A method to study composite flood hazards in data-deficient estuaries considering climate change

Through multi-source data fusion and dynamic modeling technology, the evaluation problems of non-stationarity and multivariate dependence structure of hydrological elements in climate change are solved, and the scientificity and reliability of compound flood risk assessment is improved, providing more for the estuary area Strong flood protection adaptability.

CN118760899BActive Publication Date: 2024-11-15NANJING HYDRAULIC RES INST
View PDF 8 Cites 0 Cited by

Patent Information

Application Number
CN202411254161.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-09
Publication Date
2024-11-15
Estimated Expiration
2044-09-09

AI Technical Summary

Technical Problem

The prior art is difficult to accurately characterize the non-stationarity of hydrological elements caused by climate change and the dynamic evolution of multivariable dependent structures in composite flood disasters, especially in areas with data shortage of data, which makes it difficult to accurately evaluate composite flood risk.

Method used

Multi-source heterogeneous data fusion, wind-rain-wave-flow coupled mathematical model, Bayesian non-parametric method and time-varying Copula model are used to generate joint hydrological elements through data fusion, extreme value recognition and variational inference methods. Test the distribution, conduct uncertainty analysis to assess compound flood risk.

Benefits of technology

It significantly improves the scientificity and reliability of compound flood risk assessment, enhances the flood control capacity and climate adaptability of estuary areas, can more accurately capture the impact of climate change and multivariate dependencies, and provides scientific basis for flood control and disaster reduction decisions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118760899B_ABST
    Figure CN118760899B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for studying compound flood disasters in data-scarce estuaries considering climate change, comprising the following steps: obtaining and preprocessing multi-source heterogeneous data; constructing a wind-rain-wave-current coupling mathematical model; extracting extreme value data pairs; estimating marginal distribution parameters; estimating Copula function parameters; generating joint posterior distribution; calculating joint return period and performing uncertainty analysis. This method effectively solves the difficult problem of studying compound flood disasters in data-scarce estuary areas through multi-source data fusion, dynamic model construction and non-stationarity processing. The Bayesian non-parametric method and time-varying Copula model are used to accurately describe the non-stationary characteristics of hydrological elements and their complex dependencies, significantly improving the scientificity and reliability of compound flood risk assessment, and providing strong support for flood prevention and disaster reduction decisions under the background of climate change.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to hydrological and hydrodynamic technology, in particular to a method for studying composite flood disasters in data-deficient estuaries taking into account climate change. Background Art

[0002] As global climate change intensifies, the frequency and intensity of extreme weather events have increased significantly, and the risk of compound flood disasters faced by estuaries is becoming increasingly severe. As the frontier of land-sea interaction, estuaries are affected by multiple dynamic factors such as runoff, rainfall, astronomical tides and storm surges, and are one of the most active areas for flood disasters. Under the background of climate change, factors such as rising sea levels and changes in rainfall patterns have further exacerbated the vulnerability of estuaries. Compound flood disasters are caused by the combined action of multiple hydrological factors, and their complexity and destructiveness far exceed those caused by a single factor.

[0003] At present, some progress has been made in the study of compound flood disasters. In terms of data acquisition, the application of remote sensing technology and reanalysis data has broadened the data sources; in terms of hydrodynamic simulation, the development of coupling models has improved the simulation accuracy; in terms of statistical analysis, the introduction of Copula functions has improved the modeling ability of multivariate joint distributions. However, in the face of new challenges brought by climate change, existing research still has many shortcomings. Traditional methods often assume that hydrological processes are stationary, which makes it difficult to accurately characterize the non-stationarity caused by climate change; most studies only focus on static dependency structures and ignore the time-varying characteristics of dependency relationships; in areas with data shortages, model parameter estimation and uncertainty quantification still face difficulties.

[0004] Specifically, existing research has the following technical problems. Traditional marginal distribution estimation methods (such as extreme value distribution theory) are difficult to effectively capture the non-stationarity of hydrological elements caused by climate change, especially when the distribution morphology and extreme value behavior change significantly. The Copula model cannot accurately describe the dynamic evolution of the dependency structure between multiple hydrological elements, especially in the context of changes in the mechanism of extreme events. In areas with data shortages, traditional parameter estimation methods (such as maximum likelihood estimation) are easily affected by insufficient sample size, resulting in unstable and unreliable estimation results. In addition, the existing hydrodynamic simulation methods are not accurate enough when dealing with complex terrain and extreme events, and it is difficult to accurately reproduce historical extreme events. Traditional uncertainty analysis methods are difficult to fully quantify and decompose the impact of multiple sources of uncertainty (such as model structure, parameter estimation, climate scenarios, etc.) on the final risk assessment results. These technical problems seriously restrict the accuracy and reliability of compound flood disaster risk assessment, and new methods need to be constructed to solve them. Summary of the invention

[0005] The purpose of the invention is to provide a method for studying complex flood disasters in data-deficient estuaries taking into account climate change, in order to solve at least one of the above technical problems.

[0006] Technical solution, a method for studying composite flood hazards in data-deficient estuaries considering climate change, includes the following steps:

[0007] Step S1, obtaining multi-source heterogeneous data, and preprocessing them using data fusion and standardization methods to obtain a comprehensive data set for subsequent analysis;

[0008] Among them, multi-source heterogeneous data include geographic information data, hydrological and meteorological data, and ocean dynamics data; geographic information data include global coastline remote sensing data, seafloor topography data, and nearshore water depth data; hydrological and meteorological data include runoff data from the upper reaches of the estuary, ERA5 hourly reanalysis wind speed data set, optimal typhoon path data, and hourly rainfall data from meteorological stations near the study area; ocean dynamics data include hourly tide and wave observation data, as well as sea level rise data;

[0009] Step S2, based on the comprehensive data set, construct and use the wind-rain-wave-current coupling mathematical model to calculate the long-term series hydrological dynamic elements of the study area;

[0010] Step S3, based on the long-term series hydrological dynamic data and rainfall data, using an improved extreme value identification method, extracting extreme value data pairs for compound flood disaster research, and constructing an annual extreme value data pair matrix based on the extreme value data pairs;

[0011] Step S4, based on the annual extreme value data pair matrix, using the Bayesian non-parametric method to estimate the marginal distribution parameters of rainfall, tide level and wave, and obtain the optimal marginal distribution model considering non-stationarity;

[0012] Step S5, using the annual extreme value data pair matrix and the marginal distribution model, estimating the parameters of the Copula function describing the multivariate dependency structure, and obtaining the optimal Copula model that can capture the dynamic dependency relationship;

[0013] Step S6: Based on the marginal distribution model and the Copula model, a variational inference method is used to generate a joint posterior distribution of hydrological elements, and a posterior distribution sample set of a predetermined size for uncertainty analysis is obtained;

[0014] Step S7: Calculate the joint return period under different scenarios using the posterior distribution sample set, and conduct a comprehensive uncertainty analysis to obtain the final composite flood risk assessment result.

[0015] Beneficial effect: Through multi-source data fusion, dynamic model construction, non-stationarity processing and comprehensive risk assessment, the multiple difficulties in the study of compound flood disasters in estuary areas with limited data have been effectively overcome. It not only significantly improves the scientificity and reliability of compound flood risk assessment, but also provides strong support for flood prevention and disaster reduction decisions in the context of climate change adaptation. It can enhance the flood control capacity and climate adaptability of estuary areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flow chart of the present invention.

[0017] Figure 2 It is a flow chart of step S1 of the present invention.

[0018] Figure 3 It is a flow chart of step S2 of the present invention.

[0019] Figure 4 It is a flow chart of step S3 of the present invention.

[0020] Figure 5 It is a flow chart of step S4 of the present invention.

[0021] Figure 6 It is a flow chart of step S5 of the present invention.

[0022] Figure 7 It is a flow chart of step S6 of the present invention.

[0023] Figure 8 It is a flow chart of step S7 of the present invention. DETAILED DESCRIPTION

[0024] like Figure 1 As shown in the figure, a method for studying composite flood hazards in data-deficient estuaries considering climate change includes the following steps:

[0025] Step S1, obtaining multi-source heterogeneous data, and preprocessing them using data fusion and standardization methods to obtain a comprehensive data set for subsequent analysis;

[0026] Among them, multi-source heterogeneous data include geographic information data, hydrological and meteorological data, and ocean dynamics data; geographic information data include global coastline remote sensing data, seafloor topography data, and nearshore water depth data; hydrological and meteorological data include runoff data from the upper reaches of the estuary, ERA5 hourly reanalysis wind speed data set, optimal typhoon path data, and hourly rainfall data from meteorological stations near the study area; ocean dynamics data include hourly tide and wave observation data, as well as sea level rise data;

[0027] Step S2, based on the comprehensive data set, construct and use the wind-rain-wave-current coupling mathematical model to calculate the long-term series hydrological dynamic elements of the study area;

[0028] Step S3, based on the long-term series hydrological dynamic data and rainfall data, using an improved extreme value identification method, extracting extreme value data pairs for compound flood disaster research, and constructing an annual extreme value data pair matrix based on the extreme value data pairs;

[0029] Step S4, based on the annual extreme value data pair matrix, using the Bayesian non-parametric method to estimate the marginal distribution parameters of rainfall, tide level and wave, and obtain the optimal marginal distribution model considering non-stationarity;

[0030] Step S5, using the annual extreme value data pair matrix and the marginal distribution model, estimating the parameters of the Copula function describing the multivariate dependency structure, and obtaining the optimal Copula model that can capture the dynamic dependency relationship;

[0031] Step S6: Based on the marginal distribution model and the Copula model, a variational inference method is used to generate a joint posterior distribution of hydrological elements, and a posterior distribution sample set of a predetermined size for uncertainty analysis is obtained;

[0032] Step S7: Calculate the joint return period under different scenarios using the posterior distribution sample set, and conduct a comprehensive uncertainty analysis to obtain the final composite flood risk assessment result.

[0033] In this embodiment, the problem of data scarcity in data-restricted areas is overcome by acquiring and fusing multi-source heterogeneous data. By integrating global remote sensing data, reanalysis data, and measured data, a comprehensive data set with high temporal and spatial resolution is constructed, providing a solid data foundation for subsequent analysis. This data fusion method not only improves the reliability and representativeness of the data, but also significantly enhances the accuracy of the model input.

