A short-time heavy rain forecast method and system for a near period

Abstract

The application relates to the field of weather prediction, and discloses a short-time heavy rainfall prediction method and system for a near period, which improves the quality of data by collecting and preprocessing multi-source weather data, provides a basis for subsequent extraction of multi-source data features, and redesigns a prediction model for multi-source weather data, realizes deep understanding and extraction of weather features of different source data through processing of a feature fusion module, an evolution module and a generation module, effectively captures the suddenness and locality of short-time heavy rainfall, improves the accuracy and stability of short-time heavy rainfall prediction, and designs a rain cluster precipitation area correction method, avoids the problem that deviation accumulation is prone to occur on a long time sequence due to rain cluster evolution, improves the accuracy and reliability of prediction, and improves the accuracy and reliability of short-time heavy rainfall prediction.

Description

Technical Field

[0001] This invention relates to the field of meteorological forecasting, and in particular to a method and system for forecasting short-term heavy rainfall in the near future. Background Technology

[0002] Short-duration heavy rainfall is a highly destructive type of convective weather, and is a major cause of secondary disasters such as flash floods, urban flooding, and geological disasters. Short-duration heavy rainfall is characterized by its suddenness and locality, and in actual forecasting operations, it requires extremely high levels of professional skills and rapid response capabilities from personnel. Therefore, achieving refined and intelligent forecasting in the near term is of great significance.

[0003] In existing technologies, forecasting methods for short-duration heavy precipitation mainly rely on statistical post-processing of numerical weather prediction models or extrapolation forecasts using radar observations of precipitation echoes and ground meteorological station rainfall data. However, numerical weather prediction models have limitations such as long data assimilation times, inconsistencies between the initial field and dynamic framework during the start-up phase, large errors in early forecast products, and time lags, making them difficult to apply to nowcasting. Traditional extrapolation forecasting methods, such as optical flow and cross-correlation methods, are unable to predict the generation, dissipation, and intensity changes of short-duration heavy precipitation. In contrast, nonlinear extrapolation forecasting of short-duration heavy precipitation based on artificial intelligence technologies, such as those based on convolutional long short-term memory (CNN-LSTM) and predictive recurrent neural networks (PredRNN), offers a solution. Most extrapolation forecasts are based solely on single data sources, such as radar precipitation echoes or automatic weather station rainfall data, resulting in limited improvement in forecast accuracy. Although some scholars have used artificial intelligence technology to integrate ground meteorological physical quantities, satellite, and topographic data for extrapolation, the integration methods often involve simple data overlay, lacking in-depth understanding and refinement of the meteorological characteristics of data from different sources. This leads to poor forecasting of the intensity and evolution of short-term heavy rainfall, making it difficult to effectively capture the suddenness and locality of short-term heavy rainfall. Furthermore, existing models have weak constraints on physical processes, and forecast results are prone to accumulated biases over long-term series, affecting the accuracy and reliability of extrapolation forecasts.

[0004] Therefore, designing a short-term heavy precipitation forecasting method to avoid the influence of multi-source data and long-term series, and to improve the accuracy and reliability of forecasts, has become an urgent problem to be solved. Summary of the Invention

[0005] Based on this, the present invention proposes a method and system for forecasting short-term heavy precipitation in the near term. By collecting and preprocessing multi-source meteorological data, the quality of the data is improved, providing a foundation for subsequent extraction of multi-source data features. A prediction model for multi-source meteorological data is designed, and through the processing of feature fusion, evolution, and generation modules, a deep understanding and refinement of meteorological features from different sources is achieved. This effectively captures the suddenness and locality of short-term heavy precipitation, improving the accuracy and stability of the forecast of the intensity evolution of short-term heavy precipitation. Furthermore, a regional correction method for rain cluster precipitation is designed to avoid the problem of accumulated biases in long-term series caused by rain cluster evolution, thus improving the accuracy and reliability of the forecast. The present invention improves the accuracy and reliability of short-term heavy precipitation forecasting.

[0006] This invention proposes a method for forecasting short-term heavy precipitation in the near term, comprising: Collect and preprocess multi-source meteorological data, including radar network observation data, satellite-inverted GNSS_MET water vapor data, and ground automatic weather station observation data. The preprocessed multi-source meteorological data is input into the prediction model to obtain basic forecast results. The prediction model is based on the NowcastNet deep neural network architecture and includes a feature fusion module, an evolution module, and a generation module. The basic forecast results are corrected for rain cluster precipitation areas to obtain the final forecast results. The rain cluster precipitation area correction includes rain cluster regional positioning and weak precipitation rain cluster region screening.

