A method for inland TRP feature recognition and evaluation based on data driving and numerical simulation, medium and program product

Abstract

The application discloses a kind of inland area TRP feature identification and evaluation method, medium and program product based on data driving and numerical simulation, it is related to meteorological disaster monitoring and early warning and numerical simulation technical field, by comprehensively utilizing precipitation observation, tropical cyclone path, reanalysis and other multi-source data, in combination with the radius of typhoon peripheral circulation, distance constraint and whole layer water vapor transport flux objectively identify the TRP event occurring outside typhoon circulation and connected with strong water vapor transport zone.Combining the radius of typhoon peripheral circulation, distance constraint and whole layer water vapor transport flux objectively identify the TRP event occurring outside typhoon circulation and connected with strong water vapor transport zone.Using regional numerical model and reanalysis data, the key physical fields such as whole layer water vapor flux, water vapor budget, dry intrusion and monsoon water vapor surge are diagnosed to form a set of characteristic indexes reflecting the coupling influence of typhoon, monsoon and dry cold air;On this basis, a data-driven learning algorithm is introduced to train a TRP occurrence and heavy rain intensity classification identification model, and a risk assessment product is generated through calibration and risk classification to improve the identification and evaluation capability of inland typhoon long-distance rainstorm.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of meteorological disaster monitoring and early warning and numerical simulation, and particularly relates to a method for extracting, identifying and risk assessing the characteristics of typhoon remote precipitation by combining multi-source observation data, reanalysis data and regional numerical model output results, using data-driven modeling technology, especially to a method for identifying and assessing the characteristics of typhoon remote precipitation in inland areas based on data-driven and numerical simulation, medium and program product. BACKGROUND

[0002] Typhoon remote precipitation (TRP) refers to a type of precipitation process occurring outside the typhoon circulation, with spatial separation from the typhoon body precipitation, but closely related to the typhoon through atmospheric water vapor transport and dynamic coupling. Studies have shown that TRP is an important form of summer heavy precipitation and extreme rainstorm in the inland areas of China, with strong disaster-causing, large uncertainty in the falling area, and relatively low predictability, posing a serious challenge to disaster prevention and mitigation in inland areas.

[0003] For the identification and assessment of TRP, the existing technologies mainly include: (1) subjective or subjective and objective combined weather identification, which usually relies on comprehensive experience analysis of weather situation such as typhoon position, low-level jet, water vapor channel and evolution of trough and ridge, with the problems of non-uniform standards and insufficient repeatability; (2) objective threshold-based identification method, such as using precipitation-typhoon center distance, circulation outside the band width or integrated vapor transport (IVT) threshold as a constraint, which improves objectivity, but still lacks in depicting water vapor transport path under the background of multiple typhoons, complex terrain and monsoon superposition; (3) numerical model sensitivity test can analyze the interaction of typhoon, monsoon, subtropical high and mid-high latitude systems, but has high calculation cost, limited business real-time performance, and systematic deviation in extreme precipitation area and intensity when lacking data-driven calibration and post-processing.

[0004] In recent years, data-driven and machine learning techniques have shown good potential in severe convective identification, model post-processing and ensemble prediction calibration. However, there are three challenges in directly applying machine learning to TRP scenarios: first, the fusion of multi-source data (ground precipitation observations, reanalysis data, satellite products and numerical model output, etc.) and the lack of standard procedures for feature construction make it difficult to fully extract key spatiotemporal structure features and physical quantity combinations related to TRP; second, the imbalance of training samples (TRP and non-TRP, extreme and non-extreme) and the spatiotemporal non-stationarity result in insufficient model generalization ability and interpretability; third, the coupling mechanism of data-driven results and numerical model physical consistency is not perfect, and a closed-loop system based on numerical simulation diagnosis results to dynamically test and correct data-driven identification conclusions has not been formed, making it difficult to achieve reliable risk quantification and uncertainty assessment of TRP identification results.

[0005] In summary, the existing technology still has obvious deficiencies in the objective identification, adaptive feature extraction and integration of risk assessment of TRP, especially lacking a comprehensive method system for inland areas of China that can balance the physical process reliability and data-driven identification ability. Therefore, how to integrate numerical simulation diagnosis and data-driven modeling technology based on long time series multi-source data, construct a feature recognition and quantitative evaluation method suitable for inland areas of typhoon remote rainstorm scenarios, and realize automatic identification, type division and risk grading of TRP events is a technical problem to be solved in the field. SUMMARY

[0006] (I) Invention purposes

[0007] In view of the above-mentioned defects and deficiencies in the prior art, the purpose of the present application is to provide a data-driven and numerical simulation-based inland TRP feature recognition and evaluation method, medium and program product, by constructing a database containing long time series of typhoon remote precipitation samples, combining reanalysis data and regional numerical model output, uniformly preprocessing and quantifying the multi-source physical quantities related to inland typhoon remote rainstorm such as water vapor transport, circulation pattern, dry intrusion activity, etc. On this basis, data-driven modeling technology is introduced to learn the key feature combinations and discrimination boundaries of different types of inland typhoon remote rainstorm events, forming a data-driven identification model and a risk evaluation model with physical constraints. Through the present application, automatic identification, classification and intensity grading of typhoon remote rainstorm processes can be realized for different inland areas, the comprehensive influence of typhoon position, intensity and its coupling with monsoon, trough ridge and other systems on the development of rainstorm is quantitatively described, and the objectivity, refinement and intelligence level of typhoon remote rainstorm feature recognition in inland areas are significantly improved, providing a scientific basis for extreme precipitation disaster risk assessment, early warning threshold setting and disaster prevention and mitigation decision-making in inland areas.

