Method and device for predicting rainstorm and storage medium

Abstract

The invention relates to the technical field of weather forecast, in particular to a rainstorm prediction method and device and a storage medium. The method comprises the following steps: acquiring multi-source meteorological data; extracting basic variables according to the multi-source meteorological data; identifying a water vapor conveying belt according to the basic variable; according to the water vapor conveying belt, a water vapor conveying belt track is obtained; according to the water vapor conveying belt and the water vapor conveying belt track, geometric characteristic indexes and dynamic characteristic indexes of the water vapor conveying belt are obtained; and taking the basic variables, the geometrical characteristic indexes and the dynamic characteristic indexes as input data, and performing rainstorm prediction according to a machine learning prediction model. Through the method, rainstorm prediction is realized, accuracy and timeliness of rainstorm prediction are guaranteed, dependence on artificial experience is reduced, and intelligent technical support is provided for rainstorm monitoring and early warning.

Description

Technical Field

[0001] This application relates to the field of meteorological forecasting technology, and in particular to a method, apparatus, and storage medium for predicting heavy rain. Background Technology

[0002] Heavy rain is a weather phenomenon characterized by 24-hour precipitation exceeding 50 mm. As an extreme weather event, heavy rain is often accompanied by heavy rainfall, thunderstorms, and strong winds, and is characterized by its suddenness and high destructive potential. The secondary disasters it triggers, such as floods, landslides, and mudslides, can damage critical infrastructure such as transportation, power, and agriculture. For example, urban flooding can paralyze traffic and submerge underground facilities, while heavy rain in mountainous areas can easily induce flash floods. Therefore, heavy rain forecasting is crucial. Summary of the Invention

[0003] In view of the above problems, embodiments of this application provide a method, apparatus and storage medium for predicting rainstorms, which overcomes or at least partially solves the above problems.

[0004] According to one aspect of the embodiments of this application, a method for predicting heavy rainfall is provided, comprising: acquiring multi-source meteorological data; extracting basic variables based on the multi-source meteorological data; identifying a water vapor transport belt based on the basic variables; acquiring the trajectory of the water vapor transport belt based on the water vapor transport belt; acquiring geometric and dynamic characteristic indicators of the water vapor transport belt based on the water vapor transport belt and the water vapor transport belt trajectory; and using the basic variables, the geometric and dynamic characteristic indicators as input data, and performing heavy rainfall prediction based on a machine learning prediction model.

[0005] In an alternative approach, the input data may also include climate factors, including the monsoon index, El Niño phenomenon, and Southern Oscillation index.

[0006] In one alternative approach, the method for extracting basic variables based on the multi-source meteorological data includes: extracting specific humidity, geopotential height, meridional wind component, zonal wind component, and surface air pressure based on the multi-source meteorological data; and calculating water vapor content, water vapor flux, and water vapor flux divergence based on the specific humidity, the meridional wind component, and the zonal wind component.

[0007] In one alternative approach, the atmosphere comprises isobaric surfaces from the surface to 850 hPa, isobaric surfaces from the surface to 300 hPa, and isobaric surfaces from 850 hPa to 300 hPa. The method for calculating water vapor content, water vapor flux, and water vapor flux divergence based on the specific humidity, the meridional wind component, and the zonal wind component includes: calculating the water vapor content, water vapor flux, and water vapor flux divergence of the isobaric surfaces from the surface to 850 hPa, from the surface to 300 hPa, and from the 850 hPa to 300 hPa, respectively, based on the specific humidity, the meridional wind component, and the zonal wind component.

[0008] In one alternative approach, the method for identifying the water vapor conveyor belt based on the basic variables includes: calculating the vertical water vapor flux by horizontal integration at the grid level according to the PanLu2.0 identification algorithm; setting dual thresholds for local and regional water vapor flux; using the grid where the maximum vertical water vapor flux is located as the search starting point; retrieving target grids in the neighborhood whose vertical water vapor flux exceeds the dual thresholds based on the search starting point, and including the target grids in the water vapor conveyor belt; using the target grids as the search starting point, performing the step of retrieving target grids in the neighborhood whose vertical water vapor flux exceeds the dual thresholds based on the search starting point, and including the target grids in the water vapor conveyor belt, until the water vapor conveyor belt is determined.

[0009] In one alternative approach, the method for identifying the water vapor conveyor belt based on the basic variables further includes: removing closed regions from the identified water vapor conveyor belt; and in the remaining water vapor conveyor belt, performing the step of using the grid where the maximum vertical water vapor flux is located as the search starting point.

[0010] In one optional embodiment, the method for obtaining the trajectory of the water vapor conveyor belt based on the water vapor conveyor belt includes: calculating the vertical water vapor flux by horizontal integration on a grid according to the PanLu2.0 recognition algorithm; using the grid where the maximum vertical water vapor flux is located as a reference point; extracting a first neighboring region based on a preset geographical distance from the reference point; calculating a first weighted water vapor flux direction of the first neighboring region based on the vertical water vapor flux; using a plane perpendicular to the first weighted water vapor flux direction as a cross-section of the reference point; calculating the weighted center of the cross-section; and extracting a second neighboring region based on a preset geographical distance from the weighted center of the cross-section; and calculating the second neighboring region based on the vertical water vapor flux. The second weighted water vapor flux direction; the point obtained by moving a preset distance along the second weighted water vapor flux direction is used as the reference point, and the step of extracting the first adjacent area based on the preset geographical distance from the reference point is performed until the water vapor conveyor belt is covered, thus obtaining the forward trajectory of the water vapor conveyor belt; the point obtained by moving a preset distance along the opposite direction of the second weighted water vapor flux direction is used as the reference point, and the step of extracting the first adjacent area based on the preset geographical distance from the reference point is performed until the water vapor conveyor belt is covered, thus obtaining the reverse trajectory of the water vapor conveyor belt; the forward trajectory of the water vapor conveyor belt and the reverse trajectory of the water vapor conveyor belt are integrated to obtain the trajectory of the water vapor conveyor belt.

