A short-term heavy rainfall early warning management method based on meteorological indicators

Abstract

The present invention discloses a short-term heavy rainfall early warning management method based on meteorological indicators, comprising the following steps: S1, collecting local meteorological data, establishing a meteorological data set, and classifying and preprocessing the meteorological data set; S2, determining multiple types of forecast indicators for local short-term heavy rainfall weather by performing feature analysis on the meteorological data set, and determining the weight of each type of forecast indicator; S3, based on a recurrent neural network, obtaining a neural network model for objective forecasting of short-term heavy rainfall by training the meteorological data set; S4, when performing a short-term heavy rainfall early warning, collecting real-time meteorological data and inputting the data into the neural network model, in which, for the input meteorological data, determining its matching score with each type of forecast indicator, then performing weighted addition of all matching scores to obtain a comprehensive score, and then obtaining a predicted probability of short-term heavy rainfall based on the comprehensive score.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological forecasting, and in particular to a short-term heavy rainfall early warning management method based on meteorological indicators. Background Art

[0002] Short-term heavy rainfall is a major meteorological disaster and a major challenge in weather forecasting. This phenomenon is characterized by its short duration, intense rainfall, and strong localized nature. The accumulated heavy rainfall can quickly lead to flash floods, causing waterlogging in farmland and cities, and even triggering geological disasters such as mudslides and flash floods, resulting in significant economic losses and casualties. This severe disruption severely impacts urban life and transportation, potentially causing serious hazards such as mudslides and floods, posing a significant threat to life and property. Therefore, strengthening research on its mechanisms and strengthening near-term monitoring and early warning are crucial. However, due to the suddenness, localized nature, uncertainty, randomness, and short duration of short-term heavy rainfall, forecasting has long been less than ideal. Summary of the Invention

[0003] In order to solve the problems raised in the above background technology, the technical solution adopted by the present invention is:

[0004] A short-term heavy rainfall early warning management method based on meteorological indicators includes the following steps:

[0005] S1. Collect local meteorological data, establish a meteorological dataset, and classify and preprocess the meteorological dataset;

[0006] S2. By analyzing the characteristics of the meteorological data set, determine multiple types of forecast indicators for local short-term heavy rainfall weather and determine the weight of each type of forecast indicator;

[0007] S3. Based on the recurrent neural network, a neural network model for objective forecasting of short-term heavy rainfall is obtained through training with meteorological datasets;

[0008] S4. When issuing a short-term heavy rainfall warning, real-time meteorological data is collected and input into a neural network model. In the neural network model, for the input meteorological data, its matching score with each type of forecast indicator is determined, and then all matching scores are weighted and added together to obtain a comprehensive score. The predicted probability of short-term heavy rainfall is then obtained based on the comprehensive score.

[0009] In some embodiments, in step S1 , the collected meteorological data includes micaps data, automatic weather station data, and Doppler radar product data.

[0010] In some embodiments, in step S1, local meteorological data is collected to establish a meteorological dataset, specifically:

[0011] Randomly sample meteorological data from different periods and stations over several years;

[0012] In addition, the short-term heavy rainfall weather processes that occurred in the local area over several years were collected, and the regularity, regional and temporal characteristics of the occurrence of short-term heavy rainfall were preliminarily analyzed. Based on the stations and time of the short-term heavy rainfall, meteorological data for the period from the preset time before the short-term heavy rainfall began to the end of the short-term heavy rainfall were obtained;

[0013] A meteorological data set is established based on the collected meteorological data, and in the meteorological data set, different types of meteorological data in the same period are temporally associated.

[0014] In some embodiments, in step S1, when classifying the meteorological dataset, specifically, the data in the meteorological dataset is classified according to different types, sites, and times;

[0015] When preprocessing the meteorological dataset, the size, scale, and center position of the images in the dataset are adjusted, and grayscale processing is performed.

