Two-dimensional flood risk prediction method and device, electronic equipment, medium and product

Abstract

The application relates to a two-dimensional flood risk prediction method, device, electronic equipment, medium and product, which comprises the following steps: acquiring multi-modal data in a target river basin; obtaining simulated runoff results based on a plurality of hydrological models pre-established according to the multi-modal data in the target river basin, and establishing a hydrological-machine learning coupled model based on the simulated runoff results; simulating the river basin hydrological process under a future scenario through the hydrological-machine learning coupled model to obtain a river basin hydrological process simulation result under the future scenario, and extracting flood duration and flood volume characteristic values under a plurality of warming level time windows by using a preset optimal model based on the river basin hydrological process simulation result, obtaining a two-dimensional flood risk prediction result based on the flood duration and flood volume characteristic values under the plurality of warming level time windows, thereby solving the problems of low flood risk prediction accuracy and poor applicability in the related art, and improving the flood risk prediction accuracy and applicability.

Description

Technical Field

[0001] This invention relates to the field of hydrological disaster assessment technology, and in particular to a two-dimensional flood risk prediction method, device, electronic equipment, medium and product. Background Technology

[0002] As global climate change intensifies its impact on regional ecological environments and socio-economic conditions, global warming, a major characteristic of climate change, alters the thermal and dynamic environment of the climate system, affects the global hydrological cycle, and increases extreme precipitation events and floods. In regions most severely affected by floods, the rate of warming is far higher than the global average, with temperatures potentially rising by 4°C by the end of this century, posing a serious threat to flood control, water supply, food security, energy security, and ecological environment security. A deep understanding of flood evolution and its socio-economic impacts under global warming is crucial for predicting future extreme climate disaster risks, disaster prevention and mitigation, and adaptation management.

[0003] Among the related technologies, the evolution of future floods has been studied by combining global climate model sets and watershed hydrological models. The future climate forecasts are usually represented by near-term (2021-2040), medium-term (2041-2060), and long-term (2081-2100).

[0004] However, due to the global warming effect caused by increased emissions of greenhouse gases such as carbon dioxide, the frequency and intensity of floods are exhibiting new characteristics. This challenges the applicability and accuracy of traditional methods for flood risk prediction under climate warming. Furthermore, related technologies do not consider flood risk assessment at specific warming levels and generally focus on single attributes such as flood peak, flood volume, or flood duration, failing to reflect the multivariate characteristics of floods. These issues urgently need to be addressed. Summary of the Invention

[0005] This invention provides a two-dimensional flood risk prediction method, device, electronic equipment, medium, and product to solve the problems of low accuracy and poor applicability of flood risk prediction in related technologies, and to improve the accuracy and applicability of flood risk prediction.

[0006] A first aspect of the present invention provides a two-dimensional flood risk prediction method, comprising the following steps: acquiring multimodal data within a target watershed; obtaining simulated runoff results based on multiple pre-established hydrological models using the multimodal data within the target watershed, and establishing a hydrological-machine learning coupled model based on the simulated runoff results; simulating the watershed hydrological process under future scenarios using the hydrological-machine learning coupled model to obtain simulation results of the watershed hydrological process under future scenarios; and extracting flood duration and flood volume feature values ​​under multiple warming level time windows using a preset optimal model based on the flood duration and flood volume feature values ​​under the multiple warming level time windows to obtain a two-dimensional flood risk prediction result.

[0007] Further, in some embodiments, the multimodal data includes simulation data from multiple global climate models within the target watershed, meteorological data from the ERA5 dataset, socioeconomic data, and daily flow series data from control hydrological stations in the target watershed. The step of obtaining simulated runoff results based on multiple pre-established hydrological models using the multimodal data within the target watershed, and establishing a hydrological-machine learning coupled model based on the simulated runoff results, includes: calculating relative humidity and specific humidity results based on the meteorological data from the ERA5 dataset; obtaining the simulated runoff results using multiple pre-set hydrological models based on the daily flow series data from control hydrological stations in the target watershed and the meteorological data from the ERA5 dataset; determining the simulated runoff duration of the machine learning model based on the simulated runoff results and pre-collected daily measured runoff results, according to a pre-set correlation threshold; and correcting the simulated runoff results using a long short-term memory neural network with a coupled attention mechanism to obtain the hydrological-machine learning coupled model.

[0008] Furthermore, in some embodiments, before simulating the watershed hydrological process under the future scenario using the hydrology-machine learning coupling model to obtain the simulation results of the watershed hydrological process under the future scenario, the method further includes: correcting the simulation data of the multiple global climate models based on a decoding-encoding bias correction method to obtain a series of corrected meteorological simulation results.

[0009] Furthermore, in some embodiments, the step of simulating the watershed hydrological process under future scenarios using the hydrology-machine learning coupling model to obtain the simulation results of the watershed hydrological process under future scenarios includes: simulating the watershed hydrological process under future scenarios using the hydrology-machine learning coupling model based on the corrected meteorological simulation series results to obtain the daily runoff series results under future climate change scenarios; and obtaining the watershed hydrological process simulation results under future scenarios based on the daily runoff series results under future climate change scenarios.

[0010] Furthermore, in some embodiments, before extracting flood duration and flood volume feature values ​​under multiple warming level time windows using a preset optimal model based on the watershed hydrological process simulation results, the method further includes: calculating time windows under different warming levels based on temperature data in the corrected meteorological simulation series results; fusing the corrected meteorological simulation series results with the machine learning model to obtain a fused machine learning model; determining the model parameters of the fused machine learning model based on a preset adaptive model selection mechanism, and obtaining the preset optimal model based on the determination result.

[0011] Further, in some embodiments, obtaining the two-dimensional flood risk prediction result based on the flood duration and flood volume characteristic values ​​under the multiple warming level time windows includes: constructing a marginal distribution function for flood duration and flood volume based on the flood duration and flood volume characteristic values ​​under the multiple warming level time windows, and constructing a Copula joint distribution function for flood duration and flood volume under inconsistent conditions; calculating the joint return period of flood duration and flood volume under different warming level time windows based on the marginal distribution function of flood duration and flood volume and the Copula joint distribution function of flood duration and flood volume under inconsistent conditions, and obtaining the two-dimensional flood risk prediction result.

