflection Measurement (GNSS-R) passively receives sea surface scattered echoes from L-band signals of navigation satellites. Due to the longer wavelength of the L-band, it has strong signal penetration capabilities and maintains good wind speed sensitivity even under rainfall conditions, demonstrating a natural advantage in resisting rainfall. However, GNSS-R technology has drawbacks such as low spatial resolution (typically around 25 kilometers) and sparse, strip-like observations, making it unable to meet the needs for wide-swath, high-resolution monitoring of mesoscale weather systems such as typhoons on its own.

[0004] To address the issue of decreased accuracy in scatterometer wind speed retrieval under rainfall conditions, existing technologies attempt to introduce external rainfall products (such as microwave imagers and reanalysis data) to post-process and correct the scatterometer retrieval results using physical correction models. However, sea surface rainfall systems (such as typhoons) exhibit rapid spatiotemporal evolution, and external auxiliary products and satellite scatterometer observations are typically significantly asynchronous (up to the hour), failing to accurately match the rainfall conditions at the moment of observation. This results in poor correction effects and may even introduce new errors. Furthermore, if the observations used to provide true wind speed references (such as buoy data and other satellite transit data) are not synchronized with the scatterometer observations to be corrected, the actual wind field may have changed during this period, and this uncertainty also limits the final retrieval accuracy.

[0005] Existing joint inversion methods are mostly limited to "point-to-point" correction where the GNSS-R specular reflection point and the scatterometer observation unit exactly coincide, failing to fully explore the complementary potential of the "L-band anti-rainfall characteristics" and the "wide swath characteristics of dual-frequency scatterometers." For vast areas within the scatterometer swath without GNSS-R coverage, the problem of wind speed inversion under rainfall conditions remains unresolved. This results in only a few reliable wind speed points being obtained during typhoon and other extreme weather monitoring, making it impossible to grasp the complete wind field structure of the entire storm system.

[0006] For example, Chinese invention patent CN111832175A discloses a method for measuring sea surface wind speed using a scatterometer under rainfall conditions. This method directly utilizes historical scatterometer wind speed and rainfall rate data, constructing a linear regression model with buoy wind speed as the output for correction. This method does not involve GNSS-R technology and relies on historical statistical relationships, failing to address the issues of instantaneous observation matching and spatiotemporal variations in the wind field.

[0007] For example, Chinese invention patent CN120850578A discloses a method for correcting rainfall-affected Ku-band observations from a satellite scatterometer. This method uses C-band data as a reference to correct Ku-band observations affected by rainfall. Although this method involves two bands, the C-band data it relies on may itself be affected by rainfall, and the rainfall data originates from external asynchronous products (such as GPM), resulting in spatiotemporal matching errors. More importantly, this method does not introduce an independent observation source with rainfall resistance (such as GNSS-R) as a true reference, nor does it extend the point-based correction capability to the entire scatterometer observation swath.

[0008] In summary, the existing technology has the following shortcomings:

[0009] (1) Relying on external asynchronous rainfall products or historical statistical models for correction, it is impossible to achieve instantaneous rainfall detection and correction that is strictly synchronized with scatterometer observations, resulting in errors due to time mismatch.

[0010] (2) There is a lack of an independent true source of wind speed that is strictly synchronized with the scatterometer observations and can still maintain high accuracy under rainfall conditions for training and verifying the rainfall correction model. This results in a weak physical foundation for the model and questionable reliability of the inversion results.

[0011] (3) Existing collaborative observation methods are mostly limited to “point correction” of data overlap points, and have failed to systematically extend the local high-precision anti-rainfall observation capability to the entire operational observation swath of the scatterometer, thus failing to meet the needs of sea surface wind field inversion under rainfall conditions with a wide swath and full coverage. Summary of the Invention

[0012] To address the problems existing in the prior art, the present invention aims to provide a method for rain-resistant wind speed inversion based on a GNSS-R and dual-frequency scatterometer on the same satellite platform. By designing a GNSS-R and dual-frequency scatterometer observation system on the same platform, and utilizing the high spatiotemporal synchronization of data within the platform, the rain-resistant wind speed inverted by GNSS-R is used as the ground truth to train a dual-band wind speed inversion model suitable for rainfall conditions. This model is then applied to the entire scatterometer swath, thereby achieving a complete inversion of the sea surface wind field with a wide swath and high precision under rainfall conditions.

[0013] To achieve the above-mentioned objectives, this invention provides a method for retrieving rainfall-resistant wind speeds based on the same satellite platform GNSS-R and a dual-frequency scatterometer, comprising the following steps:

[0014] Step S1: Construct a dual-band rainfall detection model. The dual-band rainfall detection model is obtained based on the backscattering characteristics of the C-band and Ku-band obtained from the same satellite platform. The dual-band rainfall detection model is used to identify the rainfall area of ​​the observation unit within the observation swath of the dual-frequency scatterometer.

