A high-precision grid-based extreme rainfall probability distribution simulation and prediction method
Patent Information
- Application Number
- CN202611264578.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-20
- Publication Date
- 2026-10-27
AI Technical Summary
[0005]有鉴于此，本发明提供一种高精度网格下极端降雨概率分布模拟预测方法，用于解决现有方法在高精度网格场景下模拟精度低、空间代表性差、未充分融合多源数据、时效性不足等问题
[0014]本发明的有益效果在于：通过将目标区域划分为高精度网格单元，并深度融合网格降水观测数据、地形数据、大气环流数据及植被覆盖数据等多源信息，解决了现有方法在高精度网格场景下模拟精度低、空间代表性差、多源数据融合不充分以及计算时效性不足的问题。
Smart Images
Figure CN122987518A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of extreme rainfall prediction technology, specifically relating to a method for simulating and predicting the probability distribution of extreme rainfall under a high-precision grid. Background Technology
[0002] As the global climate situation becomes increasingly complex, the frequency and intensity of extreme rainfall events continue to rise, posing a serious threat to urban economic and social development, people's well-being, and the ecological environment. Strengthening extreme rainfall forecasting capabilities and improving disaster prevention, mitigation, and relief levels have become urgent needs. Extreme rainfall probability distribution simulation and forecasting are core technical supports for disaster risk assessment, engineering design, and emergency dispatch; their accuracy directly determines the scientific validity and effectiveness of disaster prevention and mitigation measures.
[0003] Currently, methods for simulating and predicting the probability distribution of extreme rainfall are mainly divided into two categories: one is the traditional statistical method based on historical observation data, such as using models like the Generalized Extreme Value Distribution (GEV) and Pearson Type III distribution to fit extreme rainfall data observed at stations, thereby achieving probability prediction; the other is the numerical simulation method, which uses meteorological numerical models to simulate atmospheric circulation processes and indirectly derive the probability distribution characteristics of extreme rainfall. However, existing technologies have many shortcomings in application: First, traditional statistical methods rely on station observation data, and the uneven distribution of stations leads to insufficient representativeness of grid-scale data, especially in areas with complex terrain and sparse stations, making it difficult to accurately characterize the spatial distribution differences of extreme rainfall, resulting in large simulation prediction errors; Second, although numerical simulation methods can achieve grid-scale prediction, they are computationally complex and time-consuming, and are highly dependent on boundary conditions and initial fields, making them prone to extreme value simulation biases and difficult to balance timeliness and accuracy; Third, existing methods often do not fully integrate multi-source covariates (such as topography, vegetation, atmospheric parameters, etc.), failing to effectively capture the spatial autocorrelation and nonlinear influencing factors of extreme rainfall, resulting in insufficient accuracy and robustness of probability distribution simulation.
[0004] Therefore, there is an urgent need for a high-precision grid-based method for simulating and predicting the probability distribution of extreme rainfall to address the problems existing in current technologies. Summary of the Invention
[0005] In view of this, the present invention provides a method for simulating and predicting the probability distribution of extreme rainfall under high-precision grids, which solves the problems of low simulation accuracy, poor spatial representativeness, insufficient fusion of multi-source data, and insufficient timeliness of existing methods in high-precision grid scenarios.
[0006] To achieve the above objectives, this invention provides a method for simulating and predicting the probability distribution of extreme rainfall under a high-precision grid, comprising the following steps: Multi-source data of the target area is acquired and preprocessed, and the grid data of each high-precision grid cell is determined based on the preprocessed multi-source data to obtain a grid dataset; the multi-source data includes: grid precipitation observation data, topographic data, atmospheric circulation data, and vegetation cover data; An extreme rainfall sample set was obtained from the grid dataset; Based on a pre-defined lightweight gradient booster algorithm, the importance of the grid data of each high-precision grid cell is evaluated, and the key grid data of each high-precision grid cell is obtained and a key data matrix is generated. Using extreme rainfall sample set as the dependent variable and grid key data as the independent variable, a GEV model is constructed, and the parameters of the GEV model are estimated to obtain the GEV model for each high-precision grid cell. Determine if there are any GEV models that do not meet the preset conditions; if so, re-estimate the parameters of the GEV models that do not meet the preset conditions until all GEV models meet the preset conditions. If not, calculate the extreme rainfall value of each high-precision grid cell under different return periods according to the GEV model, and generate a spatial distribution map of the extreme rainfall return periods in the target area.
