A method and system for simulating the entire process of rainstorm and flood disaster chains in complex scenarios.
By constructing a disaster chain simulation model that combines multi-scale hydrological simulation and deep learning, the problem of insufficient simulation of rainstorm and flood disaster chains in complex scenarios has been solved, and efficient and accurate intelligent inference and early warning of disaster chains have been achieved.
Patent Information
- Application Number
- CN202611253435.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-18
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies suffer from insufficient scale adaptability, scenario limitations, low level of intelligence, and insufficient integration of physical mechanisms and artificial intelligence when simulating rainstorm and flood disaster chains in complex scenarios, making it difficult to achieve efficient and accurate full-process simulation.
A rainstorm and flash flood simulation model is constructed by coupling multi-scale hydrological simulation and deep learning methods. It combines a high-performance multi-dimensional nested flood inundation simulation model and a dam break flood simulation model that considers seepage failure. Multiple models are integrated through a standard protocol interface to construct a disaster chain simulation model. A chain-related disaster relationship model is then constructed using a deep learning network.
It has achieved accurate and efficient simulation across multiple scales and processes, breaking through the intelligent extrapolation of disaster chains in complex scenarios and improving early warning and prevention capabilities.
Smart Images

Figure CN123116798A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing technology, specifically to a method and system for simulating the entire process of a complex rainstorm and flood disaster chain. Background Technology
[0002] Due to the impact of global climate change, localized extreme rainstorm events have become more frequent and severe in recent years. Their characteristics of suddenness, large rainfall, and high destructive potential have led to a continuous increase in the risk of chain disasters triggered by rainstorms, such as flash floods, levee breaches, and dam failures. Internationally, flash flood warnings mostly focus on small watersheds, and my country's mainstream critical rainfall law exhibits a lag. At the model level, existing technologies are mostly limited to simulating single disaster processes or spatial scales. For example, traditional hydrological models lack accuracy in complex terrain, distributed physical models have low computational efficiency, and emerging artificial intelligence methods often suffer from poor interpretability and weak long-term predictive capabilities due to a lack of physical constraints. This results in significant deficiencies in current models in four aspects: scale adaptability, scene coverage, end-to-end intelligence, and the integration of physical mechanisms with artificial intelligence. Consequently, they are unable to support efficient and accurate simulation and extrapolation of the entire disaster chain.
[0003] Therefore, there is an urgent need to build an intelligent simulation and deduction system for the entire process of rainstorm and flood disaster chain, covering "rainfall-runoff-overflow-breach-inundation", to achieve accurate and efficient simulation of multiple scales and processes. Summary of the Invention
[0004] The primary objective of this application is to address at least one of the aforementioned problems by providing a method and system for simulating the entire process of a complex rainstorm and flood disaster chain.
[0005] To achieve the various objectives of this application, the following technical solution is adopted: A method for simulating the entire process of a complex rainstorm and flood disaster chain, provided to meet one of the purposes of this application, includes the following steps: Acquire historical scene data, including hydrological and meteorological data and topographic and working condition data; Using the historical scene data, a rainstorm and flash flood simulation model, a multi-dimensional nested levee inundation simulation model, and a levee breach flood simulation model were constructed. Based on standard protocol interfaces, the aforementioned rainstorm and flash flood simulation model, multi-dimensional nested levee flood inundation simulation model, and levee breach flood simulation model are integrated, and a disaster chain simulation model is constructed. Based on the historical scenario data, the disaster chain simulation model is driven to generate historical rainstorm and flood chain disaster scenarios, and an application scenario set is constructed; A chain disaster relationship model is constructed using deep learning networks, historical scene data, and application scenario sets. In one possible implementation, constructing a rainstorm and flash flood simulation model using the historical scene data includes: Couple the spatiotemporally variable distributed unit hydrograph with the DRIVE model to form a basic hydrological model; By integrating the basic hydrological model and the probabilistic statistical model, a hybrid framework is obtained, and a rainstorm and flash flood simulation model is constructed based on the hybrid framework. Based on the reference watershed control section, the parameters of the rainstorm and flash flood simulation model are calibrated to obtain the rainstorm and flash flood simulation model after parameter calibration.
[0006] In one possible implementation, after obtaining the storm flash flood simulation model with calibrated parameters, the method further includes: The two-dimensional D8 flow direction grid is reconstructed into a one-dimensional effective grid index, retaining only the effective grid points within the watershed that participate in the runoff calculation, and recording the downstream connectivity and hydrological topological order of each grid point; Parallel reconstruction of runoff routing processes from spatial, temporal, and multi-member dimensions includes: In terms of spatial dimension, based on upstream The downstream confluence topology divides the watershed grid into multiple spatial computing layers, ensuring that there is no direct upstream or downstream dependency between grid points within the same spatial computing layer, thus enabling synchronous parallel computing. Different spatial computing layers advance layer by layer according to the direction of water flow propagation, achieving spatial parallelism while satisfying the physical constraints of confluence. In the time dimension, time index parallelization is introduced, combining spatial layer numbers and time step numbers to form a spatial index. The time computation unit enables the model to advance along spatial layers and along time indices, reorganizing the serial advancement relationship into a parallel advancement structure to increase the number of computation units that can be processed simultaneously in each parallel stage. The parallel reorganization structure is the number of spatial layers + the number of time steps – 1. In the multi-member dimension, different parameter combinations, different precipitation forecast members, or different climate scenarios are regarded as independent simulation members. In addition to the spatial layer and temporal index, the member dimension is added for parallel computation, and intermediate state variables are not exchanged between members.
[0007] In one possible implementation, constructing a multi-dimensional nested flood inundation simulation model using the historical scene data includes: Floodplain data were extracted from the historical scene data using geomorphological methods. A two-dimensional hydrodynamic model was constructed based on a Godunov-type finite volume framework. Using the floodplain data as a fast baseline and model domain constraint, the rainstorm and flash flood simulation model is coupled with a two-dimensional hydrodynamic model based on a standard protocol interface and a dynamic adaptive grid to construct a multi-dimensional nested flood inundation simulation model.
[0008] In one possible implementation, the extraction of floodplain data from the historical scene data using geomorphological methods includes: A flood event segmentation method is constructed by using multi-source hydrological data and coupling excess threshold sampling and flow sequence moving variance. The flood event segmentation method is used to extract flood events and corresponding multi-peak flood peak characteristics, process initiation points and recession from the historical scene data. Based on the digital elevation model of the flood event, the confluence network of the river channel is extracted and the catchment area of each grid cell is calculated; For grid cells in the river channel, the potential maximum channel flow depth is determined using a power law relationship based on their catchment area. Neighboring units with elevations lower than their corresponding river levels are marked as floodplains, thus obtaining floodplain data.
[0009] In one possible implementation, constructing a dam breach flood simulation model using the historical scenario data includes: Construct a water level-flow rate combination scenario and obtain experimental observation data of the water level-flow rate combination scenario during a breach flood process. The experimental observation data includes piping seepage failure data, critical soil and water characteristics data, critical stress characteristics data, and three-dimensional morphology data of the breach. Based on the experimental observation data, key disaster-causing factors were identified, and a dam breach flood simulation model was constructed based on these key disaster-causing factors.