[0034] By constructing a wind-rain-wave-current coupling mathematical model, combined with adaptive grid technology, parallel computing and data assimilation methods, a high-precision simulation of the complex hydrodynamic processes of the estuary is achieved. This comprehensive model not only overcomes the limitations of traditional single models, but also can better simulate extreme events and long-term change trends, providing reliable long-term series hydrological dynamic data for areas with limited data.

[0035] In terms of extreme value identification, multi-scale decomposition, dynamic time warping and sliding window technology are used to improve the accuracy of compound extreme value event identification. This multi-angle extreme value identification method can capture the complex characteristics of compound extreme value events and provide key data support for subsequent risk analysis.

[0036] In order to solve the non-stationary problem caused by climate change, the Bayesian non-parametric method and time-varying Copula model were used. By introducing time as a covariate, a flexible model that can capture non-stationarity was constructed, accurately describing the probability distribution characteristics of hydrological elements and their changes over time. This dynamic modeling method significantly improves the model's ability to adapt to the impact of climate change.

[0037] In terms of uncertainty quantification, this method uses variational inference technology to generate the joint posterior distribution of hydrological elements, effectively solving the bottleneck of computational efficiency in high-dimensional problems. Through adaptive importance sampling, the estimation accuracy of rare events is improved, providing a reliable uncertainty analysis framework for comprehensive assessment of compound flood risks.

[0038] A composite flood risk assessment considering the impact of climate change has been achieved. The use of Latin hypercube sampling, time-varying joint return period estimation and multi-criteria decision analysis not only improves the comprehensiveness and accuracy of risk assessment, but also provides a scientific basis for flood prevention and disaster reduction decision-making.

[0039] like Figure 2 As shown, according to one aspect of the present application, step S1 is specifically:

[0040] Step S11, obtaining geographic information data, including: obtaining the latest global coastline remote sensing data, obtaining seabed topography data of a large range of sea areas from GEBCO, and extracting nearshore water depth point data from nautical charts; using spatial data processing methods to unify coordinates and resolutions to obtain a geographic information data set for the study area;

[0041] Step S12, obtaining hydrological and meteorological data, including runoff data of the upper reaches of the estuary in the study area, ERA5 hourly reanalysis wind speed data set, optimal typhoon path data, and hourly rainfall data of all meteorological stations within M kilometers near the study area in the past N years; using time series interpolation and spatial interpolation methods to perform spatiotemporal consistency processing to obtain a long-term hydrological and meteorological data set covering the study area; M and N are predetermined natural numbers;

[0042] Step S13, obtaining ocean dynamic data, including hourly tide level and wave observation data of the estuary area, and sea level rise data; using data quality control and outlier detection methods to clean and correct, to obtain a reliable ocean dynamic data set;

[0043] Step S14: align the geographic information dataset, the long-term hydrological and meteorological dataset, and the ocean dynamics dataset in time and space and unify their formats, and use a data fusion algorithm module to integrate data from different sources to obtain a unified, multi-dimensional comprehensive dataset.

[0044] In step S12, the process of performing time-space consistency processing using time series interpolation and space interpolation methods is as follows:

[0045] Step S121, receiving multi-source hydrological and meteorological data, including runoff data from the upper reaches of the estuary in the study area, ERA5 hourly reanalysis wind speed data set, optimal typhoon path data, and hourly rainfall data from meteorological stations near the study area; preprocessing the received data, including removing outliers and standardizing, to obtain a preprocessed hydrological and meteorological data set.

[0046] Step S122: Apply discrete wavelet transform to each time series in the preprocessed hydrological and meteorological data set to decompose it into approximate coefficients and detail coefficients of multiple scales; construct a multi-scale feature matrix based on the decomposition results for subsequent spatiotemporal interpolation processing.

[0047] Step S123: Use a dynamic time warping algorithm to align the time series of different sites in the multi-scale feature matrix; based on the alignment results, generate a time correspondence matrix to eliminate the time offset and scale differences between different data sources.

[0048] Step S124: construct an adaptive kernel function based on the time correspondence matrix, which considers both the time and space dimensions; use the constructed kernel function to construct an initial interpolation model for subsequent parameter optimization.

[0049] Step S125: adopting a particle swarm optimization algorithm to dynamically adjust the bandwidth parameters in the initial interpolation model; according to the optimization result, an optimal bandwidth parameter set is obtained to improve the adaptability and accuracy of the interpolation model.

[0050] Step S126, applying the optimal bandwidth parameter set to the adaptive kernel function, performing interpolation processing on the multi-scale feature matrix; applying inverse discrete wavelet transform to the interpolation result, reconstructing the complete time series, and obtaining a preliminary interpolation result.

[0051] Step S127: construct a local anomaly detection algorithm based on Mahalanobis distance, scan the preliminary interpolation results, identify potential outliers, and use local median filtering to correct them to obtain corrected interpolation results.

[0052] Step S128: construct a spatiotemporal consistency constraint function, which comprehensively considers the time gradient, space gradient and curvature; apply the constraint function to the corrected interpolation result to obtain interpolation data that satisfies the spatiotemporal consistency.

[0053] Step S129: construct an adaptive fusion strategy to dynamically adjust the weights of different interpolation methods according to the local characteristics of the data; apply the fusion strategy to the interpolation data that meets the spatiotemporal consistency to obtain the final spatiotemporal consistent hydrological and meteorological data set.

[0054] In this embodiment, by acquiring global shoreline remote sensing data, seabed topography data and nearshore water depth point data, a high-precision estuary three-dimensional terrain model is constructed, providing accurate geographical boundary conditions for subsequent hydrodynamic simulation. By integrating ERA5 reanalysis data, optimal typhoon path data and meteorological station observation data, a high-resolution meteorological driving field of a long time series is formed, which effectively captures the spatiotemporal variation characteristics of extreme weather events. By collecting tide level, wave observation data and sea level rise data, a comprehensive data set reflecting ocean dynamic processes and long-term variation trends is constructed. The data fusion algorithm module is used to integrate data from different sources, overcoming the problems of single data source and inconsistent spatiotemporal resolution in traditional methods. This multi-source data fusion method not only improves the spatiotemporal continuity and consistency of data, but also enhances the reliability and representativeness of data. In this way, this step provides an effective data acquisition and processing solution for data-limited areas, lays a solid data foundation for subsequent composite flood simulation and risk assessment, and significantly improves the scientificity and reliability of the research.

[0055] like Figure 3 As shown, according to one aspect of the present application, step S2 is specifically:

[0056] Step S21, constructing a three-dimensional numerical grid of the study area using geographic information data; using an adaptive grid refinement method to adjust the grid density according to the regional terrain conditions to obtain an optimized calculation grid;

[0057] Step S22, based on the open source three-dimensional finite volume method ocean numerical model FVCOM, combined with the computational grid, the basic hydrodynamic model and boundary conditions of the study area are established, and the model performance is optimized by using parallel computing technology; the boundary conditions include using tidal harmonic constants as the outer sea boundary conditions, using measured runoff as the upstream river boundary conditions, and using ERA5 wind pressure data and measured rainfall data as surface forcing;

[0058] Step S23: inputting the hydrological and meteorological data and the ocean dynamic data into the basic hydrodynamic model to obtain the model output result, using the data assimilation technology to fuse the model output result with the observation data, constructing a preliminary wind-rain-wave-current coupling model, and obtaining preliminary long-term series hydrological and dynamic data; wherein the long-term series hydrological and dynamic data include water level and flow velocity;

[0059] Step S24, integrating the preliminary wind-rain-wave-current coupling model with the preconfigured deep learning model, and dynamically adjusting the weights of the physical model and the data-driven model using a model hybrid strategy; evaluating the model performance through a cross-validation method, obtaining an optimized wind-rain-wave-current coupling mathematical model, and outputting an improved long-term series hydrological dynamic simulation result;

[0060] Step S25: Use the optimized wind-rain-wave-current coupling mathematical model to perform batch numerical simulations on the hydrological dynamic process in the study area over the past N years, including simulating storm surges and typhoon waves during typhoons; use distributed computing technology to accelerate large-scale computing, obtain and output hourly water level data and effective wave height data for the study area over N years as key long-term series hydrological dynamic elements;

[0061] Step S26: Use a preconfigured superposition method to superimpose the hourly water level data with the sea level rise data to consider the impact of sea level rise on the trend change of the water level; use time series decomposition technology to separate the trend term, obtain and output the corrected water level data that considers the impact of climate change as the final long-term series hydrological dynamic element.

[0062] In this embodiment, in step S21, the grid density is adjusted according to the regional terrain conditions to obtain an optimized calculation grid, specifically:

[0063] Step S211, receiving high-resolution digital elevation model (DEM) data and an initial coarse grid of the study area; applying multi-scale morphological analysis to the DEM data, including slope calculation, curvature analysis and geomorphic unit identification, to obtain a set of terrain characteristic indicators.

[0064] Step S212: construct a terrain complexity evaluation function based on the terrain feature index set; apply the function to each unit of the initial coarse grid, calculate the terrain complexity score, and obtain an initial grid complexity distribution map.

[0065] Step S213: Analyze the initial grid complexity distribution map using a pre-trained convolutional neural network; identify key areas that need to be refined, and generate a grid refinement suggestion map to guide the subsequent grid optimization process.

[0066] Step S214: Based on the grid refinement suggestion graph, an adaptive quadtree / octree subdivision algorithm is constructed; the initial coarse grid is locally refined to generate a multi-resolution grid structure to obtain a preliminarily optimized computational grid.

[0067] Step S215: Use a simplified shallow water equation model to perform rapid flow field simulation on the preliminarily optimized computational grid; analyze the simulation results, identify areas where the flow field changes dramatically, and generate a flow field sensitivity map.

[0068] Step S216: Fusing the flow field sensitivity map with the grid refinement suggestion map to obtain a comprehensive optimization index; based on the comprehensive optimization index, further adjusting the grid density to generate a secondary optimized computational grid.

[0069] Step S217, construct a mesh quality assessment function, comprehensively consider the mesh orthogonality, smoothness and aspect ratio; apply the quality assessment function to the secondary optimized computational mesh, identify low-quality mesh cells, and obtain a mesh quality distribution map.