[0007] As an embodiment of the present invention, an extreme rainfall sample set is obtained based on a grid dataset, including: Acquire rainfall intensity data for each high-precision grid cell at different return periods to obtain a rainfall intensity dataset; The rainfall intensity data in the rainfall intensity dataset are sorted, and the rainfall intensity data located after the preset quantile data are selected as extreme rainfall data; An extreme rainfall sample set was obtained based on extreme rainfall data.
[0008] As an embodiment of the present invention, based on a preset lightweight gradient boosting machine algorithm, the importance of the grid data of each high-precision grid cell is evaluated to obtain the key grid data of each high-precision grid cell and generate a key data matrix, including: A model training set is constructed based on the grid data of high-precision grid cells and the corresponding extreme rainfall sample data; An initial LGBM model is constructed and trained using the model training set. After the model training is completed, the importance evaluation index of each parameter in the grid data is calculated. The importance evaluation index includes: gain value, coverage, and frequency value. The importance assessment index is summed to obtain an importance score, and parameters with an importance score higher than the preset score value are selected as key parameters; Obtain key grid data based on key parameters and grid data; A key data matrix is generated based on the grid key data; wherein the dimension of the key data matrix is... , This represents the number of extreme rainfall samples. This represents the number of key parameters.
[0009] As an embodiment of the present invention, a GEV model is constructed using an extreme rainfall sample set as the dependent variable and grid key data as the independent variable. The parameters of the GEV model are estimated to obtain the GEV model for each high-precision grid cell, including: A GEV model was constructed using an extreme rainfall sample set as the dependent variable and grid key data as the independent variable. Based on the maximum likelihood estimation method, the model parameters of the GEV model are solved using grid key data, as shown below: In the formula, Let the objective function of the maximum likelihood estimation method be denoted as . This represents the vector formed by the model parameters. Indicates the first The number of extreme rainfall samples per high-precision grid cell. This indicates the number of high-precision grid cells. No. The first high-precision grid cell Rainfall intensity of extreme rainfall samples, Indicates model parameters, and All of these represent regression coefficients of the location parameter. and All of these represent the regression coefficients of the scaling parameter. Indicates the first The first high-precision grid cell The values of the key parameters, Indicates the number of key parameters. This represents the probability density function of the GEV model.
[0010] The GEV model is obtained based on the model parameters.
[0011] As an embodiment of the present invention, it is determined whether there are GEV models that do not meet the preset conditions; if so, the model parameters of the GEV models that do not meet the preset conditions are re-estimated until all GEV models meet the preset conditions, including: The AIC information value for each GEV model is calculated based on the model parameters, as shown below: In the formula, This represents the AIC information value of the GEV model. This indicates the number of model parameters in the GEV model. This represents the maximum value obtained when solving the objective function; Determine if there is an AIC information value for any GEV model that is greater than a preset information threshold; if so, re-estimate the model parameters of the GEV models that do not meet the preset conditions, until all GEV models meet the preset conditions.
[0012] As an embodiment of the present invention, the extreme rainfall value of each high-precision grid cell under different return periods is calculated according to the GEV model, and a spatial distribution map of the return periods of extreme rainfall is generated, including: The extreme rainfall values for each high-precision grid cell under different return periods were calculated based on the model parameters of the GEV model, as shown below: In the formula, Indicates the first A high-precision grid cell in the recurrence period Extreme rainfall values at that time; Based on the recurrence period of all high-precision grid cells The extreme rainfall values at that time were used to generate a spatial distribution map of the recurrence period of extreme rainfall.