[0010] One possible implementation involves, after constructing the disaster chain simulation model, the following: Obtain the water level-discharge observation sequence of the target watershed, and extract the flood discharge threshold of the target watershed using iterative robust regression and the over-threshold method; Based on the aforementioned flood flow threshold, the over-threshold method is used to extract flood events from the daily flow sequence; Based on the extracted flood events, the optimal flow threshold for the occurrence of flood overflow is iteratively searched, and the optimal flood overflow threshold is determined based on the generalized Pareto distribution and Kendall's rank correlation coefficient. Based on all extracted flood events, the preceding precipitation characteristic parameters and the initial soil moisture content state parameters of the watershed corresponding to each flood event are obtained. The preceding precipitation characteristic parameters and the initial soil moisture content state parameters of the watershed are used as model inputs, and whether or not a flood overflows the dike is used as the model output to construct a flood overflow early warning model, with the optimal flood overflow threshold as the trigger condition.
[0011] In one possible implementation, after constructing the disaster chain simulation model, the following steps are taken: based on the Kalman filter double assimilation algorithm, the upstream observation station flow and remote sensing soil water data are fused with the model simulation results to continuously update the parameters and state variables of the disaster chain simulation model.
[0012] In one possible implementation, after constructing the disaster chain simulation model, the process includes: based on multi-source observation data and statistical analysis methods, analyzing the main control factors and / or key watershed attributes affecting flow distribution efficiency of rainstorm to runoff under different flood occurrence mechanisms, quantifying the main control factors and key watershed attributes, and updating the disaster chain simulation model.
[0013] On the other hand, a complex scenario rainstorm and flood disaster chain full-process simulation system provided to meet one of the purposes of this application includes: The data acquisition module is used to acquire historical scene data, which includes hydrological and meteorological data and topographic and working condition data. The first construction module is used to construct a rainstorm and flash flood simulation model, a multi-dimensional nested levee flood inundation simulation model, and a levee breach flood simulation model using the historical scene data. The model integration module is used to integrate the aforementioned rainstorm and flash flood simulation model, multi-dimensional nested levee flood inundation simulation model, and levee breach flood simulation model based on standard protocol interfaces, and to construct a disaster chain simulation model. The model training module is used to drive the disaster chain simulation model to generate historical rainstorm and flood chain disaster scenarios based on the historical scene data, and to build an application scenario set; The second building module is used to construct a chain disaster relationship model using deep learning networks, historical scene data, and application scenario sets. The disaster prediction module is used to obtain the scenario prediction data output by the chain disaster relationship model in response to the input predicted meteorological data.
[0014] On another note, a complex scenario rainstorm and flood disaster chain full-process simulation device, provided to meet one of the purposes of this application, includes a central processing unit and a memory. The central processing unit is used to call and run a computer program stored in the memory to execute the steps of the complex scenario rainstorm and flood disaster chain full-process simulation method described in this application.
[0015] In another aspect, a computer-readable storage medium is provided to suit one of the purposes of this application, the computer-readable storage medium storing computer-executable instructions for causing a computer to perform the method for simulating the entire process of a complex scenario rainstorm and flood disaster chain as disclosed in any of the first aspects of this invention.
[0016] The technical solution of this application has many advantages, including but not limited to the following aspects: This application clarifies the disaster-causing mechanism of extreme rainstorm and flood disaster chains, elucidates the dynamic process of dam seepage failure leading to breach, and establishes a multi-scale rainstorm and flash flood forecasting model coupled with physical mechanisms and data-driven approaches, a high-performance multi-dimensional nested inundation model for dam overflow, and a dam breach flood simulation model considering seepage failure. It overcomes the technical bottlenecks of dynamic integration of multiple models, multi-scale matching, and multi-process coupling in disaster chains under complex scenarios. Based on standard protocol interfaces, it integrates hydrological, hydrodynamic, and breach evolution models to form a disaster chain simulation model and a chain-related disaster relationship model, realizing intelligent simulation of the entire rainstorm and flood disaster chain process. Attached Figure Description
[0017] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a schematic diagram illustrating the steps of a method for simulating the entire process of a complex rainstorm and flood disaster chain according to an embodiment of this application; Figure 2 This is a system block diagram of a method for simulating the entire process of a complex rainstorm and flood disaster chain according to an embodiment of this application; Figure 3 This is a system block diagram of the multi-dimensional nested flood inundation simulation model in the embodiments of this application; Figure 4 This is a flowchart illustrating the multi-dimensional nested flood inundation simulation model in the embodiments of this application; Figure 5 This is a method for dynamically identifying high-risk points of dam breaches in the embodiments of this application; Figure 6 This is a schematic diagram of the complex scenario rainstorm and flood disaster chain simulation system used in this application. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
[0019] The chain reaction of rainstorms and flash floods is one of the most challenging natural disasters facing the world today. Against the backdrop of intensifying climate change and increasingly significant human impacts, the frequency and intensity of extreme rainstorm events continue to rise, leading to a significant increase in the risk of chain disasters such as flash floods, levee breaches, and dam failures. Traditional hydrological models have significant limitations in dealing with complex terrain conditions and extreme rainstorm events, particularly in data-scarce regions and multi-scale coupled simulations. Although various advanced hydrodynamic models and early warning systems have been developed internationally, many bottlenecks remain in simulation accuracy, computational efficiency, and cross-scale integration. Existing rainstorm and flood disaster prediction models are still dominated by classical physical models, but in recent years, distributed hydrological models have rapidly developed towards higher resolutions internationally.
[0020] The core problems faced by this embodiment are mainly reflected in four aspects: First, the scale adaptability problem. Existing models are mostly developed for a single spatial scale, making it difficult to achieve seamless cross-scale connection from watershed to river section. Second, the scenario limitation. Most of this embodiment focuses on a single disaster type, with insufficient coverage of coupled simulation of the flash flood-overtopping-dam failure disaster chain. Third, the low level of intelligence across the entire chain. The complete process from early warning and simulation to assessment still relies heavily on manual intervention. Finally, the integration of physical mechanisms and artificial intelligence is insufficient. Existing intelligent models generally lack physical constraints and deep integration with physical models. The combination of physical constraints with physical models is insufficient, making it impossible to fully explain the flood process and the dominant mechanism, resulting in weak long-term prediction capabilities. These problems seriously restrict the accurate forecasting and effective prevention and control of rainstorm and flood disasters.
[0021] To address the aforementioned technical issues, this application proposes a method for extrapolating the entire process of rainstorm and flood disaster chains in complex scenarios. It addresses the problems of unclear mechanisms governing the occurrence and evolution of extreme rainstorm and flood disaster chains, the time-consuming and low-accuracy simulation of sudden rainstorm and flood disasters, and the lack of comprehensive and intelligent simulation technologies. The method constructs a rainstorm and flash flood simulation model that couples multi-scale hydrological simulation and deep learning methods, a high-performance multi-dimensional nested inundation simulation model for levee overflow, and a levee breach flood simulation model considering seepage failure.
[0022] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments.
[0023] The specific embodiments described below can be combined with each other, and the same or similar concepts or processes will not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.
[0024] See Figure 1 and Figure 2 This application discloses a method for simulating the entire process of a rainstorm and flood disaster chain in complex scenarios. In a typical embodiment, it includes the following steps: 101: Acquire historical scene data, which includes hydrological and meteorological data and topographic and working condition data; Among them, historical scene data consists of historical long-term experimental observation data, hydrological and meteorological data includes station-observed precipitation information, high-resolution forecast precipitation information, and other hydrological and meteorological information; topographic data includes globally consistent fine three-dimensional topographic information, landslide dam information, and land use, vegetation, and soil information.