[0008] (II) Technical Solution

[0009] To achieve the object of the present application and solve the technical problems, the present application adopts the following technical solution:

[0010] The first object of the present application is to provide a method for identifying and evaluating inland TRP based on data driving and numerical simulation, which is used for objectively identifying, evaluating the intensity and classifying the risk of the inland TRP process in the middle and high latitude regions affected by the northwest Pacific typhoon, and at least includes the following steps:

[0011] S100. Data acquisition and unified preprocessing: acquiring ground precipitation observation data, tropical cyclone path data, reanalysis data and / or numerical prediction data, and preprocessing the multi-source data according to the latitude and longitude range of the target inland region and the target period to form a unified gridded basic data set;

[0012] S200. Typhoon process matching and inland region screening: identifying the typhoon cases that have potential impact on the target inland region in the target period according to the tropical cyclone path data, establishing the spatio-temporal matching record of typhoon-inland precipitation based on the spatial relationship between the typhoon center position and the inland region, and screening to obtain the typhoon process set associated with the target inland region;

[0013] S300. Objective identification and sample labeling of TRP: based on the preset TRP identification rule, the precipitation field under each typhoon process is screened and labeled for long-distance precipitation samples, a TRP sample library of inland regions including TRP, non-TRP and different precipitation intensity level labels is constructed, and a unified gridded sample data set is formed;

[0014] S400. Numerical simulation configuration and integrated calculation: selecting several representative inland TRP cases and / or typical circulation types from the sample library, setting the integral region, initial boundary value field and physical parameterization scheme based on the regional numerical model, driving the model for numerical integration, and obtaining a high-resolution numerical model simulation output data set;

[0015] S500. Physical feature diagnosis and feature index construction: using the numerical model output data set and reanalysis data, the physical diagnosis of typhoon position and intensity evolution, main water vapor transport channel and its flux divergence, dry and cold air intrusion activity, monsoon water vapor surge intensity, and mid-high latitude trough-ridge configuration is performed, and a set of physical feature fields and feature indexes related to inland TRP is constructed;

[0016] S600. Data-driven feature extraction and identification model training: based on the unified gridded sample data and the physical feature index set, a training and verification sample set is constructed, a preset data-driven learning algorithm is used to train the model, cross-validate and optimize the parameters of the training sample, and a target identification model for identifying and evaluating the intensity of inland TRP is obtained.

[0017] S700. Rainstorm risk assessment and result output: based on the numerical prediction field of the target prediction time, reanalysis data and / or real-time observation data, an input feature set consistent with the training stage is generated and input into the target identification model to obtain the TRP occurrence probability and estimated precipitation intensity level of each grid or station in the target inland area, and the risk of the typhoon long-distance rainstorm in the inland area is assessed according to the preset risk classification threshold.

[0018] The second inventive purpose of the present application is to provide a computer program product comprising computer instructions for executing the steps of the above-mentioned data-driven and numerical simulation-based inland TRP feature identification and evaluation method of the present application.

[0019] The third inventive purpose of the present application is to provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-mentioned data-driven and numerical simulation-based inland TRP feature identification and evaluation method.

[0020] (Three) Technical effects

[0021] Compared with the prior art, the data-driven and numerical simulation-based inland TRP feature identification and evaluation method, medium and program product of the present application have the following beneficial and significant technical effects:

[0022] (1) Based on the existing TRP objective identification idea, the present application constructs a hierarchical identification process of rainband separation-typhoon peripheral distance constraint-strong IVT contact area superposition for the mid-high latitude inland area, introduces parameters such as typhoon intensity-dependent peripheral circulation radius, adaptive safety bandwidth and maximum influence radius, realizes the screening and labeling of TRP events occurring outside the typhoon circulation and closely connected with the strong water vapor transport belt, and significantly improves the consistency, repeatability and regional adaptability of TRP identification.

[0023] (2) The present application systematically constructs a complete set of physical feature index system including whole-layer water vapor flux and its divergence, box-type water vapor budget, whole-layer dry intrusion index, dynamic regional monsoon water vapor surge index and J-type / S-type water vapor transport type through the coupling diagnosis of regional numerical model and reanalysis data, quantitatively describes the comprehensive influence of typhoon, monsoon, dry and cold air and mid-high latitude trough-ridge on inland TRP rainstorm. This physical feature set not only retains key mechanism information such as water vapor transport structure, cold and warm air mass interaction and precipitation efficiency, but also inputs the data-driven model in the form of quantifiable indexes, realizing the organic unification of physical interpretability and statistical identification ability.

[0024] (3) The application introduces a data-driven learning algorithm on the basis of the TRP sample library and physical characteristic indicators, combines sample imbalance processing, cross-validation according to typhoon cases, and feature importance analysis, constructs a target recognition model suitable for TRP feature recognition and intensity classification in inland areas, and further converts the model output into a spatialized risk level product through probability calibration and a two-dimensional risk classification matrix. Compared with a method of relying only on numerical mode direct precipitation prediction or empirical index discrimination, the application can significantly improve the objectivity and stability of TRP occurrence probability and storm intensity level evaluation, and provides strong technical support for fine risk identification and disaster prevention and mitigation decision-making of TRP in inland areas. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 An implementation flowchart of the TRP feature recognition and evaluation method in inland areas based on data-driven and numerical simulation is shown. DETAILED DESCRIPTION

[0026] The application aims to provide a TRP feature recognition and evaluation method in inland areas based on data-driven and numerical simulation, a medium and a program product, which are used for objectively identifying, evaluating the intensity and classifying the risk of a TRP process in a high-latitude inland area affected by a typhoon in the northwest Pacific. In order to make the purpose, technical scheme and advantages of the application clearer, the technical scheme in the embodiments of the application will be described in more detail below in combination with the drawings in the embodiments of the application. The described embodiments are part of the embodiments of the application, not all the embodiments, and the described embodiments are exemplary and are intended to explain the application, and cannot be understood as a limitation on the application.