[0016] In some embodiments, in step S2, when performing feature analysis on a meteorological data set, a feature detection algorithm is used by combining computer processing with manual processing to extract feature points or feature areas from the image to assist in image matching, recognition, and edge detection tasks, thereby finding the small and medium-scale weather systems that cause local short-term heavy rainfall, and classifying them. The meteorological data features corresponding to the small and medium-scale weather systems of different classifications are determined as forecast indicators, and the weight of each type of forecast indicator is determined. In addition, for each type of forecast indicator at each site, several groups of typical meteorological data of the site are selected as evaluation criteria.

[0017] In some embodiments, in step S3, a neural network model corresponding to each site is established respectively, and when establishing the neural network model, a neural network model combining CNN and LSTM is adopted. During model training, different types of meteorological data of the site at the same time period are input, and characteristic information of different types of meteorological data is extracted respectively through the attention mechanism, and matched with the typical meteorological data corresponding to each type of preset forecast indicator to obtain a matching score, and then all matching scores are weightedly added to obtain a comprehensive score, and then the probability of short-term heavy rainfall is obtained based on the comprehensive score.

[0018] In some embodiments, after step S3 and before step S4, a k-fold cross-validation method is also used to verify the accuracy of the neural network model. If the accuracy is greater than a preset threshold, step S4 is entered; otherwise, step S2 is returned to, and the forecast indicators, the weights of the forecast indicators and / or the model parameters are adjusted, and the model training is re-performed until the accuracy reaches the preset threshold.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] The short-term heavy rainfall early warning management method based on meteorological indicators provided by the present invention can improve the accuracy and timeliness of short-term and near-term forecasts of short-term heavy rainfall, which is beneficial to protecting food production, and has very important practical and long-term significance for improving agricultural and industrial production, protecting people's lives and property, and promoting local economic development and social stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A schematic diagram of the process of the short-term heavy rainfall early warning management method based on meteorological indicators provided by the present invention;

[0022] Figure 2-Figure 10 It is a meteorological data diagram in a specific embodiment. DETAILED DESCRIPTION

[0023] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the following further describes how the present invention is implemented in conjunction with the accompanying drawings and specific implementation methods.

[0024] Reference Figure 1 As shown, the present invention provides a short-term heavy rainfall early warning management method based on meteorological indicators, comprising the following steps:

[0025] S1. Collect local meteorological data, establish a meteorological dataset, and classify and preprocess the meteorological dataset.

[0026] S2. By analyzing the characteristics of the meteorological data set, multiple forecast indicators for local short-term heavy rainfall weather are determined, and the weight of each forecast indicator is determined.

[0027] S3. Based on the recurrent neural network, a neural network model for objective forecasting of short-term heavy rainfall is obtained through training with meteorological datasets.

[0028] S4. When issuing a short-term heavy rainfall warning, real-time meteorological data is collected and input into a neural network model. In the neural network model, for the input meteorological data, its matching score with each type of forecast indicator is determined, and then all matching scores are weighted and added together to obtain a comprehensive score. The predicted probability of short-term heavy rainfall is then obtained based on the comprehensive score.

[0029] It is understandable that short-term heavy rainfall warnings are highly regional, so data needs to be collected separately for different regions, and the judgment criteria may also be different. The following takes the situation in Tongliao, Inner Mongolia as an example to introduce a specific embodiment of the present invention.

[0030] In step S1, local meteorological data is collected, a meteorological data set is established, and the meteorological data set is classified and preprocessed.

[0031] In this embodiment, in step S1, meteorological data of Tongliao area in Inner Mongolia for several years (2008-2023 in this embodiment) are extracted. The collected meteorological data include micaps data, automatic weather station data and Doppler radar product data.

[0032] In this embodiment, when collecting meteorological data, meteorological data from different time periods and different sites over several years are randomly sampled; in addition, short-term heavy rainfall weather processes that occurred in the local area over several years are collected, and the regularity, regional, and temporal characteristics of the occurrence of short-term heavy rainfall are preliminarily analyzed. Based on the site and time where the short-term heavy rainfall occurred, meteorological data for the period from a preset time (e.g., 24 hours) before the start of the short-term heavy rainfall to the end of the short-term heavy rainfall is obtained; a meteorological data set is established based on the collected meteorological data, and in the meteorological data set, different types of meteorological data in the same time period are temporally associated.