[0025] like Figure 1 and Figure 2 As shown, the method for simulating the entire process of a complex scenario rainstorm and flood disaster chain in this application includes step 102: using the historical scenario data to construct a rainstorm and flash flood simulation model, a multi-dimensional nested levee flood inundation simulation model, and a levee breach flood simulation model; Specifically, the storm flash flood simulation model, which couples multi-scale hydrological simulation and deep learning methods, mainly identifies key disaster-causing factors by studying the catastrophic patterns of extreme storm flash floods. It improves existing distributed hydrological models by coupling local storm rainfall-runoff dynamic unit hydrographs, constructing a multi-scale, hybrid-resolution storm flash flood simulation model. This embodiment uses a dual-objective parameter calibration method based on flood peak and flood volume, combined with a deep learning parameter transfer method based on watershed similarity, to optimize key parameters of the storm flash flood simulation model, improving simulation accuracy and efficiency.
[0026] The high-performance, multi-dimensional nested overtopping flood simulation model primarily utilizes watershed-level drainage network identification technology and sub-watershed-river network cross-section data transfer technology to construct a one-dimensional river hydrological-hydraulic coupled model. For any sudden or potential overtopping location, this embodiment employs deep learning-based complex river hydrodynamic boundary extraction technology to simulate and obtain the flow and water level processes at key cross-sections in front of the dike. A two-dimensional hydrodynamic model based on GPU parallel computing is developed, employing adaptive grid technology to simulate the surface-water network linked flood evolution process, improving computational efficiency. A nonlinear function is constructed between the river hydrodynamic field and key influencing factors such as topography, riverbed morphology, and temperature. This embodiment reflects an adaptive parameter optimization method that reflects the dynamic changes of hydrodynamic model parameters under complex scenarios, improving the accuracy of inundation simulation.
[0027] This study investigates the dynamic water pressure distribution characteristics of dams under flood impact, considering seepage failure, to clarify the seepage failure modes of dams under different flood flow rates and water levels, and to identify high-risk points for dam failure under flood impact. The physical mechanisms of dam breach formation during torrential rain and flooding are studied, and a characterization and mapping technique for dam physical properties and breach evolution dynamics in a three-dimensional scene model is established. Combining the physical mechanisms of dam seepage failure, a three-dimensional scene simulation technique for dam failure is developed, enabling dynamic extrapolation of breach morphology (top width, bottom width, and vertical depth) under flood impact. Finally, the coupling and integration technique between the hydrodynamic model and the dynamic evolution model of dam breaches is studied to simulate the flood inundation process under different breach scenarios.
[0028] like Figure 1 and Figure 2 As shown, the method for simulating the entire process of a complex scenario rainstorm and flood disaster chain in this application includes step 103: based on the standard protocol interface, integrating the rainstorm and flash flood simulation model, the multi-dimensional nested overtopping flood inundation simulation model, and the dam breach flood simulation model, and constructing a disaster chain simulation model; By coupling multi-scale hydrological simulations, key technologies such as dynamic unit hydrodynamic generation and confluence coupling, high-performance multi-dimensional nested overtopping flood simulation, and dam failure mechanism considering seepage damage have been overcome, ultimately achieving the organic integration of multiple models. This enables rapid response to simulations of different flood stages, including flash floods, overtopping, and dam failure.
[0029] like Figure 1 and Figure 2 As shown, the method for simulating the entire process of a complex rainstorm and flood disaster chain in this application includes step 104: based on the historical scenario data, driving the disaster chain simulation model to generate historical rainstorm and flood disaster scenarios, and constructing an application scenario set; Most existing studies focus on single disaster types and lack sufficient coverage of coupled simulations of flash flood-levee overflow-dam failure disaster chains. The application scenario set of rainstorm and flood disaster chains is constructed based on historical flood events. It can characterize the memory of rainstorm and flood disaster chains in different scenarios, clarify the similarity of different flood events and the correlation of chain-related disasters, and prepare for the training of subsequent chain-related disaster relationship models.
[0030] like Figure 1 and Figure 2 As shown, the method for extrapolating the entire process of a complex rainstorm and flood disaster chain in this application includes step 105: constructing a chain-generated disaster relationship model using deep learning networks, historical scene data, and application scenario sets; By using deep learning methods and application scenario sets as training sets to train the model, a relationship model between disaster-causing factors, peak flood flow and chain disasters (breach and flood) is constructed.
[0031] like Figure 1 and Figure 2 As shown, the method for extrapolating the entire process of a complex scenario rainstorm and flood disaster chain according to this application includes step 106: responding to the input predicted meteorological data, obtaining the scenario prediction data output by the chain disaster relationship model. The predicted meteorological data includes meteorological disaster-causing factors and peak flow.
[0032] When receiving forecast meteorological data that needs to be predicted, the trained disaster chain simulation model and chain-related disaster relationship model can realize the memory matching of rainstorm and flood disaster chain application scenarios, intelligent response of multiple models and multiple scenarios of flood disaster chain, and autonomous automatic simulation of rainstorm and flood disaster chain.
[0033] like Figure 3 As shown in the embodiments of this application, a storm flash flood model coupled with multi-scale hydrological simulation and deep learning methods is also constructed. Existing technologies commonly use flash flood monitoring index systems centered on critical rainfall amounts. However, this method based on critical rainfall amounts exhibits significant lag and is difficult to effectively address hydrological responses under complex terrain conditions and extreme rainstorm events. Regarding runoff generation module research, traditional models generally rely on simplified theoretical assumptions and empirical formulas, resulting in insufficient accuracy in simulating nonlinear hydrological responses under complex terrain and extreme rainstorm conditions. In terms of confluence simulation technology, traditional hydrological models mainly calculate river confluence based on kinematic wave or diffuse wave equations, but their accuracy is poor in simulating rapid confluence processes under complex mountain terrain conditions.
[0034] In this embodiment, the present application embodiment constructs a rainstorm and flash flood simulation model, including the following steps: Couple the spatiotemporally variable distributed unit hydrograph with the DRIVE model to form a basic hydrological model; By integrating the basic hydrological model and the probabilistic statistical model, a hybrid framework is obtained, and a rainstorm and flash flood simulation model is constructed based on the hybrid framework. Based on the reference watershed control section, the parameters of the rainstorm and flash flood simulation model are calibrated to obtain the rainstorm and flash flood simulation model after parameter calibration.
[0035] Specifically, the core building blocks of a spatiotemporally variable distributed unit hydrograph are the dynamic spatial distribution of runoff time (or velocity) and net rainfall. This embodiment establishes a physical process-based, rainfall-driven time-varying distributed velocity field calculation framework. This framework achieves a more physical process-compliant description by distinguishing the dynamic mechanisms of slope flow and river flow, and considers the velocity variation along the river runoff path. Simultaneously, considering the significant impact of time-varying rainfall intensity on velocity at the watershed scale, net rainfall intensity is introduced as a driving factor. Specifically, slope velocity... The estimation is based on the steady-state motion wave approximation and Manning's formula; river flow velocity Based on the Manning formula, assuming that the upstream cumulative drainage volume equals the product of the upstream drainage area and the vertical net inflow flux, the specific formula is as follows: ; ; in, The length m of the flow path within each unit is determined by the direction of water flow. The roughness coefficient of Manning is determined by the land cover type; Slope; It is the upstream catchment area ; Represents the average net rainfall intensity of the watershed (mm / h); This is the wet perimeter (m), assuming the river channel has a wide and shallow rectangle in cross-section. Approximately equal to the river channel width. After obtaining the spatially distributed velocity field, the residence time of each grid is accumulated along the water flow path ( By doing so, the confluence time field from the outlet to the entire basin can be obtained.