[0007] In summary, based on the aforementioned method for forecasting short-term heavy precipitation in the near term, the quality of the data is improved by collecting and preprocessing multi-source meteorological data, providing a foundation for subsequent extraction of multi-source data features. Furthermore, a prediction model for multi-source meteorological data is designed. Through feature fusion, evolution, and generation modules, a deep understanding and refinement of meteorological features from different sources is achieved, effectively capturing the suddenness and locality of short-term heavy precipitation. This improves the accuracy and stability of forecasting the intensity evolution of short-term heavy precipitation. Additionally, a regional correction method for rain cluster precipitation is designed to avoid the problem of accumulated biases in long-term series due to rain cluster evolution, thus improving the accuracy and reliability of the forecast. This invention improves the accuracy and reliability of short-term heavy precipitation forecasts. Specifically, the process involves collecting and preprocessing multi-source meteorological data, including radar network observation data, satellite-derived GNSS_MET water vapor data, and ground automatic weather station observation data. This improves data quality and provides a foundation for subsequent feature extraction from multi-source data. The preprocessed multi-source meteorological data is then input into a prediction model to obtain basic forecast results. The prediction model is based on the NowcastNet deep neural network architecture and includes a feature fusion module, an evolution module, and a generation module. This enables a deep understanding and refinement of meteorological features from different sources, effectively capturing the suddenness and locality of short-term heavy precipitation, and improving the accuracy and stability of forecasts regarding the intensity and evolution of short-term heavy precipitation. The basic forecast results are then corrected for rain cluster precipitation regions to obtain the final forecast results. This correction includes rain cluster regional location and weak precipitation rain cluster region filtering, avoiding the problem of accumulated biases over long-term series due to rain cluster evolution, thus improving the accuracy and reliability of the forecast. This invention improves the accuracy and reliability of short-term heavy precipitation forecasts.

[0008] Furthermore, the step of collecting and preprocessing multi-source meteorological data specifically includes: Collect radar network observation data, satellite-retrieved GNSS_MET water vapor data, and ground automatic weather station observation data; The observation data from the ground-based automatic weather station are subjected to change removal processing, which includes daily change removal and monthly change removal. The specific algorithm for change removal processing is as follows: , in, Indicates the value of the variable. Indicates the value of the variable term. Indicates the value of the non-changing term. A standardized dimensionless quantity representing the value of a variable. Represents the average variable value. This represents the average value of the non-changing term. NIndicates the total number of data items. i represents the ordinal number of the data item, and t represents the ordinal number of the time item; The radar network observation data is supplemented by a process based on radar reflectivity factor and radar-derived rainfall intensity data.

[0009] Furthermore, the step of inputting the preprocessed multi-source meteorological data into the prediction model to obtain the basic forecast results specifically includes: The prediction model includes a feature fusion module, an evolution module, and a generation module; The preprocessed multi-source meteorological data is input into the feature fusion module. The feature fusion module includes three input channel layers and one feature fusion layer. Each input channel layer is composed of a bilinear interpolation block and a double convolution block connected sequentially. The input channel layers are independent and connected in parallel. The parallel input channel layers are connected to a feature fusion layer. The feature fusion layer performs tensor splicing on the calculation results of the three input channel layers. The specific algorithm of the feature fusion module is as follows: , , , , in, and These represent the intermediate outputs of a double-layer convolution and a single-layer convolution, respectively. , , These represent the parameters of the two-dimensional convolution kernel after batch normalization of radar network observation data, satellite-retrieved GNSS_MET water vapor data, and ground automatic weather station observation data, respectively. This indicates the input data for the feature fusion module. , , These represent the bias parameters for different convolution kernels. Represents the convolution operator. Represents the ReLU activation function. Indicates batch normalization, This represents the output of the double convolution block. This represents the output of the feature fusion module. This indicates the ordinal number of the three independent variables: radar network observation data, satellite-inverted GNSS_MET water vapor data, and ground automatic weather station observation data. The evolution module calculates the rain cluster movement vector and intensity change based on the output features of the feature fusion module. The evolution module is a 4-layer Unet neural network structure, which includes an encoding layer and a decoding layer. The encoding layer includes 4 downsampling layers, and the decoding layer includes 4 upsampling layers. The upsampling layers and the downsampling layers are sequentially connected. The encoding layer and the decoding layer are sequentially connected. The upsampling layer and the downsampling layer are connected by skip connections. The generation module generates basic forecast results based on the rain cluster movement vector and intensity changes. The generation module is a 3-layer Unet neural network structure, which includes an encoding layer and a decoding layer. The encoding layer includes 3 downsampling layers, and the decoding layer includes 3 upsampling layers. The upsampling layers and the downsampling layers are sequentially connected. The encoding layer and the decoding layer are sequentially connected. The upsampling layer and the downsampling layer are connected by skip connections.

[0010] Furthermore, the step of generating the basic forecast result based on the rain cluster movement vector and intensity change by the generation module further includes: Loss optimization is performed based on piecewise weighted root mean square error, with the optimization based on effective precipitation weight adjustment. The specific algorithm for loss optimization is as follows: , , in, This represents the value of the loss function. N Indicates the total number of data items. i Indicates the ordinal number of the data item. Indicates the effective precipitation weight. This indicates the predicted precipitation value. This represents the measured precipitation value. and These represent the weight values ​​for no effective precipitation and the weight values ​​for effective precipitation, respectively.