[0013] As an embodiment of the present invention, it further includes: after all GEV models meet the preset conditions, optimizing all GEV models to obtain optimized GEV models, specifically: A grid feature vector is constructed based on the key grid data of each high-precision grid cell. The target region is divided into multiple spatial partitions by clustering based on the grid feature vectors; where the objective function during clustering is... As shown below: In the formula, Indicates the total number of spatial partitions. Indicates the first The first high-precision grid cell (belonging to the first) The grid feature vectors of (spatial partitions), Indicates the first The grid feature vector of the central grid of each spatial partition. Indicates the first The number of high-precision grid cells in each spatial partition; The neighborhood weights between adjacent grids in each spatial partition are determined as follows: In the formula, Indicates the first The high-precision grid cell and its adjacent first Neighborhood weights between high-precision grid cells Indicates the relationship with the first The number of adjacent high-precision grid cells in a high-precision grid cell. ; Based on the spatial smoothing coefficients of each spatial partition of the grid key data, as shown below: In the formula, Indicates the first Spatial smoothing coefficient of each spatial partition Indicates the baseline smoothing coefficient. Indicates the first The standard deviation of the slope values of all high-precision grid cells in each spatial partition. The maximum standard deviation of the slope values across all spatial zones; The optimization objective function is constructed and solved based on the neighborhood weights and spatial smoothing coefficients between adjacent grids in each spatial partition, resulting in the optimization model parameters for each high-precision grid cell in the spatial partition. The optimization objective function is as follows: In the formula, Indicates the first The optimization objective function for each spatial partition. Indicates determining the first The objective function value when the model parameters of a high-precision grid cell are given. , and They represent the first The standard deviation of the model parameters of all high-precision grid cells in each spatial partition. , and All indicate the first Model parameters for each high-precision grid cell; The GEV model of each high-precision grid cell is updated based on the optimized model parameters to obtain the optimized GEV model.
[0014] The beneficial effects of this invention are as follows: by dividing the target area into high-precision grid units and deeply integrating multi-source information such as grid precipitation observation data, topographic data, atmospheric circulation data and vegetation cover data, the problems of low simulation accuracy, poor spatial representativeness, insufficient fusion of multi-source data and insufficient computational timeliness of existing methods in high-precision grid scenarios are solved.
[0015] Other advantages, objectives, and features of the invention will be set forth in the following description and will be apparent to those skilled in the art in some respects, or may be learned by practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description
[0016] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration: Figure 1 This is a schematic diagram of the process of the present invention; Figure 2 This is a schematic diagram of the process for generating the key data matrix according to the present invention. Detailed Implementation
[0017] like Figures 1-2 As shown, this invention provides a method for simulating and predicting the probability distribution of extreme rainfall under a high-precision grid, comprising the following steps: Multi-source data of the target area is acquired and preprocessed, and the grid data of each high-precision grid cell is determined based on the preprocessed multi-source data to obtain a grid dataset; the multi-source data includes: grid precipitation observation data, topographic data, atmospheric circulation data, and vegetation cover data; An extreme rainfall sample set was obtained from the grid dataset; Based on a pre-defined lightweight gradient booster algorithm, the importance of the grid data of each high-precision grid cell is evaluated, and the key grid data of each high-precision grid cell is obtained and a key data matrix is generated. Using extreme rainfall sample set as the dependent variable and grid key data as the independent variable, a GEV model is constructed, and the parameters of the GEV model are estimated to obtain the GEV model for each high-precision grid cell. Determine if there are any GEV models that do not meet the preset conditions; if so, re-estimate the parameters of the GEV models that do not meet the preset conditions until all GEV models meet the preset conditions. If not, calculate the extreme rainfall value of each high-precision grid cell under different return periods according to the GEV model, and generate a spatial distribution map of the extreme rainfall return periods in the target area.
[0018] The working principle of the above technical solution is as follows: This technical solution takes the area to be predicted as the target area, acquires multi-source data of the target area and preprocesses it (such as outlier removal, missing value imputation, and standardization); simultaneously, it divides the target area into m high-precision grid cells; then, it segments the preprocessed data to determine the grid data in each high-precision grid cell, thus forming a grid dataset; the multi-source data includes: grid precipitation observation data (derived from the fusion of ground automatic rain gauges, satellite remote sensing precipitation products, and meteorological radar data), topographic data (including elevation, slope, etc.), atmospheric circulation data (including air pressure, wind speed, etc.), and vegetation cover data (obtained from the NDVI index); the grid data is the data of the corresponding high-precision grid cells after multi-source data preprocessing, such as air pressure, slope, etc.; then, the extreme rainfall sample set is determined through the grid dataset, and simultaneously... The process involves determining the number of extreme rainfall samples for each high-precision grid cell; then, based on a pre-defined lightweight gradient boosting machine algorithm, evaluating the importance of the grid data for each high-precision grid cell to obtain key grid data for each high-precision grid cell and generating a key data matrix; next, constructing a GEV model with the extreme rainfall sample set as the dependent variable and the key grid data as the independent variable, and estimating the parameters of the GEV model to obtain the GEV model for each high-precision grid cell; finally, determining whether there are any GEV models that do not meet the preset conditions; if so, re-estimating the parameters of the GEV models that do not meet the preset conditions using the above parameter estimation method until all GEV models meet the preset conditions; if there are no GEV models that do not meet the preset conditions, calculating the extreme rainfall values for each high-precision grid cell under different return periods based on the cumulative probability distribution function of the GEV model, and generating a spatial distribution map of the return periods of extreme rainfall in the target area. The beneficial effects of the above technical solution are as follows: By dividing the target area into high-precision grid units and deeply integrating multi-source information such as grid precipitation observation data, topographic data, atmospheric circulation data and vegetation cover data, the above technical solution solves the problems of low simulation accuracy, poor spatial representativeness, insufficient fusion of multi-source data and insufficient computational timeliness of existing methods in high-precision grid scenarios.