[0015] Step S2: In the rainfall area identified by the dual-band rainfall detection model, select the observation points that spatiotemporally overlap with the GNSS-R data of the same satellite platform, and use the sea surface wind speed inverted by the GNSS-R as the reference true value to train a dual-band wind speed inversion model for rainfall conditions.

[0016] Step S3: Apply the trained dual-band wind speed inversion model to the entire observation swath of the dual-frequency scatterometer. For each observation unit, if it is determined to be rainfall, the corrected wind speed is output using the dual-band wind speed inversion model; if it is determined to be non-rainfall, the wind speed is inverted using the standard geophysical model function, thereby generating a wide-swath, high-precision sea surface wind field product.

[0017] According to one technical solution of the present invention, the construction process of the dual-band rainfall detection model includes:

[0018] Step S11: Using the Ku-band backscattering coefficients obtained from the same satellite platform C-band backscattering coefficient The incident angle θ, azimuth angle φ, and rainfall intensity data provided by the microwave imager or precipitation radar on the same satellite platform are used as input sample features and supervision labels.

[0019] Step S12: Using the rainfall intensity data as a supervision label, a rainfall detection model is trained through a classification algorithm.

[0020] According to one technical solution of the present invention, in step S2, the training method of the dual-band wind speed inversion model includes:

[0021] Step S21: The training samples are selected from observation points that are determined to be rainy by the dual-band rainfall detection model and have good GNSS-R data quality, and ensure that the time difference between the GNSS-R specular reflection point and the scatterometer observation unit does not exceed 3 minutes and the center position difference does not exceed 5 kilometers.

[0022] Step S22: Use the equivalent sea surface wind speed at a height of 10 meters obtained by GNSS-R inversion as the true value of the target, and use the corresponding Ku-band and C-band backscattering coefficients, polarization information, observation incident angle and rainfall intensity level as input features;

[0023] Step S23: Train the machine learning model using a regression algorithm, construct the total loss function by superimposing the mean squared error loss function with the wind speed physical constraint term, and establish the mapping relationship between dual-band observations and actual wind speeds under rainfall conditions.

[0024] According to one technical solution of the present invention, before step S1, a data preprocessing and spatiotemporal matching process is further included, specifically including:

[0025] Quality control is performed on GNSS-R L-band reflection signals and C / Ku dual-frequency scatterometer backscattering data acquired synchronously on the same satellite platform;

[0026] The GNSS-R data and dual-frequency scatterometer data are time-aligned to ensure a time difference of ≤3 minutes, and the spatial distance between the GNSS-R specular reflection point and the scatterometer observation unit is kept within 5 kilometers by spatial indexing.

[0027] A mean aggregation algorithm was used to apply the scatterometer data to ensure that its spatial resolution was consistent with that of the GNSS-R data.

[0028] According to one technical solution of the present invention, the satellite platform is configured with:

[0029] A GNSS-R receiver employing a fully digital phased array architecture is used to achieve multi-beam synchronous reception of sea surface reflected signals from multiple navigation satellites;

[0030] The C / Ku dual-frequency scatterometer, which employs a common aperture antenna design, has the main lobe directions of the two bands aligned. Simultaneous spatial rotation observation of the antenna's field of view is achieved through mechanical scanning or platform attitude control.

[0031] The microwave imager or precipitation radar carried on the same satellite platform is used to provide instantaneous rainfall intensity information that is highly matched with the scatterometer observation time.

[0032] According to one aspect of the present invention, a rainfall-resistant wind speed inversion system based on a GNSS-R satellite platform and a dual-frequency scatterometer is proposed, comprising:

[0033] The same satellite platform observation module is used to synchronously acquire GNSS-R L-band reflection signals, C / Ku dual-frequency scatterometer backscattering data, and rainfall data from the same platform microwave imager or precipitation radar;

[0034] The data processing module is used to execute the rainfall-resistant wind speed inversion method based on the same satellite platform GNSS-R and dual-frequency scatterometer as described in any of the above technical solutions, and to generate and output wide-swath sea surface wind field products.

[0035] According to one technical solution of the present invention, the data processing module includes:

[0036] The data preprocessing and spatiotemporal matching unit is used for quality control, time alignment, spatial matching, and resolution unification of observation data;

[0037] The dual-band rainfall detection model unit is used to identify rainfall areas within the scatterometer observation swath by utilizing C-band and Ku-band backscattering characteristics and synchronous rainfall data.

[0038] The dual-band wind speed inversion model training unit is used to train a wind speed inversion model under rainfall conditions in a rainfall area, using spatiotemporally matched GNSS-R inverted wind speed as the ground truth.

[0039] The full-width wind speed correction and product generation unit is used to apply the trained model to the entire scatterometer observation swath, perform wind speed correction in the rainfall area, use the standard model to invert the non-rainfall area, and fuse them to generate the final wind field product.

[0040] According to one technical solution of the present invention, in the same satellite platform observation module, the C / Ku dual-frequency scatterometer adopts a common aperture antenna design and achieves quasi-synchronous observation of the same sea surface area by two bands through mechanical scanning, with the beam pointing switching time difference in the millisecond range.

