A method for predicting rainstorm flood process, type and behavior change

Abstract

The application discloses a rainstorm and mountain torrent process, type and behavior change prediction method, and comprises the following steps: step 1, collecting and arranging the basic data of a research area; step 2, determining the small watershed attribute type and the rainstorm and mountain torrent type of the research area; step 3, constructing a small watershed rainstorm and mountain torrent comprehensive simulation model of the research area; step 4, determining the small watershed rainstorm and mountain torrent comprehensive simulation model parameters; and step 5, predicting the small watershed rainstorm and mountain torrent process, type and behavior change. The method can quickly and efficiently determine the mapping relationship between the small watershed attribute characteristics and the flood behavior characteristic change type, realize the quick conversion from the small watershed physical attribute type to the rainstorm and mountain torrent type, accurately evaluate the small watershed rainstorm and mountain torrent comprehensive simulation precision, and predict the rainstorm and mountain torrent process, type and behavior change of the small watershed with missing data, comprehensively represent the future flood change trend of the small watershed, and provide technical support for the rainstorm and mountain torrent prediction and risk management of the small watershed.

Description

Technical Field

[0001] This invention belongs to the field of rainstorm and flash flood prediction technology, and particularly relates to a method for predicting the process, type and behavior changes of rainstorm and flash floods. Background Technology

[0002] Against the backdrop of climate change and human activities, flash floods are becoming increasingly frequent and severe, posing a prominent challenge to my country's disaster prevention and mitigation efforts. Flash floods account for approximately 70% of all flood-related deaths and disappearances in my country each year. Flash floods in my country are characterized by their numerous locations, wide distribution, sudden onset, and significant regional variations. The changing characteristics of flash floods differ significantly under different hydrological, meteorological, and underlying surface conditions, leading to diverse modes of disaster formation. Furthermore, flash floods mostly occur in small watersheds in hilly areas where flood monitoring data is scarce, limiting the accuracy of flash flood forecasting in these small watersheds and hindering the provision of effective decision-making information for disaster prevention and mitigation.

[0003] Currently, methods for simulating and predicting flash floods mainly include data-driven models, hydrological models, and hybrid models. Data-driven models use statistical relationships between rainfall and runoff to simulate and forecast flash floods. While simple to model, the physical mechanisms are unclear. They require long-term flash flood data for training or calibration and are limited by the hydrological, meteorological, and underlying surface conditions of the sample watershed, making the analysis results difficult to transfer. Hydrological models use simplified mathematical and physical equations to generalize the formation and evolution of flash floods in complex watersheds. While complex to model, they possess certain physical mechanisms. Combined with measured flash flood data from areas with available data and regional parameter analysis techniques, they can simulate and predict flash flood events in areas lacking data. Hybrid models integrate data-driven models and hydrological models, comprehensively utilizing the potential information from long-term data and the physical interpretability of hydrological models. They aim to improve the accuracy of flood event simulation and prediction, but are currently still in the exploratory stage. The methods described above primarily focus on simulating and predicting flood events. However, due to the limited water storage capacity of small watersheds in mountainous areas, and the suddenness and short confluence time of flash floods, management departments typically extract peak flow and peak time information from the flood event to implement corresponding disaster prevention and mitigation measures, neglecting the changing characteristics of the flood process. Therefore, how to quickly, efficiently, and comprehensively predict the changing characteristics of flash floods in small watersheds, especially in watersheds lacking data, is a critical and urgent problem that needs to be solved.

[0004] Therefore, given the strong heterogeneity and scarcity of data regarding the changes in flash floods caused by rainstorms in small watersheds, there is an urgent need for a simple, feasible, and universally applicable method to comprehensively predict the process, type, and behavioral changes of flash floods caused by rainstorms in small watersheds. This method would comprehensively characterize the future flood trends in small watersheds and provide decision-making information support for flash flood disaster risk management. Summary of the Invention

[0005] The purpose of this invention is to provide a method for predicting the process, type, and behavior changes of rainstorms and flash floods, so as to solve the above-mentioned technical problems.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] This invention discloses a method for predicting the process, type, and behavior changes of flash floods caused by heavy rain. The method includes the following steps:

[0008] Step 1: Collect and organize basic data for the study area: Determine the catchment boundaries of all small watersheds in the study area, including those with available data and those without. Collect and organize basic geographic information data, climate background data, predicted rainfall data, and historical rainstorm and flash flood data of the available small watersheds. Extract all attribute indicators of small watersheds and historical flood behavior characteristic indicators of available small watersheds.

[0009] Step 2: Determine the attribute types and storm flash flood types of the study area's small watersheds: First, determine the attribute types of all small watersheds based on all the attribute indicators extracted in Step 1; then, determine the storm flash flood types of the data-rich small watersheds based on the historical flood behavior characteristic indicators extracted in Step 1; next, determine the storm flash flood types of the data-scarce small watersheds based on the attribute indicators of the data-scarce small watersheds; finally, establish the membership equations of each flood behavior characteristic indicator of the small watershed for each type of storm flash flood.

[0010] Step 3: Construct a comprehensive simulation model of rainstorms and flash floods in the study area's small watersheds: First, determine the distribution of sub-watersheds in the study area based on the basic geographic information data collected and organized in Step 1; then, determine the algorithm components for the formation and evolution of rainstorms and flash floods; finally, combine the confluence topology of the sub-watersheds to form a comprehensive simulation model of rainstorms and flash floods in the study area's small watersheds.

