A historical lightning frequency reconstruction method, system, device, and medium

Abstract

The application provides a historical lightning frequency reconstruction method, system, device and medium, the method comprises the following steps: acquiring lightning frequency observation data, reanalysis meteorological data and a set of prediction factors in the reanalysis meteorological data; the target area is divided into a plurality of modeling units; a mapping relationship between the set of prediction factors and the lightning frequency observation data is constructed to obtain a lightning frequency statistical model of each modeling unit; the set of prediction factors in the historical reconstruction period is input into each lightning frequency statistical model to obtain the initial value of the historical lightning frequency of the target area and perform quantile mapping bias correction and spatial consistency correction to obtain the historical lightning frequency reconstruction value; the application is based on the lightning and meteorological data regional regression modeling of space-time matching, and the nearly 100-year historical lightning frequency data blank is inversely filled; after quantile mapping and spatial consistency correction, high-resolution gridded data is output to directly serve engineering applications such as transmission line lightning risk assessment.

Description

Technical Field

[0001] This invention belongs to the field of meteorological data analysis and lightning climate reconstruction technology, specifically involving a method, system, equipment and medium for reconstructing historical lightning frequency. Background Technology

[0002] Lightning activity is a significant manifestation of severe convective weather, posing a serious threat to the safe operation of power systems. Statistics show that lightning-induced outages account for over 50% of all outages on transmission lines, with the outage rate significantly increasing, especially during the summer months when severe convective weather is frequent. Accurately obtaining long-term, spatially continuous lightning frequency distribution data is fundamental for conducting differentiated lightning protection design for transmission lines, upgrading aging lines, and zoning lightning risk zones.

[0003] Currently, lightning frequency data mainly comes from two types of observation systems. The first type is ground-based lightning location systems, such as the Wide Area Lightning-Ground Monitoring Network (WASM), which can reliably record the time, location, and intensity of lightning strikes. The second type is spaceborne lightning detection instruments, such as NASA's LIS and OTD instruments, which can provide information on lightning climate distribution in tropical and mid-latitude regions worldwide. However, the time coverage of these systems is limited to the last two decades (approximately after 2000), and cannot meet the research needs for lightning climate conditions in more distant periods (such as the 1950s to the early 21st century).

[0004] On the other hand, reanalysis meteorological data can provide continuous, globally covered, multi-height meteorological field data since 1950. For example, the ERA5 reanalysis meteorological data released by the European Centre for Medium-Range Weather Forecasts (ECMWF) has a temporal resolution of up to 1 hour and a spatial resolution of 0.25° × 0.25°, including physical variables closely related to lightning formation such as convective effective potential energy, K-index, relative humidity, vertical velocity, and 0°C altitude. However, reanalysis meteorological data itself does not directly output lightning frequency. Lightning occurrence is the result of the combined effects of thermodynamic, dynamic, and microphysical processes, requiring inversion through indirect modeling methods.

[0005] Existing lightning retrieval methods suffer from three main shortcomings: First, statistical methods based on the number of thunderstorm days at ground-based meteorological stations have extremely uneven spatial coverage; second, lightning identification algorithms based on radar reflectivity or satellite cloud top brightness temperature have late data start times; and third, lightning parameterization schemes embedded in climate models have significant differences in applicability across different regions and cannot be directly applied to the retrieval of historical reanalysis meteorological data. Therefore, there is an urgent need to develop a historical lightning frequency reconstruction method that can establish a transferable statistical model using finite-year lightning frequency observation data and long-term reanalysis meteorological data. Summary of the Invention

[0006] To overcome the shortcomings of the prior art, this invention proposes a method for reconstructing historical lightning frequency, comprising: Acquire spatiotemporally matched lightning frequency observation data and reanalysis meteorological data within the target area, and extract a set of prediction factors from the reanalysis meteorological data; Based on the geographical features of the target area and the amount of lightning frequency observation data, the target area is divided into multiple modeling units; For each modeling unit, a regression analysis method is used to construct the mapping relationship between the predictor set and the lightning frequency observation data, so as to obtain the lightning frequency statistical model of each modeling unit; The set of predictive factors for the target area during the historical reconstruction period is input into the lightning frequency statistical model of each modeling unit to obtain the initial value of the historical lightning frequency of the target area. The initial historical lightning frequency values ​​are corrected for quantile mapping deviation and spatial consistency to obtain the reconstructed historical lightning frequency values ​​for the target area.

[0007] Preferably, based on the geographical features of the target area and the amount of lightning frequency observation data, the target area is divided into multiple modeling units, including: The target area is gridded. When the amount of lightning frequency observation data in a single grid is not less than a preset sample size threshold, the target area is divided into multiple modeling units based on the grid. When the amount of lightning frequency observation data in a single grid is lower than the preset sample size threshold, or when the target area has preset geographical features, multiple grids are merged into one partition, and the target area is divided into multiple modeling units based on the partition; wherein, the total amount of lightning frequency observation data in multiple grids within a partition is not lower than the preset sample size threshold, or multiple grids within a partition have the same preset geographical features.

[0008] Preferably, the preset geographical features include: Terrain stratification features based on altitude; where grids with altitudes within the same preset stratification range are considered to have the same terrain stratification features; Land cover characteristics based on underlying surface type; wherein, grids with the same underlying surface type are considered to have the same land cover characteristics; Climate zoning characteristics based on lightning activity features; where grids belonging to the same climate type and lightning zoning are considered to have the same climate zoning characteristics. Functional zoning features based on the characteristics of power transmission corridors; where grids located in the same geographical environment type of power transmission corridor are considered to have the same functional zoning features; Hybrid characteristics based on data availability; where grids with lightning monitoring station densities within the same preset range are considered to have the same hybrid characteristics.

[0009] Preferably, for each modeling unit, a regression analysis method is used to construct a mapping relationship between the predictor set and the lightning frequency observation data, thereby obtaining a lightning frequency statistical model for each modeling unit, including: For each modeling unit, regression analysis is performed based on a multiple linear regression model to construct the mapping relationship between the predictor set in each modeling unit and the lightning frequency observation data. The least squares method is used to solve the regression coefficients, and inequality constraints are applied to the regression coefficients according to the lightning physical mechanism to obtain the baseline model of each modeling unit. When the goodness of fit of the benchmark model is not lower than the preset goodness threshold, the benchmark model is used as the lightning frequency statistical model. When the goodness of fit of the baseline model is lower than the preset goodness threshold, regression analysis is performed based on machine learning regression algorithm to construct the mapping relationship between the prediction factor set in each modeling unit and the lightning frequency observation data, so as to obtain the enhanced model of each modeling unit, and the enhanced model is used as the lightning frequency statistical model.

[0010] Preferably, the inequality constraints include: In the predictor set, the regression coefficients of variables related to thermal instability are greater than or equal to zero, and the variables related to thermal instability include convective available potential energy and the K exponent. In the set of predictor factors, the regression coefficients of variables related to upward motion are less than or equal to zero, and the variables related to upward motion include the 500 hPa vertical velocity.

[0011] Preferably, the quantile mapping deviation correction includes: The set of predictive factors that are spatiotemporally matched with the lightning frequency observation data is input into the lightning frequency statistical model of each modeling unit to obtain lightning frequency simulation data. When the mean deviation between the simulated lightning frequency data and the observed lightning frequency data within a modeling unit significantly deviates from zero, the simulated cumulative distribution function and the observed cumulative distribution function are calculated respectively based on the simulated lightning frequency data and the observed lightning frequency data of that modeling unit. Calculate the quantile position of the initial historical lightning frequency of the modeling unit in the simulated cumulative distribution function, and use the value of the observed cumulative distribution function corresponding to the quantile position as the initial calibration value of the historical lightning frequency of the modeling unit after quantile mapping deviation correction.

[0012] Preferably, the spatial consistency correction includes: Spatial continuity checks are performed on the initial historical lightning frequency values ​​of the target area to identify aberration grids with abnormal abrupt changes; The initial historical lightning frequency values ​​in the mutation grid are processed using spatial weighted smoothing to obtain the reconstructed historical lightning frequency values; wherein, the weights in the spatial weighted smoothing process are determined according to the geographical characteristics of the mutation grid.

[0013] Preferably, determining the weights in the spatially weighted smoothing process based on the geographical characteristics of the mutation grid includes: When the mutation grid is located in a region with flat terrain or uniform underlying surface, the weights in the spatial weighted smoothing process are determined based on the spatial distance between the mutation grid and adjacent grids. When the mutation grid is located in a region with complex terrain, terrain feature similarity constraints are superimposed on the weights using digital elevation model data, so that the weights decrease as the spatial distance and differences in terrain features increase. When the mutation grid is located in the coastal-land boundary region, the weights are determined independently for the land region and the marine region, with the coastline as the boundary.