[0041] Compared with existing technologies, the rain-resistant wind speed inversion method based on the same satellite platform GNSS-R and dual-frequency scatterometer provided by this invention has the following significant technical effects:

[0042] This invention discloses a method for rain-resistant wind speed inversion based on the same satellite platform GNSS-R and a dual-frequency scatterometer. It establishes real-time rainfall detection based on dual-band synchronous data to avoid the time asynchrony problem of external auxiliary products. The wind speed inverted from the L-band rain-resistant characteristics is used as the true value to train the Ku / C dual-band rainfall detection model and the wind speed inversion model. Finally, the rain-resistant capability exhibited by GNSS-R at sparse locations is extended to the entire observation swath of the dual-frequency scatterometer.

[0043] Compared to traditional solutions that rely on external asynchronous rainfall products (updated hourly), this invention directly utilizes instantaneous rainfall data provided by the satellite platform's own payload, which is highly matched with the scatterometer observation time (time difference less than 3 minutes), as a supervisory label for the rainfall detection model and an input feature for the wind speed inversion model. This avoids correction errors caused by data time mismatch and significantly improves the real-time performance and reliability of rainfall detection and wind speed inversion.

[0044] Compared to using C-band data or asynchronous buoy data that may be affected by rainfall as a reference, this invention innovatively uses GNSS-R inversion wind speed observed on the same platform as the scatterometer as the true wind speed value under rainfall conditions. Because L-band signals have strong resistance to rainfall, and because GNSS-R observations and dual-frequency scatterometer observations are strictly synchronized in time (time difference ≤ 3 minutes), model calibration errors caused by the dynamic changes in the wind field are effectively reduced, making the trained calibration model physically more rigorous and the inversion results more reliable.

[0045] Compared to traditional point-to-point collaborative correction techniques, this invention constructs a technical chain of "three-band collaboration (L, C, Ku) - dual-band modeling (C / Ku) - global correction." By training a dual-band wind speed inversion model on sparse GNSS-R observation points and applying this model to the entire scatterometer observation swath (e.g., 1200 km), the rainfall resistance capabilities exhibited by GNSS-R at specific points are successfully "transferred" and extended to the entire coverage area of ​​operational scatterometer products. This solves the problem of wind speed correction under rainfall conditions in areas without GNSS-R coverage, significantly improving the accuracy of wind speed inversion under rainfall conditions while ensuring wide coverage.

[0046] This invention enhances the monitoring capabilities for extreme weather. By generating wide-area sea surface wind field products under rainfall conditions with complete coverage and reliable accuracy, this invention can provide high-quality data support for the forecasting, structural studies, and maritime safety assurance of extreme weather such as typhoons and squall lines, greatly improving the marine meteorological monitoring and service capabilities under complex weather conditions. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the embodiments will be briefly described below.

[0048] Figure 1 The schematic diagram illustrates the overall flowchart of a wide-area rain-resistant sea surface wind speed inversion method according to one embodiment of the present invention;

[0049] Figure 2 This schematic diagram illustrates an observation strip of the same satellite platform GNSS-R and dual-frequency scatterometer according to one embodiment of the present invention.

[0050] Figure 3 A schematic diagram illustrating the hardware structure of a GNSS-R all-digital phased array receiver according to an embodiment of the present invention;

[0051] Figure 4 This diagram illustrates the relationship between the observation area and rainfall distribution of GNSS-R and dual-frequency scatterometer on the same satellite platform according to one embodiment of the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of this application can be combined with each other. The following embodiments only illustrate several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the patent application. It should be pointed out that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of this application, and these all fall within the protection scope of this application.

[0054] like Figures 1 to 4 As shown, the present invention provides a method for retrieving rainfall-resistant wind speed based on the same satellite platform GNSS-R and a dual-frequency scatterometer, comprising the following steps:

[0055] Step S1: Construct a dual-band rainfall detection model. The dual-band rainfall detection model is obtained based on the backscattering characteristics of the C-band and Ku-band obtained from the same satellite platform. The dual-band rainfall detection model is used to identify the rainfall area of ​​the observation unit within the observation swath of the dual-frequency scatterometer.

[0056] Step S2: In the rainfall area identified by the dual-band rainfall detection model, select the observation points that spatiotemporally overlap with the GNSS-R data of the same satellite platform, and use the sea surface wind speed inverted by the GNSS-R as the reference true value to train a dual-band wind speed inversion model for rainfall conditions.

[0057] Step S3: Apply the trained dual-band wind speed inversion model to the entire observation swath of the dual-frequency scatterometer. For each observation unit, if it is determined to be rainfall, the corrected wind speed is output using the dual-band wind speed inversion model; if it is determined to be non-rainfall, the wind speed is inverted using the standard geophysical model function, thereby generating a wide-swath, high-precision sea surface wind field product.