[0036] Based on the temporal distribution of runoff, unit hydrographs were further constructed using the time-area method. To reflect the impact of spatiotemporal variability in rainfall, the unit hydrographs were constructed by tracking and accumulating net rainfall generated at different times and spatial locations. The contribution of watershed outlet runoff to the basin. Specifically, first calculate... This term represents the time before time t. The total net rainfall reaching the watershed outlet during time period t. Then, the unit hydrograph ordinate reflecting the characteristics of this rainfall event is obtained using a normalization formula. : ; This embodiment further introduces the linear reservoir method to calculate the unit hydrograph in order to simulate the water storage and attenuation effects of the watershed.
[0037] The core of the coupling between the spatiotemporally variable distributed unit line and the DRIVE model lies in, for example, Figure 3 As shown, the net rainfall intensity field output by the DRIVE model is the direct input and dynamic driving force for constructing the unit hydrograph. The net rainfall intensity field output by DRIVE is used, on the one hand, to parameterize slope and river flow velocities, generating a confluence time field reflecting the hydraulic characteristics of rainfall events; on the other hand, its spatiotemporally distributed net rainfall sequence (… The data is used as actual input and dynamically calculated in this runoff time field to construct the final unit hydrograph. This ensures that the unit hydrograph is no longer a static feature inherent to the watershed, but a dynamic response function for a specific rainfall event, thus achieving a close coupling between runoff generation and runoff confluence processes based on physical mechanisms, resulting in a basic hydrological model.
[0038] Furthermore, this embodiment proposes a hybrid framework integrating basic hydrological models and probabilistic statistical models, aiming to improve sub-seasonal to seasonal (S2S) flood forecasting capabilities by assimilating statistical hydro-climate relationships. The fusion of physical and statistical models effectively alleviates the limitations of single models, such as the precipitation error accumulation of the DRIVE model and the lack of physical processes in the Bayesian Joint Probabilistic Model (BJP model), significantly improving S2S runoff forecasting skills.
[0039] Among them, the DRIVE hydrological model, by coupling the graded main runoff path algorithm, the runoff model (DRTR) and the VIC land surface process runoff generation model, can conveniently perform continuous and refined numerical simulations of flood peak time, flow rate and inundation range and depth.
[0040] Based on the reference watershed control section, the parameters of the rainstorm and flash flood simulation model are calibrated to obtain the rainstorm and flash flood simulation model after parameter calibration.
[0041] Specifically, after the model was built, hydrological calibration was performed on major control sections with more than 5 years of data in each watershed. In the single-objective calibration scheme, the simulation accuracy of the flow time series was the sole optimization objective, and the efficiency coefficient (NSE) was used as the calibration index. The calibration algorithm employed a classic global optimization algorithm and an efficient local convergence algorithm to verify the model's stability and convergence speed under global and local optimization, respectively. Parameter calibration mainly focused on six core parameters in the VIC model that are closely related to runoff generation and underlying surface hydrological processes, including the shape characteristic parameter of infiltrated soil water storage capacity, the maximum baseflow value, the proportion of baseflow to the maximum baseflow value during nonlinear growth, the ratio of bottom soil moisture content to maximum soil moisture content (Ws) when nonlinear growth of baseflow occurs, and the thickness of the first and second soil layers. Through model calibration, the problem that the commonly used objective function based on the sum of squared deviations between measured and simulated runoff values over many years is difficult to simultaneously and accurately reflect the peak and volume characteristics of flood processes was effectively solved, and an effective parameter migration scheme based on watershed similarity was established.
[0042] In the area of stormwater and flash flood simulation, this application completed the design and development of a dynamic unit hydrograph model and achieved deep coupling with the DRIVE runoff generation module, improving the simulation accuracy and efficiency of stormwater and flood processes in mountainous watersheds. Furthermore, a hybrid framework integrating a distributed hydrological model and a probabilistic statistical model was proposed. By assimilating statistical hydro-climate relationships, the forecasting skills for sub-seasonal to seasonal runoff and floods were significantly improved, providing a feasible technical path to overcome the traditional "forecasting desert." A high-resolution distributed hydrological model was constructed based on a hierarchical main channel runoff path algorithm. Through single-objective calibration based on flow rate, the simulation accuracy of the model at major control sections was improved, laying a solid foundation for subsequent regionalized parameter studies.
[0043] like Figure 3 As shown in the embodiments of this application, a high-performance multidimensional nested flood overflow model is also constructed. As a cutting-edge direction in the current field of hydrodynamic simulation, the multidimensional nested flood overflow model achieves high-precision simulation of the flood evolution process by coupling a one-dimensional river channel model and a two-dimensional floodplain model. However, its simulation parameterization scheme is still limited by two core factors: one is the spatial resolution of the river channel topography data, and the other is the parameterization scheme of the roughness coefficient.
[0044] In this embodiment, as Figure 4 As shown, the high-performance multidimensional nested overtopping flood model constructed in this application embodiment includes: 1021. Floodplain data is extracted from the historical scene data using the geomorphological method; Specifically, in mountainous watersheds lacking detailed hydrological data, rapidly determining floodplain baselines and accurately characterizing flood dynamics is crucial for disaster prevention and mitigation. This embodiment first utilizes geomorphological methods to identify floodplains, using this as a rapid baseline constraint. Building upon this, it leverages the dynamic advantages of coupled hydrological and hydrodynamic models in characterizing rainfall-runoff generation and flood wave evolution, constructing a comprehensive analytical framework from static feature identification to dynamic process deduction.
[0045] This embodiment first identifies floodplains using geomorphological methods. The core principle is to use Leopold's hydraulic geometry law to calculate the maximum possible water depth of the river channel based on the upstream catchment area (i.e., the total area of the watershed flowing into the river section), and then determine the range that may be flooded.
[0046] This embodiment is based on the SRTM digital elevation model (DTM), extracts the runoff network and calculates the catchment area of each grid cell; for the river grid cell, the potential maximum river flow depth is estimated using a power law relationship based on its catchment area; neighboring cells with elevations lower than their corresponding river water levels are marked as floodplains, resulting in a floodplain range with a resolution of 250m. A hydrological-hydrodynamic coupled model is initially constructed based on a dynamic adaptive grid, and inundation process analysis is performed on the selected typical flood events.
[0047] 1022 A two-dimensional hydrodynamic model was constructed based on a Godunov-type finite volume framework; Specifically, the model employs two-dimensional shallow water equations that satisfy harmony properties and uses a Godunov-type finite volume framework. It considers complex mixed flow patterns and intermittent flow problems, possessing the ability to easily handle complex boundary and topographic features while maintaining stability and ease of handling wet-dry interfaces. This allows for the simulation and analysis of flood processes under complex topographical environments. Based on the numerical discretization solution of the two-dimensional shallow water equations, the model reflects the propagation and storage / discharge relationship of flood waves at the land surface, enabling dynamic forecasting of inundation range, water level, and spatial distribution of flow velocity.
[0048] 1023 uses the floodplain data as a fast baseline and model domain constraint, and couples the rainstorm and flash flood simulation model with the two-dimensional hydrodynamic model based on the standard protocol interface and dynamic adaptive grid to construct a multi-dimensional nested flood inundation simulation model.