[0070] Step S218: Develop a local mesh adjustment algorithm based on the mesh quality distribution map; perform local optimization on low-quality mesh elements, including operations such as node movement, edge exchange, and face splitting, to obtain a computational mesh with improved quality.

[0071] Step S219: construct a grid optimization objective function that balances the computational accuracy requirements and computational resource constraints; apply a genetic algorithm to globally optimize the computational grid with improved quality to obtain a final optimized computational grid.

[0072] Step S2110: comprehensively evaluate the final optimized computational grid, including calculating grid quality indicators, estimating computing resource requirements, and predicting numerical stability; based on the evaluation results, generate a grid optimization report to guide the subsequent hydrodynamic simulation process.

[0073] Step S2111: Compare and analyze the final optimized computational grid with the original terrain data; calculate the matching degree between the grid resolution and the terrain complexity, and generate a grid adaptability evaluation report.

[0074] In this embodiment, an adaptive grid refinement method is used to construct a three-dimensional numerical grid, which improves the calculation accuracy of complex terrain areas and optimizes the allocation of computing resources. Based on the open source FVCOM model, combined with parallel computing technology, the efficiency of large-scale hydrodynamic simulation is significantly improved. The model output and observation data are integrated through data assimilation technology, which not only improves the prediction accuracy of the model, but also effectively reduces the system error of the model. By integrating the deep learning model, the dynamic weight adjustment of the physical model and the data-driven model is realized, which not only ensures the physical interpretability of the model, but also improves the model's ability to capture complex nonlinear processes. By considering the impact of sea level rise, the model can reflect the long-term impact of climate change on hydrodynamic processes. This comprehensive approach not only overcomes the limitations of traditional single models, but also can better simulate extreme events and long-term change trends. In this way, this step provides an efficient and accurate hydrodynamic simulation method for estuary areas with limited data, which can generate long-term series of hydrological and dynamic element data, provide reliable data support for subsequent extreme value analysis and risk assessment, and significantly improve the scientificity and accuracy of compound flood disaster research.

[0075] like Figure 4 As shown, according to one aspect of the present application, step S3 is specifically:

[0076] Step S31, spatially process the obtained rainfall data using the inverse distance weighted interpolation method, spatially average the rainfall data of all meteorological stations within Q kilometers near the study area, and obtain a representative regional rainfall time series;

[0077] Step S32: using the variational mode decomposition method to perform multi-scale decomposition on the regional rainfall time series; using the adaptive threshold method to determine the extreme values ​​of each scale, and obtaining a multi-scale rainfall extreme value candidate set;

[0078] Step S33: applying the empirical mode decomposition method to the water level data and the effective wave height data to extract the oscillation modes of different frequencies; using the envelope method to identify the extreme points of each mode to obtain the extreme value candidate set of the water level and the wave height;

[0079] Step S34: input the rainfall extreme value candidate set and the water level and wave height extreme value candidate set into the dynamic time warping algorithm module to calculate the time correlation between extreme value events of different hydrological elements; adopt a hierarchical clustering method to classify extreme value events based on DTW distance to obtain a preliminary composite extreme value event set;

[0080] Step S35, applying a sliding window technique to the composite extreme value event set obtained in step S34, taking a hour as the window length, and calculating the cumulative effect; using a peak detection algorithm to identify the maximum value in the window, and obtaining a screened extreme value event set considering the time cumulative effect; a is a predetermined natural number;

[0081] Step S36: Based on the screened extreme value event set, the conditional probability method is used to calculate the probability of other elements having extreme values ​​within a specified time window when an extreme value of a given hydrological element occurs; the threshold is set using multivariate extreme value theory to screen out the final extreme value data pairs that meet the definition of the composite event;

[0082] Step S37, organize the final extreme value data pairs by year to form an annual extreme value data matrix; use data interpolation technology to process possible missing values ​​to obtain a complete N-row and 3-column annual extreme value data pair matrix; the hydrological elements of the 3 columns are rainfall, tide level and wave respectively.

[0083] In this embodiment, in step S32, an adaptive threshold method is used to determine the extreme values ​​of each scale, specifically:

[0084] Step S321, receiving multi-scale rainfall time series data after variational mode decomposition; applying continuous wavelet transform to the time series of each scale to obtain a time-frequency feature map for subsequent feature extraction and extreme value identification.

[0085] Step S322: Based on the time-frequency feature graph, a multidimensional feature extractor is designed; the extractor comprehensively considers local energy, instantaneous frequency and phase information, extracts feature vectors from the time-frequency feature graph of each scale, and obtains a multi-scale feature matrix.

[0086] Step S323: Use the pre-trained support vector machine model to preliminarily classify the multi-scale feature matrix; identify potential extreme value intervals and generate extreme value candidate sets to provide a basis for subsequent refined identification.

[0087] Step S324: construct an adaptive kernel density estimation model based on the extreme value candidate set; the model uses a variable width kernel function, which can automatically adjust the bandwidth according to the local data density to obtain a probability density function estimate for each scale.

[0088] Step S325: Apply the information entropy criterion to analyze the probability density function of each scale; calculate the local entropy value, identify the entropy value mutation point, and generate a preliminary extreme value threshold set.

[0089] Step S326, design a multi-scale threshold optimization algorithm, which is based on the principle of dynamic programming; taking the preliminary extreme value threshold set as input, balancing the consistency of thresholds between different scales through an iterative optimization process, and obtaining an optimized multi-scale threshold set.

[0090] Step S327: Use the optimized multi-scale threshold set to identify extreme values ​​of the original multi-scale time series; apply the peak detection algorithm to determine the extreme points on each scale and generate a multi-scale extreme point set.

[0091] Step S328: construct an extreme value importance scoring function, which comprehensively considers the amplitude, duration and frequency characteristics of the extreme value; and scores each extreme value point in the multi-scale extreme value point set to obtain an extreme value importance ranking list.

[0092] Step S329: design a multi-scale extreme value fusion algorithm based on a weighted voting mechanism; use the extreme value importance ranking list as weights to fuse extreme values ​​of different scales to obtain a final comprehensive extreme value set.

[0093] In this embodiment, the rainfall data is spatially processed by the inverse distance weighted interpolation method, which overcomes the problem of insufficient spatial representativeness caused by sparse sites. The variational mode decomposition method is used to decompose the rainfall time series at multiple scales, effectively separate extreme events of different time scales, and improve the accuracy of extreme value identification. The water level and wave height data are processed by the empirical mode decomposition method, which can capture oscillation modes of different frequencies and help identify composite extreme value events. The dynamic time warping algorithm is used to calculate the time correlation between extreme value events of different hydrological elements, which overcomes the defect of ignoring the temporal characteristics of events in traditional methods. Through the sliding window technology and peak detection algorithm, the cumulative effect of extreme value events is considered, and the accuracy of composite event identification is improved. This multi-step, multi-angle extreme value identification method can not only effectively deal with data scarcity, but also capture the complex characteristics of composite extreme value events. In this way, the step provides an advanced composite extreme value event identification method for estuary areas with limited data, which can more accurately extract key data for risk analysis and significantly improve the reliability and representativeness of subsequent statistical analysis and risk assessment.

[0094] like Figure 5 As shown, according to one aspect of the present application, step S4 is specifically:

[0095] Step S41, applying the trend test method to each hydrological element in the annual extreme value data matrix, and using the sliding window technology to calculate the statistical characteristics of each subsequence to obtain a preliminary non-stationary evaluation result;

[0096] Step S42: Based on the preliminary non-stationarity assessment results, construct a prior distribution of the Dirichlet process mixture model; adopt a hierarchical Bayesian framework to introduce time as a covariate into the Dirichlet process mixture model to obtain a Ling Bayesian non-parametric model that can capture non-stationarity;

[0097] Step S43: use the Gibbs sampling algorithm to estimate the parameters of the constructed Dirichlet process mixture model; use the adaptive Markov chain Monte Carlo method to improve the sampling efficiency and obtain the posterior distribution samples of the model parameters;

[0098] Step S44, based on the obtained distribution posterior samples, calculate the prediction distribution of the Dirichlet process mixture model; use the cross-validation method to evaluate the goodness of fit and prediction ability of the model to obtain the optimal non-parametric marginal distribution model;

[0099] Step S45, approximating the non-parametric marginal distribution model to a parameterized distribution; using a moment matching method or a maximum likelihood estimation method to fit candidate parameter distributions including a generalized extreme value distribution and a generalized Pareto distribution, to obtain a parameterized marginal distribution of each hydrological element;

[0100] Step S46, using the AIC or BIC information criterion to select a model for the candidate parameter distribution; using the bootstrap method to evaluate the uncertainty of the model selection, and obtaining the optimal marginal distribution type and parameter estimation value of each hydrological element;

[0101] Step S47: Based on the optimal marginal distribution type and parameter estimates of each hydrological element, a time-varying parameter model is constructed to allow the distribution parameters to change over time; a generalized additive model is used to fit the time trend of the parameters to obtain a final marginal distribution model that can describe non-stationarity.

[0102] In this embodiment, in step S41, the sliding window technology is used to calculate the statistical characteristics of each subsequence to obtain a preliminary non-stationary evaluation result, which is specifically:

[0103] Step S411, receiving an annual extreme value data pair matrix, which contains extreme value data of rainfall, tide level and wave; preprocessing the matrix, including missing value interpolation and outlier detection, to obtain a preprocessed extreme value data matrix.

[0104] Step S412: design an adaptive multi-scale sliding window algorithm; the algorithm dynamically adjusts the window size and sliding step size according to the local characteristics of the data, and applies the algorithm to the preprocessed extreme value data matrix to generate a multi-scale time window sequence.

[0105] Step S413: For the data in each time window, traditional statistical features are calculated, including mean, variance, skewness and kurtosis; at the same time, high-order statistical moments and quantile features are extracted to obtain a multidimensional statistical feature vector set.

[0106] Step S414: Apply Hilbert-Huang transform to decompose the data in each time window; extract the instantaneous frequency and instantaneous amplitude information, and construct a time-frequency feature matrix to capture the nonlinear and non-stationary characteristics of the data.

