A flood risk prediction method, system and electronic device

Abstract

The application discloses a flood risk prediction method, system and electronic equipment, and the method comprises the following steps: collecting meteorological hydrological data sets and training a plurality of basin hydrological models; a machine learning model is established based on the plurality of basin hydrological models and a gated recurrent unit neural network model; the machine learning model is trained to obtain basin hydrological process data under a future scenario; a plurality of global climate models are combined with the machine learning model, and the weight parameters of each combined scenario are determined; the annual average underlying surface feature parameters of the basin are obtained, a joint probability distribution function is established, and the flood risk is predicted based on the joint probability distribution function and the weight parameters of each combined scenario. The application combines a plurality of methods with basin flood, and provides important and operable reference for basin flood risk prediction.

Description

Technical Field

[0001] This invention relates to the field of flood risk prediction technology, and in particular to a flood risk prediction method, system and electronic equipment. Background Technology

[0002] Since the Industrial Revolution, emissions of greenhouse gases such as carbon dioxide in the atmosphere have increased, altering the energy balance and material cycling processes of the Earth's land and atmosphere. This has led to rising global temperatures, the large-scale disappearance of coral reefs and tropical rainforests, and frequent extreme disasters such as torrential rains and floods. To address these issues, it is crucial to strengthen scientific research on the prediction of flood risk evolution under climate change.

[0003] Currently, traditional flood risk prediction methods mainly rely on historical flood data, meteorological data, and hydrological models, which can provide some reference for flood warning and flood control decision-making. However, due to the impact of climate change and human activities, the frequency and intensity of floods are exhibiting new characteristics, challenging the applicability and accuracy of traditional methods in flood risk prediction. Current research typically uses the arithmetic mean of a global climate model ensemble to represent simulation results, without considering the impact of the uncertainties inherent in the global climate model ensemble itself. Furthermore, existing literature on predicting future flood risk generally focuses on single attributes such as flood peak, flood volume, or flood duration, failing to reflect the multivariate characteristics of floods. Therefore, there is an urgent need to develop a flood risk prediction method with high accuracy, adaptability, and ease of implementation to address the increasingly severe challenges posed by flood disasters. Summary of the Invention

[0004] This invention aims to at least partially address one of the technical problems in related technologies. Therefore, a first objective of this invention is to provide a flood risk prediction method that overcomes the deficiency in existing two-dimensional flood prediction methods that do not consider the impact of uncertainties in the global climate model ensemble itself on the prediction results.

[0005] The second objective of this invention is to provide a flood risk prediction system.

[0006] A third objective of this invention is to provide an electronic device.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0008] In a first aspect, the present invention provides a flood risk prediction method, comprising:

[0009] Step S1: Collect meteorological and hydrological data to form a meteorological and hydrological dataset;

[0010] Step S2: Determine the relative humidity and specific humidity at a set time, and train multiple watershed hydrological models based on the relative humidity, specific humidity and meteorological and hydrological datasets, and establish a machine learning model that considers the hydrological process based on the gated recurrent unit neural network model and the trained multiple watershed hydrological models.

[0011] Step S3: Obtain meteorological simulation data from the global climate model ensemble under the climate change scenario and correct the meteorological simulation data from the global climate model ensemble, and train the machine learning model using the corrected meteorological simulation data from the global climate model ensemble to obtain watershed hydrological process data under the future scenario.

[0012] Step S4: Combine multiple global climate models in the meteorological simulation data of the global climate model set with the machine learning model, and determine the weight parameters of each combined scenario in the multi-model weighted average method;

[0013] Step S5: Establish a watershed hydrothermal coupling equilibrium equation based on the watershed hydrological process data to obtain the watershed's annual average underlying surface characteristic parameters;

[0014] Step S6: Obtain the flood duration and annual maximum flood peak discharge from the watershed hydrological process data, and establish a joint probability distribution function based on the flood duration, the annual maximum flood peak discharge and the watershed annual average underlying surface characteristic parameters, and perform flood risk prediction based on the joint probability distribution function and the weight parameters of each combined scenario.

[0015] Secondly, the present invention provides a flood risk prediction system, comprising:

[0016] The data acquisition module is used to collect meteorological and hydrological data to form a meteorological and hydrological dataset.

[0017] The training module is used to determine the relative humidity and specific humidity at a set time, and to train multiple watershed hydrological models based on the relative humidity, specific humidity and the meteorological and hydrological dataset, and to establish a machine learning model that considers the hydrological process based on the gated recurrent unit neural network model and the trained multiple watershed hydrological models.

[0018] The calibration module is used to acquire meteorological simulation data of the global climate model set under the climate change scenario and calibrate the meteorological simulation data of the global climate model set, and use the calibrated meteorological simulation data of the global climate model set to train the machine learning model to obtain watershed hydrological process data under the future scenario.

[0019] The combination module is used to combine multiple global climate models in the meteorological simulation data of the global climate model set with the machine learning model, and to determine the weight parameters of each combined scenario in the multi-model weighted average method.

[0020] A module is established to establish a watershed hydrothermal coupling equilibrium equation based on the watershed hydrological process data, so as to obtain the watershed's annual average underlying surface characteristic parameters.

[0021] The prediction module is used to acquire flood duration and annual maximum flood peak discharge from the watershed hydrological process data, establish a joint probability distribution function based on the flood duration, the annual maximum flood peak discharge and the watershed's annual average underlying surface characteristic parameters, and perform flood risk prediction based on the joint probability distribution function and the weight parameters of each combined scenario.

[0022] Thirdly, the electronic device provided by the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it can implement the flood risk prediction method described above.

