Graded, partitioned and differentiated risk early warning method for mountain torrent disasters in medium and small basins

Abstract

The invention discloses a grading, zoning and differentiating risk early warning method for mountain torrent disasters in medium and small watershed. Firstly, a drainage basin risk area is determined through mountain torrent disaster data analysis; the drainage basin delay time is calculated, the section water level and the drainage basin submerging range are simulated through the simplified hydrodynamic model, and quantitative analysis is conducted on the model; according to the method, a high-precision grid database is formed by combining unstructured grid division and risk source identification, and then clustering grading is performed on multi-dimensional characteristics of grids through a Gaussian mixture model (GMM) to obtain risk levels of the grids in a drainage basin. Finally, a differential early warning threshold value and a response strategy are formulated according to a grid partition grading result, and dynamic updating and linkage issuing are achieved on a GIS platform in combination with actual measurement data. According to the method, limited real conditions of small and medium-sized basin data and precision requirements of model simulation are taken into consideration, fine setting of the early warning threshold value is realized on the basis of grading and partitioning, and the accuracy of mountain torrent risk identification is also remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of flash flood disaster warning and defense technology, and in particular to a differentiated risk warning method for flash flood disasters in small and medium-sized river basins based on classification and zoning. Background Art

[0002] Flash floods are one of the most common and serious natural disasters in mountainous and hilly areas of my country. They are characterized by rapid outbreak, short duration, and uneven impact range, posing a huge threat to villages, transportation arteries, and infrastructure in small and medium-sized river basins. The suddenness and locality of flash floods make defense work very challenging, especially in areas with complex terrain and rapid changes in rainfall intensity. Traditional flash flood warning methods often have certain limitations in terms of refined management and rapid response.

[0003] At present, flash flood disaster warning mainly relies on two methods: flood forecasting based on hydrological models and indicator judgment based on mathematical statistics. Flood forecasting based on hydrological models usually simulates the rainstorm-flood evolution process in the basin through high-precision hydrological models, and triggers warnings with simulated flow or water level thresholds. However, due to the lack of observation data in small and medium-sized basins, the calibration and verification of model parameters face great difficulties, resulting in the accuracy and reliability of forecast results being limited in many cases. The indicator judgment method based on mathematical statistics establishes empirical thresholds such as critical rainfall and critical water level, and combines real-time monitored rainfall or water level changes to determine whether to initiate a warning. Although this method is easy to operate, it cannot fully consider the spatial differences within the basin.

[0004] In addition, flash flood disasters have significant spatial and temporal differences. Different areas in the same river basin have significant differences in flood propagation speed, flooding range, and risk impact. For example, the upstream area is usually steep, floods rise and fall violently and have strong impact, but the forecast period is short; the river channel in the middle reaches is gentle, the flooding range is wide, and the spatial distribution of risks is more complex; the flood propagation time in the downstream area is longer, but the affected area is wide and the impact is serious. Traditional early warning methods use a unified threshold or a single mode, which makes it difficult to accurately match the risk characteristics of different regions, and it is impossible to scientifically distinguish between key defense areas and ordinary areas, and it is difficult to respond to sudden disasters in different scenarios in a timely manner. Summary of the invention

[0005] In view of the shortcomings of the existing technology, the present invention aims to solve the problems of insufficient regional risk assessment accuracy, inflexible response mechanism and unreasonable allocation of prevention and control resources in traditional flash flood disaster early warning methods, and proposes a differentiated early warning method for flash flood disasters in small and medium-sized watersheds. By integrating basin lag calculation, constructing hydrodynamic models, grid risk assessment and intelligent clustering algorithms, a dynamic hierarchical and partitioned early warning system is established. Differentiated early warning thresholds are set for different risk areas, and the early warning level is dynamically adjusted in combination with meteorological forecasts and real-time monitoring data to ensure that high-risk areas can respond first.

[0006] The present invention is realized by the following technical scheme: a method for early warning of graded and zoned differentiated risks of mountain torrent disasters in small and medium-sized river basins, the method comprising the following steps: Step S1, taking small and medium-sized watersheds as units, collect the underlying surface characteristics: topographic data, satellite images and meteorological and hydrological data, analyze the confluence path, watershed slope and roughness; combine the existing basic investigation results of mountain torrent disasters to clarify the preliminary distribution of risk areas in the watershed, and analyze and deduce the design rainstorm and flood information through data analysis; Step S2, after the basin area and underlying surface characteristics are determined, the basin lag time is calculated for the determined risk area, the time distance between the centroid of the flow process at the outlet section of the rainfall basin and the centroid of the corresponding net rain process is approximately regarded as the average confluence time of the basin, and the basin lag time is calculated using different formulas according to the basin area; Step S3, by constructing a two-dimensional or three-dimensional hydrodynamic model to calculate different flood frequencies at dangerous sections of key villages to determine the inundation range under different levels of water levels and flash flood disaster scenarios; Step S4, based on the unstructured grid, the flash flood risk area is divided and the risk source is identified. The land and water areas are divided on the basis of the unstructured grid, the grid scale is clarified, and the flow direction analysis and terrain filling processing are carried out on the study area based on the DEM data and the river system vector, the basin boundary is determined, and the water conservancy facilities and potential danger sources are superimposed in the grid database, thereby forming a grid unit system and risk source information matrix to support risk analysis; Step S5, based on the Gaussian clustering method GMM, combined with flood simulation results, basin lag time and regional risk source analysis, the flash flood danger area is dynamically graded and zoned. The zoning results are displayed on the GIS platform after manual verification and administrative boundary optimization, and are dynamically updated in conjunction with real-time water and rainfall data; Step S6, for the classified and zoned areas, differentiated flash flood warning thresholds are set and corresponding warning models are constructed. First, thresholds that match the inundation range and basin lag time indicators are formulated for different risk levels, and real-time linkage is achieved through meteorological forecasts and rainfall and water level monitoring data, combined with the GIS platform and emergency management system.