[0011] Step 4: Determine the parameters of the integrated simulation model for small watershed storm and flash flood: For small watersheds with available data, combine the historical storm and flash flood data collected and organized in Step 1, and use the integrated simulation model for small watershed storm and flash flood to obtain the simulated flood behavior characteristic indicators and flash flood processes; then, based on the simulated flood behavior characteristic indicators and the membership equation determined in Step 2, obtain the simulated storm and flash flood types; then evaluate the comprehensive simulation effect of storm and flash flood types, flood behavior characteristic indicators, and flash flood processes; finally, based on the storm and flash flood types of small watersheds with insufficient data determined in Step 2, determine the optimal parameter set for small watersheds with insufficient data, and thus obtain the optimal parameter set for all small watersheds in the study area;

[0012] Step 5: Predict the process, type, and behavior changes of rainstorms and flash floods in small watersheds: The predicted rainfall data collected and organized in Step 1 is used to drive the comprehensive simulation model of rainstorms and flash floods in the study area constructed in Step 3. Combined with the optimized parameter set of all small watersheds in the study area determined in Step 4, the flood behavior characteristic indicators and flash flood process of each small watershed are predicted. Then, based on the predicted flood behavior characteristic indicators of the small watersheds and the membership equation of the flood behavior characteristic indicators of the small watersheds determined in Step 2 for each type of rainstorm and flash flood, the rainstorm and flash flood type of each small watershed is predicted.

[0013] Furthermore, the basic geographic information data mentioned in step 1 includes digital elevation models, geological types, land use types, soil texture types, and water systems; the climate background data includes multi-year average precipitation, flood season precipitation, multi-year average number of rainstorm days, and multi-year average potential evapotranspiration; the small watershed attribute indicators include climate indicators, geological indicators, topographic indicators, vegetation cover indicators, soil texture indicators, water system indicators, and rainstorm indicators; and the flood behavior characteristic indicators include flood peak characteristic indicators, flood dynamic characteristic indicators, and seasonal indicators.

[0014] Furthermore, the catchment area of ​​the small watershed mentioned in step 1 does not exceed 200 km². 2 The data indicates that the small watershed is defined as a small watershed with data on no fewer than 5 rainstorms and flash floods.

[0015] Furthermore, the specific process of determining all small watershed attribute types based on all small watershed attribute indicators extracted in step 1 in step 2 is as follows: clustering algorithm is used to cluster all small watershed attribute indicators, and statistical evaluation indicators are combined to determine the optimal number of classification groups, thereby obtaining all small watershed attribute types and corresponding attribute indicator values.

[0016] The specific process of determining the type of rainstorm and flash flood in the data-rich small watershed based on the historical flood behavior characteristic indicators extracted in step 1 is as follows: the historical flood behavior characteristic indicators of the data-rich small watershed are clustered using a clustering algorithm, and the optimal number of classification groups is determined by combining statistical evaluation indicators to obtain the type of rainstorm and flash flood in the data-rich small watershed and the corresponding values ​​of the flood behavior characteristic indicators.

[0017] The specific process of determining the rainstorm and flash flood type of a small watershed with insufficient data based on its attribute indicators is as follows: the input layer is taken as the attribute types of all small watersheds and their corresponding attribute indicators, and the target layer is taken as the rainstorm and flash flood type of a small watershed with available data and its corresponding flood behavior characteristic indicators. Machine learning methods such as classification decision trees are used to establish the mapping relationship between the attribute types of small watersheds and the rainstorm and flash flood types. The attribute indicator classification thresholds of each small watershed rainstorm and flash flood type are determined. Combined with the attribute indicators of the small watershed with insufficient data, the rainstorm and flash flood type of the small watershed with insufficient data is obtained.

[0018] The specific process of establishing the membership equation of each flood behavior characteristic index of a small watershed for each type of rainstorm and flash flood is as follows: based on the frequency distribution of each flood behavior characteristic index of a small watershed in each type of rainstorm and flash flood, the membership equation of each flood behavior characteristic index of a small watershed for each type of rainstorm and flash flood is determined by frequency distribution curve fitting.

[0019] Furthermore, the clustering algorithm is dynamic K-means clustering or hierarchical clustering; the statistical evaluation index is one or more of the silhouette coefficient, C-index, and Dunn index.

[0020] Furthermore, the specific process for determining the sub-basin distribution of the study area based on the basic geographic information data of the study area collected and organized in step 1, as described in step 3, is as follows: Based on the basic geographic information data of the study area collected and organized in step 1, the data of digital elevation model, geological type, land use type, and soil texture type are overlaid and analyzed, according to a distance of 10–50 km. 2 To determine the catchment area threshold, sub-basins are divided, and the runoff topology of the sub-basins is determined.

[0021] The specific process for determining the rainstorm and flash flood formation evolution algorithm components is as follows: taking the boundaries of all small watershed catchment areas in the study area determined in step 1 as the modeling range, taking the sub-watershed as the smallest calculation unit, selecting rainstorm and flash flood formation evolution algorithms for each sub-watershed, including areal rainfall algorithm, runoff generation algorithm, confluence algorithm, gully flood evolution algorithm, and small-scale water conservancy facility flood regulation algorithm, to form rainstorm and flash flood formation evolution algorithm components;

[0022] The specific process of forming a comprehensive simulation model of rainstorm and flash flood in the study area by combining the confluence topology of sub-basins is as follows: Based on the rainstorm and flash flood formation and evolution algorithm components, combined with the confluence topology of sub-basins, the algorithm calculation order, input and output of each sub-basin are determined, the calculation network of the comprehensive simulation model of rainstorm and flash flood in the small watershed is constructed, and the comprehensive simulation model of rainstorm and flash flood in the small watershed of the study area is built.

