A multi-source data processing method, medium and program product fusing summer monsoon diurnal variation characteristics and facing LTC hourly extreme precipitation evaluation

Abstract

The application discloses a kind of multi-source data processing method, medium and program product fused with summer monsoon daily variation characteristics and facing LTC hour extreme precipitation evaluation, belong to meteorological data processing and disaster risk assessment technical field.The method obtains daily precipitation, station hourly precipitation, LTC best path, hourly reanalysis environmental field, satellite cloud picture and three-dimensional cloud inversion data, completes space-time unification, quality control and sample construction;With hourly LTC center to establish typhoon dynamic reference space, extract summer monsoon daily variation disturbance quantity, phase index and intensity evolution characteristics;Low-level convergence, water vapor flux, LTC circulation maintenance and precipitation time series characteristics are dynamically registered, and a comprehensive feature matrix is constructed, and the output hour extreme precipitation sample category, daily variation influence evaluation value and key feature contribution are output by rule discrimination or classification model.The method can improve the accuracy of multi-source meteorological data fusion representation, sample discrimination and influence interpretation.

Description

Technical Field

[0001] This invention belongs to the field of meteorological data processing and disaster risk assessment technology, and involves hourly extreme precipitation assessment of landfalling tropical cyclones (LTCs), extraction of summer monsoon diurnal variation characteristics and fusion and discrimination of multi-source spatiotemporal meteorological data. Specifically, it is a multi-source data processing method, medium and program product that integrates summer monsoon diurnal variation characteristics and is oriented towards LTC hourly extreme precipitation assessment. Background Technology

[0002] After landfalling tropical cyclones (LTCs) enter the nearshore and land boundary layer, they are easily influenced by abrupt changes in underlying surface roughness, topographic forcing, low-level water vapor transport, monsoon background circulation, and mesoscale convective systems, which can induce hourly extreme precipitation. Compared to daily or cumulative precipitation, LTC hourly extreme precipitation is characterized by its suddenness, significant locality, and rapid spatiotemporal changes. Its assessment typically relies on multi-source heterogeneous data, including typhoon track data, station hourly precipitation data, gridded reanalysis data, satellite inversion data, and numerical model output data. These data vary significantly in temporal resolution, spatial resolution, observation methods, error structures, and data formats, leading to high data processing complexity for subsequent sample construction, feature analysis, and risk assessment.

[0003] Current LTC precipitation research and operational analysis often employ fixed latitude and longitude grids, administrative divisions, observation station neighborhoods, or preset influence radii for statistical analysis of precipitation samples and environmental field variables. While this approach is applicable to static regional precipitation events, under conditions of continuous LTC movement, intensity changes, and structural evolution, fixed spatial benchmarks struggle to accurately characterize the relative relationships between precipitation samples and the typhoon center, rainband structure, and surrounding environmental field. Particularly concerning are the differences between individual typhoons, different landfall stages, and different movement paths; directly matching data and performing statistical analysis based on geographic coordinates can easily introduce spatial location biases and insufficient sample comparability.

[0004] On the other hand, the East Asian summer monsoon exhibits significant multi-timescale variations. Besides intraseasonal oscillations and synoptic-scale monsoon surges, there are also significant diurnal variations in the lower tropospheric wind field, temperature and humidity field, and water vapor transport. These diurnal variations can influence the lower-level jet stream, water vapor flux, boundary layer stability, and local convection development, modulating the peripheral rainbands of the LTC and post-landfall heavy precipitation events. However, existing multi-source data processing workflows often focus more on typhoon tracks, intensity, cumulative precipitation, or single-time environmental field variables, failing to adequately express the hourly scale of the summer monsoon's diurnal variations. While some studies have considered diurnal variations or monsoon background factors, they typically rely on statistical diagnostics, case synthesis, or physical mechanism analysis, making it difficult to establish a standardized data processing framework suitable for assessing large-sample LTC hourly extreme precipitation. Furthermore, there are problems such as spatiotemporal inconsistencies among multi-source meteorological data, discretization of station data, discontinuous orbital sampling of satellite data, and limited resolution of reanalysis data. Traditional processing methods often rely on manual screening, step-by-step interpolation, empirical threshold judgment, or statistics from a single data source, which can easily lead to defects such as insufficient repeatability of the processing flow, inconsistent sample field structure, redundant feature dimensions, and unstable labels for extreme precipitation samples.

[0005] In summary, existing technologies still suffer from problems such as strong data heterogeneity, unstable spatiotemporal correspondence under typhoon background, insufficient expression of summer monsoon diurnal variation characteristics, and low interpretability of precipitation sample characterization and discrimination when assessing LTC hourly extreme precipitation. Therefore, it is an urgent technical problem to be solved to standardize the characterization and discrimination of LTC hourly extreme precipitation data that incorporates summer monsoon diurnal variation characteristics. Summary of the Invention

[0006] (a) Purpose of the invention To address the aforementioned deficiencies and shortcomings of existing technologies, this invention aims to provide a multi-source data processing method, medium, and program product that integrates summer monsoon diurnal variation characteristics and is geared towards LTC hourly extreme precipitation assessment. By acquiring and standardizing multi-source spatiotemporal meteorological data such as typhoon paths, station hourly precipitation, and reanalysis environmental fields, a typhoon dynamic reference space is constructed. Summer monsoon diurnal variation characteristics, environmental field characteristics, and precipitation sample labels are extracted, and feature fusion, sample characterization, and classification are performed. This enables stable characterization, accurate identification, and assessment of LTC hourly extreme precipitation samples under typhoon movement background, as well as interpretable data output.

[0007] (II) Technical Solution To achieve the objective of this invention and solve its technical problems, the present invention adopts the following technical solution: In a first aspect, the present invention provides a multi-source data processing method that integrates summer monsoon diurnal variation characteristics and is geared towards assessing LTC hourly extreme precipitation. This method, based on multi-source spatiotemporal meteorological data within a typhoon dynamic reference space, characterizes, quantifies interactions, and discriminates samples regarding the impact of summer monsoon diurnal variation on hourly-scale extreme precipitation of landfalling tropical cyclones (LTCs). The method includes at least the following steps: SS1. Multi-source data acquisition and preprocessing: Acquire daily precipitation data, hourly precipitation data from stations, LTC optimal path data, and hourly reanalysis environmental field data, and perform preprocessing (such as time base unification, spatial resolution matching, and removal of missing data) to form a basic spatiotemporal dataset; SS2. LTC Precipitation Sample Construction: Based on the objective weather map analysis method, the precipitation process affected by LTC is identified. The daily scale identification results are matched with hourly precipitation data according to station, date and time to generate LTC hourly precipitation candidate sample set, sample source identifier and corresponding spatiotemporal index; SS3. Establishment of dynamic reference space: Establish a typhoon relative coordinate system with the hourly LTC center position as the dynamic origin, and transform the station position, candidate precipitation samples and environmental field grid points to the typhoon relative coordinate system hourly. Generate the dynamic characteristic area of ​​the environmental field southwest of LTC based on the relative orientation and distance. SS4. Extraction of summer monsoon diurnal variation features: Extract the wind field, temperature field, humidity field and water vapor transport field in the lower troposphere within the dynamic feature area. Calculate the diurnal variation disturbance based on the difference between the hourly values ​​and the corresponding daily average values, and generate the summer monsoon diurnal variation feature vector and its time-series index for fusion and discrimination. SS5. Interaction Feature Fusion: The summer monsoon diurnal variation feature vector is dynamically registered with the low-level convergence feature, water vapor flux feature, LTC circulation maintenance feature and precipitation time series feature to construct a comprehensive feature matrix to characterize the interaction between summer monsoon diurnal variation and LTC hourly extreme precipitation; SS6. Evaluation Data Generation and Judgment: Based on the comprehensive feature matrix, perform rule-based discrimination or classification model discrimination, and output hourly extreme precipitation sample categories, daily variation impact assessment values, key feature contribution amounts, thermal impact assessment data, dynamic impact assessment data, and spatiotemporal distribution data.