[0007] Furthermore, the step S1 includes the following sub-steps: Step S1-1, summarize and sort out the flash flood disaster points that have occurred or are potential in the basin in combination with historical documents, and preliminarily define the scope of the risk area; Step S1-2, using high-resolution DEM or satellite images, extracting key underlying surface characteristic parameters of the watershed: confluence path, slope, roughness, land use and river network density, and performing data format conversion and error correction; Step S1-3, collect and organize the rainfall observation records, rainfall station distribution, and measured runoff data of the watershed over the years; if there are radar precipitation products or weather forecast data, they are also included to provide multi-source support for subsequent flood simulation; Step S1-4, based on the acquired meteorological and hydrological data, the regional rainstorm frequency analysis method is used to deduce the basin design rainstorm intensity or rainfall process; combined with the runoff calculation or verified empirical formula, the peak flow or design flood process line under different design frequencies is obtained for subsequent model calculation and risk assessment.

[0008] Furthermore, in step S2, different formulas are used to calculate the basin delay time according to the basin area. Specifically, the basin delay time is the time distance between the centroid of the flow process at the outlet section of the rainfall basin and the centroid of the corresponding net rain process. In the basin confluence process, it can be equivalent to the average confluence time of the basin. The basin delay time corresponding to each risk area is determined according to the confluence time. When the basin area is less than or equal to 50km 2 When the basin area is greater than 50km 2 The NRCS equation is used for solution. Due to the large variation in terrain in small and medium-sized watersheds, distribution calculation based on digital elevation is used.

[0009] Furthermore, the step S3 includes the following sub-steps: Step S3-1, initialize the model and set boundary conditions, that is, initialize the hydrodynamic model according to the actual terrain characteristics and data conditions of the small and medium-sized watersheds; specifically, extract the longitudinal and cross-sectional information of the river channel of the watershed through high-resolution DEM data, adjust the section spacing to adapt to the curvature of the river channel and complex water flow; after extracting the underlying surface characteristics of the slope and river network density, specify the river width, river depth and cross-sectional shape for each section, and calculate the roughness value according to the Manning formula; the boundary conditions are set by using the designed flood peak flow or flow process curve as the upper The upstream boundary conditions; the design flood peak flow is calculated by combining the regional rainstorm intensity with the rainfall area and the runoff coefficient; the downstream boundary conditions are constant water level or free outflow conditions; when there are hydraulic structures, the downstream water level needs to be corrected by hydraulic calculation; the rainfall input is mapped to the model grid through the rainfall intensity-duration curve; the river channel and the land are coupled, the river channel area is divided into one-dimensional units, and the grid size is within 1 / 5 of the river width; the land area uses a two-dimensional unstructured triangular grid to ensure the dynamic consistency of flood overflow and river return; Step S3-2, using the finite volume method to discretize the shallow water equations to capture the nonlinear characteristics of flood propagation; when performing multi-scenario simulations under different rainfall scenarios and flood frequencies, the parallel architecture is used for minimum time control; Step S3-3, uncertainty analysis, that is, on the input side, Monte Carlo sampling is performed on the design parameters: rainstorm amount, rainfall duration and boundary flow, assuming that the rainfall intensity obeys the Chicago rain pattern distribution or normal distribution, and multiple groups of random samples are generated to quantify the uncertainty of the water level simulation results; on the parameter side, Bayesian statistics or sensitivity analysis tools are used to quantify the key parameters of roughness, bed morphology and weirs, and the parameter disturbance range is adjusted one by one to evaluate its impact on the simulation results. Through this analysis, the confidence interval of the water level simulation can be obtained; Step S3-4, after the simulation is completed, the water level process lines of key sections under different recurrence periods are generated; the water depth calculation results are superimposed on the DEM data, and the flood inundation range and flow velocity distribution map are output and presented in vector format or raster format to provide input for subsequent grid division, risk assessment and hierarchical zoning.

[0010] Furthermore, the step S4 includes the following sub-steps: Step S4-1, basin boundary determination, that is, according to the spatial distribution of flash flood warning objects, the downstream position of the basin to which the flash flood warning object belongs is selected as the outlet section, and the basin area is controlled; then, based on the DEM data of the study area, the spatial location data of the flash flood warning object, and the river system vector data, ArcGIS hydro tools are used to carry out terrain filling, flow direction analysis, water collection capacity analysis, and digital water system generation, and the basin boundary is delineated according to the basin outlet section; Use the plan layout of water conservancy projects such as river water boundary, bridge piers, and water-blocking weirs to cut the basin boundary and specify the land and water areas; Step S4-2: Discretization of watershed space and unstructured grid division, watershed characteristic analysis, land, water and river grid classification, unstructured grid division, and grid inspection and adjustment; Step S4-3, flash flood risk source identification and embedding, specifically, the potential hazard sources obtained from the investigation are superimposed on the grid database; and special marks are given to the grids where key bridges, traffic arteries and other infrastructure are located; the potential hazard sources include but are not limited to debris flow gullies, steep cliffs and weirs; Step S4-4, after completing the unstructured grid division, the DEM data, basic hydrological data and rainfall data are imported into the grid, and corresponding physical properties are assigned according to the characteristics of different areas. The characteristics of different areas include but are not limited to soil type, vegetation cover, slope, roughness and underlying surface.

[0011] Furthermore, the step S4-2 further includes the following sub-steps: Step S4-2a, watershed characteristics analysis, i.e., paying special attention to mountain torrents in small and medium-sized watersheds that are susceptible to local heavy precipitation, including precipitation distribution, slope changes in watershed terrain, and characteristics of river channels within the watershed; Step S4-2b, land, water, and river grid classification, that is, the grid size is reduced for the upstream of the mountainous area and the area with large slopes, so as to finely simulate the changes in water flow and the precipitation infiltration process; but for the low-slope area and the catchment area, the grid size is enlarged; for the river part of the mountain torrent small watershed, the river bend, and the confluence of tributaries, high-resolution grids are required for simulation; Step S4-2c, unstructured meshing, i.e. starting from the boundary of the watershed, selecting mesh nodes, performing triangulation based on the Delaunay criterion, and ensuring that the inner angle of the triangle is maximized; Step S4-2d, after the unstructured grid is divided, the grid is checked and dynamically adjusted by manual identification to ensure that the shape of each grid unit is reasonably set.