[0011] In one alternative approach, the geometric characteristics include the length, width, aspect ratio, curvature, and rotation angle of the water vapor conveyor belt; the dynamic characteristics include the average conveying speed, residence time, and source-sink distribution.

[0012] According to one aspect of the embodiments of this application, a rainstorm prediction device is provided, comprising: a first acquisition module for acquiring multi-source meteorological data; an extraction module for extracting basic variables based on the multi-source meteorological data; an identification module for identifying a water vapor conveyor belt based on the basic variables; a second acquisition module for acquiring the trajectory of the water vapor conveyor belt based on the water vapor conveyor belt; a third acquisition module for acquiring geometric and dynamic characteristic indicators of the water vapor conveyor belt based on the water vapor conveyor belt and the water vapor conveyor belt trajectory; and a rainstorm prediction module for using the basic variables, the geometric and dynamic characteristic indicators as input data, and performing rainstorm prediction based on a machine learning prediction model.

[0013] According to one aspect of the embodiments of this application, a non-volatile computer-readable storage medium is provided, the computer-readable storage medium storing computer-executable instructions, which, when executed by an electronic device, cause the electronic device to perform the method described above.

[0014] The beneficial effects of this application's embodiments are as follows: It provides a method for predicting heavy rainfall, including acquiring multi-source meteorological data; extracting basic variables based on the multi-source meteorological data; identifying water vapor transport belts based on the basic variables; acquiring the trajectory of the water vapor transport belt based on the water vapor transport belt; acquiring geometric and dynamic characteristic indicators of the water vapor transport belt based on the water vapor transport belt and its trajectory; and using the basic variables, geometric and dynamic characteristic indicators as input data, and performing heavy rainfall prediction based on a machine learning prediction model. This method achieves heavy rainfall prediction, ensuring the accuracy and timeliness of heavy rainfall prediction, while reducing reliance on human experience, providing intelligent technical support for heavy rainfall monitoring and early warning. This method combines multi-source meteorological data with machine learning technology, starting from the perspective of water vapor transport belts, to deeply explore the key characteristics of heavy rainfall occurrence and development, achieving heavy rainfall prediction and providing strong support for the scientific prevention and control of meteorological disasters.

[0015] In addition, when identifying the water vapor transport belt, this application embodiment simultaneously selects two isobaric surfaces, one at ground level and the other at 850 hPa and the other at 300 hPa, for analysis to comprehensively capture the water vapor transport characteristics in the lower and upper atmospheres and improve the accuracy of rainstorm forecasting.

[0016] Equally important, the input data for the machine learning prediction model in this application includes geometric and dynamic characteristic indicators of the water vapor transport belt. These indicators reflect the morphology, intensity, and motion characteristics of the water vapor transport belt, thereby providing a more comprehensive characterization of the key physical processes involved in rainstorms. By introducing these indicators, the model can effectively capture the spatiotemporal evolution characteristics of rainstorm events, further improving prediction accuracy.

[0017] Furthermore, the embodiments of this application integrate basic variables and characteristic indicators of the water vapor transport belt to construct multi-dimensional feature input data, which not only enhances the model's ability to represent rainstorm characteristics, but also improves its adaptability to changes in complex weather systems.

[0018] Additional aspects and advantages of this application not covered herein will be described, shown, or illustrated by practice of this application in the following description. Attached Figure Description

[0019] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0020] Figure 1 This is a schematic flowchart of a rainstorm prediction method provided in an embodiment of this application; Figure 2This is a schematic diagram of a rainstorm prediction device provided in an embodiment of this application; Figure 3 This is a schematic diagram of the hardware structure of an electronic device for performing a rainstorm prediction method provided in an embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] It should be noted that when an element is described as "fixed to" another element, it can be directly on the other element, or one or more intermediate elements may exist between them. When an element is described as "connected to" another element, it can be directly connected to the other element, or one or more intermediate elements may exist between them. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this specification are for illustrative purposes only.

[0023] Furthermore, the technical features involved in the various embodiments of this application described below can be combined with each other as long as they do not conflict with each other.

[0024] Please see Figure 1 , Figure 1 This is a flowchart illustrating a method for predicting heavy rainfall provided in an embodiment of this application. The method includes the following steps: Step S1: Obtain multi-source meteorological data.

[0025] The multi-source meteorological data includes reanalysis data, radiosonde data, satellite remote sensing data, and surface precipitation data.

[0026] The reanalysis data is a "comprehensive dataset" that integrates satellite, radiosonde, and ground observations, with broad spatial coverage (global) and long time series (decades). Examples include ERA5, the fifth-generation global atmospheric reanalysis dataset provided by the European Centre for Medium-Range Weather Forecasts (ECMWF), which boasts high spatiotemporal resolution and accuracy and is widely used in meteorological research and operational forecasting. Another example is MERRA-2, a global reanalysis dataset developed by NASA, which provides long-term atmospheric variable information.

[0027] Radiosonde data, collected vertically by a radiosonde carried by a balloon, captures key meteorological parameters such as temperature, humidity, and wind speed. It has high vertical resolution and can effectively reflect atmospheric stratification characteristics.

[0028] Satellite remote sensing data is obtained by using various sensors mounted on satellite platforms to acquire macroscopic information about the Earth's surface and atmosphere, such as AIRS water vapor products. AIRS water vapor products are high-resolution atmospheric detector data provided by the National Oceanic and Atmospheric Administration (NOAA) of the United States. They can reflect the vertical distribution of water vapor, have high spatiotemporal resolution, and are suitable for monitoring abnormal changes in water vapor before heavy rainfall.