[0008] The second objective of this invention is to provide a computer program product, including computer instructions, for executing the steps of the multi-source data processing method of the present invention, which integrates summer monsoon diurnal variation characteristics and is geared towards LTC hourly extreme precipitation assessment.

[0009] The third objective of this invention is to provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the multi-source data processing method described above, which integrates summer monsoon diurnal variation characteristics and is geared towards LTC hourly extreme precipitation assessment.

[0010] (III) Technical Effects Compared with existing technologies, the multi-source data processing method, medium, and program products of the present invention, which integrate the diurnal variation characteristics of the summer monsoon and are geared towards the assessment of LTC hourly extreme precipitation, have the following significant technical effects: (1) This invention establishes a basic spatiotemporal dataset by unifying the spatiotemporal organization of daily precipitation, hourly precipitation at stations, optimal LTC path, hourly reanalysis environmental field, satellite cloud image and three-dimensional cloud inversion data. This can reduce the inconsistency of multi-source data in terms of time reference, spatial resolution, variable hierarchy and quality status, and improve the standardization and reproducibility of LTC hourly precipitation sample construction.

[0011] (2) By constructing a typhoon dynamic reference space, this invention transforms the stations, precipitation samples and environmental field grid points into a relative coordinate system based on the hourly LTC center, and extracts the summer monsoon diurnal variation disturbance, phase index and intensity evolution characteristics in the dynamic feature area southwest of the LTC, which can improve the spatial correspondence between environmental field characteristics and hourly precipitation samples under the background of typhoon movement.

[0012] (3) By constructing a comprehensive feature matrix, this invention dynamically registers the diurnal variation characteristics of summer monsoon, low-level convergence, water vapor flux, maintenance of LTC circulation and precipitation time series characteristics, and combines rule discrimination, classification model discrimination and key feature contribution calculation, it can realize the structured characterization, category discrimination and evaluation of LTC hourly extreme precipitation samples and the interpretation and output of influencing factors. Attached Figure Description

[0013] The accompanying drawings, which form part of this specification, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. Hereinafter, embodiments of the invention will be described in detail with reference to the accompanying drawings, wherein: Figure 1 This is a flowchart illustrating the implementation of the multi-source data processing method of the present invention, which integrates the diurnal variation characteristics of the summer monsoon and is geared towards the assessment of LTC hourly extreme precipitation. Figure 2 This is a diagram illustrating the structure of multi-source spatiotemporal meteorological data acquisition and preprocessing. Figure 3 This is a flowchart of the LTC hourly precipitation candidate sample construction and extreme sample labeling process; Figure 4This is a schematic diagram of the typhoon dynamic reference space and the dynamic characteristic area of ​​the environmental field southwest of LTC; Figure 5 This is a flowchart for generating diurnal variation disturbances, phase indices, and intensity characteristics of the summer monsoon. Figure 6 This is a schematic diagram of the construction of the interaction-integrated feature matrix; Figure 7 This is a flowchart of the evaluation data generation and sample discrimination process, where: (a) is the rule discrimination process; (b) is the classification model discrimination process; and (c) is the calculation of the contribution of key features. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be described in more detail below with reference to the accompanying drawings. The described embodiments are some, but not all, embodiments of this invention, and are exemplary and intended to explain the invention, not to limit it. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0015] It should be noted that, in this invention, LTC refers to a landfalling tropical cyclone; LTC hourly extreme precipitation refers to hourly-scale heavy precipitation or extreme precipitation events occurring during the influence of LTC; and summer monsoon diurnal variation characteristics refer to the perturbations, phases, persistence, and intensity evolution of wind fields, thermal fields, humidity fields, and water vapor transport fields relative to the daily average background amount within the diurnal variation cycle under the summer monsoon background. The core improvement of this invention lies in: instead of simply statistically filtering precipitation data itself, it unifies and integrates summer monsoon diurnal variation perturbations, low-level convergence, water vapor transport, LTC circulation maintenance, and precipitation time-series responses within the hourly moving LTC dynamic reference space. This forms a comprehensive feature matrix capable of characterizing the thermodynamic-dynamic-water vapor coupling mechanism of summer monsoon diurnal variation, LTC circulation, and hourly extreme precipitation, and further generates discriminative, quantifiable, and interpretable assessment results.

[0016] Example 1: Multi-source spatiotemporal meteorological data processing for LTC hourly extreme precipitation assessment like Figure 1As illustrated, as a specific example, the multi-source data processing method provided in this embodiment of the invention, which integrates summer monsoon diurnal variation characteristics and is geared towards assessing hourly extreme precipitation in landfalling tropical cyclones (LTCs), is used to characterize, quantify the interaction of, and discriminate samples of the impact of summer monsoon diurnal variation on hourly extreme precipitation of landfalling tropical cyclones (LTCs) based on multi-source spatiotemporal meteorological data within a typhoon dynamic reference space. The technical concept of this embodiment is as follows: first, heterogeneous meteorological data are organized into a basic spatiotemporal dataset with a unified field structure; second, station precipitation samples and environmental field grid points are transformed into a typhoon dynamic reference space that moves with the LTC center; third, summer monsoon diurnal variation disturbances, phase indices, and intensity evolution characteristics are extracted; and finally, the characterization, discrimination, and assessment of LTC hourly extreme precipitation samples, as well as the output of contribution, are completed through a comprehensive feature matrix. Specifically, the above method of the present invention mainly includes the following in its implementation: SS1. Multi-source data acquisition and preprocessing: Acquire daily precipitation data, hourly precipitation data from various stations, optimal LTC path data, and hourly reanalysis environmental field data, and perform preprocessing to form a basic spatiotemporal dataset. The daily precipitation data includes daily precipitation, observation date, station number, station latitude and longitude, station altitude, and daily quality indicator. The hourly precipitation data includes hourly precipitation, observation time, station number, station latitude and longitude, station altitude, and hourly quality indicator. The optimal LTC path data includes the 6-hourly latitude and longitude position of the LTC center, maximum near-center wind speed, minimum central pressure, and intensity level, and the 6-hour interval records are completed into an hourly continuous time series using cubic spline interpolation. The hourly reanalysis environmental field data includes at least hourly wind, temperature, and humidity fields at 850 hPa, 500 hPa, and 200 hPa, as well as surface pressure and water vapor flux fields, and the horizontal resolution of the hourly reanalysis environmental field data is no less than 0.25° × 0.25°. Figure 2 As shown.

[0017] Preferably, to enhance the characterization of lower-level summer monsoon transport, boundary layer water vapor convergence, and the triggering conditions of typhoon peripheral rainbands, the hourly reanalysis environmental field data may also include one or more of the following: zonal wind, meridional wind, vertical velocity, specific humidity, relative humidity, temperature, geopotential height, divergence, vorticity, water vapor flux divergence, and precipitable water at 925 hPa, 1000 hPa, or near-surface levels. These variables are used in subsequent steps to construct lower-level convergence characteristics, water vapor flux characteristics, water vapor flux divergence characteristics, thermal disturbance characteristics, and dynamic disturbance characteristics, respectively.

[0018] The purpose of configuring the aforementioned data sources is to enable LTC precipitation samples, typhoon track status, lower-level water vapor transport, mid-level environmental configuration, and upper-level circulation background to be processed together on a unified time axis. Specifically, hourly completion of the optimal LTC track data is used to address the temporal resolution inconsistency between typhoon track data and hourly precipitation data from various stations, as well as hourly reanalysis environmental field data. The multi-height layer configuration of the hourly reanalysis environmental field data supports the structured extraction of subsequent lower-level dynamic characteristics, mid-level thermal and humidity conditions, and upper-level steering background.

[0019] Furthermore, the basic spatiotemporal dataset also includes satellite cloud imagery data and 3D cloud inversion data. The satellite cloud imagery data includes cloud top brightness temperature, cold cloud coverage, cloud cluster spatial distribution, cloud cluster expansion direction, and cloud cluster duration. The 3D cloud inversion data includes precipitation cloud top height, cloud base height, echo vertical profile, and precipitation cloud vertical structure identifier. The satellite cloud imagery data and 3D cloud inversion data are correlated with hourly precipitation data and hourly reanalysis environmental field data of the stations according to LTC number, observation time, and typhoon relative position to supplement the characterization of cloud and rain structure features corresponding to LTC hourly precipitation samples.

