A microwave water vapor atmospheric motion vector retrieval method and system based on convex optimization

Abstract

The application relates to the fields of satellite remote sensing data processing, atmospheric motion vector inversion and microwave water vapor detection, and discloses a microwave water vapor atmospheric motion vector inversion method and system based on convex optimization. The method comprises the following steps: obtaining a plurality of microwave water vapor observation data, and obtaining a multi-time-phase multi-channel brightness temperature field through registration; performing convex optimization based on a weight function of each channel to obtain guide information; inputting the multi-time-phase multi-channel brightness temperature field and the guide information into a nonlinear brightness temperature reconstruction model to generate a target pressure layer or virtual layer brightness temperature field, and determining corresponding pixels through preprocessing and displacement tracking; combining a pixel-by-pixel scanning time difference to calculate candidate wind vectors, calculating internal quality indexes according to neighborhood speed and direction consistency, and screening the candidate wind vectors to generate a microwave water vapor atmospheric motion vector product. The application improves the high-precision accuracy of the wind vector, improves the vertical resolution of the product, improves the stability of displacement estimation and optical flow tracking, and improves the independence of the product.

Description

Technical Field

[0001] This application relates to the fields of satellite remote sensing data processing, atmospheric motion vector inversion, and microwave water vapor detection technology, and in particular to a microwave water vapor atmospheric motion vector inversion method and system based on convex optimization. Background Technology

[0002] Atmospheric motion vectors are important satellite wind field products describing horizontal atmospheric motion. They can be used for weather system monitoring, numerical weather prediction data assimilation, and process analysis of typhoons, cyclones, cold air outbreaks, and high-latitude weather systems. Existing visible and infrared atmospheric motion vectors typically rely on traceable structures in cloud or water vapor images. However, their observational capabilities and altitude assignment accuracy may be limited under conditions such as cloud cover, limited illumination, uncertain cloud top height, and insufficient vertical representativeness.

[0003] Brightness temperature in the microwave water vapor absorption band can provide observational information sensitive to the atmospheric water vapor layer, offering a supplementary data source for wind field inversion under cloud influence conditions. However, several technical problems still exist in microwave water vapor brightness temperature-guided wind: First, the weighting functions of different water vapor channels have a wide vertical response range, and the response levels between adjacent channels overlap. Assigning height based solely on a single channel or a fixed channel to represent the layer makes it difficult to accurately characterize the target pressure layer, limiting the accuracy of wind vector height determination and vertical resolution. Second, the motion information in the microwave water vapor brightness temperature field usually exhibits a continuously changing brightness temperature structure. The original brightness temperature field contains local noise, missing pixels, and insufficient contrast, which reduces the traceability of motion characteristics and the stability of displacement estimation when directly used for displacement tracking. Third, existing wind vector screening processes often require constraints from external reanalysis wind fields or numerical weather prediction background fields, making product quality control dependent on external background information and affecting the independence and cross-platform applicability of microwave water vapor atmospheric motion vector products. Previous studies have reported the feasibility of using the brightness temperature of a 183.31 GHz microwave humidity probe for atmospheric motion vector inversion. Examples include the paper "Arctic Winds Retrieved from FY-3D Microwave Humidity Sounder-II 183.31 GHz Brightness Temperature Using Atmospheric Motion Vector Method" by Li et al., and related research on atmospheric motion vector inversion from polar-orbiting satellite microwave water vapor observations. These studies primarily disclose the technical approach to wind vector inversion based on the 183.31 GHz microwave water vapor brightness temperature, including observation revisit region construction, geometric registration, and conventional wind guidance verification.

[0004] Furthermore, existing patents involve satellite cloud image-based wind guidance, wind vector field measurement, or water vapor-based wind guidance inversion. For example, Chinese patent application CN103198447A discloses "A Real-time Measurement Method for Wind Vector Field Based on Satellite Cloud Images," and Chinese patent application CN111881538A discloses "An Inversion Method for Water Vapor-Based Wind Guidance." These prior art techniques demonstrate that methods such as image sequence-based wind guidance, optical flow, or wind vector calculation can be used for relevant meteorological image processing. However, these prior art techniques typically do not address the overlapping vertical weighting functions of brightness temperature in the microwave water vapor absorption band, nor do they combine the physical constraints obtained from the convex optimization of the target pressure layer response with the nonlinear brightness temperature reconstruction model. They also do not incorporate preprocessing of the target pressure layer or virtual layer brightness temperature, pixel-by-pixel scan time difference wind vector conversion, and independent internal consistency quality control of the background wind field as complementary technical steps.

[0005] Therefore, there is an urgent need for an atmospheric motion vector inversion method for microwave water vapor brightness temperature sequences at multiple or continuous times. This method would improve the accuracy of target pressure layer height determination and vertical resolution through convex optimization of the weighting function, enhance trackable motion characteristics and improve displacement tracking ability through brightness temperature preprocessing, and reduce dependence on external background field information through internal consistency quality control. This would result in microwave water vapor atmospheric motion vector products with clear vertical representativeness, stable tracking ability, and independent quality identification. Summary of the Invention

[0006] The purpose of this application is to overcome the shortcomings of not being able to improve the accuracy of target pressure layer elevation and vertical resolution through convex optimization of weighting functions, enhance trackable motion characteristics and improve displacement tracking ability through brightness temperature preprocessing, and reduce dependence on external background field information through internal consistency quality control, thereby obtaining microwave water vapor atmospheric motion vector products with clear vertical representativeness, stable tracking ability and independent quality identification. Therefore, this application provides a microwave water vapor atmospheric motion vector inversion method and system based on convex optimization.