[0023] Furthermore, the specific process of obtaining the simulated flood behavior characteristic indicators and flash flood process using the integrated simulation model of small watershed rainstorm and flash flood in step 4 is as follows: The sensitivity of the integrated simulation model of small watershed rainstorm and flash flood to the flood behavior characteristic indicators of each rainstorm and flash flood type is calculated by using the parameter sensitivity analysis method, the sensitivity parameters are screened, the optimal simulation of flood behavior characteristic indicators is taken as the objective function, the parameter automatic optimization algorithm is used to optimize the sensitivity parameters to determine the optimal parameter set, and the optimal parameter set is input into the integrated simulation model of small watershed rainstorm and flash flood to obtain the simulated flood behavior characteristic indicators and flash flood process;

[0024] The specific process of obtaining the simulated rainstorm and flash flood type based on the membership equation determined in step 2 is as follows: Based on the membership equation determined in step 2, calculate the membership degree of the simulated flood behavior characteristic index to each rainstorm and flash flood type, and the rainstorm and flash flood type with the highest membership degree is the simulated rainstorm and flash flood type.

[0025] The specific process for evaluating the comprehensive simulation effect of rainstorm and flash flood types, flood behavior characteristic indicators and flash flood processes is as follows: establish the accuracy evaluation index of the comprehensive simulation model of rainstorm and flash flood in small watersheds, and calculate the formula as shown in formula (1). Combine the simulated and measured values ​​of rainstorm and flash flood types, flood behavior characteristic indicators and flash flood processes to evaluate the comprehensive simulation effect of rainstorm and flash flood types, flood behavior characteristic indicators and flash flood processes.

[0026] f = w1f1 + w2f2 + w3f3 (1)

[0027] In the formula, f, f1, f2, and f3 represent the accuracy of the comprehensive simulation model of rainstorm and flash flood in a small watershed, the accuracy of the rainstorm and flash flood type, the accuracy of the flood behavior characteristic index, and the accuracy of the flash flood process, respectively; w1, w2, and w3 represent the weights of the rainstorm and flash flood type, the flood behavior characteristic index, and the flash flood process, respectively, and are assigned values ​​according to the importance of the three, so w1 + w2 + w3 = 1.0;

[0028] in,

[0029]

[0030] In the formula, N represents the number of historical data sessions for rainstorm and flash flood data; J represents the number of rainstorm and flash flood types; K represents the number of flood behavior characteristic indicators; μ ij The membership value is 1 if the i-th rainstorm and flash flood event belongs to the j-th rainstorm and flash flood type, and 0 otherwise; ε kj The simulation error of the k-th flood behavior characteristic index for the j-th rainstorm and flash flood type; NSE kj r is the Nash-Sutcliffe efficiency coefficient for the k-th flood behavior characteristic index of the j-th rainstorm and flash flood type; kj ε is the correlation coefficient of the k-th flood behavior characteristic index for the j-th type of rainstorm and flash flood; ij NSE represents the simulation error of the i-th rainstorm and flash flood of the j-th type; ij Let r be the Nash-Sutcliffe efficiency coefficient for the i-th rainstorm and flash flood of the j-th type; ij Let be the correlation coefficient of the i-th rainstorm and flash flood of the j-th type;

[0031] The specific process of determining the preferred parameter set for the data-deficient small watershed based on the rainstorm and flash flood types determined in step 2 is as follows: Based on the rainstorm and flash flood types of the data-deficient small watershed determined in step 2, select the preferred parameter set for the corresponding rainstorm and flash flood type of the data-deficient small watershed from the preferred parameter set for each rainstorm and flash flood type, which is the preferred parameter set for the data-deficient small watershed.

[0032] The beneficial effects of this invention are as follows: The method described in this invention combines machine learning, membership equations, and hydrological models, which can quickly and efficiently determine the mapping relationship between the attribute characteristics of small watersheds and the types of changes in flood behavior characteristics. It enables rapid conversion from the physical attribute type of small watersheds to the type of rainstorm and flash flood, accurately evaluates the accuracy of comprehensive simulation of rainstorm and flash floods in small watersheds, and predicts the process, type, and behavior changes of rainstorm and flash floods in small watersheds with missing data. It comprehensively characterizes the future flood change trend of small watersheds, providing technical support for the prediction, forecasting, and risk management of rainstorm and flash floods in small watersheds.

[0033] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0034] Figure 1 This is a flowchart of the method described in this invention;

[0035] Figure 2 This is a schematic diagram of the rainstorm and flash flood type classification regression decision tree constructed for the first small watershed attribute type in Example 1;

[0036] Figure 3 A schematic diagram of the rainstorm and flash flood type classification regression decision tree constructed for the second small watershed attribute type in Example 1;

[0037] Figure 4 This is a schematic diagram showing the distribution of the membership equations of flood rise rate and flood fall rate with respect to fast and slow flash floods in Example 1.