[0012] Furthermore, the step S5 includes the following sub-steps: Step S5-1, establishment of multidimensional data feature vector and Gaussian clustering GMM of grids, including construction of multidimensional feature vector for each grid unit, construction of risk factors and quantitative indicators of flash flood risk for each grid, definition of the number of clusters, calculation of attribution probability and fitting of data; Step S5-2, classification optimization and GIS display, that is, after completing Gaussian clustering, the classification results of the danger zone are optimized and displayed; first, the clustering results are manually checked, and the classification results that deviate from the actual situation are adjusted in combination with historical disaster records; then, the clustering results are superimposed on the administrative district boundaries, and the zoning is refined according to the township or village boundaries, so that the zoning results meet the actual management needs.

[0013] Furthermore, the step S5-1 further includes the following sub-steps: Step S5-1a, constructing a multi-dimensional feature vector for each grid unit, that is, the characteristics of each grid contain multiple dimensions of risk factors related to flash flood disasters, including hydraulic characteristic parameters, flood forecast period, and risk source information parameters; Step S5-1b, constructing a nonlinear relationship between risk factors and flash flood risk quantitative indicators of each grid; the characteristic vector of each grid is represented as a multidimensional database containing risk factors and flash flood risk determination indicator parameters; Step S5-1c, defining the number of clusters, that is, defining the number of clusters by defining the risk level, wherein the number of clusters is defined according to the basin area and the degree of differentiation; then initializing the parameters of each cluster, including: mean, covariance matrix and weight; Step S5-1d, based on the feature data of each grid: mean, covariance matrix and weight, by calculating the posterior probability that each grid unit belongs to each Gaussian distribution, thereby calculating the probability of each grid unit belonging to different clusters; Step S5-1e, optimize the partitioning result by maximizing the probability that the grid points belong to multiple Gaussian distributions so that each cluster can better fit the data. This process is achieved by maximizing the likelihood function, which is expressed as: ; Where N represents the total number of samples in the data set, that is, the number of data points; is the set of all parameters, , K is the total number of clusters (groups); Represents the i-th sample data point in the data set; represents the weight of each Gaussian distribution, For the The probability density function of a Gaussian distribution is and are the mean vector and covariance matrix respectively; Step S5-1f, alternately perform steps S5-1d and S5-1e for multiple iterations until the mean of the model is , covariance matrix And weight It becomes stable or reaches the maximum number of iterations. Each iteration updates the grid's attribution probability based on the current model and gradually optimizes the cluster division. In step S5-1g, the GMM model outputs the flash flood risk classification result of each grid, that is, the cluster to which each grid belongs.

[0014] Furthermore, the step S6 also includes the following sub-steps: Step S6-1, differentiated setting of warning thresholds, i.e., based on the hierarchical zoning results of flash flood hazard areas, combined with the flood inundation range, basin lag time, key section water level process and time required for transfer in each area, determine the zoning warning thresholds, and establish differentiated warning trigger conditions for risk areas of different levels in the zone by combining historical data and field survey results; Step S6-2, design and implementation of differentiated early warning modes, that is, after completing the setting of early warning thresholds, differentiated early warning response modes are designed in combination with the risk characteristics of different regions. By introducing weather forecasts, rainfall monitoring, water level monitoring and other multi-source real-time data, a hierarchical early warning mechanism is established, and the early warning level is dynamically adjusted; Step S6-3: Release and linkage management of warning results, that is, the warning information generated by the differentiated warning mode needs to be released within the region and linked with the emergency management system; the warning information is released through multiple channels, including text messages, broadcasts, warning terminal display screens and social media platforms, to ensure that different groups can obtain warning signals in a timely manner; and combined with the linkage function of the GIS platform and the warning system, the dynamically updated warning information is shared in real time with management departments at all levels to dispatch and coordinate resources.

[0015] Furthermore, the step S6-1 further includes the following sub-steps: Step S6-1a, defining corresponding warning indicators and threshold ranges for areas of different risk levels; Step S6-1b, using a weight analysis method to quantify the impact of different indicators to calibrate the warning threshold; the different indicators include water depth, flow velocity, rainfall, and basin hysteresis; Step S6-1c, verifying the setting of the regional warning threshold with the measured data, and calibrating the warning conditions in real time through retrospective analysis of historical flash flood events.

[0016] The beneficial effects of the present invention are as follows: The present invention realizes the hierarchical zoning and differentiated risk warning of flash flood disasters through technical means such as basin lag calculation, simplified hydrodynamic model and grid risk assessment, combined with Gaussian clustering algorithm, and significantly improves the accuracy and scientificity of risk assessment. By setting differentiated warning thresholds and dynamic adjustment mechanisms, it is possible to optimize the warning response according to different risk levels and regional characteristics, ensure priority prevention and control in high-risk areas, and improve the accuracy and timeliness of warnings. At the same time, by using meteorological forecasts, real-time monitoring data and model linkage, the warning level is dynamically updated, which significantly enhances the intelligence level of warning and emergency response capabilities. The present invention is suitable for the prevention and control of flash flood disasters in small and medium-sized basins under various terrain conditions, has wide applicability and promotion value, provides scientific support for improving the level of disaster prevention and mitigation and optimizing resource allocation, and has important social and economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of the differentiated risk warning method for mountain torrent disasters in small and medium-sized river basins according to the present invention. DETAILED DESCRIPTION

[0018] Here, exemplary embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application.

[0019] The present invention uses a hierarchical zoning and differentiated risk warning method to improve the effectiveness of flood disaster prevention. By comprehensively considering the terrain characteristics, flood propagation process and risk distribution rules within the basin, differentiated warning modes and response mechanisms are set for different regions, which can more scientifically match the warning time, transfer time and defense measures, thereby significantly improving the accuracy and timeliness of the warning. At the same time, by clarifying the danger level of different areas through hierarchical zoning, it is possible to effectively optimize resource allocation and ensure that the defense needs of key areas are met. The specific technical solution is implemented as follows: like Figure 1 As shown, the present invention provides a differentiated risk warning method for mountain torrent disasters in small and medium-sized river basins by classification and zoning, including the following steps: Step S1, taking small and medium-sized watersheds as units, collect terrain data, satellite images and meteorological and hydrological data, analyze the underlying surface characteristics such as confluence path, watershed slope and roughness. Combined with the existing basic survey results of mountain torrent disasters, clarify the preliminary distribution of risk areas in the watershed. And through data analysis and deduction of design rainstorms and design floods, lay the foundation for subsequent model calculations.