[0007] To address the aforementioned technical problems, this application provides a microwave water vapor atmospheric motion vector inversion method based on convex optimization, comprising: Microwave water vapor observation data at at least two observation times are acquired, and the multi-channel brightness temperature data in each microwave water vapor observation data are registered to obtain the registered multi-time multi-channel brightness temperature field. Based on the weighting function corresponding to each microwave water vapor observation channel, the response of the target pressure layer is optimized by convex optimization to obtain the target pressure layer guidance information. The multi-time multi-channel brightness temperature field and the target pressure layer guidance information are input into the nonlinear brightness temperature reconstruction model to obtain the target pressure layer or virtual layer brightness temperature field corresponding to each observation time. The brightness temperature field of the target pressure layer or virtual layer corresponding to each observation time is preprocessed to obtain the brightness temperature input field for displacement estimation; Displacement tracking is performed based on the brightness temperature input field at adjacent observation times to obtain the horizontal displacement field between adjacent observation times, and the corresponding pixels between adjacent observation times are determined based on the horizontal displacement field. Candidate wind vectors are calculated based on the spatial coordinate difference of the corresponding pixels and the pixel-by-pixel scanning time difference. Based on the velocity and direction differences between each candidate wind vector and other candidate wind vectors in its neighborhood, the internal quality index of each candidate wind vector is calculated. The internal quality index is compared with a preset threshold, and candidate wind vectors with internal quality indices less than the preset threshold are eliminated. Microwave water vapor atmospheric motion vector products are generated based on the retained candidate wind vectors.

[0008] According to the microwave water vapor atmospheric motion vector inversion method based on convex optimization proposed in this invention, the microwave water vapor observation data further includes: positioning data and pixel-by-pixel scan time data corresponding to the multi-channel brightness temperature data.

[0009] According to the microwave water vapor atmospheric motion vector inversion method based on convex optimization proposed in this invention, the calculation of candidate wind vectors based on the spatial coordinate difference of the corresponding pixels and the pixel-by-pixel scanning time difference includes: Obtain the pixel at the previous observation time and the projected coordinates of the corresponding pixel at the next observation time, as well as the pixel-by-pixel scan time. The projection coordinate difference is calculated based on the projection coordinates of the corresponding pixel, and the pixel-by-pixel scan time difference is calculated based on the pixel-by-pixel scan time of the corresponding pixel. The velocity component under the projected coordinates is calculated based on the ratio of the difference between the projected coordinates and the difference between the pixel-by-pixel scanning time. Based on the conversion relationship between the projected coordinates and the geographic coordinates, the zonal wind component, the meridional wind component, the wind speed, and the wind direction are obtained.

[0010] According to the microwave water vapor atmospheric motion vector inversion method based on convex optimization proposed in this invention, the target pressure layer guidance information includes: The pressure of the target pressure layer, the parallax height of the target pressure layer, and the support weight of each microwave water vapor observation channel for the target pressure layer.

[0011] According to the microwave water vapor atmospheric motion vector inversion method based on convex optimization proposed in this invention, the nonlinear brightness temperature reconstruction model is a gradient boosting decision tree regression model, which is trained in the following manner: The brightness temperature data of one of the microwave water vapor observation channels is used as the supervision target, and the brightness temperature data of the remaining microwave water vapor observation channels, channel pressure, pressure and parallax height of the target pressure layer, target channel or virtual layer identifier, and channel support weight are used as input features to construct a training sample for reconstructing one channel. The nonlinear brightness temperature reconstruction model is trained using the training samples with one channel left over.

[0012] According to the microwave water vapor atmospheric motion vector inversion method based on convex optimization proposed in this invention, the convex optimization of the response of the target pressure layer includes: A composite weighting function is constructed based on the weighting function and combination coefficients corresponding to each microwave water vapor observation channel. Its expression is: ; in, For the first Weighting functions for each microwave water vapor observation channel, For the corresponding combination coefficients, The number of microwave water vapor observation channels. For the target pressure layer; The optimized combination coefficient vectors are obtained by solving the convex optimization problem function. : ; in, , , The peak pressure of the synthetic weighting function, To smooth out the regularization coefficients, For norm, For pressure The gradient.

[0013] According to the microwave water vapor atmospheric motion vector inversion method based on convex optimization proposed in this invention, the convex optimization of the response of the target pressure layer further includes: After solving for the optimized combination coefficients, the target pressure layer response of the composite weighting function is calculated respectively. Non-target pressure layer response ; The target pressure layer response The expression is: ; in, For the neighborhood of the target pressure layer, The composite weight function is the result of convex optimization. The non-target pressure layer response The expression is: ; When the target pressure layer response quantity The response amount of the non-target pressure layer is greater than or equal to the preset target response threshold. When the target pressure layer is less than or equal to the preset non-target response threshold, the target pressure layer guidance information is constructed based on the pressure and parallax height of the target pressure layer and the channel support weight. Otherwise, adjust the pressure neighborhood of the target pressure layer, the microwave water vapor observation channel involved in constructing the synthetic weight function, or the smoothing regularization coefficient, and resolve each of the combined coefficients.

[0014] According to the present invention, a microwave water vapor atmospheric motion vector inversion method based on convex optimization is proposed, wherein the internal quality index The expression is: ; in, For spatially consistent mass components, For directional consistency mass components, and As weight; The spatial consistency mass component The expression is: ; in, The average velocity difference between the candidate wind vector and other candidate wind vectors in the neighborhood. For the first Wind speed of each candidate wind vector; The average speed difference The expression is: ; in, Indicates the neighborhood. and They represent the first The and the first One candidate wind vector, Indicates the number of candidate wind vectors in the neighborhood; The directional consistency mass component The expression is: ; in, For the first Normalization threshold for the directional difference of candidate wind vectors The average directional difference between the candidate wind vector and other candidate wind vectors in the neighborhood; The average directional difference The expression is: ; in, For the first The and the first The difference in the direction angles of the candidate wind vectors, and .

[0015] According to the microwave water vapor atmospheric motion vector inversion method based on convex optimization proposed in this invention, the preprocessing of the target pressure layer or virtual layer brightness and temperature field corresponding to each observation time includes: An effective mask is generated based on the effective physical range of brightness temperature and the effective identifier of the channel. The brightness temperature field of the target pressure layer or virtual layer within the effective mask is normalized. Median filtering and Gaussian smoothing are performed sequentially on the normalized target pressure layer or virtual layer brightness temperature field. The brightness temperature input field is obtained by performing local contrast enhancement with a cutoff threshold on the brightness temperature field of the Gaussian smoothed target pressure layer or virtual layer.