[0049] Specifically, this coupled model uses DRIVE's distributed hydrological module to simulate the runoff generation and confluence process of the watershed. The calculated inflow is used as the boundary condition of DRIVE's one-dimensional hydrodynamic module to simulate the evolution of river floods and the spatiotemporal changes of hydrodynamic elements such as water level and flow. At the same time, the surface runoff generated by the hydrological module and the river flow and water level processes calculated by the one-dimensional hydrodynamic module are used as the boundary conditions and source terms of the two-dimensional hydrodynamic model to achieve high-resolution simulation of the dynamic evolution of floodplains, surface water accumulation and inundation range.
[0050] The construction of a two-dimensional hydrodynamic model mainly includes three parts: basic data processing, computational grid generation, and boundary condition setting. The specific details are as follows: The basic data includes topographic data, used to define the model's computational domain and characterize surface and riverbed elevations. This study uses a Digital Earth Model (DEM) as the model's topographic input data. This data can accurately reflect changes in surface elevation and topographic relief, providing fundamental topographic data for constructing the computational domain and simulating hydrodynamics in the study area. Before model construction, the DEM needs necessary preprocessing. If the DEM shows roads, embankments, or other features above the riverbed, but actual waterways exist, manual correction of the local topography is required to disconnect or lower the points obstructing water flow to the corresponding connectivity elevation; otherwise, water flow in the model may be unable to pass normally due to topographic obstruction. The main model parameter is the Manning roughness coefficient. Based on high-resolution remote sensing imagery or land survey data, a land use classification map of the study area is generated through supervised classification or visual interpretation, and a lookup table of land use types and Manning roughness coefficient (n) values is established. This study references typical values for most natural rivers and sets the Manning coefficient to 0.035.
[0051] Computational mesh generation is a crucial step in constructing a two-dimensional hydrodynamic model. Its purpose is to generate an adaptive mesh system that accurately fits complex terrain boundaries, captures minute water flow channels, and achieves a balance between computational efficiency and numerical accuracy. In this study, the outer boundary line is defined based on the historical maximum inundation range, with the boundary area moderately larger than the estimated flood inundation zone to prevent non-physical "accumulation" or "rebound" of water flow at the boundary due to computational truncation.
[0052] The boundary condition settings for a two-dimensional hydrodynamic model mainly include the upstream inflow boundary and initial conditions. For river flood or flash flood simulations, the upstream boundary is typically provided by a time-series file of flow variation over time. This flow series is generally provided by measured data from upstream hydrological stations or calculated by a watershed hydrological model driven by meteorological rainfall forecasts. In this study, the upstream boundary was calculated using a hydrological model. The initial conditions mainly include the initial water depth within the computational domain at the start of the simulation, used to provide the model with the initial computational state and avoid abnormally accelerated flood wave propagation on a completely dry riverbed, which could produce spurious short-duration flood peaks. In this study, the initial conditions of the river channel were simulated using a one-dimensional channel model.
[0053] In a further implementation, floodplain data is extracted from the historical scene data using geomorphological methods, including: A flood event segmentation method is constructed by using multi-source hydrological data and coupling excess threshold sampling and flow sequence moving variance. The flood event segmentation method is used to extract flood events and corresponding multi-peak flood peak characteristics, process initiation points and recession from the historical scene data. In this embodiment, the multi-source hydrological data includes flow data from multiple hydrological stations based on time series, and runoff data from the areas where the hydrological stations are located. The Theil-Sen slope method is used to analyze these data to obtain the spatiotemporal evolution characteristics of meteorological elements and vegetation in the region.
[0054] Based on the multi-source hydrological data, a flood event segmentation method is constructed that couples excess threshold sampling and flow sequence moving variance. The flood event segmentation method is used to capture the characteristics of multi-peak flood peaks. Specifically, the flood event segmentation method that couples over-quantitative threshold sampling with flow series moving variance can filter flood events from daily runoff series. This method can accurately identify the rise and fall of floods, and is especially suitable for flood events with multiple peaks and complex morphologies, providing support for subsequent analysis of flood behavior characteristics and identification of flood types in demonstration areas.
[0055] Based on the digital elevation model of the flood event, the runoff network of the river channel is extracted and the catchment area of each grid cell is calculated. For the grid cells on the river channel, the potential maximum river flow depth is determined using a power law relationship based on their catchment area. Neighboring cells with elevations lower than their corresponding river water levels are marked as floodplains to obtain floodplain data.
[0056] Among them, floodplain data includes four major categories of characteristic indicators to characterize the complete flood process: magnitude, time domain, rate of change, and process morphology. Magnitude includes peak flow and flood volume, time includes duration of flood event and peak time, rate of change includes rise rate and precipitation rate, and process morphology includes the proportion of high flow duration.
[0057] In terms of flood inundation simulation, this embodiment, based on multi-source data collected from a thematic study, systematically analyzed the spatiotemporal evolution characteristics of meteorological elements and vegetation in the demonstration area using the Theil-Sen slope method. By constructing an innovative method coupling over-quantitative threshold sampling and flow series moving variance, this embodiment accurately identified typical historical flood events and comprehensively characterized the flood process using seven feature indicators. Based on this, this embodiment successfully constructed a two-dimensional hydrodynamic model, performing a detailed simulation of the typical flood inundation evolution process in the target watershed. This represents a crucial step towards achieving high-performance, multi-dimensional nested flood inundation simulation. The innovation of this embodiment lies in constructing a flood event segmentation method coupling over-quantitative threshold sampling and flow series moving variance, filtering flood events from diurnal runoff sequences. This method can accurately identify the rise and fall processes of floods, and is particularly suitable for multi-peak, complex-form flood events, providing support for subsequent analysis of flood behavior characteristics and flood type identification in the demonstration area. Simultaneously, by coupling the distributed hydrological model DRIVE and the hydrodynamic model, preliminary flood inundation simulation analysis of the demonstration area was achieved.
[0058] like Figure 5 As shown in the embodiments of this application, a dam-break flood model considering seepage failure is also constructed. Dam-break flood simulation models are important tools for studying the dam-break process and its disaster impact under extreme conditions. The core challenge lies in the complexity of the seepage failure mechanism. Current research still faces three major challenges: First, the multiphysics coupling mechanism is not fully elucidated, especially the dynamic interaction process of seepage-stress-deformation fields; second, it is difficult to balance computational efficiency and accuracy, with the computational resource requirements for large-scale simulations growing exponentially; and third, the scarcity of real-time monitoring data means that model validation mainly relies on historical cases, limiting the model's universality and predictive ability.
[0059] In this embodiment, the dam breach flood simulation model constructed in this application includes: Construct a water level-flow rate combination scenario and obtain experimental observation data of the water level-flow rate combination scenario during a breach flood process. The experimental observation data includes piping seepage failure data, critical soil and water characteristics data, critical stress characteristics data, and three-dimensional morphology data of the breach. Specifically, the component water level-combination scenario mainly utilizes self-developed equipment to conduct full-process observation experiments on piping occurrence, development, and dam failure. During the experiment, multiple sets of upstream water level scenarios are preset to determine the critical water level (or hydraulic gradient) for piping in different types of soil. Under the conditions of piping channel formation and continuous piping development, the deformation, instability, and eventual failure of the earth-rock dam are continuously observed. Measured data on soil temperature, noise, piping location, piping point diameter, water / sand inflow volume, and earth-rock dam settlement are obtained and analyzed using thermal infrared and ultrasonic monitoring devices and real-time image and video recording.