[0011] Furthermore, the step of correcting the basic forecast results for rain cluster precipitation areas to obtain the final forecast results specifically includes: The location of continuous rain clusters is identified in the gridded field of short-term heavy precipitation forecasts using the 8-connected neighborhood method. The specific algorithm for identifying the location of continuous rain clusters is as follows: N(R pq )={R (p±1,q) R (p,q±1) R (p,q) R (p+1,q±1) R (p-1,q±1)}, Wherein, N(R) pqR represents the grid set of short-duration heavy precipitation, which includes one grid point indicating the occurrence of the short-duration heavy precipitation and eight grid points representing the locations of neighboring occurrence grid points. pq The values ​​represent precipitation status, with p and q representing the x and y coordinates of the grid point where the forecasted short-duration heavy precipitation occurs, respectively. The principal component analysis algorithm is used to determine the direction of the rain cluster's principal axis and identify the rain cluster. Based on the direction of the guiding airflow above the rain cluster, the upstream region of the rain cluster is tracked and identified. Then, the rain cluster body and the upstream region of the rain cluster are merged to obtain the complete rain cluster region. The weak precipitation rain cluster region is then screened out from the complete rain cluster region.

[0012] Furthermore, the step of determining the principal axis direction of the rain cluster and identifying the rain cluster based on the principal component analysis algorithm specifically includes: The variance contribution rate of the eigenvectors is calculated using the principal component analysis algorithm to determine the principal axis direction of the rain cluster spatial distribution structure. Rain clusters are then identified based on the principal axis direction, with the identification based on a rectangular bounding box along the principal axis direction. The specific algorithm for determining the principal axis direction of the rain clusters is as follows: , , , = ( , ), in, and These represent the x and y coordinates of the centered rain cluster grid points, respectively. and These represent the x and y coordinates of the rain cluster grid points, respectively. and Represents the geometric center coordinates of the rain cluster grid points. Represents the covariance matrix. Represents the number of grid points. This represents the coordinate matrix of the rain cluster grid points after centralization. This indicates transpose.

[0013] Furthermore, the step of tracking and identifying the upstream region of the rain cluster based on the direction of the guiding airflow above the rain cluster, then merging the rain cluster body and the upstream region to obtain a complete rain cluster region, and then filtering out weak precipitation rain cluster regions from the complete rain cluster region, specifically includes: The direction of the steering airflow over the rain cluster is calculated based on a numerical weather prediction model. This calculation includes calculating the mean wind field and the zonal angle between the mean wind and the steering airflow. The specific algorithm for calculating the direction of the steering airflow over the rain cluster is as follows: , , , in, and These represent the zonal component and the meridional component of the mean wind, respectively. and These represent the zonal component and the meridional component of the wind vector on the isobaric surface, respectively. and These represent the zonal component and the meridional component of the wind vector, respectively. Indicates the zonal angle of the mean wind guiding the airflow; The grid points of the rain cluster region are projected from their original coordinates to the direction of the guiding airflow above the rain cluster to obtain the new coordinates of the rain cluster region grid points. The rain cluster region at the new coordinates is then marked with a rectangle to obtain the complete rain cluster region. The specific algorithm for obtaining the complete rain cluster region is as follows: , , in, and These represent the new coordinates of the grid points in the rain cluster region. and These represent the original coordinates of the grid points in the rain cluster region. Indicates the complete rain cluster area. and These represent the center coordinates of the complete rain cluster region. and These represent the distances the rectangle extends in the direction of the guiding airflow and perpendicular to the guiding airflow, respectively. and These represent the semi-major axes of the rectangle in the direction of guiding airflow and perpendicular to the direction of guiding airflow, respectively; Hourly rainfall within the complete rain cluster area is obtained to determine whether the rain cluster area is a weak precipitation rain cluster area. After removing weak precipitation rain cluster areas, the final forecast result is obtained. The specific algorithm for determining whether a rain cluster area is a weak precipitation rain cluster area is as follows: , in, This indicates the number of non-weak precipitation stations within the complete rain cluster area. Indicates the ordinal number of the site. This represents the hourly rainfall at a single station within the entire rain cluster area. Indicates the precipitation threshold. This indicates the threshold for the number of sites.

[0014] The present invention proposes a short-term heavy precipitation forecasting system for the near-term, comprising: The acquisition module is used to acquire and preprocess multi-source meteorological data, including radar network observation data, satellite-inverted GNSS_MET water vapor data, and ground automatic weather station observation data. The basic forecast module is used to input preprocessed multi-source meteorological data into the prediction model to obtain basic forecast results. The prediction model is based on the NowcastNet deep neural network architecture and includes a feature fusion module, an evolution module, and a generation module. The revised forecast module is used to correct the precipitation area of ​​the rain clusters in the basic forecast results to obtain the final forecast results. The correction of the precipitation area of ​​the rain clusters includes the localization of the rain clusters and the filtering out of the rain clusters with weak precipitation.

