Meteorology-based real-time monitoring system for extreme climate risks

Abstract

The application provides a meteorology-based extreme climate risk real-time monitoring system and relates to the technical field of meteorology.The method comprises a calibration module used for constructing an analysis loop based on the time evolution relationship of representative feature points, analyzing the contribution of element space variation and transport process by calculating the movement speed of meteorological elements in the feature influence domain relative to the regional background field, quantifying the dynamic effect of movement on the evolution of extreme climate, calculating meteorological data calibration quantity, dynamically calibrating the standardized meteorological data sequence, obtaining the calibrated meteorological element sequence, and generating an extreme climate index.The application realizes dynamic identification, quantitative evaluation and multi-level early warning of extreme climate events, and improves the real-time perception ability and emergency response efficiency of meteorological disaster risks.

Description

Technical Field

[0001] This invention relates to the field of meteorological technology, and in particular to a real-time monitoring system for extreme climate risks based on meteorology. Background Technology

[0002] In the field of meteorological monitoring, real-time and accurate risk assessment and early warning of extreme weather events are crucial for disaster prevention and mitigation. Most existing risk monitoring systems rely on standardizing meteorological data sequences from fixed-point observations and setting fixed thresholds based on historical statistics or model simulations to trigger early warnings. However, such methods may have some limitations in timeliness and accuracy when dealing with extreme weather systems that have rapid evolution and complex spatial transport characteristics.

[0003] Specifically, existing technical solutions may fail to fully analyze the dynamic changes and spatial interactions within meteorological element fields at the data processing and analysis level. For example, when monitoring a regional extreme rainstorm triggered by a strong convective system, most existing systems rely on real-time data sequences of rainfall, wind speed, etc., from one or more stations and compare them with preset static thresholds. This method reflects the instantaneous values ​​of local meteorological conditions more, but it lacks effective quantification and integrated analysis of key dynamic factors that drive the occurrence and development of extreme weather, such as the spatial variation of water vapor transport flux, the movement of strong wind centers, and their advection contribution to unstable energy. Because they fail to fully consider the dynamic contribution of the spatial movement of meteorological elements, existing systems may struggle to detect extreme weather systems in their early stages. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a meteorology-based real-time monitoring system for extreme climate risks, which enables dynamic identification, quantitative assessment and multi-level early warning of extreme climate events, thereby improving the real-time perception capability of meteorological disaster risks and the efficiency of emergency response.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] The first aspect is a real-time monitoring system for extreme climate risks based on meteorology, including:

[0007] The generation module processes real-time collected meteorological and environmental data to form standardized meteorological data sequences and set benchmark feature points, generating two independent associated feature sets; it determines the feature influence domain based on the relationship between the two associated feature sets, and selects a representative feature point within the feature influence domain.

[0008] The calibration module is used to construct an analysis loop based on the temporal evolution relationship of representative feature points. By calculating the motion velocity of meteorological elements in the feature influence domain relative to the regional background field, it analyzes the contribution of spatial variation and transport process of elements, quantifies the dynamic effect of motion on extreme climate evolution, calculates the calibration amount of meteorological data, performs dynamic calibration on the standardized meteorological data sequence, obtains the calibrated meteorological element sequence, and generates extreme climate indicators.

[0009] The setting module is used to analyze the risk level of extreme climate events based on extreme climate indicators, combined with the underlying surface feature data and disaster record data of the target area, through a dynamic risk assessment model, and to set multi-level early warning thresholds according to the preset climate state classification standards.

[0010] The push module is used to compare real-time monitored meteorological data with multi-level early warning thresholds through the cloud platform. When the real-time monitored data exceeds the multi-level early warning thresholds, it automatically generates and pushes early warning information to the designated terminal.

[0011] The monitoring module is used to build a monitoring framework based on cloud-native architecture and component-based design, and to use early warning information to achieve distributed collaborative operation and dynamic monitoring.

[0012] In a second aspect, a computer-readable storage medium storing a program that, when executed by a processor, implements the system.

[0013] The above-described solution of the present invention has at least the following beneficial effects:

[0014] By standardizing real-time meteorological and environmental data, an independent and correlated feature set is constructed, the feature influence domain is delineated, and representative feature points are selected. This enables the screening and refinement of massive heterogeneous meteorological data. Based on the temporal evolution relationship of representative feature points, an analysis loop is constructed. Combined with the regional background field, the dynamic effect of meteorological element movement on extreme climate is quantified, and the standardized data sequence is dynamically calibrated to improve the accuracy and reliability of extreme climate indicators. Combining the underlying surface characteristics of the target area, disaster record data, and dynamic risk assessment models, multi-level early warning thresholds are constructed, making the risk level classification more consistent with the actual climate impact patterns of the region. Relying on the cloud platform, real-time data is compared with early warning thresholds, and early warning information can be automatically pushed when risks are triggered, shortening the early warning response time. Based on cloud-native architecture and component-based design, a monitoring framework is constructed to achieve distributed collaborative operation and dynamic monitoring, improving the system's ability to process massive amounts of data and its anti-interference ability, and reducing system operation and maintenance costs. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a meteorological-based real-time monitoring system for extreme climate risks, provided in an embodiment of the present invention.

[0016] Figure 2 This is a schematic diagram of the process provided by an embodiment of the present invention, which compares real-time monitored meteorological data with multi-level early warning thresholds through a cloud platform, and automatically generates and pushes early warning information to a designated terminal when the real-time monitored data exceeds the multi-level early warning thresholds. Detailed Implementation

[0017] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0018] like Figure 1 As shown, embodiments of the present invention propose a meteorological-based real-time monitoring system for extreme climate risks, comprising:

[0019] The generation module processes real-time collected meteorological and environmental data to form standardized meteorological data sequences and set benchmark feature points, generating two independent associated feature sets; it determines the feature influence domain based on the relationship between the two associated feature sets, and selects a representative feature point within the feature influence domain.

[0020] The calibration module is used to construct an analysis loop based on the temporal evolution relationship of representative feature points. By calculating the motion velocity of meteorological elements in the feature influence domain relative to the regional background field, it analyzes the contribution of spatial variation and transport process of elements, quantifies the dynamic effect of motion on extreme climate evolution, calculates the calibration amount of meteorological data, performs dynamic calibration on the standardized meteorological data sequence, obtains the calibrated meteorological element sequence, and generates extreme climate indicators.

[0021] The setting module is used to analyze the risk level of extreme climate events based on extreme climate indicators, combined with the underlying surface feature data and disaster record data of the target area, through a dynamic risk assessment model, and to set multi-level early warning thresholds according to the preset climate state classification standards.

[0022] The push module is used to compare real-time monitored meteorological data with multi-level early warning thresholds through the cloud platform. When the real-time monitored data exceeds the multi-level early warning thresholds, it automatically generates and pushes early warning information to the designated terminal.

[0023] The monitoring module is used to build a monitoring framework based on cloud-native architecture and component-based design, and to use early warning information to achieve distributed collaborative operation and dynamic monitoring.

[0024] In this embodiment of the invention, by standardizing the processing of real-time meteorological and environmental data, an independent and associated feature set is constructed, the feature influence domain is delineated, and representative feature points are selected. This enables the screening and refinement of massive heterogeneous meteorological data. An analysis loop is constructed based on the temporal evolution relationship of representative feature points. The dynamic effect of meteorological element movement on extreme climate is quantified by combining the regional background field, thereby achieving dynamic calibration of the standardized data sequence and improving the accuracy and reliability of extreme climate indicators. By combining the underlying surface characteristics of the target area, disaster record data, and dynamic risk assessment models, multi-level early warning thresholds are constructed, making the risk level classification more consistent with the actual climate impact patterns of the region. The comparison between real-time data and early warning thresholds is completed by relying on the cloud platform, and early warning information can be automatically pushed when the risk is triggered, shortening the early warning response time. A monitoring framework is constructed based on cloud-native architecture and component-based design to achieve distributed collaborative operation and dynamic monitoring, improving the system's ability to process massive amounts of data and its anti-interference ability, and reducing system operation and maintenance costs.

[0025] In a preferred embodiment of the present invention, real-time collected meteorological and environmental data are processed to form a standardized meteorological data sequence and a baseline feature point is set, generating two independent associated feature sets; the feature influence domain is determined based on the relationship between the two associated feature sets, and a representative feature point is selected within the feature influence domain, which may include:

[0026] In this embodiment of the invention, quality control is performed on the real-time collected meteorological and environmental data. Based on preset threshold ranges and climatological limit rules, outliers in the original data are identified and marked, and time-series imputation is performed on missing values ​​to generate a continuous data sequence. Specifically, this includes: first, determining that the real-time collected meteorological data specifically covers temperature, precipitation, wind speed, relative humidity, air pressure, and sunshine duration; and that the environmental data specifically covers vegetation cover, soil volumetric water content, topographic elevation, surface temperature, and water coverage. All data are accompanied by a collection timestamp and the latitude and longitude coordinates of the collection location. Second, a specific threshold range is pre-set for each type of data, and this threshold range is determined by statistically analyzing the target area. The arithmetic mean and standard deviation of similar data from the same period over the past 30 years were determined, taking values ​​ranging from the arithmetic mean plus or minus three times the standard deviation. Simultaneously, climatological limit rules were established for each data category. These rules were based on the extreme maximum and minimum values ​​of similar data since records began in the target area, defining the upper and lower limits that data for each category could not exceed. Then, a quality check was performed on each piece of raw data. First, the data value was compared with the preset threshold range for the corresponding category. If the data value exceeded the threshold range, it was then compared with the climatological limit rules for the corresponding category. If both the upper and lower limits were exceeded, the data was determined to be an outlier. The data is labeled with specific anomaly information, including the original value, collection time, collection location, and anomaly type. This labeling information is stored in conjunction with the original data. Finally, time-series imputation is performed on missing values ​​in the original data. Before imputation, the data collection frequency is determined. For a single isolated missing value, three consecutive valid data points before and after the missing value are extracted. The values ​​of these six valid data points are added together and then divided by 6 to obtain the arithmetic mean. This arithmetic mean is used as the imputation value for the isolated missing value. For multiple consecutive missing values, five consecutive valid data points before and after the consecutive missing segment are extracted. The slope of the numerical change of the first five valid data points is calculated, and the value of each subsequent data point is subtracted from the value of the previous one. Based on the numerical differences, the arithmetic mean of all differences is taken as the front slope. Then, the slope of the numerical change of the last 5 valid data points is calculated in the same way as the front slope. Starting from the previous valid data value of the continuous missing segment, the preliminary imputation value of each missing value is calculated by substituting it into the front slope one by one. Starting from the next valid data value of the continuous missing segment, the preliminary imputation value of each missing value is calculated by substituting it into the back slope one by one. The two preliminary imputation values ​​of each missing value are added together and divided by 2 to obtain the final imputation value. All data after annotating outliers are integrated with the missing value data after imputation, and duplicate data are removed to form a continuous, complete, and unannotated continuous data sequence.

[0027] All data points in a continuous data sequence are arranged chronologically according to their corresponding timestamps to form a time-ordered data sequence. Max-min normalization is then applied to convert the values ​​of each meteorological element in the time-ordered data sequence to a unified range, resulting in a standardized meteorological data sequence. Specifically, this involves: First, extracting the timestamp information corresponding to each data point in the continuous data sequence, and then sorting all data points point-by-point according to their timestamps from earliest to latest. During the sorting process, it is ensured that the collection location, data value, and timestamp of each data point accurately correspond. After sorting, a time-ordered data sequence is formed, in which the data points are arranged chronologically without any time order discrepancies or data misalignments. Second, max-min normalization is performed on each meteorological element (temperature, precipitation, wind speed, relative humidity, air pressure, and sunshine duration) in the time-ordered data sequence. This processing is carried out independently for each meteorological element, without mixing different elements. Taking temperature as an example, firstly, the values ​​of all temperature data points are extracted from the time-series regularized data sequence. The maximum and minimum values ​​are then compared one by one. For each temperature data point, a calculation is performed: subtract the minimum temperature value from the data point's value to obtain the difference; subtract the minimum temperature value from the maximum temperature value to obtain the range; and then divide the difference by the range. This process transforms the temperature data point's value into the 0-1 range. Following the same calculation method, other meteorological elements in the time-series regularized data sequence, such as precipitation, wind speed, relative humidity, air pressure, and sunshine duration, are normalized to ensure that all data points for each meteorological element are converted to the unified 0-1 range. Finally, all normalized meteorological element data are integrated according to their original timestamp order to form a standardized meteorological data sequence. Each data point contains the standardized value of the corresponding meteorological element, the collection timestamp, and the latitude and longitude of the collection location.