[0020] By incorporating satellite cloud imagery and 3D cloud inversion data, this embodiment can further obtain information on the horizontal organization and vertical structure of LTC precipitation clouds, in addition to surface precipitation and the reanalysis environmental field. Cloud top brightness temperature, cold cloud coverage, and cloud cluster duration can be used to characterize the development intensity and maintenance status of deep convective cloud clusters; precipitation cloud top height, cloud base height, and echo vertical profiles can be used to characterize the vertical extension of precipitation clouds and precipitation structure. These cloud-rain structure characteristics are written as supplementary fields into the basic spatiotemporal dataset, which can improve the completeness of subsequent characterization and evaluation of hourly extreme precipitation samples.

[0021] Step SS1, in preprocessing the acquired multi-source spatiotemporal meteorological data, includes: converting daily precipitation data, hourly precipitation data from stations, LTC optimal path data, and hourly reanalysis environmental field data to the same time base and generating a unified timestamp; establishing spatial correspondences based on station latitude and longitude, reanalysis grid latitude and longitude, and LTC hourly center location, and resampling data of different spatial resolutions to a unified grid; identifying abnormal records and generating processing markers for removal, correction, interpolation, or retention based on quality indicators, missing data ratios, physical thresholds, and continuity between adjacent time periods; and writing the processed multi-source spatiotemporal meteorological data into a unified field structure according to data source, variable name, vertical level, observation time, spatial location, and quality status to generate a basic spatiotemporal dataset.

[0022] SS2. LTC precipitation sample construction: Based on objective weather map analysis, LTC-influenced precipitation processes are identified. The diurnal identification results are matched with hourly precipitation data by station, date, and time to generate a candidate sample set of LTC hourly precipitation, sample source identifiers, and corresponding spatiotemporal indexes. By separating typhoon-influenced precipitation from ordinary hourly precipitation records, the diurnal LTC-influenced precipitation process identification results are further refined to hourly samples. Clear LTC attribution, spatial relative relationships, and sample quality identifiers are established for each hourly precipitation record. This avoids the misclassification of non-LTC precipitation, localized thermal convection precipitation, or precipitation from other weather systems into LTC hourly extreme precipitation samples, thereby improving the relevance of subsequent diurnal variation modulation analysis. The identification of LTC-influenced precipitation processes based on objective weather map analysis, such as... Figure 3 As shown, it includes: S21. Path Time Series Completion: Read the center position, maximum wind speed near the center, and minimum central pressure from the optimal LTC path data. Interpolate the 6-hour path records into an hourly path sequence, and assign an LTC number and intensity status identifier to each hourly time. Specifically, the time series completion of the LTC center latitude and longitude is preferably performed using cubic spline interpolation, piecewise cubic Hermite interpolation, or linear interpolation. The completion of the maximum wind speed near the center and the minimum central pressure is performed using monotonic interpolation that maintains the extreme value trend or piecewise linear interpolation to avoid generating false intensity extreme values ​​that do not conform to the laws of physical change during the interpolation process. S22. Determination of the period of impact: Based on the optimal path position of the LTC every 6 hours, the area directly affected by LTC precipitation is delineated with the LTC center as the center and a radius of 500 km as the identification radius. According to the hourly center position, landfall time, movement path, and circulation maintenance status of the LTC, the start and end periods of the LTC's impact on the precipitation at the monitoring stations are determined, and the start and end periods are overlapped with the hourly precipitation data of the monitoring stations for time screening. Among them, the circulation maintenance status can be determined by the minimum pressure at the LTC center, the maximum wind speed near the center, the low-level closed circulation, the intensity of the low-level tangential wind, the radial inflow, or the duration and movement speed after landfall. For LTC cases that weaken after landfall but whose low-level circulation can still maintain water vapor convergence and the development of the outer rainband, the candidate precipitation records within the corresponding period are still retained. S23. Station Spatial Matching: The Haversine great circle distance formula is used to calculate the spherical distance (relative radial distance r, km) and azimuth (relative azimuth θ, °) of each station relative to the center of the LTC at the corresponding time. Based on the radial distance, azimuth, and the LTC circulation influence range, hourly precipitation records of stations within the LTC precipitation influence range are selected. The LTC circulation influence range can be determined by a fixed radius threshold, an adaptive radius threshold that varies with intensity level, a circulation radius determined based on the closed structure of the low-level wind field, or a precipitation influence radius determined based on the coverage of the outer rainband cloud cluster. This can take into account the differences in the LTC influence range at different intensities, different moving speeds, and different landfall stages. S24. Weather Situation Verification: Based on reanalysis environmental field data and combined with satellite cloud image data, identify the LTC circulation, peripheral rainband, low-level wind field (low-level convergence), and water vapor transport configuration, and verify the matching relationship between the selected station precipitation records and the weather patterns affected by the LTC. Specifically, the 850hPa or 925hPa wind field can be used to identify the low-level circulation and southwest airflow input, water vapor flux and its divergence can be used to identify water vapor transport and convergence areas, and cloud top brightness temperature and cold cloud coverage can be used to identify deep convective cloud clusters or peripheral rainbands. The above results are then verified for consistency with the hourly precipitation times and the relative position of the typhoon at the stations. S25. Determination of precipitation attribution: Based on the location of the LTC center, the spatial matching results of the station, the period of influence, the weather pattern verification results, and the continuity of precipitation records, the LTC influence attribution is determined for the hourly precipitation records, and candidate samples of LTC hourly precipitation are generated. Preferably, when an hourly precipitation record simultaneously meets the following conditions: it is within the LTC influence period, within the influence range of the LTC circulation or the outer rainband, there is low-level water vapor transport or convergence configuration at its time, and the precipitation process is continuous in adjacent time periods, the record is marked as an LTC-influenced precipitation sample; when only some conditions are met, it is set as a sample to be verified; when it is significantly inconsistent with the LTC path, cloud and rain structure, or low-level circulation configuration, it is marked as a non-LTC belonging sample and removed from the candidate sample set.

[0023] S26. Sample Field Configuration: Configure the LTC number, station number, station latitude and longitude, station altitude, date, time, precipitation, relative distance, relative orientation, attribution determination identifier, and sample quality identifier for each LTC hourly precipitation candidate sample.

[0024] Further, step SS2 also includes extreme sample labeling of the LTC hourly precipitation candidate sample set. For each station, valid LTC hourly precipitation samples that meet the quality labeling requirements and have a precipitation amount of not less than 0.1 mm within the study period are extracted. These samples are sorted in ascending order of precipitation amount, and a 99th percentile threshold is determined based on empirical distribution relationships to generate a single-station extreme precipitation judgment threshold. Samples that reach the corresponding threshold are marked as hourly extreme precipitation samples, and samples that do not reach the corresponding threshold are marked as non-extreme precipitation samples. Each sample is written with the station number, LTC number, time, precipitation amount, empirical cumulative probability, judgment threshold, and sample category identifier. Specifically, for a given station, the valid LTC hourly precipitation samples are sorted in ascending order of precipitation amount as follows: P 1, P 2, P 3,… P m ,… P N ,in N This represents the total number of valid samples at the corresponding station. m For sorting sequence numbers; according to empirical distribution relationships F m =( m -0.31) / ( N +0.38) calculate the first m The empirical cumulative probability corresponding to each sorted sample, and select those that satisfy... F m The precipitation corresponding to the smallest sort number ≥0.99 is used as the extreme precipitation threshold for that station. The purpose of using a single-station threshold is to adapt to the differences in precipitation climate background between different stations and to avoid sample labeling bias between coastal, island, inland and complex terrain stations caused by a uniform fixed threshold.