[0015] The present invention also provides a storage medium that stores one or more programs, which, when executed by a processor, implement the short-term heavy precipitation forecasting method for the near-term as described above.

[0016] The present invention also provides a computer device, the computer device including a memory and a processor, wherein: The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the short-term heavy precipitation forecasting method for the near future as described above. Attached Figure Description

[0017] Figure 1 This is a flowchart of the short-term heavy precipitation forecasting method for the immediate period proposed in the first embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the short-term heavy precipitation forecasting system for the near-term proposed in the second embodiment of the present invention; Figure 3 This is a schematic diagram of the prediction model proposed in this invention; Figure 4 This is a schematic diagram of the rain cluster identifier proposed in this invention; Figure 5 This is a schematic diagram illustrating the upstream region tracking of rain clusters proposed in this invention; Figure 6a This is based on actual short-term heavy rainfall observation results; Figure 6b The prediction results are from the original NowcastNet network model; Figure 6c The basic prediction results of the prediction model proposed in this invention Figure 6d This is the final prediction result of the short-term heavy precipitation forecasting method for the near-term proposed in the first embodiment of the present invention; Figure 6eThis is a prediction result based on existing technology; The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0018] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0019] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0021] Please see Figure 1 The diagram shows a flowchart of a short-term heavy precipitation forecasting method for the near-term proposed in the first embodiment of the present invention. This short-term heavy precipitation forecasting method for the near-term includes steps S01 to S03, wherein: Step S01: Collect multi-source meteorological data and perform preprocessing; It should be noted that in this embodiment, the multi-source meteorological data includes radar network observation data, satellite-retrieved GNSS_MET water vapor data, and ground automatic weather station observation data. The observation data from the ground-based automatic weather station are subjected to change removal processing, which includes daily change removal and monthly change removal. The specific algorithm for change removal processing is as follows: , in, Indicates the value of the variable. Indicates the value of the variable term. Indicates the value of the non-changing term. A standardized dimensionless quantity representing the value of a variable. Represents the average variable value. This represents the average value of the non-changing term. N Indicates the total number of data items. i represents the ordinal number of the data item, and t represents the ordinal number of the time item; The radar network observation data is supplemented by a process based on radar reflectivity factor and radar-derived rainfall intensity data.

[0022] Step S02: Input the preprocessed multi-source meteorological data into the prediction model to obtain basic forecast results; It should be noted that in this embodiment, the prediction model is based on the NowcastNet deep neural network architecture. The prediction model includes a feature fusion module, an evolution module, and a generation module. For the specific structure of the prediction model, please refer to [link / reference needed]. Figure 3 ; The preprocessed multi-source meteorological data is input into the feature fusion module. The feature fusion module includes three input channel layers and one feature fusion layer. Each input channel layer is composed of a bilinear interpolation block and a double convolution block connected sequentially. The input channel layers are independent and connected in parallel. The parallel input channel layers are connected to a feature fusion layer. The feature fusion layer performs tensor splicing on the calculation results of the three input channel layers. The specific algorithm of the feature fusion module is as follows: , , , , in, and These represent the intermediate outputs of a double-layer convolution and a single-layer convolution, respectively. , , These represent the parameters of the two-dimensional convolution kernel after batch normalization of radar network observation data, satellite-retrieved GNSS_MET water vapor data, and ground automatic weather station observation data, respectively. This indicates the input data for the feature fusion module. , , These represent the bias parameters for different convolution kernels. Represents the convolution operator. Represents the ReLU activation function. Indicates batch normalization, This represents the output of the double convolution block. This represents the output of the feature fusion module. This indicates the ordinal number of the three independent variables: radar network observation data, satellite-inverted GNSS_MET water vapor data, and ground automatic weather station observation data. Specifically, radar network observation data, satellite-retrieved GNSS_MET water vapor data, and ground automatic weather station observation data are processed through bilinear interpolation to generate multi-source meteorological data of the same dimension. (z∈{1,2,3}). These three parts are then fed into a double convolutional block, and the output is... The dual convolutional block is composed of two two-dimensional convolutional layers and one two-dimensional convolutional layer connected in parallel. Outputs are processed through two 2D convolutional layers and one 2D convolutional layer, respectively. and , and We get the result by arithmetic addition. . The output C of the feature fusion module is obtained by feature fusion and splicing.