[0060] Based on the experimental observation data, key disaster-causing factors were identified, and a dam breach flood simulation model was constructed based on these key disaster-causing factors.
[0061] Specifically, by using experimental observation data, we can understand the microscopic mechanism of the impact and seepage development of the dike foundation under the impact of floods in the target watershed, identify the key disaster-causing factors affecting the dike failure, and establish a dynamic identification method for high-risk points based on the fusion of multi-objective watershed indicators.
[0062] Among them, the dam breach flood simulation model is based on Smoothed Particle Hydrodynamics (SPH method) to simulate flood disasters. Traditional SPH suffers from problems such as numerical oscillations in the pressure field, instability in interface capture, and difficulties in boundary treatment, especially in complex flows such as strong impacts, large deformations, and multiphase flows. To improve numerical stability and accuracy, researchers have proposed introducing density dissipation terms and artificial viscosity terms into SPH, namely... method, It can effectively suppress pressure oscillations, improve interface capture capability, and enhance numerical stability, making it suitable for seepage dam failure problems.
[0063] In the study of dam failure mechanisms, this embodiment constructs a complete basic database, providing reliable support for indoor physical model target watershed experiments. This embodiment innovatively adopts a technical approach combining multi-scale physical model experiments with numerical simulation of the target watershed. Through a series of small-, medium-, and prototype observations, it systematically explores the entire process of seepage-instability-failure in target watersheds with different soil materials under various hydrological conditions. This embodiment focuses on revealing the microscopic mechanism of seepage development in the target watershed under flood impact, identifying key disaster-causing factors affecting dam failure, and establishing a dynamic identification method for high-risk points based on the fusion of multi-target watershed indicators. This provides important technical support for dam failure flood modeling and rapid early warning of hazards. The innovation of this embodiment lies in proposing a systematic analysis method for the dynamic water pressure distribution of dams and the characteristics of target watersheds under flood impact. By designing physical model target watersheds and numerical experiments under different flood level-flow combination scenarios, it reveals the driving process of seepage failure from a dynamic mechanism perspective, providing theoretical support for the construction of target watersheds for breach evolution models and laying a methodological foundation for subsequent model coupling and practical simulation of breach floods.
[0064] With the widespread application of watershed-scale hydrological simulations in flood forecasting, water resource management, and climate change response, higher demands are being placed on the spatiotemporal resolution and simulation capabilities of stormwater and flash flood simulation models. While high-resolution hydrological models can more accurately capture watershed topography and hydrodynamic processes, they also significantly increase computational complexity and memory requirements, making traditional serial computation extremely inefficient in large-scale, long-term time-series simulations. Especially in runoff routing processes, the strict upstream-downstream flow dependency makes parallelization difficult to achieve efficient load balancing and thread utilization, limiting the widespread application of these models in conventional computing environments.
[0065] To address the challenge of low computational efficiency in large-scale watershed simulations using high-resolution hydrological models, this application proposes a step-by-step spatiotemporal-multi-member parallel acceleration method for fine-grained hydrological simulations. This method optimizes the storm flood simulation model after parameter calibration. Through the synergistic optimization of spatial layering, temporal indexing, and multi-member parallelism, the simulation efficiency is successfully increased to up to 84 times that of traditional serial methods, providing an efficient and feasible computational solution for the practical application of large-scale fine-grained hydrological simulations. This includes: The two-dimensional D8 flow direction grid is reconstructed into a one-dimensional effective grid index, retaining only the effective grid points within the watershed that participate in the runoff calculation, and recording the downstream connectivity and hydrological topological order of each grid point; thereby reducing invalid grid loops and memory access overhead.
[0066] Parallel reconstruction of runoff routing processes from spatial, temporal, and multi-member dimensions includes: In terms of spatial dimension, based on upstream The downstream confluence topology divides the watershed grid into multiple spatial computing layers, ensuring that there is no direct upstream or downstream dependency between grid points within the same spatial computing layer, thus enabling synchronous parallel computing. Different spatial computing layers advance layer by layer according to the direction of water flow propagation, achieving spatial parallelism while satisfying the physical constraints of confluence. In the time dimension, time index parallelization is introduced, combining spatial layer numbers and time step numbers to form a spatial index. The time calculation unit enables the model to advance along spatial layers and along time indices, thus reorganizing the original serial advancement relationship of "number of spatial layers × number of time steps" into "number of spatial layers + number of time steps". The parallel progression structure of "1" significantly increases the number of computing units that can be processed simultaneously in each parallel stage, reducing thread waiting and uneven load issues.
[0067] In the multi-member dimension, different parameter combinations, different precipitation forecast members, or different climate scenarios are regarded as independent simulation members. In addition to the spatial layer and temporal index, the member dimension is added for parallel computation. Intermediate state variables are not exchanged between members. Multiple sets of simulations can be completed simultaneously in a shared memory environment with less synchronization overhead. Therefore, it is particularly suitable for tasks that require a large number of repeated model runs, such as ensemble flood forecasting, parameter calibration, and uncertainty analysis.
[0068] Ultimately, this method for accelerating refined hydrological simulation achieves joint parallelism of spatial layers, temporal indices, and multi-member arrays on a shared memory system using OpenMP. This improves the computational efficiency and resource utilization of large-scale high-resolution hydrological simulations while maintaining the physical continuity of the gridded confluence process.
[0069] This method can not only significantly shorten the simulation time, but also significantly improve the computational power of hydrological models while ensuring simulation accuracy, and has broad application prospects.
[0070] In some implementations, to address the intelligent activation problem of the overtopping flood module in the rainstorm flood disaster chain model, a data-driven overtopping flood early warning model is constructed, which determines whether the overtopping flood inundation process has occurred based on the watershed's previous precipitation and soil moisture content. Specifically, after constructing the disaster chain simulation model, the following is also included: Obtain the water level-discharge observation sequence of the target watershed; fit the water level-discharge power law relationship curve through iterative robust regression, identify the water level inflection point based on the discharge residual curvature, and determine the discharge threshold of the target watershed, which is the discharge threshold corresponding to the flood event.
[0071] Based on the flood event flow threshold, the flood event is extracted from the daily flow sequence using the over-threshold method: the peak interval between adjacent events is not less than 3 days, and if there are multiple adjacent peaks, the maximum value is retained.
[0072] Based on the extracted flood events, the optimal flow threshold for flood inundation is iteratively searched: different candidate thresholds are tried sequentially, dividing flood events into two groups: those exceeding the threshold and those below the threshold. A generalized Pareto distribution (GPD) is fitted to the events exceeding the threshold, and its goodness of fit is evaluated using the negative log-likelihood value. Simultaneously, the Kendall's rank correlation coefficient (τ) between peak flow and peak time in the events exceeding the threshold is calculated. The optimal threshold should simultaneously satisfy the highest GPD fit and Kendall's τ being close to 0; this is taken as the final optimal inundation threshold for flood inundation.
[0073] Based on all extracted flood events, the preceding precipitation characteristic parameters and the initial soil moisture content state parameters of the watershed corresponding to each flood event are obtained. The preceding precipitation characteristic parameters and the initial soil moisture content state parameters of the watershed are used as model inputs, and whether or not a flood overflows the dike is used as the model output to construct a flood overflow early warning model, with the optimal flood overflow threshold as the trigger condition.