[0023] The flood risk prediction method, system, and electronic equipment provided by this invention combine Earth system models, water-heat balance equations, watershed hydrological models, machine learning models, most probable combined scenario methods, multi-model weighted average methods, and emergent constraint methods to provide important and highly operable reference for watershed flood risk assessment and early warning under climate change scenarios, and provide engineering reference value for responding to future climate disasters and scientifically formulating emission reduction strategies.

[0024] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the flood risk prediction method provided in this embodiment of the invention.

[0026] Figure 2 This is a flowchart illustrating the flood risk prediction method provided in a specific example of the present invention.

[0027] Figure 3 This is a schematic diagram illustrating the changes in the correlation coefficient between daily measured runoff and simulated runoff under different time delays, provided in an embodiment of the present invention.

[0028] Figure 4 This is a schematic diagram of the hydrothermal balance equation provided in an embodiment of the present invention.

[0029] Figure 5 This is a schematic diagram of the flood risk prediction system provided in an embodiment of the present invention.

[0030] Figure 6 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0031] The following describes this embodiment in detail. Examples of the embodiments are shown 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 invention, and should not be construed as limiting the invention.

[0032] Figure 1 This is a schematic flowchart of the flood risk prediction method provided in an embodiment of the present invention. Figure 1 As shown, the method includes:

[0033] Step S1: Collect meteorological and hydrological data to form a meteorological and hydrological dataset;

[0034] Step S2: Determine the relative humidity and specific humidity at a set time, and train multiple watershed hydrological models based on the relative humidity, specific humidity and meteorological and hydrological datasets, and establish a machine learning model that considers the hydrological process based on the gated recurrent unit neural network model and the trained multiple watershed hydrological models.

[0035] Step S3: Obtain meteorological simulation data from the global climate model ensemble under the climate change scenario and correct the meteorological simulation data from the global climate model ensemble, and train the machine learning model using the corrected meteorological simulation data from the global climate model ensemble to obtain watershed hydrological process data under the future scenario.

[0036] Step S4: Combine multiple global climate models in the meteorological simulation data of the global climate model set with the machine learning model, and determine the weight parameters of each combined scenario in the multi-model weighted average method;

[0037] Step S5: Establish a watershed hydrothermal coupling equilibrium equation based on the watershed hydrological process data to obtain the watershed's annual average underlying surface characteristic parameters;

[0038] Step S6: Obtain the flood duration and annual maximum flood peak discharge from the watershed hydrological process data, and establish a joint probability distribution function based on the flood duration, the annual maximum flood peak discharge and the watershed annual average underlying surface characteristic parameters, and perform flood risk prediction based on the joint probability distribution function and the weight parameters of each combined scenario.

[0039] Specifically, meteorological and hydrological datasets were first collected, including meteorological data from the European Centre for Medium-Range Weather Forecasts' Generation 5 Atmospheric Reanalysis Dataset (ERA5), meteorological data from three shared socioeconomic pathway Earth system models, dynamic population and GDP data from shared socioeconomic pathway datasets, and daily flow series from watershed control hydrological stations.

[0040] Then, relative humidity and specific humidity at the set time were derived. The ERA5 reanalysis dataset was used to drive five watershed hydrological models and machine learning models, thereby establishing a machine learning model that considers hydrological processes. Based on the quantile bias correction method, a series of global climate models and meteorological simulations under climate change scenarios were obtained. The corrected simulation series was used to drive a machine learning model that considers hydrological processes, thus obtaining watershed hydrological process data under future scenarios.

[0041] Furthermore, based on the combined scenarios of global climate model sets and machine learning models that consider hydrological processes, the weight parameters of each combined scenario in the multi-model weighted average method are derived; based on the simulated hydrological series, a watershed hydrothermal coupling equilibrium equation is established to obtain the watershed's annual average underlying surface characteristic parameters.

[0042] By sampling the annual maximum values ​​of simulated hydrological series, flood duration and annual maximum peak discharge series are obtained. Then, a joint probability distribution function based on the Copula function (a probabilistic model) under inconsistent conditions is established. Based on the weight parameters of each combined scenario in the multi-model weighted average method, the socio-economic exposure caused by the increase in future flood risk is predicted. Finally, using the watershed hydrological processes under the predicted future scenarios, an emergent constraint relationship between dew point temperature variability and peak discharge variability is established. By introducing historical dew point observations, the uncertainty of future peak discharge prediction is reduced, thereby achieving accurate flood risk prediction.

[0043] See Figure 2 As shown, by collecting meteorological and hydrological datasets and inferring relative humidity and specific humidity as inputs, hydrological model simulation and machine learning model calibration are obtained. Quantile deviation correction is used to obtain simulated meteorological series, and the optimal fitting model is selected to obtain future hydrological simulation series. Furthermore, by inferring the weights of combined scenarios, a hydrothermal coupling equilibrium equation is established to obtain the underlying surface characteristic parameters. By sampling the annual maximum values ​​of the future hydrological simulation series, the flood duration and annual maximum flood peak flow are extracted. After verifying the correlation, these are used as inputs to construct a time-varying Copula joint distribution function, inferring the most likely combined scenarios, and assessing the future flood risk.

[0044] This invention combines Earth system models, water-heat balance equations, watershed hydrological models, machine learning models, most probable combination scenario methods, multi-model weighted average methods, and emergent constraint methods to provide important and highly operable references for watershed flood risk assessment and early warning under climate change scenarios, and provides engineering reference value for responding to future climate disasters and scientifically formulating emission reduction strategies.

[0045] Based on the above embodiments, step S1 includes:

[0046] Daily flow series were collected from watershed control hydrological stations, and meteorological data were obtained from the European Centre for Medium-Range Weather Forecasts' Generation 5 Atmospheric Reanalysis dataset (ERA5).