[0023] The evolution module calculates the rain cluster movement vector and intensity change based on the output features of the feature fusion module. The evolution module is a 4-layer Unet neural network structure, which includes an encoding layer and a decoding layer. The encoding layer includes 4 downsampling layers, and the decoding layer includes 4 upsampling layers. The upsampling layers and the downsampling layers are sequentially connected. The encoding layer and the decoding layer are sequentially connected. The upsampling layer and the downsampling layer are connected by skip connections. The generation module generates basic forecast results based on the rain cluster movement vector and intensity changes. The generation module is a 3-layer Unet neural network structure, which includes an encoding layer and a decoding layer. The encoding layer includes 3 downsampling layers, and the decoding layer includes 3 upsampling layers. The upsampling layers and the downsampling layers are sequentially connected. The encoding layer and the decoding layer are sequentially connected. The upsampling layer and the downsampling layer are connected by skip connections.

[0024] Loss optimization is performed based on piecewise weighted root mean square error, with the optimization algorithm adjusted according to effective precipitation weights. The specific algorithm for loss optimization is as follows: , , in, This represents the value of the loss function. N Indicates the total number of data items. i Indicates the ordinal number of the data item. Indicates the effective precipitation weight. This indicates the predicted precipitation value. This represents the measured precipitation value. and These represent the weight values ​​for no effective precipitation and the weight values ​​for effective precipitation, respectively.

[0025] Step S03: Correct the precipitation area of ​​the rain clusters in the basic forecast results to obtain the final forecast results; It should be noted that in this embodiment, the correction of the rain cluster precipitation area includes rain cluster area localization and weak precipitation rain cluster area screening. The location of continuous rain clusters is identified in the forecast short-term heavy precipitation grid field according to the 8-connected neighborhood method. The specific algorithm for continuous rain cluster location identification is as follows: N(R pq )={R (p±1,q) R (p,q±1) R (p,q) R (p+1,q±1) R (p-1,q±1)}, Wherein, N(R) pq R represents the grid set of short-duration heavy precipitation, which includes one grid point indicating the occurrence of the short-duration heavy precipitation and eight grid points representing the locations of neighboring occurrence grid points. pq The values ​​represent precipitation status, with p and q representing the x and y coordinates of the grid point where the forecasted short-duration heavy precipitation occurs, respectively. The principal component analysis algorithm is used to determine the principal axis direction of the rain clusters and identify the rain clusters. For details, please refer to [link to relevant documentation]. Figure 4 ; Based on the direction of the guiding airflow above the rain cluster, the upstream region of the rain cluster is tracked and identified. Then, the rain cluster body and its upstream region are merged to obtain the complete rain cluster region. Weak precipitation rain cluster regions are then filtered out from this complete rain cluster region. For details, please refer to [link to relevant documentation]. Figure 5 .

[0026] The variance contribution rate of the eigenvectors is calculated using the principal component analysis algorithm to determine the principal axis direction of the rain cluster spatial distribution structure. Rain clusters are then identified based on the principal axis direction, with the identification based on a rectangular bounding box along the principal axis direction. The specific algorithm for determining the principal axis direction of the rain clusters is as follows: , , , =( , ), in, and These represent the x and y coordinates of the centered rain cluster grid points, respectively. and These represent the x and y coordinates of the rain cluster grid points, respectively. and Represents the geometric center coordinates of the rain cluster grid points. Represents the covariance matrix. Represents the number of grid points. This represents the coordinate matrix of the rain cluster grid points after centralization. This indicates transpose.

[0027] The direction of the steering airflow over the rain cluster is calculated based on a numerical weather prediction model. This calculation includes calculating the mean wind field and the zonal angle between the mean wind and the steering airflow. The specific algorithm for calculating the direction of the steering airflow over the rain cluster is as follows: , , , in, and These represent the zonal component and the meridional component of the mean wind, respectively. and These represent the zonal component and the meridional component of the wind vector on the isobaric surface, respectively. and These represent the zonal component and the meridional component of the wind vector, respectively. Indicates the zonal angle of the mean wind guiding the airflow; The grid points of the rain cluster region are projected from their original coordinates to the direction of the guiding airflow above the rain cluster to obtain the new coordinates of the rain cluster region grid points. The rain cluster region at the new coordinates is then marked with a rectangle to obtain the complete rain cluster region. The specific algorithm for obtaining the complete rain cluster region is as follows: , , in, and These represent the new coordinates of the grid points in the rain cluster region. and These represent the original coordinates of the grid points in the rain cluster region. Indicates the complete rain cluster area. and These represent the center coordinates of the complete rain cluster region. and These represent the distances the rectangle extends in the direction of the guiding airflow and perpendicular to the guiding airflow, respectively. and These represent the semi-major axes of the rectangle in the direction of guiding airflow and perpendicular to the direction of guiding airflow, respectively; Hourly rainfall within the complete rain cluster area is obtained to determine whether the rain cluster area is a weak precipitation rain cluster area. After removing weak precipitation rain cluster areas, the final forecast result is obtained. The specific algorithm for determining whether a rain cluster area is a weak precipitation rain cluster area is as follows: , in, This indicates the number of non-weak precipitation stations within the complete rain cluster area. Indicates the ordinal number of the site. This represents the hourly rainfall at a single station within the entire rain cluster area. Indicates the precipitation threshold. Indicates the threshold number of sites; The proposed solution is compared with observational data and existing technologies. The comparison results are shown in Table 1 below. For specific prediction results, please refer to [link / reference]. Figure 6a , Figure 6b , Figure 6c , Figure 6d , Figure 6e ; Therefore, it can be seen that Scheme 1 (the original NowcastNet network model) and Scheme 2 (using only the prediction model proposed in this invention) predict a relatively scattered distribution of short-term heavy precipitation areas. Besides the actual areas of short-term heavy precipitation, there are also scattered distributions in other areas within the province, resulting in false alarms. Scheme 2's predicted area is more concentrated than Scheme 1's. As shown in Table 1, Scheme 2 has a lower false alarm rate and a higher TS score. This invention significantly improves the forecast performance, with the predicted area highly consistent with the actual area. Scheme 4 (existing technology - the National Meteorological Center's operational product "Fenglei") predicts a smaller area, and the overall forecast is significantly weaker than the actual situation. Analysis of Table 1 shows that this invention has excellent forecast performance. Among the four schemes, it has the highest TS score, the lowest false alarm rate, and the BIAS score closest to 1. Therefore, compared with existing technologies, this invention integrates multi-source observation data, improving the prediction capability of short-term heavy precipitation. Compared with advanced existing technologies, the technical solution of this invention still has a significant advantage in the prediction of short-term heavy precipitation.