[0028] Based on meteorological data sequences, extreme value identification and trend reversal identification are performed. According to preset meteorological element numerical thresholds, extreme high or low points in the sequence are selected as baseline feature points. Simultaneously, reversal points where the trend of meteorological element changes significantly are identified in the sequence, and these reversal points are also set as baseline feature points. Specifically, this includes: First, based on standardized meteorological data sequences, extreme value identification and trend reversal identification are carried out simultaneously without interference. During extreme value identification, specific numerical thresholds are preset for each type of meteorological element, including extreme high and extreme low thresholds. The extreme high threshold is set as the value of the meteorological element in the standardized data sequence. The 95th percentile value is defined as the value at the 95th percentile after sorting from smallest to largest. The extreme low value threshold is set as the 5th percentile value of all values ​​in the standardized data sequence of the meteorological element, i.e., the value at the 5th percentile after sorting from smallest to largest. Each data point value for each type of meteorological element is compared one by one. If a data point value is higher than the extreme high value threshold of the corresponding meteorological element, it is determined to be an extreme high value point; if a data point value is lower than the extreme low value threshold of the corresponding meteorological element, it is determined to be an extreme low value point. All identified extreme high and extreme low points are included in the baseline feature point candidate range, and the value and corresponding information of each candidate point are recorded. Meteorological element type, data collection timestamp, and data collection location latitude and longitude; during trend reversal identification, for each type of meteorological element's standardized data sequence, adjacent data points are compared one by one in chronological order. The difference between the value of the subsequent data point and the value of the preceding data point is calculated. The trend is determined based on the sign of the difference: a positive difference indicates an upward trend, a negative difference indicates a downward trend, and a difference of 0 indicates a stable trend. The trend of three consecutive adjacent data points is then statistically analyzed, i.e., the trend of two consecutive differences. If the first two consecutive differences are both positive (continuous upward trend), and the subsequent adjacent difference turns negative (downward trend), then the data point where the trend changes is identified, i.e., the third phase. The preceding data point of an adjacent data point is identified as the turning point from rising to falling. If the first two consecutive differences are both negative (continuous downward trend), and the subsequent adjacent differences turn positive (upward trend), then the data point where the trend changes is identified as the turning point from falling to rising. If a stable trend appears in the consecutive differences and then the trend changes, the data point at the point where the trend changes is also identified as the turning point. Finally, all extreme high value points, extreme low value points, and all turning points are uniformly set as benchmark feature points. Each benchmark feature point is assigned a unique identifier and its corresponding value, meteorological element type, collection timestamp, collection location latitude and longitude, and feature type (extreme high value / extreme low value / turning point) are recorded.

[0029] Centered on the position of each benchmark feature point in the time series, fixed-length data subsequences are extracted forward and backward to form the analysis time window corresponding to the benchmark feature point. From the data subsequence of each time window, dynamic features including mean, variance, trend, and fluctuation frequency are extracted. The dynamic features of all benchmark feature points are then summarized to form the first association feature set. Specifically, this includes: First, for each benchmark feature point, locating its corresponding time position in the standardized meteorological data series, i.e., the time sequence position corresponding to the collection timestamp. Centered on this time position, fixed forward and backward truncation lengths are pre-set, both set to 12 hours, which can be adjusted according to the data collection frequency. The number of data points extracted should be no less than 20. According to the set time length, all data points within the corresponding time period are extracted from the standardized meteorological data sequence to form a forward data subsequence, and all data points within the corresponding time period are extracted to form a backward data subsequence. The forward and backward data subsequences are then integrated in chronological order, and duplicate data points are removed to form a unique analysis time window for each benchmark feature point. Each analysis time window is associated with a unique identifier for the corresponding benchmark feature point. Next, for each data subsequence within the analysis time window, four dynamic features—mean, variance, trend, and fluctuation frequency—are extracted one by one. The extraction process is based on the values ​​of all data points within the window. When extracting the mean... To extract the variance, first sum the values ​​of all data points within the window, then divide the sum by the total number of data points to obtain the mean. To extract the variance, first calculate the mean of all data points within the window (using the same method as mean extraction), then subtract the mean from the value of each data point to obtain the difference between each data point and the mean. Square each difference to obtain the square value, then sum all the squares to obtain the sum of squares. Divide the sum of squares by the total number of data points within the window to obtain the variance. To extract the trend, first find the value of the first data point (starting value) and the value of the last data point (ending value) within the window, then subtract the starting value from the ending value to obtain the trend. The trend difference value is used to determine the overall trend. A positive trend difference indicates an upward trend, a negative trend indicates a downward trend, and a zero trend indicates a stable trend. Simultaneously, the difference between the values ​​of all adjacent data points within the window is calculated. The sum of all differences is divided by the total number of differences to obtain the average difference. The larger the absolute value of the average difference, the more drastic the trend, and vice versa. When extracting the fluctuation frequency, the mean of all data points within the window is first calculated (using the same method as mean extraction). Each data point's value within the window is then compared to the mean, recording the number of times the value changes from below the mean to above the mean, or from above the mean to below the mean (each crossing is counted as one fluctuation). The more fluctuations, the higher the fluctuation frequency.Simultaneously, the time interval of each fluctuation (the time difference between two adjacent crossings) is calculated. All time intervals are summed and divided by the number of fluctuations to obtain the average fluctuation interval, which helps characterize the fluctuation frequency. Finally, the four dynamic features corresponding to each benchmark feature point—mean, variance, trend (including trend direction and average difference), and fluctuation frequency (including the number of fluctuations and average fluctuation interval)—are summarized one by one according to the unique identifier of the benchmark feature point to form the first associated feature set. Each entry in the first associated feature set is completely associated with all dynamic feature information and basic identifier information of the corresponding benchmark feature point.

[0030] For each benchmark feature point, a geographic spatial range with a preset radius is defined centered on the benchmark feature point. Within this spatial range, environmental element information is extracted based on the processed raw environmental data and remote sensing inversion data, and spatial interpolation is used to generate the spatial distribution characteristics of the environmental elements. The spatial distribution characteristics of all benchmark feature points are then summarized to form a second associated feature set. Specifically, this includes: First, for each benchmark feature point, its corresponding spatial location information is extracted. A fixed geographic spatial radius is preset, centered on the latitude and longitude coordinates. This radius is set according to the terrain characteristics of the target area: 50 kilometers for plains, 30 kilometers for mountainous areas, and 40 kilometers for coastal areas. A circular geographic spatial range is delineated using this radius, and the latitude and longitude boundaries (easternmost, westernmost, southernmost, and northernmost coordinates) of this range are clearly defined. This circular range is the spatial analysis range corresponding to the benchmark feature point, and each spatial analysis range is associated with a unique identifier for the corresponding benchmark feature point. Next, raw environmental data (vegetation coverage, soil volumetric water content, topographic elevation, surface temperature, and water coverage) and remote sensing inversion data (total water vapor column, surface albedo, and aerosol optical thickness) that have undergone quality control are acquired. From these data, all environmental element data within the delineated geographic spatial range are extracted. During the extraction process, latitude and longitude coordinate comparisons are used to ensure that all extracted environmental element data fall within the delineated geographic spatial range. Within a defined spatial range, data exceeding the range is discarded. The extracted environmental element data is then organized and categorized by type, with each environmental element data point associated with a corresponding collection timestamp, consistent with the collection timestamp of the baseline feature point and its spatial location (latitude and longitude). Next, spatial interpolation is performed on the categorized environmental element data using a distance-inverse weighted method. Specifically, the defined geographic space is divided into 1km × 1km grids, resulting in several grid points, each with unique latitude and longitude coordinates. For each grid point, all discrete environmental element data points within a certain surrounding range are extracted, and the straight-line distance between each discrete data point and the grid point is calculated. Divide 1 by the straight-line distance to obtain the distance weight. The closer the distance, the greater the weight. Add the distance weights of all discrete data points one by one to obtain the total weight. Multiply the environmental element value of each discrete data point by its corresponding distance weight to obtain the weighted value of each discrete data point. Add all the weighted values ​​one by one to obtain the weighted sum. Divide the weighted sum by the total weight to obtain the interpolated value of the environmental element for that grid point. Perform interpolation calculations on all grid points within the geographic space in the above manner to generate continuous spatial distribution data of environmental elements covering the entire geographic space. This data can clearly reflect the numerical distribution pattern and spatial variation characteristics of environmental elements within the spatial range.Finally, the spatial distribution characteristics of various environmental elements corresponding to each benchmark feature point are summarized one by one according to the unique identifier of the benchmark feature point to form a second associated feature set. Each entry in the second associated feature set is completely associated with all spatial distribution feature information and basic identifier information of all environmental elements corresponding to the benchmark feature point.

[0031] The coupling strength index between the first and second associated feature sets is calculated to obtain a quantitative correlation measurement result. Specifically, this includes: First, establishing a one-to-one correspondence between the first and second associated feature sets, using the unique identifier of a baseline feature point as the correlation basis. The dynamic characteristics (mean, variance, trend, and fluctuation frequency) of each baseline feature point in the first associated feature set are matched one-to-one with the spatial distribution characteristics of environmental elements in the second associated feature set. This involves matching the spatial distribution mean, spatial distribution variance, spatial trend, and spatial fluctuation frequency of each environmental element to form multiple sets of feature correspondences. For example, the mean temperature corresponds to the spatial distribution mean of vegetation cover, the variance of temperature corresponds to the spatial distribution variance of vegetation cover, the trend of wind speed corresponds to the spatial trend of soil moisture, and the frequency of humidity fluctuation corresponds to the spatial fluctuation frequency of total water vapor column. Each set of correspondences is associated with the unique identifier of the same baseline feature point. Second, for each set of feature correspondences, the correlation coefficient between the two is calculated. The calculation process involves extracting all data point values ​​of the two features in the set of correspondences, including the time-series data of the dynamic features. For grid data points with point and spatial distribution characteristics, ensuring a consistent number of data points, multiply each data point value of the first characteristic with each corresponding data point value of the second characteristic one by one to obtain several product values. Sum all these product values ​​to obtain a total product. Simultaneously, calculate the sum of all data point values ​​for the first characteristic and the sum of all data point values ​​for the second characteristic, and multiply these two sums to obtain a total product. Divide this total product by the total product to obtain the correlation coefficient of the corresponding feature set. The correlation coefficient ranges from 0 to 1; the closer the value is to 1, the stronger the correlation between the two sets of features. Then, count the total number of sets of corresponding features and sum the correlation coefficients of all sets to obtain a total correlation coefficient. Divide this total correlation coefficient by the total number of sets of corresponding features to obtain a coupling strength index. This index is the quantitative result of the correlation between the first and second sets of correlated features, and can intuitively and comprehensively reflect the coupling and matching of the dynamic characteristics of meteorological elements and the spatial distribution characteristics of environmental elements. The closer the index value is to 1, the stronger the overall correlation between the two sets of features.

[0032] Based on the correlation measurement results, the overlapping influence areas of the first and second correlation feature sets in spatial and temporal dimensions are identified to obtain a preliminary overlapping spatiotemporal range. Specifically, this includes: First, identifying the overlapping influence areas in the spatial dimension by extracting the latitude and longitude of the collection location corresponding to each benchmark feature point in the first correlation feature set, and the geographic spatial range (latitude and longitude boundaries of the circular range) corresponding to each benchmark feature point in the second correlation feature set; superimposing the geographic spatial ranges (circular ranges) of all benchmark feature points onto the same geographic coordinate system, comparing the overlap of each circular range one by one, and identifying areas where multiple circular ranges overlap, i.e., areas simultaneously covered by two or more circular ranges; integrating all overlapping areas, removing duplicate coverage, forming the overlapping influence area in the spatial dimension, and clarifying the latitude and longitude boundaries (easternmost, westernmost, southernmost, and northernmost coordinates) of this area; Second, identifying the overlapping influence areas in the temporal dimension by extracting the latitude and longitude of the first correlation feature set, and the second correlation feature set. The analysis time window (start and end timestamps) corresponding to each benchmark feature point in the first association feature set is collected, as well as the analysis time window corresponding to each benchmark feature point in the second association feature set (consistent with the first association feature set). The analysis time windows of all benchmark feature points are superimposed on the same time axis, and the overlap of each time window is compared one by one to find the time periods where multiple time windows overlap, that is, the time periods covered by two or more time windows at the same time. All overlapping time periods are integrated, and the overlapping parts are removed to form the overlapping influence area in the time dimension, and the start and end timestamps of the time period are determined. Finally, the overlapping influence area in the spatial dimension is integrated with the overlapping influence area in the time dimension to form a preliminary overlapping spatiotemporal range that includes the overlapping spatial range (latitude and longitude boundaries) and the overlapping time period (start and end timestamps). This range can completely cover the core area and time period in which the first association feature set and the second association feature set work together in the spatiotemporal dimension, and focus the key range for subsequent analysis.