[0019] In one embodiment, obtaining an extreme rainfall sample set based on a grid dataset includes: Acquire rainfall intensity data for each high-precision grid cell at different return periods to obtain a rainfall intensity dataset; The rainfall intensity data in the rainfall intensity dataset are sorted, and the rainfall intensity data located after the preset quantile data are selected as extreme rainfall data; An extreme rainfall sample set was obtained based on extreme rainfall data.
[0020] The working principle and beneficial effects of the above technical solution are as follows: This technical solution extracts rainfall intensity data corresponding to different return periods from each high-precision grid cell in its historical rainfall time series to form a rainfall intensity dataset; then, the rainfall intensity data in each grid cell are sorted in ascending order, and based on the core idea of extreme events in extreme value theory that they are located at the tail of the distribution, the rainfall intensity data located after the preset quantile (generally 95%) is selected as extreme rainfall data; finally, the extreme rainfall data selected from each grid cell are collected to form an extreme rainfall sample set for subsequent GEV model construction (the samples in the extreme rainfall sample set include grid number, observation period, rainfall intensity, cumulative rainfall during the period, rainfall duration, grid basic geographic information, etc.).
[0021] In one embodiment, based on a preset lightweight gradient boosting machine algorithm, the importance of the grid data of each high-precision grid cell is evaluated to obtain the key grid data of each high-precision grid cell and generate a key data matrix, including: A model training set is constructed based on the grid data of high-precision grid cells and the corresponding extreme rainfall sample data; An initial LGBM model is constructed and trained using the model training set. After the model training is completed, the importance evaluation index of each parameter in the grid data is calculated. The importance evaluation index includes: gain value, coverage, and frequency value. The importance assessment index is summed to obtain an importance score, and parameters with an importance score higher than the preset score value are selected as key parameters; Obtain key grid data based on key parameters and grid data; A key data matrix is generated based on the grid key data; wherein the dimension of the key data matrix is... , This represents the number of extreme rainfall samples. This represents the number of key parameters.
[0022] The working principle and beneficial effects of the above technical solution are as follows: First, for each high-precision grid cell, its corresponding grid data is used as a feature variable, and its corresponding extreme rainfall sample data is used as a target variable to construct a model training set; then, an initial LGBM model is constructed and trained using the model training set. After the model converges, the importance evaluation index of each parameter in the grid data is calculated, including gain value (the amount of loss reduction brought by the feature when splitting tree nodes), coverage (the proportion of samples where the feature is used for splitting), and frequency value (the number of times the feature is used as a splitting node); the above three indices are summed to obtain an importance score, and parameters with scores higher than the preset score value are selected as key parameters; finally, based on the selected key parameters, the corresponding grid key data is extracted from the original grid data, and a dimension is constructed according to the number of extreme rainfall samples and the number of key parameters. The data within the matrix is a key data matrix of grid data after multi-source data preprocessing, providing a concise and highly interpretable input for subsequent GEV model construction.
[0023] This technical solution introduces a lightweight gradient boosting machine algorithm to assess the importance of parameters in grid data, enabling automatic and objective screening of key influencing factors. This solves the bias problem caused by relying on subjective experience or simple correlation analysis to select variables in traditional methods. It uses three indicators—gain, coverage, and frequency—to comprehensively evaluate parameter importance, reflecting the contribution of features to extreme rainfall prediction from different dimensions. This avoids the one-sidedness of single-indicator evaluation and improves the accuracy and robustness of key variable screening. The screened key data matrix significantly reduces data dimensionality and removes redundant information, which not only alleviates the computational burden of subsequent GEV model construction and improves the timeliness of large-scale high-precision grid modeling, but also enhances the model's interpretability and generalization ability by retaining the variables most explanatory to extreme rainfall. This provides accurate and efficient feature inputs for extreme rainfall simulation in different geographical regions and under different underlying surface conditions.