[0107] Step S415: design a composite feature extractor based on the multidimensional statistical feature vector set and the time-frequency feature matrix; the extractor uses a combination of principal component analysis and independent component analysis to extract the most recognizable feature combination to obtain a feature set after dimensionality reduction.

[0108] Step S416, construct a dynamic information entropy calculation model; the model combines Shannon entropy, Renyi entropy and permutation entropy, calculates information entropy on different time scales for the feature set after dimensionality reduction, and generates a multi-scale entropy value sequence.

[0109] Step S417, design a non-stationary index calculation function; the function comprehensively considers the time change rate of the entropy value, the mutation detection result and the long-term trend, analyzes the multi-scale entropy value sequence, and obtains a preliminary non-stationary index.

[0110] Step S418, apply the recursive quantization analysis (RQA) technology to construct a phase space reconstruction model; reconstruct the preliminary non-stationary index sequence, extract the dynamic system characteristics, including determinism, hierarchy and complexity, and obtain the dynamic system feature set.

[0111] Step S419, design a fuzzy inference system based on the Mamdani model; take the dynamic system feature set as input, and obtain the screened non-stationary evaluation results through a predefined fuzzy rule set.

[0112] Step S4110, construct a multivariate joint non-stationarity assessment model; the model considers the interaction between rainfall, tide level and waves, uses Granger causality test and mutual information analysis to evaluate the dynamic correlation between variables, and obtains a comprehensive non-stationarity assessment result.

[0113] In this embodiment, a preliminary assessment of non-stationarity is achieved through trend testing and sliding window technology, providing a basis for subsequent modeling. The Dirichlet process mixture model is used as a prior distribution, and time is introduced as a covariate to construct a flexible Bayesian non-parametric model that can capture non-stationarity, overcoming the limitations of traditional parameter models on the assumption of distribution form. The efficiency and accuracy of parameter estimation are improved through Gibbs sampling and adaptive Markov chain Monte Carlo methods. The goodness of fit and predictive ability of the model are evaluated by cross-validation methods to ensure the reliability of the model. By approximating the non-parametric model to a parameterized distribution and adopting a time-varying parameter model, an accurate description of non-stationarity is achieved. This comprehensive method can not only effectively deal with parameter estimation problems in the case of scarce data, but also accurately capture the non-stationary characteristics caused by climate change. In this way, this step provides an advanced marginal distribution estimation method for estuary areas with limited data, which can more accurately describe the probability distribution characteristics of hydrological elements and their changes over time, and significantly improve the reliability and applicability of subsequent composite risk analysis.

[0114] like Figure 6 As shown, according to one aspect of the present application, step S5 is specifically:

[0115] Step S51: Based on the final marginal distribution model, the annual extreme value data is transformed by probability integral; the empirical distribution function method is used to transform the data to obtain uniformly distributed samples in the interval [0,1], i.e., transformed samples;

[0116] Step S52: Apply a dynamic correlation analysis method to the transformed samples; use time series decomposition technology to extract the time-varying characteristics of the correlation, and obtain a preliminary evaluation of the dependence structure over time, which is recorded as the evaluation result;

[0117] Step S53: construct a dynamic Vine Copula structure based on the evaluation results; use an adaptive algorithm to dynamically select the structure of the Vine tree to obtain a Copula framework for describing multi-dimensional dependencies;

[0118] Step S54: for the dynamic Vine Copula structure, the local likelihood estimation method is used to estimate its parameters; and the kernel smoothing technique is used to process the time effect to obtain the estimation result of the Copula parameters that continuously change with time, which is recorded as the time-varying parameter;

[0119] Step S55: construct a conditional Copula model using time-varying parameters; use a sliding window technique to estimate the conditional Copula at each time point to obtain a dynamic Copula model for capturing the instantaneous dependency structure;

[0120] Step S56, performing model diagnosis on the dynamic Copula model; using time-varying goodness of fit test and residual analysis method to evaluate the fitting effect of the model, and obtaining the adaptability evaluation result of the Copula model;

[0121] Step S57: Based on the adaptability evaluation result, a plurality of candidate Copula models are synthesized by using a model averaging method; a Bayesian model averaging method is used, and the uncertainty of the model is taken into consideration to obtain the final optimal Copula model and its parameter estimation.

[0122] In this embodiment, in step S53, an adaptive algorithm is used to dynamically select the structure of the Vine tree to obtain a Copula framework for describing multi-dimensional dependencies, specifically:

[0123] Step S531, receiving a multidimensional hydrological extreme value data set and a preliminary non-stationarity assessment result; preprocessing the data, including marginal distribution conversion and time window division, to obtain a standardized multidimensional time series data matrix.

[0124] Step S532, construct a dynamic mutual information calculation module; this module uses a kernel density estimation method to calculate the mutual information of different variable pairs in each time window and generate a time-varying mutual information matrix sequence.

[0125] Step S533, design an adaptive tree structure generator; based on the time-varying mutual information matrix sequence, use the maximum spanning tree algorithm to construct an initial Vine tree structure to obtain a candidate Vine structure set.

[0126] Step S534, implement a multi-objective evaluation function; the function comprehensively considers likelihood value, complexity penalty and non-stationary adaptability, performs multi-dimensional evaluation on the candidate Vine structure set, and generates a structure score vector.

[0127] Step S535, calling a dynamic structure adjustment algorithm; the algorithm is based on the principle of simulated annealing, and dynamically adjusts the Vine tree structure according to the structure scoring vector, including adding, deleting and exchanging edges, to obtain an optimized Vine structure sequence.

[0128] Step S536, construct an adaptive pair-copula selection module; combine the Bayesian information criterion and cross-validation to dynamically select the most suitable binary Copula function for each node in the Vine structure to generate a complete Vine Copula model.

[0129] Step S537, design a parameter estimation optimizer; use an adaptive differential evolution algorithm to estimate the parameters of the complete VineCopula model to obtain a preliminary model parameter set.

[0130] Step S538, implement the model diagnosis and adjustment module; use the KS test and Anderson-Darling test to evaluate the goodness of fit of the model, make local adjustments to the parts that do not meet the requirements, and obtain the screened Vine Copula model.

[0131] In this embodiment, multidimensional data is converted into uniformly distributed samples through probability integral transformation, laying the foundation for subsequent Copula modeling. The dynamic correlation analysis method and time series decomposition technology are used to capture the time-varying characteristics of the dependency structure and overcome the limitations of the traditional static Copula model. By constructing a dynamic Vine Copula structure, flexible modeling of high-dimensional complex dependencies is achieved, and the applicability of the model is improved. The local likelihood estimation and kernel smoothing techniques are used to estimate the continuous changes of Copula parameters over time, accurately reflecting the dynamic evolution of the dependency structure. The reliability and stability of the model are improved through model diagnosis and Bayesian model averaging methods. This dynamic Copula modeling method can not only handle multivariate dependencies under data scarcity, but also accurately describe the evolution of dependency structures caused by climate change. It provides an advanced multivariate joint distribution modeling method for estuary areas with limited data, which can more accurately characterize the complex interactions between hydrological elements and their changes over time, and significantly improve the accuracy and reliability of compound flood risk assessment.

[0132] like Figure 7 As shown, according to one aspect of the present application, step S6 is specifically:

[0133] Step S61, combining the marginal distribution model and the optimal Copula model, using a hierarchical modeling method to integrate the parameters of the marginal distribution and the dependent structure into a unified Bayesian framework, and obtaining a global parameter model;

[0134] Step S62: Based on the global parameter model, the optimization target of the approximate posterior distribution is constructed using the Kullback-Leibler divergence minimization principle to obtain a mathematical expression of variational inference, which is the objective function;

[0135] Step S63, using the stochastic gradient descent method to optimize the objective function, to obtain a parameter estimate of an approximate posterior distribution, that is, to obtain a stochastic variational inference result;

[0136] Step S64: Based on the random variational inference results, an adaptive importance sampling technique is used to improve the sampling efficiency of rare events, and a preliminary posterior sample set is obtained;

[0137] Step S65, using the effective sample size and Gelman-Rubin statistics to evaluate the sample quality, using an adaptive sampling strategy to adjust the generator parameters, and obtaining an optimized posterior sample generation scheme;

[0138] Step S66, using the optimized posterior sample generation scheme, adopting a parallel computing method to generate several groups of posterior samples for each hydrological element, and obtaining a high-quality sample set of a predetermined size;

[0139] Step S67: reconstruct the continuous posterior distribution using the kernel density estimation method, extract interval estimates of different confidence levels using the quantile regression technique, and obtain a complete description of the posterior distribution and its summary statistics, which is the posterior sample.

[0140] In this embodiment, step S64 is specifically as follows:

[0141] Step S641, receiving the results of random variational inference, including parameter estimation and uncertainty quantification of the approximate posterior distribution; preprocessing these results, including parameter normalization and uncertainty calibration, to obtain a standardized variational inference result set.

[0142] Step S642: construct an initial proposal distribution based on the standardized variational inference result set; parameterize the proposal distribution using a mixed Gaussian model to obtain an initialized mixed Gaussian model parameter set.

[0143] Step S643, design an adaptive stratified sampling strategy; divide the parameter space into multiple subspaces, each subspace corresponds to a different degree of rarity, and obtain a stratified sampling architecture diagram.

[0144] Step S644: For each subspace, apply the temperature annealing technique to adjust the proposed distribution; gradually reduce the temperature parameter so that the proposed distribution gradually approaches the target posterior distribution, and generate a series of intermediate distributions.

[0145] Step S645, implement a cross-layer adaptive importance sampler; the sampler can dynamically adjust the sampling ratio between different layers, perform sampling based on the intermediate distribution, and obtain a preliminary sample set.

[0146] Step S646, construct a meta-learning optimization module; use a recurrent neural network to learn the pattern in the sampling process, dynamically adjust the sampling parameters, including the temperature reduction rate and the probability of transition between levels, to obtain an optimized sampling strategy.

[0147] Step S647, implement a rare event detector; use an anomaly detection algorithm to identify potential rare events in the sample set, focus on sampling these events, and obtain an enhanced rare event sample set.

[0148] Step S648, design a multi-scale resampling mechanism; resample at different parameter scales to capture rare event features of different granularities and generate a multi-scale sample set.