[0012] The two-dimensional flood risk prediction method provided by the present invention first collects basic meteorological and hydrological data of the watershed, then derives humidity-related data, calibrates the hydrological model, and constructs a machine learning model; then, it obtains a series of meteorological simulations under climate change scenarios through decoding-encoding bias correction, driving the model to simulate future hydrological processes; based on the optimal model, it extracts flood feature values ​​and constructs a joint probability distribution function; finally, it obtains the two-dimensional flood risk prediction result, solving the problems of low accuracy and poor applicability of flood risk prediction in related technologies, and improving the accuracy and applicability of flood risk prediction.

[0013] A second aspect of the present invention provides a two-dimensional flood risk prediction device, wherein the device comprises: an acquisition module for acquiring multimodal data within a target watershed; a mathematical modeling module for obtaining simulated runoff results based on multiple pre-established hydrological models according to the multimodal data within the target watershed, and establishing a hydrological-machine learning coupled model based on the simulated runoff results; and a prediction module for simulating the watershed hydrological process under future scenarios using the hydrological-machine learning coupled model, obtaining simulation results of the watershed hydrological process under future scenarios, and extracting flood duration and flood volume feature values ​​under multiple warming level time windows using a preset optimal model based on the simulation results of the watershed hydrological process, and obtaining two-dimensional flood risk prediction results based on the flood duration and flood volume feature values ​​under the multiple warming level time windows.

[0014] Furthermore, in some embodiments, the mathematical modeling module is specifically used for: calculating relative humidity and specific humidity results based on meteorological data from the ERA5 dataset; obtaining the simulated runoff results based on daily flow series data from the target watershed control hydrological station and meteorological data from the ERA5 dataset through multiple preset hydrological models; determining the simulated runoff duration of the machine learning model based on the simulated runoff results and pre-collected daily measured runoff results according to a preset correlation threshold; and correcting the simulated runoff results based on a long short-term memory neural network with a coupled attention mechanism to obtain the hydrology-machine learning coupled model.

[0015] Furthermore, in some embodiments, before simulating the watershed hydrological process under the future scenario using the hydrology-machine learning coupled model to obtain the simulation results of the watershed hydrological process under the future scenario, the prediction module is also used to: correct the simulation data of the multiple global climate models based on the decoding-encoding bias correction method to obtain the corrected meteorological simulation series results.

[0016] Furthermore, in some embodiments, the prediction module is also used to: simulate the watershed hydrological process under future scenarios based on the corrected meteorological simulation series results, and obtain the daily runoff series results under future climate change scenarios; and obtain the watershed hydrological process simulation results under the future scenarios based on the daily runoff series results under the future climate change scenarios.

[0017] Furthermore, in some embodiments, before extracting flood duration and flood volume feature values ​​under multiple warming level time windows using a preset optimal model based on the watershed hydrological process simulation results, the prediction module is further configured to: calculate time windows under different warming levels based on temperature data in the corrected meteorological simulation series results; fuse the corrected meteorological simulation series results with the machine learning model to obtain a fused machine learning model; determine the model parameters of the fused machine learning model based on a preset adaptive model selection mechanism, and obtain the preset optimal model based on the determination result.

[0018] Furthermore, in some embodiments, the prediction module is also used to: construct marginal distribution functions for flood duration and flood volume based on the flood duration and flood volume characteristic values ​​under the multiple warming level time windows, and construct a Copula joint distribution function for flood duration and flood volume under inconsistent conditions; calculate the joint return period of flood duration and flood volume under different warming level time windows based on the marginal distribution function of flood duration and flood volume and the Copula joint distribution function of flood duration and flood volume under inconsistent conditions, and obtain the two-dimensional flood risk prediction result.

[0019] The two-dimensional flood risk prediction device provided in this embodiment of the invention first collects basic meteorological and hydrological data of the watershed, then derives humidity-related data, calibrates the hydrological model, and constructs a machine learning model; then, it obtains a series of meteorological simulations under climate change scenarios through decoding-encoding bias correction, driving the model to simulate future hydrological processes; based on the optimal model, it extracts flood feature values ​​and constructs a joint probability distribution function; finally, it obtains the two-dimensional flood risk prediction result, solving the problems of low accuracy and poor applicability of flood risk prediction in related technologies, and improving the accuracy and applicability of flood risk prediction.

[0020] A third aspect of the present invention provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described two-dimensional flood risk prediction method.

[0021] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the two-dimensional flood risk prediction method as described in the above embodiments.

[0022] A fifth aspect of the present invention provides a computer program product, including a computer program that is executed to implement the two-dimensional flood risk prediction method as described in any of the preceding claims.

[0023] Therefore, the present invention has the following advantages:

[0024] (1) This invention fully considers the non-consistent characteristics of hydrological series under the influence of climate change and human underlying surface activities, constructs a hydrological-machine learning coupled model with attention mechanism to realize high reliability simulation of runoff, and constructs a time-varying Copula model considering the non-consistency of hydrological series. It has strong physical meaning and statistical basis and can effectively characterize the changing characteristics of future floods under climate warming.

[0025] (2) This invention combines a multi-model climate model, a hydrological-machine learning coupled model with an attention mechanism, and the most probable combination scenario method with different global warming levels. It can provide important and highly operable reference for basin flood risk assessment and early warning under changing environments, and provide engineering reference value for responding to future climate disasters and scientifically formulating emission reduction strategies. Attached Figure Description

[0026] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0027] Figure 1 This is a flowchart of a two-dimensional flood risk prediction method provided according to an embodiment of the present invention;

[0028] Figure 2 This is a schematic diagram illustrating the change in the correlation coefficient between measured daily runoff and simulated runoff under different time lags, according to a specific embodiment of the present invention.