[0028] In summary, based on the aforementioned method for forecasting short-term heavy precipitation in the near term, the quality of data is improved by collecting and preprocessing multi-source meteorological data, providing a foundation for subsequent extraction of multi-source data features. Furthermore, a prediction model for multi-source meteorological data is designed. Through feature fusion, evolution, and generation modules, a deep understanding and refinement of meteorological features from different sources is achieved, effectively capturing the suddenness and locality of short-term heavy precipitation. This improves the accuracy and stability of forecasting the intensity evolution of short-term heavy precipitation. Additionally, a regional correction method for rain cluster precipitation is designed to avoid the problem of accumulated biases in long-term series due to rain cluster evolution, thus improving the accuracy and reliability of the forecast. Therefore, this invention improves the accuracy and reliability of short-term heavy precipitation forecasts. Specifically, the process involves collecting and preprocessing multi-source meteorological data, including radar network observation data, satellite-derived GNSS_MET water vapor data, and ground-based automatic weather station observation data. This improves data quality and provides a foundation for subsequent extraction of multi-source data features. The preprocessed multi-source meteorological data is then input into a prediction model to obtain basic forecast results. The prediction model is based on the NowcastNet deep neural network architecture and includes a feature fusion module, an evolution module, and a generation module. This enables a deep understanding and refinement of meteorological features from different sources, effectively capturing the suddenness and locality of short-term heavy precipitation, and improving the accuracy and stability of forecasts regarding the intensity and evolution of short-term heavy precipitation. The basic forecast results are then corrected for rain cluster precipitation regions to obtain the final forecast results. This correction includes rain cluster regional location and weak precipitation rain cluster region filtering, avoiding the problem of accumulated biases in long-term series due to rain cluster evolution, thus improving the accuracy and reliability of the forecast. This invention improves the accuracy and reliability of short-term heavy precipitation forecasts.

[0029] Please see Figure 2 The diagram shows a schematic representation of a short-term heavy rainfall forecasting system for the near-term period proposed in the second embodiment of the present invention. The system includes: The acquisition module 10 is used to acquire multi-source meteorological data and perform preprocessing. The multi-source meteorological data includes radar network observation data, satellite-inverted GNSS_MET water vapor data, and ground automatic weather station observation data. The basic forecast module 20 is used to input preprocessed multi-source meteorological data into the prediction model to obtain basic forecast results. The prediction model is based on the NowcastNet deep neural network architecture and includes a feature fusion module, an evolution module, and a generation module. The revised forecast module 30 is used to correct the precipitation area of ​​the rain clusters in the basic forecast results to obtain the final forecast results. The correction of the precipitation area of ​​the rain clusters includes the location of the rain clusters and the screening of the rain clusters with weak precipitation.

[0030] The present invention also proposes a computer storage medium storing one or more programs that, when executed by a processor, implement the aforementioned method for forecasting short-term heavy rainfall in the near future.

[0031] The present invention also proposes a computer device, including a memory and a processor, wherein the memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory to implement the above-mentioned method for short-term heavy precipitation forecasting in the near future.