[0014] Based on the same inventive concept, the present invention also provides a historical lightning frequency reconstruction system, comprising: The data extraction module is used to acquire spatiotemporally matched lightning frequency observation data and reanalysis meteorological data within the target area, and to extract a set of prediction factors from the reanalysis meteorological data. The partitioning module is used to divide the target area into multiple modeling units based on the geographical features of the target area and the amount of lightning frequency observation data; The modeling module is used to construct the mapping relationship between the predictor set and the lightning frequency observation data for each modeling unit by using regression analysis, so as to obtain the lightning frequency statistical model for each modeling unit. The inversion module is used to input the set of prediction factors for the target area during the historical reconstruction period into the lightning frequency statistical model of each modeling unit to obtain the initial value of the historical lightning frequency of the target area. The correction module is used to perform quantile mapping deviation correction and spatial consistency correction on the initial value of the historical lightning frequency to obtain the reconstructed value of the historical lightning frequency in the target area.

[0015] Preferably, the partitioning module is specifically used for: The target area is gridded. When the amount of lightning frequency observation data in a single grid is not less than a preset sample size threshold, the target area is divided into multiple modeling units based on the grid. When the amount of lightning frequency observation data in a single grid is lower than the preset sample size threshold, or when the target area has preset geographical features, multiple grids are merged into one partition, and the target area is divided into multiple modeling units based on the partition; wherein, the total amount of lightning frequency observation data in multiple grids within a partition is not lower than the preset sample size threshold, or multiple grids within a partition have the same preset geographical features.

[0016] Preferably, the preset geographical features include: Terrain stratification features based on altitude; where grids with altitudes within the same preset stratification range are considered to have the same terrain stratification features; Land cover characteristics based on underlying surface type; wherein, grids with the same underlying surface type are considered to have the same land cover characteristics; Climate zoning characteristics based on lightning activity features; where grids belonging to the same climate type and lightning zoning are considered to have the same climate zoning characteristics. Functional zoning features based on the characteristics of power transmission corridors; where grids located in the same geographical environment type of power transmission corridor are considered to have the same functional zoning features; Hybrid characteristics based on data availability; where grids with lightning monitoring station densities within the same preset range are considered to have the same hybrid characteristics.

[0017] Preferably, the modeling module is specifically used for: For each modeling unit, regression analysis is performed based on a multiple linear regression model to construct the mapping relationship between the predictor set in each modeling unit and the lightning frequency observation data. The least squares method is used to solve the regression coefficients, and inequality constraints are applied to the regression coefficients according to the lightning physical mechanism to obtain the baseline model of each modeling unit. When the goodness of fit of the benchmark model is not lower than the preset goodness threshold, the benchmark model is used as the lightning frequency statistical model. When the goodness of fit of the baseline model is lower than the preset goodness threshold, regression analysis is performed based on machine learning regression algorithm to construct the mapping relationship between the prediction factor set in each modeling unit and the lightning frequency observation data, so as to obtain the enhanced model of each modeling unit, and the enhanced model is used as the lightning frequency statistical model.

[0018] Preferably, the inequality constraints include: In the predictor set, the regression coefficients of variables related to thermal instability are greater than or equal to zero, and the variables related to thermal instability include convective available potential energy and the K exponent. In the set of predictor factors, the regression coefficients of variables related to upward motion are less than or equal to zero, and the variables related to upward motion include the 500 hPa vertical velocity.

[0019] Preferably, the correction module includes a quantile mapping deviation correction unit, used for: The set of predictive factors that are spatiotemporally matched with the lightning frequency observation data is input into the lightning frequency statistical model of each modeling unit to obtain lightning frequency simulation data. When the mean deviation between the simulated lightning frequency data and the observed lightning frequency data within a modeling unit significantly deviates from zero, the simulated cumulative distribution function and the observed cumulative distribution function are calculated respectively based on the simulated lightning frequency data and the observed lightning frequency data of that modeling unit. Calculate the quantile position of the initial historical lightning frequency of the modeling unit in the simulated cumulative distribution function, and use the value of the observed cumulative distribution function corresponding to the quantile position as the initial calibration value of the historical lightning frequency of the modeling unit after quantile mapping deviation correction.

[0020] Preferably, the correction module further includes a spatial consistency correction unit, the spatial consistency correction unit comprising: An anomaly identification subunit is used to perform spatial continuity checks on the initial calibration values ​​of historical lightning frequencies in the target area and identify anomalous grids with anomalous changes. The smoothing subunit is used to perform spatial weighted smoothing on the initial calibration values ​​of historical lightning frequencies in the mutation grid to obtain the reconstructed values ​​of historical lightning frequencies. The weight determination sub-unit is used to determine the weights in the spatial weighted smoothing process based on the geographical characteristics of the mutation grid.

[0021] Preferably, the weight determination subunit is specifically used for: When the mutation grid is located in a region with flat terrain or uniform underlying surface, the weights in the spatial weighted smoothing process are determined based on the spatial distance between the mutation grid and adjacent grids. When the mutation grid is located in a region with complex terrain, terrain feature similarity constraints are superimposed on the weights using digital elevation model data, so that the weights decrease as the spatial distance and differences in terrain features increase. When the mutation grid is located in the coastal-land boundary region, the weights are determined independently for the land region and the marine region, with the coastline as the boundary.

[0022] Based on the same inventive concept, the present invention also provides a computer device, comprising: one or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, a historical lightning frequency reconstruction method as described above is implemented.

[0023] Based on the same inventive concept, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed, it implements the historical lightning frequency reconstruction method as described above.

[0024] Compared with the closest existing technology, the present invention has the following beneficial effects: This invention provides a method, system, device, and medium for reconstructing historical lightning frequency. The method includes: acquiring spatiotemporally matched lightning frequency observation data and reanalysis meteorological data within a target area, and extracting a predictive factor set from the reanalysis meteorological data; dividing the target area into multiple modeling units based on its geographical features and the amount of lightning frequency observation data; for each modeling unit, using regression analysis to construct a mapping relationship between the predictive factor set and the lightning frequency observation data, obtaining a lightning frequency statistical model for each modeling unit; inputting the predictive factor set of the target area during the historical reconstruction period into the lightning frequency statistical model of each modeling unit to obtain an initial historical lightning frequency value for the target area; and performing quantile mapping deviation correction and spatial consistency correction on the initial historical lightning frequency value to obtain a reconstructed historical lightning frequency value for the target area. This invention achieves this by acquiring spatiotemporally matched lightning observation and reanalysis meteorological data. This method analyzes meteorological data and extracts a set of predictive factors. Based on geographical features and the amount of observational data, the target area is divided into multiple modeling units. Regression analysis is used to construct the mapping relationship between predictive variable factors and lightning frequency, achieving refined modeling applicable to different geographical regions. The predictive factor set from the historical reconstruction period is input into the models of each modeling unit, extending the effective lightning frequency data length from nearly 20 years to nearly 100 years, filling the gap in long-term historical lightning frequency data. Finally, quantile mapping bias correction is used to eliminate the systematic bias caused by the evolution of the meteorological data assimilation system during reanalysis. Combined with spatial consistency correction, the spatial rationality of the data is optimized, significantly improving the reconstruction accuracy. The overall method forms a complete technical process from data processing and modeling to correction output, outputting spatially continuous, high-resolution, regionally gridded lightning frequency reconstruction data, which can directly serve engineering applications such as transmission line lightning strike risk assessment and differentiated lightning protection design, and has significant practical value. Attached Figure Description

[0025] Figure 1 A schematic diagram of a historical lightning frequency reconstruction method provided by the present invention; Figure 2 The heatmap of the correlation coefficient matrix between variable factors provided by this invention; Figure 3A scatter plot of observed lightning frequency data and simulated lightning frequency data during the verification period of the lightning frequency statistical model provided by this invention; Figure 4 A schematic diagram of a historical lightning frequency reconstruction system provided by the present invention; Figure 5 This is a schematic diagram of an electronic device structure provided by the present invention. Detailed Implementation

[0026] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0027] Example 1: This invention provides a method for reconstructing historical lightning frequencies, such as... Figure 1 As shown, it includes: S1. Obtain spatiotemporally matched lightning frequency observation data and reanalysis meteorological data within the target area, and extract a set of prediction factors from the reanalysis meteorological data; S2. Based on the geographical features of the target area and the amount of lightning frequency observation data, the target area is divided into multiple modeling units; S3. For each modeling unit, a regression analysis method is used to construct the mapping relationship between the set of predictive factors and the lightning frequency observation data, so as to obtain the lightning frequency statistical model of each modeling unit. S4. Input the set of prediction factors for the target area during the historical reconstruction period into the lightning frequency statistical model of each modeling unit to obtain the initial value of the historical lightning frequency of the target area. S5. Perform quantile mapping deviation correction and spatial consistency correction on the initial value of the historical lightning frequency to obtain the reconstructed value of the historical lightning frequency in the target area.