[0029] Figure 3 This is a schematic diagram of a hydrology-machine learning coupled model structure provided according to a specific embodiment of the present invention;

[0030] Figure 4 A schematic diagram of the structure of a memory unit in a Long Short-Term Memory (LSTM) neural network model provided according to a specific embodiment of the present invention;

[0031] Figure 5 A schematic diagram of the structure of a Convolutional Attention Module (CBAM) according to a specific embodiment of the present invention;

[0032] Figure 6 A flowchart of a two-dimensional flood risk prediction method according to a specific embodiment of the present invention;

[0033] Figure 7 This is a block diagram of a two-dimensional flood risk prediction device provided according to an embodiment of the present invention;

[0034] Figure 8 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of the present invention. Detailed Implementation

[0035] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0036] The following description, with reference to the accompanying drawings, outlines a two-dimensional flood risk prediction method, apparatus, electronic device, medium, and product according to embodiments of the present invention. Addressing the issues of low accuracy and poor applicability in flood risk prediction in related technologies mentioned in the background section, the present invention provides a two-dimensional flood risk prediction method. This method first collects basic meteorological and hydrological data of the watershed, then derives humidity-related data, calibrates the hydrological model, and constructs a machine learning model. Next, it obtains a series of meteorological simulations under climate change scenarios through decoding-encoding bias correction, driving the model to simulate future hydrological processes. Based on the optimal model, flood feature values ​​are extracted, and a joint probability distribution function is constructed. Finally, a two-dimensional flood risk prediction result is obtained, solving the problems of low accuracy and poor applicability in flood risk prediction in related technologies, and improving the accuracy and applicability of flood risk prediction.

[0037] Specifically, Figure 1This is a flowchart of a two-dimensional flood risk prediction method provided according to an embodiment of the present invention.

[0038] like Figure 1 As shown, this two-dimensional flood risk prediction method includes the following steps:

[0039] In step S101, multimodal data within the target watershed is acquired.

[0040] Among them, multimodal data within the target watershed refers to a collection of data with various data types and representations collected through multiple different sensors, observation methods, and data sources within a specific defined watershed area.

[0041] Specifically, the multimodal data includes simulation data from multiple global climate models within the target watershed, meteorological data from the ERA5 dataset, socioeconomic data, and daily flow series data from the control hydrological stations in the target watershed. In this embodiment of the invention, the daily flow series from the control hydrological stations in the watershed are collected, and meteorological data such as precipitation, 2m air temperature, 2m dew point temperature, wind speed, air pressure, shortwave radiation, and longwave radiation are obtained from the ERA5 reanalysis dataset.

[0042] For example, this embodiment of the invention uses a watershed as the research unit. First, daily flow series from watershed control hydrological stations are collected, and then hourly data from ERA5 are acquired. ERA5 is the fifth-generation atmospheric reanalysis dataset from the European Centre for Medium-Range Weather Forecasts (ECMWF), with a spatial resolution of 0.25°, providing hourly meteorological data covering the globe since 1979. This embodiment acquires hourly precipitation, 2-meter air temperature, 2-meter dew point temperature, wind speed, air pressure, shortwave radiation, and longwave radiation data from the ERA5 dataset for the study area from 1985 to 2014. After time-scale transformation, daily series are obtained, and finally, watershed-averaged daily meteorological series are obtained using the Thiessen polygon method.

[0043] Furthermore, daily meteorological data from M global climate models (GCMs) were collected. For predicting future climate scenarios, the latest M global climate models (GCMs) released in the Coupled Model Inter-comparison Project Phase 6 (CMIP6) were used. CMIP6 employs a matrix framework of the Shared Socioeconomic Pathway (SSP) and the Representative Concentration Pathway (RCP). The scenarios selected in this invention include three scenarios (SSP245, SSP370, and SSP585) covering historical and future periods. The selected meteorological variables are daily precipitation, daily average temperature, daily maximum temperature, daily minimum temperature, specific humidity, relative humidity, wind speed, shortwave radiation, and longwave radiation data. Simultaneously, annual-scale potential evapotranspiration data from the GCMs under the three SSP scenarios were acquired. The historical period was set as 1985-2014, and the future period as 2015-2100.

[0044] Furthermore, population, GDP, and land use data, as well as vulnerability indices, were collected from shared socioeconomic pathway datasets for the study basin. To assess the potential socioeconomic risks posed by flood events, population, GDP, and land use data for three shared socioeconomic pathways—moderate development (SSP2), competitive development (SSP3), and conventional development (SSP5)—were considered and combined with corresponding greenhouse gas emission scenarios (RCPs). Output data were generated using three matrix frameworks: SSP245, SSP370, and SSP585. Several international institutions provide population and GDP simulation data for shared socioeconomic pathways. This invention uses open-source projection datasets based on relevant policies. This product considers results from previous population and economic censuses and annual statistical yearbooks, and, based on the Cobb-Douglas model and the population-development-environment analysis model, projects the socioeconomic indices for the relevant region from 2010 to 2100. The Land-Use Harmonization2 (LUH2) dataset is a global land use dataset with a spatial resolution of 0.25°. It provides land use patterns, potential land use transitions, and key agricultural management information over a period of 850–2100 years and is widely used for socioeconomic risk assessment of extreme hydrological events.

[0045] Furthermore, the vulnerability coefficients of population, GDP, and agriculture to floods are obtained, denoted as Vul POP Vul GDP and Vul ALDBased on this coefficient, socioeconomic risks can be further assessed.

[0046] It should be noted that after obtaining raster-scale population, GDP and land use data in the embodiments of the present invention, the Thiessen polygon method is used to estimate the average population, GDP and farmland area of ​​the watershed under future climate change scenarios.

[0047] In step S102, based on multimodal data within the target watershed, simulated runoff results are obtained using multiple pre-established hydrological models, and a hydrological-machine learning coupled model is established based on the simulated runoff results.

[0048] Among them, the pre-established hydrological models are hydrological models built based on the multidimensional characteristics of different watersheds, and the hydrological-machine learning coupling model is a deep learning model built by considering hydrological data and machine learning models, which is used to predict flood risk.

[0049] In some embodiments, the multimodal data includes simulation data from multiple global climate models within the target watershed, meteorological data from the ERA5 dataset, socioeconomic data, and daily flow series data from control hydrological stations in the target watershed. Based on the multimodal data within the target watershed, simulated runoff results are obtained using multiple pre-established hydrological models. A hydrological-machine learning coupled model is then established based on the simulated runoff results, including: calculating relative humidity and specific humidity results based on meteorological data from the ERA5 dataset; obtaining simulated runoff results using multiple pre-set hydrological models based on daily flow series data from control hydrological stations in the target watershed and meteorological data from the ERA5 dataset; determining the simulated runoff duration of the machine learning model based on the simulated runoff results and pre-collected daily measured runoff results, according to a pre-set correlation threshold; and correcting the simulated runoff results using a long short-term memory neural network with a coupled attention mechanism to obtain the hydrological-machine learning coupled model.