[0149] Step S649: Apply particle filtering technology to dynamically update the multi-scale sample set; evaluate the sample weights according to the likelihood function, perform importance resampling, and obtain an updated sample set.

[0150] In this embodiment, the parameters of the marginal distribution and the dependent structure are integrated into a unified Bayesian framework through a hierarchical modeling method, and the consistency estimation of the global parameters is achieved. The approximate posterior distribution is constructed by minimizing the Kullback-Leibler divergence principle, which overcomes the computational efficiency bottleneck of the traditional MCMC method in high-dimensional problems. The objective function is optimized by the stochastic gradient descent method, which improves the convergence speed and stability of variational inference. The sampling efficiency of rare events is improved by using the adaptive importance sampling technology, which solves the problem of inaccurate estimation of extreme events by traditional methods. Through parallel computing and kernel density estimation methods, a high-quality posterior distribution sample set is generated. This uncertainty quantification method based on variational inference can not only effectively deal with parameter uncertainty in the case of scarce data, but also accurately characterize the probabilistic characteristics of extreme events. It provides an advanced uncertainty quantification method for estuary areas with limited data, which can more comprehensively evaluate the sources of uncertainty and their impacts of compound flood risks, and significantly improve the reliability and credibility of risk assessment results.

[0151] like Figure 8 As shown, according to one aspect of the present application, step S7 is specifically:

[0152] Step S71, using the Latin hypercube sampling method to generate parameter combinations, simulating a predetermined hydrological scenario, obtaining a sample distribution of a joint return period and outputting it as a simulation result;

[0153] Step S72: for the simulation results, use the numerical integration method to solve the multi-dimensional integral, calculate different types of joint return periods, and obtain the time-varying joint return period estimation considering non-stationarity;

[0154] Step S73: construct a confidence interval based on the time-varying joint return period estimation; use the quantile method to calculate the interval boundaries of different confidence levels to obtain the uncertainty range of the return period estimation; calculate the relative uncertainty index to obtain a standardized uncertainty measure;

[0155] Step S74: Based on the uncertainty measurement, the Sobol index method is used to quantify the contribution of each hydrological element to the overall uncertainty, and a sensitivity analysis is performed to obtain the decomposition results of the uncertainty sources.

[0156] Step S75, construct a multi-criteria decision analysis framework, integrate return period estimation and uncertainty decomposition; use the analytic hierarchy process to determine the weight of each criterion, use the TOPSIS method to comprehensively evaluate different scenarios, and obtain the risk ranking that considers multiple factors, which is the comprehensive evaluation;

[0157] Step S76: Based on the comprehensive evaluation results, use visualization technology to display the risk distribution, adopt scenario analysis method to explore the impact of climate change on future risks, and obtain the final composite flood disaster risk assessment report and decision-making recommendations.

[0158] In this embodiment, the Latin hypercube sampling method is used to generate parameter combinations, which improves the efficiency and coverage of scenario simulation. The multidimensional integral is solved by the numerical integration method, and the time-varying joint return period estimation considering non-stationarity is realized, overcoming the limitations of the traditional static return period analysis. The confidence interval is constructed by the quantile method, and the uncertainty range of the return period estimation is accurately quantified. Moreover, the contribution of each hydrological element to the overall uncertainty is quantified by the Sobol index method, and the accurate decomposition of the uncertainty source is achieved. A multi-criteria decision analysis framework is adopted, and the return period estimation and uncertainty decomposition results are comprehensively considered to provide a comprehensive risk assessment and decision-making recommendation. This comprehensive analysis method can not only effectively deal with the risk assessment problem under the condition of scarce data, but also accurately reflect the long-term impact of climate change on compound flood risks. In this way, this step provides an advanced compound flood risk assessment method for estuary areas with limited data, which can more comprehensively and accurately assess current and future flood risks, provide a reliable scientific basis for flood prevention and disaster reduction decisions, and significantly improve the flood control capacity and climate adaptability of estuary areas.

[0159] According to one aspect of the present application, in step S34, a hierarchical clustering method is used to classify extreme value events based on DTW distance to obtain a preliminary set of composite extreme value events, specifically:

[0160] Step S341, calculating the DTW distance for each pair of hydrological elements, using the FastDTW algorithm to improve the calculation efficiency, especially for long time series, generating three DTW distance matrices, each of which is a×a in size, where a is the number of extreme value events; each pair of hydrological elements includes rainfall-water level, rainfall-wave height, and water level-wave height;

[0161] Step S342: Treat each extreme value event as a separate cluster, create a priority queue to store the distances between clusters, and sort them from small to large distances; initialize a union-find data structure to track the merging of clusters;

[0162] Step S343, take out the cluster pair with the minimum distance from the priority queue; merge the two clusters, update and query; calculate the distance between the new cluster and all other clusters: new distance = (size of cluster A * distance between A and C + size of cluster B * distance between B and C) / (size of A + size of B); add the newly calculated distance to the priority queue; until all extreme value events are merged into one cluster:

[0163] Step S344: for each cluster number k, obtain several clusters according to the cluster tree cutting, calculate the silhouette coefficient of each sample, obtain the silhouette coefficients under different cluster numbers and calculate the average silhouette coefficient, and select the cluster number k with the largest average silhouette coefficient as the optimal cluster number;

[0164] Step S345: cut the cluster tree according to the optimal number of clusters, and assign a unique identifier to each cluster; for each cluster, calculate the cluster center using the average value of the original time series, and find the event with the smallest DTW distance to the cluster center as the representative event of the cluster;

[0165] Step S346, creating an empty composite extreme value event set, for each pair of representative events of hydrological elements: checking whether the time of occurrence of the events overlaps, if so, combining these events into a composite event; adding the combined composite event to the composite extreme value event set;

[0166] Step S347, calculate the internal evaluation indicators of clustering, including the Davies-Bouldin index and the Calinski-Harabasz index; visualize the clustering results, generate a clustering dendrogram and a multidimensional scaling diagram; generate a clustering analysis report, the clustering analysis includes the clustering results of each hydrological element, a representative event list, a set of composite extreme value events, clustering quality evaluation indicators, as well as a dendrogram and an MDS diagram.

[0167] In this embodiment, the FastDTW algorithm is used to calculate the distance matrix between hydrological elements, which significantly improves the computational efficiency of long time series and overcomes the defect that the traditional Euclidean distance cannot handle the problem of time series misalignment. By creating a priority queue and a union-find data structure, an efficient hierarchical clustering process is realized, and the time complexity of the algorithm is effectively reduced. By dynamically calculating the distance between the new cluster and other clusters, the accuracy and flexibility of the clustering process are guaranteed. The silhouette coefficient is used as an evaluation index to adaptively determine the optimal number of clusters, avoiding the subjectivity that may be caused by artificially setting the number of clusters. By identifying representative events and checking time overlap, accurate identification of composite extreme value events is achieved. It can not only effectively handle data scarcity, but also capture the complex time relationship between different hydrological elements. By calculating internal evaluation indicators and visualizing clustering results, a comprehensive clustering quality assessment is provided, which enhances the interpretability and credibility of the results.

[0168] According to one aspect of the present application, in step S54, for the dynamic Vine Copula structure, a local likelihood estimation method is used to estimate its parameters, specifically:

[0169] S541, define the window size and sliding step size, create a time window set for the time series; for each time window, extract the corresponding multivariate extreme value data subset;

[0170] S542, for the data subset in each time window: using the empirical distribution function to convert the original data into pseudo observations, for each candidate Copula family, constructing the log-likelihood function, and selecting the optimal Copula family using the Akaike information criterion, that is, selecting the Copula family with the smallest Akaike information criterion value as the optimal model for the window;

[0171] S543, for each parameter, construct a time series and apply local polynomial regression to obtain a smoothed time function of the parameter;

[0172] S544. For each time point and parameter, generate B Bootstrap samples and calculate the Bootstrap standard error:

[0173] S545. For each time point, estimate the parameters of each candidate Copula family, calculate the Akaike Information Criterion value of each Copula family, and select the Copula family with the smallest AIC as the optimal model at that time point.

[0174] S546. Divide the time series into several intervals, in each interval the Copula family remains unchanged, and use the hidden Markov model to describe the transformation of the Copula family; generate a time-varying Copula model report, which includes the optimal Copula family and corresponding parameters for each time window, the time-varying function graph of the parameters, the information of the Copula family transformation model, and the model diagnostic information.

[0175] In this embodiment, the sliding time window technology is used to accurately capture the time-varying characteristics of the dependency structure, overcoming the limitation of the traditional static Copula model that ignores time evolution. The empirical distribution function is used to convert the original data into pseudo-observations, which effectively avoids the influence of marginal distribution selection on Copula modeling and improves the robustness of the model. By constructing the log-likelihood function and using the Akaike information criterion to select the optimal Copula family, the dynamic optimization of the model structure is achieved, and the adaptability of the model to complex dependencies is enhanced. Using the local polynomial regression technology, the smooth time function of the Copula parameters is obtained, which accurately describes the continuous change process of the dependency structure. The confidence interval of the parameter estimation is constructed by the Bootstrap method, and the uncertainty of the model estimation is quantified. It can not only handle high-dimensional multivariate dependencies under the condition of scarce data, but also accurately describe the evolution of dependency structures caused by climate change. By using the hidden Markov model to describe the transformation of the Copula family, the model's ability to capture mutations in the dependency structure is further enhanced.

[0176] In another embodiment of the present application, step S2 further includes a process of converting the sea level rise data into the wind-rain-wave-current coupling model, specifically:

[0177] Based on the sea level rise prediction data from authoritative organizations such as the IPCC, a sea level rise time series for the study area is constructed.

[0178] Use piecewise linear regression or polynomial fitting methods to build a temporal trend model of sea level rise.

[0179] At each simulation time step, the amount of sea level rise is calculated based on the time trend model.

[0180] The calculated sea level rise is superimposed on the open boundary water level conditions of the model.

[0181] Updates the bathymetric information of the model grid to reflect the effects of sea level rise on the terrain.

[0182] The coupled model was rerun to obtain the hydrological dynamic simulation results taking into account the impact of sea level rise.

[0183] In another embodiment of the present application, in step S4, the process of processing missing data is specifically as follows:

[0184] Perform missing value detection on the annual extreme value data matrix to identify the location and proportion of missing data.