[0058] In this embodiment, firstly, a rainfall detection model is constructed using dual-frequency scatterometer data and synchronous precipitation data from the same platform to achieve real-time rainfall identification over a wide swath. Next, within the identified rainfall area, a C / Ku dual-band wind speed retrieval model specifically designed for rainfall conditions is trained using reliable ground truth wind speeds retrieved from the same platform's GNSS-R with spatiotemporal height matching. Finally, the trained model is extended to the entire scatterometer swath to achieve wind speed correction for all rainfall areas. This process effectively solves the core problems of asynchronous rainfall detection, unreliable ground truth references, and limited correction range in traditional methods, systematically extending the anti-rainfall advantages of GNSS-R to the wide-swath observation capabilities of the scatterometer.

[0059] By utilizing synchronized data within the platform, the timeliness consistency of rainfall detection and model training input features was ensured, avoiding errors introduced by external data delays. Using rainfall-resistant GNSS-R synchronized wind speed as the sole ground truth source provided a highly reliable learning target for the machine learning model, improving the model's physical plausibility and inversion accuracy. Through the "model training-full-domain application" framework, the capability expansion from finite overlapping points to the entire observation surface was achieved, ultimately producing a spatially continuous, fully covered, and significantly more accurate wide-swath sea surface wind field product under rainfall conditions, laying a solid foundation for operational marine meteorological applications.

[0060] In some embodiments of the present invention, the construction process of the dual-band rainfall detection model includes:

[0061] Step S11, Dataset Construction: Utilize backscattering observation data from the C / Ku dual-frequency scatterometer acquired on the same satellite platform, including Ku-band backscattering coefficients. C-band backscattering coefficient The data includes observational geometric parameters such as the incident angle θ and azimuth angle φ. Simultaneously, rainfall intensity data provided by a microwave imager or precipitation radar on the same platform, synchronized in time and altitude (preferably with a time difference of less than 1 minute), is used as a supervisory label y.

[0062] The constructed dataset should cover typical rainfall intensity ranges (e.g., 0-10 mm / h) and wind speed ranges (e.g., 0-25 m / s), and be divided into training, validation, and test sets in a certain ratio (e.g., 7:2:1).

[0063] Step S12, Model Training: Train the model using a machine learning classification model (such as random forest, support vector machine, or deep neural network). The input feature set can be represented as:

[0064] { , , , ,θ,φ};

[0065] This includes backscattering coefficients and geometric information for different polarizations in both bands. The supervision label y is the instantaneous rainfall rate R (mm / h) provided by the precipitation load on the same platform. It can be a continuous value or discretized into multiple rainfall intensity levels (e.g., no rain, 0-2, 2-4, ..., 8-10 mm / h) for multi-class classification training. The training objective is to enable the model to accurately determine whether rainfall exists and its intensity level in the corresponding observation unit based on the dual-frequency scatterometer observations.

[0066] In this embodiment, the present invention constructs a rainfall detection model. By directly using instantaneous precipitation data from the same platform as labels, it ensures strict temporal consistency between the monitoring signal and the scatterometer observations, solving the label inaccuracy problem caused by the reliance on asynchronous external products in traditional methods. By comprehensively utilizing backscattering information from both C and Ku bands and dual polarizations as features, and considering the different response characteristics of different frequencies and polarizations to rainfall, this enriches the information dimensions of rainfall identification, enabling more sensitive and accurate detection of rainfall events, especially weak rainfall or the initial stage of rainfall. The use of machine learning methods can automatically learn the complex nonlinear relationship between rainfall and backscattering features from the data, avoiding the shortcomings of traditional threshold methods or simple physical models, and improving the accuracy and robustness of rainfall detection.

[0067] In some embodiments of the present invention, step S2, the training method of the dual-band wind speed inversion model includes:

[0068] Step S21, Training Sample Selection: From a large amount of observation data, select sample points that simultaneously meet the following conditions for model training:

[0069] (1) The dual-band rainfall detection model determines that "it is raining";

[0070] (2) The GNSS-R observation data on the same platform are of good quality (e.g., high signal-to-noise ratio, far from the shore at the mirror point, etc.).

[0071] (3) The time difference between the GNSS-R mirror reflection point and the dual-frequency scatterometer observation unit shall not exceed 3 minutes, and the spatial distance between their centers shall not exceed 5 kilometers.

[0072] Step S22, Dataset Construction: The equivalent sea surface wind speed at a height of 10 meters obtained by GNSS-R inversion is used as the target ground value, and the corresponding Ku-band and C-band backscattering coefficients, polarization information, observation incident angle, and rainfall intensity level are used as input features.

[0073] Based on the selected sample points, the equivalent sea surface wind speed at a height of 10 meters obtained by GNSS-R inversion was used as the ground truth for supervision. The input feature set is: { , , , ,θ,φ, };

[0074] In addition to the dual-band dual-polarization backscattering coefficient and geometric parameters, the rainfall intensity level output by the rainfall detection model was also included. As an important feature.

[0075] For supervised truth values In rainfall areas where the three bands overlap, wind speed can be retrieved using quality-controlled L-band GNSS-R observation data through a geophysical function model (GMF) that correlates NBRCS and LES characteristic parameters, serving as the ground truth for wind speed monitoring under rainfall conditions. .