[0024] In one embodiment, a GEV model is constructed using an extreme rainfall sample set as the dependent variable and grid key data as the independent variable. The parameters of the GEV model are then estimated to obtain the GEV model for each high-precision grid cell, including: A GEV model was constructed using an extreme rainfall sample set as the dependent variable and grid key data as the independent variable. Based on the maximum likelihood estimation method, the model parameters of the GEV model are solved using grid key data, as shown below: In the formula, Let the objective function of the maximum likelihood estimation method be denoted as . This represents the vector formed by the model parameters. Indicates the first The number of extreme rainfall samples per high-precision grid cell. This indicates the number of high-precision grid cells. No. The first high-precision grid cell Rainfall intensity of extreme rainfall samples, and All of these represent regression coefficients of the location parameter. and All of these represent the regression coefficients of the scaling parameter. Indicates the first The first high-precision grid cell The values of the key parameters, Indicates the number of key parameters. This represents the probability density function of the GEV model.
[0025] The GEV model is obtained based on the model parameters.
[0026] The working principle and beneficial effects of the above technical solution are as follows: First, the key parameters of each high-precision grid cell are used as covariates. The location parameters and scale parameters of the GEV model are expressed as linear and exponential functions of the key parameters, respectively. The location parameters are expressed in linear form to reflect the direct influence of key factors on the central trend of extreme rainfall, while the scale parameters are expressed in exponential form to ensure their non-negativity and reflect the nonlinear effect of key factors on the variability of extreme rainfall. Then, the objective function of the joint maximum likelihood estimation is constructed. By maximizing the joint likelihood function of all extreme rainfall samples in all grid cells, the regression coefficients of the location parameters, the regression coefficients of the scale parameters, and the shape parameters are solved simultaneously to obtain the GEV model of each high-precision grid cell. The cumulative distribution function and probability density function of the GEV model for each high-precision grid cell are as follows: In the formula, Indicates the first The cumulative distribution function of a GEV model with high-precision grid cells. Indicates the first The probability density function of a GEV model with a high-precision grid cell. Indicates extreme rainfall intensity. Indicates the first Position parameters of a GEV model with high-precision grid cells. Indicates the first Scale parameters of a GEV model with high-precision grid cells Indicates the first Shape parameters of a GEV model with a high-precision mesh element; By incorporating key grid data as covariates into the parameter expression of the GEV model, direct coupling between extreme rainfall extreme value distribution and environmental factors such as local topography, atmospheric circulation, and vegetation cover is achieved. This enables the model to adapt to the local characteristics of different grid cells, significantly improving its ability to characterize the spatial differences in the probability distribution of extreme rainfall. The maximum likelihood estimation method is used to jointly estimate the model parameters of all grid cells, fully utilizing global sample information. Compared to the traditional site-by-site independent fitting method, this yields more robust parameter estimation results even on grid cells with limited sample sizes. The parameterization design, employing linear functions for location parameters and exponential functions for scale parameters, ensures that the non-negativity constraint of the scale parameter satisfies the theoretical requirements of extreme value distribution while also considering the differentiated influence mechanisms of different environmental factors on the center location and fluctuation amplitude of extreme rainfall, enhancing the physical interpretability of the model. Finally, this embodiment constructs a GEV model with spatial consistency and local adaptability for each high-precision grid cell, providing a solid mathematical foundation for subsequent calculations of extreme rainfall values under different return periods.
[0027] In one embodiment, it is determined whether there are GEV models that do not meet the preset conditions; if so, the model parameters of the GEV models that do not meet the preset conditions are re-estimated until all GEV models meet the preset conditions, including: The AIC information value for each GEV model is calculated based on the model parameters, as shown below: In the formula, This represents the AIC information value of the GEV model. This indicates the number of model parameters in the GEV model. This represents the maximum value obtained when solving the objective function; Determine if there is an AIC information value for any GEV model that is greater than a preset information threshold; if so, re-estimate the model parameters of the GEV models that do not meet the preset conditions, until all GEV models meet the preset conditions.