[0185] For short-term missing values ​​(such as a single year), multiple imputation is used to generate multiple sets of possible filling values.

[0186] For long-term (multiple consecutive years) missing data, time series forecasting methods (such as ARIMA or Prophet) are used to fill them by combining historical data trends and climate model predictions.

[0187] The imputed data are checked for consistency to ensure that the imputed values ​​are consistent with the known data in terms of statistical characteristics.

[0188] In the subsequent analysis, the uncertainty introduced by data filling was considered and its impact on the final results was evaluated through sensitivity analysis.

[0189] In another embodiment of the present application, climate change scenarios are also included in the calculation of the joint return period, specifically:

[0190] Based on the climate change scenarios of IPCC and other institutions (such as RCP4.5 and RCP8.5), a climate change prediction dataset for the study area is constructed.

[0191] Use statistical downscaling or dynamical downscaling methods to adjust global climate model results to the scale of the study area.

[0192] The downscaled climate data are bias-corrected to ensure consistency with historical observation data.

[0193] The corrected climate change data are used as input and the above steps are recalculated to obtain the joint distribution of hydrological elements considering climate change.

[0194] Based on the joint distribution of hydrological elements, the conditional probability method is used to calculate the joint return period under different climate change scenarios.

[0195] By comparing and analyzing the joint return periods under different climate change scenarios, the impact of climate change on compound flood risks is assessed.

[0196] Construct sensitivity indicators of climate change impacts and quantify the response of the joint return period to climate change.

[0197] In another embodiment of the present application, it may also be:

[0198] Step 1: Collect information.

[0199] ①. Collect the latest OpeanStreetMap global coastline remote sensing data;

[0200] ②. Collect GEBCO seabed topography data of a large area of ​​sea; collect nearshore chart data, extract water depth point data from S57 chart data, and merge it with GEBCO data to form a new data file *.xyz.

[0201] ③. Runoff data of the upper reaches of the estuary in the study area;

[0202] ④. Collect the ERA5 hourly reanalysis wind speed data set; collect the best typhoon path data;

[0203] ⑤. Collect hourly rainfall data from all weather stations within 100 kilometers of the study area for the past 70 years;

[0204] ⑥. Collect and study hourly tide level and wave observation data in the estuary area.

[0205] ⑦. Sea level rise data.

[0206] Step 2: Establish a wind-rain-wave-current coupling mathematical model to calculate the hourly water level and effective wave height in the study area. Based on the topographic coastline data ①② in step 1, the open source three-dimensional finite volume method ocean numerical model FVCOM is used to construct a wind-rain-wave-current coupling mathematical model, and batch numerical simulations of the water level and waves in the study area for nearly 70 years are performed, including storm surges and typhoon waves during typhoons, to generate hourly water level and effective wave height data in the study area. Among them, the outer sea boundary is driven by harmonic constants, and the upstream river boundary is driven by measured runoff (step 1 data ③). The external force input to the model is era5 wind pressure data (step 1 data ④) and measured rainfall data (step 1 data ⑤). This step can calculate the hourly water level and effective wave height data in the study area. Among them, the hourly water level at the research site is directly linearly superimposed with the sea level rise data (data ⑦), so that the impact of sea level rise on the trend change of water level can be considered.

[0207] Step 3: Identification of extreme events: Identify extreme data pairs for complex flood disaster research.

[0208] According to the definition, two (or more) disaster factors in a composite event occur simultaneously (exceeding a certain threshold) or the time interval between them is short. That is, before the impact of the previous disaster factor has completely subsided, another disaster factor also has an over-threshold event. Therefore, the input data of the statistical model in the composite event study is a series of extreme value data pairs with a certain time delay (which can be 0). There are generally two methods to identify extreme value data: one is to determine the extreme values ​​in the time series of each disaster factor separately, and then pair the extreme values ​​according to the time delay, but the number of extreme value data pairs obtained by this method is generally small; the other more commonly used method uses the AM or POT method to identify the extreme value event in the first variable (preferred variable), and then sets a window period (advance or lag) to identify the maximum value of another accompanying variable within the window period.

[0209] Spatial processing of rainfall data: The rainfall data of all meteorological stations within 120km near the study area are averaged to obtain a new rainfall time series for studying the current coastal compound flood. As shown in the figure, when studying the compound flood disaster in the Yangtze River Estuary, the easternmost side of Chongming Island is selected as the center and the circular area with a radius of 120km is drawn, and the rainfall observation data of all meteorological stations in the area are averaged.

[0210] Temporal processing of rainfall data: In the traditional method, rainfall data is generally processed by first calculating the daily cumulative rainfall, and then calculating the maximum value of each year as the annual rainfall extreme value. The present invention uses a sliding window summation method to process rainfall data. That is, first determine the length of the sliding window (24h), then sum the data in the window to obtain a new 24-hour cumulative rainfall time series, and finally calculate the maximum value of all data in a year. The advantage of this method is that it does not interfere with the division of the window artificially, so that the maximum value of the cumulative rainfall within 24 hours can be well captured.

[0211] Assume that the original data sequence is {x1, x2, ..., x n}, the sliding window size is k, and the new data sequence after the sliding window summation is {y1, y2, ..., y n-k+1}. Each element y in the new data sequence i Calculated by the following formula: ;

[0212] The value range of i is 1≤i≤n-k+1. Generally, k is 24 and n is 365.25×24.

[0213] The annual maximum 24-hour cumulative rainfall is: Pre = max (y im ), where y im Represents the i-th rainfall data in the m-th year, and the value range of year m is 1, 2, 3, ..., K.

[0214] The above method can obtain the maximum 24-hour cumulative rainfall time series, including extreme rainfall values ​​and their occurrence time.

[0215] Based on the occurrence time of annual extreme rainfall, the highest tidal level can be found within ±24 hours to obtain the annual extreme tidal level time series and its occurrence time corresponding to the annual extreme rainfall.

[0216] Based on the time when the annual extreme tide occurs each year, the corresponding effective wave height at this moment can be found to obtain the annual extreme effective wave height corresponding to the annual extreme tide.

[0217] Through this step, we can get the annual extreme value data pairs of rainfall, tide level and wave. If we have 70 years of hourly data of the three, the dimension of the annual extreme value data finally obtained will be 70 rows and 3 columns.

[0218] Step 4: Use the stationary / non-stationary method to find the parameters of rainfall, tide level and wave data. In the context of global warming, climate change leads to the non-stationarity of hydrological factors. Six common hydrological variable distributions, namely normal distribution, exponential distribution, logistic distribution, gamma distribution, Pearson type III distribution and GEV distribution, are selected to fit the observation sequence of each variable respectively to find the distribution that best suits the marginal data. In order to consider non-stationarity, it is assumed that the location parameter of each marginal distribution changes over time. Therefore, the location parameter becomes: μ=μ0+μ1t; where t represents time (year), μ0 and μ1 are linear regression parameters. This formula takes into account the trend of rainfall, sea level and wave time series. Although scale parameters and shape parameters are also non-stationary, the changes are small. It is generally believed that only the location parameters will change over time. Therefore, this embodiment only considers the non-stationarity of the location parameter, which captures the mean change of hydrological variables. Finally, the Kolmogorov-Smirnov (KS) test is used to check the validity of each marginal distribution. Estimating the parameters of the marginal distribution through stationary / non-stationary methods is the basis for the subsequent calculation of the impact of non-stationarity on compound flood disasters.

[0219] Step 5. Estimate the copula parameters between rainfall, tide and wave extreme value data. Based on the rainfall, tide and wave data and the marginal distribution parameters of step 4, six commonly used copula functions, including Gaussian Copula, Student t Copula, Clayton Copula, Gumbel Copula, Frank Copula and Joe Copula, are selected as alternative copula functions. The maximum likelihood method is used to find the optimal copula type and its parameters. Finally, the Akaike information criterion (AIC) statistic is used to evaluate the goodness of fit of the copula function. During the test, the smaller the AIC value, the better the fitting effect. The expression of AIC is: AIC=-2Ln(l max ), K is the number of estimated parameters, l max is the value of the maximum likelihood function.

[0220] Step 6: Generate the posterior distribution of rainfall, tide level and waves using the Markov chain Monte Carlo method (MCMC) based on Bayesian theorem. Based on the marginal distribution and parameters in step 4 and the copula function type and parameters in step 5, the Markov chain Monte Carlo method (MCMC) based on Bayesian theorem is used to generate the posterior distribution of rainfall, tide level and waves under stationary / non-stationary scenarios (10,000 groups each).

[0221] Consider two random variables, the random variables and Cumulative joint distribution function (such as rainfall-runoff, rainfall-tide, or wave-tide) It can be expressed as: ;in and is the continuous marginal cumulative distribution, C(•) is the copula function, θ represents the copula parameter, represents the parameters of the joint cumulative distribution function, which include the parameters of the marginal distribution and the copula. Assuming that the joint probability distribution is continuous, the joint density function f XY can be written as:

[0222] ;

[0223] where f X and f Y is the marginal density function, and C(•) is the copula probability density function: ;

[0224] The Markov Chain Monte Carlo method (MCMC) based on Bayes' theorem allows the incorporation of prior knowledge through prior distributions, which greatly facilitates the estimation of univariate and bivariate model parameters. MCMC can systematically sample from multidimensional probability distributions and obtain the posterior distribution of model parameters. The prior distribution represents the assumption about the parameter value before incorporating new data, which is especially applicable when the measured data is short. Bayes' theorem can be expressed mathematically as:

[0225] ;

[0226] That is the posterior distribution, which provides the distribution of the model parameters given new data; P(φ) is the prior distribution; is the likelihood function, which quantifies the probability of an observation given specific values ​​of the model parameters; while P(Y obs ) is a normalization constant that ensures that the posterior probability distribution integrates to unity. In practice, we calculate P(Y obs ) is difficult because it is usually an integral without a closed-form solution. Based on the following formula, the likelihood function can be calculated as follows:

[0227] ;

[0228] Where n is the number of data we use for a particular site. Regarding the prior distribution, non-informative uniform distribution was used as the copula parameter. For the location and scale parameters of the marginal distribution, non-informative normal distribution was used, while for the shape parameter, a normal distribution with a standard deviation of 3 was applied to ensure the convergence of the model. After selecting the best marginal distribution, copula function, and prior distribution, the posterior distribution becomes based on the following formula:

[0229] ;

[0230] Step 7: Calculate the stationary / non-stationary joint return period, and quantify the impact of the non-stationarity of each disaster factor (rainfall, tide level, wave) and the dependence of the three on the composite flood return period. Based on steps 4 to 6, the joint return period under different scenarios can be calculated, and the impact of the non-stationarity of hydrological variables and the dependence between hydrological variables on the uncertainty of the composite flood return period can be quantified. Assuming that the hydrological elements are stationary, taking the co-occurrence (and) return period as an example, the calculation method of the double variable and return period is as follows: ;in, It is the average time between two consecutive compound flood disasters.