[0033] Within the initial overlapping spatiotemporal range, the spatial distribution and temporal variation characteristics of the correlation measurement results are analyzed to extract core areas and time periods where the correlation consistently exceeds a preset threshold. Specifically, this includes: First, a correlation threshold is preset. This threshold is determined based on historical correlation measurement data of the target area, using statistical analysis of data from the same period over the past 10 years. The value is taken as the 80th percentile of the historical correlation measurement data, i.e., the value at the 80th percentile after sorting from smallest to largest. This value is used to determine the strength of the correlation; a value higher than this threshold indicates a stronger correlation. Second, within the initial overlapping spatiotemporal range, the spatial distribution and temporal variation characteristics of the correlation measurement results are analyzed at 1 km × 1 The system divides spatial units into kilometer-scale grids and time units into one-hour periods, forming several spatiotemporal units. Each spatiotemporal unit corresponds to one spatial grid and one time unit. The system extracts and compares the correlation metrics (coupling strength index) within each spatiotemporal unit, comparing the index with a preset correlation threshold. If the index value is higher than the threshold, the spatiotemporal unit is marked as a valid spatiotemporal unit, and its spatial grid coordinates and time period are recorded. Then, the core region is extracted, and the valid time corresponding to each spatial grid unit is calculated. The number of units is calculated by taking the proportion of valid time units to the total number of time units within the initially overlapping spatiotemporal range of the spatial grid unit (valid duration percentage). If the valid duration percentage of a spatial grid unit reaches 70% or more, then the spatial grid unit is included in the core region candidate range. All eligible spatial grid units are integrated, and discrete single grid units (isolated grid units) are removed to form a continuous core region, and the latitude and longitude boundaries of the core region are defined. At the same time, core time periods are extracted, and the number of valid spatial grid units corresponding to each time unit is counted. The proportion of the number of valid spatial grid units to the total number of spatial grid units within the initially overlapping spatiotemporal range of the time unit (valid spatial percentage) is calculated. If the valid spatial percentage of a time unit reaches 70% or more, then the time unit is included in the core time period candidate range. All eligible time units are integrated, and discrete single time units (isolated time units) are removed to form a continuous core time period, and the start and end timestamps of the core time period are defined. Finally, the core region and the core time period together constitute a core region and time period with a correlation degree that is consistently higher than a preset threshold. Both focus on areas with strong and consistently stable correlation.

[0034] Based on the core area and time period, the final spatiotemporal boundaries are delineated, and the characteristic influence domain is determined. Within the characteristic influence domain, anomaly indicators reflecting the degree of deviation of meteorological elements from the norm, sensitivity indicators reflecting the degree of response of the natural environment to meteorological changes, and event frequency indicators based on past observation data are acquired and calculated. Specifically, this includes: First, delineating the spatiotemporal boundaries of the characteristic influence domain, using the latitude and longitude coordinates of the outermost spatial grid unit of the core area as the spatial boundary (easternmost, westernmost, southernmost, and northernmost coordinates), ensuring that the core area is completely contained within the spatial boundary, while extending outward by 5 kilometers to avoid missing data at the edge of the core area; second, using the start and end timestamps of the core time period as the time boundary, ensuring that the core time period is completely contained within the time boundary. The analysis extends the data by one hour both forward and backward to avoid missing edge data during the core period. The extended spatial and temporal boundaries are then integrated to form the final spatiotemporal boundary, which covers the characteristic influence domain. This domain includes the complete core area and core time period, as well as the surrounding auxiliary analysis area. Next, an anomaly index is calculated within the characteristic influence domain, obtaining real-time values ​​for each spatiotemporal unit of meteorological elements (temperature, precipitation, wind speed, etc.) within the domain. Simultaneously, the normal values ​​and historical averages for each meteorological element in the target area over the same period (same time period, same spatial range) over the past 30 years are obtained. These values ​​are calculated by adding the values ​​of similar data from the same period over the past 30 years and dividing by 30. For each meteorological element, each spatiotemporal unit... The numerical difference is obtained by subtracting the historical average value from the real-time value. Simultaneously, the numerical fluctuation range of the meteorological element over the same period over the past 30 years is obtained, with the historical maximum value minus the historical minimum value. The anomaly index of the meteorological element in that spatiotemporal unit is obtained by dividing the numerical difference by the numerical fluctuation range. The anomaly index of all meteorological elements is obtained by summing the anomaly indexes of all meteorological elements and dividing by the total number of meteorological element categories. Then, the sensitivity index is calculated to obtain the real-time values ​​of each environmental element (vegetation cover, soil moisture, etc.) in each spatiotemporal unit within the characteristic influence domain, as well as the meteorological driving factors corresponding to each environmental element, such as the real-time values ​​of temperature and precipitation for vegetation cover, and precipitation and evaporation for soil moisture in each spatiotemporal unit. Calculate the real-time numerical change of each environmental element (the current spatiotemporal unit value minus the previous spatiotemporal unit value), and calculate the real-time numerical change of its corresponding meteorological driving factor (calculated in the same way as the environmental element). Divide the real-time numerical change of the environmental element by the real-time numerical change of the corresponding meteorological driving factor to obtain the sensitivity sub-index of the environmental element. Add up the sensitivity sub-indices of all environmental elements and divide by the total number of environmental element categories to obtain the overall sensitivity index of the characteristic influence domain. Finally, calculate the event occurrence frequency index, retrieve the observation data of the target area for the past 30 years, and screen out extreme climate events that occur within the range of spatiotemporal characteristics similar to the characteristic influence domain, with a spatial overlap of ≥80% and a time period similarity of ≥80%.The frequency of extreme climate events meeting the criteria was calculated. Simultaneously, the total number of observations coinciding with the characteristic impact area over the same period over the past 30 years was also calculated, with each year counted as one complete observation, totaling 30 events over 30 years. The frequency index of events was obtained by dividing the number of extreme climate events by the total number of observations.

[0035] The anomaly index, sensitivity index, and event frequency index are normalized and then weighted and fused according to preset weights to generate a comprehensive evaluation value. Based on the spatial distribution of the comprehensive evaluation value within the feature influence domain, the spatial location where the comprehensive evaluation value reaches its peak is located as a representative feature point. Specifically, this includes: First, normalizing the three indicators separately using maximum-minimum value normalization, and conducting the process independently without overlap. For the anomaly index, the maximum and minimum values ​​of the index in all spatiotemporal units within the feature influence domain are found. The minimum anomaly value is then subtracted from the anomaly index value in each spatiotemporal unit to obtain the result. To obtain the anomaly difference, subtract the minimum anomaly from the maximum anomaly to get the anomaly range. Divide the anomaly difference by the anomaly range to convert the anomaly index value to the 0-1 range. Using the same calculation method, normalize the sensitivity index and the event frequency index to ensure that both index values ​​are converted to the 0-1 range. Next, pre-set the weights for the three normalized indicators. The weights are determined based on the importance of each indicator to extreme climate risk identification, with the anomaly index having a weight of 0.4, the sensitivity index having a weight of 0.3, and the event frequency index having a weight of 0.3. The sum is 1. The three normalized indicators are then weighted and fused. The calculation method is as follows: multiply the normalized anomaly index value by 0.4 to obtain the anomaly weighted value; multiply the normalized sensitivity index value by 0.3 to obtain the sensitivity weighted value; multiply the normalized event frequency index value by 0.3 to obtain the event frequency weighted value. The anomaly weighted value, sensitivity weighted value, and event frequency weighted value are then added together to obtain the comprehensive evaluation value for each spatiotemporal unit. Finally, the spatial distribution pattern of the comprehensive evaluation value is generated, and the average of the comprehensive evaluation values ​​for each spatial grid unit within the characteristic influence domain during the core time period is taken. The average comprehensive evaluation value of each spatial grid unit is obtained by summing the comprehensive evaluation values ​​of all spatiotemporal units and dividing by the number of time units. The average comprehensive evaluation value of each spatial grid unit is then plotted according to the coordinates of the spatial grid units to form a spatial distribution pattern of the comprehensive evaluation values. Finally, representative feature points are located, and the average comprehensive evaluation value of each spatial grid unit in the spatial distribution pattern is compared one by one to find the spatial grid unit with the largest average comprehensive evaluation value (reaching the peak value). The central latitude and longitude coordinates of the grid unit are extracted, and the spatial location point corresponding to the central latitude and longitude coordinates is determined as the representative feature point. This point can reflect the core characteristics of extreme climate risk within the feature influence domain.

[0036] By implementing quality control and missing value imputation, outlier data is eliminated and missing data is filled, ensuring the integrity and reliability of basic data and avoiding deviations in analysis results due to data problems. Through time series regularization and independent normalization processing, the standardization and unification of data of different meteorological elements are achieved.

[0037] In a preferred embodiment of the present invention, an analysis loop is constructed based on the temporal evolution relationship of representative feature points. By calculating the motion velocity of meteorological elements within the feature influence domain relative to the regional background field, the contribution of spatial variation and transport processes of the elements is analyzed, the dynamic effect of motion on extreme climate evolution is quantified, meteorological data calibration is calculated, and the standardized meteorological data sequence is dynamically calibrated to obtain the calibrated meteorological element sequence, generating extreme climate indicators, which may include:

[0038] In this embodiment of the invention, a representative feature point is used as the spatiotemporal starting point. The evolution trajectories of corresponding meteorological elements are traced forward and backward along the time axis, respectively. The trajectories obtained from the forward and backward trajectories are then connected to form an analysis loop. Specifically, this includes: First, locking the core information of the representative feature point, determining its spatial location as latitude and longitude coordinates, determining the type of core meteorological element corresponding to the feature point, matching it according to the type of extreme climate event: extreme rainstorm events correspond to precipitation elements, extreme high temperature events correspond to temperature elements, and extreme strong wind events correspond to wind speed elements. The collection timestamp of the feature point is recorded, and this point is used as the unique spatiotemporal starting reference point for trajectory tracing. All tracing operations are carried out around this reference point. Second... The time boundaries for trajectory tracking are determined. The forward tracking time range is defined as the period from the start of the characteristic influence domain's time boundary to the timestamp of representative feature point collection. The backward tracking time range is defined as the period from the timestamp of representative feature point collection to the end of the characteristic influence domain's time boundary, ensuring the tracking range completely covers the entire time period of the characteristic influence domain without any time gaps. Then, trajectory tracking is carried out in stages. During forward trajectory tracking, starting from the spatial coordinates of the representative feature point, the values ​​of core meteorological elements and the spatial locations of corresponding high-value centers (or low-value centers for extreme low-value events) are extracted sequentially along the time axis. For each time period, the core meteorological data of all spatial grid points within the characteristic influence domain are first statistically analyzed. For element values, the grid point with the largest (or smallest) value is selected. This grid point is the core center for that time period. Its latitude and longitude coordinates, core meteorological element value, and corresponding timestamp are recorded. The core centers of each time period are connected sequentially from late to early to form a continuous trajectory segment extending forward. Each node on the trajectory segment is fully associated with the time period, value, and coordinate information. When tracing the trajectory backward, starting from the spatial coordinates of a representative feature point, the spatial location and related information of the core meteorological element core centers are extracted for each time period along the time axis using the same time intervals and core center selection criteria as for forward tracing. The core centers of each time period are then connected sequentially from early to late to form a continuous trajectory segment extending backward. Continuous trajectory segments; finally, the trajectory connection is closed to form an analysis loop. First, the starting node of the forward trajectory segment, i.e., the core center coordinates corresponding to the starting time of the characteristic influence domain time boundary, and the ending node of the backward trajectory segment, i.e., the core center coordinates corresponding to the ending time of the characteristic influence domain time boundary, are extracted. These two nodes are connected by a straight line, so that the forward trajectory segment, the backward trajectory segment, and the connecting straight line together form a complete closed trajectory line. The entire spatiotemporal range enclosed by this closed trajectory line, including the time period and spatial region corresponding to all nodes on the trajectory line, is the analysis loop. Each analysis loop is individually associated with the corresponding representative feature point information and core meteorological element type to ensure the uniqueness and relevance of the loop.