[0050] Specifically, this invention derives relative humidity and specific humidity from meteorological data in the ERA5 dataset, and quantitatively describes the saturated vapor pressure e using the Clausius-Clapeyron thermodynamic equation. sat Nonlinear relationship with temperature T:

[0051]

[0052] Among them, T0 and e s0 L is the integration constant, taken as 273.16 K and 611 Pa respectively; v As the latent heat of vaporization, take 2.5 × 10⁻⁶. 6 Jkg -1 ;R v Let be the water vapor gas constant, taken as 461 J kg. -1 K -1 .

[0053] Dew point temperature characterizes the temperature at which air, under constant water vapor content and pressure, cools to water vapor saturation. Substituting this into the Clausius-Clapeyron equation, it can measure the actual water vapor pressure. ERA5 2m air temperature (T) 2m ) and dew point temperature (T dew Substitute them respectively

[0054] The Clausius-Clapeyron thermodynamic equation is used to derive the near-surface relative humidity: RH = e sat (T dew ) / e sat (T 2m );

[0055] Specific humidity q is the ratio of water vapor mass to the total mass of the air mass, calculated using ERA5 near-surface pressure p and dew point temperature.

[0056]

[0057] Furthermore, based on daily runoff data observed at hydrological stations and daily precipitation, daily maximum temperature, and daily minimum temperature series from the ERA5 dataset, four hydrological models (including XAJ, GR4J, HBV, and SIMHYD) were driven to obtain preliminary simulated runoff. Among them, XAJ is the Xin'anjiang model, GR4J is a conceptual hydrological model developed by the French National Institute of Hydrometeorology and Hydrology (SHMI), mainly used for runoff simulation in small and medium-sized watersheds, HBV is a distributed hydrological model developed by the Royal Institute of Technology in Sweden (KTH), which can be used to simulate hydrological processes in watersheds of different scales, and SIMHYD is a conceptual hydrological model developed by the Commonwealth Scientific and Industrial Research Organisation (CSIRO) of Australia.

[0058] Furthermore, statistical analysis is conducted on the daily runoff processes in the preliminary simulated runoff results and the measured daily runoff processes to determine the lag time affecting the measured daily runoff. An appropriate correlation threshold is selected to determine the simulated runoff duration for establishing a machine learning model with the measured runoff.

[0059] Figure 2 This is a schematic diagram illustrating the variation of the correlation coefficient between measured daily runoff and simulated runoff at different time lags, according to a specific embodiment of the present invention. Figure 2 As shown, the correlation coefficient between simulated runoff and measured runoff generally decreases gradually with the extension of the lag time. A suitable correlation threshold is selected, for example, 0.5 is determined as the simulated runoff duration of the machine learning model.

[0060] Furthermore, a Long Short-Term Memory (Attention-LSTM) neural network model with coupled attention mechanism is used to correct the preliminary simulated runoff results in order to quantitatively evaluate the coefficient T. sThe highest objective function is used to calibrate the fitting model, thereby establishing a hydrology-machine learning coupled model.

[0061] Figure 3 This is a schematic diagram of a hydrology-machine learning coupled model structure provided according to a specific embodiment of the present invention, such as... Figure 3 As shown, a Long Short-Term Memory (LSTM) neural network model with a three-layer neural network architecture is constructed to generalize the water storage and regulation effects of dams, reservoirs, or water diversion projects on the watershed and improve the accuracy of hydrological simulation. In this embodiment, the neural network interval simulation mean method is used to run the neural network model independently multiple times and take the average value as the final simulation result to reduce uncertainty.

[0062] Furthermore, such as Figure 4 As shown, Figure 4 This diagram illustrates the structure of a memory unit in a Long Short-Term Memory (LSTM) neural network model according to a specific embodiment of the present invention. To address the gradient explosion and vanishing problems caused by Nonlinear Autoregressive External Input Pattern (NARX) dynamic neural networks during deep learning (with ≥2 hidden layers), the LSTM network introduces storage units—namely, input gates, forget gates, internal feedback connections, and output gates—into the hidden layers of the NARX neural network. This allows for selective memorization of current information or forgetting of past information (such as rainfall-runoff mapping), thereby enhancing the long-term memory capacity of the NARX neural network. In short, the LSTM network replaces each hidden layer in the NARX dynamic neural network with a storage unit possessing memory functionality, referred to as an LSTM unit. The input and output layers are identical to those in the NARX dynamic neural network.

[0063] Furthermore, to better focus on key meteorological variables affecting runoff simulation and improve simulation accuracy, this embodiment of the invention adds an Attention layer to the LSTM to enhance the LSTM's contextual modeling capability when processing time series data. Figure 5 This is a schematic diagram of the structure of a Convolutional Attention Module (CBAM) according to a specific embodiment of the present invention, as shown below. Figure 5 As shown, the Convolutional Attention Module (CBAM) of this invention can dynamically allocate attention according to different time steps of the input, thereby better capturing long-term dependencies and key information. CBAM is a modular attention mechanism, including channel attention and spatial attention. For the output of the LSTM, the Attention layer first calculates channel attention. The role of the channel attention module is to assign a weight to the features at each time step according to the importance of each channel. The formulas for average pooling and max pooling are:

[0064] M avg =AvgPool(Ht );

[0065] M max =MaxPool(H t );

[0066] Where AvgPool and MaxPool are global average pooling and global max pooling, respectively, H t It is the output feature (i.e., hidden state) of LSTM at time t.

[0067] Furthermore, the two pooling results are concatenated and passed to a shared fully connected layer to obtain the channel attention weights:

[0068] M channel =σ(W ch [M avg M max ]+b ch ):

[0069] Among them, W ch and b ch The weights and biases are represented respectively, and σ is the sigmoid activation function.

[0070] Furthermore, for the features at each time step, the spatial attention module enhances the importance of the model's focus by weighting each spatial location. After performing average pooling and max pooling along the channel dimension, the results are concatenated and a convolutional operation is used to generate the spatial attention map. The output of the convolutional layer generates spatial attention weights using a sigmoid activation function.

[0071] M spatial =σ(Conv([S avg ,S max ]));

[0072] Among them, S avg To perform average pooling results along the channel dimension, S max To perform max pooling in the channel dimension, Conv represents the convolution operation, and σ is the sigmoid activation function.