[0029] Surface precipitation data is used to correlate water vapor and precipitation, and is key to verifying the "practicality" of water vapor statistics. Examples of surface precipitation data include GPCP, a global precipitation climate dataset developed under the leadership of the National Oceanic and Atmospheric Administration (NOAA). GPCP integrates various satellite and ground observation data, providing long-term, high-precision global precipitation distribution information, suitable for analyzing precipitation intensity and spatial distribution characteristics during heavy rainfall events. Another example is CMORPH, a global high spatiotemporal resolution precipitation estimation dataset developed by NOAA.

[0030] Step S2: Extract basic variables based on the multi-source meteorological data.

[0031] After obtaining multi-source meteorological data, the next step is to extract the basic variables closely related to the occurrence of rainstorms based on these data.

[0032] Before extracting basic variables, it is generally necessary to preprocess the acquired multi-source meteorological data, including data cleaning, format standardization, and spatiotemporal matching. For example, data quality control (removing outliers, such as "abrupt values" caused by radiosonde malfunctions); spatiotemporal matching (unifying data from different sources to the same temporal resolution (e.g., monthly scale) and the same spatial resolution (e.g., 1°×1°) to ensure data comparability); and coordinate system transformation (using a unified latitude and longitude coordinate system to avoid projection errors).

[0033] The fundamental variables include specific humidity, geopotential height, meridional wind component, zonal wind component, and surface air pressure, as well as water vapor content, water vapor flux, and water vapor flux divergence. These fundamental variables comprehensively reflect the distribution and transport of water vapor in the atmosphere. Atmospheric water vapor is not only the main source of surface humidity but also a crucial basis for precipitation, occupying a vital position in the global climate system. Studies have shown a close correlation between changes in atmospheric water vapor and precipitation intensity across Eurasia; an increase in water vapor significantly enhances precipitation intensity, especially during heavy precipitation events. In other words, atmospheric water vapor plays an indispensable and crucial role in the formation of rainstorms. These fundamental variables are important bases for predicting rainstorms.

[0034] In some embodiments, the method for extracting basic variables based on the multi-source meteorological data includes: Step S21: Based on the multi-source meteorological data, extract specific humidity, geopotential height, meridional wind component, zonal wind component, and surface air pressure.

[0035] Among them, specific moisture It is the mass of water vapor per unit mass of air (g / kg), and is a microscopic indicator that directly reflects the water vapor content (the higher the specific humidity, the more humid the air).

[0036] Among them, the geopotential height H is the height of the pressure surface (gpm, geopotential meter), which is used to determine the vertical level (e.g., the geopotential height of 850hPa is about 1500gpm, corresponding to 1.5km altitude).

[0037] Among them, the longitudinal / zonal air volume ( / ) is something that is windy ( North and South ( The 0.5% m / s component is a dynamic index for calculating water vapor transport (the stronger the wind, the stronger the water vapor transport).

[0038] Among them, surface air pressure (ps) is the lower bound of the vertical integral (the water vapor integral from the ground to the upper atmosphere must start from ps).

[0039] Step S22: Calculate the water vapor content, water vapor flux, and water vapor flux divergence based on the specific humidity, the meridional wind component, and the zonal wind component.

[0040] Among them, water vapor content (atmospheric precipitable water) Definition of ): The total amount of water vapor in a certain layer of atmosphere per unit area (mm), equivalent to the thickness of that layer after all the water vapor in that layer has been condensed into water; Calculation method: By vertical integration of specific humidity The formula is:

[0041] in, The acceleration due to gravity (9.8 m / s²) 2 ), The pressure is in hPa. barosphere The specific moisture content, This refers to the top pressure of the layer (e.g., 850 hPa or 300 hPa). This refers to surface air pressure (ps).

[0042] Water vapor flux is the amount of water vapor passing through a unit area per unit time (kg / m·s), reflecting the intensity and direction of water vapor transport; it can be used to analyze the source and path of water vapor (e.g., water vapor in the East Asian monsoon region mainly comes from the South China Sea and the Indian Ocean and is transported by the southwest wind).

[0043] The calculation method is: combined with specific humidity and air volume ( / It is divided into latitudinal flux (east-west direction) and longitudinal flux (north-south direction):

[0044] Water vapor flux divergence; spatial variation rate of water vapor flux (kg / m³) 2 ·s), reflects the accumulation (convergence) or dispersion (divergence) of water vapor in a certain area; it is a "precursor indicator" of precipitation (negative divergence = water vapor convergence, which means that water vapor is accumulating in the area and is likely to form precipitation; positive divergence = water vapor divergence, which means that water vapor is escaping and is not likely to form precipitation).

[0045] Calculation method: Calculate the spatial derivatives (longitude) of the zonal flux and the meridional flux respectively. ),latitude( Then sum them up:

[0046] It is worth noting that this application divides the atmosphere into isobaric surfaces from the surface to 850 hPa, isobaric surfaces from the surface to 300 hPa, and isobaric surfaces from 850 hPa to 300 hPa. The method for calculating water vapor content, water vapor flux, and water vapor flux divergence based on the specific humidity, the meridional wind component, and the zonal wind component includes: calculating the water vapor content, water vapor flux, and water vapor flux divergence of the isobaric surfaces from the surface to 850 hPa, from the surface to 300 hPa, and from 850 hPa to 300 hPa based on the specific humidity, the meridional wind component, and the zonal wind component, respectively.

[0047] Step S3: Identify the water vapor conveyor belt based on the aforementioned basic variables.

[0048] A water vapor transport belt is a channel or area in the atmosphere that carries large amounts of water vapor and transports it from one region to another. Water vapor transport belts are the creators of rainstorms.

[0049] In some embodiments, the method for identifying the water vapor conveyor belt based on the basic variables includes: Step S31: Calculate the vertical water vapor flux by integrating the grid horizontally according to the PanLu2.0 recognition algorithm.

[0050] PanLu2.0 recognition algorithm is a water vapor conveyor belt detection method based on the principles of dynamic threshold and spatial continuity. The grid is the basic unit of geographic space (such as a 0.25°×0.25° latitude and longitude grid), and each grid corresponds to a specific geographic coordinate point.