[0039] Within the characteristic influence domain, the advection velocity vector of meteorological element fields relative to the regional climate background field at each spatial grid point is calculated during the time period covered by the analysis loop. Simultaneously, the diffusion rate scalar of meteorological element fields at each spatial grid point within the time period is calculated. Specifically, this includes: first, completing basic data preparation and preprocessing by dividing the characteristic influence domain into uniform spatial grids according to a fixed standard of 1 km × 1 km, assigning unique latitude and longitude coordinates to each grid point to ensure grid coverage of the entire characteristic influence domain without overlap or omission; second, extracting the time-by-time values ​​of core meteorological elements at each spatial grid point within the time period covered by the analysis loop, i.e., meteorological element field data, with each value associated with the corresponding grid point coordinates and time period timestamp. To ensure data continuity in time and spatial integrity, a regional climate background field is constructed. This background field data is calculated based on the time-period averages of core meteorological elements at each spatial grid point in the target region over the past 30 years. Specifically, for each spatial grid point and corresponding time period, the core meteorological element values ​​for that grid point and time period over the past 30 years are extracted. These 30 values ​​are then summed one by one, and the sum is divided by 30 (the total number of years over the past 30 years) to obtain the background field average for that grid point and time period. The background field averages for all grid points and all time periods are integrated to form complete regional climate background field data. Secondly, the advection velocity vector (including the east-west component and the south-north component) is calculated grid-by-grid and time-period by time period. (The magnitude and direction of the vector are clearly defined). When calculating the east-west component, first extract a spatial grid point, denoted as grid A, and its eastern adjacent grid point, denoted as grid B. Grid points located at the same latitude as grid A, but with a longitude increase of 1 km, are used to calculate the core meteorological element field values ​​and regional climate background field average values ​​for the same time period. The differences between the element field values ​​and background field average values ​​of grid A and grid B are calculated respectively: the element field value of grid A minus the background field average value of grid A, and the element field value of grid B minus the background field average value of grid B. Subtracting the difference in grid A from the difference in grid B yields the change in the east-west direction difference. Dividing this change in difference by the horizontal distance between grid A and grid B gives the element values ​​in the east-west direction. The gradient is multiplied by the corresponding time interval, determined by the data acquisition frequency (e.g., 10 minutes for every 10 minutes of data collection). This yields the east-west advection velocity component, with positive values ​​indicating eastward transport and negative values ​​indicating westward transport. When calculating the north-south component, grid A and its adjacent grid points to its north are extracted and designated as grid C. The element field values ​​and background field mean values ​​of the grid points located at the same longitude as grid A, but 1 km higher in latitude, are calculated using the same method as the east-west component: first, the difference, the change in the difference, and the north-south gradient are calculated. Then, these are multiplied by the time interval to obtain the north-south advection velocity component, with positive values ​​indicating northward transport and negative values ​​indicating southward transport.The east-west component is combined with the south-north component to form the advection velocity vector for grid A during that time period. The vector magnitude is calculated by multiplying the east-west component value by itself (square operation) and adding the south-north component value multiplied by itself (square operation), obtaining the sum of the two squared values, and taking the square root of this sum, which is the vector magnitude. The vector direction is determined by the combination of the positive and negative signs of the two components; for example, if the east-west component is positive and the south-north component is positive, the direction is northeast. Then, the diffusion rate scalar is calculated grid by grid and time period by time. For grid A, its four adjacent grid points (east, south, west, and north) are extracted, namely grids B, grid D, grid E, and grid C. Grid D is the south adjacent grid point, and grid E is the west adjacent grid point, representing the core meteorological element field values ​​for the same time period. The values ​​of each adjacent grid point are calculated and compared with the values ​​of grid A. The difference in values, i.e., the value of adjacent grid points minus the value of grid A, is calculated by summing the absolute values ​​of the four differences one by one. This sum is then divided by the average distance between the four adjacent grid points and grid A, which is fixed at 1 kilometer (since the distance between adjacent grid points and grid A is always 1 kilometer). This sum is then divided by the corresponding time interval to obtain the diffusion rate scalar for grid A during that time interval. The larger the scalar value, the more intense the spatial diffusion of the core meteorological element at that grid point. Finally, the above calculation process is repeated for each spatial grid point and each time interval within the analysis loop coverage period to calculate the advection velocity vector and diffusion rate scalar, forming a complete vector field dataset and scalar field dataset. All data are associated with the corresponding grid point coordinates, time interval timestamps, and core meteorological element types to ensure data traceability and verifiability.

[0040] Based on the advection velocity vector and diffusion rate scalar, the total rate of change of meteorological elements is calculated, and decomposed into the contribution of change caused by the advection transport process and the contribution of change caused by the local diffusion process, obtaining the contribution ratio of each. Specifically, this includes: First, calculating the total rate of change of meteorological elements grid by grid and time period by time period. For a certain spatial grid point (grid A) and a certain time period (time period T), the core meteorological element field values ​​of grid A in time period T (current values) and the core meteorological element field values ​​of the previous time period (time period T-1) are extracted. The current value is subtracted from the previous value to obtain the change in value of the grid point between the two time periods. The change in value is divided by the time interval between the two time periods to obtain the change in value of grid A in time period T. The total rate of change of core meteorological elements is calculated. A positive rate of change indicates an upward trend in the values ​​of meteorological elements during that period; a negative rate indicates a downward trend; and a larger absolute value indicates a more drastic change. Next, the contribution of the advection transport process is calculated. The magnitude, east-west component, and south-north component of the advection velocity vector at grid A during time period T are extracted, along with the difference between the core meteorological element field values ​​at that grid point during time period T and the regional climate background field mean (element deviation value). The advection velocity vector magnitude is multiplied by the element deviation value to obtain the baseline value of the advection effect. The consistency between the direction of the advection velocity vector and the total trend of meteorological element change is determined. If the total rate of change is positive, the vector direction is towards that grid point; if they are opposite, the direction is reversed. The direction is determined based on the consistency. The coefficient is set to 1 when the directions are consistent and -1 when the directions are opposite. The directional coefficient is multiplied by the baseline value of advection to obtain the contribution value of the advection transport process. A positive value indicates that advection transport promotes the development of meteorological element values ​​towards the current trend; a negative value indicates that advection transport inhibits the current trend. Then, the contribution of local diffusion is calculated by extracting the diffusion rate scalar of grid A at time T and the core meteorological element field value of that grid point at time T. The diffusion rate scalar is multiplied by the element field value to obtain the baseline value of diffusion. The consistency between the diffusion direction and the overall trend of meteorological element change is determined. If the overall rate of change is positive, the diffusion direction is the convergence of element values ​​towards grid A, indicating a consistent direction; if it is divergent, the direction is opposite. The consistency is then used to determine the overall trend of the meteorological element change. A diffusion direction coefficient is defined, with a coefficient of 1 when the directions are consistent and a coefficient of -1 when the directions are opposite. The base value of diffusion is multiplied by the diffusion direction coefficient to obtain the contribution value of the change caused by the local diffusion process. The positive or negative meaning of this value is consistent with that of the advection transport contribution value: a positive value promotes change, and a negative value inhibits change. Next, the rationality of the contribution value is verified by adding the advection transport contribution value and the local diffusion contribution value to obtain the composite contribution value. The difference between the composite contribution value and the total rate of change is calculated. If the absolute value of this difference does not exceed 10% of the absolute value of the total rate of change, the calculation of the advection transport and local diffusion contribution values ​​is considered valid. If it exceeds the allowable error range, the direction coefficient is readjusted, such as adjusting coefficient 1 to 0.9 and coefficient -1 to -0.9. Recalculate the two contribution values ​​until the difference meets the allowable error requirement; finally, calculate the contribution ratio of the two processes. Divide the absolute value of the advection transport change contribution value by the absolute value of the composite change contribution value to obtain the advection transport contribution ratio; divide the absolute value of the local diffusion change contribution value by the absolute value of the composite change contribution value to obtain the local diffusion contribution ratio. Add the two ratios together to obtain 1, since the composite change contribution value is the algebraic sum of the two contribution values, and its absolute value is equal to the sum of the absolute values ​​of the two contribution values. Calculate the contribution ratio for each spatial grid point and each time period, and record the corresponding ratio data.

[0041] Based on the proportion of advection transport contribution and the intensity of the advection velocity vector, the dynamic weight of atmospheric motion on the formation and evolution of extreme climate events at representative characteristic points is quantified. Specifically, this includes: First, extracting core computational data, focusing on spatial grid points within the analysis loop that contain representative characteristic points, denoted as core grid points (i.e., 1 km × 1 km grids containing representative characteristic points). Extracting the proportion of advection transport contribution of each core grid point in each time period covered by the analysis loop, and the corresponding advection velocity vector intensity (i.e., the calculated magnitude of the advection velocity vector). Then, counting the total number of time periods for each core grid point within the analysis loop coverage period; for example, covering 9 hours with each time period lasting 10 minutes, the total number of time periods is 54. First, determine the duration of each time period. Second, calculate the average weighted advection transport contribution ratio of the core grid points. For each time period, calculate its duration percentage, which is the duration of that time period divided by the total duration of the time period. For example, if a single time period is 10 minutes and the total duration is 540 minutes, then the duration percentage is 10 / 540. Multiply the advection transport contribution ratio of each time period by the duration percentage of the corresponding time period to obtain the weighted advection transport contribution ratio of that time period. Add up the weighted advection transport contribution ratios of all time periods within the analysis loop coverage period to obtain the average weighted advection transport contribution ratio of the core grid points. This ratio reflects the average influence weight of advection transport on the changes in meteorological elements of the core grid points throughout the entire analysis period. Simultaneously, calculate the core... The average advection velocity vector intensity of the grid points is calculated by summing the advection velocity vector intensities of the core grid points for each time period. Dividing this sum by the total number of time periods yields the average advection velocity vector intensity, which reflects the average strength of atmospheric advection throughout the analysis period. Next, the dynamic weighting of atmospheric motion is quantified by multiplying the average weighted advection transport contribution ratio of the core grid points by the average advection velocity vector intensity to obtain the basic dynamic weighting. The basic dynamic weighting of all spatial grid points within the characteristic influence domain is extracted. Each grid point's basic weight is calculated using the above process, and the maximum value among all basic weights is selected as the maximum basic weight within the characteristic influence domain. The dynamic action weight is obtained by dividing the basic weight of the core grid point by the maximum basic weight within the characteristic influence domain. The weight value is strictly controlled between 0 and 1. The closer the weight value is to 1, the stronger the dynamic driving effect of atmospheric motion on the formation and evolution of extreme climate events at representative characteristic points. Finally, the rationality of the dynamic action weight is verified by comparing the dynamic action weight of the core grid point with the changing trend of the core meteorological element values ​​of the grid point. If the weight value is high and the core meteorological element values ​​show a rapid upward (or downward) trend in the corresponding period, while the advection transport contribution ratio remains at a high level and the advection velocity vector intensity remains strong, then the weight quantification result is deemed valid.If the weight values ​​do not match the changing trends of the elements, the calculation method for the duration proportion should be readjusted, and the average weighted contribution ratio, average velocity intensity, and dynamic action weights should be recalculated until the weights can truly reflect the dynamic driving contribution of atmospheric motion.