[0047] Meteorological data were collected from five preset global climate models in the sixth phase of the Coupled Model Intercomparison Project (CPIP).

[0048] Obtain population and GDP data from the shared socioeconomic pathway dataset.

[0049] Specifically, daily flow series from watershed control hydrological stations are collected, and meteorological data such as precipitation, air pressure, 2m air temperature, 2m dew point temperature, shortwave radiation, and longwave radiation are obtained from the ERA5 reanalysis dataset. Using the watershed as the research unit, daily flow series from watershed control hydrological stations are first collected, followed by hourly data from ERA5. 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 of the invention obtains hourly precipitation, air pressure, 2m air temperature, 2m dew point temperature, 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.

[0050] Meteorological data from five GCMs (Global Climate Models) under three shared socioeconomic pathways were collected. To predict future climate scenarios, the five latest global climate models released in the Coupled Model Intercomparison Project Phase VI (CMIP6) were used. CMIP6 uses a matrix framework based on shared socioeconomic pathways. To assess the potential socioeconomic risks of flood events, population and GDP data from three shared socioeconomic pathways—Sustainable Development (SSP1), Competitive Development (SSP3), and Conventional Development (SSP5)—were considered and combined with corresponding greenhouse gas emission scenarios. Output data from three matrix frameworks—SSP126, SSP370, and SSP585—were used. The meteorological variables selected in this embodiment are daily precipitation, daily average temperature, daily maximum temperature, daily minimum temperature, specific humidity, relative humidity, shortwave radiation, and longwave radiation data. The historical period is set as 1985-2014, and the future period as 2015-2100.

[0051] Population and GDP data from a shared socioeconomic path dataset were obtained, and the output data were processed using three matrix frameworks: SSP126, SSP370, and SSP585. This embodiment of the invention uses a population policy-based projection dataset published by the Collaborative Innovation Center for Meteorological Disaster Forecasting and Assessment of a university's School of Geography. This dataset considers the results of previous national population and economic censuses, as well as annual statistical yearbooks. Economic data was projected using the Cobb-Douglas model and the Population-Environment-Development (PED) model to project domestic socioeconomic indices from 2010 to 2100. This dataset has been widely used to assess the socioeconomic impacts of extreme hydrological events.

[0052] Based on the above embodiments, step S2 includes:

[0053] By determining the latent heat of vaporization constant, water vapor gas constant, first integral constant, and second integral constant in the Clausius-Clapeyron thermodynamic equation, the nonlinear function of saturated water vapor pressure versus air temperature is obtained.

[0054] The relative humidity is obtained from the surface air temperature and dew point temperature in ERA5, as well as the nonlinear function of the saturated water vapor pressure and air temperature.

[0055] The specific humidity is obtained based on the dew point temperature, the nonlinear function of the saturated water vapor pressure and air temperature, and the near-ground pressure in ERA5.

[0056] Based on the relative humidity, specific humidity, daily flow series of the watershed control hydrological stations, and precipitation, air pressure, surface air temperature, dew point temperature, surface downflow shortwave radiation, and surface downflow longwave radiation data in ERA5, the multiple watershed hydrological models are trained to obtain preliminary simulated daily runoff.

[0057] The preliminary simulated daily runoff is corrected using the gated recurrent unit neural network model, and an objective function is established based on the Kling-Gupta efficiency coefficient to obtain the machine learning model that considers the hydrological process.

[0058] Specifically, this embodiment of the invention, based on the collection of meteorological and hydrological data sets, derives relative humidity and specific humidity using the Clausius-Clapeyron thermodynamic equation; it employs ERA5 reanalysis data to drive five hydrological models and a machine learning model, thereby establishing a machine learning model that considers hydrological processes, specifically including:

[0059] The first step is to deduce relative humidity and specific humidity based on meteorological data from the ERA5 dataset;

[0060] The Clausius-Clapeyron thermodynamic equation can quantitatively describe the saturated water vapor pressure e. sat Nonlinear relationship with temperature T:

[0061]

[0062] Among them, T0 and e s0 is the first and second integral constants; Lv is the latent heat of vaporization constant; exp represents the exponential function; Rv is the water vapor gas constant.

[0063] Dew point temperature characterizes the temperature of air cooled to water vapor saturation under constant water vapor content and pressure. Substituting it into the Clausius-Clapeyron equation, it can measure actual water vapor pressure. The surface air temperature at 2m in ERA5 (T) is then used to measure the actual water vapor pressure. 2m ) and dew point temperature (T dew Substituting these values ​​into formula (1), we can derive the near-ground relative humidity RH = e sat (T dew ) / e sat (T 2m ), where e sat (T dew ) represents T dew The saturated vapor pressure at the following value, e sat (T 2m ) represents T 2m The saturated water vapor pressure below.

[0064] 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, as shown in the following formula:

[0065]

[0066] The second step is to use relative humidity, specific humidity, daily runoff data from hydrological stations, and precipitation (pr), air pressure (ps), and surface air temperature (T2m) from the ERA5 reanalysis dataset. 2m ), dew point temperature (T) dew The surface downflow shortwave radiation (RSDS) and surface downflow longwave radiation (RLDS) series were used to drive five hydrological models (including GR4J, HBV, IHACRES, XAJ, and SIMHYD) to obtain preliminary simulated daily runoff.