[0028] The working principle and beneficial effects of the above technical solution are as follows: The goodness of fit and complexity of the model are evaluated by calculating the AIC information value of each GEV model, and this is used to determine whether the model meets the preset conditions. Specifically, for each high-precision grid cell GEV model whose model parameters have been estimated, the maximum value of the objective function is estimated based on the number of model parameters and the maximum likelihood value to calculate the AIC information value. Then, it is determined whether the AIC information value of each model is greater than the preset prediction information threshold. If there is a model that does not meet the conditions, the parameters of the model are re-estimated, and the above judgment process is repeated until all grid cell GEV models meet the preset condition that the AIC information value is lower than the threshold. By introducing the AIC information criterion as an objective quantitative indicator of GEV model quality, a unified standard for model selection and evaluation is achieved, avoiding the arbitrariness of judging the quality of models based on subjective experience. By setting an iterative optimization mechanism, parameters of models that do not meet the preset conditions are re-estimated until convergence, effectively ensuring that the GEV model of each high-precision grid cell meets the statistically significant reliability standard, avoiding model failure caused by complex local terrain, abnormal samples, or improper initial parameter values.
[0029] In one embodiment, the extreme rainfall value of each high-precision grid cell under different return periods is calculated based on the GEV model, generating a spatial distribution map of the extreme rainfall return periods, including: The extreme rainfall values for each high-precision grid cell under different return periods were calculated based on the model parameters of the GEV model, as shown below: In the formula, Indicates the first A high-precision grid cell in the recurrence period Extreme rainfall values at that time; Based on the recurrence period of all high-precision grid cells The extreme rainfall values at that time were used to generate a spatial distribution map of the recurrence period of extreme rainfall.
[0030] The working principle and beneficial effects of the above technical solution are as follows: For each high-precision grid cell, the location parameters, scale parameters, and shape parameters of its GEV model are substituted into the calculation formula for extreme rainfall values. When the shape parameter is not zero, the quantile function form of the generalized extreme value distribution is adopted; when the shape parameter is zero, it degenerates into the quantile function form of the Gumbel distribution. Through this piecewise function, the return period T is converted into the corresponding extreme rainfall quantile, thereby obtaining the extreme rainfall value of the grid cell under a specific return period. Finally, the calculation results of all high-precision grid cells are collected and rendered according to the grid spatial location to generate a spatial distribution map of the return period of extreme rainfall.
[0031] In one embodiment, the method further includes: optimizing all GEV models after all GEV models meet preset conditions to obtain optimized GEV models, specifically: Based on the K-Means++ algorithm, a grid feature vector is constructed from the key grid data of each high-precision grid cell. The target region is divided into multiple spatial partitions by clustering based on the grid feature vectors; where the objective function during clustering is... As shown below: In the formula, Indicates the total number of spatial partitions. Indicates the first The first high-precision grid cell (belonging to the first) The grid feature vectors of (spatial partitions), Indicates the first The grid feature vector of the central grid of each spatial partition. Indicates the first The number of high-precision grid cells in each spatial partition; The neighborhood weights between adjacent grids in each spatial partition are determined as follows: In the formula, Indicates the first The high-precision grid cell and its adjacent first Neighborhood weights between high-precision grid cells Indicates the relationship with the first The number of adjacent high-precision grid cells in a high-precision grid cell. ; Based on the spatial smoothing coefficients of each spatial partition of the grid key data, as shown below: In the formula, Indicates the first Spatial smoothing coefficient of each spatial partition Indicates the baseline smoothing coefficient. Indicates the first The standard deviation of the slope values of all high-precision grid cells in each spatial partition. The maximum standard deviation of the slope values across all spatial zones; The optimization objective function is constructed and solved based on the neighborhood weights and spatial smoothing coefficients between adjacent grids in each spatial partition, resulting in the optimization model parameters for each high-precision grid cell in the spatial partition. The optimization objective function is as follows: In the formula, Indicates the first The optimization objective function for each spatial partition. Indicates determining the first The objective function value when the model parameters of a high-precision grid cell are given. , and They represent the first The standard deviation of the model parameters of all high-precision grid cells in each spatial partition. , and All indicate the first Model parameters for each high-precision grid cell; The GEV model of each high-precision grid cell is updated based on the optimized model parameters to obtain the optimized GEV model.