[0076] Step S23, Model Training and Loss Function: The machine learning model is trained using a regression algorithm. The total loss function is constructed by superimposing the mean squared error loss function with the wind speed physical constraint term, and the mapping relationship between dual-band observations and actual wind speeds under rainfall conditions is established.

[0077] The model is trained using regression algorithms (such as random forest regression, gradient boosting trees, or neural networks). The loss function is designed to be based on the mean squared error (MSE) plus an additional physical constraint loss term. For example, penalties can be imposed on non-physical extreme wind speed values ​​output by the model (such as negative wind speeds or wind speeds that are outside the reasonable range).

[0078] The total loss function can be expressed as: ;

[0079] in, This is a tradeoff coefficient. By optimizing this loss function, a robust mapping relationship is established between dual-band observations and actual wind speeds under rainfall conditions.

[0080] In this embodiment, training samples are selected through strict spatiotemporal matching conditions to ensure a high degree of consistency between the true wind speed values ​​provided by GNSS-R and scatterometer observations. This minimizes training noise caused by variations in the wind field itself, ensuring that the model learns the true patterns of the impact of rainfall on backscattering, rather than obfuscated signals from wind field changes. Explicitly incorporating rainfall intensity levels into the input features allows the model to perceive the differences in the impact of different rainfall intensities on the signal, enabling more targeted wind speed retrieval and improving the model's adaptability and retrieval accuracy. Adding a physical constraint term to the loss function guides the model to follow fundamental physical laws, avoiding erroneous outputs that might result from purely data-driven approaches, and enhancing the model's reliability and generalization ability.

[0081] In some embodiments of the present invention, a data preprocessing and spatiotemporal matching process is included before step S1, specifically including:

[0082] (a) Preprocessing and quality control of GNSS-R L-band reflection signals and C / Ku dual-frequency scattering backscattering data acquired synchronously on the same satellite platform.

[0083] For GNSS-R data, based on indicators such as signal-to-noise ratio, data quality identifier, and distance of specular reflection points from the shore, low-quality data affected by instrument noise, transmission interference, or nearshore land pollution are eliminated. For scatterometer data, based on its original quality identifier, abnormal data points caused by instrument malfunctions or other problems are mainly eliminated, while those points that may be affected by rainfall but are still valid observations are retained.

[0084] Quality control removes unreliable noisy data, providing clean, high-quality input for subsequent model training, which is beneficial for obtaining a high-precision model.

[0085] (b) Time and spatial location matching: The GNSS-R data and the dual-frequency scatterometer data are time-aligned to ensure that the time difference is ≤3 minutes, and the spatial distance between the GNSS-R specular reflection point and the scatterometer observation unit is within 5 kilometers through spatial indexing;

[0086] Based on the precise timestamps of the satellite platform, GNSS-R data and dual-frequency scatterometer data are time-aligned to ensure that the time difference between matched data pairs does not exceed 3 minutes, in order to accommodate the dynamic rate of change of marine meteorological parameters (such as wind field and rainfall). Spatially, using the latitude and longitude coordinates of the GNSS-R specular reflection point as a reference, spatial indexing algorithms (such as nearest neighbor search) are used to locate the corresponding scatterometer observation unit in its region. The center distance between the observation areas of the two is required to be within 5 kilometers to ensure the consistency of spatial observation objects.

[0087] By setting time (≤3 minutes) and space (≤5 kilometers) matching thresholds, the physical state of the same sea surface and the same moment observed by the GNSS-R and the scatterometer is guaranteed to the maximum extent. This enables the use of GNSS-R as the true value to correct the scatterometer, thereby solving the spatiotemporal mismatch problem that is difficult to overcome in traditional multi-satellite and multi-source data collaboration.

[0088] (c) Spatial resolution uniformity: The scatterometer data is subjected to a mean aggregation algorithm to ensure that its spatial resolution is consistent with that of the GNSS-R data.

[0089] Since the spatial resolution of scatterometers (e.g., about 10 km) is typically higher than that of GNSS-R (e.g., about 25 km), direct matching would lead to a mismatch in information scale. Therefore, for each scatterometer observation unit group that has completed location matching, algorithms such as mean aggregation are used to aggregate multiple high-resolution scatterometer observations into a low-resolution observation that matches the GNSS-R observation scale, thereby eliminating the potential impact of spatial resolution differences on subsequent collaborative modeling.

[0090] By unifying the resolution, the scale effect problem caused by the inherent resolution differences between different sensors is solved, enabling data from different sources to be effectively fused and model-learned on the same spatial scale, thus improving the physical consistency and accuracy of collaborative inversion.