[0016] This invention also proposes a microwave water vapor atmospheric motion vector inversion system based on convex optimization, used to implement any of the above-described methods, comprising: The data acquisition module is used to acquire microwave water vapor observation data at at least two observation times; The registration module registers the multi-channel brightness temperature data in each microwave water vapor observation data to obtain the registered multi-time multi-channel brightness temperature field. The vertical characterization and target pressure layer brightness temperature reconstruction module is used to perform convex optimization on the response of the target pressure layer based on the weight function corresponding to each microwave water vapor observation channel to obtain the target pressure layer guidance information; the multi-time multi-channel brightness temperature field and the target pressure layer guidance information are input into the nonlinear brightness temperature reconstruction model to obtain the target pressure layer or virtual layer brightness temperature field corresponding to each observation time. The brightness temperature preprocessing module is used to preprocess the brightness temperature field of the target pressure layer or virtual layer corresponding to each observation time to obtain the brightness temperature input field for displacement estimation. The displacement estimation and wind vector conversion module is used to perform displacement tracking based on the brightness temperature input field at adjacent observation times, obtain the horizontal displacement field between adjacent observation times, and determine the corresponding pixels between adjacent observation times based on the horizontal displacement field; calculate candidate wind vectors based on the spatial coordinate difference of the corresponding pixels and the pixel-by-pixel scan time difference; and An internal consistency quality control module is used to calculate the internal quality index of each candidate wind vector based on the velocity and direction differences between each candidate wind vector and other candidate wind vectors in its neighborhood, compare the internal quality index with a preset threshold, eliminate candidate wind vectors whose internal quality index is less than the preset threshold, and generate microwave water vapor atmospheric motion vector products based on the retained candidate wind vectors.

[0017] Compared with the prior art, the present invention has at least the following beneficial effects: First, by optimizing the weighting function convexly, the main response of the synthesized weighting function is concentrated near the target pressure layer, thereby improving the height determination accuracy of the microwave water vapor wind vector. Second, the target pressure layer or virtual layer brightness temperature field is generated by the nonlinear brightness temperature reconstruction model guided by convex optimization, which increases the invertible vertical layers and improves the vertical resolution of microwave water vapor atmospheric motion vector products. Third, brightness temperature preprocessing enhances the brightness temperature motion characteristics that can be continuously tracked over time, thereby improving the stability of displacement estimation and optical flow tracking. Fourth, wind vector conversion is performed by using the pixel-by-pixel scanning time difference to improve the rationality of wind speed calculation under asynchronous scanning observation conditions; Fifth, reduce reliance on external reanalysis wind fields or numerical forecast background fields through internal consistency quality control, thereby improving the independence and cross-platform applicability of wind vector products. Attached Figure Description

[0018] Figure 1 The diagram shown is a schematic diagram of the microwave water vapor atmospheric motion vector inversion process provided in an embodiment of the present invention; Figure 2 The diagram shown is a schematic diagram of the revisit or continuous observation coverage area and registration provided in an embodiment of the present invention; Figure 3 The figure shown is a schematic diagram of virtual layer brightness temperature reconstruction guided by convex optimization of weight function according to an embodiment of the present invention; Figure 4 The diagram shown is a schematic diagram of the brightness temperature preprocessing link provided in an embodiment of the present invention; Figure 5 The diagram shown is a schematic diagram of optical flow tracing and wind vector conversion provided in an embodiment of the present invention; Figure 6 The diagram shown is a schematic diagram of quality control and product output provided by an embodiment of the present invention; Figure 7(a) shows a schematic diagram of the original physical channel brightness temperature and its wind vector inversion results provided in the embodiment of the present invention; Figure 7(b) shows a schematic diagram of the target response brightness temperature and wind vector inversion results at 450 hPa provided by the embodiment of the present invention; Figure 7(c) shows a schematic diagram of the target response brightness temperature and wind vector inversion results at 500 hPa provided by an embodiment of the present invention. Detailed Implementation

[0019] The present invention will be further described below with reference to the accompanying drawings and embodiments. It should be understood that the described embodiments are only illustrative of some implementations of the present invention and are not intended to limit the scope of protection of the present invention. Where there is no conflict, the technical features in the following embodiments can be combined with each other.

[0020] Example 1 The purpose of this invention is to provide a microwave water vapor atmospheric motion vector inversion method and system based on weighted function convex optimization-guided virtual layer brightness temperature reconstruction, target pressure layer brightness temperature preprocessing, and internal consistency quality control. This method aims to improve the height accuracy, vertical resolution, displacement tracking capability, and quality control independence of microwave water vapor brightness temperature-guided wind. To achieve the above objective, this invention provides a microwave water vapor atmospheric motion vector inversion method and system based on convex optimization. The method includes: acquiring microwave water vapor observation data at at least two observation times or continuous observation periods. In this embodiment, the microwave water vapor observation data is observation data in the microwave water vapor absorption frequency band, including brightness temperature, location, and time data, and performing registration to form a registered multi-time multi-channel brightness temperature field; performing convex optimization of the target pressure layer response based on the weight function of multiple microwave water vapor channels to determine the target pressure layer pressure, the target pressure layer parallax height, and the support weight of each real water vapor channel for the target pressure layer; Using the brightness temperature of the real water vapor channel, the pressure of the target pressure layer, the parallax height of the target pressure layer, and the channel support weight as input features, a nonlinear brightness temperature reconstruction model trained with leave-one-out reconstruction samples from the real water vapor channel is used to generate the brightness temperature field of the target pressure layer or virtual layer. The brightness temperature field of the target pressure layer or virtual layer is then subjected to effective value screening, normalization, filtering, and local contrast enhancement to obtain the brightness temperature input field for displacement estimation. Displacement tracking is performed based on the brightness temperature input field at consecutive observation times to obtain the horizontal displacement field between adjacent observation times, and the corresponding pixels at consecutive observation times are determined based on the horizontal displacement field. The horizontal wind vector is calculated based on the spatial coordinate difference of the corresponding pixels and the pixel-by-pixel scanning time difference. Internal quality indicators are constructed based on neighborhood velocity consistency and direction consistency to perform internal consistency quality control on the horizontal wind vector, outputting a microwave water vapor atmospheric motion vector product.

[0021] In the convex optimization of the target pressure layer response: Let the first... The weighting function for each water vapor channel is: The channel brightness temperature is The combination coefficient is The target pressure layer is The combined weight function is then... And a linear reference brightness temperature can be obtained. ,in, , This represents the vector of coefficients for each combination. By solving the convex optimization problem and satisfy To obtain the combination coefficients ,in The central or peak pressure of the composite weighting function. To smooth the regularization coefficients, the convex optimization results are used to determine the target pressure layer response, the target pressure layer parallax height, and the channel support weight, so that the main response of the synthetic weight function is concentrated near the target pressure layer, thereby reducing the layer deviation caused by the single channel peak height determination and improving the height determination accuracy of the microwave water vapor wind vector.