[0033] When collecting data on short-term heavy rainfall events, we searched and screened hourly precipitation data to identify any such events. For a preliminary analysis of the patterns, geographic locations, and temporal characteristics of these events, we can analyze and summarize them from the perspectives of spatial distribution, interannual variation, and diurnal variation, as detailed below.

[0034] (1) Spatial distribution

[0035] Analysis of the frequency of short-duration heavy rainfall across various regions reveals that the highest number of such events occurs in Zhalute Banner in the northern part of Tongliao Prefecture, Inner Mongolia, followed by Naiman Banner (including Qinglongshan Station), then Kezuozhong Banner and Kulun Banner, with the lowest number occurring in Horqin District and Holingol City. In Tongliao, short-duration heavy rainfall events are primarily concentrated in the mountainous areas of the north and south, with relatively few in the plains. Holingol City experiences relatively few such events due to its smaller area than other banners and counties and its relatively small number of regional automatic rainfall stations. The distribution of short-duration heavy rainfall across the city is closely linked to the region's terrain, which encompasses rocky hills and mountains along the southern foothills of the Greater Khingan Mountains in the north, shallow mountains and loess hills along the edge of the western Liaoning Mountains in the south, and the sandy alluvial plains of the Xiliao River basin in the center. (Tongliao City's terrain is high in the south and north, low and flat in the center, forming a saddle-shaped pattern.) Dynamic uplift and uneven thermal convection in the mountainous areas are key factors in the frequent occurrence of such heavy rainfall.

[0036] (2) Interannual variation

[0037] If one or more of the national and regional stations in Tongliao experience short-term heavy rainfall on a single day, this is considered a short-term heavy rainfall day. The annual average number of short-term heavy rainfall days in the years studied was 15.2. If a single station experiences short-term heavy rainfall on a single day is counted as a station-day, the annual average is 40.6 station-days. The study found that while the number of short-term heavy rainfall days and station-days decreased in 2009 and 2010, they have increased since 2011, with the increase in station-days being more pronounced than that of short-term heavy rainfall days. This is partly due to the deployment of numerous new regional automatic stations in Tongliao in recent years, resulting in more refined spatial capabilities for detecting short-term heavy rainfall. It also reflects the increasing frequency of extreme weather and climate events as global warming continues, expanding the scope of impact of short-term heavy rainfall events.

[0038] (3) Seasonal changes

[0039] Tongliao experiences short bursts of heavy rainfall from April to October, but they are most concentrated from June to September, accounting for 98% of all days. This percentage is concentrated in June and July, accounting for 65%. The earliest occurrence is on April 23rd, and the latest on October 10th. July is the most frequent month for heavy rainfall, followed by June. The peak of heavy rainfall in June and July precedes the peak of rainfall in Tongliao (July and August). This phenomenon is closely related to the characteristics of heavy rainfall and the seasonal advance and retreat of the subtropical high pressure.

[0040] (4) Diurnal variation

[0041] Data from all Tongliao observation stations show that short bursts of heavy rainfall occurred at all times, primarily between 11:00 and 23:00, accounting for 76% of the total. Short bursts of heavy rainfall occurred between 12:00 and 20:00, accounting for 64% of the total. This indicates that short bursts of heavy rainfall in Tongliao are highly concentrated in the afternoon and evening. This suggests that during the afternoon and evening periods, when daily temperature fluctuations are greatest, turbulence-induced atmospheric stratification instability is more likely to lead to severe convective weather. The occurrence of short bursts of heavy rainfall is closely positively correlated with the occurrence of severe convective weather.