[0231] Assuming that the hydrological elements are stable, taking the co-occurrence (and) recurrence period as an example, the calculation method of the co-occurrence (and) recurrence period of the three variables of rainfall, tide level and wave is as follows: ;

[0232] Assuming that the hydrological elements are stable, taking the co-occurrence (and) recurrence period as an example, the calculation method of the bivariate and recurrence period is as follows: ;

[0233] When considering the non-stationarity of hydrological elements, taking the co-occurrence (and) return period as an example, the calculation method of the bivariate and return period is as follows: ;

[0234] When considering the non-stationarity of hydrological elements, taking the co-occurrence (and) recurrence period as an example, the calculation method of the three-variable and recurrence period is as follows: ;

[0235] To calculate the uncertainty of the return period, based on the sampling results of the MCMC method, the 5th, 50th and 95th quantiles of the return period values ​​are first calculated. Then the symmetric range of the return period is calculated:

[0236] ;

[0237] The contribution of the i-th factor to the uncertainty of the composite flood hazard can be calculated as follows: ;

[0238] In short, there are few studies on the non-stationarity of estuarine and coastal composite flood disasters in the prior art, especially in estuarine areas with limited hydrological data. In the context of climate change, hydrological factors show strong non-stationarity. First, based on the physical process, the present invention establishes a wind-rain-wave-current coupling mathematical model, and calculates the hourly water level and effective wave height of the study area in the context of sea level rise; secondly, the sliding window correlation idea is used to calculate the 24h cumulative rainfall, and on this basis, the extreme data pairs used for composite flood disaster research are identified; again, the copula function is used to generate the posterior distribution of rainfall, tide level and wave based on the Markov chain Monte Carlo method (MCMC) method of Bayesian theorem; finally, the return period and uncertainty of coastal composite floods are calculated, and the impact of the non-stationarity and dependence of each hydrological factor on the uncertainty of the return period is quantified. This method has certain theoretical value and practical significance for the risk assessment of composite disasters in estuaries caused by climate change, including sea level rise and changes in rainfall patterns.

[0239] It should be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

Claims

1. A method to study composite flood hazards in data-deficient estuaries considering climate change, It is characterized in that The steps include: Step S1, obtaining multi-source heterogeneous data, and preprocessing them using data fusion and standardization methods to obtain a comprehensive data set for subsequent analysis; Among them, multi-source heterogeneous data include geographic information data, hydrological and meteorological data, and ocean dynamics data; geographic information data include global coastline remote sensing data, seafloor topography data, and nearshore water depth data; hydrological and meteorological data include runoff data from the upper reaches of the estuary, ERA5 hourly reanalysis wind speed data set, optimal typhoon path data, and hourly rainfall data from meteorological stations near the study area; ocean dynamics data include hourly tide and wave observation data, as well as sea level rise data; Step S2, based on the comprehensive data set, construct and use the wind-rain-wave-current coupling mathematical model to calculate the long-term series hydrological dynamic elements of the study area; Step S3, based on the long-term series hydrological dynamic data and rainfall data, using an improved extreme value identification method, extracting extreme value data pairs for compound flood disaster research, and constructing an annual extreme value data pair matrix based on the extreme value data pairs; Step S4, based on the annual extreme value data pair matrix, using the Bayesian non-parametric method to estimate the marginal distribution parameters of rainfall, tide level and wave, and obtain the optimal marginal distribution model considering non-stationarity; Step S5, using the annual extreme value data pair matrix and the marginal distribution model, estimating the parameters of the Copula function describing the multivariate dependency structure, and obtaining the optimal Copula model that can capture the dynamic dependency relationship; Step S6: Based on the marginal distribution model and the Copula model, a variational inference method is used to generate a joint posterior distribution of hydrological elements, and a posterior distribution sample set of a predetermined size for uncertainty analysis is obtained; Step S7, using the posterior distribution sample set to calculate the joint return period under different scenarios, and perform a comprehensive uncertainty analysis to obtain the final composite flood risk assessment result; The step S6 is specifically as follows: Step S61, combining the marginal distribution model and the optimal Copula model, using a hierarchical modeling method to integrate the parameters of the marginal distribution and the dependent structure into a unified Bayesian framework, and obtaining a global parameter model; Step S62: Based on the global parameter model, the optimization target of the approximate posterior distribution is constructed using the Kullback-Leibler divergence minimization principle to obtain a mathematical expression of variational inference, which is the objective function; Step S63, using the stochastic gradient descent method to optimize the objective function, to obtain a parameter estimate of an approximate posterior distribution, that is, to obtain a stochastic variational inference result; Step S64: Based on the random variational inference results, an adaptive importance sampling technique is used to improve the sampling efficiency of rare events, and a preliminary posterior sample set is obtained; Step S65, using the effective sample size and Gelman-Rubin statistics to evaluate the sample quality, using an adaptive sampling strategy to adjust the generator parameters, and obtaining an optimized posterior sample generation scheme; Step S66, using the optimized posterior sample generation scheme, adopting a parallel computing method to generate several groups of posterior samples for each hydrological element, and obtaining a high-quality sample set of a predetermined size; Step S67: reconstruct the continuous posterior distribution using the kernel density estimation method, extract interval estimates of different confidence levels using the quantile regression technique, and obtain a complete description of the posterior distribution and its summary statistics, which is the posterior sample.

2. A method for studying composite flood disasters in data-deficient estuaries taking into account climate change as claimed in claim 1, It is characterized in that The step S1 is specifically as follows: Step S11, obtaining geographic information data, including: obtaining the latest global coastline remote sensing data, obtaining seabed topography data of a large range of sea areas from GEBCO, and extracting nearshore water depth point data from nautical charts; using spatial data processing methods to unify coordinates and resolutions to obtain a geographic information data set for the study area; Step S12, obtaining hydrological and meteorological data, including runoff data of the upper reaches of the estuary in the study area, ERA5 hourly reanalysis wind speed data set, optimal typhoon path data, and hourly rainfall data of all meteorological stations within M kilometers near the study area in the past N years; using time series interpolation and spatial interpolation methods to perform spatiotemporal consistency processing to obtain a long-term hydrological and meteorological data set covering the study area; M and N are predetermined natural numbers; Step S13, obtaining ocean dynamic data, including hourly tide level and wave observation data of the estuary area, and sea level rise data; using data quality control and outlier detection methods to clean and correct, to obtain a reliable ocean dynamic data set; Step S14: align the geographic information dataset, the long-term hydrological and meteorological dataset, and the ocean dynamics dataset in time and space and unify their formats, and use a data fusion algorithm module to integrate data from different sources to obtain a unified, multi-dimensional comprehensive dataset.

3. A method for studying composite flood disasters in data-deficient estuaries taking into account climate change as claimed in claim 2, It is characterized in that The step S2 is specifically as follows: Step S21, constructing a three-dimensional numerical grid of the study area using geographic information data; using an adaptive grid refinement method to adjust the grid density according to the regional terrain conditions to obtain an optimized calculation grid; Step S22, based on the open source three-dimensional finite volume method ocean numerical model FVCOM, combined with the computational grid, the basic hydrodynamic model and boundary conditions of the study area are established, and the model performance is optimized by using parallel computing technology; the boundary conditions include using tidal harmonic constants as the outer sea boundary conditions, using measured runoff as the upstream river boundary conditions, and using ERA5 wind pressure data and measured rainfall data as surface forcing; Step S23: inputting the hydrological and meteorological data and the ocean dynamic data into the basic hydrodynamic model to obtain the model output result, using the data assimilation technology to fuse the model output result with the observation data, constructing a preliminary wind-rain-wave-current coupling model, and obtaining preliminary long-term series hydrological and dynamic data; wherein the long-term series hydrological and dynamic data include water level and flow velocity; Step S24: Integrate the preliminary wind-rain-wave-current coupling model with the preconfigured deep learning model, and dynamically adjust the weights of the physical model and the data-driven model using a model hybrid strategy; evaluate the model performance through a cross-validation method to obtain an optimized wind-rain-wave-current coupling mathematical model, and output an improved long-term series hydrological dynamic simulation result; Step S25: Use the optimized wind-rain-wave-current coupling mathematical model to perform batch numerical simulations on the hydrological dynamic process in the study area over the past N years, including simulating storm surges and typhoon waves during typhoons; use distributed computing technology to accelerate large-scale computing, obtain and output hourly water level data and effective wave height data for the study area over N years as key long-term series hydrological dynamic elements; Step S26: Use a preconfigured superposition method to superimpose the hourly water level data with the sea level rise data to consider the impact of sea level rise on the trend change of the water level; use time series decomposition technology to separate the trend term, obtain and output the corrected water level data that considers the impact of climate change as the final long-term series hydrological dynamic element.