[0027] Embodiment 1: TRP feature recognition and evaluation method in inland areas

[0028] As a specific embodiment, as shown in the drawings, Figure 1 The TRP feature recognition and evaluation method in inland areas based on data-driven and numerical simulation provided by the embodiments of the application is constructed by data preparation, typhoon process screening, TRP objective recognition, numerical simulation and physical diagnosis, data-driven modeling and risk evaluation for a TRP process in a high-latitude inland area affected by a typhoon in the northwest Pacific. Specifically, in this embodiment, the method can include the following steps in sequence when implemented:

[0029] S100. Data acquisition and unified pretreatment:

[0030] The ground precipitation observation data, tropical cyclone path data, reanalysis data and / or numerical prediction data are obtained, and the multi-source data is preprocessed according to the longitude and latitude range of the target inland area and the target period to form a unified gridded basic data set. As preferred, the ground precipitation observation data includes hourly or daily precipitation data of the meteorological station within the target inland area and at least 500 kilometers around the target inland area; the tropical cyclone path data at least includes 6-hour interval records of typhoon number, central position longitude and latitude, central pressure, maximum wind speed and moving direction; the reanalysis data at least includes three-dimensional grid data of temperature field, geopotential height field, wind field and specific humidity field in the ERA5 or NCEP reanalysis data set, with a time resolution of not less than 6 hours and vertical levels covering 1000-100 hPa; the numerical prediction data is prediction product data based on a global numerical prediction model or a regional numerical model.

[0031] In addition, step S100 also includes: performing site height correction, missing value filling and outlier removal on the ground precipitation observation data, and using a time stamp alignment method to uniformly resample the hourly or daily precipitation observation data and the reanalysis data and the numerical prediction data; by using a bilinear interpolation or a nearest neighbor interpolation method, the multi-source data with different spatial resolutions are unified to a preset regular longitude and latitude grid, and a buffer extension processing is performed on the edge grid points, so that the outside of the typhoon circulation and the long-distance water vapor transport channel have continuous and smooth spatiotemporal expression in the basic data set, thereby improving the accuracy and stability of subsequent TRP identification and feature diagnosis.

[0032] S200. Typhoon process matching and inland area screening:

[0033] According to the tropical cyclone path data, a typhoon case that has a potential impact on the target inland area in the target period is identified, a spatiotemporal matching record of typhoon-inland precipitation is established based on the spatial relationship between the typhoon center position and the inland area, and a typhoon process set associated with the target inland area is screened.

[0034] As preferred, the typhoon process matching and inland area screening includes: based on the tropical cyclone path data, setting a typhoon generation area, a minimum central pressure or a maximum wind speed intensity threshold, and removing tropical depressions, extratropical cyclones and cases with extremely weak impact on the target inland area; in the time dimension, a matching time window is constructed by extending a preset number of hours forward and backward from the time when the typhoon is closest to the target inland area; in the spatial dimension, only the typhoon processes within a preset distance range and in the matching time window are retained, which have a significant precipitation response to the inland area, so as to reduce the interference of invalid typhoon cases on the sample library construction.

[0035] It should be noted that the specific typhoon generating area, intensity threshold, matching time window and spatial distance threshold in step S200 can be adjusted and optimized according to the geographical location, business demand and data characteristics of different inland areas, and can also be calibrated through historical case back calculation or sensitivity test; the selection and value change of the above parameters do not affect the basic idea of the present application about matching and screening of typhoon process and inland area precipitation. In addition, when establishing the space-time matching record of typhoon-inland precipitation, the present application also considers the sustained influence effect after the typhoon changes. When the typhoon enters the middle and high latitudes and weakens into a warm cyclone or a warm low pressure, although its circulation structure changes significantly, it can still cause long-distance precipitation in the inland area through the interaction of the residual vortex and the westerly trough ridge system in the middle and high latitudes. Therefore, when screening the typhoon process set, not only the influence period of the active period of the typhoon is retained, but also the matching time window is extended backward to 24-48 hours after the typhoon completely dissipates, so as to completely capture the potential precipitation contribution of the whole process of the typhoon life history to the target inland area, and ensure the completeness and representativeness of the TRP sample library.

[0036] S300. TRP objective recognition and sample labeling:

[0037] Based on the preset TRP recognition rule, the long-distance precipitation samples of each typhoon process are screened and labeled, and the inland TRP sample library including TRP, non-TRP and different precipitation intensity level labels is constructed, and a unified gridded sample data set is formed. Specifically, in the implementation of the objective recognition of TRP, the following sub-steps can be included:

[0038] S301. Precipitation rainband separation and typhoon body rainband recognition: the connected domain clustering and morphological processing method is used to divide the precipitation field into several independent precipitation rainband units, and the typhoon body precipitation rainband is identified according to the high-intensity precipitation connected body in the area where the typhoon center position and its maximum wind speed radius are located, and the body rainband and its adjacent several grid points are excluded from the subsequent long-distance precipitation judgment;

[0039] S302. Potential TRP candidate rainband screening: for each non-body precipitation rainband, the spherical distance Dts between each precipitation grid point in the rainband and the typhoon center is calculated, and the corresponding typhoon peripheral circulation radius D1 is determined according to the current intensity of the typhoon, when D1 + ΔD≤Dts≤D0, the rainband is marked as a potential TRP candidate rainband, wherein ΔD is a preset safety bandwidth, and D0 is a preset maximum influence radius;

[0040] S303. Typhoon strong water vapor transport path and contact area identification: On the whole layer water vapor transport flux IVT field corresponding to the time of the typhoon, the dominant water vapor transport path is tracked step by step along the maximum direction of IVT vector with the typhoon center as the starting point, a number of rectangular contact areas with preset latitude and longitude range are constructed around each path point, and the grid point set with IVT greater than the preset strong water vapor transport judgment threshold IVT0 is extracted in each contact area to form a strong water vapor transport belt mask extending outward from the typhoon center;

[0041] S304. Spatial superposition of candidate rain belt and strong water vapor transport belt and TRP determination: The potential TRP candidate rain belt is spatially superimposed on the strong water vapor transport belt mask on a unified grid, and only when the candidate rain belt is located in the north quadrant of the typhoon center, there is continuous overlap with the strong water vapor transport belt, and there is at least one to several low precipitation or no precipitation grid points as spatial spacing between the candidate rain belt and the typhoon body rain belt, the candidate rain belt is marked as a far distance precipitation rain belt meeting the TRP condition, and the grid points or stations contained in the rain belt are marked as TRP samples, and are included in the inland area TRP sample library.