[0042] In addition, when classifying the meteorological dataset, specifically, the data in the meteorological dataset is classified according to different types, sites and times; when preprocessing the meteorological dataset, the size, scale, and center position of the images in the dataset are adjusted, and grayscale processing is performed to facilitate the subsequent establishment and training of the recurrent neural network.

[0043] In step S2, by performing feature analysis on the meteorological data set, multiple types of forecast indicators for local short-term heavy rainfall weather are determined, and the weight of each type of forecast indicator is determined.

[0044] In this embodiment, when performing feature analysis on a meteorological data set, a feature detection algorithm is used by combining computer processing with manual processing to extract feature points or areas of certain importance or representativeness from the image to assist in tasks such as image matching, recognition, and edge detection. The small and medium-scale weather systems that cause local short-term heavy rainfall are identified and classified. The meteorological data features corresponding to the small and medium-scale weather systems of different classifications are determined as forecast indicators, and the weight of each type of forecast indicator is determined. In addition, for each type of forecast indicator at each site, several groups of typical meteorological data of the site are selected as evaluation criteria.

[0045] In this embodiment, they are classified into: mesoscale convergence shear line, ground mesoscale convergence line, mesoscale low-vortex circulation, low-level jet and ultra-low-level jet, thunderstorm cell, thunderstorm group (including squall line), gust front, mesocyclone or mesoscale convergence zone (headwind zone), etc.

[0046] The vast majority of short-term heavy rainfall in the Tongliao region is directly caused by small and medium-scale systems embedded within synoptic-scale systems, within a specific large-scale circulation pattern. Automatic weather station intensified wind data and Doppler radar products offer higher temporal and spatial resolution, enabling better monitoring of the activity of these small and medium-scale systems. Using these data, we have developed a classification of the small and medium-scale systems responsible for these short-term heavy rainfall events:

[0047] (1) Mesoscale convergence shear line

[0048] Long-term convergence of small and medium-scale winds enhances surface convergence lift and is a mesoscale system and triggering condition for localized heavy rainfall. During short bursts of heavy rainfall in the Tongliao region, wind discontinuities or points of maximum cyclonic curvature, known as convergence lines, are often observed in the lower and middle troposphere. In addition to convergent shear in wind direction, convergence in wind speed is also present.

[0049] (2) Ground mesoscale convergence lines

[0050] Using the hourly wind field data from encrypted automatic stations, we analyzed the effect of ground wind field on short-term heavy rainfall. The results showed that during the short-term heavy rainfall, the ground was always affected by mesoscale convergence lines with a length of tens to 200 km. The convergence lines showed both wind direction and wind speed convergence, which corresponded well to the heavy rainfall area.

[0051] (3) Mesoscale low-vortex circulation

[0052] During the short-term heavy rainfall in Tongliao area, mesoscale low-vortex circulation is often analyzed in the wind field in the middle and lower layers of the troposphere.

[0053] (4) Low-level jets and ultra-low-level jets

[0054] During the short-term heavy rainfall in Tongliao area, low-level jets and ultra-low-level jets are often analyzed in the lower troposphere wind field.

[0055] (5) Thunderstorm cell

[0056] like Figure 2 As shown in the figure, it is the reflectivity factor map at 14:07 on July 5, 2019, corresponding to the short-term heavy rainfall process in Zhalute Banner; Figure 3 As shown, this is the reflectivity factor map at 11:44 on August 21, 2015, corresponding to the short-term heavy rainfall process in Zhalute Banner.

[0057] (6) Thunderstorm groups (including squall lines)

[0058] A thunderstorm cluster is composed of multiple individual thunderstorms, such as Figure 4 As shown in the figure, it is the reflectivity factor map at 12:40 on August 20, 2015, corresponding to the short-term heavy rainfall process in Zhalute Banner; sometimes multiple thunderstorm cells are arranged linearly to form a squall line system, such as Figure 5 As shown, this is the reflectivity factor map at 21:42 on July 25, 2019, corresponding to the short-term heavy rainfall process in Kezuohou Banner.