[0038] Figure 5 This is a schematic diagram of the predicted flood process in two small watersheds in Example 1. Detailed Implementation

[0039] This invention discloses a method for predicting the process, type, and behavioral changes of rainstorm flash floods, such as... Figure 1 As shown, the method includes the following steps:

[0040] Step 1: Collect and organize basic data for the study area. Determine the catchment boundaries of all small watersheds in the study area, including those with available data and those without. Collect and organize basic geographic information data, climate background data, predicted rainfall data, and historical storm and flash flood data for the available small watersheds. Basic geographic information data includes digital elevation models, geological types, land use types, soil texture types, and river systems; climate background data includes multi-year average precipitation, flood season precipitation, multi-year average number of storm days, and multi-year average potential evapotranspiration. Extract attribute indicators for all small watersheds and historical flood behavior characteristic indicators for available small watersheds. The catchment area of ​​each small watershed should not exceed 200 km². 2 Small watersheds with available data are those with no fewer than 5 rainstorm and flash flood data. Small watershed attribute indicators include climate indicators, geological indicators, topographic indicators, vegetation cover indicators, soil texture indicators, water system indicators, and rainstorm indicators. Flood behavior characteristic indicators include flood peak characteristics indicators, flood dynamic characteristics indicators, and seasonality indicators.

[0041] Step 2: Determine the attribute types and storm surge types of the study area's small watersheds: First, determine the attribute types of all small watersheds based on all the attribute indicators extracted in Step 1. Specifically, use clustering algorithms (such as dynamic K-means clustering or hierarchical clustering) to cluster the attribute indicators of all small watersheds, and combine them with statistical evaluation indicators (using one or more of the silhouette coefficient, C-index, and Dunn index) to determine the optimal number of classification groups, thus obtaining all small watershed attribute types and their corresponding attribute indicator values.

[0042] Then, based on the historical flood behavior characteristic indicators of the available small watershed extracted in step 1, the rainstorm and flash flood type of the available small watershed is determined. The specific process is as follows: the historical flood behavior characteristic indicators of the available small watershed are clustered using a clustering algorithm (such as dynamic K-means clustering or hierarchical clustering), and the optimal number of classification groups is determined by combining statistical evaluation indicators (using one or more of the silhouette coefficient, C index, and Dunn index) to obtain the rainstorm and flash flood type and the corresponding flood behavior characteristic indicator values ​​of the available small watershed.

[0043] Then, the rainstorm and flash flood type of the data-deficient watershed is determined by combining the attribute indicators of the data-deficient watershed. The specific process is as follows: the attribute types of all watersheds and their corresponding attribute indicators are taken as the input layer, and the rainstorm and flash flood types of the data-available watersheds and their corresponding flood behavior characteristic indicators are taken as the target layer. Machine learning methods such as classification decision trees are used to establish the mapping relationship between the attribute types of watersheds and the rainstorm and flash flood types. The attribute indicator classification thresholds of each watershed rainstorm and flash flood type are determined. Combined with the attribute indicators of the data-deficient watersheds, the rainstorm and flash flood types of the data-deficient watersheds are obtained.

[0044] Finally, the membership equations of each flood behavior characteristic index in the small watershed for each type of rainstorm and flash flood are established. The specific process is as follows: based on the frequency distribution of each flood behavior characteristic index in the small watershed for each type of rainstorm and flash flood in the available data, the membership equations of each flood behavior characteristic index in the small watershed for each type of rainstorm and flash flood are determined by frequency distribution curve fitting.

[0045] Step 3: Construct a comprehensive simulation model of flash floods and torrential rains in the study area's small watersheds. First, based on the basic geographic information data collected and organized in Step 1, determine the distribution of sub-watersheds in the study area. Specifically, based on the basic geographic information data collected and organized in Step 1, overlay analysis is performed on the digital elevation model, geological type, land use type, and soil texture type data, according to a range of 10–50 km. 2 Using the catchment area threshold as a guide, sub-basins are divided, and the confluence topology of the sub-basins is determined.

[0046] Then, the rainstorm and flash flood formation evolution algorithm components are determined. The specific process is as follows: taking the boundaries of all small watershed catchment areas in the study area determined in step 1 as the modeling range, and taking the sub-watershed as the smallest calculation unit, rainstorm and flash flood formation evolution algorithms are selected for each sub-watershed, including areal rainfall algorithm, runoff generation algorithm, confluence algorithm, gully flood evolution algorithm, and small-scale water conservancy facility flood regulation algorithm, to form the rainstorm and flash flood formation evolution algorithm components.

[0047] Finally, by combining the confluence topology of the sub-basins, a comprehensive simulation model of rainstorm and flash flood in the study area is formed. The specific process is as follows: based on the rainstorm and flash flood formation and evolution algorithm components and combined with the confluence topology of the sub-basins, the algorithm calculation order, input and output of each sub-basin are determined, the calculation network of the comprehensive simulation model of rainstorm and flash flood in the small watershed is constructed, and the comprehensive simulation model of rainstorm and flash flood in the small watershed of the study area is built.

[0048] Step 4: Determine the parameters of the integrated simulation model for small watershed rainstorms and flash floods: First, for small watersheds with available data, combined with the historical rainstorm and flash flood data collected and organized in Step 1, the sensitivity of the integrated simulation model for small watershed rainstorms and flash floods to the flood behavior characteristic indicators of each rainstorm and flash flood type is calculated using the parameter sensitivity analysis method. Sensitive parameters are screened, and with the optimal simulation of flood behavior characteristic indicators as the objective function, the automatic parameter optimization algorithm is used to optimize the sensitive parameters and determine the optimal parameter set. The optimal parameter set is then input into the integrated simulation model for small watershed rainstorms and flash floods to obtain the simulated flood behavior characteristic indicators and flash flood process.

[0049] Then, combining the simulated flood behavior characteristic indicators, and based on the membership equation determined in step 2, the membership degree of the simulated flood behavior characteristic indicators to each rainstorm and flash flood type is calculated. The rainstorm and flash flood type with the highest membership degree is the simulated rainstorm and flash flood type.