[0022] Furthermore, in solving the combination coefficients Then, the response of the target pressure layer is calculated. Non-target pressure layer response ,in For the target pressure layer neighborhood; when Not less than the preset target response threshold and If the target pressure layer, the target pressure layer, the parallax height of the target pressure layer, and the channel support weight obtained by normalizing the absolute value of the combination coefficient are used as the target pressure layer guidance information; otherwise, the target pressure layer neighborhood, the water vapor channels involved in the combination, or the smoothing regularization coefficient are adjusted and the solution is recalculated. The nonlinear brightness temperature reconstruction model is a gradient boosting decision tree regression model. During the training phase, the nonlinear brightness temperature reconstruction model is trained using leave-one-out training samples constructed based on real microwave water vapor channels, enabling the nonlinear brightness temperature reconstruction model to learn the nonlinear mapping relationship between the brightness temperatures of different channels. Leave-one training sample: Each time, use the brightness temperature of a real water vapor channel as the supervision target, and use the brightness temperature of the other real water vapor channels, channel pressure, target pressure, parallax height of the target pressure layer, target channel or virtual layer identifier, and channel support weight as input features; During the inference phase, the brightness temperature of the real water vapor channels at multiple time points and the guiding information of the target pressure layer are input into the nonlinear brightness temperature reconstruction model, which outputs the brightness temperature field of the target pressure layer or virtual layer. This allows the construction of the target pressure layer or virtual water vapor layer brightness temperature input between real physical water vapor channels, increasing the invertible vertical hierarchy and improving the vertical resolution of microwave water vapor atmospheric motion vector products.

[0023] Furthermore, the brightness temperature preprocessing includes: generating an effective mask based on the effective physical range of brightness temperature and the effective channel identifier; and processing the brightness temperature field within the effective mask. Normalization is performed; median filtering is then applied to the normalized brightness temperature field. ; in, For spatial grid points or target cell locations , For the observation time, For the cell location within the neighborhood, To take the median of the effective brightness temperature values ​​within the local neighborhood window, For A local neighborhood window centered on the subject; Perform Gaussian smoothing ,in For a two-dimensional Gaussian kernel; then for Execute with truncation threshold Local contrast enhancement, resulting in a bright temperature input field The brightness temperature preprocessing is used to suppress invalid pixels, isolated noise and local anomalous disturbances, and enhance the motion characteristics that can be continuously tracked over time in the brightness temperature field, thereby improving the stability of subsequent displacement estimation and optical flow tracing.

[0024] Furthermore, the displacement tracking and horizontal wind vector calculation based on the brightness temperature input field includes: analyzing the brightness temperature input field at the previous observation time. Brightness temperature input field at the next observation time Establish the correspondence between brightness temperature and structure to satisfy ; in, Indicates the same brightness temperature structure from arrive In the unified analysis of displacement components in the horizontal and vertical directions of the grid, the unit is usually grid cell, which can be a non-integer. Dense displacement estimation is used to obtain the displacement components of each effective grid point on the uniform analysis grid. and And determine the pixel of the previous moment. Pixel corresponding to the next time step ; Read pixel-by-pixel scan time and projected coordinates , Read pixel-by-pixel scan time and projected coordinates , ; Calculate the pixel-by-pixel scan time difference Spatial displacement , And calculate the velocity components in the projected coordinates. , Then, based on the conversion relationship between projected coordinates and geographic coordinates, the zonal wind component, meridional wind component, wind speed, and wind direction are obtained.

[0025] Furthermore, the internal quality index includes a spatial consistency quality component and a directional consistency quality component; for the first... Candidate wind vectors ,in, For this is the first The zonal wind component of each candidate wind vector, positive for eastward, is expressed in m / s. For this is the first The meridional wind component of each candidate wind vector, with positive for north, is expressed in m / s. In its neighborhood Internal computation , ; in, , For the neighborhood The number of candidate wind vectors within; Computational spatial consistency mass component and direction-consistent mass components And calculate internal quality indicators. ,in, For the first The wind speed of each candidate wind vector. The normalized threshold for the direction difference. and As weight; when If the wind vector is less than a preset threshold, the corresponding wind vector is discarded; otherwise, it is written into the product. The internal quality index is based on the neighborhood velocity consistency and direction consistency among candidate wind vectors for quality control, and does not use the external reanalysis wind field or numerical forecast background field as a necessary screening condition, thereby reducing the dependence of the microwave water vapor atmospheric motion vector product generation process on background field information.

[0026] This invention also provides an atmospheric motion vector inversion system based on microwave water vapor brightness temperature sequences. The system is structured with functional modules based on the aforementioned method flow. These modules include: a data acquisition module for acquiring brightness temperature data, positioning information, and pixel-by-pixel scan time information in the microwave water vapor absorption band; a registration module for generating a unified analysis grid and forming a multi-time, multi-channel brightness temperature field; a vertical characterization and target pressure layer brightness temperature reconstruction module for solving a convex optimization problem of the weighting function, determining the target pressure layer guidance information, and generating the target pressure layer or virtual layer brightness temperature field through a nonlinear brightness temperature reconstruction model; a brightness temperature preprocessing module for performing effective value screening, normalization, filtering, and local contrast enhancement; a displacement estimation and wind vector conversion module for estimating the horizontal displacement field and calculating the wind vector based on the pixel-by-pixel scan time difference; and an internal consistency quality control module for generating quality indicators based on neighborhood velocity consistency and direction consistency and outputting microwave water vapor atmospheric motion vector products.

[0027] Example 2 This embodiment is an atmospheric motion vector inversion based on the 183GHz water vapor brightness temperature sequence of the FY-3E / F binary satellites.