[0028] Considering that existing lightning inversion methods suffer from discontinuous spatial coverage, late data start times, and significant differences in applicability across different regions, and cannot be directly applied to the inversion of historical reanalysis meteorological data, this invention acquires spatiotemporally matched lightning observation and reanalysis meteorological data and extracts a set of predictive factors. Based on geographical features and the amount of observational data, the target area is divided into multiple modeling units. Regression analysis is used to construct a mapping relationship between predictive variable factors and lightning frequency, achieving refined modeling applicable to different geographical regions. The predictive factor set from the historical reconstruction period is input into the models of each modeling unit, extending the effective lightning frequency data length from nearly twenty years to nearly a century, filling the gap in long-term historical lightning frequency data. Finally, quantile mapping bias correction eliminates systemic bias caused by the evolution of the reanalysis meteorological data assimilation system, and spatial consistency correction optimizes the spatial rationality of the data, significantly improving reconstruction accuracy. The overall method forms a complete technical process from data processing and modeling to correction output, outputting spatially continuous, high-resolution, regionally gridded lightning frequency reconstruction data, which can directly serve engineering applications such as transmission line lightning strike risk assessment and differentiated lightning protection design, and has significant practical value. The following provides a detailed explanation of each step.

[0029] Given that lightning-induced outages account for an extremely high proportion of all outages in transmission lines, and that existing lightning observation systems only cover the last two decades, they cannot meet the needs of lightning climate state research from the last century to the beginning of this century. In S1 above, acquiring spatiotemporally matched lightning frequency observation data and reanalysis meteorological data within the target area can achieve precise alignment of multi-source data on spatial and temporal scales. This provides a physically meaningful foundational dataset for subsequently constructing high-precision statistical models, overcoming the shortcomings of traditional daily thunderstorm statistical methods, such as low spatial resolution and incomplete coverage.

[0030] In this embodiment, in S1, spatiotemporally matched lightning frequency observation data and reanalysis meteorological data within the target area are acquired to obtain a spatiotemporally matched, lightning-meteorological synchronized basic dataset, specifically including: S1.1 Obtaining lightning frequency observation data: Obtaining the target area within a continuous time window T obs Wide-area lightning monitoring data is collected, recording the occurrence time (accurate to the second), latitude and longitude coordinates, polarity, and current amplitude of each lightning event. The data sources include ground-based lightning location systems and spaceborne lightning detection instruments. Based on this data, spatiotemporal aggregation and statistical analysis are performed to obtain the number of lightning occurrences per unit area per unit time, which is used as the lightning frequency observation data. Continuous time window T obs This period is designated as a time when the lightning observation system is operating stably and the data quality is reliable. Quality control is performed on the acquired lightning data, removing records with excessive positioning errors or abnormal current amplitudes.

[0031] S1.2 Obtaining Reanalysis Meteorological Data: Obtaining data within the same time window T as the lightning frequency observation data. obs The reanalysis meteorological data refers to reanalysis meteorological data of the same target area. This reanalysis meteorological data is a global atmospheric reanalysis product (such as ERA5 released by ECMWF), generated by fusing historical observation data with numerical models through data assimilation technology, providing continuous and spatially complete meteorological field data since 1950. The reanalysis meteorological data includes at least the thermodynamic, dynamic, and microphysical variables related to lightning generation. Specifically, these variables include, but are not limited to: convective effective potential energy (CAPE, unit J / kg), representing atmospheric instability energy; the K-index (dimensionless), representing atmospheric stability; 500 hPa relative humidity (RH500, unit %), representing mid-level humidity; 500 hPa vertical velocity (ω500, unit Pa / s), representing dynamic lifting conditions; and 0°C layer height (H0°C, unit m), representing the height at which ice phase processes begin. All of the above variables have a clear causal relationship with lightning occurrence.

[0032] S1.3 Spatiotemporal Grid Matching: Lightning frequency observation data and reanalysis meteorological data are matched to the same spatiotemporal grid to obtain the initial basic dataset. Spatially, a latitude and longitude grid covering the target area is established, with a uniform spatial resolution set. Spatial interpolation methods (including but not limited to bilinear interpolation, nearest neighbor interpolation, or inverse distance weighted interpolation) are used to resample the reanalysis meteorological data to the lightning observation grid. Temporally, an hourly aggregation strategy is adopted, and for each grid i and each hour t, the average values ​​of lightning frequency and meteorological variable factors are calculated respectively. Lightning Frequency The calculation formula is: ; in, Let be the number of ground flashes (i.e., lightning frequency) counted for grid i in hour t. Let be the area of ​​grid i.

[0033] Grid area is calculated based on spherical geometry: = R 2 ·Δφ·Δλ·cosφi, where R is the average radius of the Earth (6371 km), Δφ and Δλ are the latitudinal and meridional grid step sizes (in radians), respectively, and φi is the latitude of the grid center.

[0034] S1.4 Data Cleaning: The initial base dataset is cleaned to remove the following samples: grids with zero lightning observation frequency and an hourly average CAPE below a preset threshold (these grids have no physical possibility of generating lightning); high-altitude grids with terrain height exceeding a preset threshold and a consistently zero lightning frequency; thus forming the final base dataset S={(X i,t ,L i,tobs )}, where X i,t It is a variable factor vector.

[0035] Considering the potential for strong multicollinearity among different meteorological variables, and that statistical correlation does not equate to causation, this embodiment may include the following when extracting the predictor set in S1: S1.5 Correlation Analysis: Calculate the variance inflation factor (VIF) among the variables, check for multicollinearity, and remove variables whose VIF exceeds a preset threshold (preferably 10).

[0036] The Pearson correlation coefficient and partial correlation coefficient between each meteorological variable factor after removal and the lightning frequency are calculated under a spatiotemporal grid to quantitatively evaluate the predictive ability of each variable factor.

[0037] like Figure 2 The figure shows a heatmap of the Pearson correlation coefficient matrix between the various variable factors and the frequency of lightning in this invention. The figure displays the pairwise correlation coefficients between six factors: CAPE, K-index, RH500 (500 hPa relative humidity), ω500 (500 hPa vertical velocity), H0°C (0°C altitude), and lightning frequency. The heatmap uses a red-blue color scheme, with red indicating a positive correlation and blue indicating a negative correlation; darker colors indicate stronger correlations. Each cell is labeled with the specific correlation coefficient value. The graph shows that CAPE has the highest correlation coefficient with lightning frequency, reaching 0.68, indicating that convective available potential energy is the most important thermodynamic factor affecting lightning occurrence. The K-index has a correlation coefficient of 0.56 with lightning frequency, reflecting the significant impact of atmospheric stability on lightning generation. ω500 is negatively correlated with lightning frequency (-0.50), indicating that the stronger the upward motion (the larger the negative value of ω5000), the higher the lightning frequency, consistent with the physical mechanism of lightning. RH500 has a moderate positive correlation with lightning frequency (0.46), indicating that mid-level humidity has a significant impact on the charge separation process within clouds. The correlation between H0°C and lightning frequency is relatively weak (0.27), but still has physical significance, reflecting the influence of the initiation height of the ice phase process on lightning generation. This thermogram statistically verifies the rationality and physical interpretability of the selected predictive factors, providing a quantitative basis for the selection of the predictive factor set.

[0038] S1.6 Physical Mechanism Verification: Based on the principles of lightning physics, verify the causal relationship between each variable factor and lightning generation to ensure that the selected variable factors have clear physical meaning.

[0039] The principles of lightning physics include: convective effective potential energy determines the intensity of updrafts and cloud top height, serving as the energy basis for convection development; the K-index comprehensively reflects the configuration of warm and humid conditions in the lower layer and dry and cold conditions in the middle layer, and is closely related to thunderstorm potential; 500 hPa relative humidity affects the charge separation efficiency of mixed-phase cloud regions; 500 hPa vertical velocity reflects the dynamic lifting intensity that triggers convection; and the 0°C layer height determines the starting height of ice phase processes within the cloud and the vertical range of the charge generation zone.

[0040] S1.7 Determine the predictor set: Based on correlation analysis and physical verification, determine the predictor set X = {x1, x2, ..., x} for lightning frequency modeling. m}, where m is the number of predictor factors. The predictor factor set includes at least the five core variable factors mentioned above, specifically including convective available potential energy (CAPE), K index, 500 hPa relative humidity (RH500), 500 hPa vertical velocity (ω500), and 0°C layer height (H0°C), and supplementary variable factors can be added according to the characteristics of the target area.

[0041] The above process, by calculating the variance inflation factor among the variables and verifying the physical mechanism, can quantitatively evaluate the predictive ability of each selected variable, ensuring that each factor input into the model has a clear causal basis in the physics of lightning, and avoiding the model output from violating meteorological laws.

[0042] Considering the significant altitude differences, complex underlying surfaces, and substantial variations in lightning characteristics with altitude in the target region (such as the Qinghai-Tibet Plateau), using a single model across the entire region would lead to a decrease in goodness of fit. In S2 above, based on the geographical characteristics of the target region and the amount of lightning frequency observation data, the target region is divided into multiple modeling units. Differentiated modeling strategies can be adopted for different regional characteristics. By increasing the sample size within each partition, the robustness of the model coefficient estimation is improved, enabling the transferable application of the statistical model to regions with different geographical features.

[0043] In this embodiment, S2 includes: The target area is gridded. When the amount of lightning frequency observation data in a single grid is not less than a preset sample size threshold, the target area is divided into multiple modeling units based on the grid. When the amount of lightning frequency observation data in a single grid is lower than the preset sample size threshold, or when the target area has preset geographical features, multiple grids are merged into one partition, and the target area is divided into multiple modeling units based on the partition; wherein, the total amount of lightning frequency observation data in multiple grids within a partition is not lower than the preset sample size threshold, or multiple grids within a partition have the same preset geographical features.