[0042] Further, in step S300, the sample labeling further comprises: according to the daily precipitation and the process cumulative precipitation threshold, the TRP event is intensity graded and labeled, and the intensity grading at least includes moderate to heavy rain level, heavy rain level, heavy rain level and heavy rain level, wherein the daily precipitation of moderate to heavy rain level is 10-50 mm, the daily precipitation of heavy rain level is 50-100 mm, the daily precipitation of heavy rain level is 100-250 mm, and the daily precipitation of heavy rain level is not less than 250 mm; and in space, the regions exceeding the same intensity threshold in the continuous grid points or adjacent stations are connected and clustered, which are regarded as the spatial unit of the same TRP event, and further, the adjacent time event is implemented in the time dimension Process merging, finally forming a sample label system with TRP properties and heavy rain intensity level properties, providing unified and fine-grained supervision information for subsequent data-driven identification model.

[0043] It should be noted that the distance threshold, typhoon peripheral circulation radius, safety bandwidth, maximum influence radius and IVT threshold and other parameters used by the TRP in step S300 to identify can be set by region or adaptively adjusted according to data resolution, typhoon climate characteristics and business standards; the rainstorm intensity classification threshold can also be revised according to different national or industry standards. In addition, the present application emphasizes the dual constraints of physical connectivity and spatial isolation when labeling TRP samples, avoiding misjudgment of the typhoon body rain belt expansion as long-distance precipitation; in the missing or sparse station network area, reanalysis or high-resolution numerical grid precipitation can be used as a substitute basis, but the data source and confidence level should be explicitly labeled in the label; when multiple typhoons coexist or interact with the westerly short wave, resulting in the superposition of multiple source and sink transport paths, a hierarchical labeling strategy of dominant path identification and secondary path labeling should be used to ensure the interpretability and consistency of the training data.

[0044] S400. Numerical simulation configuration and integrated calculation:

[0045] Select several representative inland TRP cases and / or typical circulation types from the sample library, set the integral region, initial boundary field and physical parameterization scheme based on the regional numerical model, drive the model for numerical integration, and obtain the model output data set.

[0046] As preferred, in step S400, a multi-nested regional numerical weather prediction model is used, which has a horizontal resolution that is gradually increased from the outer layer to the inner layer, and the innermost layer region covers the target inland TRP landing area and the key water vapor transport channel; when setting the initial boundary field, the typhoon vortex structure initialization and relocation technology is introduced to assimilate and correct the typhoon position, intensity and radial wind field in the large-scale field, so as to ensure that the typhoon circulation in the model is consistent with the actual observation; in the selection of physical parameterization scheme, the combination of schemes suitable for strong precipitation simulation is adopted for radiation, deep and shallow convection, cloud microphysics and planetary boundary layer processes, and high time resolution three-dimensional dynamic, thermal and water vapor diagnostic variables are output during the integration process, providing a high-quality data basis for subsequent physical feature diagnosis.

[0047] Further, step S400 also includes model result physical consistency verification: comparing the typhoon path, central pressure evolution, maximum wind speed and TRP landing area of the numerical simulation with the corresponding tropical cyclone path data, reanalysis data and / or precipitation observation, calculating the position deviation, intensity deviation and / or precipitation distribution deviation; when the error index exceeds the preset threshold, adjust the model parameterization scheme or the initial field and re-integrate until the simulated typhoon structure, water vapor transport form and TRP spatial distribution meet the preset consistency requirements.

[0048] It should be noted that the area numerical model in step S400 can use WRF model, or other similar function area numerical weather prediction model; the grid resolution, the number of nested layers, the integration time and the combination of physical parameterization scheme can be flexibly configured according to the computing resources, research purposes and prediction time limit. For example, three layers of nested grid can be set, the outer grid resolution is 27 kilometers covering the East Asian region, the middle grid resolution is 9 kilometers covering the eastern region, and the inner grid resolution is 3 kilometers focusing on the target inland region; the initial field and boundary condition uses ERA5 reanalysis data or GFS global forecast data, and the time interval is 6 hours; the physical parameterization scheme selects WSM6 microphysical scheme, Kain-Fritsch cumulus convection scheme, YSU boundary layer scheme and RRTMG radiation scheme; the model integration time is set to 24 hours before the typhoon enters the research domain to 48 hours after it leaves, and the output time interval is 1 hour; the baseline simulation result is obtained through the control test, which is used as the reference baseline for comparison and analysis of subsequent sensitivity test, to ensure that the numerical simulation can reasonably reproduce the main physical characteristics of TRP process.

[0049] It should also be pointed out that the numerical simulation configuration needs to consider both the computing cost and the physical integrity, and the convection scale precipitation and topographic forcing should be prioritized in the innermost grid; when there are regional gaps in assimilated data or large initialization uncertainty in typhoon vortex, the ensemble initial value and boundary disturbance can be used to quantify the prediction distribution, thereby providing uncertainty quantification information in subsequent risk assessment; the sensitivity test of model parameters should use unified evaluation index and stopping criterion to avoid overfitting of single instance and ensure the robustness of the scheme under different years and circulation patterns.

[0050] S500. Physical feature diagnosis and characteristic index construction:

[0051] Using the numerical model output data set and reanalysis data, the physical diagnosis of typhoon position and intensity evolution, main water vapor transport channel and its flux divergence, dry and cold air intrusion activity, monsoon water vapor surge intensity, and mid-high latitude trough ridge configuration is performed, and the physical feature field and characteristic index set related to inland TRP are constructed. As an optimization, the physical feature diagnosis and characteristic index construction at least includes the following sub-steps when implemented:

[0052] S501. Water vapor transport structure and water vapor flux divergence diagnosis: Based on the model output and the corresponding reanalysis data at the same time, the whole layer water vapor flux and water vapor flux divergence in the target inland TRP landing area and its upstream region are calculated. The flux divergence is decomposed into the wind field convergence term and the humidity advection term, respectively, to represent the contribution of dynamic convergence and humidity transport to the formation of TRP. The path profile is constructed along the dominant water vapor transport path from the typhoon center to the TRP landing area, and the path integral IVT intensity, path length, maximum IVT and its occurrence position are extracted as the basic characteristics of the description of the spatial organization structure of the wet side water vapor transport capacity;

[0053] S502. Box water vapor budget and precipitation efficiency diagnosis: A fixed or time-updated evaluation area is constructed in the TRP landing area and its vicinity. The area integral and boundary line integral of water vapor flux in the evaluation area are calculated to calculate the water vapor input and output flux of each side boundary, as well as the whole layer water vapor content change rate in the region. Combined with the observed or simulated precipitation flux, the water vapor balance relationship is obtained, and the net water vapor input flux, precipitation efficiency and vertical integral wet layer thickness index are constructed to quantitatively describe the differences in water vapor budget structure of different TRP events.