[0059] (7) Gust front

[0060] In many cases, strong echoes are accompanied by gust fronts, such as Figure 6 As shown in the figure, it is the reflectivity factor map at 21:58 on July 25, 2019, corresponding to the short-term heavy rainfall process in Kezuohou Banner; Figure 7 As shown, this is the reflectivity factor map at 17:10 on July 17, 2010, corresponding to the short-term heavy rainfall process in Kezuozhong Banner.

[0061] (8) Mesocyclone or mesoscale convergence zone (headwind zone)

[0062] Sometimes, obvious and strong convergence field features can be detected on the radial velocity map, reaching the characteristics of "mesocyclone", such as: Figure 8 As shown in the figure, it is the radial velocity map at 13:09 on August 1, 2011, corresponding to the short-term heavy rainfall process in Kezuozhong Banner; sometimes there will be "headwind areas", such as Figure 9 As shown in Figure 1, it is the radial velocity map at 18:18 on July 22, 2009, corresponding to the short-term heavy rainfall process in Naiman Banner; Figure 10 As shown, this is the radial velocity map at 12:02 on July 5, 2019, corresponding to the short-term heavy rainfall process in Kezuohou Banner.

[0063] The above images of different classifications can be stored as typical meteorological data after preprocessing and used as subsequent evaluation criteria.

[0064] In step S3, based on a recurrent neural network, a neural network model for objective forecasting of short-term heavy rainfall is obtained through training of a meteorological dataset.

[0065] In this embodiment, a neural network model corresponding to each site is established respectively, and when establishing the neural network model, a CNN combined with LSTM neural network model can be used. Moreover, during model training, different types of meteorological data of the site at the same time period are input, and the characteristic information of different types of meteorological data is extracted respectively through the attention mechanism, and matched with the typical meteorological data corresponding to each type of preset forecast indicator to obtain a matching score, and then all matching scores are weightedly added to obtain a comprehensive score, and then the probability of short-term heavy rainfall is obtained based on the comprehensive score.

[0066] In this embodiment, after step S3 and before step S4, the k-fold cross-validation method is also used to verify the accuracy of the neural network model. If the accuracy is greater than the preset threshold, step S4 is entered; otherwise, step S2 is returned to adjust the forecast indicators, the weights of the forecast indicators and / or the model parameters, and the model training is re-performed until the accuracy reaches the preset threshold.

[0067] When verifying the accuracy of the neural network model, the verification criteria are as follows: In each k-fold cross-validation run, for different types of meteorological data input over the same time period: if the meteorological data corresponds to an actual occurrence of short-term heavy rainfall, and the model calculates a probability of short-term heavy rainfall greater than 70%, the model is considered to have accurately identified the short-term heavy rainfall; otherwise, the model is considered to have missed the prediction. If the meteorological data does not correspond to an actual occurrence of short-term heavy rainfall, and the model calculates a probability of short-term heavy rainfall greater than 70%, the model is considered to have misjudged the prediction. When both the missed and misjudgment rates are less than a preset threshold (e.g., 10%), the model is considered to have passed the accuracy test.

[0068] In this embodiment, in step S4, when issuing a short-term heavy rainfall warning, real-time meteorological data is collected and input into the neural network model corresponding to the site. In the neural network model, for the input meteorological data, its matching score with each type of forecast indicator is determined, and then all matching scores are weighted and added together to obtain a comprehensive score, and then the predicted probability of short-term heavy rainfall is obtained based on the comprehensive score.

[0069] In summary, the short-term heavy rainfall warning management method based on meteorological indicators provided by the present invention can improve the accuracy and timeliness of short-term and near-term forecasts of short-term heavy rainfall, which is beneficial to protecting food production, and has very important practical and long-term significance for improving agricultural and industrial production, protecting people’s lives and property, and promoting local economic development and social stability.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions of the present invention, which should all be covered by the scope of the claims of the present invention.