[0050] Then, the comprehensive simulation effect of rainstorm and flash flood type, flood behavior characteristic index and flash flood process is evaluated. The specific process is as follows: establish the accuracy evaluation index of the comprehensive simulation model of rainstorm and flash flood in small watershed, and calculate the formula as shown in formula (1). Combine the simulated value and measured value of rainstorm and flash flood type, flood behavior characteristic index and flash flood process to evaluate the comprehensive simulation effect of rainstorm and flash flood type, flood behavior characteristic index and flash flood process.

[0051] f = w1f1 + w2f2 + w3f3 (1)

[0052] In the formula, f, f1, f2, and f3 represent the accuracy of the comprehensive simulation model of rainstorm and flash flood in a small watershed, the accuracy of the rainstorm and flash flood type, the accuracy of the flood behavior characteristic index, and the accuracy of the flash flood process, respectively; w1, w2, and w3 represent the weights of the rainstorm and flash flood type, the flood behavior characteristic index, and the flash flood process, respectively, and are assigned values ​​according to the importance of the three, so w1 + w2 + w3 = 1.0;

[0053] in,

[0054]

[0055] In the formula, N represents the number of historical data sessions for rainstorm and flash flood data; J represents the number of rainstorm and flash flood types; K represents the number of flood behavior characteristic indicators; μ ij The membership value is 1 if the i-th rainstorm and flash flood event belongs to the j-th rainstorm and flash flood type, and 0 otherwise; ε kj The simulation error of the k-th flood behavior characteristic index for the j-th rainstorm and flash flood type; NSE kj r is the Nash-Sutcliffe efficiency coefficient for the k-th flood behavior characteristic index of the j-th rainstorm and flash flood type; kj ε is the correlation coefficient of the k-th flood behavior characteristic index for the j-th type of rainstorm and flash flood; ij NSE represents the simulation error of the i-th rainstorm and flash flood of the j-th type; ij Let r be the Nash-Sutcliffe efficiency coefficient for the i-th rainstorm and flash flood of the j-th type; ij Let be the correlation coefficient of the i-th rainstorm and flash flood of the j-th type.

[0056] Finally, based on the rainstorm and flash flood types of the data-deficient small watersheds determined in step 2, the optimal parameter set for the corresponding rainstorm and flash flood type of the data-deficient small watershed is selected from the optimal parameter set for each rainstorm and flash flood type. This is the optimal parameter set for the data-deficient small watershed, and thus the optimal parameter set for all small watersheds in the study area is obtained.

[0057] Step 5: Predict the process, type, and behavior changes of rainstorms and flash floods in small watersheds: The predicted rainfall data collected and organized in Step 1 is used to drive the comprehensive simulation model of rainstorms and flash floods in the study area constructed in Step 3. Combined with the optimized parameter set of all small watersheds in the study area determined in Step 4, the flood behavior characteristic indicators and flash flood process of each small watershed are predicted. Then, based on the predicted flood behavior characteristic indicators of the small watersheds and the membership equation of the flood behavior characteristic indicators of the small watersheds determined in Step 2 for each type of rainstorm and flash flood, the rainstorm and flash flood type of each small watershed is predicted.

[0058] Example 1

[0059] This embodiment is a specific application example of the above method.

[0060] This embodiment discloses a method for predicting the process, type, and behavior changes of rainstorm flash floods, including the following steps:

[0061] Step 1: Collect and organize basic data for the study area: Determine the catchment boundaries of all small watersheds in the study area, including 20 small watersheds (16 with available data and 4 without data), with a total catchment area of ​​53 km². 2 ~150km 2 We collected and organized basic geographic information data, climate background data, predicted rainfall data for the next 6 hours, and historical hourly rainstorm and flash flood data for 95 events in 16 data-available small watersheds. The basic geographic information data included a 1:50,000 digital elevation model, 1:5,000,000 geological types, 1:250,000 land use types, 1:500,000 soil texture types, and 1:100,000 river systems. The climate background data included multi-year average precipitation, flood season precipitation, multi-year average number of rainstorm days, and multi-year average potential evapotranspiration. We extracted attribute indicators for all small watersheds (as shown in Table 1) and historical flood behavior characteristic indicators for 16 data-available small watersheds (as shown in Table 2).

[0062] Table 1 List of attribute indicators for small watersheds

[0063]

[0064] Table 2 List of flood behavior characteristic indicators

[0065]

[0066] Step 2: Determine the attribute types and flash flood types of the study area's small watersheds: Based on all the attribute indicators extracted in Step 1, principal component analysis and dynamic K-means clustering were used to cluster all the attribute indicators of the small watersheds. Combined with the silhouette coefficient, the optimal number of classification groups was determined to be 2, thus determining 2 types of small watershed attribute indicators and their corresponding attribute indicator values. The first type is characterized by arid climate, flat terrain, strong infiltration, predominantly grassland, sparse river network, meandering river channels, and predominantly short-duration heavy rainfall. The second type is characterized by humid climate, steep terrain, predominantly forested land, high saturation, dense river network, straight river channels, and predominantly long-duration uniform rainfall. Principal component analysis and dynamic K-means clustering were used to cluster the flood behavior characteristic indicators of 95 events. Combined with the silhouette coefficient, the optimal number of classification groups was determined to be 2, thus determining 2 types of flash floods and their corresponding flood behavior characteristic indicator values ​​for the data-rich small watersheds. The first type of flash flood is a rapid flash flood, and the second is a slow flash flood.