[0042] Based on the dynamic action weights, meteorological data calibration quantities are determined, and the standardized meteorological data sequence is dynamically corrected in time and space to obtain a calibrated meteorological element sequence. Based on the calibrated meteorological element sequence, key characteristic parameters representing extreme climate events are extracted to generate extreme climate indicators for risk monitoring. Specifically, this includes: first, determining the meteorological data calibration quantities grid-by-grid and time-by-time; for the core meteorological element values ​​of each spatial grid point and each time period in the standardized meteorological data sequence, first extracting the dynamic action weights corresponding to that grid point, then extracting the difference (element deviation value) between the core meteorological element field value of that grid point and the regional climate background field mean value of that time period, and using the dynamic action weights... The weights are multiplied by the element deviation values ​​to obtain the basic calibration quantity. Then, the anomaly sub-index corresponding to the grid point for the given time period is extracted, i.e., the calculated anomaly sub-index of the grid point for the given spatiotemporal unit, reflecting the degree to which meteorological elements deviate from normal at the grid point for the given time period. The anomaly sub-index is multiplied by the basic calibration quantity to obtain the preliminary spatiotemporal dynamic calibration quantity. Based on the magnitude of the dynamic action weight, the preliminary calibration quantity is adjusted to ensure that the calibration quantity matches the intensity of the dynamic action. For grid points with a dynamic action weight in the range of 0-0.3 (weak dynamic action, small influence of atmospheric motion), the preliminary calibration quantity is multiplied by an adjustment coefficient of 0.8 (to weaken the calibration effect and avoid over-calibration). For grid points with weights in the range of 0.3-0.7 (moderate dynamic effects, moderate influence from atmospheric motion), the initial calibration remains unchanged. For grid points with weights in the range of 0.7-1 (strong dynamic effects, significant influence from atmospheric motion), the initial calibration is multiplied by an adjustment factor of 1.2 (strengthening the calibration effect and accurately correcting deviations caused by dynamic effects). The adjusted data is the final spatiotemporal dynamic calibration, which fully reflects the differences in dynamic effects and the degree of meteorological element anomalies at different spatiotemporal locations (different grid points, different time periods), completely avoiding the deviation problems caused by traditional unified calibration. Secondly, the standardized meteorological data sequence is subjected to spatiotemporal dynamic... For dynamic correction, a calibrated meteorological element sequence is generated. The core meteorological element values ​​for each data point in the standardized meteorological data sequence, corresponding to each grid point and each time period, are extracted. These values ​​are then added to the corresponding spatiotemporal dynamic calibration values ​​to obtain the corrected core meteorological element values. During the correction process, the continuity of corrected values ​​for adjacent grid points in the same time period must be rigorously verified. The difference between the corrected values ​​of two adjacent grid points is calculated, along with the difference between the original values ​​of these two grid points. If the difference after correction exceeds 20% of the difference before correction, the adjustment coefficient is fine-tuned. For example, the adjustment coefficient for strong dynamic effects is changed from 1.2 to 1.1, and the adjustment coefficient for weak dynamic effects is changed from 0.8 to 0.9. Recalculate the calibration values ​​and correction values ​​until the difference between the correction values ​​of adjacent grid points meets the continuity requirement. Integrate all corrected values ​​that meet the continuity requirement according to the timestamp order and spatial grid coordinate order of the original data sequence, remove duplicate data points, and form a complete calibrated meteorological element sequence. This sequence can more realistically and accurately reflect the actual changes of meteorological elements during the evolution of extreme climate events. Then, based on the calibrated meteorological element sequence, extract three types of key characteristic parameters characterizing extreme climate events. When extracting event intensity parameters, focus on the calibrated meteorological element sequence within the analysis loop coverage period, and statistically analyze the core grid points and the grids where representative characteristic points are located. The maximum value of core meteorological elements within a given time period is used to extract the mean value of core grid points in the regional climate background field during the same period. Subtracting the mean value from the calibrated maximum value yields the intensity parameter of the extreme climate event, reflecting its severity. When extracting the duration parameter, the number of consecutive time periods within the statistical analysis loop coverage period where the core meteorological element values ​​exceed the extreme threshold (i.e., the set extreme high or low threshold for the corresponding meteorological element) is multiplied by the duration of each time period to obtain the duration parameter of the extreme climate event, reflecting its duration. When extracting the spatial extent parameter, the statistical analysis loop coverage area (characteristics...) Within the impact domain, the number of spatial grid points whose core meteorological element values ​​exceed the extreme threshold is calculated. This number is multiplied by the area of ​​each grid point, fixed at 1 square kilometer. Since the grid is 1 km × 1 km, this yields the spatial extent parameter of the extreme climate event's impact, reflecting the breadth of the event's influence. Finally, key characteristic parameters are integrated to generate an extreme climate index. The three key characteristic parameters—event intensity, duration, and spatial extent of impact—are weighted and fused. The weights of the three parameters are pre-set based on the importance of extreme climate risk monitoring: intensity parameter weight 0.5, duration parameter weight 0.3, and spatial extent of impact parameter weight 0.2, with a total weight of 1. The intensity parameter is multiplied by 0.5 to obtain an intensity-weighted value, the duration parameter is multiplied by 0.3 to obtain a duration-weighted value, and the impact spatial range parameter is multiplied by 0.2 to obtain a range-weighted value. These three weighted values ​​are then summed to obtain a comprehensive extreme climate index. Simultaneously, specific extreme climate indicators are generated for each type of key characteristic parameter: intensity-specific indicators, duration-specific indicators, and range-specific indicators. The comprehensive index and these three types of specific indicators together constitute an extreme climate indicator system for risk monitoring. Each indicator is clearly associated with its corresponding spatiotemporal range and core meteorological element type, providing accurate and reliable indicator support for extreme climate risk level assessment and early warning threshold setting.

[0043] By refining the construction process of the regional climate background field, calculating the advection velocity vector and diffusion rate scalar of each spatial grid point, the dynamic characteristics and spatial distribution patterns of meteorological element movement are clearly analyzed, thereby improving the accuracy of meteorological data.

[0044] In a preferred embodiment of the present invention, based on extreme climate indicators and combined with underlying surface feature data and disaster record data of the target area, a dynamic risk assessment model is used to analyze the risk level of extreme climate events, and multi-level warning thresholds are set according to preset climate state classification standards, which may include:

[0045] In this embodiment of the invention, extreme climate indicators, geospatial underlying surface feature data of the target area, and past disaster record data are spatiotemporally aligned and fused to form a standardized comprehensive assessment dataset. Specifically, this includes: first, conducting data collection to determine the specific content and source of various data types. Extreme climate indicators include quantitative data corresponding to key characteristic parameters such as the intensity, duration, and spatial extent of previously generated extreme climate events. Geospatial underlying surface feature data of the target area includes the region's topography, vegetation cover, soil type and moisture characteristics, water system distribution density, population density distribution, and infrastructure distribution. The data includes geospatial data such as location and density, and land use type; and historical disaster records, including the occurrence time, impact range, economic losses, affected area, number of casualties, and infrastructure damage of various extreme weather events that have occurred in the target area throughout history. Secondly, a spatiotemporal alignment operation is performed. In terms of time dimension, the three types of data are unified and standardized to the same time granularity. If the time granularity of extreme weather indicators is hourly, the underlying surface feature data is annual static data, and the historical disaster records are event-level, then the underlying surface feature data is associated with the corresponding time range by year, and the historical disaster records are... The data is broken down into hourly time units to ensure that the corresponding information of the three types of data is contained within the same time unit. Spatially, using a standard geographic grid of the target area as a benchmark, all three types of data are matched into this grid system. For extreme climate indicators, the indicator values ​​within each grid are statistically analyzed. For underlying surface feature data, the underlying surface attributes corresponding to each grid are determined. For past disaster record data, the disaster impact range is mapped to the corresponding grid, and the historical disaster situation of each grid is marked to achieve accurate spatial matching of the three types of data. Then, data standardization processing is carried out. For extreme climate indicators, different types of indicators are standardized. The indicators are standardized based on their historical statistical range in the target area to ensure that the values ​​of different indicators are within the same comparable range. Specifically, the standardized value of a certain extreme climate indicator is calculated as follows: (current original value of the indicator - historical minimum value of the indicator) ÷ (historical maximum value of the indicator - historical minimum value of the indicator). This method eliminates the dimensional differences between different indicators. For categorical data in geospatial underlying surface feature data, a coding method is used for standardization, assigning a unique standardized code to each type. For continuous data, the same standardization method as for extreme climate indicators is used to convert the values ​​to the same comparable range.For past disaster records, continuous data such as economic losses and affected areas are converted according to the overall economic scale and total area of ​​the target region. The number of casualties is converted according to the total population of the corresponding grid. Categorized data is also standardized and coded. Finally, data fusion is performed, using the aligned spatiotemporal grid as the core. Standardized extreme climate indicators, standardized underlying surface characteristic data, and standardized past disaster records corresponding to each grid unit are linked and integrated to form a standardized comprehensive assessment dataset where each record corresponds to a spatiotemporal grid unit and contains complete information on all three types of data. During the fusion process, it is necessary to ensure that all types of data within the same spatiotemporal grid unit correspond one-to-one, with no missing or mismatched data. For a small number of missing data points, the average of the same type of spatiotemporal grid units is used to supplement and improve the dataset, ensuring its completeness.

[0046] The comprehensive assessment dataset is input into the dynamic risk assessment model to calculate the quantitative value of the probability of extreme climate events causing disasters in the target area, as well as the quantitative value of the potential impact represented by the vulnerability and exposure of disaster-bearing bodies. Specifically, the dynamic risk assessment model is constructed using a multilayer perceptron structure, consisting of an input layer, hidden layers, and an output layer. The specific construction process is as follows: The input parameters of the input layer are various standardized data from the comprehensive assessment dataset, specifically including standardized values ​​of intensity, duration, and impact spatial range in extreme climate indicators; standardized values ​​of terrain coding, vegetation cover, population density, infrastructure density, and land use type coding in underlying surface feature data; and standardized values ​​of historical economic loss percentage, historical affected area percentage, and historical casualty percentage in past disaster record data. Each input parameter corresponds to one neuron in the input layer. Three hidden layers are set, with the number of neurons in each layer determined by the number of input parameters. For example, if there are 12 input parameters, the number of neurons in each layer is set to 24. Each neuron in each layer uses the ReLU activation function to enhance the model's nonlinear fitting ability. Layers are fully connected. The model is connected in a specific manner, with the output value of each neuron serving as the input value for all neurons in the next layer. The output layer has two neurons, corresponding to the quantified value of the probability of disaster and the quantified value of the potential impact, respectively. The output layer uses the Sigmoid activation function to ensure the output result is between 0 and 1, facilitating quantitative assessment. After the model is built, it is trained using comprehensive historical assessment data from the target area over the past 30 years, including samples of extreme weather disasters and samples of no disasters. During training, the connection weights and biases of neurons in each layer are continuously adjusted until the model's prediction error reaches the preset requirements. The dynamic risk assessment model is constructed and calibrated. Next, the quantification value of the probability of disaster is calculated. The completed comprehensive assessment dataset is then fully input into the trained dynamic risk assessment model. After receiving various input parameters, the model's input layer passes them to the first hidden layer. Each neuron in the first hidden layer performs a weighted summation of the received input parameters, that is, each input parameter is multiplied by its corresponding connection weight and then summed, plus the neuron's bias value, and the result is output through the ReLU activation function. The output of the first hidden layer is passed to the second hidden layer, and the above weighted summation and activation operations are repeated, sequentially passing to the third hidden layer.The output of the third hidden layer is passed to the first neuron of the output layer. This neuron performs a weighted summation of the received input parameters, multiplying each input parameter by its corresponding connection weight and adding the sum, along with a bias value. The result is then output through the Sigmoid activation function. This result is the quantified value of the probability of an extreme weather event causing disaster in the target area. During the calculation of this quantified value, the model focuses on the matching degree between extreme weather indicators and historical disaster records. If the standardized value of the current extreme weather indicator has a high similarity to the standardized values ​​of indicators that led to disasters in the past, and the corresponding area has a high frequency of historical disasters, the quantified value of the probability of disaster will be correspondingly higher. This is achieved through the weight allocation trained by the model. For example, a higher weight corresponding to the frequency of historical disasters has a greater impact on the final quantified value. Finally, the quantified value of the potential impact degree is calculated. This quantified value is obtained by weighted fusion of the quantified value of the vulnerability of the disaster-bearing body and the quantified value of the exposure of the disaster-bearing body. The specific calculation process is as follows: After receiving the output of the third hidden layer, the second neuron of the model's output layer first extracts the vulnerability and exposure degree of the disaster-bearing body respectively. The weighted results corresponding to the relevant input parameters include: vulnerability-related parameters such as standardized population density, standardized infrastructure density, land use type coding, and standardized vegetation cover from the underlying surface feature data, and standardized historical economic loss ratio from past disaster record data; and exposure-related parameters such as topographic coding, standardized water system distribution density, and standardized population density from the underlying surface feature data, and standardized impact spatial range from extreme climate indicators. The model first performs a weighted summation of vulnerability-related parameters (each parameter multiplied by its corresponding vulnerability weight and then summed) to obtain a quantified vulnerability value for the affected body; then it performs a weighted summation of exposure-related parameters (each parameter multiplied by its corresponding exposure weight and then summed) to obtain a quantified exposure value for the affected body; finally, the vulnerability quantified value is multiplied by the vulnerability weight coefficient, and the exposure quantified value is multiplied by the exposure weight coefficient. The sum of the vulnerability weight coefficient and the exposure weight coefficient is 1, determined through training with historical data. The result, after processing with the Sigmoid activation function, is the quantified value of the potential impact.