[0067] The third step involves statistical analysis of the preliminary simulated daily runoff process and the measured daily runoff process from the second step to determine the lag time affecting the measured daily runoff. A schematic diagram illustrating the changes in the correlation coefficient between the measured daily runoff and the simulated runoff at different lag times is shown below. Figure 3 As shown, the correlation coefficient between simulated runoff and measured runoff generally decreases gradually with the extension of the lag time. Further, an appropriate correlation threshold is selected to determine the duration of simulated runoff for establishing a machine learning model with measured runoff, for example, 0.5.

[0068] The fourth step uses a gated recurrent unit neural network model to correct the simulated daily runoff process obtained in the second step. The Kling-Gupta efficiency coefficient (KGE) is taken as the objective function to calibrate the fitting model, thereby establishing a machine learning model that considers the hydrological process.

[0069] By constructing a gated recurrent unit neural network model with a three-layer neural network architecture, the water storage and regulation effects of dams, reservoirs, or water diversion projects on watersheds are generalized, improving the accuracy of hydrological simulation. This embodiment of the invention uses the neural network interval simulation mean method, independently running the neural network model multiple times and taking the average value as the final simulation result to reduce uncertainty.

[0070] Furthermore, the Gated Recurrent Unit (GRU) neural network model controls the flow of information through gating units, including update gates and reset gates. The reset gate determines whether to combine past states with the current input, while the update gate controls the degree of mixing of the new state. These gate operations allow the GRU to selectively forget previous information while retaining important information. Compared to Long Short-Term Memory (LSTM) neural networks, the GRU performs better but has fewer parameters and is easier to train; its input and output layers are the same as those of LSTM neural networks.

[0071] In this embodiment of the invention, 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 data obtained from ERA products, along with simulated runoff series and measured runoff series, are used as inputs. After calibrating the GRU model, the equation for correcting the simulated runoff series using this model can be expressed as:

[0072] R cor (t)=F GRU [QM(t),QM(t-1),QM(t-2),…,QM(tN lag (3)

[0073] Among them, R cor QM(t) represents the corrected runoff at time t, and QM(t) represents the input variables of the calibrated GRU model, including the daily runoff series simulated by the hydrological model and the watershed average meteorological data derived by ERA5; QM(tN) lag ) represents tN lag Simulated runoff and measured meteorological series of hydrological models at various times, N lag The time delay is determined by the GRU model; FGRU represents the GRU model.

[0074] The GRU model was further trained using the minimum batch gradient descent method to optimize its parameters, with the highest KGE coefficient as the target for calibration.

[0075]

[0076] Where r represents the Pearson linear correlation coefficient between the simulated and measured series; α represents the ratio of the variances of the simulated and measured series; and β represents the ratio of the means of the simulated and measured series. The KGE efficiency coefficient ranges from (-∞, 1), and when KGE = 1, it indicates that the simulated and measured series are in perfect agreement.

[0077] In summary, the embodiments of the present invention establish a machine learning model that considers hydrological processes by coupling a hydrological model and a machine learning model, denoted as the TM model.

[0078] Based on the above embodiments, step S3 includes:

[0079] The differences between the output variables and observed meteorological variables in the global climate model ensemble meteorological simulation data at each quantile are calculated, and the differences are removed from the output future scenarios of the global climate model ensemble meteorological simulation data to obtain the corrected global climate model ensemble meteorological simulation data.

[0080] The machine learning model is driven by the corrected global climate model ensemble meteorological simulation data to output watershed hydrological process data under the future scenario.

[0081] This invention, based on a global climate model ensemble and quantile bias correction method, obtains a series of meteorological simulations under climate change scenarios and drives a machine learning model considering hydrological processes to simulate watershed hydrological processes under future scenarios, specifically including:

[0082] The first step is to obtain a series of meteorological simulations under climate change scenarios based on a global climate model ensemble and quantile bias correction methods.

[0083] The differences between the output variables of GCMs and the observed meteorological variables at each quantile (0.01-0.99) are calculated, and these differences are removed from each quantile of the future scenarios output by GCMs to obtain the future climate predictions of GCMs.

[0084] Corrections for air temperature (as well as specific humidity, relative humidity, wind speed, shortwave radiation, and longwave radiation) are as follows:

[0085]

[0086] The correction for precipitation is as follows:

[0087]

[0088] Where T and P represent air temperature (specific humidity, relative humidity, wind speed, shortwave radiation, and longwave radiation) and precipitation, respectively; adj represents the corrected series; obs represents the observed data; ref and fut represent the historical reference period and the future forecast period, respectively; d represents daily data; and Q... p Represents each quantile.

[0089] The second step uses calibrated simulated meteorological data to drive the watershed hydrological model and machine learning model to simulate watershed hydrological processes under future scenarios. The meteorological data under the climate change scenario calibrated in the first step are input into the calibrated watershed hydrological model and machine learning model to simulate the daily runoff series under the future climate change scenario.

[0090] Based on the above embodiments, step S4 includes:

[0091] The sum of the weight parameters for each combined scenario is determined to be 1 based on the total number of multiple combined scenarios and the weight parameter of any combined scenario.

[0092] The similarity between any given scenario and other scenarios is determined by the Euclidean difference and similarity radius of different scenario combinations, and the independence weight parameter is obtained based on the similarity between any given scenario and other scenarios.

[0093] Obtain the relative actual observation error and model quality radius of any combination scenario, and obtain the skill weight parameters based on the relative actual observation error and model quality radius of any combination scenario.

[0094] This invention employs a combined scenario (GCM-TM) based on M global climate models and a machine learning model considering hydrological processes to derive the weighting parameters of the multi-model weighted average method, specifically including:

[0095] The first step is to obtain the weight parameters of the combined scenarios by normalizing the independence weight parameters and skill weight parameters under the combined scenarios. The weight parameters of the combined scenarios satisfy:

[0096]

[0097] Where i represents the combined scenario, w o This represents the weighting parameter for the combined scenario, and M represents the number of global climate models.