[0091] In some embodiments of the present invention, the satellite platform is configured with:

[0092] A GNSS-R receiver used to achieve multi-beam synchronous reception of sea surface reflected signals from multiple navigation satellites;

[0093] like Figure 3 As shown, the GNSS-R receiver preferably adopts a fully digital phased array architecture, for example, consisting of 32 antenna elements in 4 rows × 8 columns. Each element is equipped with an independent radio frequency front-end and analog-to-digital conversion channel. Through digital beamforming technology, it can achieve high-gain, multi-beam synchronous reception of sea surface reflected signals from multiple navigation satellites (such as GPS, BeiDou, and Galileo), which is beneficial to improving signal quality and spatial sampling rate.

[0094] The GNSS-R receiver uses a fully digital phased array, enabling simultaneous, high-gain reception of reflected signals from multiple navigation satellites. This not only improves the efficiency of GNSS-R observations and the accuracy of wind speed inversion, but also increases the probability of spatial overlap with scatterometer observation points over a wide area, providing more effective samples for model training.

[0095] The C / Ku dual-band scatterometer, employing a shared-aperture antenna design, allows the C-band and Ku-band RF front-ends to share the same reflector antenna or phased array antenna aperture, ensuring strict alignment of the main lobe pointing of the antennas in both bands. Simultaneous spatial and temporal observations of the same sea surface area are achieved through rolling scans of the satellite platform or mechanical rotation of the antenna itself. Due to the shared-aperture and cooperative scanning design, the switching time difference between the beam pointing of the two bands can be controlled within milliseconds, enabling quasi-synchronous observations under constant wind field and rainfall conditions.

[0096] The dual-frequency scatterometer adopts a common aperture design, which helps to ensure strict spatial consistency between C-band and Ku-band observations. This, in turn, facilitates the construction of high-quality dual-band features for rainfall detection and wind speed inversion, thereby reducing beam pointing errors that may be caused by traditional split antennas.

[0097] Microwave imagers or precipitation radars carried on the same satellite platform provide instantaneous rainfall intensity information that closely matches the scatterometer observation times. Rainfall data serves as key supervisory labels for training dual-band rainfall detection models and as important input features for wind speed retrieval models. Observations from the same platform ensure a high degree of spatiotemporal overlap between precipitation data and scatterometer observations, such as... Figure 2 and Figure 4 As shown, this provides reliable rainfall environment information for the model.

[0098] Integrating precipitation measurement payloads on the same platform is beneficial for achieving "synchronous rainfall detection," and it can provide spatiotemporal matching accuracy that traditional cross-platform and cross-task data collaboration cannot achieve.

[0099] In some embodiments of the present invention, step S3 specifically includes:

[0100] Step S31, Rainfall Status Determination: During operational use or historical data inversion, for each independent observation unit within the entire observation swath of the scatterometer, the trained and fixed "dual-band rainfall detection model" is first used, based on the C-band and Ku-band observation values ​​of that unit ( , (θ, φ) to determine whether it is in a rainfall state and the rainfall intensity level.

[0101] Step S32, Differentiated Wind Speed ​​Inversion: Execute different wind speed inversion procedures based on the rainfall determination results:

[0102] If the observation unit is determined to be "rainless", then the standard geophysical model function that is compatible with the scatterometer is used to perform conventional wind speed inversion for that observation unit.

[0103] If the condition is determined to be "rainy" (and the rainfall intensity is within the model's applicable range, such as ≤10 mm / h), then the C-band and Ku-band observations of this unit ( , The wind speed values ​​(θ, φ) and their rainfall intensity levels are input into the trained "dual-band wind speed inversion model", which directly outputs the sea surface wind speed value after rainfall impact correction.

[0104] Step S33, Product Fusion and Output: All wind speed units obtained from the above inversion (including wind speeds in rainless areas retrieved by GMF and wind speeds in rainy areas corrected by the model) are spatially stitched and fused to form a spatially continuous sea surface wind field grid product covering the entire observation swath. The final product is output in standard scientific data formats such as NetCDF. The product includes not only the wind speed field but also information such as wind direction, rainfall status, data quality, observation time, and location, facilitating subsequent applications.

[0105] In some embodiments of the present invention, a regular (e.g., quarterly or annual) model update mechanism is established to retrain or incrementally learn the rainfall detection model and wind speed inversion model using the accumulated new observation data, so as to adapt to the climate characteristics changes in different sea areas and seasons and ensure the long-term stability of product accuracy.

[0106] (1) Regular model updates: Set the model update cycle (e.g., every quarter or every six months). Using newly accumulated observation data from the same platform, retrain or incrementally learn the dual-band rainfall detection model and wind speed inversion model according to the above process. This can adapt to the climate and marine environment characteristics of different sea areas (e.g., tropical typhoon areas, mid-latitude storm areas) and different seasons, and prevent the model performance from drifting over time.

[0107] (2) Continuous accuracy assessment: Establish an automated verification process. Cross-validate the wind field products retrieved by the system with international marine buoy networks (such as NDBC), coastal meteorological stations, reanalysis wind field products (such as ERA5), and other independent observation data such as spaceborne lidar and radar altimeters. Continuously monitor key indicators such as root mean square error (RMSE), bias, and scattering index (SI) of the retrieved wind speed, especially their performance in different rainfall intensity ranges (0-2, 2-5, 5-10 mm / h) and wind speed ranges (low, medium, and high wind speeds).