[0047] Based on the quantified probability value and the quantified potential impact value, a pre-set risk matrix is ​​used for comprehensive evaluation to classify the corresponding risk levels. Simultaneously, based on long-term climate statistics and climate state classification standards, threshold boundaries for one or more key extreme climate indicators that trigger different levels of warnings are determined, and multi-level warning thresholds matching each risk level are set. Specifically, this includes: first, constructing a pre-set risk matrix. The horizontal axis of the risk matrix represents the level classification of the quantified probability of disaster, and the vertical axis represents the level classification of the quantified potential impact value. The horizontal axis is divided into four levels: low probability (quantified probability value 0-0.25), medium probability (0.25-0.5), high probability (0.5-0.75), and very high probability (0.75-1.0); the vertical axis is divided into four levels: small impact (quantified potential impact value 0-0.25), medium impact (0.25-0.5), large impact (0.5-0.75), and extremely large impact (0.75-1.0). Each cell in the risk matrix corresponds to a risk level, specifically, low probability + small impact = low risk. Risk is assessed using the following risk levels: Low probability + medium impact, medium probability + small impact = relatively low risk; Low probability + large impact, medium probability + medium impact, high probability + small impact = medium risk; Medium probability + large impact, high probability + medium impact, very high probability + small impact = relatively high risk; High probability + large impact, very high probability + medium impact, very high probability + large impact = very high risk. This completes the pre-construction of the risk matrix. Next, a comprehensive risk level assessment is performed. The calculated quantified values ​​of the probability of disaster and the potential impact are extracted and mapped to the level intervals on the horizontal and vertical axes of the risk matrix, respectively. The risk matrix cell corresponding to the intersection of two levels is identified. The risk level corresponding to this cell is the risk level of the current extreme weather event in the target area. For example, if the quantified value of the probability of disaster is 0.6 (corresponding to a high probability level) and the quantified value of the potential impact is 0.7 (corresponding to a large impact level), then the risk level corresponding to the intersection of the two is relatively high risk; if the quantified value of the probability of disaster is 0.8 (corresponding to a very high probability level) and the quantified value of the potential impact is 0...A value of 9 (corresponding to a level of extreme impact) indicates an extremely high risk level. During the assessment, if the quantified value falls precisely at the threshold between two levels, the higher level will be applied to ensure the rigor of the risk assessment. Then, long-term climate statistics are collected, and climate state classification standards are determined. Long-term climate statistics include historical data on extreme climate indicators for the target area over the past 30 years, such as the maximum 24-hour precipitation during heavy rainstorms, the number of consecutive days with the highest temperatures during high temperatures, and the minimum temperature during cold waves. Historical climate state data includes average temperature, average precipitation, and the occurrence of extreme climate events for each year. The climate state classification criteria are preset to four levels: normal climate state, slightly abnormal climate state, abnormal climate state, and extreme abnormal climate state. The classification is based on the statistical results of long-term climate data. For example, a climate indicator within ±1 standard deviation of its multi-year average is considered a normal state, within ±1-2 standard deviations is a slightly abnormal state, within ±2-3 standard deviations is an abnormal state, and exceeding ±3 standard deviations is an extreme abnormal state. Finally, multi-level early warning thresholds are set. First, key extreme climate indicators are screened from extreme climate indicators, with the screening criterion being that they play a leading role in causing disasters by extreme climate events. Indicators such as 24-hour precipitation corresponding to heavy rain, consecutive days of high temperatures corresponding to high temperatures, and maximum wind speed corresponding to typhoons are used. Each type of extreme climate event corresponds to 1-3 key extreme climate indicators. For each selected key extreme climate indicator, based on long-term climate statistics and climate state classification standards, threshold boundaries corresponding to different risk levels are determined. The low-risk level corresponds to the critical value between normal and slightly abnormal climate states; that is, the maximum value of the key indicator in a normal state is used as the lower limit of the low-risk warning threshold, and the maximum value in a slightly abnormal state is used as the upper limit. The relatively low-risk level corresponds to the critical value between slightly abnormal and abnormal climate states; that is, the maximum value in a slightly abnormal state is used as the lower limit, and the maximum value in an abnormal state is used as the upper limit. The medium-risk, relatively high-risk, and extremely high-risk levels correspond to the internal range of abnormal climate states, the critical value between abnormal and extreme anomalies, and the internal range of extreme anomalies, respectively. Upper and lower limits of thresholds are determined for each level. Simultaneously, it is ensured that the warning threshold corresponding to each risk level matches the risk level of that level; the higher the risk level, the more extreme the value of the key extreme climate indicator corresponding to the warning threshold. This ultimately forms a multi-level warning threshold system that matches each risk level one-to-one.

[0048] Achieving in-depth integration and standardized processing of multi-source data ensures the integrity of the data used for risk analysis, making risk level classification more targeted, accurately reflecting the risk characteristics of extreme weather events under different temporal and spatial conditions, and improving the pertinence of disaster prevention and mitigation work.

[0049] like Figure 2As shown, in a preferred embodiment of the present invention, real-time meteorological data is compared with multi-level early warning thresholds via a cloud platform. When the real-time monitoring data exceeds the multi-level early warning thresholds, an early warning message is automatically generated and pushed to a designated terminal. This may include:

[0050] In this embodiment of the invention, the data stream processing service of the cloud platform continuously accesses real-time meteorological observation data or short-term forecast data streams from the target area and processes the data streams to form a unified real-time monitoring dataset. Specifically, this includes: first, the data stream processing service of the cloud platform establishes a dedicated data access channel, which operates 24 hours a day, continuously monitoring and accessing real-time meteorological observation data and short-term forecast data streams sent by various data sources within the target area. The real-time meteorological observation data originates from ground meteorological observation stations, upper-air sounding stations, radar monitoring stations, and satellite remote sensing monitoring facilities deployed within the target area. The system includes various data types, such as temperature, precipitation, wind speed, wind direction, air pressure, humidity, and visibility. Short-term forecast data streams originate from 0-6 hour forecasts issued by regional meteorological forecast centers, and their data types are consistent with real-time meteorological observation data, supplementing the temporal continuity of real-time observation data. After the data streams are accessed, they undergo format standardization. Addressing the differences in data formats from different data sources, the data stream processing service converts various formats into a unified structured data format according to a pre-defined unified data format standard, ensuring that all data field definitions and types are consistent. Unit identifiers must be consistent. For example, all precipitation data should be converted to millimeters, and wind speed data to meters per second. Then, data quality verification should be performed. For the converted structured data, each data item's value should be checked against preset reasonable thresholds. For example, the reasonable threshold for temperature data is set to -60℃ to 60℃, and the reasonable threshold for hourly precipitation data is set to 0 mm to 200 mm. If a data item's value exceeds the reasonable threshold, it is marked as abnormal data. Simultaneously, missing data in the data series should be checked. For a single missing data point in a continuous time series, [further steps should be taken]. The interpolated data is calculated by adding the values ​​of two adjacent valid data points and dividing by 2 to fill in the missing positions. For two or more consecutive missing data segments, they are marked as invalid segments and stored separately, not included in subsequent real-time monitoring datasets. Finally, data normalization is performed: valid data that has passed quality verification is sorted chronologically according to the data collection timestamps to form a time-ordered data sequence. Simultaneously, the data is spatially categorized according to the administrative region or grid division of the target area, ensuring that each administrative region or grid unit corresponds to a complete meteorological element data sequence. After these three steps of format standardization, quality verification, and data normalization, a unified real-time monitoring dataset is finally formed, containing valid meteorological element data for each spatial unit and time point within the target area.

[0051] The system performs real-time matching and hierarchical comparison between real-time monitoring datasets and multi-level early warning thresholds. When a key indicator in the real-time monitoring dataset exceeds any preset early warning threshold, a corresponding level of early warning trigger signal is generated. Specifically, this involves: first, determining the key indicator data in the real-time monitoring dataset. The key indicator data is determined based on the type of extreme weather event. For example, for rainstorm events, the key indicator data includes hourly precipitation, 6-hour cumulative precipitation, and 24-hour cumulative precipitation; for typhoon events, the key indicator data includes maximum wind speed, average wind speed, and air pressure. Each extreme weather event type corresponds to a specific list of key indicators, and the key indicators must be completely consistent with the indicator types of the preset multi-level early warning thresholds. Then, real-time matching is performed, matching each key indicator in the real-time monitoring dataset with the preset multi-level early warning thresholds, such as blue, yellow, orange, and red warning thresholds, ensuring that each key indicator corresponds to its specific multi-level early warning threshold system and avoiding mismatches between different indicators and different threshold systems. For example, the hourly precipitation data in the real-time monitoring dataset is matched with the hourly precipitation data corresponding to a rainstorm event. The system matches hourly precipitation warning thresholds with different levels, but does not correlate them with wind speed warning thresholds. After matching, the system compares the data level by level, from the lowest to the highest level. First, it compares the key indicator data with the lowest level (e.g., blue) warning threshold to determine if the key indicator data exceeds that level. If not, normal monitoring continues, and the system continues to compare data at the next time point. If the data exceeds the threshold, a warning for that level is preliminarily triggered, and the system continues to compare the key indicator data with the next higher level (e.g., yellow) warning threshold. If the key indicator data exceeds the yellow warning threshold, the system continues to compare the orange warning threshold, and so on, until all levels of warning thresholds are compared. During the comparison process, if a key indicator in the real-time monitoring dataset exceeds any preset warning threshold (i.e., the key indicator data exceeds that level), a warning trigger signal for the corresponding level is immediately generated. The warning trigger signal contains the following information: warning level, name of the key indicator that triggered the warning, real-time value of the key indicator, data acquisition time, spatial range of the warning trigger, and timestamp of the warning trigger. All information is generated in a structured format.

[0052] Based on the warning trigger signal, a preset warning information template matching the warning level, event type, and affected area is automatically invoked. The template is then populated with relevant intensity, location, and evolution trend information from the real-time monitoring data, dynamically generating warning information. Specifically, this involves: first, parsing the generated warning trigger signal to extract the warning level and the name of the key indicator that triggered the warning. The type of extreme weather event is determined by combining the key indicator name; for example, if the key indicator is hourly precipitation, the event type is heavy rain; if the key indicator is maximum wind speed, the event type is typhoon. Simultaneously, the spatial range information triggering the warning is extracted to clarify the affected area. Based on the parsed warning level, event type, and affected area information, the cloud platform automatically searches the preset warning information template library and invokes a warning information template that perfectly matches the above three pieces of information. The warning information template library stores exclusive templates for different event types, warning levels, and affected areas, and the templates have reserved space for... Fixed filling areas are provided for intensity information, location information, evolution trend information, and defense suggestion. The content of defense suggestions for different templates varies according to the event type and warning level. For example, the defense suggestion for a blue rainstorm warning focuses on reminding people to carry rain gear when going out, while the defense suggestion for a red rainstorm warning focuses on reminding people to evacuate and take shelter. Subsequently, intensity, location, and evolution trend information related to the warning event are extracted from the real-time monitoring dataset. In terms of intensity information extraction, the real-time values ​​of the key indicators that triggered the warning are extracted, and the values ​​of the indicators at the three most recent consecutive time nodes are also extracted, such as the values ​​at each node every 10 minutes in the last 30 minutes. The change range of the indicator is calculated, that is, the current value minus the value at the earliest time node. The real-time value and the change range are used together as intensity information. In terms of location information extraction, the spatial range (administrative region or grid unit) that triggered the warning is converted into easily understandable geographical description information, and key geographical markers in the area are added as location information.In terms of extracting evolution trend information, based on the numerical change pattern of the key indicator in the real-time monitoring dataset over the most recent five consecutive time nodes, and combined with the forecast values ​​of the indicator for the next two time nodes in the short-term forecast data, the evolution trend of the indicator is determined. If the value continues to increase, it is determined to be an strengthening trend; if the value remains stable, it is determined to be a maintaining trend; and if the value continues to decrease, it is determined to be a weakening trend. Simultaneously, the expected rate of change of the indicator in the future is calculated, i.e., the difference between the future forecast value and the current value divided by the time interval. The evolution trend judgment result and the expected rate of change are combined as evolution trend information. Finally, the extracted intensity information, location information, and evolution trend information are filled into the corresponding fill positions of the early warning information template. At the same time, the preset defense suggestions in the template are called according to the event type and early warning level. After the template is filled, a complete early warning message is automatically generated. The generated early warning message must ensure that the language is easy to understand, the information is complete and accurate, and includes core content such as event type, early warning level, location, intensity, evolution trend, and defense suggestions, without any missing or ambiguous information.