[0098] The second step is to calculate the independence weight parameters and skill weight parameters for the combined scenarios, respectively. The independence weight parameter w for combined scenario i... u (i) can be represented as:

[0099]

[0100] Wherein, S(δ) i,j The similarity score () represents the similarity between the current scenario and other scenarios, with a value ranging from (0,1). The calculation formula is as follows:

[0101]

[0102] Where, δ i,j The Euclidean difference for different scenarios is obtained by standardizing the average difference of simulated runoff under scenarios i and j; D u The similarity radius is obtained by iteratively calculating the difference between simulated and measured runoff under the optimal combination scenario.

[0103] Skill weight parameter w for combined scenario i q (i) can be represented as:

[0104]

[0105] Where, δ i,obs The error of combined scenario i relative to actual observation is represented by D, which is obtained by standardizing the root mean square error of simulated runoff relative to measured runoff; q The mass radius of the model is represented by the difference between simulated and measured runoff under the optimal combination scenario; when D q When the value is close to 0, only the optimal combination scenario has a high skill weight.

[0106] Based on the above embodiments, step S5 includes:

[0107] Annual precipitation and annual runoff are calculated based on the water balance equation, and the actual evapotranspiration data at the annual scale under the climate scenario are obtained.

[0108] A preset time window is determined, and preset characteristic parameters are determined using the least squares method. The watershed hydrothermal coupling equilibrium equation is constructed using the potential evapotranspiration data output by the global climate model, the preset characteristic parameters, and the actual evapotranspiration data to obtain the watershed's annual average underlying surface characteristic parameters.

[0109] This invention establishes a watershed hydrothermal coupling equilibrium equation based on simulated meteorological and hydrological series to obtain the watershed's annual average underlying surface characteristic parameters, specifically including the following steps:

[0110] The first step involves calculating the actual annual evapotranspiration under the climate scenario using the Budyko hydrothermal balance equation (a model based on hydrothermal balance theory). The formula used is ET = Py - Ry, where ET is the actual evapotranspiration data, Py is the annual precipitation, and Ry is the annual runoff. A schematic diagram of this hydrothermal balance equation is shown below. Figure 4 As shown.

[0111] The second step involves selecting a fixed time window and using the least squares method to determine the characteristic parameter w of the hydrothermal coupling equilibrium equation. The annual average hydrothermal coupling equilibrium equation is as follows:

[0112]

[0113] PET represents potential evapotranspiration data output from global climate models.

[0114] Based on the above embodiments, step S6 includes:

[0115] A joint probability distribution function is constructed based on the Copula function, which combines the flood duration and the annual maximum peak flow. The flood attribute combination scenario is determined by the principle of maximizing the joint probability distribution function, and the maximum probability weight function is constructed based on the flood attribute combination scenario.

[0116] Based on the maximum possible weight function, the Lagrange multiplier method is used to solve the joint return period of flood duration and annual maximum peak flow under non-uniform conditions.

[0117] The return period of each combined scenario in the joint return period is obtained by using the Newton-Raphson iteration method. The return period of each combined scenario is then weighted and averaged according to the weight parameters of each combined scenario to obtain the socio-economic exposure caused by the increased risk of future floods.

[0118] This invention uses the parameter w, i.e., the annual average underlying surface characteristic parameter of the watershed, in the hydrothermal coupling equilibrium equation as a covariate to establish a joint probability distribution function based on Copula under inconsistent conditions. Based on the weight parameters of the combined scenario of global climate model output and fitted model, a multi-model weighted average method is used to assess the socioeconomic exposure caused by the increased risk of future floods. Specifically, the following steps are included:

[0119] The first step is to obtain the flood duration and annual maximum peak flow using the annual maximum sampling method, and then construct a joint probability distribution function of the flood duration and annual maximum peak flow based on the Copula function. Based on the principle of maximizing the joint probability density function, the combination scenario of two-dimensional flood attributes is selected, and the maximum possible weight function is constructed.

[0120] For any climate scenario, the Gumbel Copula function (a probability distribution function) is selected as the joint probability distribution function of the annual maximum flood peak discharge and flood duration, and the parameter θ of the Copula function is replaced with a time-varying parameter.

[0121]

[0122] in, Let Copula be the joint distribution function. The parameter is time-varying and ranges from (1,∞); u t v t Let be the probability density functions of the marginal distributions of flood duration D and peak flow Q, respectively, where t is time.

[0123] Based on the definition of a Copula function, a non-consistent two-variable Copula function can be expressed as:

[0124]

[0125] Among them, F t (d t ,q t ) represents the time-varying joint distribution function of flood duration D and peak flow Q; C represents the joint distribution function, and Let represent the cumulative density function and time-varying parameter of the time-varying marginal distributions of variables D and Q, respectively. Further, the parameters of the time-varying Copula function are expressed using covariates w as follows:

[0126]

[0127] in, Represents the connective function of a Copula function, when At that time (for GH Copula), b0 and b1 are the parameters of the model, respectively.

[0128] Furthermore, the maximum possible weight function is:

[0129]

[0130] Among them, H t (d t ,q t (d*, q*) represents the time-varying probability density function of variables D and Q, and (d*, q*) represents a certain joint return period T. or The most probable combination of flood duration d and annual maximum peak flow q; μ is the average interval between flood events; T or For the recurrence period of "or".

[0131] The second step is to calculate the joint return period of flood duration and annual maximum peak flow under non-consistent conditions based on the most likely combination scenario.