[0108] By establishing a periodic model update mechanism, the system acquires adaptive and learning evolution capabilities, enabling it to track long-term changes in the ocean-atmosphere environment and sensor status, ensuring the long-term stability of product accuracy, and avoiding performance degradation problems that may occur with static models.

[0109] According to one aspect of the present invention, a rainfall-resistant wind speed inversion system based on a GNSS-R satellite platform and a dual-frequency scatterometer is proposed, comprising:

[0110] The same satellite platform observation module is used to synchronously acquire GNSS-R L-band reflection signals, C / Ku dual-frequency scatterometer backscattering data, and rainfall data from the same platform microwave imager or precipitation radar;

[0111] The data processing module is used to execute the rainfall-resistant wind speed inversion method based on the same satellite platform GNSS-R and dual-frequency scatterometer as described in any of the above technical solutions, and to generate and output wide-swath sea surface wind field products.

[0112] In some embodiments of the present invention, the data processing module includes:

[0113] The data preprocessing and spatiotemporal matching unit is used for quality control, time alignment, spatial matching, and resolution unification of observation data;

[0114] The dual-band rainfall detection model unit is used to identify rainfall areas within the scatterometer observation swath by utilizing C-band and Ku-band backscattering characteristics and synchronous rainfall data.

[0115] The dual-band wind speed inversion model training unit is used to train a wind speed inversion model under rainfall conditions in a rainfall area, using spatiotemporally matched GNSS-R inverted wind speed as the ground truth.

[0116] The full-width wind speed correction and product generation unit is used to apply the trained model to the entire scatterometer observation swath, perform wind speed correction in the rainfall area, use the standard model to invert the non-rainfall area, and fuse them to generate the final wind field product.

[0117] In some embodiments of the present invention, in the same satellite platform observation module, the C / Ku dual-frequency scatterometer adopts a common aperture antenna design and achieves quasi-synchronous observation of the same sea surface area by two bands through mechanical scanning, with the beam pointing switching time difference in the millisecond range.

[0118] In practical applications, through cross-validation with the International Buoy Network and reanalysis wind field products, the rain-resistant wind speed inversion method based on the same satellite platform GNSS-R and dual-frequency scatterometer of this invention can maintain a root mean square error (RMSE) of the retrieved sea surface wind speed at a height of 10 meters within 2 m / s under the condition of rainfall intensity ≤ 10 mm / h, and the absolute value of the bias is less than 0.5 m / s. Compared with the inversion results of the uncorrected scatterometer, the accuracy is significantly improved.

[0119] This invention provides a rainfall-resistant wind speed inversion method based on the same satellite platform GNSS-R and dual-frequency scatterometer. It utilizes the observation data of the same satellite platform itself as the real-time basis for rainfall identification, and the rainfall detection and wind speed observation within the platform can achieve strict time matching, realizing high-precision and real-time rainfall area identification, avoiding the time asynchrony problem caused by relying on external auxiliary rainfall products.

[0120] This invention uses GNSS-R inverted wind speed observed on the same satellite platform as the scatterometer as the true wind speed reference under rainfall conditions. Furthermore, the true wind speed provided by the L-band is highly synchronized with the observations of the dual-frequency scatterometer in time, effectively reducing model calibration errors caused by dynamic changes in wind speed.

[0121] This invention constructs a technical link of "three-band collaboration - dual-band modeling - global correction", which applies the trained model to the entire swath of the scatterometer to solve the problem of wind speed correction of dual-frequency scatterometers under rainfall conditions in areas without GNSS-R coverage. It systematically extends the correction capability based on the synchronous true value to all rainfall areas within the entire observation swath of the scatterometer, realizing wide-swath high-precision rain-resistant wind field monitoring.

[0122] The above description is merely one embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to the technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for retrieving wind speed under rainfall conditions based on a GNSS-R satellite platform and a dual-frequency scatterometer, characterized in that, Includes the following steps: Step S1: Construct a dual-band rainfall detection model. The dual-band rainfall detection model is obtained based on the backscattering characteristics of the C-band and Ku-band obtained from the same satellite platform. The dual-band rainfall detection model is used to identify the rainfall area of ​​the observation unit within the observation swath of the dual-frequency scatterometer. Step S2: In the rainfall area identified by the dual-band rainfall detection model, select the observation points that spatiotemporally overlap with the GNSS-R data of the same satellite platform, and use the sea surface wind speed inverted by the GNSS-R as the reference true value to train a dual-band wind speed inversion model for rainfall conditions. Step S3: Apply the trained dual-band wind speed inversion model to the entire observation swath of the dual-frequency scatterometer. For each observation unit, if it is determined to be rainfall, the corrected wind speed is output using the dual-band wind speed inversion model. If the wind speed is determined to be non-rainfall, a standard geophysical model function is used to invert the wind speed, thereby generating a wide-area, high-precision sea surface wind field product.