[0020] Step S2, after clarifying the basin area and underlying surface characteristics, calculate the basin delay time for the determined risk area, and approximate the time distance between the flow process centroid of the outlet section of the rainfall basin and the centroid of the corresponding net rain process as the average confluence time of the basin. Use different formulas to calculate the basin delay time according to the basin area.

[0021] Step S3, by constructing a two-dimensional or three-dimensional hydrodynamic model to calculate different flood frequencies at dangerous sections of key villages, determine the inundation range under different levels of water levels and flash flood disaster scenarios.

[0022] Step S4, divide the flash flood risk area and identify the risk source based on the unstructured grid. Divide the land and water areas on the basis of the unstructured grid, clarify the grid scale, conduct flow direction analysis and terrain filling in the study area based on DEM (digital elevation model data) data and river system vectors, determine the basin boundary, and superimpose water conservancy facilities such as bridges and dams and potential danger sources in the grid database, so as to form a grid unit system and risk source information matrix to support risk analysis.

[0023] Step S5: Based on the Gaussian Mixture Model (GMM), combined with flood simulation results, basin lag time, and regional risk source analysis, the flash flood risk areas are dynamically graded and zoned. The zoning results are displayed on the GIS platform after manual verification and administrative boundary optimization, and are dynamically updated in conjunction with real-time water and rainfall data.

[0024] Step S6: Differentiate the flash flood warning thresholds for the classified and zoned areas and build corresponding warning models. First, set thresholds that match indicators such as the flooding range and basin lag time for different risk levels, and link them with meteorological forecasts, rainfall, and water level monitoring data in real time, and combine the GIS platform with the emergency management system.

[0025] Furthermore, the step S1 is specifically as follows: Step S1-1, combining historical documents, existing investigation reports and field surveys, summarize and sort out the existing or potential flash flood disaster points in the basin, and preliminarily define the scope of the risk area; Step S1-2, using high-resolution DEM or satellite images, extracting key underlying surface characteristic parameters such as the drainage path, slope, roughness, land use and river network density of the basin, and performing necessary format conversion and error correction on the data; Step S1-3, collect and organize the rainfall observation records, rainfall station distribution, and measured runoff data of the hydrological station in the basin over the years; if there are radar precipitation products or weather forecast data, they are also included to provide multi-source support for subsequent flood simulation; Step S1-4, based on the acquired meteorological and hydrological data, adopt regional rainstorm frequency analysis methods (such as P-III distribution, Lognormal, etc.) to deduce the basin design rainstorm intensity or rainfall process; combine appropriate runoff calculation or verified empirical formula to obtain the peak flow or design flood process line under different design frequencies, laying a quantitative foundation for model calculation and risk assessment.

[0026] Furthermore, the step S2 is specifically as follows: The basin lag time is the time distance between the centroid of the flow process at the outlet section of the rainfall basin and the centroid of the corresponding net rain process. In the basin confluence process, it can be equivalent to the average confluence time of the basin. The basin lag time corresponding to each risk area is determined according to the confluence time. Optionally, the drainage area is less than or equal to 50km 2 The design flood reasoning formula of Zhejiang Province is used for iterative calculation; Optionally, the catchment area is greater than 50km 2 The NRCS equation is used to solve the problem, and the distribution calculation is adopted based on the digital elevation, considering the large regional variation of terrain in small and medium-sized watersheds. Furthermore, the step S3 is specifically as follows: Step S3-1, initialize the model and set boundary conditions: Initialize the hydrodynamic model according to the actual terrain characteristics and data conditions of small and medium-sized watersheds. First, extract the longitudinal and cross-sectional information of the river channel in the basin through high-resolution DEM data, and adjust the section spacing to 50 to 200 meters to adapt to the curvature of the river channel and the complexity of the water flow. After extracting the underlying surface features such as slope and river network density, specify the river width, river depth and cross-sectional shape (such as rectangle, trapezoid or composite section) for each section to calculate the roughness value according to the Manning formula. The typical roughness value range is 0.020 to 0.025 for sandy riverbed and 0.030 to 0.035 for stone riverbed. The specific value can be adjusted according to the on-site investigation.

[0027] In the boundary condition setting, the design peak flow or flow process curve is used as the upstream boundary condition. The design peak flow can be calculated by combining the regional rainstorm intensity with the rainfall area and the runoff coefficient. The downstream boundary condition usually adopts a constant water level or free outflow condition, but if there are hydraulic structures such as weirs, the downstream water level needs to be corrected by hydraulic calculation. The rainfall input part is mapped to the model grid through the rainfall intensity-duration curve. The river channel and the land are coupled. The river channel area is divided into one-dimensional units with a grid size of less than 1 / 5 of the river width; the land area adopts a two-dimensional unstructured triangular grid with a grid size of 10 to 30 meters to ensure the dynamic consistency of flood overflow and river backflow.

[0028] Step S3-2, the finite volume method is used to discretize the shallow water equations to capture the nonlinear characteristics of flood propagation. When performing multi-scenario simulations under different rainfall scenarios and flood frequencies, the parallel architecture can significantly reduce the computing time. For example, if the basin area is less than 50 square kilometers, the time to complete a simulation is controlled within 1 hour.

[0029] Step S3-3, uncertainty analysis. Specifically, on the input side, Monte Carlo sampling is performed on parameters such as the design rainstorm volume, rainfall duration, and boundary flow, assuming that the rainfall intensity follows the Chicago rain pattern distribution or normal distribution, and multiple groups of random samples are generated to quantify the uncertainty of the water level simulation results; On the parameter side, Bayesian statistics or sensitivity analysis tools are used to quantify key parameters such as roughness, bed morphology and weirs, and the parameter disturbance range is adjusted one by one to evaluate their impact on the simulation results. This analysis can be used to obtain the confidence interval of the water level simulation.