[0044] Specifically, for areas with sufficient sample size (i.e., the number of samples in a single grid exceeds the preset sample size threshold; since the reanalysis meteorological data is continuous and massive, the sample size here is represented by the amount of lightning frequency observation data), a grid-by-grid modeling strategy is adopted. For sparse areas with insufficient sample size (i.e., the number of samples in a single grid is less than the preset sample size threshold), or for areas with special geographical features, a zonal modeling strategy is adopted.

[0045] The preset sample size threshold is the minimum number of valid samples required to ensure the robustness of the statistical model parameter estimation. It can be determined in one of the following ways: First, it is determined based on statistical rules of thumb: the preset sample size threshold is set to 10 to 30 times the number of predictor variable factors to ensure that the model has sufficient degrees of freedom to solve the parameters; Second, dynamic determination based on cross-validation: Within the current grid or candidate partition, gradually increase the number of samples participating in training, calculate the mean squared error (MSE) of the model during the validation period after each increase in samples, and when the rate of decrease of MSE is less than the preset convergence condition (such as 5%), the corresponding number of samples is the preset sample size threshold for that region.

[0046] In practical engineering applications, considering that the core predictive variable factors selected in this invention are five, and that lightning activity is a sparse event, the preset sample size threshold is preferably 100 to 200 valid lightning hour samples. When the number of valid lightning hour samples in a single grid reaches this threshold, independent modeling is adopted grid-by-grid; otherwise, a merging mechanism is triggered to merge adjacent grids with the same geographical features into partitions until the total sample size within a partition meets the threshold requirement. Through the above methods, an optimal balance can be achieved between the risk of model overfitting and spatial resolution.

[0047] In this embodiment, the preset geographical features include: Terrain stratification features based on altitude; where grids with altitudes within the same preset stratification range are considered to have the same terrain stratification features; Land cover characteristics based on underlying surface type; wherein, grids with the same underlying surface type are considered to have the same land cover characteristics; Climate zoning characteristics based on lightning activity features; where grids belonging to the same climate type and lightning zoning are considered to have the same climate zoning characteristics. Functional zoning features based on the characteristics of power transmission corridors; where grids located in the same geographical environment type of power transmission corridor are considered to have the same functional zoning features; Hybrid characteristics based on data availability; where grids with lightning monitoring station densities within the same preset range are considered to have the same hybrid characteristics.

[0048] The specific strategies for zoning based on the aforementioned preset geographical features include: (I) Topographic stratification based on altitude The target area was divided into several tiers based on altitude: low altitude (<1000m), medium altitude (1000-3000m), and high altitude (>3000m). For high-altitude areas with dense power transmission line traversal, such as the Qinghai-Tibet Plateau and western Sichuan, the area was further subdivided into: river valleys (2500-3500m, mainly including southeastern Tibet and western Sichuan river valleys), plateau areas (3500-4500m, mainly including the Lhasa River Valley and eastern Nagqu), and high-altitude areas (4500-5500m, mainly including the Ngari region, Hoh Xil, and Tanggula Pass). Different altitude tiers were further divided into sub-regions to establish statistical models that characterize the impact of altitude on convection triggering and lightning generation.

[0049] (ii) Land cover feature zoning based on underlying surface type The target area is divided into four land cover types: urban built-up areas, farmland / grassland, forest / shrubland, water bodies / wetlands, and bare land / desert. Different underlying surface types have different surface thermal properties and roughness, affecting the triggering and development of local convection. Transmission line corridors often traverse multiple underlying surface types; zonal modeling can improve the model's adaptability to different surface types.

[0050] (III) Climate regionalization based on lightning activity characteristics Using a combination of Köppen-Geiger climate classification and lightning climate zoning, the target region was divided into: tropical monsoon climate zone (South China, southern Yunnan), subtropical humid climate zone (Central China, East China, eastern Southwest China), temperate monsoon climate zone (North China, Northeast China), plateau climate zone (Qinghai-Tibet Plateau and surrounding areas), and arid and semi-arid zone (Northwest China). Significant differences exist in the seasonal distribution and triggering mechanisms of lightning activity within each zone, allowing for targeted optimization of the zoning model.

[0051] (iv) Functional zoning based on the characteristics of power transmission corridors To meet the application needs of the power industry, transmission line corridors are divided into zones based on their geographical characteristics: open plains, hilly areas, mountainous canyon areas, and coastal-land-border areas. Among these, the mountainous canyon areas and the coastal-land-border areas are high-risk regions for transmission line tripping due to lightning strikes and require special attention. For the coastal-land-border areas, the sea-land temperature difference and distance from the sea should be added as supplementary predictive factors; for the mountainous canyon areas, local elevation differences and valley orientation should be added as supplementary predictive factors.

[0052] (v) Hybrid partitioning based on data availability For areas with dense monitoring stations, independent modeling is adopted on a grid-by-grid basis; for sparsely monitored areas, they are merged into a regional model; for extremely sparse areas (such as the desert hinterland or uninhabited areas), spatial proximity weighted regression or transfer learning methods are used to transfer the model relationships of neighboring areas or source domains (such as the eastern plains) to the target domain.

[0053] Considering that lightning is the result of a combination of thermal, dynamic, and microphysical processes, and that there are nonlinear relationships and threshold effects among different variables, regression analysis is used in S3 above to construct a mapping relationship. This effectively captures the quantitative relationship between predictor variables and lightning frequency, and adaptively selects the benchmark or enhanced model based on the goodness of fit, thereby improving the model's prediction accuracy in complex terrain areas.

[0054] In this embodiment, S3 includes: For each modeling unit, regression analysis is performed based on a multiple linear regression model to construct the mapping relationship between the predictor set in each modeling unit and the lightning frequency observation data. The least squares method is used to solve the regression coefficients, and inequality constraints are applied to the regression coefficients according to the lightning physical mechanism to obtain the baseline model of each modeling unit. When the goodness of fit of the benchmark model is not lower than the preset goodness threshold, the benchmark model is used as the lightning frequency statistical model. When the goodness of fit of the baseline model is lower than the preset goodness threshold, regression analysis is performed based on machine learning regression algorithm to construct the mapping relationship between the prediction factor set in each modeling unit and the lightning frequency observation data, so as to obtain the enhanced model of each modeling unit, and the enhanced model is used as the lightning frequency statistical model.

[0055] The specific construction process of the above benchmark model includes: S3.1 Modeling Time Period Division: Divide the continuous time window T obs The continuous time window T divided into training periods train and the continuous time window T of the verification period valid The training period is used for solving model parameters and training, while the validation period is used for independent evaluation of model performance. The preferred ratio of training to validation time is 3:1 to 4:1. The specific division method can be adjusted according to the amount of data, but it is necessary to ensure that the validation period contains enough samples to perform statistical significance tests.

[0056] S3.2 Baseline Model Establishment: Multiple linear regression is used as the baseline modeling method. For each modeling unit, a linear relationship between lightning frequency and predictor factors is established. ; in Simulated data for lightning frequency estimation by the model. There are 1 to m predictors. The regression coefficients are for 1 to m predictors. These are the basic regression coefficients.

[0057] The regression coefficients are solved using the least squares method, and the objective function is: ; Among them, T train For the continuous time window of the model training period, This is data on lightning frequency observation. This is simulated data on lightning frequency.

[0058] Based on the physical mechanism of lightning, inequality constraints are applied to the regression coefficients. In this embodiment, the inequality constraints include: In the predictor set, the regression coefficients of variables related to thermal instability are greater than or equal to zero, including convective available potential energy and the K exponent; in the predictor set, the regression coefficients of variables related to upward motion are less than or equal to zero, including the 500 hPa vertical velocity.

[0059] The above constraints ensure that the model conforms to the physical law that "the greater the unstable energy and the stronger the upward motion, the higher the frequency of lightning".

[0060] For linear regression models whose goodness-of-fit R² is lower than a preset goodness-of-fit threshold (preferably 0.60) during the validation period, an augmentation model is constructed using a machine learning regression algorithm. The specific construction process of the above augmentation model includes: The machine learning regression algorithms mentioned include, but are not limited to: Random Forest, Multilayer Perceptron, Gradient Boosting Tree (XGBoost / LightGBM), Support Vector Regression (SVR), or Artificial Neural Network (ANN). Taking Random Forest as an example, its core idea is to reduce the risk of overfitting by constructing multiple decision trees and taking the average. The prediction expression of the Random Forest model is: ; Where N tree T represents the number of decision trees. k (x) represents the predicted value of the k-th decision tree. Model hyperparameters (including the number of trees, maximum depth, minimum number of leaf samples, and number of feature samples) are optimized through cross-validation. The advantage of machine learning methods lies in their ability to automatically capture the nonlinear relationship and interaction between predictors and lightning frequency, making them particularly suitable for regions where variables such as CAPE exhibit threshold effects.