[0032] The working principle of the above technical solution is as follows: This technical solution clusters and partitions the feature vectors constructed from the key grid data (such as vegetation cover, atmospheric circulation factors, elevation, and slope) of each high-precision grid cell, so that grid cells within the same partition have similar geographical backgrounds. Then, spatial neighborhood relationships are independently constructed for each partition, neighborhood weights are determined, and corresponding objective functions are constructed to collaboratively optimize the GEV model parameters (location parameters, scale parameters, and shape parameters) within the spatial partition. The objective function is solved using existing algorithms such as coordinate ascent. The spatial smoothing coefficient specific to each partition is adaptively set according to the standard deviation of the slope within the partition—the coefficient is smaller in areas with complex terrain to preserve local details, and larger in areas with flat terrain to enhance smoothing stability, thereby eliminating sample noise while preserving local heterogeneity. Finally, through the progressive processing of "partition clustering—adaptive optimization within partitions" described above, spatially continuous and physically reasonable GEV model parameters are obtained, providing high-quality input for the subsequent generation of spatial distribution maps of extreme rainfall recurrence periods. The beneficial effects of the above technical solution are as follows: By using clustering, grid cells with similar key data and terrain features are grouped into the same partition, ensuring that grid cells within the same partition have similar terrain, vegetation, and atmospheric circulation backgrounds. Based on this, parameter co-optimization is performed, avoiding the confusion of physical mechanisms that may be caused by cross-partition smoothing, and better reflecting the differentiated response characteristics of different geographical units to extreme rainfall. At the same time, by using clustering to group sparse, high-precision grid cells into partitions with similar physical properties, the spatial neighborhood information of grid cells with sufficient samples within the partition can be used during the optimization process within the partition, effectively mitigating the parameter estimation bias caused by insufficient samples and improving the estimation accuracy of extreme rainfall in sparse data areas.
[0033] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.
Claims
1. A method for simulating and predicting the probability distribution of extreme rainfall under a high-precision grid, characterized in that, Includes the following steps: Multi-source data of the target area is acquired and preprocessed, and the grid data of each high-precision grid cell is determined based on the preprocessed multi-source data to obtain a grid dataset; the multi-source data includes: grid precipitation observation data, topographic data, atmospheric circulation data, and vegetation cover data; An extreme rainfall sample set was obtained from the grid dataset; Based on a pre-defined lightweight gradient booster algorithm, the importance of the grid data of each high-precision grid cell is evaluated, and the key grid data of each high-precision grid cell is obtained and a key data matrix is generated. Using extreme rainfall sample set as the dependent variable and grid key data as the independent variable, a GEV model is constructed, and the parameters of the GEV model are estimated to obtain the GEV model for each high-precision grid cell. Determine if there are any GEV models that do not meet the preset conditions; if so, re-estimate the parameters of the GEV models that do not meet the preset conditions until all GEV models meet the preset conditions. If not, calculate the extreme rainfall value of each high-precision grid cell under different return periods according to the GEV model, and generate a spatial distribution map of the extreme rainfall return periods in the target area.
2. The method for simulating and predicting the probability distribution of extreme rainfall under a high-precision grid according to claim 1, characterized in that, The extreme rainfall sample set was obtained from the grid dataset, including: Acquire rainfall intensity data for each high-precision grid cell at different return periods to obtain a rainfall intensity dataset; The rainfall intensity data in the rainfall intensity dataset are sorted, and the rainfall intensity data located after the preset quantile data are selected as extreme rainfall data; An extreme rainfall sample set was obtained based on extreme rainfall data.
3. The method for simulating and predicting the probability distribution of extreme rainfall under a high-precision grid according to claim 1, characterized in that, Based on a pre-defined lightweight gradient boosting machine algorithm, the importance of the grid data in each high-precision grid cell is evaluated, resulting in key grid data for each high-precision grid cell and generating a key data matrix, including: A model training set is constructed based on the grid data of high-precision grid cells and the corresponding extreme rainfall sample data; An initial LGBM model is constructed and trained using the model training set. After the model training is completed, the importance evaluation index of each parameter in the grid data is calculated. The importance evaluation index includes: gain value, coverage, and frequency value. The importance assessment index is summed to obtain an importance score, and parameters with an importance score higher than the preset score value are selected as key parameters; Obtain key grid data based on key parameters and grid data; A key data matrix is generated based on the grid key data; wherein the dimension of the key data matrix is... , This represents the number of extreme rainfall samples. This represents the number of key parameters.