[0030] Step S3-4: After the simulation is completed, the water level process lines of key sections under different return periods (such as 5 years, 20 years, and 50 years) are generated. The water depth calculation results are superimposed on the DEM data, and the flood inundation range and flow velocity distribution map are output in vector or raster format. These results provide accurate input for subsequent grid division, risk assessment, and grading and zoning.

[0031] Furthermore, the step S4 is specifically as follows: Step S4-1, basin boundary determination. According to the spatial distribution of flash flood warning objects (Si), the downstream location of the basin to which the flash flood warning object belongs is selected as the outlet section, and the controlled basin area is guaranteed to be less than 200km2 as much as possible. Based on the DEM data of the study area, the spatial location data of the flash flood warning object, and the vector data of the river system, ArcGIS hydro tools are used to carry out terrain filling, flow direction analysis, water collection capacity analysis, digital water system generation, etc., and the basin boundary is delineated according to the basin outlet section.

[0032] The basin boundaries are cut using the plan layout of water area boundaries, bridge piers, water-blocking weirs and other water conservancy projects to detail the land and water areas.

[0033] Step S4-2, watershed spatial discretization and unstructured grid division, specifically includes the following sub-steps: Step S4-2a, basin characteristics analysis. Mountain torrents in small and medium basins are usually affected by local heavy precipitation, and the terrain is complex and the slope is large. Therefore, special attention should be paid to precipitation distribution, slope changes of basin terrain, and characteristics of rivers in the basin.

[0034] Step S4-2b, land, water, and river grid classification. For the upper reaches of mountainous areas and areas with large slopes, it is recommended to use a smaller grid size (such as about 10 meters) to accurately simulate the changes in water flow and precipitation infiltration. For low-slope areas and catchment areas, the grid can be slightly larger (no more than 30 meters) to improve calculation efficiency. For the river part of the small watershed of mountain torrents, the grid should be finer, usually no more than 5-10 meters, especially in areas with large changes in water flow such as sharp bends in the river and confluences of tributaries, and higher resolution grids are required for simulation.

[0035] Step S4-2c, unstructured grid division. Starting from the basin boundary, select grid nodes and perform triangulation based on the Delaunay criterion (ensuring that the internal angle of the triangle is maximized). Preferably, in key areas such as flood propagation paths and runoff convergence areas, use smaller grid refinement simulation accuracy to ensure accurate capture of the basin's hydrological processes.

[0036] Step S4-2d, after the unstructured grid is divided, the grid is checked and dynamically adjusted by manual identification to ensure that each grid unit has a reasonable shape and is not too narrow or extreme.

[0037] Step S4-3, flash flood risk source identification and embedding. Specifically, the potential hazard sources obtained from the investigation (such as debris flow gully mouths, steep cliffs, and weirs) are superimposed on the grid database; further, special marks are given to the grids where key bridges, traffic arteries and other infrastructure are located.

[0038] Step S4-4, after completing the unstructured grid division, import the DEM, basic hydrological data, rainfall data, etc. into the grid, and assign corresponding physical properties (such as soil type, vegetation cover, slope, roughness, underlying surface, etc.) according to the characteristics of different areas.

[0039] Furthermore, the step S5 is specifically as follows: Step S5-1, establishing a multidimensional data feature vector and Gaussian clustering (GMM) of the grid, specifically includes the following sub-steps: Step S5-1a, construct a multi-dimensional feature vector for each grid unit. The features of each grid contain multiple dimensions of risk factors related to flash flood disasters, including hydraulic characteristic parameters, flood forecast period, risk source information parameters, etc.

[0040] Specifically, the calculation results of the hydrodynamic model in step S3 can provide grid-level hydraulic characteristic parameters, such as dynamic hydrological data such as maximum water depth, maximum flow velocity and flood arrival time; the basin lag time in step S2 can calculate the flood forecast period of each grid area; the regional risk source information in step S4 (such as population density, distribution of important facilities and historical disaster points, etc.) can further supplement the spatial and attribute integrity of the data. These features are used to create a multidimensional data vector of risk factors for each grid unit.

[0041] Step S5-1b, construct a nonlinear relationship between risk factors (hydraulic characteristic parameters, risk source information) and the quantitative indicators of flash flood risk of each grid (different flood frequencies determine different levels of water levels at dangerous sections of key villages, flood forecast periods, and inundation ranges under flash flood disaster scenarios); ultimately, the characteristic vector of each grid can be represented as a multidimensional database containing parameters such as risk factors and flash flood risk determination indicators.

[0042] Step S5-1c, define the number of clusters. Assume that we have three risk levels (high risk, medium risk, low risk), so the number of clusters is K=3 (the number of clusters is defined according to the basin area and the degree of differentiation, and can also be 4, 5, 6, etc.). Then, initialize the parameters of each cluster, including: mean, covariance matrix, weight.

[0043] Step S5-1d, based on the characteristic data (mean, covariance matrix, weight) of each grid, by calculating the posterior probability that each grid unit belongs to each Gaussian distribution, the probability of each grid unit belonging to different clusters (risk levels) is calculated.

[0044] Step S5-1e, optimize the partitioning result by maximizing the probability that the grid points belong to multiple Gaussian distributions, so that each cluster can better fit the data. This is achieved by maximizing the likelihood function, which is expressed as: ; Where N represents the total number of samples in the data set, that is, the number of data points; is the set of all parameters, , K is the total number of clusters (groups); Represents the i-th sample data point in the data set; represents the weight of each Gaussian distribution, For the The probability density function of a Gaussian distribution is and are the mean vector and covariance matrix respectively; Step S5-1f, alternately perform steps S5-1d and S5-1e for multiple iterations until the mean of the model is , covariance matrix And weight It becomes stable or reaches the maximum number of iterations. Each iteration updates the probability of the grid belonging to the current model and gradually optimizes the division of clusters. In step S5-1g, the GMM model will output the flash flood risk classification results of each grid, that is, which cluster each grid belongs to (high risk, medium risk or low risk). For example, the grids are divided into three categories: high risk, medium risk and low risk. The high-risk area is characterized by high water depth (such as ≥1.5m), fast flow rate (such as ≥2.5m / s), short arrival time (such as ≤30 minutes), and high population density and facility exposure, while the low-risk area is characterized by shallow water depth, slow flow rate and long arrival time.