[0051] The full name of vertical water vapor flux is vertical integral horizontal water vapor flux, which is the core quantitative indicator of water vapor transport belts. It is defined as the mass of water vapor passing through a vertical plane of a unit width per unit time, with units of kg / m·s. Its physical meaning is to describe the intensity of horizontal water vapor transport in the atmosphere, and it is a key basis for identifying water vapor transport belts.

[0052] The formulas for calculating the zonal and meridional components of the vertical water vapor flux are as follows:

[0053] in, This represents the zonal component of the vertical water vapor flux. This represents the meridional component of the vertical water vapor flux. The acceleration due to gravity (9.8 m / s²) 2 ), The pressure is in hPa. barosphere The specific moisture content, This refers to the top pressure of the layer (e.g., 850 hPa or 300 hPa). This refers to surface air pressure (ps). barosphere Zonal wind volume, barosphere Meridional air volume.

[0054] The vertical water vapor flux of each grid cell is calculated using the zonal and meridional components of the vertical water vapor flux. The direction of the vertical water vapor flux can also be calculated using the following formula:

[0055] in, This refers to the vertical water vapor flux. This represents the zonal component of the vertical water vapor flux. This represents the meridional component of the vertical water vapor flux. The direction perpendicular to the water vapor flux indicates the direction of water vapor transport, and its value ranges from 0° to 360°.

[0056] In some embodiments, both ground level to 850 hPa and ground level to 300 hPa are selected simultaneously. (Pa is 850 hPa or 300 hPa), which is a feature of the PanLu2.0 algorithm, and its significance lies in: From the surface to 850 hPa: This reflects the transport of water vapor in the lower layers (such as water vapor advection within the boundary layer) and is the focus of traditional atmospheric river identification; From the surface to 300 hPa: This reflects water vapor transport at deeper levels (such as water vapor transport in the middle troposphere) and can capture more complete water vapor transport belt structures (such as water vapor transport around typhoons and water vapor transport accompanied by upper-level jet streams).

[0057] Step S32: Set dual thresholds for local and regional water vapor flux.

[0058] By setting dual thresholds for local and regional water vapor flux, the core and peripheral areas of a water vapor transport belt can be effectively distinguished. The local threshold identifies the significance of water vapor flux within each grid cell, while the regional threshold determines the continuity and extent of water vapor transport between grid cells. The dual thresholds are set based on historical data statistics and a dynamic adjustment mechanism, ensuring high sensitivity while avoiding noise interference. Combining the local and regional threshold conditions, a set of grid cells matching the characteristics of a water vapor transport belt is selected, thus achieving accurate identification of the belt. For example, in practical implementation, setting the local water vapor flux threshold to a fixed percentile (e.g., 95%) ensures that only the grid cells with the strongest vertical water vapor flux are retained; while the regional water vapor flux threshold is calculated using a sliding window method, where the average vertical water vapor flux of the grid cells within the window must exceed a certain historical threshold to ensure sufficient spatial continuity of the identified area.

[0059] Step S33: Use the grid cell containing the maximum vertical water vapor flux as the search starting point.

[0060] The core area of ​​a water vapor conveyor belt is usually the place where water vapor is most concentrated in the region, that is, the grid where the maximum vertical water vapor flux is located. Therefore, using this grid as the search starting point can effectively locate the core area of ​​the water vapor conveyor belt.

[0061] Step S34: Based on the search starting point, retrieve target grids whose neighborhood vertical water vapor flux exceeds the dual thresholds, and incorporate the target grids into the water vapor conveyor belt.

[0062] Step S35: Using the target grid as the search starting point, perform the step of retrieving target grids whose neighborhood vertical water vapor flux exceeds the dual thresholds based on the search starting point, and incorporating the target grids into the water vapor conveyor belt, until the water vapor conveyor belt is determined.

[0063] In steps S34 and S35, starting from the search starting point, target gratings that simultaneously exceed both local and regional thresholds within the neighborhood (e.g., 8-neighborhood or 4-neighborhood) are retrieved and included in the water vapor transport belt range. Then, using these target gratings as "new starting points," the above process is repeated until no gratings meet the conditions. This iterative approach "from core to edge" can gradually and accurately delineate the boundary of the water vapor transport belt (avoiding boundaries that are too wide or too narrow).

[0064] The identified water vapor conveyor belts may contain closed regions (i.e., isolated small grid groups not connected to other high-value areas). These regions are usually not continuous water vapor conveyor belts, but rather noise or local disturbances, and therefore need to be eliminated. Specifically, see step S36.

[0065] Step S36: Remove the closed area in the identified water vapor conveyor belt.

[0066] Step S37: In the remaining water vapor conveyor belt, perform the step of using the grid where the maximum vertical water vapor flux is located as the search starting point.

[0067] After removing closed regions, the main structure of the water vapor conveyor belt becomes clearer. At this point, in the remaining water vapor conveyor belt, the search starting point is again the grid cell containing the maximum vertical water vapor flux, ensuring the identification process always revolves around the core area of ​​the conveyor belt. Next, steps S34 and S35 are executed again until no grid cell meeting the criteria is included in the new water vapor conveyor belt. By iterating this process repeatedly, interfering regions can be gradually eliminated, retaining the truly continuous and physically meaningful water vapor conveyor belt.

[0068] Step S4: Obtain the trajectory of the water vapor conveyor belt based on the water vapor conveyor belt.

[0069] A water vapor transport belt refers to the path that an air mass carrying water vapor takes over a certain period of time. Water vapor in the atmosphere is not stationary; it moves with air currents. These air currents, carrying large amounts of water vapor, act like a "conveyor belt," transporting the water vapor from its source to other regions.