[0028] This embodiment uses the brightness temperature observation of the 183GHz water vapor absorption band by two low-Earth orbit satellites, FY-3E and FY-3F, as an example to illustrate a specific implementation of the present invention. Unlike existing studies on 183GHz microwave water vapor wind guidance based on continuous multi-orbit revisit observations by a single low-Earth orbit satellite, this embodiment uses adjacent time observation data obtained by the FY-3E and FY-3F satellites in their respective orbits to construct a dual-satellite revisit brightness temperature sequence. The channels may include at least two of the following: 183.31±1GHz, 183.31±1.8GHz, 183.31±3GHz, 183.31±4.5GHz, and 183.31±7GHz. The specific center frequency, bandwidth, and number of channels are determined by the actual microwave humidity detector configuration. This embodiment only uses the 183GHz water vapor absorption band as an example and does not limit the claims to this specific frequency or satellite platform.

[0029] In this embodiment, the revisit observation matching, common coverage area determination, positioning registration, and necessary geometric preprocessing can be completed using existing satellite brightness temperature registration procedures. Their purpose is to provide spatiotemporally consistent multi-moment brightness temperature inputs for subsequent inversion. This embodiment differs from existing 183GHz microwave water vapor wind guidance research in that it performs convex optimization of the target pressure layer response based on multiple water vapor channel weight functions. The convex optimization results are used as physical guidance information such as the target pressure layer, parallax height, and channel support weights. Combined with a nonlinear brightness temperature reconstruction model obtained from leave-one-out training of real water vapor channels, a brightness temperature field for the target pressure layer or virtual layer is generated. Effective value screening, normalization, filtering, and local contrast enhancement are performed on the target pressure layer or virtual layer brightness temperature field to form a brightness temperature input field for displacement estimation. Furthermore, an internal quality index independent of the external background wind field is constructed based on neighborhood velocity consistency and direction consistency.

[0030] like Figure 1 The diagram shown illustrates the microwave water vapor atmospheric motion vector inversion process. Figure 2 The diagram illustrates the process of microwave water vapor atmospheric motion vector inversion. This embodiment reads first-order brightness temperature data obtained from adjacent observation times of FY-3E and FY-3F. Unlike existing studies on 183GHz microwave water vapor wind guidance based on continuous multi-orbit revisits of a single low-orbit satellite, this embodiment utilizes adjacent observation times formed by the two satellites in their respective orbits to construct a dual-satellite revisit brightness temperature sequence. The read data includes the latitude and longitude positioning, scan time, scan line number, pixel sequence number, channel identifier, and valid data identifier for each pixel. The scan time uses the actual scan time of each pixel, rather than the nominal transit time of the entire orbit or the entire image. A common coverage area is determined based on the positioning data and valid brightness temperature range of the corresponding orbits of the two satellites, and a unified analysis grid is established within this common coverage area. Subsequently, brightness temperature data, scan time fields, and positioning information from different observation times and different 183GHz water vapor channels are mapped to this unified analysis grid, forming a registered multi-time, multi-channel brightness temperature field.

[0031] like Figure 3 As shown, this embodiment takes multiple real water vapor channels near the 183 GHz water vapor absorption band as examples. First, the weighting function corresponding to each channel is calculated or read, and the atmospheric layer near 500 hPa is taken as the target pressure layer. For adjacent water vapor channels such as 183.31±1.8 GHz and 183.31±3 GHz, a synthetic weighting function is first constructed based on the weighting function, and the support relationship of each real channel to the 500 hPa target pressure layer is determined through target pressure layer response convex optimization. The convex optimization result is not directly used as the linear synthesis result of the final virtual layer brightness temperature, but is used to provide physical guidance information such as the target pressure layer, the parallax height of the target pressure layer, and the channel support weight; the linear convex optimization brightness temperature in the cache can be retained as a diagnostic comparison result.

[0032] In the brightness-temperature reconstruction model training phase, this embodiment utilizes real water vapor channels to construct leave-one-out reconstruction samples. Specifically, each time, the brightness temperature of one real water vapor channel is hidden as a supervision target, and the brightness temperatures of the remaining real water vapor channels, channel pressure, target pressure, disparity height of the target pressure layer, target channel identifier, and channel support weights are used as input features to train the nonlinear brightness-temperature reconstruction model. The model can employ a gradient boosting tree regression model, or other regression models capable of characterizing the nonlinear relationship between multi-channel brightness temperatures.

[0033] During the virtual layer inference stage, the actual water vapor channel brightness temperatures at times before and after FY-3E and FY-3F, along with the target pressure layer pressure at 500 hPa, disparity height, and channel support weights obtained through convex optimization of the weighting function, are input into the trained brightness temperature reconstruction model to generate virtual water vapor layer brightness temperature fields near 500 hPa at the time points before and after FY-3E. Subsequent brightness temperature preprocessing and optical flow wind guidance both use this nonlinearly reconstructed virtual layer brightness temperature field as one of the inputs; the linearly convex optimized virtual brightness temperature is only used for diagnostic comparison or model checking.

[0034] like Figure 4 As shown, the brightness temperature field of the physical channel representative layer or the virtual water vapor layer brightness temperature field near 500 hPa generated by the nonlinear brightness temperature reconstruction model is preprocessed. Specifically, an effective mask is first generated based on the effective brightness temperature range and effective data identifiers to remove invalid pixels, and masking or neighborhood interpolation is performed on the missing measurement areas; then, the brightness temperature values ​​are normalized to a uniform grayscale range; next, a 3×3 median filter is used to suppress isolated anomalous pixels, and Gaussian smoothing is used to reduce local noise; finally, local contrast enhancement is performed to obtain the brightness temperature input field for optical flow displacement estimation. The above preprocessing parameters can be adjusted according to the sensor footprint size, analysis grid spacing, and target wind field scale.

[0035] like Figure 5 As shown, the preprocessed brightness temperature fields from the previous and next time moments are input into the dense optical flow algorithm to calculate the horizontal displacement component of each effective grid point. and The displacement of the optical flow output is used to determine the grid points at the previous time step. The corresponding position in the next moment .when and When the integer value is not an integer, the pixel-by-pixel scan time and projected coordinates of the corresponding position at the next time step can be obtained through bilinear interpolation. Then, the data is read... The pixel-by-pixel scan time and projection coordinates, and The pixel-by-pixel scan time and projected coordinates are used to calculate the pixel-by-pixel scan time difference and the corresponding spatial displacement. This step uses a pixel-level scan time field, rather than the uniform transit time of the entire image, thus more accurately reflecting the differences in scan geometry and observation time sequence. After obtaining the spatial displacement and pixel-by-pixel scan time difference of the corresponding pixel, the velocity component under the projected coordinates is calculated, and the zonal wind component, meridional wind component, wind speed, and wind direction are obtained according to the conversion relationship between projected coordinates and geographic coordinates used in this embodiment. If the pixel-by-pixel scan time difference is less than a preset lower limit, the corresponding pixel falls outside the effective range, the displacement amplitude exceeds the preset range, or the projected coordinates are unavailable, the corresponding candidate wind vector is marked as invalid.