[0045] Step S5-2, classification optimization and GIS display: After completing Gaussian clustering, the classification results of the danger zone are optimized and displayed. First, the clustering results are manually checked, and the classification results that deviate from the actual situation are adjusted in combination with expert experience or historical disaster records. For example, when the risk level of certain grids does not match their historical disaster records, they can be reclassified according to the actual situation. Subsequently, the clustering results are superimposed on the administrative district boundaries, and the zoning is refined according to the township or village boundaries so that the zoning results meet the actual management needs.

[0046] The zoning results are intuitively presented in different colors on the GIS platform to distinguish high, medium and low risk areas. Through the linkage with real-time rainfall monitoring and water level monitoring data, the classification of dangerous areas can be dynamically updated. For example, when an extreme rainfall event occurs, real-time data triggers the model to recalculate or call pre-simulation results to quickly adjust the zoning level to ensure the timeliness of the early warning response. Zoning results include vector-formatted dangerous area layer files and dynamically updated reports to support flood control command and emergency decision-making.

[0047] Through the above process, the flash flood hazard zone classification and zoning method based on Gaussian clustering achieved scientific division and dynamic adjustment, provided technical support for the precise prevention and control of hazardous areas, and enhanced applicability and response efficiency in actual management.

[0048] Furthermore, the step S6 is specifically as follows: Step S6-1, differentiated setting of warning thresholds: Based on the hierarchical zoning results of flash flood hazard areas, comprehensive consideration is given to the flood inundation range, basin lag time, water level process of key sections, and time required for transfer in each area to determine the zoning warning thresholds. By combining historical data and field survey results, differentiated warning trigger conditions for risk areas of different levels within the zone are established. Specifically, it includes the following sub-steps: Step S6-1a, define the corresponding warning indicators and threshold ranges for different risk level areas (level I, level II, and level III). For example, for level I high-risk areas, priority is given to the characteristics of short flood arrival time, and strict rainfall and water level thresholds are set, such as immediately starting a warning when the rainfall intensity exceeds the critical value (such as 50 mm / h) or the cross-section water level exceeds 1.5 meters; for level II and level III areas, looser trigger conditions can be set, focusing on the flood overflow range and personnel transfer time.

[0049] Step S6-1b, using a weight analysis method (such as the analytic hierarchy process or the entropy weight method) to quantify the impact of different indicators (such as water depth, flow velocity, rainfall, and basin hysteresis) to calibrate the warning threshold.

[0050] Step S6-1c, verifying the setting of the regional warning threshold with the measured data, and calibrating the scientificity and applicability of the warning conditions through retrospective analysis of historical flash flood events.

[0051] Step S6-2, design and implementation of differentiated early warning mode: After completing the setting of early warning thresholds, design differentiated early warning response modes based on the risk characteristics of different regions. By introducing multi-source real-time data such as weather forecasts, rainfall monitoring, and water level monitoring, a hierarchical early warning mechanism is established, and the early warning level is adjusted dynamically.

[0052] For Level I high-risk areas, the focus is on monitoring the impact of short-term heavy rainfall, and a rapid response mechanism is set up in combination with radar precipitation forecasts and measured data from rainfall monitoring stations. When the accumulated rainfall or monitored water level approaches the warning threshold, an early warning is triggered and information is transmitted downstream simultaneously. In this process, high-frequency data acquisition equipment and communication networks can be used to ensure the timely transmission and execution of warning signals.

[0053] For Level II areas, a joint trigger mode is used to determine whether to initiate an early warning through a comprehensive analysis of rainfall and water level. When the water level at the monitoring station exceeds the threshold continuously and the accumulated rainfall exceeds the warning line, the early warning state is entered.

[0054] For Level III areas, the main trigger condition is the accumulated rainfall, and the start time is evaluated in combination with the external flooding range to provide a sufficient response window.

[0055] In order to enhance the flexibility of early warning response, a dynamic adjustment mechanism can be introduced into the early warning mode. When extreme rainfall events or abnormal hydrological changes occur, real-time monitoring data is combined with model simulation results to quickly adjust the warning level and trigger conditions. For example, when the monitoring data exceeds the preset threshold range, the early warning system automatically upgrades the warning level of the corresponding area and prompts the upgrade of defense measures.

[0056] Step S6-3, release and linkage management of warning results: The warning information generated by the differentiated warning mode needs to be quickly released within the region and linked with the emergency management system. The release of warning information can be carried out through multiple channels, including text messages, broadcasts, warning terminal display screens, social media platforms, etc., to ensure that different groups can obtain warning signals in a timely manner.

[0057] In high-risk areas, early warning information should include specific flood arrival time, expected inundation area, and evacuation route recommendations, to guide personnel and facilities in relevant areas to enter emergency status in advance; In medium and low risk areas, the information released can focus on preventive measures and dynamic attention recommendations; Furthermore, by combining the linkage function of the GIS platform and the early warning system, the dynamically updated early warning information will be shared with management departments at all levels in real time, so that the flood control command center can quickly dispatch resources and coordinate the forces of all parties.

[0058] Through the above steps, the setting of differentiated warning thresholds, model design and release management have achieved precise prevention and control of different regions and different risk levels, providing reliable guarantees for the effective defense against mountain torrents disasters.