[0061] The training process may include: The first step is to start from the training period T train The data includes bootstrap samples extracted with replacement to generate training subsets for each decision tree; the second step involves randomly selecting features from all m predictor variables at each split node, and searching for the optimal split point based on the mean squared error minimization criterion; the third step repeats the splitting process until the stopping condition is met. Model hyperparameters are optimized through cross-validation, specifically including: the number of decision trees N. tree (Preferred 500 trees), maximum depth (preferred 10), minimum number of leaf samples (preferred 3).

[0062] Figure 3 This is a scatter plot of observed and simulated values ​​for a specific grid during the independent validation period (2021-2024) of the model established in this invention. Each point in the plot represents an hourly sample of one grid. The horizontal axis represents observed lightning frequency data, and the vertical axis represents simulated lightning frequency data. The unit is fl / km. 2 The figure shows the 1:1 reference line (represented by a gray dashed line) and the fitted line (represented by a red solid line) obtained by linearly fitting the scatter data, with the fitted equation being y = 0.869x + 0.009. Key statistical indicators are also marked in the figure: sample size N = 5500, correlation coefficient R = 0.946, and coefficient of determination R0. 2 =0.894, Root Mean Square Error (RMSE) = 0.79 fl / km 2 / h. As can be seen from the figure, the scatter points are densely distributed on both sides of the 1:1 reference line, and the fitted line is very close to the 1:1 reference line, indicating that the model simulation values ​​and observed values ​​have good consistency. The high correlation coefficient and determination coefficient indicate that the statistical model established in this invention can effectively capture the quantitative relationship between lightning frequency and meteorological predictive factors, and the low root mean square error verifies that the model has high prediction accuracy.

[0063] Considering the lack of lightning observations during the historical reconstruction period but the availability of complete reanalysis data, in step S4 above, the set of predictive variables for the target area during the historical reconstruction period is input into the lightning frequency statistical model of each modeling unit. This yields the initial historical lightning frequency values ​​for the target area, enabling the statistical model to be transferred and inverted over historical time series. This extends the effective lightning frequency data length from nearly twenty years back to nearly a century, filling the gap in long-term historical lightning frequency data.

[0064] In this embodiment, S4 specifically includes: Obtain the historical reconstruction period T of the target area histReanalysis meteorological data (preferably from 1950 to 2010) is used to extract a set of variable factors of the same type as the predictive factor set in step S1, which are then used as historical reanalysis variable factors. These historical reanalysis variable factors are input into the statistical model trained in step S3, and initial values ​​of historical lightning frequency are calculated for each modeling unit and each hour. : ; Where y represents the historical hour, y∈T hist f(·) is the statistical model of lightning frequency obtained through training. Let be the set of predictors for grid i in historical hours y. Calculate hourly for each modeling unit, and detect and correct outliers. The outlier criterion is that the value exceeds the historical series mean of that unit plus or minus k times the standard deviation (k is preferably 5). The correction method uses linear interpolation or spline interpolation over several days before and after (preferably 5 days before and after).

[0065] For the goodness-of-fit R of the linear regression model during the validation period 2 For modeling units with frequencies below a preset threshold (preferably 0.60), an enhanced model is used. Historical reanalysis variable factors are input into the enhanced model, and the initial values ​​of historical lightning frequency are calculated for each modeling unit and each hour.

[0066] Considering that the evolution of the assimilation system can lead to systematic bias when migrating from recent modeling periods to historical inversion periods, the quantile mapping bias correction performed on the initial historical lightning frequency values ​​in S5 above can establish a mapping relationship between the simulated values ​​and the cumulative distribution function of the observed values, effectively eliminating systematic bias and reducing RMSE.

[0067] In this embodiment, the quantile mapping deviation correction in S5 above includes: The set of predictive factors that are spatiotemporally matched with the lightning frequency observation data is input into the lightning frequency statistical model of each modeling unit to obtain lightning frequency simulation data. When the mean deviation between the simulated lightning frequency data and the observed lightning frequency data within a modeling unit significantly deviates from zero, the simulated cumulative distribution function and the observed cumulative distribution function are calculated respectively based on the simulated lightning frequency data and the observed lightning frequency data of that modeling unit. Calculate the quantile position of the initial historical lightning frequency of the modeling unit in the simulated cumulative distribution function, and use the value of the observed cumulative distribution function corresponding to the quantile position as the initial calibration value of the historical lightning frequency of the modeling unit after quantile mapping deviation correction.

[0068] Specifically, the simulation bias of each modeling unit during the training period is calculated using lightning frequency simulation data. i,t : Bias i,t=L i,t obs -L i,t sim ; Among them, L i,t obs For lightning frequency observation data, L i,t sim For training period T train Simulated data on lightning frequency; Analyze the statistical distribution characteristics of the bias (mean, variance, skewness) and its relationship with predictors, spatial location, and time. Identify whether systematic bias exists (i.e., the mean of the bias deviates significantly from zero).

[0069] The core idea of ​​quantile mapping is to establish a cumulative distribution function mapping relationship between simulated values ​​and observed values ​​during the training period, and then apply this mapping to the initial values ​​of historical lightning frequencies.

[0070] The specific steps include: for each modeling unit, based on the training period T... train Lightning frequency simulation data and lightning frequency observation data L i,t obs Calculate the simulated cumulative distribution function F respectively. sim train and the observed cumulative distribution function F obs train For the historical reconstruction period T hist Initial value L for each historical lightning frequency i,y recon,raw Calculate its quantile position p = F in the simulated cumulative distribution function. sim train (L i,y recon,raw ); The value of the corresponding quantile p in the observed cumulative distribution function is taken as the initial calibration value of the historical lightning frequency of the modeling unit after quantile mapping bias correction. : ; The cumulative distribution function is calculated using the percentile method, taking Nquant quantiles (Nquant is preferably 51, i.e., 2% intervals from 0% to 100%). For simulated values ​​that exceed the training period range (i.e., p is close to 0 or 1), linear extrapolation is used, with the extrapolation magnitude not exceeding 20% ​​of the extreme values ​​during the training period.

[0071] Considering that spatially adjacent grids may exhibit anomalous jumps due to abrupt terrain changes or local noise, spatial consistency correction is performed in S5 above. This optimizes the spatial rationality of the data, outputting high-resolution gridded data that is spatially continuous and without gaps, thus avoiding unreasonable lightning risk abrupt changes along the transmission line corridor.

[0072] In this embodiment, the spatial consistency correction in S5 above includes: Spatial continuity checks are performed on the initial historical lightning frequency values ​​of the target area to identify aberration grids with abnormal abrupt changes; The initial historical lightning frequency values ​​in the mutation grid are processed using spatial weighted smoothing to obtain the reconstructed historical lightning frequency values; wherein, the weights in the spatial weighted smoothing process are determined according to the geographical characteristics of the mutation grid.

[0073] In this embodiment, determining the weights in the spatial weighted smoothing process based on the geographical features of the mutation grid includes: When the mutation grid is located in a region with flat terrain or uniform underlying surface, the weights in the spatial weighted smoothing process are determined based on the spatial distance between the mutation grid and adjacent grids. When the mutation grid is located in a region with complex terrain, terrain feature similarity constraints are superimposed on the weights using digital elevation model data, so that the weights decrease as the spatial distance and differences in terrain features increase. When the mutation grid is located in the coastal-land boundary region, the weights are determined independently for the land region and the marine region, with the coastline as the boundary.

[0074] When the abrupt grid is located in a region with flat terrain or a uniform underlying surface, Gaussian kernel function weighted smoothing is specifically used to determine the weights in the spatial weighted smoothing process: For abrupt grid changes, a Gaussian kernel-weighted average is used for local smoothing, with the weights... The expression is: ; Where d ij L represents the spatial distance between abruptly changed grids and adjacent grids, and L is the spatial attenuation scale (preferably 100 km). The corrected historical lightning frequency reconstruction value. for: ; in, Let i be the set of neighboring grids. The lightning frequency reconstruction value of grid j before correction.

[0075] When the abrupt grid is located in a terrain-complex area (such as a high mountain canyon area, the Qinghai-Tibet Plateau, etc.), terrain feature similarity constraints are superimposed on the weights using digital elevation model (DEM) data, causing the weights to decrease as spatial distance and terrain feature differences increase. A contour-constrained Kriging interpolation method is used, where the weights consider not only spatial distance but also terrain feature similarity. The core idea is as follows: ① Establish the statistical relationship between "lightning frequency and altitude": By statistically analyzing each altitude level, determine the baseline curve of lightning frequency variation with altitude.

[0076] ② When the initial reconstructed value of a grid deviates from the predicted value based on the altitude baseline curve by more than a preset threshold (such as 2 standard deviations), correction is triggered.

[0077] ③ During calibration, the values ​​of adjacent grids on the same terrain feature line are preferentially referenced, rather than simply weighted by distance. Specifically, for the grid i to be calibrated, its neighborhood weights consider not only spatial distance but also terrain feature similarity: ; Where z i z j H represents the elevation of each grid cell, and H is the vertical attenuation scale (preferably 300m). The indicator function takes a value of 1 when grids i and j are located in the same contour region, and 0.5 otherwise. This method can effectively avoid the problem of incorrect smoothing of mountain top grids and valley grids at the foot of mountains, thus maintaining the physical rationality of terrain features.