[0073] Furthermore, channel attention and spatial attention are applied to the hidden state sequence of the LSTM, respectively. Meteorological data such as daily precipitation, daily average temperature, daily maximum temperature, daily minimum temperature, specific humidity, relative humidity, wind speed, shortwave radiation, and longwave radiation obtained from ERA products, along with simulated and measured runoff series, are used as inputs. After calibrating the LSTM model, an attention layer is added to further correct the equations of the simulated runoff series:

[0074] R cor (t)=FAttention-LSTM [QM(t),QM(t-1),QM(t-2),…,QM(tN)];

[0075] Among them, R cor (t) represents the corrected runoff at time t, QM(t) represents the input variables for calibrating the Attention-LSTM model, including the daily runoff series simulated by four hydrological models and the watershed average meteorological data derived by ERA5; QM(t-1) represents the simulated runoff and meteorological series at time t-1, and N represents the lag time determined by the Attention-LSTM model; F Attention-LSTM This represents the Attention-LSTM model.

[0076] Furthermore, the Attention-LSTM model is trained using minimum batch gradient descent to optimize its parameters, thereby quantitatively evaluating the coefficient T. s The highest is the target rate-fit model:

[0077]

[0078] Where KGE() is the Kling-Gupta efficiency coefficient, Q obs and Q sim This represents daily runoff data from both simulated and measured series. and This represents the monthly flood duration data for both simulated and measured series. and This represents the monthly-scale flood discharge data corresponding to the simulated and measured series; α, β, and γ represent their respective weighting parameters, satisfying α + β + γ = 1. T s The value range is (-∞, 1], when T s When = 1, it means that the simulation series and the measured series are in perfect agreement.

[0079] Therefore, this embodiment of the invention establishes a hydrology-machine learning coupled model by coupling a hydrological model and a machine learning model, denoted as HLM.

[0080] In step S103, the watershed hydrological process under future scenarios is simulated using a hydrological-machine learning coupled model to obtain the simulation results of the watershed hydrological process under future scenarios. Based on the simulation results of the watershed hydrological process, the flood duration and flood volume feature values ​​under multiple warming level time windows are extracted using a preset optimal model. Based on the flood duration and flood volume feature values ​​under multiple warming level time windows, a two-dimensional flood risk prediction result is obtained.

[0081] Among them, the hydrology-machine learning coupling model is a machine learning model of hydrological processes. The future scenario is a hypothesis or simulation of the climate change, socio-economic development and other conditions that the basin may face in the future period after the node, based on current research and analysis, with a preset time point as the node. Flood duration and flood volume are two important characteristic quantities in the flood process, and their marginal distribution functions are used to describe the probability distribution of their respective variables.

[0082] Furthermore, in some embodiments, the watershed hydrological processes under future scenarios are simulated using a hydrological-machine learning coupled model to obtain simulation results of the watershed hydrological processes under future scenarios, including: simulating the watershed hydrological processes under future scenarios using a hydrological-machine learning coupled model based on the corrected meteorological simulation series results to obtain a series of daily runoff results under future climate change scenarios; and obtaining simulation results of the watershed hydrological processes under future scenarios based on the series of daily runoff results under future climate change scenarios.

[0083] Specifically, the corrected meteorological simulation series results are input into the hydrology-machine learning coupled model to simulate the daily runoff series under future climate change scenarios, and obtain the simulation results of watershed hydrological processes under future scenarios.

[0084] Furthermore, in some embodiments, before extracting flood duration and flood volume characteristics under multiple warming level time windows using a preset optimal model based on the results of watershed hydrological process simulation, the method further includes: calculating time windows under different warming levels based on temperature data in the corrected meteorological simulation series results; fusing the corrected meteorological simulation series results with a machine learning model to obtain a fused machine learning model; determining the model parameters of the fused machine learning model based on a preset adaptive model selection mechanism, and obtaining a preset optimal model based on the determined results.

[0085] Specifically, taking 1985-2014 as the historical period and 2015-2100 as the future period, the temperature rise relative to pre-industrial levels in a future year is defined as:

[0086]

[0087] in, Let T be the global average temperature for future year t, which is the ensemble average of the M group of global climate models. his ΔT represents the global average temperature over a historical period. his This represents the temperature rise relative to pre-industrial levels during historical periods.

[0088] Furthermore, ΔT(t) can be calculated. 1.5℃ ) = 1.5℃, ΔT(t) 2.0℃ ) = 2.0℃ and ΔT(t)3.0℃ The year t when ) = 3.0℃ 1.5℃ t 2.0℃ and t 3.0℃ And using this year as the midpoint, we obtain 30-year time windows under different warming levels.

[0089] Furthermore, the meteorological simulation series under the M-group climate change scenarios are combined with machine learning models (GCM-HLM). The optimal model is dynamically selected based on an adaptive model selection mechanism. Considering the combination of the M-group meteorological simulation series and four HLM models, for each simulated watershed hydrological process, the baseflow separation method is used to extract the duration and flood volume of measured runoff and the simulated runoff for the corresponding period. The error metric formula for both is defined as follows:

[0090]

[0091] DTW stands for Dynamic Time Warping, used to calculate the similarity between time series. obs (t) and S represents the duration of measured runoff and simulated runoff output by the i-th hydrological machine learning model under the k-th global climate model, respectively. obs (t) and represents the measured runoff and the simulated runoff output by the i-th hydrological machine learning model under the k-th global climate model, respectively, where n is the time step.

[0092] Furthermore, considering both duration error and flood volume error, the overall error evaluation function is obtained as follows:

[0093]

[0094] Here, α is a weighting parameter used to adjust the relative importance of duration error and flood volume error.

[0095] Furthermore, the optimal model is adaptively selected based on minimizing error:

[0096]

[0097] Furthermore, in some embodiments, based on the flood duration and flood volume characteristic values ​​under multiple warming level time windows, a two-dimensional flood risk prediction result is obtained, including: constructing a marginal distribution function of flood duration and flood volume based on the flood duration and flood volume characteristic values ​​under multiple warming level time windows, and constructing a Copula joint distribution function of flood duration and flood volume under inconsistent conditions; calculating the joint return period of flood duration and flood volume under different warming level time windows based on the marginal distribution function of flood duration and flood volume and the Copula joint distribution function of flood duration and flood volume under inconsistent conditions, and obtaining a two-dimensional flood risk prediction result.