4. A method for studying composite flood disasters in data-deficient estuaries taking into account climate change as claimed in claim 3, It is characterized in that The step S3 is specifically as follows: Step S31, spatially process the obtained rainfall data using the inverse distance weighted interpolation method, spatially average the rainfall data of all meteorological stations within Q kilometers near the study area, and obtain a representative regional rainfall time series; Q is a predetermined value; Step S32: using the variational mode decomposition method to perform multi-scale decomposition on the regional rainfall time series; using the adaptive threshold method to determine the extreme values ​​of each scale, and obtaining a multi-scale rainfall extreme value candidate set; Step S33: applying the empirical mode decomposition method to the water level data and the effective wave height data to extract the oscillation modes of different frequencies; using the envelope method to identify the extreme points of each mode to obtain the extreme value candidate set of the water level and the wave height; Step S34, inputting the rainfall extreme value candidate set and the water level and wave height extreme value candidate set into the dynamic time warping algorithm module to calculate the time correlation between extreme value events of different hydrological elements; A hierarchical clustering method is used to classify extreme value events based on DTW distance, and a preliminary set of composite extreme value events is obtained; Step S35, applying the sliding window technique to the composite extreme value event set obtained in step S34, taking a hour as the window length, and calculating the cumulative effect; The peak detection algorithm is used to identify the maximum value in the window, and a set of extreme value events that are screened considering the time accumulation effect is obtained; a is a predetermined natural number; Step S36: Based on the screened extreme value event set, the conditional probability method is used to calculate the probability of other elements having extreme values ​​within a specified time window when an extreme value of a given hydrological element occurs; the threshold is set using multivariate extreme value theory to screen out the final extreme value data pairs that meet the definition of the composite event; Step S37, organize the final extreme value data pairs by year to form an annual extreme value data matrix; use data interpolation technology to process possible missing values ​​to obtain a complete N-row and 3-column annual extreme value data pair matrix; the hydrological elements of the 3 columns are rainfall, tide level and wave respectively.

5. A method for studying composite flood disasters in data-deficient estuaries taking into account climate change as claimed in claim 4, It is characterized in that The step S4 is specifically as follows: Step S41, applying the trend test method to each hydrological element in the annual extreme value data matrix, and using the sliding window technology to calculate the statistical characteristics of each subsequence to obtain a preliminary non-stationary evaluation result; Step S42: Based on the preliminary non-stationarity assessment results, construct a prior distribution of the Dirichlet process mixture model; adopt a hierarchical Bayesian framework to introduce time as a covariate into the Dirichlet process mixture model to obtain a Ling Bayesian non-parametric model that can capture non-stationarity; Step S43: use the Gibbs sampling algorithm to estimate the parameters of the constructed Dirichlet process mixture model; use the adaptive Markov chain Monte Carlo method to improve the sampling efficiency and obtain the posterior distribution samples of the model parameters; Step S44, based on the obtained distribution posterior samples, calculate the prediction distribution of the Dirichlet process mixture model; use the cross-validation method to evaluate the goodness of fit and prediction ability of the model to obtain the optimal non-parametric marginal distribution model; Step S45, approximating the non-parametric marginal distribution model to a parameterized distribution; using a moment matching method or a maximum likelihood estimation method to fit candidate parameter distributions including a generalized extreme value distribution and a generalized Pareto distribution, to obtain a parameterized marginal distribution of each hydrological element; Step S46, using the AIC or BIC information criterion to select a model for the candidate parameter distribution; using the bootstrap method to evaluate the uncertainty of the model selection, and obtaining the optimal marginal distribution type and parameter estimation value of each hydrological element; Step S47: Based on the optimal marginal distribution type and parameter estimates of each hydrological element, a time-varying parameter model is constructed to allow the distribution parameters to change over time; a generalized additive model is used to fit the time trend of the parameters to obtain a final marginal distribution model that can describe non-stationarity.

6. A method for studying composite flood disasters in data-deficient estuaries taking into account climate change as claimed in claim 5, It is characterized in that The step S5 is specifically as follows: Step S51: Based on the final marginal distribution model, the annual extreme value data is transformed by probability integral; the empirical distribution function method is used to transform the data to obtain uniformly distributed samples in the interval [0,1], i.e., transformed samples; Step S52: Apply a dynamic correlation analysis method to the transformed samples; use time series decomposition technology to extract the time-varying characteristics of the correlation, and obtain a preliminary evaluation of the dependence structure over time, which is recorded as the evaluation result; Step S53: construct a dynamic Vine Copula structure based on the evaluation results; use an adaptive algorithm to dynamically select the structure of the Vine tree to obtain a Copula framework for describing multi-dimensional dependencies; Step S54: for the dynamic Vine Copula structure, estimate its parameters using a local likelihood estimation method; The kernel smoothing technique is used to process the time effect, and the estimation result of the Copula parameter that changes continuously with time is obtained, which is recorded as the time-varying parameter; Step S55: construct a conditional Copula model using time-varying parameters; use a sliding window technique to estimate the conditional Copula at each time point to obtain a dynamic Copula model for capturing the instantaneous dependency structure; Step S56, performing model diagnosis on the dynamic Copula model; using time-varying goodness of fit test and residual analysis method to evaluate the fitting effect of the model, and obtaining the adaptability evaluation result of the Copula model; Step S57: Based on the adaptability evaluation result, a plurality of candidate Copula models are synthesized by using a model averaging method; a Bayesian model averaging method is used, and the uncertainty of the model is taken into consideration to obtain the final optimal Copula model and its parameter estimation.

7. A method for studying composite flood disasters in data-deficient estuaries taking into account climate change as claimed in claim 6, It is characterized in that The step S7 is specifically as follows: Step S71, using the Latin hypercube sampling method to generate parameter combinations, simulating a predetermined hydrological scenario, obtaining a sample distribution of a joint return period and outputting it as a simulation result; Step S72: for the simulation results, use the numerical integration method to solve the multi-dimensional integral, calculate different types of joint return periods, and obtain the time-varying joint return period estimation considering non-stationarity; Step S73: construct a confidence interval based on the time-varying joint return period estimation; use the quantile method to calculate the interval boundaries of different confidence levels to obtain the uncertainty range of the return period estimation; calculate the relative uncertainty index to obtain a standardized uncertainty measure; Step S74: Based on the uncertainty measurement, the Sobol index method is used to quantify the contribution of each hydrological element to the overall uncertainty, and a sensitivity analysis is performed to obtain the decomposition result of the uncertainty source; Step S75, construct a multi-criteria decision analysis framework, integrate return period estimation and uncertainty decomposition; use the analytic hierarchy process to determine the weight of each criterion, use the TOPSIS method to comprehensively evaluate different scenarios, and obtain the risk ranking that considers multiple factors, which is the comprehensive evaluation; Step S76: Based on the comprehensive evaluation results, use visualization technology to display the risk distribution, adopt scenario analysis method to explore the impact of climate change on future risks, and obtain the final composite flood disaster risk assessment report and decision-making recommendations.

8. A method for studying composite flood disasters in data-deficient estuaries taking into account climate change as claimed in claim 7, It is characterized in that In step S34, a hierarchical clustering method is used to classify extreme value events based on DTW distance to obtain a preliminary set of composite extreme value events, specifically: Step S341, calculating the DTW distance for each pair of hydrological elements, using the FastDTW algorithm to improve the calculation efficiency, especially for long time series, generating three DTW distance matrices, each of which is a×a in size, where a is the number of extreme value events; each pair of hydrological elements includes rainfall-water level, rainfall-wave height, and water level-wave height; Step S342: Treat each extreme value event as a separate cluster, create a priority queue to store the distances between clusters, and sort them from small to large distances; initialize a union-find data structure to track the merging of clusters; Step S343, take out the cluster pair with the minimum distance from the priority queue; merge the two clusters, update and query; calculate the distance between the new cluster and all other clusters: new distance = (size of cluster A * distance between A and C + size of cluster B * distance between B and C) / (size of A + size of B); add the newly calculated distance to the priority queue; until all extreme value events are merged into one cluster: Step S344: for each cluster number k, obtain several clusters according to the cluster tree cutting, calculate the silhouette coefficient of each sample, obtain the silhouette coefficients under different cluster numbers and calculate the average silhouette coefficient, and select the cluster number k with the largest average silhouette coefficient as the optimal cluster number; Step S345, cutting the cluster tree according to the optimal number of clusters, and assigning a unique identifier to each cluster; For each cluster, the cluster center is calculated using the average value of the original time series, and the event with the smallest DTW distance to the cluster center is found as the representative event of the cluster; Step S346, creating an empty composite extreme value event set, for each pair of representative events of hydrological elements: checking whether the time of occurrence of the events overlaps, if so, combining these events into a composite event; adding the combined composite event to the composite extreme value event set; Step S347, calculating the internal evaluation indicators of clustering, including the Davies-Bouldin index and the Calinski-Harabasz index; visualizing the clustering results, generating a clustering dendrogram and a multidimensional scaling diagram; Generate a cluster analysis report, which includes the clustering results of each hydrological element, a list of representative events, a set of composite extreme events, clustering quality assessment indicators, as well as a dendrogram and MDS diagram.

9. A method for studying composite flood disasters in data-deficient estuaries taking into account climate change as claimed in claim 8, It is characterized in that In step S54, for the dynamic Vine Copula structure, the local likelihood estimation method is used to estimate its parameters, specifically: S541, define the window size and sliding step size, create a time window set for the time series; for each time window, extract the corresponding multivariate extreme value data subset; S542, for the data subset in each time window: using the empirical distribution function to convert the original data into pseudo observations, for each candidate Copula family, constructing the log-likelihood function, and selecting the optimal Copula family using the Akaike information criterion, that is, selecting the Copula family with the smallest Akaike information criterion value as the optimal model for the window; S543, for each parameter, construct a time series and apply local polynomial regression to obtain a smoothed time function of the parameter; S544. For each time point and parameter, generate B Bootstrap samples and calculate the Bootstrap standard error: S545, for each time point, estimate the parameters of each candidate Copula family, calculate the Akaike Information Criterion value of each Copula family, and select the Copula family with the smallest AIC as the optimal model at that time point; S546. Divide the time series into several intervals, in each interval the Copula family remains unchanged, and use the hidden Markov model to describe the transformation of the Copula family; generate a time-varying Copula model report, which includes the optimal Copula family and corresponding parameters for each time window, the time-varying function graph of the parameters, the information of the Copula family transformation model, and the model diagnostic information.

Citation Information

Patent Citations

  • Flood combined probability analysis method and application thereof based on Copula

    CN106202788A

  • Most probable combination method of design flood peak amount condition based on Copula function

    CN108875130A

  • Uncertainty analysis method for basin non-uniform design flood under changing conditions

    CN110598315A

  • Rainstorm high-risk zoning method based on hydrometeorological region linear moment frequency analysis

    CN112347652A

  • Multi-station flood process stochastic simulation method considering historical extreme flood characteristics

    CN113946940A