[0067] The input layer uses two types of small watershed attribute types and their corresponding attribute indicators, while the target layer uses two types of data-available small watershed storm and flash flood types and their corresponding flood behavior characteristic indicators. The Classification and Regression Decision Tree (CART) method is used to establish the mapping relationship between small watershed attribute types and storm and flash flood types. The threshold values ​​for the attribute indicators corresponding to each small watershed storm and flash flood type are determined. The CART trees constructed for the first and second small watershed attribute types are shown below. Figure 2 and Figure 3 As shown. Figure 2 To illustrate the judgment rules, the following steps are taken: First, determine if the multi-year average potential evapotranspiration (ET) of the small watershed is greater than or equal to 800 mm. If ET ≥ 800 mm, then further determine if the topographic relief (TR) is greater than or equal to 300 m. If TR ≥ 300 m, it can be identified as a slow flash flood. If TR < 300 m, then further determine if the grassland area ratio (Gr) is greater than or equal to 0.4. If Gr ≥ 0.4, it can be identified as a fast flash flood; if Gr < 0.4, it can be identified as a slow flash flood. Then, combining the attribute indicators of the remaining four watersheds with missing data, the types of rainstorm flash floods in the four watersheds with missing data are obtained. Among them, the rainstorm flash floods formed in three watersheds with missing data are fast flash floods, and the rainstorm flash floods formed in one watershed with missing data are slow flash floods.

[0068] Based on the frequency distribution of flood behavior characteristic indicators from 95 events in 16 small watersheds with available data during rapid and slow flash floods, the membership equations of 10 flood behavior characteristic indicators for the two types of rainstorm flash floods were determined using the Gamma frequency distribution curve (as shown in formula (5)). Taking the flood rise rate and flood fall rate as examples, the distribution of their membership equations is as follows: Figure 4 As shown.

[0069]

[0070] In the formula, x is a flood behavior characteristic index; α is a shape parameter; β is a scale parameter; and Γ(·) is a gamma function.

[0071] Step 3: Construct a comprehensive simulation model of rainstorms and flash floods in the study area's small watersheds: First, based on the basic geographic information data of the study area collected and organized in Step 1, overlay analysis is performed on the digital elevation model, geological type, land use type, and soil texture type data, according to a range of 10–50 km. 2 Using the catchment area threshold, a total of 140 sub-basins were divided, and the confluence topology of the sub-basins was determined.

[0072] Then, taking the boundaries of 20 small watershed catchment areas as the modeling scope and the sub-watershed as the smallest calculation unit, a rainstorm flash flood formation evolution algorithm was selected for each sub-watershed. Specifically, the Thiessen polygon method was selected for areal rainfall calculation, the three-source full-storage runoff generation method was selected for runoff generation calculation, the distributed unitline method was selected for runoff calculation, and the dynamic Muskingan method was selected for channel flood evolution calculation, thus forming a rainstorm flash flood formation evolution algorithm component.

[0073] Finally, by combining the confluence topology of the sub-basins, the algorithm calculation order, input and output of each sub-basin are determined, and the calculation network of the comprehensive simulation model of small watershed rainstorm and flash flood is constructed, thus forming the comprehensive simulation model of small watershed rainstorm and flash flood in the study area.

[0074] Step 4: Determine the parameters of the integrated simulation model for small watershed storm and flash flood: For 16 small watersheds with available data, and combining the 95 storm and flash flood data collected in Step 1, the LH-OAT parameter sensitivity analysis method is used to calculate the sensitivity of the integrated simulation model for small watershed storm and flash flood to flood behavior characteristic indicators of two types of storm and flash flood. Sensitive parameters are selected respectively. Taking the optimal simulation of 10 flood behavior characteristic indicators as the objective function, the NSGA-II parameter automatic optimization algorithm is used to optimize the sensitive parameters and determine the optimal parameter set. The optimal parameter set is input into the integrated simulation model for small watershed storm and flash flood to obtain the simulated flood behavior characteristic indicators and flash flood process. According to the membership equation determined in Step 2, the membership degree of the simulated flood behavior characteristic indicators to each storm and flash flood type is calculated. The storm and flash flood type with the highest membership degree is the simulated storm and flash flood type. The accuracy of the integrated simulation model for small watershed storm and flash flood, the accuracy of storm and flash flood types, the accuracy of flood behavior characteristic indicators, and the accuracy of flash flood processes were calculated using formulas (1) to (4), as shown in Table 3. The integrated simulation effect of storm and flash flood types, flood behavior characteristic indicators, and flash flood processes was evaluated. Based on the storm and flash flood types of the small watersheds with missing data determined in step 2, the optimal parameter set for the small watersheds with missing data was determined, resulting in the optimal parameter set for all small watersheds in the study area.

[0075] Table 3 Comprehensive Simulation Assessment Indicators for Rainstorm and Flash Flood in Small Watersheds

[0076] 0.78 0.72 0.76 0.85

[0077] Step 5: Predict the process, type, and behavioral changes of flash floods caused by heavy rainfall in small watersheds:

[0078] The predicted rainfall data collected and organized in step 1 drives the comprehensive simulation model of rainstorm and flash flood in the study area constructed in step 3. Combined with the optimized parameter set for all small watersheds in the study area determined in step 4, the model predicts the flood behavior characteristics and flash flood processes of each small watershed. Then, based on the predicted flood behavior characteristics and the membership equation of each rainstorm and flash flood type determined in step 2, the predicted rainstorm and flash flood types for each small watershed are determined. In this embodiment, the predicted rainstorm and flash flood types for small watershed A and small watershed B in the study area are slow flash flood and fast flash flood, respectively. The predicted flood behavior characteristics are shown in Table 4, and the predicted flood processes are as follows: Figure 5 As shown.