[0036] like Figure 6 As shown, internal consistency quality control is performed on candidate wind vectors. This embodiment does not use external background wind fields such as ERA5 as necessary screening conditions. Instead, a neighborhood window is established around the candidate wind vector to compare the velocity and direction differences between the wind vector and its neighbors. First, candidate wind vectors with insufficient brightness temperature ratio, abnormal per-pixel scan time differences, excessive displacement amplitude, or corresponding pixels exceeding limits are eliminated. For the remaining candidate wind vectors, neighborhood velocity consistency index and neighborhood direction consistency index are calculated respectively, and a comprehensive quality index is obtained accordingly. When the comprehensive quality index is less than a preset threshold, the corresponding wind vector is eliminated; when the comprehensive quality index meets the threshold, it is organized into microwave water vapor atmospheric motion vector products. Each product record includes at least one of the following: longitude, latitude, projected coordinates, corresponding pixel scan time before and after, per-pixel scan time difference, channel identifier, identifier representing a pressure layer or virtual layer, zonal wind component, meridional wind component, wind speed, wind direction, quality index, and valid data identifier.

[0037] As shown in Figures 7(a), 7(b), and 7(c), after brightness temperature registration, target pressure layer or virtual layer brightness temperature reconstruction, brightness temperature preprocessing, optical flow tracing, pixel-by-pixel scan time difference wind vector conversion, and internal consistency quality control, the brightness temperature target response construction result and the corresponding atmospheric motion vector product can be generated. Figure 7(a) shows the original physical channel brightness temperature and its wind vector result; Figure 7(b) shows the brightness temperature and its wind vector result at a target response of 450 hPa; and Figure 7(c) shows the brightness temperature and its wind vector result at a target response of 500 hPa. The arrows indicate the direction and relative magnitude of the wind vector after quality control, and the gray background represents the corresponding brightness temperature structure.

[0038] Example 3 A microwave water vapor atmospheric motion vector inversion method based on convex optimization includes: Acquire microwave water vapor observation data at at least two observation times, register the multi-channel brightness temperature data in each microwave water vapor observation data, and obtain the registered multi-time multi-channel brightness temperature field; Based on the weighting function corresponding to each microwave water vapor observation channel, the response of the target pressure layer is optimized by convex optimization to obtain the target pressure layer guidance information. The multi-time, multi-channel brightness temperature field and target pressure layer guidance information are input into the nonlinear brightness temperature reconstruction model to obtain the target pressure layer or virtual layer brightness temperature field corresponding to each observation time. The brightness temperature field of the target pressure layer or virtual layer corresponding to each observation time is preprocessed to obtain the brightness temperature input field for displacement estimation; Displacement tracking is performed based on the brightness temperature input field at adjacent observation times to obtain the horizontal displacement field between adjacent observation times, and the corresponding pixels between adjacent observation times are determined based on the horizontal displacement field. Candidate wind vectors are calculated based on the spatial coordinate difference of corresponding pixels and the pixel-by-pixel scanning time difference. Based on the velocity and direction differences between each candidate wind vector and other candidate wind vectors in its neighborhood, the internal quality index of each candidate wind vector is calculated. The internal quality index is compared with a preset threshold, and candidate wind vectors with internal quality indices less than the preset threshold are eliminated. Microwave water vapor atmospheric motion vector products are generated based on the retained candidate wind vectors.

[0039] According to the microwave water vapor atmospheric motion vector inversion method based on convex optimization proposed in this invention, the microwave water vapor observation data also includes: positioning data and pixel-by-pixel scan time data corresponding to multi-channel brightness temperature data.

[0040] According to the present invention, a microwave water vapor atmospheric motion vector inversion method based on convex optimization is proposed, which calculates candidate wind vectors based on the spatial coordinate difference of corresponding pixels and the pixel-by-pixel scanning time difference, including: Obtain the pixel at the previous observation time and the projected coordinates of the corresponding pixel at the next observation time, as well as the pixel-by-pixel scan time. Calculate the projection coordinate difference based on the projection coordinates of the corresponding pixel, and calculate the pixel-by-pixel scan time difference based on the pixel-by-pixel scan time of the corresponding pixel. The velocity component under the projected coordinates is calculated based on the ratio of the difference between the projected coordinates and the difference in the pixel-by-pixel scanning time. Based on the conversion relationship between the projected coordinates and the geographic coordinates, the zonal wind component, the meridional wind component, the wind speed, and the wind direction are obtained.

[0041] According to the microwave water vapor atmospheric motion vector inversion method based on convex optimization proposed in this invention, the target pressure layer guidance information includes: The pressure and parallax height of the target pressure layer, as well as the support weight of each microwave water vapor observation channel to the target pressure layer.

[0042] According to the microwave water vapor atmospheric motion vector inversion method based on convex optimization proposed in this invention, the nonlinear brightness temperature reconstruction model is trained in the following manner: The brightness temperature data of one microwave water vapor observation channel is used as the supervision target, and the brightness temperature data of the other microwave water vapor observation channels, channel pressure, pressure and parallax height of the target pressure layer, target channel or virtual layer identifier, and channel support weight are used as input features to construct a training sample for reconstructing one channel. A nonlinear brightness-temperature reconstruction model was trained using training samples with one channel left open.

[0043] According to the present invention, a microwave water vapor atmospheric motion vector inversion method based on convex optimization is proposed, which performs convex optimization on the response of the target pressure layer, including: A composite weighting function is constructed based on the weighting function and combination coefficients corresponding to each microwave water vapor observation channel. Its expression is: ; in, For the first Weighting functions for each microwave water vapor observation channel, For the corresponding combination coefficients, The number of microwave water vapor observation channels. For the target pressure layer; The optimized combination coefficient vectors are obtained by solving the convex optimization problem function. : ; in, , , The peak pressure of the synthesized weighting function, To smooth out the regularization coefficients, For norm, For pressure The gradient.