[0032] Those skilled in the art will understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain stored, communicated, propagated, or transmitted programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0033] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0034] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0035] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0036] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for forecasting short-term heavy precipitation in the near term, characterized in that, include: Collect and preprocess multi-source meteorological data, including radar network observation data, satellite-inverted GNSS_MET water vapor data, and ground automatic weather station observation data. The preprocessed multi-source meteorological data is input into the prediction model to obtain basic forecast results. The prediction model is based on the NowcastNet deep neural network architecture and includes a feature fusion module, an evolution module, and a generation module. The basic forecast results are corrected for rain cluster precipitation areas to obtain the final forecast results. The rain cluster precipitation area correction includes rain cluster regional positioning and weak precipitation rain cluster region screening.

2. The method for forecasting short-term heavy precipitation in the near term according to claim 1, characterized in that, The steps of collecting and preprocessing multi-source meteorological data specifically include: Collect radar network observation data, satellite-retrieved GNSS_MET water vapor data, and ground automatic weather station observation data; The observation data from the ground-based automatic weather station are subjected to change removal processing, which includes daily change removal and monthly change removal. The specific algorithm for change removal processing is as follows: ， in, Represents the value of the variable. Indicates the value of the variable term. Indicates the value of the non-changing term. A standardized dimensionless quantity representing the value of a variable. Represents the average variable value. This represents the average value of the non-changing term. N Indicates the total number of data items. i represents the ordinal number of the data item, and t represents the ordinal number of the time item; The radar network observation data is supplemented by a process based on radar reflectivity factor and radar-derived rainfall intensity data.

3. The method for forecasting short-term heavy precipitation in the near future according to claim 1, characterized in that, The step of inputting the preprocessed multi-source meteorological data into the prediction model to obtain the basic forecast results specifically includes: The prediction model includes a feature fusion module, an evolution module, and a generation module; The preprocessed multi-source meteorological data is input into the feature fusion module. The feature fusion module includes three input channel layers and one feature fusion layer. Each input channel layer is composed of a bilinear interpolation block and a double convolution block connected sequentially. The input channel layers are independent and connected in parallel. The parallel input channel layers are connected to a feature fusion layer. The feature fusion layer performs tensor splicing on the calculation results of the three input channel layers. The specific algorithm of the feature fusion module is as follows: ， ， ， ， in, and These represent the intermediate outputs of a double-layer convolution and a single-layer convolution, respectively. , , These represent the parameters of the two-dimensional convolution kernel after batch normalization of radar network observation data, satellite-retrieved GNSS_MET water vapor data, and ground automatic weather station observation data, respectively. This indicates multi-source meteorological data. , , These represent the bias parameters for different convolution kernels. Represents the convolution operator. Represents the ReLU activation function. Indicates batch normalization, This represents the output of the double convolution block. This represents the output of the feature fusion module. This indicates the ordinal number of the three independent variables: radar network observation data, satellite-inverted GNSS_MET water vapor data, and ground automatic weather station observation data. The evolution module calculates the rain cluster movement vector and intensity change based on the output features of the feature fusion module. The evolution module is a 4-layer Unet neural network structure, which includes an encoding layer and a decoding layer. The encoding layer includes 4 downsampling layers, and the decoding layer includes 4 upsampling layers. The upsampling layers and the downsampling layers are sequentially connected. The encoding layer and the decoding layer are sequentially connected. The upsampling layer and the downsampling layer are connected by skip connections. The generation module generates basic forecast results based on the rain cluster movement vector and intensity changes. The generation module is a 3-layer Unet neural network structure, which includes an encoding layer and a decoding layer. The encoding layer includes 3 downsampling layers, and the decoding layer includes 3 upsampling layers. The upsampling layers and the downsampling layers are sequentially connected. The encoding layer and the decoding layer are sequentially connected. The upsampling layer and the downsampling layer are connected by skip connections.

4. The method for forecasting short-term heavy precipitation in the near term according to claim 3, characterized in that, The step of generating the basic forecast result based on the rain cluster movement vector and intensity change in the generation module further includes: Loss optimization is performed based on piecewise weighted root mean square error, with the optimization algorithm adjusted according to effective precipitation weights. The specific algorithm for loss optimization is as follows: ， ， in, This represents the value of the loss function. N Indicates the total number of data items. i Indicates the ordinal number of the data item. Indicates the effective precipitation weight. This indicates the predicted precipitation value. This represents the measured precipitation value. and These represent the weight values ​​for no effective precipitation and the weight values ​​for effective precipitation, respectively.