[0079] Table 4. Characteristics of Predicted Flood Behavior in Two Small Watersheds

[0080]

[0081]

[0082] Finally, it should be noted that the above description is only used to illustrate the technical solution of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred arrangement, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for predicting the process, type, and behavioral changes of flash floods caused by heavy rain, characterized in that, The method includes the following steps: Step 1: Collect and organize basic data for the study area: Determine the catchment boundaries of all small watersheds in the study area, including those with available data and those without. Collect and organize basic geographic information data, climate background data, predicted rainfall data, and historical rainstorm and flash flood data of the available small watersheds. Extract all attribute indicators of small watersheds and historical flood behavior characteristic indicators of available small watersheds. Step 2: Determine the attribute types and storm flash flood types of the study area's small watersheds: First, determine the attribute types of all small watersheds based on all the attribute indicators extracted in Step 1; then, determine the storm flash flood types of the data-rich small watersheds based on the historical flood behavior characteristic indicators extracted in Step 1; next, determine the storm flash flood types of the data-scarce small watersheds based on the attribute indicators of the data-scarce small watersheds; finally, establish the membership equations of each flood behavior characteristic indicator of the small watershed for each type of storm flash flood. The specific process of determining the rainstorm and flash flood type of a small watershed with insufficient data based on its attribute indicators is as follows: the input layer is taken as the attribute types of all small watersheds and their corresponding attribute indicators, and the target layer is taken as the rainstorm and flash flood type of a small watershed with available data and its corresponding flood behavior characteristic indicators. A machine learning method is used to establish the mapping relationship between the attribute types of small watersheds and the rainstorm and flash flood types. The attribute indicator classification thresholds of each small watershed's rainstorm and flash flood type are determined. Combined with the attribute indicators of the small watershed with insufficient data, the rainstorm and flash flood type of the small watershed with insufficient data is obtained. Step 3: Construct a comprehensive simulation model of rainstorms and flash floods in the study area's small watersheds: First, determine the distribution of sub-watersheds in the study area based on the basic geographic information data collected and organized in Step 1; then, determine the algorithm components for the formation and evolution of rainstorms and flash floods; finally, combine the confluence topology of the sub-watersheds to form a comprehensive simulation model of rainstorms and flash floods in the study area's small watersheds. Step 4: Determine the parameters of the integrated simulation model for small watershed storm and flash flood: For small watersheds with available data, combine the historical storm and flash flood data collected and organized in Step 1, and use the integrated simulation model for small watershed storm and flash flood to obtain the simulated flood behavior characteristic indicators and flash flood processes; then, based on the simulated flood behavior characteristic indicators and the membership equation determined in Step 2, obtain the simulated storm and flash flood types; then evaluate the comprehensive simulation effect of storm and flash flood types, flood behavior characteristic indicators, and flash flood processes; finally, based on the storm and flash flood types of small watersheds with insufficient data determined in Step 2, determine the optimal parameter set for small watersheds with insufficient data, and thus obtain the optimal parameter set for all small watersheds in the study area; The specific process of determining the preferred parameter set of the data-deficient small watershed based on the rainstorm and flash flood types determined in step 2 is as follows: Based on the rainstorm and flash flood types of the data-deficient small watershed determined in step 2, select the preferred parameter set of the corresponding rainstorm and flash flood type of the data-deficient small watershed from the preferred parameter set of each rainstorm and flash flood type, which is the preferred parameter set of the data-deficient small watershed. Step 5: Predict the process, type, and behavior changes of rainstorms and flash floods in small watersheds: The predicted rainfall data collected and organized in Step 1 is used to drive the comprehensive simulation model of rainstorms and flash floods in the study area constructed in Step 3. Combined with the optimized parameter set of all small watersheds in the study area determined in Step 4, the flood behavior characteristic indicators and flash flood process of each small watershed are predicted. Then, based on the predicted flood behavior characteristic indicators of the small watersheds and the membership equation of the flood behavior characteristic indicators of the small watersheds determined in Step 2 for each type of rainstorm and flash flood, the rainstorm and flash flood type of each small watershed is predicted.

2. The method for predicting the process, type, and behavior changes of rainstorm flash floods according to claim 1, characterized in that, The basic geographic information data mentioned in step 1 includes digital elevation models, geological types, land use types, soil texture types, and water systems; the climate background data includes multi-year average precipitation, flood season precipitation, multi-year average number of rainstorm days, and multi-year average potential evapotranspiration; the small watershed attribute indicators include climate indicators, geological indicators, topographic indicators, vegetation cover indicators, soil texture indicators, water system indicators, and rainstorm indicators; the flood behavior characteristic indicators include flood peak characteristic indicators, flood dynamic characteristic indicators, and seasonal indicators.

3. The method for predicting the process, type, and behavior changes of rainstorm flash floods according to claim 1, characterized in that, The catchment area of ​​the small watershed mentioned in step 1 does not exceed 200 km². 2 The data indicates that the small watershed is defined as a small watershed with data on no fewer than 5 rainstorms and flash floods.