[0054] S503. Dry and cold air intrusion and whole layer dry intrusion index construction: In the dry and cold air source area and intrusion channel upstream of TRP, based on temperature, specific humidity, potential vorticity and vertical velocity field, dry intrusion air mass with high potential vorticity, low specific humidity and significant sinking motion is identified. The isobaric whole layer dry intrusion index and / or isentropic whole layer dry intrusion index are constructed by vertically integrating and time integrating the dry intrusion signal along the isobaric surface or isentropic surface, which are used to represent the intensity, intrusion depth and duration of dry and cold air intrusion. The maximum value, time lag and spatial overlap area characteristics of the dry intrusion index within a certain time window before and after the occurrence of TRP rainstorm are extracted as physical indicators to describe the dynamic-thermal inhibition and triggering conditions on the cold side.

[0055] S504. Dynamic regional monsoon water vapor surge index construction: In the monsoon water vapor transport area on the south or southwest side of the typhoon, the monsoon water vapor surge evaluation area is dynamically defined according to the longitude and latitude of the typhoon center, 7-level wind circle radius and central pressure, and the vertical integral of the meridional wind field weighted by specific humidity is performed in the evaluation area to construct the dynamic monsoon water vapor surge index, which is used to reflect the intensity and time evolution of the monsoon water vapor surge associated with the typhoon. The increase, peak value and duration of the index before and after the formation of TRP are extracted as indicators to describe the warm side monsoon water vapor supply capacity.

[0056] Further, in step S500, the TRP event water vapor transport pattern is also classified: by calculating the geometric relationship and azimuth angle distribution of the typhoon center, the target inland TRP drop zone centroid and the dominant water vapor transport path, the water vapor transport pattern of the TRP event is divided into J-type transport type characterized by northwest block water vapor convergence of the typhoon and S-type transport type characterized by long-distance S-shaped water vapor transport belt. The typical values and combination characteristics of physical quantities such as whole-layer water vapor flux divergence, dry intrusion index, dynamic monsoon water vapor surge index and precipitation efficiency under different transport types are counted respectively, and the transport type category label and multi-source diagnostic index are used to form a physical characteristic index set, which provides an input feature space with clear physical meaning and type distinguishing ability for data-driven feature extraction and recognition model training in step S600.

[0057] It should be noted that the construction of physical characteristic indexes is based on reproducible evidence, and the threshold values of each index should be determined by distribution test and discrimination analysis of historical samples, and should be adapted in different seasons and regions. When the isentropic surface tracking or MPV calculation is limited by the resolution of the data layer, mixed layer thickness correction and filtering smoothing can be used to reduce numerical noise. The type classification (J type, S type) is a diagnostic classification, and the boundary samples can be labeled as mixed type and participate in the model training in the form of soft label to reduce the classification bias caused by hard threshold. In addition, the water vapor flux divergence, box water vapor budget, whole-layer dry intrusion index, dynamic monsoon water vapor surge index and water vapor transport type classification given in step S500 are the preferred embodiments, and other dynamic-thermal diagnostic quantities such as vertical wind shear, convective available potential energy and potential vorticity flux can also be introduced according to specific research needs, or the existing indexes can be combined, normalized and dimensionally reduced to construct a feature index set more suitable for specific regions or specific business objectives.

[0058] S600. Data-driven feature extraction and recognition model training:

[0059] The training and verification sample set is constructed based on the unified gridded sample data and the physical characteristic index set, a preset data-driven learning algorithm is used for model training, cross-validation and parameter optimization on the training samples, and a target recognition model for inland TRP feature recognition and intensity evaluation is obtained. As an optimization, the data-driven learning algorithm at least includes an integrated tree model, a gradient boosting decision tree model and / or a deep neural network model, and further includes: performing class imbalance processing on TRP and non-TRP samples, different rainstorm intensity level samples, using sample resampling, class weight adjustment or synthetic minority over-sampling method to improve the weight of minority class samples in training; using K-fold cross-validation strategy divided by year or typhoon instance to test the stability of the model, and selecting the optimal model structure and hyperparameter group by considering the accuracy, recall rate, F1 score and area under the receiver operating characteristic curve; and identifying the physical characteristics that contribute most to TRP feature recognition and intensity evaluation through feature importance ranking or sensitivity analysis, providing a basis for the physical interpretability of the model results.

[0060] It should be noted that the data-driven learning algorithm used in step S600 is not limited to the integrated tree model, the gradient boosting decision tree model or the deep neural network model, but can also use support vector machines, graph neural networks, integrated learning frameworks or multi-model weighted fusion algorithms; the class imbalance processing method, the cross-validation strategy and the evaluation index system can also be adjusted according to the sample size and the business emphasis.

[0061] In addition, in order to improve the business application value and generalization ability of the target recognition model, the present application adopts a double verification strategy based on typhoon activity year and geographical zoning during model training. Specifically, the historical TRP samples are divided into training period and independent verification period according to the occurrence year, ensuring that the typhoon instances in the verification period are completely independent of the training samples, avoiding the overfitting risk caused by the cross-occurrence of the same typhoon samples in the training set and the verification set. At the same time, for inland TRPs in different geographical zones such as North China, Northeast China and the middle and lower reaches of the Yangtze River, the recognition accuracy and intensity evaluation bias of the model in each zone are evaluated, and for the case where the recognition performance of a certain zone is significantly low, the model is optimized by increasing the weight of representative samples in the zone or introducing zone-specific local terrain and climate background characteristic variables, ensuring the universality and stability of the model for different types and regions of TRP events.