[0070] In some embodiments, the method for obtaining the trajectory of the water vapor conveyor belt based on the water vapor conveyor belt includes: Step S41: Calculate the vertical water vapor flux by integrating the grid horizontally according to the PanLu2.0 recognition algorithm. Step S41 is the same as step S31 above, and will not be repeated here.

[0071] Step S42: Use the grid cell containing the maximum vertical water vapor flux as the reference point.

[0072] The core area of ​​a water vapor conveyor belt is usually the place where water vapor is most concentrated in the region, that is, the grid where the maximum vertical water vapor flux is located. Therefore, using this grid as a reference point can effectively locate the core of the water vapor conveyor belt trajectory.

[0073] Step S43: Extract the first neighboring region based on the preset geographical distance from the reference point.

[0074] The preset geographical distance can be set according to actual needs. For example, grids within 30km, 50km or 100km of the reference point can be used as the first nearest area. The preset geographical distance is not a simple numerical straight-line distance, but a distance measurement determined in the actual geographical environment by comprehensively considering the geographical features of the Earth's surface (such as topography, landforms and other factors).

[0075] Step S44: Calculate the first weighted water vapor flux direction of the first adjacent region based on the vertical water vapor flux.

[0076] The vertical water vapor flux of each grid cell has a certain directionality, as described above. Within the first adjacent region, based on the magnitude of the vertical water vapor flux of each grid (as described above) The direction of water vapor transport in the first adjacent region can be calculated using a weighted average method, which is the direction of the first weighted water vapor flux.

[0077] Step S45: Use a plane perpendicular to the direction of the first weighted water vapor flux as the cross section of the reference point.

[0078] It is understood that the cross-section passes through the reference point.

[0079] Step S46: Calculate the weighted center of gravity of the cross section, and extract the second neighboring region based on the preset geographical distance from the weighted center of gravity. Calculate the second weighted water vapor flux direction of the second neighboring region according to the vertical water vapor flux.

[0080] The weighted center of gravity of the cross-section is calculated by weighting the vertical water vapor flux of each grid cell in the cross-section according to its magnitude and direction. Then, using this weighted center of gravity as a new reference point, and based on a preset geographical distance (e.g., 30km, 50km, or 100km) from it, a second neighboring region is extracted. This second neighboring region also considers the influence of geographical features and matches the actual terrain conditions.

[0081] The method for calculating the second weighted water vapor flux direction of the second adjacent region based on the vertical water vapor flux can refer to the method for calculating the first weighted water vapor flux direction of the first adjacent region based on the vertical water vapor flux, and will not be repeated here.

[0082] Step S47: The point obtained by moving a preset distance along the second weighted water vapor flux direction is used as the reference point. The step of extracting the first neighboring area based on the preset geographical distance from the reference point is performed until the water vapor conveyor belt is covered, and the forward trajectory of the water vapor conveyor belt is obtained.

[0083] The preset distance can be 100km.

[0084] It is understandable that after using the point obtained by moving a preset distance along the second weighted water vapor flux direction as the reference point, the step of extracting the first neighboring region based on the preset geographical distance from the reference point is not simply executing step S43, but rather executing a cyclical process from steps S43 to S47 until the entire range of the water vapor transport belt is covered, obtaining the complete forward trajectory of the water vapor transport belt. By cyclically executing the above steps, the main direction of water vapor transport can be continuously tracked, and the trajectory can be gradually extended to the entire range of the transport belt. This method fully considers the influence of terrain on the water vapor transport path, ensuring that the trajectory generation is closer to the actual transport process. The final generated forward trajectory can provide important data support for meteorological analysis, climate research, and disaster early warning.

[0085] Step S48: Move a point obtained by moving a preset distance in the direction opposite to the direction of the second weighted water vapor flux as the reference point, and perform the step of extracting the first neighboring area based on the preset geographical distance from the reference point until the water vapor conveyor belt is covered, and obtain the reverse trajectory of the water vapor conveyor belt.

[0086] In step S48, the preset distance can also be 100km. After moving the preset distance along the direction opposite to the second weighted water vapor flux direction to obtain a new reference point, the first neighboring region within a preset geographical distance range around this reference point is extracted (step S43). Steps S43 to S47 are repeated until the water vapor transport belt is covered, thus gradually tracing back the initial path of water vapor transport. By tracing in the opposite direction, not only can the incoming path of water vapor transport be obtained, but the rationality of the forward trajectory can also be further verified, improving the accuracy of the overall trajectory construction.

[0087] Step S49: Integrate the forward trajectory of the water vapor conveyor belt and the reverse trajectory of the water vapor conveyor belt to obtain the water vapor conveyor belt trajectory.

[0088] A complete water vapor conveyor belt track includes both the forward trajectory and the reverse trajectory of the water vapor conveyor belt. By integrating the forward and reverse trajectories of the water vapor conveyor belt, a complete water vapor conveyor belt track can be formed.

[0089] Step S5: Based on the water vapor conveyor belt and its trajectory, obtain the geometric and dynamic characteristic indicators of the water vapor conveyor belt.

[0090] The geometric characteristics include the length, width, aspect ratio, curvature, and rotation angle of the water vapor conveyor belt; the dynamic characteristics include the average conveying speed, residence time, and source-sink distribution.

[0091] Length: The extended distance of the water vapor conveyor belt track. It is not the straight-line distance from the start to the end of the water vapor conveyor belt track, but the actual extended length along the track path.

[0092] Width: The span of the high-value area of ​​water vapor flux perpendicular to the direction of trajectory extension (e.g., taking a vertical cross-section of a point on the trajectory and calculating the grid width where the vertical water vapor flux is greater than the threshold), reflecting the "thickness" of the water vapor conveyor belt.

[0093] Length-to-width ratio (L / W): The ratio of length to width, which reflects the "slenderness" of the conveyor belt (the larger the ratio, the more belt-like the conveyor belt).

[0094] Curvature: The ratio of the length of the track to the straight-line distance from the start to the end of the water vapor conveyor belt track, reflecting the degree of curvature of the track (the larger the ratio, the more obvious the curvature).