[0132] To solve the most probable combination problem using the Lagrange multiplier method, the following solution equations are constructed:

[0133]

[0134] in, Represent the Lagrangian functions of variables D and Q. This represents the time-varying Copula joint probability density function. Let λ represent the time-varying marginal distribution probability density functions of variables D and Q, respectively. t This represents the Lagrange multiplier corresponding to time state t.

[0135] The third step, based on the results of a weighted average of multiple models, estimates the socioeconomic exposure caused by increased flood risk under future climate change.

[0136] The Newton-Raphson iteration method was used to solve formula (16) to obtain the flood duration and annual maximum flood peak discharge (D) corresponding to a certain return period (Th) in the historical period (1985-2014). h Q h Furthermore, using a 30-year sliding window (consistent with the historical period length), the time-varying marginal distribution and Copula function for the future period (2015–2100) are constructed, and (D) h Q h Substituting the time-varying distribution function of the k-th sliding window of the future time period into the values, the new return period T is calculated. f (k).

[0137] After obtaining the return periods for each combination scenario, the weighted average method is used to obtain the average return period of the k-th sliding window under the M combination scenarios.

[0138]

[0139] Among them, w o (i) represents the weight parameters of combined scenario i; This represents the recurrence period of the k-th sliding window under combined scenario i.

[0140] like This indicates that the two-dimensional flood risk increases in the k-th window, and vice versa. For the k-th time window, the parameters of the marginal and joint distributions are derived using data from 15 years before and after the center point, and the socioeconomic exposure in future periods is measured by the following formula:

[0141]

[0142]

[0143] Among them, E pop and E GDP Population and GDP exposure to increased two-dimensional flood risk are represented respectively, POP k and GDP k Let be the population and GDP in year k, respectively; I(·) is the indicator function. The time period is recorded as 1, otherwise 0; N1 and N2 represent the start and end years of the research period, respectively.

[0144] Based on the above embodiments, the method further includes establishing an emergent constraint relationship between dew point temperature variability and peak flow variability using the watershed hydrological process data, and reducing the uncertainty of future peak flow prediction by introducing historical dew point observations, so as to achieve accurate flood risk prediction. The method for achieving accurate flood risk prediction specifically includes:

[0145] The variability of historical measured annual dew point temperature, the variability of annual dew point temperature in historical periods of multiple global climate models, and the corresponding variability of annual maximum flood peak flow in future periods were calculated using linear regression equations.

[0146] An emergent constraint relationship between the dew point temperature variability and the annual maximum flood peak flow variability is established, and the historical measured dew point temperature variability is used to constrain the future annual maximum flood peak flow variability in order to reduce the uncertainty of future flood peak prediction.

[0147] This invention's embodiment is based on an emergence constraint method, which uses historical measured dew point trends to constrain future peak flow trends. Specifically, it includes the following steps:

[0148] The first step uses the least squares method to calculate the trend of historically measured annual average dew point temperature, the trend of annual average dew point temperature in historical periods of multiple global climate models, and the trend of the corresponding annual maximum flood peak flow in future periods.

[0149] The second step involves establishing an emergence constraint equation for the historical annual average dew point temperature trend and the future annual maximum flood peak flow trend, driven by global climate model data. The historical observed annual average dew point trend is then used to replace the climate model's annual average dew point term in the equation, yielding the future annual maximum flood peak trend under measured constraints. The emergence constraint relationship is represented by a linear regression equation as follows:

[0150]

[0151]

[0152]

[0153]

[0154] Where, x l This represents the annual average dew point trend for the l-th historical period of the global climate model. Representing the emergence constraint through x l The predicted trend of the largest flood peak in future years, y lThis represents the trend of the maximum flood peak in future years calculated using the l-th global climate model, where a and b are the slope and intercept of the linear emergence constraint equation, respectively; s is the least squares error, and σ is the maximum flood peak value. x and σ y (x) represents the variance of x and the prediction variance of y for a given x, respectively. l The average value, N is x l The number of samples.

[0155] It can be seen that existing methods typically use the arithmetic mean or median of a set of combined scenarios directly into future periods to assess flood risk, without considering the contribution of different combined scenarios to the overall fitting result, which is somewhat unreasonable. This invention not only considers the inconsistencies caused by future climate change and underlying surface factors, but also the uncertainties inherent in the combined scenarios themselves, quantifying the weight of the combined scenarios in the overall fitting result, thus reducing errors in future risk assessment. Furthermore, this invention considers the emergent constraint relationship between historical dew point temperature variability and future predicted peak flow under climate model data forcing, and introduces measured dew point temperature variability to reduce the uncertainty in peak flow prediction.

[0156] The flood risk prediction system provided by the present invention is described below. The flood risk prediction system described below can be referred to in correspondence with the flood risk prediction method described above.

[0157] Figure 5 This is a schematic diagram of the flood risk prediction system provided in an embodiment of the present invention, as shown below. Figure 5 As shown, the system includes a data acquisition module 51, a training module 52, a correction module 53, a combination module 54, a setup module 55, and a prediction module 56 connected in sequence.