[0078] When the mutation grid is located in the land-sea interface region, the weights for smoothing are determined independently for the land and marine regions, using the coastline as the boundary, to avoid unrealistic smoothing between the land and sea areas. After each correction, the spatial variability function before and after correction is calculated to evaluate the smoothness. Simultaneously, independent verification points (such as lightning observation station data or historical lightning trip records) are selected to verify the correction effect. If the correction causes the error at the verification point to increase beyond a preset threshold, the correction is rolled back or the smoothing parameters are adjusted to obtain the final reconstructed historical lightning frequency values.

[0079] The corrected historical lightning frequency reconstruction values ​​are output in a standard data format, including but not limited to NetCDF-4, HDF5, GeoTIFF, or CSV. NetCDF-4 format is preferred due to its self-descriptive nature and good support for multidimensional data.

[0080] The output data should contain at least three dimensions: longitude, latitude, and time. The variable name should be defined as `lightning_density`, and the unit should be flashes per square kilometer per hour (fl / km² / h). Auxiliary metadata should also be output, including but not limited to: the modeling method identifier for each modeling unit (linear regression or machine learning type), regression coefficients (for linear models), correction parameters (for QM correction), and spatial reference information. The longitude range should cover the minimum to maximum longitude of the target area, the latitude range should cover the minimum to maximum latitude of the target area, and the time range should cover the historical reconstruction period T. hist The start and end years. The spatial resolution is consistent with the set grid resolution.

[0081] This embodiment uses the Qinghai-Tibet Plateau region (longitude 75°E-105°E, latitude 25°N-40°N) as the application area, and demonstrates in detail the specific implementation methods and parameter selections for each step of the present invention.

[0082] Obtain lightning monitoring data from the State Grid wide-area lightning monitoring system for the target area from 2011 to 2024. This data records the occurrence time (accurate to the second), latitude and longitude coordinates, polarity, and current amplitude of each ground flash event.

[0083] We acquired ERA5 reanalysis meteorological data released by the European Centre for Medium-Range Weather Forecasts (ECMWF) during the same period, with hourly temporal resolution and spatial resolution of 0.25° × 0.25°, covering 37 vertical layers. The following variables were extracted: convective available potential energy (CAPE, J / kg), K-index (dimensionless), 500 hPa relative humidity (RH500, %), 500 hPa vertical velocity (ω500, Pa / s), and 0°C layer height (H0°C, m).

[0084] A 0.1° × 0.1° latitude and longitude grid was established, comprising approximately 2400 grids. Bilinear interpolation was used to resample the reanalysis meteorological data to the lightning observation grid. The number of ground flashes within each grid was counted hourly, and the grid area was calculated using spherical geometry and converted to lightning frequency.

[0085] Grids with zero lightning frequency and hourly average CAPE less than 100 J / kg, as well as grids in glacier areas with an altitude greater than 5500 meters, were excluded.

[0086] Pearson correlation coefficients between each variable factor and lightning frequency were calculated based on lightning samples from 2011 to 2024. Results showed: CAPE correlation coefficients were 0.55–0.72 (Eastern region) and 0.45–0.60 (Western region); K-index was 0.40–0.55; RH500 was 0.35–0.50; ω500 was -0.40 to -0.55; and H0°C was 0.25–0.40. The VIF values ​​for all variable factors were less than 5, indicating no significant multicollinearity. The causal rationality of each variable factor was verified based on the physical mechanism of lightning. The predictor factor set was determined to consist of five variable factors: CAPE, K-index, RH500, ω500, and H0°C.

[0087] The training period was from 2011 to 2020, and the validation period was from 2021 to 2024. Considering the large altitude differences and significant variations in lightning characteristics with altitude on the Qinghai-Tibet Plateau, a terrain-layered modeling strategy was adopted. Furthermore, taking into account that transmission lines mainly run along river valleys and mountain passes, this invention employs functional zoning based on the characteristics of the transmission corridor. The river valley region (2500-3500m, mainly including the river valleys of southeastern Tibet and western Sichuan) has relatively open terrain, making a significant contribution to CAPE (Capacity for Earth Events), and is dominated by convective thunderstorms. In this region, valley floor width and the angle between the valley direction and the prevailing wind direction are added as supplementary predictive factors.

[0088] The plateau region (3500-4500m, mainly including the Lhasa River Valley and eastern Nagqu) has a flat and open terrain, with lightning activity primarily driven by thermal convection and exhibiting significant diurnal variation. A standard prediction factor set was used for this region.

[0089] In high-altitude areas (4500-5500m, mainly including Ali Prefecture, Hoh Xil, and Tanggula Pass), the terrain is narrow, and the airflow convergence effect is significant, with lightning activity concentrated under specific wind directions. In this region, the orientation of the mountain pass and the elevation difference between the mountains on either side are added as supplementary predictive factors.

[0090] A multiple linear regression equation was used, with physical constraints applied: β1≥0, β2≥0, β3≥0, β4≤0.

[0091] For the validation period R of linear regression 2 For regions with a value below 0.6, random forest regression is used with the following parameters: 500 decision trees, maximum depth 10, minimum number of leaf samples 3, and number of feature samples 3.

[0092] The ERA5 historical reanalysis variables from 1950 to 2010 were input into the model to obtain the original reconstructed values. Outlier detection was performed on each grid, and values ​​exceeding the mean ± 5 standard deviations were replaced with linear interpolations of the preceding and following values, each 5 days prior.

[0093] The quantile mapping method was used for bias correction, with each grid calculated independently, using 51 quantiles (0%, 2%, ..., 100%). Simultaneously, terrain trend consistency correction was performed: for grids that violated the physical law that "the higher the altitude, the lower the lightning frequency," adjustments were made by referencing adjacent grids of similar altitude.

[0094] In this embodiment, a contour line constraint correction method is adopted to address the complex terrain of the Qinghai-Tibet Plateau. Specifically, the study area is divided into three topographic zones: the steep southern slope zone of the Himalayas, the gentle plateau surface zone, and the undulating Tanggula-Kunlun Mountains zone. A baseline curve for "lightning frequency - altitude" is calculated for each zone. Correction is triggered when the initial reconstructed value of a grid deviates from the predicted value of the baseline curve by more than two standard deviations. During correction, adjacent grids along the same topographic feature line (such as a river valley or a ridge) are preferentially referenced. After correction, the consistency of the negative correlation between terrain and lightning increases from 0.58 to 0.71, and the smoothness of the lightning frequency profile along the transmission corridor improves by approximately 35%.

[0095] Output a NetCDF-4 format dataset with the variable factor name "lightning_density" and the unit "fl / km² / h" (number of lightning strikes per hour per square kilometer). It includes annual data from 1950 to 2010, with spatial dimensions of longitude 75-105°E (60 grid points) and latitude 25-40°N (30 grid points). Also output the modeling method identifier and regression coefficient metadata.

[0096] In summary, addressing the shortcomings of existing lightning observation data (short time span, low spatial resolution of traditional daily thunderstorm statistics methods, and the difficulty of retrospectively applying existing lightning parameterization schemes to historical meteorological data reanalysis due to reliance on real-time observations), this invention provides a complete method for reconstructing historical lightning frequencies, achieving the following beneficial effects: (1) Filling long-term historical data gaps and breaking through the limitations of time retrospection. This invention constructs a mapping relationship between predictive variable factors and lightning frequency, inputs meteorological data from the historical reconstruction period into the statistical model, and realizes the reconstruction of lightning frequency grid by grid for 61 years from 1950 to 2010. This extends the effective historical data length from nearly 20 years to more than 80 years, and the data length is more than 4 times longer than the existing observation records, providing irreplaceable basic data for the analysis of long-term lightning climate characteristics.

[0097] (2) Achieving high-resolution spatially continuous output, overcoming the shortcomings of traditional statistical methods. This invention divides modeling units based on geographical features and the amount of observation data, and eliminates anomalous abrupt changes through spatial consistency correction, finally outputting gridded reconstructed data with a preferred resolution of 0.1°×0.1°. This data is spatially continuous and without missing data, completely overcoming the shortcomings of uneven spatial coverage and low resolution in traditional thunderstorm daily statistical methods.

[0098] (3) Balancing physical interpretability and inversion accuracy to ensure model robustness. This invention is based on variable factors with clear physical significance of lightning, and applies inequality constraints based on thermodynamic and dynamic mechanisms in the regression solution to ensure that the model output conforms to the basic laws of lightning generation; at the same time, machine learning enhancement model is introduced to effectively capture nonlinear threshold effects and improve the prediction accuracy of complex terrain areas.

[0099] (4) Dual correction reduces systemic bias and ensures the reliability of reconstructed data. In response to the systematic bias caused by the evolution of the assimilation system of reanalysis data, this invention adopts quantile mapping bias correction combined with spatial consistency correction, which effectively eliminates mean shift and local spatial jump caused by model migration, and reduces the root mean square error (RMSE) during the validation period by an average of 22%-30%, significantly improving the credibility of the reconstructed data.