2. The rain-resistant wind speed inversion method based on GNSS-R and a dual-frequency scatterometer on the same satellite platform as described in claim 1, characterized in that, The construction process of the dual-band rainfall detection model includes: Step S11: Using the Ku-band backscattering coefficients obtained from the same satellite platform C-band backscattering coefficient The incident angle θ, azimuth angle φ, and rainfall intensity data provided by the microwave imager or precipitation radar on the same satellite platform are used as input sample features and supervision labels. Step S12: Using the rainfall intensity data as a supervision label, a rainfall detection model is trained through a classification algorithm.

3. The rain-resistant wind speed inversion method based on GNSS-R and a dual-frequency scatterometer on the same satellite platform as described in claim 2, characterized in that, In step S2, the training method for the dual-band wind speed inversion model includes: Step S21: The training samples are selected from observation points that are determined to be rainy by the dual-band rainfall detection model and have good GNSS-R data quality, and ensure that the time difference between the GNSS-R specular reflection point and the scatterometer observation unit does not exceed 3 minutes and the center position difference does not exceed 5 kilometers. Step S22: Use the equivalent sea surface wind speed at a height of 10 meters obtained by GNSS-R inversion as the true value of the target, and use the corresponding Ku-band and C-band backscattering coefficients, polarization information, observation incident angle and rainfall intensity level as input features; Step S23: Train the machine learning model using a regression algorithm, construct the total loss function by superimposing the mean squared error loss function with the wind speed physical constraint term, and establish the mapping relationship between dual-band observations and actual wind speeds under rainfall conditions.

4. The rain-resistant wind speed inversion method based on GNSS-R and a dual-frequency scatterometer on the same satellite platform as described in claim 1, characterized in that, Before step S1, a data preprocessing and spatiotemporal matching process is also included, specifically including: Quality control is performed on GNSS-R L-band reflection signals and C / Ku dual-frequency scatterometer backscattering data acquired synchronously on the same satellite platform; The GNSS-R data and dual-frequency scatterometer data are time-aligned to ensure a time difference of ≤3 minutes, and the spatial distance between the GNSS-R specular reflection point and the scatterometer observation unit is kept within 5 kilometers by spatial indexing. A mean aggregation algorithm was used to apply the scatterometer data to ensure that its spatial resolution was consistent with that of the GNSS-R data.

5. The rain-resistant wind speed inversion method based on GNSS-R and a dual-frequency scatterometer on the same satellite platform as described in claim 1, characterized in that, The same satellite platform is equipped with: A GNSS-R receiver employing a fully digital phased array architecture is used to achieve multi-beam synchronous reception of sea surface reflected signals from multiple navigation satellites; The C / Ku dual-frequency scatterometer, which employs a common aperture antenna design, has the main lobe directions of the two bands aligned. Simultaneous spatial rotation observation of the antenna's field of view is achieved through mechanical scanning or platform attitude control. The microwave imager or precipitation radar carried on the same satellite platform is used to provide instantaneous rainfall intensity information that is highly matched with the scatterometer observation time.

6. A rainfall-resistant wind speed inversion system based on the same satellite platform GNSS-R and a dual-frequency scatterometer, characterized in that, include: The same satellite platform observation module is used to synchronously acquire GNSS-R L-band reflection signals, C / Ku dual-frequency scatterometer backscattering data, and rainfall data from the same platform microwave imager or precipitation radar; The data processing module is used to execute the rain-resistant wind speed inversion method based on the same satellite platform GNSS-R and dual-frequency scatterometer as described in any one of claims 1-5, and to generate and output wide-swath sea surface wind field products.

7. The wide-swath, rain-resistant sea surface wind speed inversion system based on the same satellite platform GNSS-R and a dual-frequency scatterometer as described in claim 6, characterized in that, The data processing module includes: The data preprocessing and spatiotemporal matching unit is used for quality control, time alignment, spatial matching, and resolution unification of observation data; The dual-band rainfall detection model unit is used to identify rainfall areas within the scatterometer observation swath by utilizing C-band and Ku-band backscattering characteristics and synchronous rainfall data. The dual-band wind speed inversion model training unit is used to train a wind speed inversion model under rainfall conditions in a rainfall area, using spatiotemporally matched GNSS-R inverted wind speed as the ground truth. The full-width wind speed correction and product generation unit is used to apply the trained model to the entire scatterometer observation swath, perform wind speed correction in the rainfall area, use the standard model to invert the non-rainfall area, and fuse them to generate the final wind field product.

8. The wide-swath, rain-resistant sea surface wind speed inversion system based on a GNSS-R satellite platform and a dual-frequency scatterometer as described in claim 6 or 7, characterized in that, In the same satellite platform observation module, the C / Ku dual-frequency scatterometer adopts a common aperture antenna design and achieves quasi-synchronous observation of the same sea surface area by two bands through mechanical scanning, with the beam pointing switching time difference in the millisecond range.

Images