[0098] Among them, the Copula joint distribution function of flood duration and flood volume under non-uniform conditions is the combined distribution function that combines the non-uniform marginal distribution functions of flood duration and flood volume. The joint return period of flood duration and flood volume is the combination of the joint probability density function with the largest value on the return period contour line. The two-dimensional flood risk prediction results include the socio-economic risk prediction results of the increase in future flood risk under various warming levels.

[0099] Specifically, the 95th percentile of daily runoff is used to identify the annual maximum flood event; the duration and volume of this flood event are calculated, and let X represent the flood characteristics (duration D and volume S). In this invention, the gamma distribution, generalized extreme value distribution, inverse Gaussian distribution, and Weiber distribution function are used as candidate marginal distributions for flood duration and volume, and the Kolmogorov-Smirnov (KS) statistic is used for goodness-of-fit testing.

[0100]

[0101] Among them, F n (x) represents the observed sequence value, F n (x) represents the theoretical sequence value.

[0102] Furthermore, the marginal distribution function is optimized based on the AIC criterion:

[0103] AIC = 2k - 2ln(L);

[0104] Where k is the number of model parameters and L is the likelihood function.

[0105] Furthermore, in this embodiment of the invention, Gumbel copula, Frank copula, Clayton copula, t-copula, and Gaussian copula are used as candidate functions for the joint distribution of flood duration and flood volume. After using the KS goodness-of-fit test, a Copula joint distribution function for flood duration and flood volume under inconsistent conditions is constructed according to the AIC criterion. Based on the definition of the Copula function, the inconsistent two-variable Copula function can be expressed as:

[0106]

[0107] Among them, F t (d t ,s t ) represents the time-varying joint distribution function of D and S; and Let represent the time-varying marginal distribution function and time-varying parameter of variables D and S, respectively.

[0108] Furthermore, the OR return period is used as a metric for floods, and it is defined as follows:

[0109]

[0110] in, The time-varying OR joint return period is in years.

[0111] Furthermore, the most probable combination of flood duration and flood volume refers to the combination that maximizes the joint probability density function on the return period isoline (d). * (t),s * (t) can be solved by constructing the following equation:

[0112]

[0113] Among them, f t (d t ,s t ) represents the density function of the time-varying joint distribution function of duration D and flood volume S; The density function representing the time-varying Copula joint distribution function; and They represent and The density function of the marginal distribution function.

[0114] Therefore, given the joint return period T or Using the above method, the historical period (1985-2014) in T... or By determining the most probable combination of flood duration and volume for each year under the given conditions, we obtain the characteristic values ​​of flood duration and volume for historical periods.

[0115] Furthermore, the obtained historical flood duration and flood volume characteristics are substituted into the most probable combination model for different future warming periods to calculate the new return period T of the 30-year window under different warming levels. f (ΔT), where T is the recurrence period of a given historical period. h The future socioeconomic risks corresponding to global warming ΔT can be measured by the following formula:

[0116]

[0117] Among them, E pop E GDP and E ALD POPs respectively represent the risk to the population, GDP, and farmland affected by floods. k GDP k and ALD kLet be the population, GDP, and farmland area in year k within a 30-year window corresponding to global warming ΔT; I(·) is the indicator function, and T... h -T f When (ΔT)>0, it is recorded as 1, indicating an increased risk of two-dimensional flooding; conversely, it is recorded as 0, indicating a decreased risk. N1 and N2 represent the start and end years of the study period, respectively.

[0118] Furthermore, the comprehensive socio-economic risks under different warming levels are assessed and predicted. Based on the socio-economic risk analysis results at different warming levels, the comprehensive socio-economic risks under the warming level corresponding to the global temperature rise ΔT are as follows:

[0119]

[0120] SocioRisk represents comprehensive socioeconomic risk; E POP (max), E GDP (max) and E ALD (max) represents the maximum values ​​of the population, GDP, and farmland risk affected by floods obtained from M global climate models, respectively.

[0121] Furthermore, in some embodiments, before simulating the watershed hydrological process under future scenarios using a hydrology-machine learning coupled model to obtain the simulation results of the watershed hydrological process under future scenarios, the method further includes: correcting multiple global climate model simulation data based on a decoding-encoding bias correction method to obtain a series of corrected meteorological simulation results.

[0122] Specifically, the monthly observation data of the ERA5 reanalysis dataset from 1985 to 2014 are denoted as follows: The simulated data of GCMs for each month in the historical period are denoted as follows: Let the simulated data of GCMs for each month in a given future period be denoted as Derivation separately and The interannual trend line is obtained by decoding the trend term caused by human climate forcing; then the trend line is shifted so that the sum of the trend terms for each year is 0; for series where the interannual trend is not significant at the 0.05 confidence level, the trend terms for each year are reset to 0; finally, the trend terms for each series are removed to obtain three new series. and This refers to a data series caused by internal climate variability.

[0123] Furthermore, the historical periods under different quantiles were calculated. and situation period The deviation was determined, and a pseudo-observation series for the scenario period was obtained based on the actual observations.

[0124]

[0125] Where p represents different quantiles, taking 100 equidistant quantiles in the range of 0.01 to 0.99.

[0126] Furthermore, based on Scenario simulation series The frequency distribution curve is used to correct the deviation, and the corrected future series is obtained based on the quantile mapping. Finally, the trend term subtracted in the first step is superimposed by the encoding method to obtain the corrected future scenario.

[0127] Therefore, the trend of each month is preserved and the bias correction method is applied on the daily scale. Considering that the frequency distribution function of each month may have some differences, the data of each GCM in the 12 months are corrected separately, and the daily series of meteorological data of M GCMs under three scenarios (SSP245, SSP370 and SSP585) in the historical period and future period are obtained. Finally, the corrected meteorological simulation series results are obtained.

[0128] To enable those skilled in the art to better understand the two-dimensional flood risk prediction method of the present invention, the two-dimensional flood risk prediction method will be explained and described below in conjunction with specific embodiments.

[0129] Figure 6 A flowchart of a two-dimensional flood risk prediction method according to a specific embodiment of the present invention.