[0053] The warning information is distributed in real time to relevant emergency command terminals, professional department response terminals, and public information release platform terminals according to preset push rules. Specifically, this includes: first, determining the preset push rules, which are based on factors such as terminal type, the responsibilities of the department to which the terminal belongs, the warning level, and the affected area. Specific rules include: different warning levels correspond to different push terminal ranges; different professional department response terminals push warning information related to their respective responsibilities; the same warning information has different push priorities for different terminals; for terminals within the affected area, only warning information relevant to that area is pushed, and warning information from outside the area is not pushed. Before pushing, the generated warning information undergoes a final verification to check whether the core content of the warning information is complete and accurate, and whether the format of the warning information meets the reception requirements of each terminal. If there are errors or format inconsistencies, the process immediately returns to the previous step for correction. After correction, the verification is repeated until the warning information meets the push requirements. After successful verification, it is distributed in real time according to the push rules: for emergency command terminals, through dedicated... The system uses encrypted communication channels to push warning information, ensuring the security and confidentiality of the transmission. After the push is completed, the receiving terminal automatically sends a confirmation message, and the cloud platform records the push time and reception status. For terminals handled by specialized departments, the warning information is pushed through the government intranet or a dedicated communication link, depending on the department type. Each specialized department terminal only receives warning information relevant to its responsibilities, and a push log is recorded synchronously after each push. For terminals on public information release platforms, the warning information is converted into a text or audio format suitable for public reception and pushed precisely according to the affected area, while ensuring the timeliness of the push information. The time from the generation of the warning information to the public terminal receiving it does not exceed the preset time. After the push is completed, the cloud platform summarizes and records the push status of all terminals, including the push terminal name, push time, reception status, and information reading status, forming a complete push ledger for easy traceability and verification. At the same time, for terminals that fail to receive the warning information, a second push is automatically performed. If the second push is still unsuccessful, an alarm message is immediately sent to the operation and maintenance management terminal to remind operation and maintenance personnel to troubleshoot the fault.

[0054] It achieves unified integration of meteorological data of different types and formats, removes abnormal data and fills in missing data, ensuring the integrity, accuracy and uniformity of real-time monitoring datasets, avoiding early warning deviations caused by data inconsistencies or data quality issues, and can determine the matching relationship between key indicator data and early warning thresholds at all levels, ensuring the timely generation of early warning trigger signals. This avoids the omission of low-level early warnings and prevents the delayed triggering of high-level early warnings, thus improving the timeliness of early warning triggering.

[0055] In a preferred embodiment of the present invention, a monitoring framework is constructed based on cloud-native architecture and component-based design, and distributed collaborative operation and dynamic monitoring are achieved using early warning information, which may include:

[0056] In this embodiment of the invention, a monitoring framework is constructed based on a cloud-native architecture using containerization and microservices. Within this framework, data collection, data processing, data calibration, risk assessment, and early warning generation functions are encapsulated as independent and elastically scalable microservice components and deployed accordingly. Specifically, this includes: First, building a containerized deployment environment by constructing a container cluster based on a mainstream container engine. The cluster contains multiple node servers, each configured with preset CPU, memory, and storage resources. A container orchestration tool is also deployed to automate container deployment, scheduling, scaling, and management. After the container cluster is built, a basic support layer for the cloud-native architecture is constructed based on this cluster. This basic support layer includes a service registry, a configuration center, a load balancer, a log collection component, and a monitoring and alerting component. The service registry records the registration information and network addresses of all microservice components; the configuration center manages the runtime configuration parameters of all components; the load balancer distributes client requests and data traffic; the log collection component aggregates the runtime logs of each component; and the monitoring and alerting component monitors the resources of each component in real time. After assessing the usage and operational status, the data acquisition function was independently encapsulated. The complete business process of the data acquisition function was analyzed, including data source access, data reception, preliminary data verification, and data forwarding. All business logic in this process was broken down into multiple independently runnable sub-logic units, each corresponding to a dedicated piece of functional code. All sub-logic units were integrated and packaged into an independent container image. This image contained all the dependencies and configuration files required for the data acquisition function to run, and was marked as the data acquisition microservice component. During the encapsulation process, basic parameters related to elastic scaling were configured for this component, including the minimum number of running instances, the maximum number of running instances, and scaling trigger conditions. The scaling trigger conditions were set as follows: when the CPU utilization of this component exceeds a preset threshold (e.g., 70%) for 5 consecutive minutes, a scaling-up operation is triggered, with the number of instances added each time being the current number of running instances multiplied by 0.5 and rounded up; when the CPU utilization is less than a preset threshold (e.g., 30%) for 10 consecutive minutes, a scaling-down operation is triggered, with the number of instances reduced each time being the current number of running instances multiplied by 0.Rounding down from 3 ensures the component can flexibly adjust the number of running instances based on data acquisition pressure. Following the same encapsulation process described above, data processing, data calibration, risk assessment, and early warning generation functions are independently encapsulated. When encapsulating the data processing function, business logic including data format conversion, data quality verification, data normalization, and data regularization is included, packaged into a data processing microservice component. It is also configured with elastic scaling parameters, with scaling triggering conditions based on memory usage and data processing throughput. Specifically, scaling is triggered when memory usage exceeds 75% for 5 consecutive minutes or data processing throughput exceeds the preset processing limit for 5 consecutive minutes. When encapsulating the data calibration function, business logic including analysis loop construction, motion feature calculation, dynamic action quantification, and dynamic data correction is included, packaged into a data calibration microservice component. Scaling triggering conditions are based on CPU usage and calibration task completion time. When encapsulating the risk assessment function, business logic including multi-source data fusion, disaster probability calculation, potential impact assessment, and risk level classification is included, packaged into a risk assessment microservice group. The early warning generation function is encapsulated into a microservice component, encompassing business logic such as threshold comparison, early warning trigger signal generation, early warning information template invocation, and early warning information filling. Both the early warning generation and early warning generation microservices are configured with corresponding elastic scaling parameters according to their respective business pressure characteristics. After all microservice components are encapsulated, they are deployed to a container cluster using container orchestration tools. First, the container images of each component are uploaded to the cluster's image repository. The service registry registers each component, recording its service name, network address, and interface information. The configuration center distributes the runtime parameters of each component to the corresponding component. The load balancer configures the traffic distribution strategy for each component, ensuring that data requests are distributed to different running instances of each component according to a preset ratio. After deployment, each microservice component runs independently, achieving configuration synchronization and service discovery through the service registry and configuration center. Based on the resource scheduling capabilities of the container cluster, elastic scaling of each component is achieved, ultimately building a complete cloud-native architecture monitoring framework. Each functional component within the framework can be independently extended and maintained without affecting others.

[0057] Utilizing early warning information as internal event trigger signals drives dynamic updates to the risk assessment model and simultaneously triggers adaptive adjustments to early warning thresholds. Specifically, this involves: First, establishing an early warning information parsing channel. This channel establishes a real-time connection with the output of the early warning generation function, continuously receiving generated early warning information. The early warning information is then structured and parsed to extract core information, including the early warning level, extreme weather event type, event occurrence time, affected area, real-time values ​​of key indicators, event evolution trends, and the impact on disaster-bearing entities. This core information is then organized into standardized event trigger data, serving as the internal event trigger signal to drive subsequent updates and adjustments, ensuring that the trigger signal contains sufficient... The update is based on the event's trigger signal, initiating a dynamic update process for the risk assessment model. First, the extreme weather event type, real-time values ​​of key indicators, and event evolution trend data are extracted from the trigger signal. Combined with historical data on similar events in the event's location, a model update dataset is constructed. For core assessment parameters in the risk assessment model, such as disaster probability assessment parameters and vulnerability assessment parameters, a parameter correction algorithm is used for dynamic adjustment. Specifically, the correction method involves comparing the current event's real-time key indicator values ​​with the average values ​​of key indicators from historical similar events, calculating the difference, multiplying this difference by a preset correction coefficient to obtain the parameter adjustment amount, and then adding the original assessment parameters. The adjusted amount yields the updated assessment parameters. For the weight parameters in the model, adjustments are made based on the impact of the current event on the affected entities. If the impact of the current event on a certain type of affected entity is greater than the historical average, the weight parameter corresponding to that type of affected entity is increased by multiplying the original weight by a preset ratio; conversely, the weight parameter is appropriately reduced to ensure that the weight parameters accurately reflect the actual situation of the affected entities in the current region. Simultaneously, while driving the risk assessment model update, an adaptive adjustment process for the warning threshold is initiated. This involves first extracting the warning level, real-time values ​​of key indicators, event evolution trends, and underlying surface characteristics of the affected area from the trigger signal, and combining this with historical warning effect feedback data, i.e., historical... To assess the adaptability of warning thresholds under the same warning level and event type, and to determine whether there are insufficient or excessive warnings, a threshold adjustment dataset is constructed. For different warning levels, the thresholds are adaptively adjusted as follows: For the key indicator threshold that triggers the current warning, the difference between the real-time value of the current key indicator and the threshold of that level is calculated. If the difference is greater than a preset deviation threshold, meaning the current indicator is far beyond the threshold and the warning is triggered too late, the warning threshold for that level is lowered by multiplying the original threshold by a preset reduction ratio. If the difference is less than a negative deviation threshold, meaning the current indicator has just reached the threshold and the warning is triggered too early, the warning threshold for that level is raised by multiplying the original threshold by a preset increase ratio.Simultaneously, thresholds are adjusted based on underlying surface characteristics data of the affected areas. For example, in mountainous areas with complex terrain, the risk of disaster from extreme events of the same intensity is higher due to their weaker disaster-bearing capacity. Therefore, the warning thresholds for each level in these areas are appropriately lowered by multiplying the original threshold by a regional correction coefficient. The regional correction coefficient is set according to the region's disaster-bearing capacity; the weaker the disaster-bearing capacity, the smaller the correction coefficient. In densely populated urban areas, to avoid excessive warnings affecting normal production and life, the warning thresholds are appropriately raised by multiplying the original threshold by the regional correction coefficient. After the threshold adjustments are completed, the updated risk assessment model parameters and the adaptively adjusted warning thresholds are synchronously stored in the configuration center and distributed to the corresponding risk assessment microservice component and warning generation microservice component. This ensures that risk assessment and warning generation are both based on the updated parameters and thresholds, achieving dynamic adaptation between the model and the thresholds.