[0100] (5) A complete technical process is formed, with significant engineering practical value. This invention constructs a complete technical system from multi-source data acquisition, spatiotemporal matching, partitioned modeling to deviation correction output, supporting engineering deployment. The output high-resolution gridded lightning frequency reconstruction data can be directly applied to actual engineering scenarios such as transmission line lightning strike risk assessment, differentiated lightning protection design, and lightning fire analysis, providing accurate data support for disaster prevention and mitigation decision-making.

[0101] Example 2: Based on the same inventive concept, this invention also provides a historical lightning frequency reconstruction system, such as... Figure 4 As shown, it includes: The data extraction module is used to acquire spatiotemporally matched lightning frequency observation data and reanalysis meteorological data within the target area, and to extract a set of prediction factors from the reanalysis meteorological data. The partitioning module is used to divide the target area into multiple modeling units based on the geographical features of the target area and the amount of lightning frequency observation data; The modeling module is used to construct the mapping relationship between the predictor set and the lightning frequency observation data for each modeling unit by using regression analysis, so as to obtain the lightning frequency statistical model for each modeling unit. The inversion module is used to input the set of prediction factors for the target area during the historical reconstruction period into the lightning frequency statistical model of each modeling unit to obtain the initial value of the historical lightning frequency of the target area. The correction module is used to perform quantile mapping deviation correction and spatial consistency correction on the initial value of the historical lightning frequency to obtain the reconstructed value of the historical lightning frequency in the target area.

[0102] In this embodiment, the partitioning module is specifically used for: The target area is gridded. When the amount of lightning frequency observation data in a single grid is not less than a preset sample size threshold, the target area is divided into multiple modeling units based on the grid. When the amount of lightning frequency observation data in a single grid is lower than the preset sample size threshold, or when the target area has preset geographical features, multiple grids are merged into one partition, and the target area is divided into multiple modeling units based on the partition; wherein, the total amount of lightning frequency observation data in multiple grids within a partition is not lower than the preset sample size threshold, or multiple grids within a partition have the same preset geographical features.

[0103] In this embodiment, the preset geographical features include: Terrain stratification features based on altitude; where grids with altitudes within the same preset stratification range are considered to have the same terrain stratification features; Land cover characteristics based on underlying surface type; wherein, grids with the same underlying surface type are considered to have the same land cover characteristics; Climate zoning characteristics based on lightning activity features; where grids belonging to the same climate type and lightning zoning are considered to have the same climate zoning characteristics. Functional zoning features based on the characteristics of power transmission corridors; where grids located in the same geographical environment type of power transmission corridor are considered to have the same functional zoning features; Hybrid characteristics based on data availability; where grids with lightning monitoring station densities within the same preset range are considered to have the same hybrid characteristics.

[0104] In this embodiment, the modeling module is specifically used for: For each modeling unit, regression analysis is performed based on a multiple linear regression model to construct the mapping relationship between the predictor set in each modeling unit and the lightning frequency observation data. The least squares method is used to solve the regression coefficients, and inequality constraints are applied to the regression coefficients according to the lightning physical mechanism to obtain the baseline model of each modeling unit. When the goodness of fit of the benchmark model is not lower than the preset goodness threshold, the benchmark model is used as the lightning frequency statistical model. When the goodness of fit of the baseline model is lower than the preset goodness threshold, regression analysis is performed based on machine learning regression algorithm to construct the mapping relationship between the prediction factor set in each modeling unit and the lightning frequency observation data, so as to obtain the enhanced model of each modeling unit, and the enhanced model is used as the lightning frequency statistical model.

[0105] In this embodiment, the inequality constraints include: In the predictor set, the regression coefficients of variables related to thermal instability are greater than or equal to zero, and the variables related to thermal instability include convective available potential energy and the K exponent. In the set of predictor factors, the regression coefficients of variables related to upward motion are less than or equal to zero, and the variables related to upward motion include the 500 hPa vertical velocity.

[0106] In this embodiment, the correction module includes a quantile mapping deviation correction unit, used for: The set of predictive factors that are spatiotemporally matched with the lightning frequency observation data is input into the lightning frequency statistical model of each modeling unit to obtain lightning frequency simulation data. When the mean deviation between the simulated lightning frequency data and the observed lightning frequency data within a modeling unit significantly deviates from zero, the simulated cumulative distribution function and the observed cumulative distribution function are calculated respectively based on the simulated lightning frequency data and the observed lightning frequency data of that modeling unit. Calculate the quantile position of the initial historical lightning frequency of the modeling unit in the simulated cumulative distribution function, and use the value of the observed cumulative distribution function corresponding to the quantile position as the initial calibration value of the historical lightning frequency of the modeling unit after quantile mapping deviation correction.

[0107] In this embodiment, the correction module further includes a spatial consistency correction unit, which includes: An anomaly identification subunit is used to perform spatial continuity checks on the initial calibration values ​​of historical lightning frequencies in the target area and identify anomalous grids with anomalous changes. The smoothing subunit is used to perform spatial weighted smoothing on the initial calibration values ​​of historical lightning frequencies in the mutation grid to obtain the reconstructed values ​​of historical lightning frequencies. The weight determination sub-unit is used to determine the weights in the spatial weighted smoothing process based on the geographical characteristics of the mutation grid.

[0108] In this embodiment, the weight determination subunit is specifically used for: When the mutation grid is located in a region with flat terrain or uniform underlying surface, the weights in the spatial weighted smoothing process are determined based on the spatial distance between the mutation grid and adjacent grids. When the mutation grid is located in a region with complex terrain, terrain feature similarity constraints are superimposed on the weights using digital elevation model data, so that the weights decrease as the spatial distance and differences in terrain features increase. When the mutation grid is located in the coastal-land boundary region, the weights are determined independently for the land region and the marine region, with the coastline as the boundary.

[0109] Example 3 like Figure 5 As shown, the present invention also provides an electronic device, which may be a computer device, a microcontroller device, a smart mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, processor, and transceiver component are connected via a bus; the memory can be used to store executable programs, and an exemplary executable program may include instructions; the processor is used to execute the instructions stored in the memory. The memory can also be used to store data, which can be accessed and / or modified when instructions are executed.

[0110] The processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and it is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the storage medium to implement the corresponding method flow or corresponding function, so as to realize the steps of the historical lightning frequency reconstruction method in the above embodiments.

[0111] Example 4 Based on the same inventive concept, this invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory). This readable storage medium is a memory device within an electronic device used to store programs and data. It is understood that the storage medium here can include both built-in storage media within the electronic device and extended storage media supported by the electronic device. The storage medium provides storage space, which stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more executable programs (including program code). It should be noted that the storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. Loading and executing one or more instructions stored in the storage medium by the processor can implement the steps of the historical lightning frequency reconstruction method described in the above embodiments.

[0112] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0113] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0114] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0115] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit its protection scope. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that after reading the present invention, they can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the application, but these changes, modifications or equivalent substitutions are all within the protection scope of the claims of the present invention.

Claims

1. A method for reconstructing historical lightning frequencies, characterized in that, include: Acquire spatiotemporally matched lightning frequency observation data and reanalysis meteorological data within the target area, and extract a set of prediction factors from the reanalysis meteorological data; Based on the geographical features of the target area and the amount of lightning frequency observation data, the target area is divided into multiple modeling units; For each modeling unit, a regression analysis method is used to construct the mapping relationship between the predictor set and the lightning frequency observation data, so as to obtain the lightning frequency statistical model of each modeling unit; The set of predictive factors for the target area during the historical reconstruction period is input into the lightning frequency statistical model of each modeling unit to obtain the initial value of the historical lightning frequency of the target area. The initial historical lightning frequency values ​​are corrected for quantile mapping deviation and spatial consistency to obtain the reconstructed historical lightning frequency values ​​for the target area.

2. The method as described in claim 1, characterized in that, Based on the geographical features and lightning frequency observation data of the target area, the target area is divided into multiple modeling units, including: The target area is gridded. When the amount of lightning frequency observation data in a single grid is not less than a preset sample size threshold, the target area is divided into multiple modeling units based on the grid. When the amount of lightning frequency observation data in a single grid is lower than the preset sample size threshold, or when the target area has preset geographical features, multiple grids are merged into one partition, and the target area is divided into multiple modeling units based on the partition; wherein, the total amount of lightning frequency observation data in multiple grids within a partition is not lower than the preset sample size threshold, or multiple grids within a partition have the same preset geographical features.

3. The method as described in claim 2, characterized in that, The preset geographical features include: Terrain stratification features based on altitude; where grids with altitudes within the same preset stratification range are considered to have the same terrain stratification features; Land cover characteristics based on underlying surface type; wherein, grids with the same underlying surface type are considered to have the same land cover characteristics; Climate zoning characteristics based on lightning activity features; where grids belonging to the same climate type and lightning zoning are considered to have the same climate zoning characteristics. Functional zoning features based on the characteristics of power transmission corridors; where grids located in the same geographical environment type of power transmission corridor are considered to have the same functional zoning features; Hybrid characteristics based on data availability; where grids with lightning monitoring station densities within the same preset range are considered to have the same hybrid characteristics.