[0062] S700. Rainstorm risk assessment and result output:

[0063] Based on the numerical prediction field, reanalysis data and / or real-time observation data of the target prediction time, the input feature set is generated which is isomorphic with the training stage, and is input into the target identification model to obtain the TRP occurrence probability and estimated precipitation intensity level of each grid or station in the inland area of the target. According to the preset risk classification threshold, the risk of the typhoon rainstorm in the inland area is evaluated. As an optimization, the rainstorm risk assessment and result output include: probability calibration is performed on the TRP occurrence probability output by the target identification model, and the consistency of the probability prediction is evaluated and corrected by using the reliability curve and Brier score index; a two-dimensional risk classification matrix is constructed according to the occurrence probability interval and the estimated precipitation intensity level, and the risk level is divided into low risk, medium risk, high risk and extremely high risk levels, and the connectivity analysis is performed on the adjacent high-risk grids in space to generate a continuous high-risk area patch; when the result is output, in addition to providing a spatial risk classification map, the key physical feature diagnostic field and type information of the corresponding TRP event are also output, which facilitates the business personnel to review and decision support for the high-risk area from the physical mechanism angle.

[0064] It should be noted that when the rainstorm risk assessment result is output, not only a static spatial risk classification map is provided, but also a dynamic time evolution sequence and uncertainty representation information are generated. For a prediction period of 24-72 hours in the future, the TRP occurrence probability and risk level evolution of each time are output at 6-hour or 12-hour intervals to form a spatiotemporal continuous risk warning product. At the same time, through the ensemble prediction or model integration method, the uncertainty range of the prediction result is quantified, and for the area where the TRP occurrence probability is close to the critical point of the risk classification threshold, the probability interval and the possible risk of grade jump are clearly marked, providing a more comprehensive reference basis for business decision-making. In addition, the real-time diagnostic value of the key physical characteristics and its comparison with the historical statistical threshold are also attached in the output result, which is helpful for quickly understanding the physical cause and abnormal degree of the current TRP event, and realizing the organic combination of quantitative evaluation and physical explanation.

[0065] Embodiment 2: Application example

[0066] On the basis of the above embodiment 1, this embodiment 2 takes the TRP process of extreme rainstorm in a certain province of North China-Huanghe River Basin caused by the remote influence of typhoon 'Yanyan' No. 6 in June 2021 as an example to illustrate the application of the method in actual business. In step S100, 20 ° N-45 ° N, 105 ° E-125 ° E are selected as the target inland and its upstream and downstream regions, covering North China-Huanghe River Basin and the middle and lower reaches of the Yangtze River in the meridian direction, and the period is from July 17 to 31, 2021; the hourly or daily surface precipitation observation, typhoon best path data and ERA5 reanalysis data of the region are obtained, and the time alignment, quality control and uniform gridding interpolation are completed according to the method of embodiment 1. In step S200, the typhoon path is taken as the constraint to build the space-time matching record of typhoon-inland precipitation in the above period, and to determine the typhoon process which has a significant remote precipitation influence on the province and the surrounding inland area from July 19 to 21.

[0067] In step S300, the grid precipitation field is first subjected to rainband separation and typhoon body rainband identification, and then subjected to constraint screening of potential TRP candidate rainbands such as typhoon peripheral circulation radius D1, maximum influence radius D0 and safety belt width ΔD, and the use of whole layer water vapor transport flux IVT field to track the dominant strong water vapor transport belt from the typhoon to North China-Huanghe River Basin to form a strong IVT water vapor transport belt mask. The spatial superposition of the potential TRP rainband and the strong transport belt is carried out, and only the rainband located in the north quadrant of the typhoon, continuously overlapped with the transport belt and spatially spaced from the typhoon body precipitation is reserved as the TRP rainband; then the extreme rainstorm event is marked according to the daily precipitation threshold, and sample units with TRP attribute and intensity level label are formed.

[0068] In steps S400 and S500, taking the extreme rainstorm as a representative example, a multi-nested regional numerical model test is constructed, which covers the large-scale circulation environment in the outer layer and the province and the TRP landing area in the inner layer, and carries out control integration by using ERA5 as the initial boundary value field. After the simulation results can better reproduce the typhoon path, intensity and North China-Huanghe River Basin remote rainstorm belt distribution, the whole layer water vapor flux and divergence, box water vapor budget, dry intrusion index and dynamic monsoon water vapor surge index are further diagnosed based on the model output and ERA5 data, revealing the long-distance S-shaped water vapor transport channel formed by the coupling of strong southwest monsoon and typhoon during the typhoon period, and the three-dimensional circulation structure under the configuration of upstream dry cold air and upper-level jet, and accordingly the construction and extraction of physical characteristic indexes of this embodiment are completed.

[0069] In steps S600 and S700, the identified TRP data set of China in 1979-2020 is combined with the above-mentioned set of physical feature indicators, used to train a target recognition model based on an integrated tree or gradient boosting, and the TRP process in 2021 is taken as an independent test case. After the input features of the forecast field and the reanalysis field around 0000 UTC on July 20, 2021 are constructed and input, the target model is input, the area where the TRP occurrence probability of the province and its surrounding inland area significantly increases (such as greater than 0.7) can be obtained, and the corresponding heavy rain or heavy rain intensity grade is given, the high-risk patch spatial distribution obtained has good consistency with the actual result, indicating that the method of the present application can realize effective identification and risk assessment of inland TRP extreme rainstorm events in actual business environment.

[0070] Through the above embodiments, the purposes of the present application are completely and effectively achieved. Those skilled in the art can understand that the present application includes but is not limited to the contents described in the drawings and the above specific embodiments. Although the present application has been described in relation to the presently preferred embodiments thereof, it is to be understood that the application is not limited to the disclosed embodiments, and any modification not deviating from the functional and structural principles of the present application will be included in the scope of the claims.