[0095] Rotation angle: The angle of change of the water vapor conveyor belt track (such as the angle between a certain section of the track and the initial direction), reflecting the "turning" characteristics of the conveyor belt (such as the rotation of the outer conveyor belt of a typhoon).

[0096] Average conveying speed: the ratio of the length of the water vapor conveyor belt track to the conveying time, where the conveying time can be calculated by the timestamps of each point on the water vapor conveyor belt track, reflecting the average moving speed of the water vapor.

[0097] Residence time: The duration of water vapor in the water vapor transport belt (such as the time it takes for a clump of water vapor to enter and leave the water vapor transport belt). It can be calculated by tracking the trajectory of water vapor particles and reflects the degree of "retention" of water vapor in the water vapor transport belt (the longer the residence time, the easier it is to form heavy precipitation).

[0098] The source-sink area includes the source area and the sink area.

[0099] Source region: The area from which water vapor originates (e.g., the starting point of a reverse trajectory tracing, i.e., the initial location of the water vapor, usually an ocean or humid region); Sinking area: The region where water vapor converges (such as the endpoint of a forward trajectory, i.e., the final landing location of the water vapor, which is usually a precipitation area).

[0100] The source / sink area can be determined by the starting point (end point in the reverse direction) and the ending point (end point in the forward direction) of the water vapor conveyor belt trajectory, reflecting the "source-destination" path of water vapor.

[0101] Step S6: Using the basic variables, geometric feature indicators, and dynamic feature indicators as input data, perform rainstorm prediction based on the machine learning prediction model.

[0102] The machine learning prediction model is obtained by optimizing the training samples. The training samples include historical precipitation events, multi-source meteorological data related to the historical precipitation events, basic variables proposed based on the multi-source meteorological data as described in the embodiments of this application, geometric characteristic indicators of the water vapor transport belt, and dynamic characteristic indicators.

[0103] In some embodiments, the input data further includes climate factors, including the monsoon index, El Niño phenomenon, and Southern Oscillation index. The samples used in the machine learning prediction model obtained from the training samples may also include the aforementioned monsoon index.

[0104] El Niño refers to the phenomenon of abnormally warm sea surface temperatures across a large area of ​​the central and eastern equatorial Pacific Ocean. Generally, an El Niño event is defined as a sea surface temperature in this region that is 0.5°C higher than the average temperature for the same period in previous years for more than three consecutive months. El Niño causes global climate anomalies. In the Pacific coastal regions, the west coast of South America may experience torrential rains and floods, while Australia, Indonesia, and other areas may experience droughts. Globally, it also affects the intensity and path of monsoons, altering precipitation patterns and leading to extreme rainfall events in some areas while other areas experience less rainfall. It may also influence the formation location and frequency of tropical cyclones.

[0105] The Southern Oscillation (SOI) is a seesaw phenomenon in air pressure changes between the tropical Pacific and tropical Indian Oceans. Specifically, when air pressure decreases in the eastern tropical Pacific (during El Niño), air pressure rises in the western tropical Pacific; conversely, when air pressure rises in the eastern tropical Pacific (during La Niña), air pressure decreases in the western tropical Pacific. This inverse relationship in air pressure is measured by the SOI index, which measures sea-level pressure anomalies. The SOI also has a significant impact on global climate, closely linked to El Niño, jointly influencing global atmospheric circulation and precipitation distribution. For example, during the warm phase of the El Niño-SOI (El Niño event), the Asian summer monsoon weakens, potentially leading to more precipitation in southern my country and less precipitation in northern China; while during the cold phase (La Niña event), the situation may be reversed, with the Asian summer monsoon strengthening, potentially increasing precipitation in the north and decreasing precipitation in the south.

[0106] The machine learning prediction model can be constructed using algorithms such as random forest, support vector machine, or deep neural network. By continuously adjusting the model parameters through training samples, the model can accurately identify the key feature combination of precipitation occurrence, and then output a prediction of rainstorm based on the newly input basic variables, geometric feature indicators, dynamic feature indicators, and climate factors.

[0107] The forecast for heavy rainfall includes 24-hour and 72-hour precipitation. It is understood that in some embodiments, the probability of heavy rainfall in either the 24-hour or 72-hour period can also be predicted.

[0108] In some embodiments, SCI clustering analysis is performed on historical weather data to train a clustering model and multiple weather situation categories (e.g., Type_0: no rain, Type_1: persistent light rain, Type_2: short-term severe convection, Type_3: regional rainstorm). Then, the basic variables, geometric feature indices, dynamic feature indices, and climate factors are input, and SCI clustering inference is used to obtain the weather situation category for the current basic variables, geometric feature indices, dynamic feature indices, and climate factors. Spatial features are extracted using a CNN convolutional neural network. Subsequently, the weather situation information is fused with the spatial features extracted by the CNN convolutional neural network, and the fused features are processed using a Transformer Encoder (the encoder part of the transformer) to output precipitation (regression) for different forecast periods (24 hours, 72 hours). This regression prediction can also be converted into a final rainstorm event forecast using a thresholding method.

[0109] In some embodiments, RMSE (root mean square error), MAE (mean absolute error), and R can also be used.2 Indicators such as (coefficient of determination) were used to evaluate the effectiveness of forecasts for 24-hour and 72-hour precipitation, respectively.

[0110] Among them, SCI clustering reasoning is a clustering method based on similarity measurement. By extracting latent patterns from historical meteorological data, it provides a basis for classifying current input data, thereby enhancing the ability of machine learning prediction models to distinguish extreme weather events. Combining it with convolutional neural networks (CNN) can effectively extract spatial feature information from meteorological data, extract data-driven local detail features (such as spatial "high value centers" and temporal "mutation points"), and reflect small-scale variations in water vapor transport (such as local water vapor convergence in a certain area may trigger localized heavy rainfall).