[0158] The acquisition module 51 is used to collect meteorological and hydrological data to form a meteorological and hydrological dataset; the training module 52 is used to determine the relative humidity and specific humidity at a set time, and train multiple watershed hydrological models based on the relative humidity, specific humidity, and the meteorological and hydrological dataset, and establish a machine learning model considering hydrological processes based on a gated recurrent unit neural network model and the trained multiple watershed hydrological models; the correction module 53 is used to acquire meteorological simulation data from a global climate model ensemble under a climate change scenario and correct the meteorological simulation data from the global climate model ensemble, and train the machine learning model using the corrected meteorological simulation data from the global climate model ensemble to obtain the watershed hydrological data under the future scenario. The data processing module 54 is used to combine multiple global climate models from the meteorological simulation data of the global climate model set with the machine learning model, and determine the weight parameters of each combined scenario in the multi-model weighted average method; the establishment module 55 is used to establish the watershed hydrothermal coupling equilibrium equation based on the watershed hydrological process data to obtain the watershed's annual average underlying surface characteristic parameters; the prediction module 56 is used to obtain the flood duration and annual maximum flood peak discharge from the watershed hydrological process data, and establish a joint probability distribution function based on the flood duration, the annual maximum flood peak discharge and the watershed's annual average underlying surface characteristic parameters, and perform flood risk prediction based on the joint probability distribution function and the weight parameters of each combined scenario.

[0159] It should be noted that the specific implementation of the flood risk prediction system in this embodiment of the invention can be found in the specific implementation of the flood risk prediction method described above. To avoid redundancy, it will not be repeated here.

[0160] Figure 6 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 6As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840. The processor 810, communication interface 820, and memory 830 communicate with each other via the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute a flood risk prediction method. This method includes: collecting meteorological and hydrological data to form a meteorological and hydrological dataset; determining the relative humidity and specific humidity at a set time, and training multiple watershed hydrological models based on the relative humidity, specific humidity, and the meteorological and hydrological dataset; establishing a machine learning model considering hydrological processes based on a gated recurrent unit neural network model and the trained multiple watershed hydrological models; acquiring meteorological simulation data from a global climate model ensemble under a climate change scenario and correcting the global climate model ensemble meteorological simulation data; and training a machine learning model using the corrected global climate model ensemble meteorological simulation data. The machine learning model is described to obtain watershed hydrological process data under future scenarios; multiple global climate models from the global climate model set meteorological simulation data are combined with the machine learning model, and the weight parameters of each combined scenario in the multi-model weighted average method are determined; a watershed hydrothermal coupling equilibrium equation is established based on the watershed hydrological process data to obtain the watershed's annual average underlying surface characteristic parameters; flood duration and annual maximum flood peak discharge are obtained from the watershed hydrological process data, and a joint probability distribution function is established based on the flood duration, the annual maximum flood peak discharge, and the watershed's annual average underlying surface characteristic parameters; and flood risk prediction is performed based on the joint probability distribution function and the weight parameters of each combined scenario.

[0161] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0162] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0163] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0164] In summary, the flood risk prediction method, system, and electronic equipment provided by this invention combine Earth system models, water-heat balance equations, watershed hydrological models, machine learning models, most probable combined scenario methods, multi-model weighted average methods, and emergent constraint methods. This provides important and highly operable reference data for watershed flood risk assessment and early warning under climate change scenarios, and offers engineering reference value for responding to future climate disasters and scientifically formulating emission reduction strategies.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A flood risk prediction method, characterized in that, include: Step S1: Collect meteorological and hydrological data to form a meteorological and hydrological dataset; Step S2: Determine the relative humidity and specific humidity at a set time, and train multiple watershed hydrological models based on the relative humidity, specific humidity and meteorological and hydrological datasets, and establish a machine learning model that considers the hydrological process based on the gated recurrent unit neural network model and the trained multiple watershed hydrological models. Step S3: Obtain meteorological simulation data from the global climate model ensemble under the climate change scenario and correct the meteorological simulation data from the global climate model ensemble, and train the machine learning model using the corrected meteorological simulation data from the global climate model ensemble to obtain watershed hydrological process data under the future scenario. Step S4: Combine multiple global climate models from the meteorological simulation data of the global climate model ensemble with the machine learning model to form various combined scenarios, and determine the weight parameters of each combined scenario in the multi-model weighted average method, specifically including: The similarity between any combination scenario and other combination scenarios is determined by the Euclidean difference and similarity radius of different combination scenarios, and the independence weight parameter is obtained based on the similarity between any combination scenario and other combination scenarios. Obtain the relative actual observation error and model quality radius of any combination scenario, and obtain the skill weight parameters based on the relative actual observation error and model quality radius of any combination scenario. The combined scenario weight parameters are obtained by normalizing the independence weight parameters and skill weight parameters under the combined scenarios. Step S5: Establish a watershed hydrothermal coupling equilibrium equation based on the watershed hydrological process data to obtain the watershed's annual average underlying surface characteristic parameters; Step S6: Obtain flood duration and annual maximum peak flow from the watershed hydrological process data, and establish a joint probability distribution function based on the flood duration, the annual maximum peak flow, and the watershed's annual average underlying surface characteristic parameters. Then, perform flood risk prediction based on the joint probability distribution function and the weight parameters of each combined scenario. In the flood risk prediction, the watershed hydrological process data is used to establish an emergent constraint relationship between dew point temperature variability and peak flow variability. Historical dew point observations are introduced to reduce the uncertainty of future peak flow prediction, thereby achieving accurate flood risk prediction. Specifically, the method for achieving accurate flood risk prediction includes: using linear regression equations to calculate historical measured annual average dew point temperature variability, annual average dew point temperature variability from multiple global climate models for historical periods, and their corresponding future annual maximum peak flow variability; establishing an emergent constraint relationship between dew point temperature variability and annual maximum peak flow variability, and using historical measured dew point temperature variability to constrain future annual maximum peak flow variability, thereby reducing the uncertainty of future peak flow prediction.

2. The flood risk prediction method as described in claim 1, characterized in that, Step S1 includes: Daily flow series were collected from watershed control hydrological stations, and meteorological data from the European Centre for Medium-Range Weather Forecasts' Generation 5 Atmospheric Reanalysis dataset (ERA5) were obtained. Meteorological data were collected from five preset global climate models in the sixth phase of the Coupled Model Intercomparison Project (CPIP). Obtain population and GDP data from the shared socioeconomic pathway dataset.