[0059] The preferred embodiments of the present invention are described in detail above; however, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A differentiated risk warning method for flash flood disasters in small and medium-sized river basins based on classification and zoning, characterized in that: The method comprises the following steps: Step S1, collect the underlying surface characteristics with small and medium-sized watersheds as units; clarify the preliminary distribution of risk areas in the watershed, and analyze and infer the design rainstorm and flood information through data analysis; Step S2, calculating the basin delay time for the determined risk area, and using different formulas to calculate the basin delay time according to the basin area; Step S3, by constructing a two-dimensional or three-dimensional hydrodynamic model to calculate different flood frequencies at dangerous sections of key villages to determine the inundation range under different levels of water levels and flash flood disaster scenarios; Step S4, dividing the flash flood risk area and identifying the risk source based on the unstructured grid, dividing the land and water areas on the basis of the unstructured grid, clarifying the grid scale and determining the basin boundary, and then obtaining the grid unit system and risk source information matrix; Step S5, dynamically classifying and zoning flash flood hazard areas based on flood simulation results, basin lag time, and regional risk source analysis; Step S6, for the classified and zoned areas, differentiated flash flood warning thresholds are set and corresponding warning modes are constructed, and the release and linkage management of warning results are carried out.

2. According to claim 1, a differentiated risk warning method for mountain torrent disasters in small and medium-sized river basins is characterized by: The step S1 comprises the following sub-steps: Step S1-1, summarize and sort out the flash flood disaster points that have occurred or are potential in the basin in combination with historical documents, and preliminarily define the scope of the risk area; Step S1-2, using high-resolution DEM or satellite images, extracting key underlying surface characteristic parameters of the watershed: confluence path, slope, roughness, land use and river network density, and performing data format conversion and error correction; Step S1-3, collect and organize the rainfall observation records, rainfall station distribution, and measured runoff data of the watershed over the years; if there are radar precipitation products or weather forecast data, they are also included to provide multi-source support for subsequent flood simulation; Step S1-4, based on the acquired meteorological and hydrological data, a regional rainstorm frequency analysis method is used to deduce the design rainstorm intensity or rainfall process of the basin; Combined with runoff calculations or verified empirical formulas, peak flows or design flood process lines under different design frequencies are obtained for subsequent model calculations and risk assessments.

3. The method for differentiated risk warning of flash flood disasters in small and medium-sized river basins according to claim 1 is characterized in that: In step S2, different formulas are used to calculate the basin delay time according to the basin area. Specifically, the basin delay time is the time distance between the centroid of the flow process at the outlet section of the rainfall basin and the centroid of the corresponding net rain process. In the basin confluence process, it can be equivalent to the average confluence time of the basin. The basin delay time corresponding to each risk area is determined according to the confluence time; when the basin area is less than or equal to 50km 2 When the basin area is greater than 50km 2 The NRCS equation is used for solution. Due to the large variation in terrain in small and medium-sized watersheds, distribution calculation based on digital elevation is used.

4. The method for differentiated early warning of flash flood disasters in small and medium-sized river basins according to claim 1 is characterized in that: The step S3 comprises the following sub-steps: Step S3-1, initialize the model and set boundary conditions, that is, initialize the hydrodynamic model according to the actual terrain characteristics and data conditions of the small and medium-sized watersheds; specifically, extract the longitudinal and cross-sectional information of the river channel of the watershed through high-resolution DEM data, adjust the section spacing to adapt to the curvature of the river channel and complex water flow; after extracting the underlying surface characteristics of the slope and river network density, specify the river width, river depth and cross-sectional shape for each section, and calculate the roughness value according to the Manning formula; the boundary conditions are set by using the designed flood peak flow or flow process curve as the upper The upstream boundary conditions; the design flood peak flow is calculated by combining the regional rainstorm intensity with the rainfall area and the runoff coefficient; the downstream boundary conditions are constant water level or free outflow conditions; when there are hydraulic structures, the downstream water level needs to be corrected by hydraulic calculation; the rainfall input is mapped to the model grid through the rainfall intensity-duration curve; the river channel and the land are coupled, the river channel area is divided into one-dimensional units, and the grid size is within 1 / 5 of the river width; the land area uses a two-dimensional unstructured triangular grid to ensure the dynamic consistency of flood overflow and river return; Step S3-2, using the finite volume method to discretize the shallow water equations to capture the nonlinear characteristics of flood propagation; when performing multi-scenario simulations under different rainfall scenarios and flood frequencies, the parallel architecture is used for minimum time control; Step S3-3, uncertainty analysis, that is, on the input side, Monte Carlo sampling is performed on the design parameters: rainstorm amount, rainfall duration and boundary flow, assuming that the rainfall intensity obeys the Chicago rain pattern distribution or normal distribution, and multiple groups of random samples are generated to quantify the uncertainty of the water level simulation results; on the parameter side, Bayesian statistics or sensitivity analysis tools are used to quantify the key parameters of roughness, bed morphology and weirs, and the parameter disturbance range is adjusted one by one to evaluate its impact on the simulation results. Through this analysis, the confidence interval of the water level simulation can be obtained; Step S3-4, after the simulation is completed, the water level process lines of key sections under different recurrence periods are generated; the water depth calculation results are superimposed on the DEM data, and the flood inundation range and flow velocity distribution map are output and presented in vector format or raster format to provide input for subsequent grid division, risk assessment and hierarchical zoning.

5. The method for differentiated early warning of flash flood disasters in small and medium-sized river basins according to claim 1 is characterized in that: The step S4 comprises the following sub-steps: Step S4-1, basin boundary determination, that is, according to the spatial distribution of flash flood warning objects, the downstream position of the basin to which the flash flood warning object belongs is selected as the outlet section, and the basin area is controlled; then, based on the DEM data of the study area, the spatial location data of the flash flood warning object, and the river system vector data, ArcGIS hydro tools are used to carry out terrain filling, flow direction analysis, water collection capacity analysis, and digital water system generation, and the basin boundary is delineated according to the basin outlet section; Use the plan layout of water conservancy projects such as river water boundary, bridge piers, and water-blocking weirs to cut the basin boundary and specify the land and water areas; Step S4-2: Discretization of watershed space and unstructured grid division, watershed characteristic analysis, land, water and river grid classification, unstructured grid division, and grid inspection and adjustment; Step S4-3, flash flood risk source identification and embedding, specifically, the potential hazard sources obtained from the investigation are superimposed on the grid database; and special marks are given to the grids where key bridges, traffic arteries and other infrastructure are located; the potential hazard sources include but are not limited to debris flow gullies, steep cliffs and weirs; Step S4-4, after completing the unstructured grid division, the DEM data, basic hydrological data and rainfall data are imported into the grid, and corresponding physical properties are assigned according to the characteristics of different areas. The characteristics of different areas include but are not limited to soil type, vegetation cover, slope, roughness and underlying surface.