[0044] According to the microwave water vapor atmospheric motion vector inversion method based on convex optimization proposed in this invention, the response of the target pressure layer is convexly optimized, and the method further includes: After solving for the optimized combination coefficients, the target pressure layer response of the composite weighting function is calculated respectively. Non-target pressure layer response ; Target pressure layer response The expression is: ; in, For the neighborhood of the target pressure layer, The composite weight function is the result of convex optimization. Non-target pressure layer response The expression is: ; When the target pressure layer response Response quantity greater than or equal to the preset target response threshold, and not the response quantity of the target pressure layer. When the target pressure layer is less than or equal to the preset non-target response threshold, target pressure layer guidance information is constructed based on the pressure and parallax height of the target pressure layer and the channel support weight. Otherwise, adjust the pressure neighborhood of the target pressure layer, the microwave water vapor observation channel involved in constructing the synthetic weight function, or the smoothing regularization coefficient, and resolve each combination coefficient.

[0045] This invention proposes a microwave water vapor atmospheric motion vector inversion method based on convex optimization, with internal quality indicators... The expression is: ; in, For spatially consistent mass components, For directional consistency mass components, and As weight; Spatial consistency mass component The expression is: ; in, The average velocity difference between a candidate wind vector and other candidate wind vectors in its neighborhood. For the first Wind speed of each candidate wind vector; Average speed difference The expression is: ; in, Indicates the neighborhood. and They represent the first The and the first One candidate wind vector, Indicates the number of candidate wind vectors in the neighborhood; Orientational consistency mass component The expression is: ; in, For the first Normalization threshold for the directional difference of candidate wind vectors This represents the average directional difference between a candidate wind vector and other candidate wind vectors in its neighborhood. Average directional difference The expression is: ; in, For the first The and the first The difference in the direction angles of the candidate wind vectors, and .

[0046] According to the proposed microwave water vapor atmospheric motion vector inversion method based on convex optimization, the brightness and temperature fields of the target pressure layer or virtual layer corresponding to each observation time are preprocessed, including: An effective mask is generated based on the effective physical range of brightness temperature and the effective identifier of the channel. Normalize the brightness temperature field of the target pressure layer or virtual layer within the effective mask; Median filtering and Gaussian smoothing are performed sequentially on the normalized target pressure layer or virtual layer brightness temperature field. The brightness temperature field of the Gaussian smoothed target pressure layer or virtual layer is enhanced with a local contrast enhancement with a cutoff threshold to obtain the brightness temperature input field.

[0047] Example 4 A microwave water vapor atmospheric motion vector inversion system based on convex optimization, used to implement the method described in any of the above embodiments, includes: The data acquisition module is used to acquire microwave water vapor observation data at at least two observation times; The registration module registers the multi-channel brightness temperature data in each microwave water vapor observation data to obtain the registered multi-time multi-channel brightness temperature field; The vertical characterization and target pressure layer brightness temperature reconstruction module is used to perform convex optimization on the response of the target pressure layer based on the weight function corresponding to each microwave water vapor observation channel to obtain the target pressure layer guidance information; the multi-time multi-channel brightness temperature field and the target pressure layer guidance information are input into the nonlinear brightness temperature reconstruction model to obtain the target pressure layer or virtual layer brightness temperature field corresponding to each observation time. The brightness temperature preprocessing module is used to preprocess the brightness temperature field of the target pressure layer or virtual layer corresponding to each observation time to obtain the brightness temperature input field for displacement estimation. The displacement estimation and wind vector conversion module is used to perform displacement tracking based on the brightness temperature input field at adjacent observation times, obtain the horizontal displacement field between adjacent observation times, and determine the corresponding pixels between adjacent observation times based on the horizontal displacement field; and calculate candidate wind vectors based on the spatial coordinate difference of the corresponding pixels and the pixel-by-pixel scanning time difference. The internal consistency quality control module is used to calculate the internal quality index of each candidate wind vector based on the velocity and direction differences between each candidate wind vector and other candidate wind vectors in its neighborhood. The internal quality index is compared with a preset threshold, and candidate wind vectors with internal quality indices less than the preset threshold are eliminated. Microwave water vapor atmospheric motion vector products are generated based on the retained candidate wind vectors.

[0048] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit it. Although this application has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of this application do not depart from the spirit and scope of the technical solutions of this application, and should all be covered within the scope of the claims of this application.

Claims

1. A microwave water vapor atmospheric motion vector inversion method based on convex optimization, comprising: Microwave water vapor observation data at at least two observation times are acquired, and the multi-channel brightness temperature data in each microwave water vapor observation data are registered to obtain the registered multi-time multi-channel brightness temperature field. Based on the weighting function corresponding to each microwave water vapor observation channel, the response of the target pressure layer is optimized by convex optimization to obtain the target pressure layer guidance information. The multi-time multi-channel brightness temperature field and the target pressure layer guidance information are input into the nonlinear brightness temperature reconstruction model to obtain the target pressure layer or virtual layer brightness temperature field corresponding to each observation time. The brightness temperature field of the target pressure layer or virtual layer corresponding to each observation time is preprocessed to obtain the brightness temperature input field for displacement estimation; Displacement tracking is performed based on the brightness temperature input field at adjacent observation times to obtain the horizontal displacement field between adjacent observation times, and the corresponding pixels between adjacent observation times are determined based on the horizontal displacement field. Candidate wind vectors are calculated based on the spatial coordinate difference of the corresponding pixels and the pixel-by-pixel scanning time difference. Based on the velocity and direction differences between each candidate wind vector and other candidate wind vectors in its neighborhood, the internal quality index of each candidate wind vector is calculated. The internal quality index is compared with a preset threshold, and candidate wind vectors with internal quality indices less than the preset threshold are eliminated. Microwave water vapor atmospheric motion vector products are generated based on the retained candidate wind vectors.

2. The microwave water vapor atmospheric motion vector inversion method according to claim 1, characterized in that, The microwave water vapor observation data also includes: positioning data and pixel-by-pixel scan time data corresponding to the multi-channel brightness temperature data.