4. The method according to any one of claims 1-3, characterized in that, For each modeling unit, a regression analysis method is used to construct a mapping relationship between the predictor set and the lightning frequency observation data, thereby obtaining a lightning frequency statistical model for each modeling unit, including: For each modeling unit, regression analysis is performed based on a multiple linear regression model to construct the mapping relationship between the predictor set in each modeling unit and the lightning frequency observation data. The least squares method is used to solve the regression coefficients, and inequality constraints are applied to the regression coefficients according to the lightning physical mechanism to obtain the baseline model of each modeling unit. When the goodness of fit of the benchmark model is not lower than the preset goodness threshold, the benchmark model is used as the lightning frequency statistical model. When the goodness of fit of the baseline model is lower than the preset goodness threshold, regression analysis is performed based on machine learning regression algorithm to construct the mapping relationship between the prediction factor set in each modeling unit and the lightning frequency observation data, so as to obtain the enhanced model of each modeling unit, and the enhanced model is used as the lightning frequency statistical model.

5. The method as described in claim 4, characterized in that, The inequality constraints include: In the predictor set, the regression coefficients of variables related to thermal instability are greater than or equal to zero, and the variables related to thermal instability include convective available potential energy and the K exponent. In the set of predictor factors, the regression coefficients of variables related to upward motion are less than or equal to zero, and the variables related to upward motion include the 500 hPa vertical velocity.

6. The method as described in claim 2 or 3, characterized in that, The quantile mapping bias correction includes: The set of predictive factors that are spatiotemporally matched with the lightning frequency observation data is input into the lightning frequency statistical model of each modeling unit to obtain lightning frequency simulation data. When the mean deviation between the simulated lightning frequency data and the observed lightning frequency data within a modeling unit significantly deviates from zero, the simulated cumulative distribution function and the observed cumulative distribution function are calculated respectively based on the simulated lightning frequency data and the observed lightning frequency data of that modeling unit. Calculate the quantile position of the initial historical lightning frequency of the modeling unit in the simulated cumulative distribution function, and use the value of the observed cumulative distribution function corresponding to the quantile position as the initial calibration value of the historical lightning frequency of the modeling unit after quantile mapping deviation correction.

7. The method as described in claim 6, characterized in that, The spatial consistency correction includes: Spatial continuity checks are performed on the initial historical lightning frequency values ​​of the target area to identify aberration grids with abnormal abrupt changes; The initial historical lightning frequency values ​​in the mutation grid are processed using spatial weighted smoothing to obtain the reconstructed historical lightning frequency values; wherein, the weights in the spatial weighted smoothing process are determined according to the geographical characteristics of the mutation grid.

8. The method as described in claim 7, characterized in that, The step of determining the weights in the spatially weighted smoothing process based on the geographical features of the mutation grid includes: When the mutation grid is located in a region with flat terrain or uniform underlying surface, the weights in the spatial weighted smoothing process are determined based on the spatial distance between the mutation grid and adjacent grids. When the mutation grid is located in a region with complex terrain, terrain feature similarity constraints are superimposed on the weights using digital elevation model data, so that the weights decrease as the spatial distance and differences in terrain features increase. When the mutation grid is located in the coastal-land boundary region, the weights are determined independently for the land region and the marine region, with the coastline as the boundary.

9. A historical lightning frequency reconstruction system, characterized in that, include: The data extraction module is used to acquire spatiotemporally matched lightning frequency observation data and reanalysis meteorological data within the target area, and to extract a set of prediction factors from the reanalysis meteorological data. The partitioning module is used to divide the target area into multiple modeling units based on the geographical features of the target area and the amount of lightning frequency observation data; The modeling module is used to construct the mapping relationship between the predictor set and the lightning frequency observation data for each modeling unit by using regression analysis, so as to obtain the lightning frequency statistical model for each modeling unit. The inversion module is used to input the set of prediction factors for the target area during the historical reconstruction period into the lightning frequency statistical model of each modeling unit to obtain the initial value of the historical lightning frequency of the target area. The correction module is used to perform quantile mapping deviation correction and spatial consistency correction on the initial value of the historical lightning frequency to obtain the reconstructed value of the historical lightning frequency in the target area.

10. The system as described in claim 9, characterized in that, The partitioning module is specifically used for: The target area is gridded. When the amount of lightning frequency observation data in a single grid is not less than a preset sample size threshold, the target area is divided into multiple modeling units based on the grid. When the amount of lightning frequency observation data in a single grid is lower than the preset sample size threshold, or when the target area has preset geographical features, multiple grids are merged into one partition, and the target area is divided into multiple modeling units based on the partition; wherein, the total amount of lightning frequency observation data in multiple grids within a partition is not lower than the preset sample size threshold, or multiple grids within a partition have the same preset geographical features.

11. The system as claimed in claim 10, characterized in that, The preset geographical features include: Terrain stratification features based on altitude; where grids with altitudes within the same preset stratification range are considered to have the same terrain stratification features; Land cover characteristics based on underlying surface type; wherein, grids with the same underlying surface type are considered to have the same land cover characteristics; Climate zoning characteristics based on lightning activity features; where grids belonging to the same climate type and lightning zoning are considered to have the same climate zoning characteristics. Functional zoning features based on the characteristics of power transmission corridors; where grids located in the same geographical environment type of power transmission corridor are considered to have the same functional zoning features; Hybrid characteristics based on data availability; where grids with lightning monitoring station densities within the same preset range are considered to have the same hybrid characteristics.

12. The system according to any one of claims 9-11, characterized in that, The modeling module is specifically used for: For each modeling unit, regression analysis is performed based on a multiple linear regression model to construct the mapping relationship between the predictor set in each modeling unit and the lightning frequency observation data. The least squares method is used to solve the regression coefficients, and inequality constraints are applied to the regression coefficients according to the lightning physical mechanism to obtain the baseline model of each modeling unit. When the goodness of fit of the benchmark model is not lower than the preset goodness threshold, the benchmark model is used as the lightning frequency statistical model. When the goodness of fit of the baseline model is lower than the preset goodness threshold, regression analysis is performed based on machine learning regression algorithm to construct the mapping relationship between the prediction factor set in each modeling unit and the lightning frequency observation data, so as to obtain the enhanced model of each modeling unit, and the enhanced model is used as the lightning frequency statistical model.

13. The system as described in claim 12, characterized in that, The inequality constraints include: In the predictor set, the regression coefficients of variables related to thermal instability are greater than or equal to zero, and the variables related to thermal instability include convective available potential energy and the K exponent. In the set of predictor factors, the regression coefficients of variables related to upward motion are less than or equal to zero, and the variables related to upward motion include the 500 hPa vertical velocity.

14. The system as described in claim 10 or 11, characterized in that, The correction module includes a quantile mapping deviation correction unit, used for: The set of predictive factors that are spatiotemporally matched with the lightning frequency observation data is input into the lightning frequency statistical model of each modeling unit to obtain lightning frequency simulation data. When the mean deviation between the simulated lightning frequency data and the observed lightning frequency data within a modeling unit significantly deviates from zero, the simulated cumulative distribution function and the observed cumulative distribution function are calculated respectively based on the simulated lightning frequency data and the observed lightning frequency data of that modeling unit. Calculate the quantile position of the initial historical lightning frequency of the modeling unit in the simulated cumulative distribution function, and use the value of the observed cumulative distribution function corresponding to the quantile position as the initial calibration value of the historical lightning frequency of the modeling unit after quantile mapping deviation correction.

15. The system as described in claim 14, characterized in that, The correction module further includes a spatial consistency correction unit, which includes: An anomaly identification subunit is used to perform spatial continuity checks on the initial calibration values ​​of historical lightning frequencies in the target area and identify anomalous grids with anomalous changes. The smoothing subunit is used to perform spatial weighted smoothing on the initial calibration values ​​of historical lightning frequencies in the mutation grid to obtain the reconstructed values ​​of historical lightning frequencies. The weight determination sub-unit is used to determine the weights in the spatial weighted smoothing process based on the geographical characteristics of the mutation grid.

16. The system as described in claim 15, characterized in that, The weight determination sub-unit is specifically used for: When the mutation grid is located in a region with flat terrain or uniform underlying surface, the weights in the spatial weighted smoothing process are determined based on the spatial distance between the mutation grid and adjacent grids. When the mutation grid is located in a region with complex terrain, terrain feature similarity constraints are superimposed on the weights using digital elevation model data, so that the weights decrease as the spatial distance and differences in terrain features increase. When the mutation grid is located in the coastal-land boundary region, the weights are determined independently for the land region and the marine region, with the coastline as the boundary.

17. An electronic device, characterized in that, include: At least one processor and memory; The memory and processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, the historical lightning frequency reconstruction method as described in any one of claims 1 to 8 is implemented.

18. A readable storage medium, characterized in that, It contains an executable program, which, when executed, implements the historical lightning frequency reconstruction method as described in any one of claims 1 to 8.

Images