[0025] SS3. Establishment of Dynamic Reference Space: like Figure 4 As shown, a typhoon relative coordinate system is established with the hourly LTC center position as the dynamic origin. The station positions, candidate precipitation samples, and environmental field grid points are hourly transformed to the typhoon relative coordinate system. Based on relative azimuth and distance, a dynamic characteristic region of the environmental field southwest of the LTC is generated. This step is one of the key technical means that distinguishes this invention from fixed geographic grid statistical methods. Since the LTC moves continuously before and after landfall, directly statistically analyzing environmental field variables within a fixed latitude and longitude area easily mixes environmental signals from different typhoon life stages, different circulation quadrants, and different relative positions. This embodiment establishes a dynamic reference space that moves hourly with the LTC, enabling a unified expression of low-level water vapor input, low-level convergence, thermal disturbances, and precipitation station positions relative to the LTC center. This more accurately characterizes the modulation effect of summer monsoon diurnal variations on the southwest side or outer rainband of the LTC.

[0026] The typhoon relative coordinate system uses the hourly LTC center position as the origin and the due north direction or the LTC movement direction as the azimuth reference direction. It converts the station position, candidate precipitation samples, and environmental field grid points into relative radial distance, relative azimuth, meridional displacement, and zonal displacement relative to the origin. The meridional and zonal displacements are calculated using spherical distance or map projection distance. The relative radial distance is synthesized from the meridional and zonal displacements. The relative azimuth is determined by the meridional displacement, zonal displacement, and azimuth reference direction. The relative azimuth is used to divide the samples into the northeast, southeast, southwest, and northwest quadrants, forming a dynamic reference space that is updated synchronously with the LTC movement process.

[0027] Preferably, for monitoring stations or environmental field grid points P Its relative to the time-to-time LTC center C t The meridional displacement, latitudinal displacement, radial distance, and azimuth angle can be denoted as Δ. x t Δ y t , r t and θ t ,in r t From Δ x t With Δ y t Synthesized θ t This is used to determine the relative limits of the point located at the center of the LTC. If the LTC movement direction is used as the reference direction, then... θ t Convert to a rotational azimuth angle relative to the direction of movement to distinguish the left, right, front, and rear areas of the direction of travel.

[0028] Generating the dynamic characteristic region of the environmental field on the southwest side of the LTC is another core technical point of this step. Its physical basis lies in the fact that the low-level jet stream of the East Asian summer monsoon carries a large amount of warm and moist water vapor, which is mainly transported from the southwest side of the LTC to the LTC core region. This is a key external dynamic and thermal forcing source that triggers hourly extreme precipitation during the LTC landfall phase. Specifically, when generating the dynamic characteristic region of the environmental field on the southwest side of the LTC, based on the relative azimuth and relative radial distance in the typhoon's relative coordinate system, reanalysis environmental field grid points located in the southwest quadrant of the LTC center and within a preset radial range are selected as candidate grid points. These candidate grid points are then screened based on the LTC circulation scale, movement direction, and spatial resolution of the reanalysis data. The screened candidate grid points are weighted according to their distance from the LTC center, the area represented by the grid point, or the validity of environmental field variables. Hourly outputs are then used to characterize the summer monsoon input, low-level water vapor transport, low-level convergence, environmental thermal conditions, and peripheral circulation configuration on the southwest side of the LTC, and these regional statistics are written into the dynamic characteristic region index. Preferably, in the typhoon relative coordinate system, a fan-shaped characteristic area on the southwest side of the LTC is defined with an azimuth range of 210°±60° (i.e., 150° to 270°); the radial range of the characteristic area is determined by two boundaries: the inner boundary is fixed at 200km (effectively excluding the area of ​​strong convection interference from the LTC eyewall and inner rainband), and the radius of the outer boundary is dynamically calculated based on the current intensity of the LTC, with the outer radius (km) = 500 + k × V max ,in V max The maximum wind speed (m / s) near the center of the LTC, and the proportionality coefficient. k Rated based on historical LTC data (recommended) k =2.0~3.0), the outer radius is limited to the range of 500~1000 km, so as to realize the adaptive dynamic expansion and contraction of the spatial range of the feature region with the intensity of LTC.

[0029] The southwest side of the LTC was chosen as the dynamic feature area because, during the impact of landfalling typhoons in my country, this region typically corresponds to a significant source of southwest monsoon input, moisture transport, and low-level wind disturbances in the environmental field. This embodiment does not select a fixed geographical area, but rather generates a dynamic feature area that updates as the typhoon moves, based on the hourly LTC center location, relative azimuth, and relative distance. This improves the spatial accuracy of the correspondence between the diurnal variation characteristics of the summer monsoon and hourly precipitation samples from the LTC.

[0030] SS4. Extraction of diurnal variation characteristics of the summer monsoon: like Figure 5As shown, the wind field, temperature field, humidity field and water vapor transport field of the lower troposphere are extracted in the dynamic feature area. The daily variation disturbance is calculated by the difference between the hourly value and the corresponding daily average value, and the summer monsoon daily variation feature vector and its time series index are generated for fusion and discrimination.

[0031] This step expands the diurnal variation of the summer monsoon from a univariate hourly curve into a multidimensional feature vector that includes wind field disturbances, thermal disturbances, moisture transport disturbances, disturbance phases, and disturbance persistence. This vector is then extracted within a dynamic feature region that updates as the LTC moves. Therefore, the obtained diurnal variation disturbance is no longer a simple diurnal cycle of the background field in a fixed region, but rather reflects the direct effect of the diurnal variation of the summer monsoon input and the environmental field relative to the LTC on the hourly precipitation response.

[0032] Specifically, the extraction of diurnal variation characteristics of the summer monsoon includes at least the following sub-steps: S41. Low-level variable extraction: Extract zonal wind, meridional wind, temperature, relative humidity or specific humidity, water vapor flux and surface air pressure variables at heights of 850hPa, 925hPa and / or 1000hPa within the dynamic feature area, and assign height layer, grid point, time, data source and effective layer identifier to each variable. S42. Diurnal variation cycle division: Divide a continuous 24 hours into a diurnal variation cycle according to the local solar time, and assign each hourly time to the corresponding diurnal variation cycle, so that environmental field variables, dynamic characteristic grid points and LTC hourly precipitation candidate samples correspond to each other in the same diurnal variation time frame. S43. Calculation of daily average background amount: For each daily change cycle, each variable, each height layer and each dynamic feature area, calculate the regional weighted daily average based on the effective grid point set. The regional weighted daily average is determined by the area represented by the grid point and the distance weight from the grid point to the LTC center. S44. Generation of daily variation disturbances: The difference between each hourly variable value and the corresponding regional weighted daily average value is used to generate disturbances in wind field, temperature, humidity, water vapor transport and air pressure. The disturbances are written into the disturbance data table according to LTC number, sample time, relative azimuth, radial distance and height layer identifier. S45. Phase index extraction: Identify the peak times of positive disturbances, peak times of negative disturbances, and times of disturbance sign transitions in the hourly disturbance sequence within the same diurnal variation cycle, and generate diurnal variation phase indices to characterize the strengthening, weakening, and turning processes of summer monsoon diurnal variation; S46. Intensity normalization: Based on the standard deviation, interquartile range, or range after outlier removal of the same variable in the study area or the same LTC process, normalize each disturbance quantity to generate standardized disturbance intensity that can be compared across cases, time periods, and variables. S47. Calculation of intensity evolution characteristics: Based on the standardized disturbance intensity, and combined with the wind field disturbance, humidity disturbance and water vapor transport disturbance, calculate the regional average disturbance intensity, disturbance duration and disturbance intensification rate to generate dynamic summer monsoon diurnal variation intensity characteristics. S48. Time-series feature encoding: Based on the disturbance amount, diurnal variation phase index and dynamic summer monsoon diurnal variation intensity characteristics, the summer monsoon diurnal variation disturbance sequence is encoded according to LTC number, diurnal variation period, dynamic feature area number and sample time.