Claims

1. A short-term heavy rainfall early warning management method based on meteorological indicators, characterized in that: The steps include: S1. Collect local meteorological data, establish a meteorological dataset, and classify and preprocess the meteorological dataset; S2. By analyzing the characteristics of the meteorological data set, determine multiple types of forecast indicators for local short-term heavy rainfall weather and determine the weight of each type of forecast indicator; In step S2, when performing feature analysis on the meteorological data set, a feature detection algorithm is used to extract feature points or feature areas from the image through a combination of computer and manual processing to assist in image matching, recognition, and edge detection tasks, thereby identifying and classifying the small and medium-scale weather systems that cause local short-term heavy rainfall. The meteorological data features corresponding to the small and medium-scale weather systems of different classifications are determined as forecast indicators, and the weight of each type of forecast indicator is determined. In addition, for each type of forecast indicator at each station, several groups of typical meteorological data at the station are selected as evaluation criteria; The classification of small and medium-scale weather systems includes: mesoscale convergence shear line, surface mesoscale convergence line, mesoscale low-vortex circulation, low-level jet and ultra-low-level jet, thunderstorm cell, thunderstorm group, gust front, mesocyclone or mesoscale convergence area; S3. Based on the recurrent neural network, a neural network model for objective forecasting of short-term heavy rainfall is obtained through training with meteorological datasets; S4. When issuing a short-term heavy rainfall warning, real-time meteorological data is collected and input into a neural network model. In the neural network model, for the input meteorological data, its matching score with each type of forecast indicator is determined, and then all matching scores are weighted and added together to obtain a comprehensive score. The predicted probability of short-term heavy rainfall is then obtained based on the comprehensive score.

2. The short-term heavy rainfall early warning management method based on meteorological indicators according to claim 1 is characterized in that: In step S1 , the collected meteorological data include micaps data, automatic weather station data, and Doppler radar product data.

3. The short-term heavy rainfall early warning management method based on meteorological indicators according to claim 2 is characterized in that: In step S1, local meteorological data is collected to establish a meteorological dataset. Specifically: Randomly sample meteorological data from different periods and stations over several years; In addition, the short-term heavy rainfall weather processes that occurred in the local area over several years were collected, and the regularity, regional and temporal characteristics of the occurrence of short-term heavy rainfall were preliminarily analyzed. Based on the stations and time of the short-term heavy rainfall, meteorological data for the period from the preset time before the short-term heavy rainfall began to the end of the short-term heavy rainfall were obtained; A meteorological data set is established based on the collected meteorological data, and in the meteorological data set, different types of meteorological data in the same period are temporally associated.

4. The short-term heavy rainfall early warning management method based on meteorological indicators according to claim 3 is characterized in that: In step S1, when classifying the meteorological data set, specifically, the data in the meteorological data set is classified according to different types, sites and times; When preprocessing the meteorological dataset, the size, scale, and center position of the images in the dataset are adjusted, and grayscale processing is performed.

5. The short-term heavy rainfall early warning management method based on meteorological indicators according to claim 4 is characterized in that: In step S3, a neural network model corresponding to each site is established respectively, and when establishing the neural network model, a CNN combined with LSTM neural network model is adopted. During model training, different types of meteorological data of the site at the same time period are input, and the characteristic information of different types of meteorological data is extracted respectively through the attention mechanism, and matched with the typical meteorological data corresponding to each type of preset forecast indicator to obtain a matching score, and then all matching scores are weightedly added to obtain a comprehensive score, and then the probability of short-term heavy rainfall is obtained according to the comprehensive score.

6. The short-term heavy rainfall early warning management method based on meteorological indicators according to claim 5 is characterized in that: After step S3 and before step S4, the k-fold cross-validation method is also used to verify the accuracy of the neural network model. If the accuracy is greater than the preset threshold, step S4 is entered; otherwise, step S2 is returned to adjust the forecast indicators, the weights of the forecast indicators and / or the model parameters, and the model is retrained until the accuracy reaches the preset threshold.

Images