[0058] By leveraging event-driven mechanisms and service calls between components, distributed collaborative operation across regions and departments is achieved, enabling dynamic and continuous monitoring of extreme weather risks. Specifically, this includes: First, constructing an event-driven mechanism with generated internal event trigger signals as the core event source. Simultaneously, the operational status information of each microservice component, data source access signals from cross-regional monitoring nodes, and cross-departmental business demand signals are all incorporated into the event system, establishing a unified event bus. The event bus is responsible for the reception, forwarding, subscription, and trigger management of all events. Each microservice component, cross-regional monitoring node, and cross-departmental terminal interacts with events through the event bus. Configuration of each component and... The event subscription and triggering rules for each node are as follows: The data acquisition microservice component subscribes to data source access events from cross-regional monitoring nodes. Upon receiving such an event, it automatically triggers the data acquisition function, initiating data access and preliminary processing from the corresponding data source. The data processing microservice component subscribes to the data acquisition component's acquisition completion event. Upon receiving such an event, it automatically invokes its own data processing function to standardize the accessed data. The data calibration microservice component subscribes to the data processing component's processing completion event, triggering the data calibration process. The risk assessment microservice component subscribes to the data calibration component's calibration completion event and early warning information trigger event, combining both information to initiate... The system dynamically assesses risks and generates early warnings. A microservice component subscribes to the risk assessment component's completion event, triggering early warning generation. Simultaneously, cross-department terminals subscribe to events related to their responsibilities, automatically receiving relevant information and triggering terminal alerts when events occur. Standardized service call interfaces are established between the various microservice components. Each component provides its own dedicated service call interface, containing clearly defined input parameters, output parameters, and calling protocols, ensuring cross-regional and cross-network service calls between different components. For example, when the risk assessment microservice component needs to call calibrated data from the data calibration component, it queries the data calibration component's network through the service registry. The network address sends a data retrieval request to its service interface according to a preset calling protocol. The request includes input parameters such as the time range, spatial range, and indicator type of the required data. After receiving the request, the data calibration component retrieves the corresponding calibrated data, organizes it according to the output parameter format, and returns it to the risk assessment component to complete the service call. Cross-regional monitoring nodes also realize service calls through standardized interfaces. For example, when the risk assessment component of the monitoring node in location A needs to call the historical data of the monitoring node in location B, it initiates a call request through the cross-regional service gateway. The service gateway is responsible for routing forwarding and communication encryption to ensure the security and stability of cross-regional data interaction.A distributed collaborative operation system is constructed, integrating monitoring resources from different regions and departments into a unified collaborative framework. Resource sharing and business collaboration are achieved through an event bus and service call interfaces. At the cross-regional level, monitoring nodes in each region share local real-time monitoring data, risk assessment results, and early warning information in real time. When an extreme weather event occurs in a certain region, monitoring nodes in surrounding areas automatically receive event information, initiate local collaborative monitoring, and predict in advance whether the event will spread to their respective regions. At the cross-departmental level, terminals and microservice components of relevant departments such as emergency command departments, meteorological departments, water resources departments, and transportation departments are interconnected. After the meteorological department's early warning generation component generates early warning information, it automatically calls the information receiving service of the emergency command department and the flood control monitoring service of the water resources department to simultaneously push early warning information, driving each department to initiate corresponding emergency response and special monitoring work. To achieve dynamic and continuous monitoring, a 7×2 system is established. A 4-hour uninterrupted monitoring mechanism ensures that each microservice component reports its operational status data to the monitoring and alarm component in real time. The monitoring and alarm component analyzes this data in real time. If the operational status data of a component exceeds the preset normal range, an abnormal event is immediately generated and pushed to the operations and maintenance management terminal via the event bus. The component's backup instance is also automatically invoked to ensure uninterrupted business operations. Simultaneously, the monitoring framework periodically initiates full-process inspection tasks, invoking the self-inspection services of each component to check whether the entire process of data collection, processing, calibration, evaluation, and early warning is running normally. If a process breakpoint is detected, an abnormal event is immediately generated, triggering an automatic repair mechanism. If automatic repair fails, an alarm is pushed to operations and maintenance personnel. Through this event-driven, service-invoking, cross-domain and cross-departmental collaboration, and uninterrupted monitoring, dynamic and continuous monitoring of extreme weather risks is ultimately achieved, ensuring timely detection of risk changes and efficient collaborative handling.

[0059] By building a cloud-native architecture and encapsulating and deploying microservice components, the monitoring framework achieves independence and elasticity for each function. Each component can dynamically adjust its operating resources according to business pressure. At the same time, the independent deployment and maintenance characteristics reduce the impact of single-function failures on the overall framework, thereby improving the stability and scalability of the monitoring framework.

[0060] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0061] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A meteorological-based real-time monitoring system for extreme climate risks, characterized in that: include: The generation module processes real-time collected meteorological and environmental data to form standardized meteorological data sequences and set benchmark feature points, generating two independent associated feature sets; it determines the feature influence domain based on the relationship between the two associated feature sets, and selects a representative feature point within the feature influence domain. The calibration module is used to construct an analysis loop based on the temporal evolution relationship of representative feature points. By calculating the motion velocity of meteorological elements in the feature influence domain relative to the regional background field, it analyzes the contribution of spatial variation and transport process of elements, quantifies the dynamic effect of motion on extreme climate evolution, calculates the calibration amount of meteorological data, performs dynamic calibration on the standardized meteorological data sequence, obtains the calibrated meteorological element sequence, and generates extreme climate indicators. This includes using representative feature points as spatiotemporal starting points, tracing the evolution trajectories of corresponding meteorological elements forward and backward along the time axis, and connecting the forward and backward trajectories to form an analysis loop; within the characteristic influence domain, calculating the advection velocity vector of the meteorological element field at each spatial grid point relative to the regional climate background field during the time period covered by the analysis loop; and simultaneously calculating the diffusion rate scalar of the meteorological element field at each spatial grid point during the time period. Based on the advection velocity vector and the diffusion rate scalar, the total rate of change of meteorological elements is calculated, and the total rate of change is decomposed into the contribution of change caused by the advection transport process and the contribution of change caused by the local diffusion process, and the contribution ratio of each is obtained. According to the contribution ratio of advection transport, combined with the intensity of the advection velocity vector, the dynamic weight of atmospheric motion on the formation and evolution of extreme climate events at representative characteristic points is quantified. Based on the dynamic action weight, the calibration amount of meteorological data is determined, and the standardized meteorological data sequence is dynamically corrected in time and space to obtain the calibrated meteorological element sequence. Based on the calibrated meteorological element sequence, key characteristic parameters characterizing extreme climate events are extracted to generate extreme climate indicators for risk monitoring. The setting module is used to analyze the risk level of extreme climate events based on extreme climate indicators, combined with the underlying surface feature data and disaster record data of the target area, through a dynamic risk assessment model, and to set multi-level early warning thresholds according to the preset climate state classification standards. The push module is used to compare real-time monitored meteorological data with multi-level early warning thresholds through the cloud platform. When the real-time monitored data exceeds the multi-level early warning thresholds, it automatically generates and pushes early warning information to the designated terminal. The monitoring module is used to build a monitoring framework based on cloud-native architecture and component-based design, and to use early warning information to achieve distributed collaborative operation and dynamic monitoring.

2. The meteorological-based real-time monitoring system for extreme climate risks according to claim 1, characterized in that, The real-time collected meteorological and environmental data are processed to form a standardized meteorological data sequence and a baseline feature point is set, generating two independent associated feature sets, including: Quality control is performed on real-time collected meteorological and environmental data. Based on preset threshold ranges and climatological limit rules, outliers in the raw data are identified and marked, and time series imputation is performed on missing values ​​to generate continuous data sequences. All data points in the continuous data sequence are arranged in chronological order according to the timestamps corresponding to the data points to form a time-ordered data sequence. Through maximum and minimum value normalization, the values ​​of each meteorological element in the time-ordered data sequence are transformed to a unified range to obtain a standardized meteorological data sequence. Based on meteorological data sequences, extreme value identification and trend reversal identification are performed. According to the preset meteorological element numerical thresholds, extreme high or low value points in the sequence are selected as benchmark feature points. At the same time, the reversal points where the trend of meteorological element change in the sequence changes significantly are identified and set as benchmark feature points. Centered on the position of each benchmark feature point in the time series, data subsequences of fixed time lengths are extracted forward and backward to form the analysis time window corresponding to the benchmark feature point; dynamic features including mean, variance, trend and fluctuation frequency are extracted from the data subsequence of each time window, and the dynamic features of all benchmark feature points are summarized to form the first association feature set; For each reference feature point, a geographic spatial range with a preset radius is defined centered on the reference feature point. Within the spatial range, environmental element information is extracted based on the processed original environmental data and remote sensing inversion data, and spatial interpolation is used to generate the spatial distribution characteristics of the environmental elements. The spatial distribution characteristics of all reference feature points are summarized to form the second associated feature set.

3. The meteorological-based real-time monitoring system for extreme climate risks according to claim 2, characterized in that, The feature influence domain is determined based on the relationship between two associated feature sets, and a representative feature point is selected within the feature influence domain, including: Calculate the coupling strength index between the first associated feature set and the second associated feature set to obtain a quantitative association measurement result; Based on the correlation measurement results, the overlapping influence areas of the first correlation feature set and the second correlation feature set in the spatial and temporal dimensions are identified, and a preliminary overlapping spatiotemporal range is obtained. Within the initial overlapping spatiotemporal range, the spatial distribution and temporal variation characteristics of the correlation measurement results are analyzed to extract the core regions and time periods where the correlation degree is consistently higher than the preset threshold. Based on the core area and time period, the final spatiotemporal boundaries are delineated, and the characteristic influence domain is determined. Within the characteristic influence domain, anomaly index reflecting the degree of deviation of meteorological elements from the normal, sensitivity index reflecting the degree of response of the natural environment to meteorological changes, and event occurrence frequency index based on past observation data are obtained and calculated respectively. The anomaly index, sensitivity index, and event frequency index are normalized and then weighted and fused according to preset weights to generate a comprehensive evaluation value. Based on the spatial distribution of the comprehensive evaluation value within the feature influence domain, the spatial location point where the comprehensive evaluation value reaches its peak is located and used as a representative feature point.

4. The meteorological-based real-time monitoring system for extreme climate risks according to claim 1, characterized in that, The key characteristic parameters include event intensity, duration, or spatial range of influence.

5. The meteorological-based real-time monitoring system for extreme climate risks according to claim 3, characterized in that, Based on extreme climate indicators, and combined with underlying surface characteristic data and disaster record data of the target area, a dynamic risk assessment model is used to analyze the risk level of extreme climate events. Multi-level warning thresholds are set according to preset climate state classification standards, including: Extreme climate indicators, geospatial surface feature data of the target area, and past disaster records are spatiotemporally aligned and fused to form a standardized comprehensive assessment dataset. The comprehensive assessment dataset is input into the dynamic risk assessment model to calculate the quantitative value of the probability of extreme climate events causing disasters in the target area, as well as the quantitative value of the potential impact represented by the vulnerability and exposure of the disaster-bearing body. Based on the quantitative values ​​of probability and potential impact, a pre-set risk matrix is ​​used for comprehensive evaluation to classify the corresponding risk levels. At the same time, based on long-term climate statistics and climate state classification standards, threshold boundaries for one or more key extreme climate indicators that trigger different levels of warnings are determined, and multi-level warning thresholds matching each risk level are set.

6. The meteorological-based real-time monitoring system for extreme climate risks according to claim 5, characterized in that, The cloud platform compares real-time meteorological data with multi-level early warning thresholds in real time. When the real-time monitoring data exceeds the multi-level early warning thresholds, an early warning message is automatically generated and pushed to designated terminals, including: Through the cloud platform's data stream processing service, real-time meteorological observation data or short-term forecast data streams of the target area are continuously accessed and processed to form a unified real-time monitoring dataset. The real-time monitoring dataset is matched and compared with multi-level early warning thresholds in real time; when the key indicator data in the real-time monitoring dataset exceeds any preset level of early warning threshold, an early warning trigger signal of the corresponding level is generated. Based on the warning trigger signal, the system automatically calls the preset warning information template that matches the warning level, event type, and affected area; and fills the template with relevant intensity, location, and evolution trend information from the real-time monitoring data to dynamically generate warning information. The warning information will be distributed in real time to relevant emergency command terminals, professional department handling terminals, and public information release platform terminals according to preset push rules.

7. The meteorological-based real-time monitoring system for extreme climate risks according to claim 6, characterized in that, The early warning information includes the specific event type, quantitative intensity, affected area, and recommended defense measures.

8. The meteorological-based real-time monitoring system for extreme climate risks according to claim 7, characterized in that, A monitoring framework is built based on cloud-native architecture and component-based design, and early warning information is used to achieve distributed collaborative operation and dynamic monitoring, including: Based on a cloud-native architecture using containerization and microservices, a monitoring framework is built. Within the monitoring framework, data collection, data processing, data calibration, risk assessment, and early warning generation functions are encapsulated as independent and elastically scalable microservice components and deployed. The early warning information is used as an internal event trigger signal to drive the risk assessment model to be dynamically updated and to simultaneously trigger the adaptive adjustment of the early warning threshold. By leveraging event-driven mechanisms and service calls between components, distributed collaborative operation across regions and departments can be achieved, enabling dynamic and continuous monitoring of extreme climate risks.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the system as described in any one of claims 1 to 8.

Images