5. The method for forecasting short-term heavy precipitation in the near term according to claim 1, characterized in that, The step of correcting the basic forecast results for rain cluster precipitation areas to obtain the final forecast results specifically includes: The location of continuous rain clusters is identified in the gridded field of short-term heavy precipitation forecasts using the 8-connected neighborhood method. The specific algorithm for identifying the location of continuous rain clusters is as follows: N(R pq )={R (p±1,q) ,R (p,q±1) ,R (p,q) ,R (p+1,q±1) ,R (p-1,q±1) }, Wherein, N(R) pq R represents the grid set of short-duration heavy precipitation, which includes one grid point indicating the occurrence of the short-duration heavy precipitation and eight grid points representing the locations of neighboring occurrence grid points. pq The values ​​represent precipitation status, with p and q representing the x and y coordinates of the grid point where the forecasted short-duration heavy precipitation occurs, respectively. The principal component analysis algorithm is used to determine the direction of the rain cluster's principal axis and identify the rain cluster. Based on the direction of the guiding airflow above the rain cluster, the upstream region of the rain cluster is tracked and identified. Then, the rain cluster body and the upstream region of the rain cluster are merged to obtain the complete rain cluster region. The weak precipitation rain cluster region is then screened out from the complete rain cluster region.

6. The method for forecasting short-term heavy precipitation in the near term according to claim 5, characterized in that, The step of determining the principal axis direction of rain clusters and identifying rain clusters based on the principal component analysis algorithm specifically includes: The variance contribution rate of the eigenvectors is calculated using the principal component analysis algorithm to determine the principal axis direction of the rain cluster spatial distribution structure. Rain clusters are then identified based on the principal axis direction, with the identification based on a rectangular bounding box along the principal axis direction. The specific algorithm for determining the principal axis direction of the rain clusters is as follows: ， ， ， =（ ， )， in, and These represent the x and y coordinates of the centered rain cluster grid points, respectively. and These represent the x and y coordinates of the rain cluster grid points, respectively. and Represents the geometric center coordinates of the rain cluster grid points. Represents the covariance matrix. Represents the number of grid points. This represents the coordinate matrix of the rain cluster grid points after centralization. This indicates transpose.

7. The method for forecasting short-term heavy precipitation in the near term according to claim 6, characterized in that, The steps of tracking and identifying the upstream region of the rain cluster based on the direction of the guiding airflow above the rain cluster, merging the rain cluster body and the upstream region to obtain a complete rain cluster region, and then filtering out weak precipitation rain cluster regions from the complete rain cluster region specifically include: The direction of the steering airflow over the rain cluster is calculated based on a numerical weather prediction model. This calculation includes calculating the mean wind field and the zonal angle between the mean wind and the steering airflow. The specific algorithm for calculating the direction of the steering airflow over the rain cluster is as follows: ， ， ， in, and These represent the zonal component and the meridional component of the mean wind, respectively. and These represent the zonal component and the meridional component of the wind vector on the isobaric surface, respectively. and These represent the zonal component and the meridional component of the wind vector, respectively. Indicates the zonal angle of the mean wind guiding the airflow; The grid points of the rain cluster region are projected from their original coordinates to the direction of the guiding airflow above the rain cluster to obtain the new coordinates of the rain cluster region grid points. The rain cluster region at the new coordinates is then marked with a rectangle to obtain the complete rain cluster region. The specific algorithm for obtaining the complete rain cluster region is as follows: ， ， in, and These represent the new coordinates of the grid points in the rain cluster region. and These represent the original coordinates of the grid points in the rain cluster region. Indicates the complete rain cluster area. and These represent the center coordinates of the complete rain cluster region. and These represent the distances the rectangle extends in the direction of the guiding airflow and perpendicular to the guiding airflow, respectively. and These represent the semi-major axes of the rectangle in the direction of guiding airflow and perpendicular to the direction of guiding airflow, respectively; Hourly rainfall within the complete rain cluster area is obtained to determine whether the rain cluster area is a weak precipitation rain cluster area. After removing weak precipitation rain cluster areas, the final forecast result is obtained. The specific algorithm for determining whether a rain cluster area is a weak precipitation rain cluster area is as follows: ， in, This indicates the number of non-weak precipitation stations within the complete rain cluster area. Indicates the ordinal number of the site. This represents the hourly rainfall at a single station within the entire rain cluster area. Indicates the precipitation threshold. This indicates the threshold for the number of sites.

8. A short-term heavy precipitation forecasting system for the near future, characterized in that, include: The acquisition module is used to acquire and preprocess multi-source meteorological data, including radar network observation data, satellite-inverted GNSS_MET water vapor data, and ground automatic weather station observation data. The basic forecast module is used to input preprocessed multi-source meteorological data into the prediction model to obtain basic forecast results. The prediction model is based on the NowcastNet deep neural network architecture and includes a feature fusion module, an evolution module, and a generation module. The revised forecast module is used to correct the precipitation area of ​​the rain clusters in the basic forecast results to obtain the final forecast results. The correction of the precipitation area of ​​the rain clusters includes the localization of the rain clusters and the filtering out of the rain clusters with weak precipitation.

9. A storage medium, characterized in that, The storage medium stores one or more programs that, when executed by a processor, implement the short-term heavy precipitation forecasting method for the near-term as described in any one of claims 1-7.

10. A computer device, characterized in that, The computer device includes a memory and a processor, wherein: The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the short-term heavy precipitation forecasting method for the near-term as described in any one of claims 1-7.

Images