4. The method for predicting the process, type, and behavior changes of rainstorm flash floods according to claim 1, characterized in that, The specific process for determining all small watershed attribute types based on all small watershed attribute indicators extracted in step 1 in step 2 is as follows: use a clustering algorithm to cluster all small watershed attribute indicators, combine statistical evaluation indicators to determine the optimal number of classification groups, and obtain all small watershed attribute types and corresponding attribute indicator values. The specific process of determining the type of rainstorm and flash flood in the data-rich small watershed based on the historical flood behavior characteristic indicators extracted in step 1 is as follows: the historical flood behavior characteristic indicators of the data-rich small watershed are clustered using a clustering algorithm, and the optimal number of classification groups is determined by combining statistical evaluation indicators to obtain the type of rainstorm and flash flood in the data-rich small watershed and the corresponding values ​​of the flood behavior characteristic indicators. The specific process of establishing the membership equation of each flood behavior characteristic index of a small watershed for each type of rainstorm and flash flood is as follows: based on the frequency distribution of each flood behavior characteristic index of a small watershed in each type of rainstorm and flash flood, the membership equation of each flood behavior characteristic index of a small watershed for each type of rainstorm and flash flood is determined by frequency distribution curve fitting.

5. The method for predicting the process, type, and behavior changes of rainstorm flash floods according to claim 4, characterized in that, The clustering algorithm is dynamic K-means clustering or hierarchical clustering; the statistical evaluation index is one or more of the silhouette coefficient, C-index, and Dunn index.

6. The method for predicting the process, type, and behavior changes of rainstorm flash floods according to claim 1, characterized in that, The specific process for determining the sub-basin distribution of the study area based on the basic geographic information data collected and organized in Step 1, as described in Step 3, is as follows: Based on the basic geographic information data of the study area collected and organized in Step 1, overlay analysis is performed on the digital elevation model, geological type, land use type, and soil texture type data, according to a distance of 10–50 km. 2 To determine the catchment area threshold, sub-basins are divided, and the runoff topology of the sub-basins is determined. The specific process for determining the rainstorm and flash flood formation evolution algorithm components is as follows: taking the boundaries of all small watershed catchment areas in the study area determined in step 1 as the modeling range, taking the sub-watershed as the smallest calculation unit, selecting rainstorm and flash flood formation evolution algorithms for each sub-watershed, including areal rainfall algorithm, runoff generation algorithm, confluence algorithm, gully flood evolution algorithm, and small-scale water conservancy facility flood regulation algorithm, to form rainstorm and flash flood formation evolution algorithm components; The specific process of forming a comprehensive simulation model of rainstorm and flash flood in the study area by combining the confluence topology of sub-basins is as follows: Based on the rainstorm and flash flood formation and evolution algorithm components, combined with the confluence topology of sub-basins, the algorithm calculation order, input and output of each sub-basin are determined, the calculation network of the comprehensive simulation model of rainstorm and flash flood in the small watershed is constructed, and the comprehensive simulation model of rainstorm and flash flood in the small watershed of the study area is built.

7. The method for predicting the process, type, and behavior changes of rainstorm flash floods according to claim 1, characterized in that, The specific process of obtaining the simulated flood behavior characteristic indicators and flash flood process using the integrated simulation model of small watershed rainstorm and flash flood in step 4 is as follows: The sensitivity of the integrated simulation model of small watershed rainstorm and flash flood to the flood behavior characteristic indicators of each rainstorm and flash flood type is calculated by using the parameter sensitivity analysis method, the sensitivity parameters are screened, the optimal simulation of flood behavior characteristic indicators is taken as the objective function, the parameter automatic optimization algorithm is used to optimize the sensitivity parameters to determine the optimal parameter set, and the optimal parameter set is input into the integrated simulation model of small watershed rainstorm and flash flood to obtain the simulated flood behavior characteristic indicators and flash flood process; The specific process of obtaining the simulated rainstorm and flash flood type based on the membership equation determined in step 2 is as follows: Based on the membership equation determined in step 2, calculate the membership degree of the simulated flood behavior characteristic index to each rainstorm and flash flood type, and the rainstorm and flash flood type with the highest membership degree is the simulated rainstorm and flash flood type. The specific process for evaluating the comprehensive simulation effect of rainstorm and flash flood types, flood behavior characteristic indicators, and flash flood processes is as follows: Establish an accuracy evaluation index for the comprehensive simulation model of rainstorm and flash floods in a small watershed, calculated using the formula (1). Combine the simulated and measured values ​​of rainstorm and flash flood types, flood behavior characteristic indicators, and flash flood processes to evaluate the comprehensive simulation effect of these factors. （1） In the formula, f, f1, f2, and f3 represent the accuracy of the comprehensive simulation model of rainstorm and flash flood in a small watershed, the accuracy of the rainstorm and flash flood type, the accuracy of the flood behavior characteristic index, and the accuracy of the flash flood process, respectively; w1, w2, and w3 represent the weights of the rainstorm and flash flood type, the flood behavior characteristic index, and the flash flood process, respectively, and are assigned values ​​according to the importance of the three, so w1 + w2 + w3 = 1.0; in, （2） （3） （4） In the formula, N represents the number of historical data sessions for rainstorm and flash flood data; J represents the number of rainstorm and flash flood types; K represents the number of flood behavior characteristic indicators; μ ij The membership value is 1 if the i-th rainstorm and flash flood event belongs to the j-th rainstorm and flash flood type, and 0 otherwise; ε kj The simulation error of the k-th flood behavior characteristic index for the j-th rainstorm and flash flood type; NSE kj r is the Nash-Sutcliffe efficiency coefficient for the k-th flood behavior characteristic index of the j-th rainstorm and flash flood type; kj ε is the correlation coefficient of the k-th flood behavior characteristic index for the j-th type of rainstorm and flash flood; ij NSE represents the simulation error of the i-th rainstorm and flash flood of the j-th type; ij Let r be the Nash-Sutcliffe efficiency coefficient for the i-th rainstorm and flash flood of the j-th type; ij Let be the correlation coefficient of the i-th rainstorm and flash flood of the j-th type.

Images