[0111] The macroscopic physical pattern features (such as class center vectors) obtained by SCI clustering inference are fused in parallel with the local spatiotemporal features (such as feature maps output by convolutional layers) obtained by CNN (such as concatenation, weighted summation or attention fusion) to form a more comprehensive feature representation (such as a joint feature vector of "macroscopic pattern + local details"), avoiding the loss of information from a single feature (such as using only macroscopic patterns may ignore local rainstorm trigger points, and using only local features may ignore the influence of the overall pattern).

[0112] The parallel fusion of features simultaneously incorporates the physical meaning (macroscopic pattern) of SCI clustering inference and the data details (local features) of CNN, ensuring interpretability while improving feature expressiveness (better predicting rainstorms).

[0113] Transformer's self-attention mechanism can capture the dependencies between features at different spatiotemporal locations (such as the relationship between "convergence of water vapor transport belt in a certain area at the current moment" and "precipitation in the downstream area in the next 24 hours"; the relationship between "local features at the bend of the transport belt" and "overall pattern").

[0114] For forecasting precipitation over a 24-hour or 72-hour period, Transformer can effectively process time-series data (such as water vapor characteristics over the past 6 hours → precipitation over the next 24 hours) and capture temporal evolution patterns (such as how changes in water vapor flux over time affect the probability of precipitation).

[0115] The features obtained after parallel fusion contain information on both macroscopic and local dimensions, as well as physical and data aspects. However, modeling spatiotemporal relationships is required to predict precipitation (e.g., how current water vapor transport belt convergence leads to future heavy rainfall). The self-attention mechanism of the Transformer Encoder excels at handling such spatiotemporal dependencies, thus enabling effective precipitation prediction, such as 24-hour or 72-hour heavy rainfall prediction.

[0116] This application provides a method for predicting heavy rainfall, including acquiring multi-source meteorological data; extracting basic variables from the multi-source meteorological data; identifying a water vapor transport belt based on the basic variables; acquiring the trajectory of the water vapor transport belt based on the water vapor transport belt; acquiring geometric and dynamic characteristic indicators of the water vapor transport belt based on the water vapor transport belt and its trajectory; and using the basic variables, geometric and dynamic characteristic indicators as input data, and performing heavy rainfall prediction based on a machine learning prediction model. This method achieves heavy rainfall prediction, ensuring accuracy and timeliness while reducing reliance on human experience, providing intelligent technical support for heavy rainfall monitoring and early warning. This method combines multi-source meteorological data with machine learning technology, starting from the perspective of the water vapor transport belt, to deeply explore the key characteristics of heavy rainfall occurrence and development, achieving heavy rainfall prediction and providing strong support for the scientific prevention and control of meteorological disasters.

[0117] In addition, when identifying the water vapor transport belt, this application embodiment simultaneously selects two isobaric surfaces, one at ground level and the other at 850 hPa and the other at 300 hPa, for analysis to comprehensively capture the water vapor transport characteristics in the lower and upper atmospheres and improve the accuracy of rainstorm forecasting.

[0118] Equally important, the input data for the machine learning prediction model in this application includes geometric and dynamic characteristic indicators of the water vapor transport belt. These indicators reflect the morphology, intensity, and motion characteristics of the water vapor transport belt, thereby providing a more comprehensive characterization of the key physical processes involved in rainstorms. By introducing these indicators, the model can effectively capture the spatiotemporal evolution characteristics of rainstorm events, further improving prediction accuracy.

[0119] Furthermore, the embodiments of this application integrate basic variables and characteristic indicators of the water vapor transport belt to construct multi-dimensional feature input data, which not only enhances the model's ability to represent rainstorm characteristics, but also improves its adaptability to changes in complex weather systems.

[0120] Example 2 Please see Figure 2 , Figure 2This is a schematic diagram of a rainstorm prediction device provided in an embodiment of this application. The rainstorm prediction device 40 includes: a first acquisition module 41 for acquiring multi-source meteorological data; an extraction module 42 for extracting basic variables based on the multi-source meteorological data; an identification module 43 for identifying a water vapor conveyor belt based on the basic variables; a second acquisition module 44 for acquiring the trajectory of the water vapor conveyor belt based on the water vapor conveyor belt; a third acquisition module 45 for acquiring the geometric and dynamic characteristic indicators of the water vapor conveyor belt based on the water vapor conveyor belt and its trajectory; and a rainstorm prediction module 46 for using the basic variables, the geometric and dynamic characteristic indicators as input data and performing rainstorm prediction based on a machine learning prediction model. This rainstorm prediction device 40 achieves rainstorm prediction, ensuring the accuracy and timeliness of rainstorm prediction, while reducing reliance on human experience, providing intelligent technical support for rainstorm monitoring and early warning. The technical effects of this rainstorm prediction device 40 are consistent with the aforementioned method embodiments, and will not be repeated here.

[0121] Example 3 Please see Figure 3 , Figure 3 This is a schematic diagram of the hardware structure of an electronic device 50 for performing a rainstorm prediction method according to an embodiment of this application. The electronic device 50 includes one or more processors 51 and a memory 52. Figure 3 Let's take a memory as an example.

[0122] The processor 51 and the memory 52 can be connected via a bus or other means. In this embodiment, the connection via a bus is taken as an example.

[0123] The memory 52, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the rainstorm prediction method in the embodiments of this application. The processor 51 executes various functional applications and data processing of the rainstorm prediction device by running the non-volatile software programs, instructions, and modules stored in the memory 52, thereby implementing the rainstorm prediction method of the above-described method embodiments.

[0124] The memory 52 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the rainstorm forecasting device. Furthermore, the memory 52 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 52 may optionally include memory remotely located relative to the processor 51, and this remote memory may be connected to the rainstorm forecasting device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0125] The one or more modules are stored in the memory 52, and when executed by the one or more processors 51, they execute the rainstorm prediction method in any of the above method embodiments.