[0033] Furthermore, step SS4 introduces the Ensemble Empirical Mode Decomposition (EEMD) method to perform multi-timescale adaptive decomposition enhancement processing on the time series of diurnal variation disturbances of various atmospheric elements. The specific method is as follows: EEMD decomposition is performed on the time series of diurnal variation disturbances of various elements in the southwest characteristic area using a white noise ensemble (ensemble number ≥ 200, noise standard deviation taken as 0.2 times the signal standard deviation), effectively suppressing mode aliasing effects and obtaining several Intrinsic Mode Function (IMF) components and residual trend terms; Hilbert transform is performed on each IMF component to calculate the hourly instantaneous amplitude. A ( t and instantaneous frequency f ( t Identify and retain components whose instantaneous frequencies fall within the range of 24h±4h (corresponding to the diurnal cycle mode) and 12h±2h (corresponding to the semi-diurnal harmonic mode); extract the time of occurrence of the daily peak value of the diurnal cycle component ( T peak Reflecting the diurnal variation forced phase preference), diurnal amplitude ( A peak The three extended characteristic parameters (reflecting the intensity of diurnal variation) and the energy proportion of the half-diurnal harmonic component (reflecting the degree of nonlinear harmonic modulation) are incorporated into the enhanced summer monsoon diurnal variation characteristic vector, enabling the comprehensive characteristic matrix to capture the multi-frequency, non-stationary modulation characteristics of the impact of summer monsoon diurnal variation on LTC hourly extreme precipitation.

[0034] Furthermore, the diurnal variation feature vector of the summer monsoon includes wind field disturbance, thermal disturbance, water vapor transport disturbance, disturbance phase, and disturbance persistence components. The wind field disturbance components include zonal wind disturbance, meridional wind disturbance, wind speed disturbance amplitude, and wind direction disturbance angle; the thermal disturbance components include temperature disturbance, humidity disturbance, and surface pressure disturbance; the water vapor transport disturbance components include water vapor flux disturbance, lower-level water vapor input intensity, and water vapor flux divergence disturbance; the disturbance phase components include disturbance peak times (positive disturbance peak times and negative disturbance peak times), disturbance trough times, and peak-trough phase difference; the disturbance persistence components include disturbance continuous intensification duration, disturbance continuous weakening duration, and disturbance intensification rate; and each component is encoded into a summer monsoon diurnal variation feature vector with a unified field structure.

[0035] SS5. Interaction Feature Fusion: like Figure 6 As shown, the diurnal variation feature vector of the summer monsoon is dynamically registered with the low-level convergence feature, water vapor flux feature, LTC circulation maintenance feature and precipitation time series feature to construct a comprehensive feature matrix to characterize the interaction between the diurnal variation of the summer monsoon and LTC hourly extreme precipitation.

[0036] In this embodiment, the low-level convergence characteristics are calculated by negative divergence or radial inflow of the horizontal wind field in the lower troposphere within the dynamic characteristic region; the water vapor flux characteristics are calculated by coupling the horizontal wind field in the lower troposphere with specific humidity data within the dynamic characteristic region, forming water vapor flux intensity, water vapor flux divergence, and water vapor transport direction characteristics; the LTC circulation maintenance characteristics include the LTC center minimum pressure, near-center maximum wind speed, moving speed, moving direction, duration after landfall, relative topographic position, and statistics of tangential and radial winds in the typhoon's relative coordinate system; the precipitation time series characteristics include hourly precipitation, previous cumulative precipitation, precipitation enhancement rate, precipitation duration, and the number of times peak precipitation occurs; and variable identifiers, time identifiers, spatial location identifiers, and sample category identifiers are configured for each of the above characteristics to form the basic characteristics fused with the summer monsoon diurnal variation characteristic vector.

[0037] Furthermore, low-level convergence characteristics may include negative horizontal divergence values ​​at 850 hPa, 925 hPa, or 1000 hPa, radial inflow, low-level wind speed enhancement, low-level tangential wind enhancement, and upward motion intensity; water vapor flux characteristics may include zonal water vapor flux, meridional water vapor flux, overall or low-level water vapor flux modulus, water vapor flux divergence, water vapor transport direction angle, and water vapor input intensity from the dynamic characteristic region southwest of the LTC; LTC circulation maintenance characteristics may include the rate of change of the central minimum pressure, the rate of change of the near-center maximum wind speed, low-level circulation closure, post-landfall circulation maintenance duration, and circulation intensity decay rate; precipitation time series characteristics may include cumulative precipitation in the first 1 hour, the first 3 hours, and the first 6 hours, pre-peak precipitation enhancement rate, post-peak decay rate, and precipitation process duration. The basic variables for thermal impact assessment may include lower-level temperature disturbances, humidity disturbances, changes in saturation difference, equivalent potential temperature disturbances, and changes in water vapor conditions; the basic variables for dynamic impact assessment may include lower-level wind field disturbances, horizontal convergence, wind speed enhancement, radial inflow, vertical motion, LTC circulation maintenance, and water vapor transport. These variables are labeled with thermal variable identifiers, dynamic variable identifiers, water vapor variable identifiers, and LTC state variable identifiers respectively in the integrated feature matrix to output thermal impact assessment data, dynamic impact assessment data, and water vapor transport impact assessment data in SS6.

[0038] Furthermore, dynamic registration includes temporal registration, spatial registration, and lag correlation registration. Temporal registration extracts summer monsoon diurnal variation feature vectors, low-level convergence features, and water vapor flux features from the synchronous times, preset leading time sets, and preset lag time sets of LTC hourly precipitation candidate samples. Spatial registration matches environmental field statistics within the corresponding dynamic feature areas based on the relative azimuth, radial distance, meridional displacement, and zonal displacement of the station in the typhoon relative coordinate system. Lag correlation registration uses the peak times of hourly extreme precipitation samples as a benchmark to calculate the time offset, duration, and rate of change of each input feature relative to the precipitation peak, and retains the leading, synchronous, or lag relationship identifiers of summer monsoon diurnal variation features relative to LTC hourly precipitation samples.

[0039] When constructing the comprehensive feature matrix, the following sub-matrices are first established: summer monsoon diurnal variation, low-level convergence, water vapor transport, LTC circulation maintenance, precipitation time series, and lag correlation. Then, matrix row and column indices are established using LTC number, station number, sample time, relative azimuth, radial distance, height layer identifier, and sample category identifier. The sub-matrices are then spliced ​​together and subjected to missing measurement masking according to a unified field structure to form a comprehensive feature matrix that includes environmental field diurnal variation characteristics, typhoon circulation background characteristics, water vapor transport characteristics, precipitation response characteristics, and interaction time series characteristics. This matrix is ​​used to evaluate the thermodynamic-dynamic-water vapor coupling interaction between summer monsoon diurnal variation and LTC hourly extreme precipitation.

[0040] Through step SS5, this embodiment can transform the temporal relationship of whether the diurnal variation of the summer monsoon leads, is synchronous with, or lags behind the LTC hourly extreme precipitation into the time offset, duration, and rate of change fields in the lag correlation submatrix; at the same time, it unifies the low-level convergence, water vapor flux, circulation maintenance, and precipitation time-series characteristics into the comprehensive feature matrix. This matrix serves as both the data input for the SS6 rule discrimination and the feature basis for classification model discrimination and the calculation of key feature contributions.

[0041] SS6. Evaluation Data Generation and Judgment: like Figure 7 As shown, based on the comprehensive feature matrix, rule-based discrimination or classification model discrimination is performed, and the hourly extreme precipitation sample category, daily variation impact assessment value, key feature contribution, thermal impact assessment data, dynamic impact assessment data, and spatiotemporal distribution data are output.