[0130] like Figure 6 As shown, this embodiment of the invention first collects basic meteorological and hydrological data of the watershed, then derives relative humidity and specific humidity, calibrates the watershed hydrological model and machine learning model, and then obtains a series of meteorological simulations under M climate change scenarios based on a decode-encoder bias correction method, and drives a hydrological-machine learning coupled model with an attention mechanism to simulate the watershed hydrological process under future scenarios; based on the corrected global climate model set, it derives time windows under different warming levels, and dynamically selects the best model according to an adaptive model selection mechanism; based on the simulated hydrological process and the annual maximum sampling method, it extracts flood duration and flood volume feature values ​​under different warming level time windows, and establishes a joint probability distribution function based on Copula under inconsistent conditions; based on the most likely combination scenario, it calculates the joint return period of flood duration and flood volume under inconsistent conditions, and assesses the impact of climate change and underlying human activities on the future flood situation of the watershed; finally, based on a shared socioeconomic path dataset, it derives the comprehensive socioeconomic risk of increased future flood risk.

[0131] The two-dimensional flood risk prediction method provided by the present invention first collects basic meteorological and hydrological data of the watershed, then derives humidity-related data, calibrates the hydrological model, and constructs a machine learning model; then, it obtains a series of meteorological simulations under climate change scenarios through decoding-encoding bias correction, driving the model to simulate future hydrological processes; based on the optimal model, it extracts flood feature values ​​and constructs a joint probability distribution function; finally, it obtains the two-dimensional flood risk prediction result, solving the problems of low accuracy and poor applicability of flood risk prediction in related technologies, and improving the accuracy and applicability of flood risk prediction.

[0132] Next, the two-dimensional flood risk prediction device according to an embodiment of the present invention is described with reference to the accompanying drawings.

[0133] Figure 7 This is a block diagram of a two-dimensional flood risk prediction device provided according to an embodiment of the present invention.

[0134] like Figure 7 As shown, the two-dimensional flood risk prediction device 10 includes: an acquisition module 100, a mathematical modeling module 200, and a prediction module 300.

[0135] The acquisition module 100 is used to acquire multimodal data within the target watershed; the mathematical modeling module 200 is used to obtain simulated runoff results based on multiple pre-established hydrological models using the multimodal data within the target watershed, and to establish a hydrological-machine learning coupled model based on the simulated runoff results; the prediction module 300 is used to simulate the watershed hydrological process under future scenarios using the hydrological-machine learning coupled model, to obtain the simulation results of the watershed hydrological process under future scenarios, and to extract flood duration and flood volume feature values ​​under multiple warming level time windows using a preset optimal model based on the watershed hydrological process simulation results, and to obtain two-dimensional flood risk prediction results based on the flood duration and flood volume feature values ​​under multiple warming level time windows.

[0136] Furthermore, in some embodiments, the mathematical modeling module 200 is specifically used to: calculate relative humidity and specific humidity results based on meteorological data from the ERA5 dataset; and obtain simulated runoff results based on daily flow series data from the target watershed control hydrological station and meteorological data from the ERA5 dataset through multiple preset hydrological models; determine the simulated runoff duration of the machine learning model based on the simulated runoff results and pre-collected daily measured runoff results, according to a preset correlation threshold; and correct the simulated runoff results based on a long short-term memory neural network with a coupled attention mechanism to obtain a hydrological-machine learning coupled model.

[0137] Furthermore, in some embodiments, before simulating the watershed hydrological process under future scenarios using a hydrological-machine learning coupled model to obtain the simulation results of the watershed hydrological process under future scenarios, the prediction module 300 is also used to: correct multiple global climate model simulation data based on a decoding-encoding bias correction method to obtain a series of corrected meteorological simulation results.

[0138] Furthermore, in some embodiments, the prediction module 300 is also used to: simulate the watershed hydrological process under future scenarios based on the corrected meteorological simulation series results, and obtain the daily runoff series results under future climate change scenarios; and obtain the watershed hydrological process simulation results under future scenarios based on the daily runoff series results under future climate change scenarios.

[0139] Furthermore, in some embodiments, before extracting flood duration and flood volume feature values ​​under multiple warming level time windows using a preset optimal model based on the results of watershed hydrological process simulation, the prediction module 300 is also used to: calculate time windows under different warming levels based on temperature data in the corrected meteorological simulation series results; fuse the corrected meteorological simulation series results with the machine learning model to obtain a fused machine learning model; determine the model parameters of the fused machine learning model based on a preset adaptive model selection mechanism, and obtain a preset optimal model based on the determination result.

[0140] Furthermore, in some embodiments, the prediction module 300 is also used to: construct marginal distribution functions of flood duration and flood volume based on flood duration and flood volume characteristic values ​​under multiple warming level time windows, and construct a Copula joint distribution function of flood duration and flood volume under inconsistent conditions; calculate the joint return period of flood duration and flood volume under different warming level time windows based on the marginal distribution function of flood duration and flood volume and the Copula joint distribution function of flood duration and flood volume under inconsistent conditions, and obtain two-dimensional flood risk prediction results.

[0141] It should be noted that the foregoing explanation of the two-dimensional flood risk prediction method embodiment also applies to the two-dimensional flood risk prediction device of this embodiment, and will not be repeated here.

[0142] The two-dimensional flood risk prediction device provided in this embodiment of the invention first collects basic meteorological and hydrological data of the watershed, then derives humidity-related data, calibrates the hydrological model, and constructs a machine learning model; then, it obtains a series of meteorological simulations under climate change scenarios through decoding-encoding bias correction, driving the model to simulate future hydrological processes; based on the optimal model, it extracts flood feature values ​​and constructs a joint probability distribution function; finally, it obtains the two-dimensional flood risk prediction result, solving the problems of low accuracy and poor applicability of flood risk prediction in related technologies, and improving the accuracy and applicability of flood risk prediction.

[0143] Figure 8 This is a schematic diagram of an electronic device provided according to an embodiment of the present invention. The electronic device may include:

[0144] The memory 801, the processor 802, and the computer program stored on the memory 801 and capable of running on the processor 802.

[0145] When the processor 802 executes the program, it implements the two-dimensional flood risk prediction method provided in the above embodiments.

[0146] Furthermore, the electronic device also includes:

[0147] Communication interface 803 is used for communication between memory 801 and processor 802.