4. The method for simulating and predicting the probability distribution of extreme rainfall under a high-precision grid according to claim 1, characterized in that, A GEV model is constructed using an extreme rainfall sample set as the dependent variable and grid key data as the independent variable. The parameters of the GEV model are estimated to obtain the GEV model for each high-precision grid cell, including: A GEV model was constructed using an extreme rainfall sample set as the dependent variable and grid key data as the independent variable. Based on the maximum likelihood estimation method, the model parameters of the GEV model are solved using grid key data, as shown below: In the formula, Let the objective function of the maximum likelihood estimation method be denoted as . This represents the vector formed by the model parameters. Indicates the first The number of extreme rainfall samples per high-precision grid cell. This indicates the number of high-precision grid cells. No. The first high-precision grid cell Rainfall intensity of extreme rainfall samples, and All of these represent regression coefficients of the location parameters. and All of these represent the regression coefficients of the scaling parameter. Indicates the first The first high-precision grid cell The values of the key parameters, Indicates the number of key parameters. This represents the probability density function of the GEV model. The GEV model is obtained based on the model parameters.
5. The method for simulating and predicting the probability distribution of extreme rainfall under a high-precision grid according to claim 1, characterized in that, Determine if any GEV models do not meet the preset conditions; if so, re-estimate the model parameters of the GEV models that do not meet the preset conditions until all GEV models meet the preset conditions, including: The AIC information value for each GEV model is calculated based on the model parameters, as shown below: In the formula, This represents the AIC information value of the GEV model. This indicates the number of model parameters in the GEV model. This represents the maximum value obtained when solving the objective function; Determine if there is an AIC information value for any GEV model that is greater than a preset information threshold; if so, re-estimate the model parameters of the GEV models that do not meet the preset conditions, until all GEV models meet the preset conditions.
6. The method for simulating and predicting the probability distribution of extreme rainfall under a high-precision grid according to claim 1, characterized in that, Extreme rainfall values for each high-precision grid cell under different return periods are calculated based on the GEV model, generating a spatial distribution map of extreme rainfall return periods, including: The extreme rainfall values for each high-precision grid cell under different return periods were calculated based on the model parameters of the GEV model, as shown below: In the formula, Indicates the first A high-precision grid cell in the recurrence period Extreme rainfall values at that time; Based on the recurrence period of all high-precision grid cells The extreme rainfall values at that time were used to generate a spatial distribution map of the recurrence period of extreme rainfall.
7. The method for simulating and predicting the probability distribution of extreme rainfall under a high-precision grid according to claim 4, characterized in that, This also includes: once all GEV models meet the preset conditions, optimizing all GEV models to obtain optimized GEV models, specifically: A grid feature vector is constructed based on the key grid data of each high-precision grid cell. The target region is divided into multiple spatial partitions by clustering based on the grid feature vectors; where the objective function during clustering is... As shown below: In the formula, Indicates the total number of spatial partitions. Indicates the first The first high-precision grid cell (belonging to the first) The grid feature vectors of (spatial partitions), Indicates the first The grid feature vector of the central grid of each spatial partition. Indicates the first The number of high-precision grid cells in each spatial partition; The neighborhood weights between adjacent grids in each spatial partition are determined as follows: In the formula, Indicates the first The high-precision grid cell and its adjacent first Neighborhood weights between high-precision grid cells Indicates the relationship with the first The number of adjacent high-precision grid cells in a high-precision grid cell. ; Based on the spatial smoothing coefficients of each spatial partition of the grid key data, as shown below: In the formula, Indicates the first Spatial smoothing coefficient of each spatial partition Represents the baseline smoothing coefficient. Indicates the first The standard deviation of the slope values of all high-precision grid cells in each spatial partition. The maximum standard deviation of the slope values across all spatial zones; The optimization objective function is constructed and solved based on the neighborhood weights and spatial smoothing coefficients between adjacent grids in each spatial partition, resulting in the optimization model parameters for each high-precision grid cell in the spatial partition. The optimization objective function is as follows: In the formula, Indicates the first The optimization objective function for each spatial partition. Indicates determining the first The objective function value when the model parameters of a high-precision grid cell are given. , and They represent the first The standard deviation of the model parameters of all high-precision grid cells in each spatial partition. , and All indicate the first Model parameters for each high-precision grid cell; The GEV model of each high-precision grid cell is updated based on the optimized model parameters to obtain the optimized GEV model.