[0042] Preferably, the rule-based discrimination includes a hierarchical discrimination process based on extreme precipitation constraints, diurnal variation disturbance constraints, LTC circulation maintenance constraints, and thermodynamic-dynamic-water vapor co-constraints. Specifically: first, it determines whether candidate samples meet hourly extreme precipitation conditions based on hourly precipitation absolute thresholds, regional quantile thresholds, or station historical extreme value thresholds; then, it determines whether the diurnal variation disturbances of low-level convergence, water vapor flux, water vapor flux divergence, low-level temperature and humidity disturbances, and vertical motion disturbances within the dynamic feature region of the comprehensive feature matrix reach their corresponding enhancement thresholds; subsequently, it determines whether the LTC center minimum... The study assesses whether the indices of air pressure, maximum near-center wind speed, low-level tangential wind, radial inflow, moving speed, post-landfall duration, and low-level circulation intensity meet the conditions for sustained circulation influence. Finally, based on whether enhanced low-level convergence, enhanced water vapor transport, enhanced thermal and humid conditions, and enhanced upward motion occur simultaneously or within a preset leading time period, it generates hourly extreme precipitation sample categories: those significantly affected by enhanced summer monsoon diurnal variation, those weakly affected by summer monsoon diurnal variation, and those with insignificant diurnal variation influence. Corresponding diurnal variation impact assessment, dynamic impact assessment, thermal impact assessment, and water vapor transport impact assessment data are output. Specifically, extreme precipitation constraints are used to exclude candidate samples that do not reach hourly extreme precipitation intensity; diurnal variation disturbance constraints are used to determine whether the summer monsoon diurnal variation shows significant enhancement before and after precipitation; LTC circulation maintenance constraints are used to confirm that the hourly precipitation sample is still under the effective influence of LTC circulation or the outer rainband; and thermal-dynamic-water vapor synergistic constraints are used to determine whether enhanced thermal and humid conditions, enhanced low-level convergence, enhanced water vapor transport, and enhanced upward motion collectively constitute a coupling environment conducive to the occurrence of hourly extreme precipitation.

[0043] The purpose of this rule-based discrimination process is to incorporate the diurnal variation characteristics of the summer monsoon, the lower-level dynamic environment, water vapor transport conditions, and the maintenance state of the LTC circulation into a stratified judgment, thereby avoiding sample discrimination based solely on precipitation or a single environmental variable. By setting four types of constraints—extreme precipitation, diurnal variation disturbance, circulation maintenance, and synergistic enhancement—this embodiment can structurally determine whether hourly extreme precipitation samples are significantly modulated by the diurnal variation of the summer monsoon.

[0044] Preferably, when a classification model is chosen for discrimination, it includes supervised learning discrimination based on the comprehensive feature matrix using labeled samples, wherein: The comprehensive feature matrix is ​​divided into training samples, validation samples, and samples to be discriminated based on LTC cases, years, regions, or precipitation processes. The model inputs include summer monsoon diurnal variation feature vectors, low-level convergence, water vapor flux, water vapor flux divergence, LTC circulation maintenance, precipitation time series, relative azimuth, radial distance, and lag correlation features. Random forest, gradient boosting tree, support vector machine, logistic regression, or a combination thereof are used to establish an hourly extreme precipitation sample discrimination model. The output of the hourly extreme precipitation sample discrimination model includes at least the sample class probability, class confidence, diurnal variation impact assessment value, thermal impact assessment value, dynamic impact assessment value, water vapor transport impact assessment value, and initial ranking of key features. The model discrimination threshold, class probability threshold, or feature weight parameters are calibrated based on the validation sample set to reduce the impact of sample differences between different LTC cases on the discrimination results.

[0045] The classification model described above is not a replacement for rule-based discrimination, but rather provides an alternative computational approach for sample discrimination under large sample conditions. By dividing the comprehensive feature matrix into training samples, validation samples, and samples to be discriminated, and outputting class probabilities, class confidence scores, and diurnal variation impact assessment values, this embodiment can transform the assessment of the impact of summer monsoon diurnal variation on LTC hourly extreme precipitation into quantifiable, calibrable, and comparable model output results.

[0046] Furthermore, the contribution of key features is obtained by performing interpretability analysis on the input features in the comprehensive feature matrix. The interpretability analysis includes at least one of sensitivity perturbation, feature permutation, model weight decomposition, tree model node gain statistics, or local sample contribution decomposition. Specifically, the contribution weights of summer monsoon diurnal variation features, low-level convergence features, water vapor flux features, water vapor flux divergence features, LTC circulation maintenance features, precipitation time series features, relative azimuth features, radial distance features, and lag correlation features to the sample discrimination results, sample category probabilities, and diurnal variation impact assessment values ​​are calculated. The contribution weights are then grouped and summarized according to thermodynamic variables, dynamic variables, water vapor variables, LTC state variables, precipitation response variables, pressure layers, relative azimuth partitions, radial distance partitions, and time lag orders to generate key feature contribution quantities that can distinguish between thermodynamic influences, dynamic influences, water vapor transport influences, and circulation maintenance influences. At the same time, the time series variation, spatial distribution, and ranking results of the key feature contribution quantities are output to form hourly extreme precipitation discrimination results with physical interpretability.

[0047] By calculating the contribution of key features, this embodiment can further explain the main data sources and major influencing factors of the sample discrimination results. Specifically, the thermal impact assessment data can be obtained by summarizing fields such as lower-level temperature disturbance, humidity disturbance, air pressure disturbance, and saturation difference changes; the dynamic impact assessment data can be obtained by summarizing lower-level wind field disturbance, lower-level convergence, radial inflow, tangential wind, and circulation maintenance characteristics; and the water vapor transport impact assessment data can be obtained by summarizing water vapor flux, water vapor flux divergence, and water vapor input direction characteristics. The above results can be written into the sample attribute table and spatiotemporal distribution data table, providing data support for subsequent sample verification, model updates, and result interpretation.

[0048] In summary, this embodiment enables a complete data processing workflow, from multi-source data standardization, LTC hourly precipitation sample construction, typhoon dynamic reference space establishment, summer monsoon diurnal variation feature extraction, interaction feature fusion, to sample characterization and discrimination assessment output. This workflow supports LTC hourly extreme precipitation sample category discrimination and outputs diurnal variation impact assessment values, key feature contribution amounts, thermal impact assessment data, dynamic impact assessment data, and spatiotemporal distribution data, thereby improving the structure, reproducibility, and interpretability of multi-source meteorological data processing results.

[0049] Example 2: Sample discrimination and contribution output verification of a typical LTC process Based on Example 1 above, Example 2 selects a landfalling tropical cyclone process as the verification object to illustrate the sample discrimination results and key feature contributions of the method described in Example 1 of this invention. This LTC process contains 2864 hourly precipitation samples, of which 2718 are valid after quality control; 143 hourly extreme precipitation samples and 2575 non-extreme precipitation samples are obtained by labeling according to single-station empirical distribution thresholds. After SS3 processing, all samples are converted to the typhoon dynamic reference space, and summer monsoon diurnal variation characteristics are extracted within the dynamic feature region southwest of the LTC. The sample discrimination results of a typical LTC process are shown in Table 1 below.

[0050] Table 1. Sample discrimination results of a typical LTC process As shown in Table 1, the classification model, after comprehensively utilizing the diurnal variation characteristics of the summer monsoon, low-level convergence, water vapor flux, LTC circulation maintenance, and precipitation time series characteristics, demonstrates superior ability to identify hourly extreme precipitation samples compared to rule-based discrimination. In particular, the recall rate increased from 0.895 to 0.951, indicating that the comprehensive feature matrix can more fully capture the diurnal variation signals of the environmental field before and after the occurrence of extreme samples.

[0051] Table 2 Output Results of Key Feature Contribution Furthermore, the contribution of key features to the classification model output was calculated, yielding the results shown in Table 2. Table 2 shows that the summer monsoon diurnal variation feature has the highest contribution, indicating its strong discriminative role in the assessment of hourly extreme precipitation in this LTC. Specifically, the peak of positive disturbance occurred 2 hours earlier on average than the peak of hourly precipitation, the water vapor flux disturbance enhancement rate was 38.5%, and the low-level convergence disturbance enhancement rate was 26.2%, indicating a significant leading or synchronous relationship between the enhancement of summer monsoon diurnal variation and the occurrence of hourly extreme precipitation in the dynamic characteristic region southwest of the LTC.