[0148] The memory 801 is used to store computer programs that can run on the processor 802.

[0149] The memory 801 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0150] If the memory 801, processor 802, and communication interface 803 are implemented independently, then the communication interface 803, memory 801, and processor 802 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be divided into address buses, data buses, control buses, etc. For ease of representation, Figure 8 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0151] Optionally, in a specific implementation, if the memory 801, processor 802, and communication interface 803 are integrated on a single chip, then the memory 801, processor 802, and communication interface 803 can communicate with each other through an internal interface.

[0152] The processor 802 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention.

[0153] In addition, embodiments of the present invention also provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the two-dimensional flood risk prediction method as described in the above embodiments.

[0154] In addition, embodiments of the present invention also provide a computer program product, including a computer program that is executed to implement the two-dimensional flood risk prediction method as described in the above embodiments.

[0155] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0156] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0157] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of the invention pertain.

[0158] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0159] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it includes one or a combination of the steps of the method embodiments.

Claims

1. A two-dimensional flood risk prediction method, characterized in that, Includes the following steps: Acquire multimodal data within the target watershed; Based on multimodal data within the target watershed, simulated runoff results are obtained using multiple pre-established hydrological models, and a hydrological-machine learning coupled model is established based on the simulated runoff results. The hydrological-machine learning coupled model is used to simulate the watershed hydrological process under future scenarios, obtaining simulation results. Based on these simulation results, a preset optimal model is used to extract flood duration and volume feature values ​​under multiple warming level time windows. Based on these flood duration and volume feature values ​​under the multiple warming level time windows, a two-dimensional flood risk prediction result is obtained. Before simulating the watershed hydrological processes under future scenarios using the hydrology-machine learning coupled model to obtain the simulation results, the method further includes: using a decode-encode bias correction method to correct multiple global climate model simulation data to obtain a series of corrected meteorological simulation results. Before extracting flood duration and flood volume characteristics under multiple warming level time windows using a preset optimal model based on the simulation results of the watershed hydrological process, the method further includes: calculating time windows under different warming levels based on temperature data in the corrected meteorological simulation series results; fusing the corrected meteorological simulation series results with the hydrological-machine learning coupled model to obtain a fused hydrological-machine learning coupled model; determining the model parameters of the fused hydrological-machine learning coupled model based on a preset adaptive model selection mechanism, and obtaining the preset optimal model based on the determination results.

2. The two-dimensional flood risk prediction method according to claim 1, characterized in that, The multimodal data includes simulation data from multiple global climate models within the target watershed, meteorological data from the ERA5 dataset, socioeconomic data, and daily flow series data from control hydrological stations in the target watershed. The process involves obtaining simulated runoff results based on multiple pre-established hydrological models using the multimodal data within the target watershed, and establishing a hydrological-machine learning coupled model based on the simulated runoff results, including: The relative humidity and specific humidity results are calculated based on the meteorological data of the ERA5 dataset; and the simulated runoff results are obtained based on the daily flow series data of the target watershed control hydrological station and the meteorological data of the ERA5 dataset through multiple preset hydrological models. Based on the simulated runoff results and the pre-collected daily measured runoff results, the simulated runoff duration of the machine learning model is determined according to a preset correlation threshold. A long short-term memory neural network based on a coupled attention mechanism is used to correct the simulated runoff results, resulting in the hydrology-machine learning coupled model.

3. The two-dimensional flood risk prediction method according to claim 1, characterized in that, The simulation of watershed hydrological processes under future scenarios using the hydrology-machine learning coupled model yields simulation results of watershed hydrological processes under future scenarios, including: Based on the corrected meteorological simulation results, the watershed hydrological processes under future scenarios are simulated using the hydrology-machine learning coupled model to obtain a series of daily runoff results under future climate change scenarios. Based on the daily runoff series results under the future climate change scenario, the simulation results of watershed hydrological processes under the future scenario are obtained.

4. The two-dimensional flood risk prediction method according to claim 1, characterized in that, The two-dimensional flood risk prediction results, obtained based on the flood duration and flood volume characteristics under the multiple warming level time windows, include: Based on the flood duration and flood volume characteristic values ​​under the multiple warming level time windows, the marginal distribution functions of flood duration and flood volume are constructed, and the Copula joint distribution function of flood duration and flood volume under non-uniform conditions is constructed. Based on the marginal distribution function of flood duration and flood volume and the Copula joint distribution function of flood duration and flood volume under the non-uniform conditions, the joint return period of flood duration and flood volume under different warming level time windows is calculated to obtain the two-dimensional flood risk prediction results.

5. A two-dimensional flood risk prediction device, characterized in that, The device includes: The acquisition module is used to acquire multimodal data within the target watershed. The mathematical modeling module is used to obtain simulated runoff results based on multiple pre-established hydrological models using multimodal data within the target watershed, and to establish a hydrological-machine learning coupled model based on the simulated runoff results. The prediction module is used to simulate the watershed hydrological processes under future scenarios using the hydrology-machine learning coupled model, obtain simulation results of the watershed hydrological processes under future scenarios, and, based on the simulation results, extract flood duration and flood volume feature values ​​under multiple warming level time windows using a preset optimal model. Based on the flood duration and flood volume feature values ​​under the multiple warming level time windows, a two-dimensional flood risk prediction result is obtained. Before simulating the watershed hydrological processes under future scenarios using the hydrology-machine learning coupled model to obtain the simulation results, the prediction module is further used to: correct multiple global climate model simulation data based on a decode-encoder bias correction method to obtain a series of corrected meteorological simulation results, wherein... Before extracting flood duration and flood volume characteristics under multiple warming level time windows based on the simulation results of the watershed hydrological process using a preset optimal model, the prediction module is further configured to: calculate time windows under different warming levels based on temperature data in the corrected meteorological simulation series results; fuse the corrected meteorological simulation series results with the hydrological-machine learning coupled model to obtain a fused hydrological-machine learning coupled model; determine the model parameters of the fused hydrological-machine learning coupled model based on a preset adaptive model selection mechanism, and obtain the preset optimal model based on the determination results.

6. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the two-dimensional flood risk prediction method as described in any one of claims 1-4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the two-dimensional flood risk prediction method as described in any one of claims 1-4.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the two-dimensional flood risk prediction method as described in any one of claims 1-4.

Images