[0074] The input features include two categories: previous precipitation features and initial watershed state features. Previous precipitation features include daily precipitation, cumulative precipitation, maximum precipitation, and previous precipitation index for the previous 1 to 7 days. Initial watershed state features are provided by a storm flood simulation model, including previous soil moisture content and its 7-day average. Historical observed precipitation and its driven hydrological model soil moisture content simulation results are used as ideal forecast inputs. An L1 regularized logistic regression algorithm is employed, with L1 penalty used to compress redundant variable coefficients to zero, selecting a subset of key disaster-causing features to mitigate multicollinearity. When there are sufficient urban flooding event samples during the training period (≥20 positive samples), a gradient boosting tree (GBDT) is used to capture nonlinear patterns, and a grid search on the validation set is used to determine the linear combination weights of logistic regression and GBDT (the sum of the weights is 1). When samples are insufficient, only regularized logistic regression is retained. Threshold decision-making is optimized on an independent validation set, using the F2 score as a benchmark and tilting towards the recall rate of urban flooding to determine two-level early warning thresholds for normal conditions and urban flooding, prioritizing the ability to capture flooding events.
[0075] This early warning module establishes a benchmark for evaluating early warning capabilities driven by pure precipitation, laying the foundation for subsequent coupled numerical precipitation forecasting.
[0076] In some implementations, due to the difficulty of fully characterizing the complex impacts of human activities (such as irrigation, reservoir / dam scheduling, and the activation of flood storage areas) by physical process models, various uncertainties are inevitably introduced during flood simulation, further increasing the difficulty of simulating watershed runoff generation and confluence processes. To improve the accuracy of flood simulation, this application's embodiments use a Kalman filter dual assimilation algorithm to fuse upstream observation station flow and remote sensing soil water data with model simulation results, continuously updating the parameters and state variables of the disaster chain simulation model. The results show that assimilating flow observations from tributary stations of different levels can improve the simulation effect of downstream stations to a certain extent, with KGE increases exceeding 0.4.
[0077] In some implementations, this embodiment conducts a systematic and in-depth analysis focusing on the core scientific question of "the relationship between rainstorm conversion efficiency and climate characteristics." First, the occurrence mechanisms of flood events in the target area of this embodiment were identified and classified, clearly distinguishing different types of rainstorm-induced flood processes. Subsequently, representative watersheds were selected for quantitative analysis, comprehensively considering multiple factors such as meteorology, topography, soil, and underlying surface, identifying 10 key influencing factors that play a dominant role in flood formation. Based on multi-source observation data and statistical analysis methods, the main controlling factors of rainstorm-to-runoff conversion efficiency under different flood occurrence mechanisms were systematically analyzed, and the relative importance and differences in the effects of each factor were further quantified.
[0078] In some implementations, this embodiment conducted in-depth data-driven research focusing on the core scientific question of "how watershed characteristics affect flood event dynamics." It analyzed key watershed attributes influencing flow distribution and quantified their relative importance and nonlinear impact relationships. By constructing a complete analytical framework, it systematically identified and explained 13 key watershed attributes related to the baseflow parameter N (a larger value indicates a greater proportion of surface runoff in the total flow). The results of this embodiment not only reveal the dominant influence of watershed area but also clearly characterize the complex nonlinear relationships between climate, soil, and other factors and baseflow dynamics, providing a solid scientific basis and practical methodology for understanding and predicting hydrological behavior in data-free areas.
[0079] The main control factors and key watershed attributes are quantified and the disaster chain simulation model is updated, adjusting or increasing / decreasing the weights of the corresponding main control factors and key watershed attributes.
[0080] Please see Figure 6 According to one aspect of this application, a system for simulating the entire process of a complex scenario of rainstorm and flood disaster chain is provided, the system comprising: Data acquisition module 201 is used to acquire historical scene data, which includes hydrological and meteorological data and topographic and working condition data; The first construction module 202 is used to construct a rainstorm and flash flood simulation model, a multi-dimensional nested levee flood inundation simulation model, and a levee breach flood simulation model using the historical scene data. The model integration module 203 is used to integrate the rainstorm and flash flood simulation model, the multi-dimensional nested levee flood inundation simulation model, and the levee breach flood simulation model based on the standard protocol interface, and to construct a disaster chain simulation model. The model training module 204 is used to drive the disaster chain simulation model to generate historical rainstorm and flood chain disaster scenarios based on the historical scene data, and to construct an application scenario set; The second building module 205 is used to build a chain disaster relationship model using deep learning networks, historical scene data and application scenario sets; The disaster prediction module 206 is used to obtain the scenario prediction data output by the chain disaster relationship model in response to the input predicted meteorological data.
[0081] Another embodiment of this application provides a device for simulating the entire process of a complex scenario rainstorm and flood disaster chain. This device includes a processor, a computer-readable storage medium, a memory, and a network interface connected via a system bus. The computer-readable, non-volatile storage medium of this device stores an operating system, a database, and computer-readable instructions. The database can store information sequences. When the computer-readable instructions are executed by the processor, the processor can implement a method for simulating the entire process of a complex scenario rainstorm and flood disaster chain.
[0082] The processor of this complex-scenario rainstorm and flood disaster chain simulation device provides computing and control capabilities, supporting the operation of the entire device. The device's memory can store computer-readable instructions, which, when executed by the processor, enable the processor to perform the complex-scenario rainstorm and flood disaster chain simulation method of this application. The device's network interface is used for communication with a terminal.
[0083] In this embodiment, the processor is used to execute... Figure 6 The system defines the specific functions of each module, and the memory stores the program code and various data required to execute these modules or submodules. The network interface is used to enable data transmission between user terminals or servers.
[0084] The non-volatile readable storage medium in this embodiment stores the program code and data required to execute all modules in the complex scenario rainstorm and flood disaster chain simulation system of this application. The server can call the program code and data of the server to execute the functions of all modules.
[0085] This application also provides a non-volatile readable storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the complex scenario rainstorm and flood disaster chain simulation method of any embodiment of this application.
[0086] This application also provides a computer program product, including a computer program / instructions that, when executed by one or more processors, implement the steps of the method described in any embodiment of this application.
[0087] The key features of the embodiments described in this application are: First, we completed the design of a rainstorm and flash flood model based on a spatiotemporally variable distributed unit line and deeply coupled it with the DRIVE runoff generation module. We constructed a 1km / 3h resolution distributed model in the target area and carried out single-objective, runoff component dual-objective, and "flow-remote sensing evapotranspiration" dual-objective calibration. This significantly improved the simulation accuracy of flood processes at control sections and proposed a new calibration scheme that takes into account water balance constraints.
[0088] Second, based on long-series meteorological and NDVI data, typical flood events in the demonstration area were identified, and behavioral characteristics such as flood peak magnitude, duration, and rise and fall rate were extracted. A two-dimensional hydrodynamic model was constructed and typical flood inundation simulations of the target watershed were carried out. The results were highly consistent with those of the geomorphological floodplain, laying the foundation for subsequent GPU-accelerated multi-dimensional nested overtopping flood simulations.
[0089] Third, starting from the multi-physics coupling of seepage-stress-deformation, a numerical model of SPH seepage-break dam considering seepage softening was constructed, and the model was completed. Key algorithms such as artificial stress and boundary treatment were implemented and verified by typical examples. At the same time, an indoor physical test scheme and observation index system for the entire process of piping and embankment failure were designed to provide data support for revealing the seepage failure mechanism of embankment foundation and the evolution law of breach.