3. The microwave water vapor atmospheric motion vector inversion method according to claim 2, characterized in that, The calculation of candidate wind vectors based on the spatial coordinate difference of the corresponding pixels and the pixel-by-pixel scan time difference includes: Obtain the pixel at the previous observation time and the projected coordinates of the corresponding pixel at the next observation time, as well as the pixel-by-pixel scan time. The projection coordinate difference is calculated based on the projection coordinates of the corresponding pixel, and the pixel-by-pixel scan time difference is calculated based on the pixel-by-pixel scan time of the corresponding pixel. The velocity component under the projected coordinates is calculated based on the ratio of the difference between the projected coordinates and the difference between the pixel-by-pixel scanning time. Based on the conversion relationship between the projected coordinates and the geographic coordinates, the zonal wind component, the meridional wind component, the wind speed, and the wind direction are obtained.

4. The microwave water vapor atmospheric motion vector inversion method according to claim 1, characterized in that, The target pressure layer guidance information includes: The pressure of the target pressure layer, the parallax height of the target pressure layer, and the support weight of each microwave water vapor observation channel for the target pressure layer.

5. The microwave water vapor atmospheric motion vector inversion method according to claim 4, characterized in that, The nonlinear brightness temperature reconstruction model is a gradient boosting decision tree regression model, which is trained in the following manner: The brightness temperature data of one of the microwave water vapor observation channels is used as the supervision target, and the brightness temperature data of the remaining microwave water vapor observation channels, channel pressure, pressure and parallax height of the target pressure layer, target channel or virtual layer identifier, and channel support weight are used as input features to construct a training sample for reconstructing one channel. The nonlinear brightness temperature reconstruction model is trained using the training samples with one channel left over.

6. The microwave water vapor atmospheric motion vector inversion method according to claim 4, characterized in that, The convex optimization of the response to the target pressure layer includes: A composite weighting function is constructed based on the weighting function and combination coefficients corresponding to each microwave water vapor observation channel. Its expression is: ； in, For the first Weighting functions for each microwave water vapor observation channel, For the corresponding combination coefficients, The number of microwave water vapor observation channels. For the target pressure layer; The optimized combination coefficient vectors are obtained by solving the convex optimization problem function. : ； in, , , The peak pressure of the synthetic weighting function, To smooth out the regularization coefficients, For norm, For pressure The gradient.

7. The microwave water vapor atmospheric motion vector inversion method according to claim 6, characterized in that, The convex optimization of the response to the target pressure layer also includes: After solving for the optimized combination coefficients, the target pressure layer response of the composite weighting function is calculated respectively. Non-target pressure layer response ; The target pressure layer response The expression is: ； in, For the neighborhood of the target pressure layer, The composite weight function is the result of convex optimization. The non-target pressure layer response The expression is: ； When the target pressure layer response quantity The response amount of the non-target pressure layer is greater than or equal to the preset target response threshold. When the target pressure layer is less than or equal to the preset non-target response threshold, the target pressure layer guidance information is constructed based on the pressure and parallax height of the target pressure layer and the channel support weight. Otherwise, adjust the pressure neighborhood of the target pressure layer, the microwave water vapor observation channel involved in constructing the synthetic weight function, or the smoothing regularization coefficient, and resolve each of the combined coefficients.

8. The microwave water vapor atmospheric motion vector inversion method according to claim 1, characterized in that, The internal quality indicators The expression is: ； in, For spatially consistent mass components, For directional consistency mass components, and As weight; The spatial consistency quality component The expression is: ； in, The average velocity difference between the candidate wind vector and other candidate wind vectors in the neighborhood. For the first Wind speed of each candidate wind vector; The average speed difference The expression is: ； in, Indicates the neighborhood. and They represent the first The and the first One candidate wind vector, Indicates the number of candidate wind vectors in the neighborhood; The directional consistency mass component The expression is: ； in, For the first Normalization threshold for the directional difference of candidate wind vectors The average directional difference between the candidate wind vector and other candidate wind vectors in the neighborhood; The average directional difference The expression is: ； in, For the first The and the first The difference in the direction angles of the candidate wind vectors, and .

9. The microwave water vapor atmospheric motion vector inversion method according to claim 1, characterized in that, The preprocessing of the target pressure layer or virtual layer brightness and temperature field corresponding to each observation time includes: An effective mask is generated based on the effective physical range of brightness temperature and the effective identifier of the channel. The brightness temperature field of the target pressure layer or virtual layer within the effective mask is normalized. Median filtering and Gaussian smoothing are performed sequentially on the normalized target pressure layer or virtual layer brightness temperature field. The brightness temperature input field is obtained by performing local contrast enhancement with a cutoff threshold on the brightness temperature field of the Gaussian smoothed target pressure layer or virtual layer.

10. A microwave water vapor atmospheric motion vector inversion system based on convex optimization, used to implement the method described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to acquire microwave water vapor observation data at at least two observation times; The registration module registers the multi-channel brightness temperature data in each microwave water vapor observation data to obtain the registered multi-time multi-channel brightness temperature field. The vertical characterization and target pressure layer brightness temperature reconstruction module is used to perform convex optimization on the response of the target pressure layer based on the weight function corresponding to each microwave water vapor observation channel, so as to obtain the target pressure layer guidance information. The multi-time multi-channel brightness temperature field and the target pressure layer guidance information are input into the nonlinear brightness temperature reconstruction model to obtain the target pressure layer or virtual layer brightness temperature field corresponding to each observation time. The brightness temperature preprocessing module is used to preprocess the brightness temperature field of the target pressure layer or virtual layer corresponding to each observation time to obtain the brightness temperature input field for displacement estimation. The displacement estimation and wind vector conversion module is used to perform displacement tracking based on the brightness temperature input field at adjacent observation times, obtain the horizontal displacement field between adjacent observation times, and determine the corresponding pixels between adjacent observation times based on the horizontal displacement field; calculate candidate wind vectors based on the spatial coordinate difference of the corresponding pixels and the pixel-by-pixel scan time difference; and An internal consistency quality control module is used to calculate the internal quality index of each candidate wind vector based on the velocity and direction differences between each candidate wind vector and other candidate wind vectors in its neighborhood, compare the internal quality index with a preset threshold, eliminate candidate wind vectors whose internal quality index is less than the preset threshold, and generate microwave water vapor atmospheric motion vector products based on the retained candidate wind vectors.

Images