3. The flood risk prediction method as described in claim 1, characterized in that, Step S2 includes: By determining the latent heat of vaporization constant, water vapor gas constant, first integral constant, and second integral constant in the Clausius-Clapeyron thermodynamic equation, the nonlinear function of saturated water vapor pressure versus air temperature is obtained. The relative humidity is obtained from the surface air temperature and dew point temperature in ERA5, as well as the nonlinear function of saturated water vapor pressure and air temperature. The specific humidity is obtained based on the dew point temperature, the nonlinear function of the saturated water vapor pressure and air temperature, and the near-ground air pressure in ERA5. Based on the relative humidity, specific humidity, daily flow series of the watershed control hydrological stations, and precipitation, air pressure, surface air temperature, dew point temperature, surface downflow shortwave radiation and surface downflow longwave radiation data in ERA5, the multiple watershed hydrological models are trained to obtain preliminary simulated daily runoff. The preliminary simulated daily runoff is corrected using the gated recurrent unit neural network model, and an objective function is established based on the Kling-Gupta efficiency coefficient to obtain the machine learning model that considers the hydrological process.

4. The flood risk prediction method as described in claim 1, characterized in that, Step S3 includes: Calculate the difference between the output variable and the observed meteorological variable in the meteorological simulation data of the global climate model ensemble at each quantile, and remove the difference value from the output future scenario of the meteorological simulation data of the global climate model ensemble to obtain the corrected meteorological simulation data of the global climate model ensemble. The machine learning model is driven by the corrected global climate model ensemble meteorological simulation data to output watershed hydrological process data under the future scenario.

5. The flood risk prediction method as described in claim 1, characterized in that, Step S5 includes: Based on the water balance equation, annual precipitation and annual runoff are calculated to obtain actual evapotranspiration data at the annual scale under the climate scenario. A preset time window is determined, and preset characteristic parameters are determined using the least squares method. The watershed hydrothermal coupling equilibrium equation is constructed using the potential evapotranspiration data output by the global climate model, the preset characteristic parameters, and the actual evapotranspiration data to obtain the watershed's annual average underlying surface characteristic parameters.

6. The flood risk prediction method as described in claim 1, characterized in that, Step S6 includes: A joint probability distribution function is constructed based on the Copula function, which combines the flood duration and the annual maximum peak flow. The flood attribute combination scenario is determined by the principle of maximizing the joint probability distribution function. The maximum possible weight function is constructed based on the flood attribute combination scenario. Based on the maximum possible weight function, the Lagrange multiplier method is used to solve the joint return period of flood duration and annual maximum peak flow under non-uniform conditions; The return period of each combined scenario in the joint return period is obtained by using the Newton-Raphson iteration method. The return period of each combined scenario is then weighted and averaged according to the weight parameters of each combined scenario to obtain the socio-economic exposure caused by the increased risk of future floods.

7. A flood risk prediction system, characterized in that, include: The data acquisition module is used to collect meteorological and hydrological data to form a meteorological and hydrological dataset. The training module is used to determine the relative humidity and specific humidity at a set time, and to train multiple watershed hydrological models based on the relative humidity, specific humidity and the meteorological and hydrological dataset, and to establish a machine learning model that considers the hydrological process based on the gated recurrent unit neural network model and the trained multiple watershed hydrological models. The calibration module is used to acquire meteorological simulation data of the global climate model set under the climate change scenario and calibrate the meteorological simulation data of the global climate model set, and use the calibrated meteorological simulation data of the global climate model set to train the machine learning model to obtain watershed hydrological process data under the future scenario. The combination module is used to combine multiple global climate models from the meteorological simulation data of the global climate model ensemble with the machine learning model to form various combined scenarios, and to determine the weight parameters of each combined scenario in the multi-model weighted average method, specifically including: The similarity between any combination scenario and other combination scenarios is determined by the Euclidean difference and similarity radius of different combination scenarios, and the independence weight parameter is obtained based on the similarity between any combination scenario and other combination scenarios. Obtain the relative actual observation error and model quality radius of any combination scenario, and obtain the skill weight parameters based on the relative actual observation error and model quality radius of any combination scenario. The combined scenario weight parameters are obtained by normalizing the independence weight parameters and skill weight parameters under the combined scenarios. A module is established to establish a watershed hydrothermal coupling equilibrium equation based on the watershed hydrological process data, so as to obtain the watershed's annual average underlying surface characteristic parameters. The prediction module is used to acquire flood duration and annual maximum peak flow from the watershed hydrological process data, and to establish a joint probability distribution function based on the flood duration, the annual maximum peak flow, and the watershed's annual average underlying surface characteristic parameters. It then performs flood risk prediction based on the joint probability distribution function and the weight parameters of each combined scenario. In this flood risk prediction, the watershed hydrological process data is used to establish an emergent constraint relationship between dew point temperature variability and peak flow variability. Historical dew point observations are introduced to reduce the uncertainty of future peak flow prediction, thereby achieving accurate flood risk prediction. Specifically, the method for achieving accurate flood risk prediction includes: using linear regression equations to calculate historical measured annual average dew point temperature variability, annual average dew point temperature variability from multiple global climate models for historical periods, and their corresponding future annual maximum peak flow variability; establishing an emergent constraint relationship between dew point temperature variability and annual maximum peak flow variability, and using historical measured dew point temperature variability to constrain future annual maximum peak flow variability to reduce the uncertainty of future peak flow prediction.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the flood risk prediction method as described in any one of claims 1 to 6.

Images