Claims
1. A complex scene storm flood disaster chain whole process deduction method, characterized in that, Includes the following steps: Acquire historical scene data, including hydrological and meteorological data and topographic and working condition data; Using the historical scene data, a rainstorm and flash flood simulation model, a multi-dimensional nested levee inundation simulation model, and a levee breach flood simulation model were constructed. Based on standard protocol interfaces, the aforementioned rainstorm and flash flood simulation model, multi-dimensional nested levee flood inundation simulation model, and levee breach flood simulation model are integrated, and a disaster chain simulation model is constructed. Based on the historical scenario data, the disaster chain simulation model is driven to generate historical rainstorm and flood chain disaster scenarios, and an application scenario set is constructed; A chain disaster relationship model is constructed using deep learning networks, historical scene data, and application scenario sets. In response to the input predicted meteorological data, the scenario prediction data output by the chain disaster relationship model is obtained.
2. The complex scene rainstorm flood disaster chain whole process deduction method according to claim 1, characterized in that, The process of constructing a rainstorm and flash flood simulation model using the historical scene data includes: Couple the spatiotemporally variable distributed unit hydrograph with the DRIVE model to form a basic hydrological model; By integrating the basic hydrological model and the probabilistic statistical model, a hybrid framework is obtained, and a rainstorm and flash flood simulation model is constructed based on the hybrid framework. Based on the reference watershed control section, the parameters of the rainstorm and flash flood simulation model are calibrated to obtain the rainstorm and flash flood simulation model after parameter calibration.
3. The complex scene rainstorm flood disaster chain whole process deduction method according to claim 2, characterized in that, After obtaining the storm flash flood simulation model with calibrated parameters, it also includes: The two-dimensional D8 flow direction grid is reconstructed into a one-dimensional effective grid index, retaining only the effective grid points within the watershed that participate in the runoff calculation, and recording the downstream connectivity and hydrological topological order of each grid point; Parallel reconstruction of runoff routing processes from spatial, temporal, and multi-member dimensions includes: In the spatial dimension, according to the upstream The downstream confluence topological relationship divides the drainage grid into multiple spatial computing layers, so that there is no direct upstream and downstream dependence between the grid points in the same spatial computing layer, thereby enabling synchronous parallel computing. Different spatial computing layers are pushed in layers according to the water flow propagation direction, while meeting the confluence physical constraints and realizing spatial parallelism. In the time dimension, introduce time index parallelization, space layer number and time step number together constitute space Time calculation unit, make the model can advance along the space layer and advance along the time index, reorganize the serial advance relationship into parallel advance structure, increase the number of calculation units that can be processed simultaneously at each parallel stage, and the parallel reorganization structure is space layer number + time step number -1; In the multi-member dimension, different parameter combinations, different precipitation forecast members, or different climate scenarios are regarded as independent simulation members. In addition to the spatial layer and temporal index, the member dimension is added for parallel computation, and intermediate state variables are not exchanged between members.
4. The complex scene rainstorm flood disaster chain whole process deduction method according to claim 2, characterized in that, The construction of a multi-dimensional nested flood inundation simulation model using the historical scene data includes: Floodplain data were extracted from the historical scene data using geomorphological methods. A two-dimensional hydrodynamic model was constructed based on a Godunov-type finite volume framework. Using the floodplain data as a fast baseline and model domain constraint, the rainstorm and flash flood simulation model is coupled with a two-dimensional hydrodynamic model based on a standard protocol interface and a dynamic adaptive grid to construct a multi-dimensional nested flood inundation simulation model.
5. The complex scene rainstorm flood disaster chain whole process deduction method according to claim 4, characterized in that, The extraction of floodplain data from the historical scene data using geomorphological methods includes: A flood event segmentation method is constructed by using multi-source hydrological data and coupling excess threshold sampling and flow sequence moving variance. The flood event segmentation method is used to extract flood events and corresponding multi-peak flood peak characteristics, process initiation points and recession from the historical scene data. Based on the digital elevation model of the flood event, the confluence network of the river channel is extracted and the catchment area of each grid cell is calculated; For grid cells in the river channel, the potential maximum channel flow depth is determined using a power law relationship based on their catchment area. Neighboring units with elevations lower than their corresponding river levels are marked as floodplains, thus obtaining floodplain data.
6. The complex scene rainstorm flood disaster chain whole process deduction method according to claim 1, characterized in that, The construction of the dam breach flood simulation model using the historical scenario data includes: Construct a water level-flow rate combination scenario and obtain experimental observation data of the water level-flow rate combination scenario during a breach flood process. The experimental observation data includes piping seepage failure data, critical soil and water characteristics data, critical stress characteristics data, and three-dimensional morphology data of the breach. Based on the experimental observation data, key disaster-causing factors were identified, and a dam breach flood simulation model was constructed based on these key disaster-causing factors.
7. The complex scene rainstorm flood disaster chain whole process deduction method according to claim 5, characterized in that, After constructing the disaster chain simulation model, it includes: Obtain the water level-discharge observation sequence of the target watershed, and extract the flood discharge threshold of the target watershed using iterative robust regression and the over-threshold method; Based on the aforementioned flood flow threshold, the over-threshold method is used to extract flood events from the daily flow sequence; Based on the extracted flood events, the optimal flow threshold for the occurrence of flood overflow is iteratively searched, and the optimal flood overflow threshold is determined based on the generalized Pareto distribution and Kendall's rank correlation coefficient. Based on all extracted flood events, the preceding precipitation characteristic parameters and the initial soil moisture content state parameters of the watershed corresponding to each flood event are obtained. The preceding precipitation characteristic parameters and the initial soil moisture content state parameters of the watershed are used as model inputs, and whether or not a flood overflows the dike is used as the model output to construct a flood overflow early warning model, with the optimal flood overflow threshold as the trigger condition.
8. The complex scene rainstorm flood disaster chain whole process deduction method according to claim 1, characterized in that, After constructing the disaster chain simulation model, the following steps are taken: based on the Kalman filter double assimilation algorithm, the upstream observation station flow and remote sensing soil water data are fused with the model simulation results to continuously update the parameters and state variables of the disaster chain simulation model.
9. The complex scene rainstorm flood disaster chain whole process deduction method according to claim 1, characterized in that, After constructing the disaster chain simulation model, the process includes: based on multi-source observation data and statistical analysis methods, analyzing the main control factors and / or key watershed attributes affecting flow distribution efficiency of rainstorm to runoff under different flood occurrence mechanisms, quantifying the main control factors and key watershed attributes, and updating the disaster chain simulation model.
10. A complex scene rainstorm flood disaster chain whole process deduction system, characterized in that, The system is used to execute the method for simulating the entire process of a complex scenario rainstorm and flood disaster chain as described in any one of claims 1-9. The system includes: The data acquisition module is used to acquire historical scene data, which includes hydrological and meteorological data and topographic and working condition data. The first construction module is used to construct a rainstorm and flash flood simulation model, a multi-dimensional nested levee flood inundation simulation model, and a levee breach flood simulation model using the historical scene data. The model integration module is used to integrate the aforementioned rainstorm and flash flood simulation model, multi-dimensional nested levee flood inundation simulation model, and levee breach flood simulation model based on standard protocol interfaces, and to construct a disaster chain simulation model. The model training module is used to drive the disaster chain simulation model to generate historical rainstorm and flood chain disaster scenarios based on the historical scene data, and to build an application scenario set; The second building module is used to construct a chain disaster relationship model using deep learning networks, historical scene data, and application scenario sets. The disaster prediction module is used to obtain the scenario prediction data output by the chain disaster relationship model in response to the input predicted meteorological data.