[0052] This embodiment further summarizes the contributions of key features into three categories: thermal, dynamic, and water vapor transport. Dynamic impact assessment data accounts for 36.9%, water vapor transport impact assessment data accounts for 34.1%, thermal impact assessment data accounts for 21.8%, and the remainder is contributed by precipitation time series and location index. The above results demonstrate that this invention can not only output hourly extreme precipitation sample categories but also further provide diurnal variation impact assessment values, key feature contributions, and sub-assessment data for thermal, dynamic, and water vapor transport, thereby achieving a structured, quantifiable, and interpretable assessment of the impact of summer monsoon diurnal variation on LTC hourly extreme precipitation processes.

[0053] The objectives of this invention have been fully and effectively achieved through the above embodiments. Those skilled in the art will understand that this invention includes, but is not limited to, the contents described in the accompanying drawings and the specific embodiments described above. Although the invention has been described with reference to what is currently considered the most practical and preferred embodiments, it should be understood that the invention is not limited to the disclosed embodiments, and any modifications that do not depart from the functional and structural principles of the invention will be included within the scope of the claims.

Claims

1. A multi-source data processing method that integrates summer monsoon diurnal variation characteristics and is geared towards assessing LTC hourly extreme precipitation, characterized in that, It should include at least the following steps: SS1. Acquire daily precipitation data, hourly precipitation data from stations, LTC optimal path data, and hourly reanalysis environmental field data, and perform preprocessing to form a basic spatiotemporal dataset; SS2. Based on the objective weather map analysis method, identify the precipitation process affected by LTC, match the daily scale identification results with hourly precipitation data according to station, date and time, and generate a candidate sample set of LTC hourly precipitation, sample source identifier and corresponding spatiotemporal index; SS3. Establish a typhoon relative coordinate system with the hourly LTC center location as the dynamic origin, and hourly transform the station location, candidate precipitation samples and environmental field grid points to the typhoon relative coordinate system. Generate the dynamic characteristic area of ​​the environmental field southwest of LTC based on the relative orientation and distance. SS4. Extract the wind field, temperature field, humidity field and water vapor transport field of the lower troposphere within the dynamic feature area, calculate the daily variation disturbance based on the difference between the hourly value and the corresponding daily average value, and generate the summer monsoon daily variation feature vector and its time series index for fusion discrimination; SS5. Dynamically register the summer monsoon diurnal variation feature vector with low-level convergence features, water vapor flux features, LTC circulation maintenance features, and precipitation time series features to construct a comprehensive feature matrix to characterize the interaction between summer monsoon diurnal variation and LTC hourly extreme precipitation; SS6. Based on the comprehensive feature matrix, perform rule-based discrimination or classification model discrimination, and output the hourly extreme precipitation sample category, daily variation impact assessment value, key feature contribution, thermal impact assessment data, dynamic impact assessment data, and spatiotemporal distribution data.

2. The method according to claim 1, characterized in that, In step SS1, the daily precipitation data includes daily precipitation at the station, observation date, station number, station latitude and longitude, station altitude, and daily quality indicator; the hourly precipitation data includes hourly precipitation, observation time, station number, station latitude and longitude, station altitude, and hourly quality indicator; the LTC optimal path data includes the 6-hourly latitude and longitude position of the LTC center, the maximum wind speed near the center, the minimum air pressure at the center, and the intensity level, and the 6-hour interval records are completed into an hourly continuous time series through cubic spline interpolation; the hourly reanalysis environmental field data includes at least hourly wind field, temperature field, humidity field at 850hPa, 500hPa, and 200hPa, as well as surface air pressure field and water vapor flux field, and the horizontal resolution of the hourly reanalysis environmental field data is not less than 0.25°×0.25°.

3. The method according to claim 1, characterized in that, In step SS2, when identifying the LTC-affected precipitation process based on objective weather map analysis, at least the following sub-steps are included: S21. Path timing completion: Read the center position, maximum wind speed near the center and minimum air pressure near the center from the LTC optimal path data, interpolate the 6-hour path records into an hourly path sequence, and configure an LTC number and intensity status identifier for each hourly time. S22. Determination of the period of impact: Based on the hourly center location, landfall time, movement path and circulation status of LTC, determine the start and end periods of LTC's impact on the station's precipitation, and perform time overlap screening with the hourly precipitation data of the station; S23. Station Spatial Matching: Calculate the radial distance and azimuth of each station relative to the LTC center at the corresponding time. Based on the radial distance, azimuth, and the influence range of the LTC circulation, select hourly precipitation records of stations within the LTC precipitation influence range. S24. Weather pattern verification: Based on reanalysis environmental field data and combined with satellite cloud image data, identify the LTC circulation, peripheral rainband, low-level wind field and water vapor transport configuration, and verify the matching relationship between the filtered station precipitation records and the weather patterns affected by LTC. S25. Determination of precipitation attribution: Based on the location of the LTC center, the spatial matching results of the station, the period of influence, the weather pattern verification results, and the continuity of precipitation records, the LTC influence attribution is determined for the hourly precipitation records, and candidate samples of LTC hourly precipitation are generated. S26. Sample Field Configuration: Configure the LTC number, station number, station latitude and longitude, station altitude, date, time, precipitation, relative distance, relative orientation, attribution determination identifier, and sample quality identifier for each LTC hourly precipitation candidate sample.

4. The method according to claim 1, characterized in that, In step SS3, the typhoon relative coordinate system uses the hourly LTC center position as the coordinate origin and the due north direction or the LTC movement direction as the azimuth reference direction. The station position, candidate precipitation samples, and environmental field grid points are converted into relative radial distance, relative azimuth, meridional displacement, and zonal displacement relative to the coordinate origin. The meridional and zonal displacements are calculated by spherical distance or map projection distance. The relative radial distance is synthesized from the meridional and zonal displacements. The relative azimuth is determined by the meridional displacement, zonal displacement, and azimuth reference direction. The relative azimuth is used to divide the samples into the northeast, southeast, southwest, and northwest quadrants to form a dynamic reference space that is updated synchronously with the LTC movement process.

5. The method according to claim 1 or 4, characterized in that, In step SS3, when generating the dynamic characteristic region of the environmental field on the southwest side of the LTC, based on the relative azimuth and relative radial distance in the typhoon's relative coordinate system, reanalysis environmental field grid points located in the southwest quadrant of the LTC center and within a preset radial range are selected as candidate grid points. The candidate grid points are then screened in conjunction with the LTC circulation scale, movement direction, and spatial resolution of the reanalysis data. The screened candidate grid points are weighted according to their distance from the LTC center, the area represented by the grid point, or the validity of environmental field variables. Hourly outputs are then used to characterize the summer monsoon input, low-level water vapor transport, low-level convergence, environmental thermal conditions, and peripheral circulation configuration on the southwest side of the LTC. These regional statistics are then written into the dynamic characteristic region index.