6. The method for differentiated early warning of flash flood disasters in small and medium-sized river basins according to claim 5 is characterized in that: The step S4-2 also includes the following sub-steps: Step S4-2a, watershed characteristics analysis, i.e., paying special attention to mountain torrents in small and medium-sized watersheds that are susceptible to local heavy precipitation, including precipitation distribution, slope changes in watershed terrain, and characteristics of river channels within the watershed; Step S4-2b, land, water, and river grid classification, that is, the grid size is reduced for the upstream of the mountainous area and the area with a large slope; but for the low-slope area and the catchment area, the grid size is enlarged; for the river part of the mountain torrent small watershed, the river bend, and the confluence of tributaries, high-resolution grids are required for simulation; Step S4-2c, unstructured meshing, i.e. starting from the boundary of the watershed, selecting mesh nodes, performing triangulation based on the Delaunay criterion, and ensuring that the inner angle of the triangle is maximized; Step S4-2d, after the unstructured grid is divided, the grid is checked and dynamically adjusted by manual identification to ensure that the shape of each grid unit is reasonably set.

7. The method for differentiated early warning of flash flood disasters in small and medium-sized river basins according to claim 1 is characterized in that: The step S5 comprises the following sub-steps: Step S5-1, establishment of multidimensional data feature vector and Gaussian clustering GMM of grids, including construction of multidimensional feature vector for each grid unit, construction of risk factors and quantitative indicators of flash flood risk for each grid, definition of the number of clusters, calculation of attribution probability and fitting of data; Step S5-2, classification optimization and GIS display, that is, after completing Gaussian clustering, the classification results of the danger zone are optimized and displayed; first, the clustering results are manually checked, and the classification results that deviate from the actual situation are adjusted in combination with historical disaster records; Subsequently, the clustering results are superimposed on the administrative district boundaries, and the divisions are refined according to township or village boundaries to make the zoning results meet actual management needs.

8. The method for differentiated early warning of flash flood disasters in small and medium-sized river basins according to claim 7 is characterized in that: The step S5-1 further comprises the following sub-steps: Step S5-1a, constructing a multi-dimensional feature vector for each grid unit, that is, the characteristics of each grid contain multiple dimensions of risk factors related to flash flood disasters, including hydraulic characteristic parameters, flood forecast period, and risk source information parameters; Step S5-1b, constructing a nonlinear relationship between risk factors and flash flood risk quantitative indicators of each grid; the characteristic vector of each grid is represented as a multidimensional database containing risk factors and flash flood risk determination indicator parameters; Step S5-1c, defining the number of clusters, that is, defining the risk level and then defining the number of clusters, wherein the number of clusters is defined according to the watershed area and the degree of differentiation; Then initialize the parameters of each cluster, including: mean, covariance matrix and weight; Step S5-1d, based on the feature data of each grid: mean, covariance matrix and weight, by calculating the posterior probability that each grid unit belongs to each Gaussian distribution, thereby calculating the probability of each grid unit belonging to different clusters; Step S5-1e, optimize the partitioning result by maximizing the probability that the grid points belong to multiple Gaussian distributions so that each cluster can better fit the data. This process is achieved by maximizing the likelihood function, which is expressed as: ; Where N represents the total number of samples in the data set, that is, the number of data points; is the set of all parameters, , K is the total number of clusters, Represents the i-th sample data point in the data set; represents the weight of each Gaussian distribution, For the The probability density function of a Gaussian distribution is and are the mean vector and covariance matrix respectively; Step S5-1f, alternately perform steps S5-1d and S5-1e for multiple iterations until the mean of the model is , covariance matrix And weight It becomes stable or reaches the maximum number of iterations. Each iteration updates the probability of the grid belonging to the current model and gradually optimizes the division of clusters. In step S5-1g, the GMM model outputs the flash flood risk classification result of each grid, that is, the cluster to which each grid belongs.

9. The method for differentiated early warning of flash flood disasters in small and medium-sized river basins according to claim 1 is characterized in that: The step S6 also includes the following sub-steps: Step S6-1, differentiated setting of warning thresholds, i.e., based on the hierarchical zoning results of flash flood hazard areas, combined with the flood inundation range, basin lag time, key section water level process and time required for transfer in each area, determine the zoning warning thresholds, and establish differentiated warning trigger conditions for risk areas of different levels in the zone by combining historical data and field survey results; Step S6-2, design and implementation of differentiated early warning modes, that is, after completing the setting of early warning thresholds, differentiated early warning response modes are designed in combination with the risk characteristics of different regions. By introducing weather forecasts, rainfall monitoring, water level monitoring and other multi-source real-time data, a hierarchical early warning mechanism is established, and the early warning level is dynamically adjusted; Step S6-3: Release and linkage management of warning results, i.e., the warning information generated by the differentiated warning mode needs to be released within the region and linked with the emergency management system; the warning information is released through multiple channels, including text messages, broadcasts, warning terminal display screens and social media platforms; By combining the linkage function of the GIS platform and the early warning system, dynamically updated early warning information can be shared with management departments at all levels in real time.

10. A differentiated risk early warning method for mountain torrent disasters in small and medium-sized river basins according to claim 9, characterized in that: The step S6-1 further comprises the following sub-steps: Step S6-1a, defining corresponding warning indicators and threshold ranges for areas of different risk levels; Step S6-1b, using a weight analysis method to quantify the impact of different indicators to calibrate the warning threshold; the different indicators include water depth, flow velocity, rainfall, and basin hysteresis; Step S6-1c, verifying the setting of the regional warning threshold with the measured data, and calibrating the warning conditions in real time through retrospective analysis of historical flash flood events.

Images