Claims

1. A method for TRP feature identification and evaluation in inland areas based on data-driven and numerical simulation, characterized in that, At least the following steps are included: S100. Acquire surface precipitation observation data, tropical cyclone track data, reanalysis data and / or numerical forecast data, and preprocess the multi-source data according to the target inland area and target time period; S200. Identify individual typhoon cases that have a potential impact on the target inland area within the target time period, and establish a spatiotemporal matching record of typhoon-inland precipitation based on the spatial relationship between the typhoon center location and the inland area. S300. Based on the preset TRP identification rules, samples of precipitation fields in each typhoon process are screened and labeled to construct an inland TRP sample library containing TRP, non-TRP and different precipitation intensity level labels, and form a unified gridded sample dataset. S400. Select several representative inland TRP cases and / or typical circulation patterns from the sample library, set the integration region, initial boundary value field and physical parameterization scheme based on the regional numerical model, drive the model to perform numerical integration, and obtain a high-resolution numerical model simulation output dataset. S500. Using model output and reanalysis data, physical diagnosis is performed on the evolution of typhoon location and intensity, major water vapor transport channels and their flux divergence, dry and cold air intrusion activities, monsoon water vapor surge intensity, and mid-to-high latitude trough and ridge configuration, and a set of physical characteristic fields and characteristic indicators related to inland TRP is constructed. S600. Based on the unified gridded sample data and physical feature index set, a training and validation sample set is constructed. The training sample is trained, cross-validated and optimized using a preset data-driven learning algorithm to obtain a target recognition model for TRP feature recognition and intensity assessment in inland areas. S700. Based on the numerical forecast field, reanalysis data and / or real-time observation data of the target forecast time, generate an input feature set isomorphic to the training phase and input it into the target recognition model to obtain the probability of TRP occurrence and the estimated precipitation intensity level in the inland area of ​​the target.

2. The method according to claim 1, characterized in that, In step S100, the surface precipitation observation data includes hourly or daily precipitation data from meteorological stations within at least 500 kilometers of the target inland area; the tropical cyclone track data includes at least 6-hour interval records of typhoon number, center location latitude and longitude, central pressure, maximum wind speed, and direction of movement; the reanalysis data includes at least three-dimensional gridded data of temperature field, geopotential height field, wind field, and specific humidity field from the ERA5 or NCEP reanalysis dataset, with a temporal resolution of not less than 6 hours and a vertical layer coverage of 1000-100 hPa; and the numerical forecast data is forecast product data based on global numerical forecast models or regional numerical models.

3. The method according to claim 1 or 2, characterized in that, Step S100 also includes: correcting station height, filling missing values ​​and removing outliers for surface precipitation observation data, and using timestamp alignment to uniformly resample hourly or daily precipitation observation data with reanalysis data and numerical forecast data; using bilinear interpolation or nearest neighbor interpolation to unify multi-source data with different spatial resolutions onto a preset regular latitude and longitude grid, and using buffer expansion processing for edge grid points.

4. The method according to claim 1, characterized in that, In step S200, typhoon process matching and inland area screening include: based on tropical cyclone path data, setting the typhoon formation area, minimum central pressure or maximum wind speed, and eliminating tropical depressions, extratropical cyclones, and cases with extremely weak impact on the target inland area; in the time dimension, constructing a matching time window by extending a preset number of hours forward and backward, with the moment when the typhoon is closest to the target inland area as the center; in the spatial dimension, screening is performed based on the minimum spherical distance between the typhoon center and the centroid or boundary of the target inland area, retaining only typhoon processes within the preset distance range and with significant precipitation response in the inland area within the matching time window.

5. The method according to claim 1, characterized in that, In step S300, the objective identification of TRP includes at least the following sub-steps during implementation: S301. The precipitation field is divided into several independent precipitation rainband units using the connected component clustering method. The typhoon body precipitation rainband is identified based on the high-intensity precipitation connected components within the area where the typhoon center is located and its maximum wind speed radius is located. The body rainband and its adjacent grid points are removed from the subsequent TRP determination. S302. For each non-body precipitation rainband, calculate the spherical distance Dts between each precipitation grid point in the rainband and the typhoon center, and determine the corresponding outer circulation radius D1 of the typhoon based on the current intensity of the typhoon. When Dts satisfies D1+ΔD≤Dts≤D0, mark the rainband as a potential TRP candidate rainband, where ΔD is the preset safe bandwidth and D0 is the preset maximum influence radius. S303. On the whole layer water vapor transport flux field corresponding to the typhoon time, the dominant water vapor transport path is traced along the direction of maximum IVT vector starting from the typhoon center. Several connection areas with preset latitude and longitude ranges are constructed around each path point. In each connection area, grid point sets with IVT greater than the preset strong water vapor transport judgment threshold IVT0 are extracted to form a strong water vapor transport belt mask extending outward from the typhoon center. S304. Potential TRP candidate rainbands are spatially overlaid with a strong water vapor transport belt mask on a unified grid. Only when the candidate rainband is located entirely in the quadrant north of the typhoon center, has continuous overlap with the strong water vapor transport belt, and has at least one low-precipitation or no-precipitation grid point as a spatial interval between the candidate rainband and the typhoon body rainband, is the candidate rainband marked as a long-distance precipitation rainband that meets the TRP conditions. Based on this, the grid points or stations contained in the rainband are labeled as TRP samples and included in the inland TRP sample library.

6. The method according to claim 1 or 5, characterized in that, In step S300, sample labeling further includes: classifying TRP events into intensity levels based on daily precipitation and cumulative precipitation thresholds. The intensity levels include at least moderate to heavy rain, rainstorm, heavy rainstorm, and extremely heavy rainstorm. The daily precipitation for moderate to heavy rain is 10-50 mm, for rainstorm is 50-100 mm, for heavy rainstorm is 100-250 mm, and for extremely heavy rainstorm is not less than 250 mm. Spatially, connectivity clustering is performed on regions exceeding the same intensity threshold in continuous grid points or adjacent stations, treating them as spatial units of the same TRP event. Furthermore, adjacent time-series events are merged in the temporal dimension, ultimately forming a sample labeling system that simultaneously possesses TRP attributes and rainstorm intensity level attributes.