6. The method according to claim 1, characterized in that, Step SS4, the extraction of diurnal variation characteristics of the summer monsoon, includes at least: S41. Low-level variable extraction: Extract zonal wind, meridional wind, temperature, relative humidity or specific humidity, water vapor flux and surface air pressure variables at heights of 850hPa, 925hPa and / or 1000hPa within the dynamic feature area, and assign height layer, grid point, time, data source and effective layer identifier to each variable. S42. Diurnal variation cycle division: Divide a continuous 24 hours into a diurnal variation cycle according to the local solar time, and assign each hourly time to the corresponding diurnal variation cycle, so that environmental field variables, dynamic characteristic grid points and LTC hourly precipitation candidate samples correspond to each other in the same diurnal variation time frame. S43. Calculation of daily average background amount: For each daily change cycle, each variable, each height layer and each dynamic feature area, calculate the regional weighted daily average based on the effective grid point set. The regional weighted daily average is determined by the area represented by the grid point and the distance weight from the grid point to the LTC center. S44. Generation of daily variation disturbances: The difference between each hourly variable value and the corresponding regional weighted daily average value is used to generate disturbances in wind field, temperature, humidity, water vapor transport and air pressure. The disturbances are written into the disturbance data table according to LTC number, sample time, relative azimuth, radial distance and height layer identifier. S45. Phase index extraction: Identify the peak times of positive disturbances, peak times of negative disturbances, and times of disturbance sign transitions in the hourly disturbance sequence within the same diurnal variation cycle, and generate diurnal variation phase indices to characterize the strengthening, weakening, and turning processes of summer monsoon diurnal variation; S46. Intensity normalization: Based on the standard deviation, interquartile range, or range after outlier removal of the same variable in the study area or the same LTC process, normalize each disturbance quantity to generate standardized disturbance intensity that can be compared across cases, time periods, and variables. S47. Calculation of intensity evolution characteristics: Based on the standardized disturbance intensity, and combined with the wind field disturbance, humidity disturbance and water vapor transport disturbance, calculate the regional average disturbance intensity, disturbance duration and disturbance intensification rate to generate dynamic summer monsoon diurnal variation intensity characteristics. S48. Time-series feature encoding: Based on the disturbance amount, diurnal variation phase index and dynamic summer monsoon diurnal variation intensity characteristics, the summer monsoon diurnal variation disturbance sequence is encoded according to LTC number, diurnal variation period, dynamic feature area number and sample time.

7. The method according to claim 1 or 6, characterized in that, In step SS4, the summer monsoon diurnal variation feature vector includes at least wind field disturbance, thermal disturbance, water vapor transport disturbance, disturbance phase, and disturbance persistence components. The wind field disturbance component includes zonal wind disturbance, meridional wind disturbance, wind speed disturbance amplitude, and wind direction disturbance angle. The thermal disturbance component includes temperature disturbance, humidity disturbance, and surface pressure disturbance. The water vapor transport disturbance component includes water vapor flux disturbance, lower-level water vapor input intensity, and water vapor flux divergence disturbance. The disturbance phase component includes disturbance peak frequency, disturbance trough frequency, and peak-trough phase difference. The disturbance persistence component includes disturbance continuous enhancement duration, disturbance continuous weakening duration, and disturbance enhancement rate. Each component is encoded into a summer monsoon diurnal variation feature vector with a unified field structure.

8. The method according to claim 1, characterized in that, In step SS5, the low-level convergence characteristics are calculated by negative divergence or radial inflow of the horizontal wind field in the lower troposphere within the dynamic characteristic region; the water vapor flux characteristics are calculated by coupling the horizontal wind field in the lower troposphere within the dynamic characteristic region with specific humidity data, forming water vapor flux intensity, water vapor flux divergence, and water vapor transport direction characteristics; the LTC circulation maintenance characteristics include the LTC center minimum pressure, near-center maximum wind speed, moving speed, moving direction, duration after landfall, relative topographic position, and statistics of tangential and radial winds in the typhoon's relative coordinate system; the precipitation time series characteristics include hourly precipitation, previous cumulative precipitation, precipitation enhancement rate, precipitation duration, and the number of times peak precipitation occurs; and variable identifiers, time identifiers, spatial location identifiers, and sample category identifiers are configured for each characteristic to form the basic characteristics fused with the summer monsoon diurnal variation characteristic vector.

9. The method according to claim 1 or 8, characterized in that, In step SS5, when constructing the comprehensive feature matrix, firstly, sub-matrices for summer monsoon diurnal variation, low-level convergence, water vapor transport, LTC circulation maintenance, precipitation time series, and lag correlation are established respectively; then, matrix row and column indexes are established using LTC number, station number, sample time, relative azimuth, radial distance, height layer identifier, and sample category identifier; the sub-matrices are spliced ​​and missing measurement masking is performed according to a unified field structure to form a comprehensive feature matrix that includes diurnal variation features of the environmental field, background features of typhoon circulation, water vapor transport features, precipitation response features, and interaction time series features.

10. The method according to claim 1, characterized in that, In step SS6, the rule discrimination includes a hierarchical discrimination process based on extreme precipitation constraints, diurnal variation disturbance constraints, LTC circulation maintenance constraints, and thermodynamic-dynamic-water vapor co-constraints, wherein: First, based on the absolute threshold of hourly precipitation, the regional quantile threshold, or the historical extreme value threshold of the station, it is determined whether the candidate samples meet the conditions for hourly extreme precipitation. Then, it is determined whether the diurnal variation disturbances of low-level convergence, water vapor flux, water vapor flux divergence, low-level temperature and humidity disturbances, and vertical motion disturbances in the dynamic feature area of ​​the comprehensive feature matrix reach the corresponding enhancement thresholds. Subsequently, it is determined whether the minimum pressure at the LTC center, the maximum wind speed near the center, the low-level tangential wind, the radial inflow, the movement speed, the duration after landfall, and the low-level circulation intensity indicators meet the conditions for the continuous influence of circulation. Finally, based on whether the enhancement of low-level convergence, the enhancement of water vapor transport, the enhancement of humid and hot conditions, and the enhancement of upward motion occur simultaneously or within a preset leading period, hourly extreme precipitation sample categories are generated that are significantly affected by the enhancement of summer monsoon diurnal variation, weakly affected by the summer monsoon diurnal variation, or have no significant diurnal variation influence. The corresponding diurnal variation impact assessment, dynamic impact assessment, thermal impact assessment, and water vapor transport impact assessment data are output.

11. The method according to claim 1, characterized in that, In step SS6, the classification model discrimination includes supervised learning discrimination based on the comprehensive feature matrix using labeled samples, including: The comprehensive feature matrix is ​​divided into training samples, validation samples, and samples to be discriminated based on LTC cases, years, regions, or precipitation processes. The model inputs include summer monsoon diurnal variation feature vectors, low-level convergence, water vapor flux, water vapor flux divergence, LTC circulation maintenance, precipitation time series, relative azimuth, radial distance, and lag correlation features. Random forest, gradient boosting tree, support vector machine, logistic regression, or a combination thereof are used to establish an hourly extreme precipitation sample discrimination model. The output of the hourly extreme precipitation sample discrimination model includes at least the sample class probability, class confidence, diurnal variation impact assessment value, thermal impact assessment value, dynamic impact assessment value, water vapor transport impact assessment value, and initial ranking of key features. The model discrimination threshold, class probability threshold, or feature weight parameters are calibrated based on the validation sample set to reduce the impact of sample differences between different LTC cases on the discrimination results.

12. The method according to claim 1, 10, or 11, characterized in that, In step SS6, the contribution of key features is obtained by performing interpretability analysis on the input features in the comprehensive feature matrix. This interpretability analysis includes at least one of sensitivity perturbation, feature permutation, model weight decomposition, tree model node gain statistics, or local sample contribution decomposition. The contribution weights of summer monsoon diurnal variation characteristics, low-level convergence characteristics, water vapor flux characteristics, water vapor flux divergence characteristics, LTC circulation maintenance characteristics, precipitation temporal characteristics, relative azimuth characteristics, radial distance characteristics, and lag correlation characteristics to sample discrimination results, sample category probabilities, and diurnal variation impact assessment values ​​are calculated separately. These contribution weights are then grouped and summarized according to thermodynamic variables, dynamic variables, water vapor variables, LTC state variables, precipitation response variables, pressure layers, relative azimuth partitions, radial distance partitions, and time lag orders to generate key feature contribution quantities that can distinguish between thermodynamic influences, dynamic influences, water vapor transport influences, and circulation maintenance influences. Simultaneously, the temporal variation, spatial distribution, and ranking results of the key feature contribution quantities are output to form hourly extreme precipitation discrimination results with physical interpretability.

13. A computer program product, characterized in that, Includes computer instructions for performing the steps of the multi-source data processing method according to any one of claims 1 to 12, which integrates summer monsoon diurnal variation characteristics and is geared towards LTC hourly extreme precipitation assessment.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When a computer program is executed by a processor, it implements the steps of the multi-source data processing method according to any one of claims 1 to 12, which integrates summer monsoon diurnal variation characteristics and is geared towards LTC hourly extreme precipitation assessment.

Images