[0126] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0127] This application provides a non-volatile computer-readable storage medium storing computer-executable instructions, which are executed by an electronic device using the rainstorm prediction method in any of the above method embodiments.

[0128] This application provides a computer program product, including a computing program stored on a non-volatile computer-readable storage medium. The computer program includes program instructions, which, when executed by a computer, cause the computer to perform the rainstorm prediction method in any of the above method embodiments.

[0129] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0130] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software and a general-purpose hardware platform, or it can be implemented using hardware. Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and not to limit them; under the concept of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of this application as described above, which are not provided in detail for the sake of brevity; although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method of predicting a storm, characterized by, The method comprises the following steps: acquiring multi-source meteorological data; extracting basic variables according to the multi-source meteorological data; identifying a water vapor transport belt according to the basic variables; acquiring a water vapor transport belt trajectory according to the water vapor transport belt; acquiring geometric characteristic indexes and dynamic characteristic indexes of the water vapor transport belt according to the water vapor transport belt and the water vapor transport belt trajectory; performing rainstorm prediction according to a machine learning prediction model by taking the basic variables, the geometric characteristic indexes and the dynamic characteristic indexes as input data.

2. The method of claim 1, wherein, The input data further comprises climate factors, and the climate factors comprise monsoon indexes, El Nino phenomenon and Southern Oscillation Index.

3. The method of claim 1, wherein, The method for extracting basic variables according to the multi-source meteorological data comprises the following steps: extracting specific humidity, potential height, meridional wind component, zonal wind component and surface air pressure according to the multi-source meteorological data; calculating water vapor content, water vapor flux and water vapor flux divergence according to the specific humidity, the meridional wind component and the zonal wind component.

4. The method of claim 3, wherein, The method for calculating water vapor content, water vapor flux and water vapor flux divergence according to the specific humidity, the meridional wind component and the zonal wind component comprises the following steps: respectively calculating water vapor content, water vapor flux and water vapor flux divergence of the atmospheric layer from the surface to the 850 hPa isobaric surface, the atmospheric layer from the surface to the 300 hPa isobaric surface and the atmospheric layer from the 850 hPa isobaric surface to the 300 hPa isobaric surface according to the specific humidity, the meridional wind component and the zonal wind component.

5. The method according to any one of claims 1 to 4, characterized in that, The method for identifying a water vapor transport belt according to the basic variables comprises the following steps: integrating and calculating vertical water vapor flux at a grid level according to a PanLu2.0 identification algorithm; setting local and regional water vapor flux double thresholds; taking a grid where a maximum value of the vertical water vapor flux is located as a search starting point; searching for target grids where a neighborhood vertical water vapor flux exceeds the double thresholds according to the search starting point, and adding the target grids into the water vapor transport belt; taking the target grids as the search starting point, and performing the step of searching for target grids where a neighborhood vertical water vapor flux exceeds the double thresholds according to the search starting point, and adding the target grids into the water vapor transport belt, until the water vapor transport belt is determined.

6. The method of claim 5, wherein, The method for identifying a water vapor transport belt according to the basic variables further comprises the following steps: removing closed areas in the determined water vapor transport belt; performing the step of taking a grid where a maximum value of the vertical water vapor flux is located as a search starting point in the remaining water vapor transport belt.

7. The method according to any one of claims 1 to 4, characterized in that, The method for acquiring a water vapor transport belt trajectory according to the water vapor transport belt comprises the following steps: integrating and calculating vertical water vapor flux at a grid level according to a PanLu2.0 identification algorithm; taking a grid where a maximum value of the vertical water vapor flux is located as a reference point; extracting a first adjacent area with a preset geographical distance from the reference point as a reference; calculating a first weighted water vapor flux direction of the first adjacent area according to the vertical water vapor flux; taking a plane perpendicular to the first weighted water vapor flux direction as a cross section of the reference point; The weighted center of gravity of the section is calculated, and a second adjacent area is extracted based on a preset geographical distance from the weighted center of gravity; and a second weighted water vapor flux direction of the second adjacent area is calculated according to the vertical water vapor flux; A point obtained by moving a preset distance along the second weighted water vapor flux direction is taken as the reference point, and the step of extracting the first adjacent area based on a preset geographical distance from the reference point is performed until the water vapor transport belt is covered, to obtain a forward trajectory of the water vapor transport belt; A point obtained by moving a preset distance along a direction opposite to the second weighted water vapor flux direction is taken as the reference point, and the step of extracting the first adjacent area based on a preset geographical distance from the reference point is performed until the water vapor transport belt is covered, to obtain a reverse trajectory of the water vapor transport belt; The forward trajectory of the water vapor transport belt and the reverse trajectory of the water vapor transport belt are integrated to obtain the water vapor transport belt trajectory.

8. The method according to any one of claims 1 to 4, characterized in that, The geometric feature indicators include length, width, length-width ratio, bending degree, and rotation angle of the water vapor transport belt; and the dynamic feature indicators include average transport speed, residence time, and source-sink area distribution.

9. A storm prediction device characterized by comprising: Comprise: The first acquisition module is configured to acquire multi-source meteorological data; The extraction module is configured to extract basic variables according to the multi-source meteorological data; The identification module is configured to identify a water vapor transport belt according to the basic variables; The second acquisition module is configured to acquire a water vapor transport belt trajectory according to the water vapor transport belt; The third acquisition module is configured to acquire geometric feature indicators and dynamic feature indicators of the water vapor transport belt according to the water vapor transport belt and the water vapor transport belt trajectory; The rainstorm prediction module is configured to take the basic variables, the geometric feature indicators, and the dynamic feature indicators as input data, and perform rainstorm prediction according to a machine learning prediction model.

10. A non-transitory computer readable storage medium, comprising: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by the electronic device, the electronic device performs the method in any one of claims 1-8. The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by the electronic device, the electronic device performs the method in any one of claims 1-8.

Images