7. The method according to claim 1, characterized in that, In step S400, a multi-nested regional numerical weather prediction model is adopted, with its horizontal resolution progressively increasing from the outermost layer to the innermost layer. The innermost region covers the target inland TRP landing area and key water vapor transport channels. When setting the initial boundary value field, the typhoon vortex structure initialization and repositioning technology is introduced to assimilate and correct the typhoon position, intensity, and radial wind field in the large-scale field. In the selection of physical parameterization schemes, a combination of schemes suitable for heavy precipitation simulation is adopted for radiation, shallow and deep convection, cloud microphysics, and planetary boundary layer processes. During the integration process, high temporal resolution three-dimensional dynamic, thermodynamic, and water vapor diagnostic variables are output.

8. The method according to claim 1 or 7, characterized in that, Step S400 also includes a physical consistency check of model results: the typhoon track, central pressure evolution, maximum wind speed, and TRP landing area simulated in numerical simulation are compared with the tropical cyclone track data, reanalysis data, and / or precipitation observations at the corresponding time, and the positional deviation, intensity deviation, and / or precipitation distribution deviation are calculated; when the error index exceeds the preset threshold, the model parameterization scheme or initial field is adjusted and re-integrated until the simulated typhoon structure, water vapor transport pattern, and TRP spatial distribution reach the preset consistency with the observations.

9. The method according to claim 1, characterized in that, In step S500, the physical feature diagnosis and feature index construction shall include at least the following sub-steps during implementation: S501. Based on the model output and the corresponding time-time reanalysis data, calculate the whole-layer water vapor flux and water vapor flux divergence of the target inland TRP landing area and its upstream region, decompose the flux divergence into wind field convergence term and humidity advection term; and construct a path profile along the dominant water vapor transport path from the typhoon center to the TRP landing area, and extract the path integral IVT intensity, path length, maximum IVT and its occurrence location. S502. Construct a fixed or time-updated assessment area in and around the TRP landing area, perform area integration and boundary line integration on the water vapor flux within the assessment area, calculate the water vapor input and output fluxes at each side boundary, as well as the rate of change of the whole layer water vapor content within the area, and obtain the water vapor budget balance relationship by combining the observed or model-simulated precipitation flux, thereby constructing the net water vapor input flux, precipitation efficiency, and vertically integrated wet layer thickness index. S503. In the mid-to-high latitude dry and cold air source area and intrusion channel upstream of TRP, based on temperature, specific humidity, potential vorticity, and vertical velocity field, identify dry intrusion air masses with high potential vorticity, low specific humidity, and significant sinking motion. By performing vertical integration and time integration along isobaric or isentropic surfaces on the dry intrusion signal, construct the isobaric whole-layer dry intrusion index and / or isentropic whole-layer dry intrusion index, and extract the maximum value, time delay, and spatial overlap area characteristics of the dry intrusion index within a certain time window before and after the TRP rainstorm. S504. In the monsoon water vapor transport area on the south or southwest side of the typhoon, the monsoon water vapor surge assessment area is dynamically defined based on the typhoon center latitude and longitude, the radius of the 7-level wind circle and the central air pressure. The dynamic monsoon water vapor surge index is constructed by vertically integrating the wet-weighted meridional wind field within the assessment area and extracting the increase, peak value and duration of the index before and after the formation of TRP.

10. The method according to claim 1, characterized in that, Step S500 also includes classifying the water vapor transport patterns of TRP events: by calculating the geometric relationship and azimuth distribution of the typhoon center, the centroid of the target inland TRP landing area, and the dominant water vapor transport path, the water vapor transport patterns of TRP events are classified into J-type transport patterns characterized by blocky water vapor convergence on the northwest side of the typhoon and S-type transport patterns characterized by long-distance S-shaped water vapor transport belts. The typical values ​​and combination characteristics of whole-layer water vapor flux divergence, dry intrusion index, dynamic monsoon water vapor surge index, and precipitation efficiency under different transport patterns are statistically analyzed. The transport pattern category labels and multi-source diagnostic indicators are combined to form a set of physical characteristic indicators.

11. The method according to claim 1, characterized in that, In step S600, the data-driven learning algorithm includes an ensemble tree model, a gradient boosting decision tree model, and / or a deep neural network model, and further includes: class imbalance processing of TRP and non-TRP samples, and samples of different rainstorm intensity levels; using sample resampling, class weight adjustment, or synthetic minority class oversampling methods to increase the weight of minority class samples in training; using a K-fold cross-validation strategy divided by year or typhoon cases to test the stability of the model; comprehensively considering evaluation indicators such as accuracy, recall, F1 score, and area under the receiver operating characteristic curve to select the optimal model structure and hyperparameter set; and identifying the physical features that contribute most to TRP feature recognition and intensity assessment through feature importance ranking or sensitivity analysis.

12. The method according to claim 1, characterized in that, In step S700, the rainstorm risk assessment and result output includes: performing probability calibration on the TRP occurrence probability output by the target identification model, evaluating and correcting the consistency of the probability forecast using reliability curves and Brier score indicators; constructing a two-dimensional risk classification matrix based on the occurrence probability interval and the estimated precipitation intensity level, classifying the risk level into low risk, medium risk, high risk, and extremely high risk levels, and performing connectivity analysis on spatially adjacent high-risk grid points to generate continuous high-risk area patches; in the result output, in addition to providing a spatial risk classification map, it also outputs the key physical characteristic diagnostic field and type information of the corresponding TRP event.

13. A computer program product, characterized in that, Includes computer instructions for performing the steps of the method according to any one of claims 1 to 12.

14. A computer-readable storage medium storing a computer program thereon, characterized in that, When a computer program is executed by a processor, it implements the steps of the method described in any one of claims 1 to 12.

Images