Hydrological and hydrodynamic coupling method for flood simulation and prediction in alpine mountainous area

Abstract

The application discloses a hydrological and hydrodynamic coupling alpine mountain flood simulation and prediction method, and belongs to the technical field of disaster prediction. The method specifically comprises the following steps: S1, multi-source basic data collection and preprocessing: for the target alpine mountain basin, collect multi-type and multi-scale basic data; S2, deep learning correction and fusion of satellite precipitation data: for the local overestimation and underestimation problems of satellite precipitation data, a deep learning algorithm is used for hour-scale correction and fusion. By collecting satellite remote sensing, reanalysis, ground observation and geographic space four kinds of core data, covering all factors of 'precipitation-runoff-terrain-underlying surface' required for alpine mountain flood simulation, avoiding the one-sidedness of traditional modeling caused by the lack of data types, using 3 sigma criterion to remove precipitation and runoff outliers, unifying data format and space-time scale, forming a standardized data set, avoiding the interference of abnormal values, incompatible format and space-time mismatch on subsequent model input.

Description

A Method for Flood Simulation and Forecasting in High-Altitude Cold Regions Based on Hydrological-Hydrodynamic Coupling Technical Field

[0001] This invention relates to the field of disaster forecasting technology, and in particular to a method for flood simulation and forecasting in high-altitude and cold mountainous areas based on hydrological and hydrodynamic coupling. Background Technology

[0002] The high-altitude mountainous areas of Tibet are located in the core area of ​​the Qinghai-Tibet Plateau, with extremely complex underlying surface conditions. These include widespread glaciers, permafrost development, and significant topographical differences. Furthermore, the runoff generation and confluence processes of torrential rains and floods are significantly influenced by the combined effects of glacial meltwater and permafrost freeze-thaw cycles, resulting in mechanisms that differ markedly from those of ordinary mountain floods. Simultaneously, the region suffers from a sparse number of meteorological and hydrological observation stations, leading to a scarcity of measured precipitation and runoff data. This results in three core problems with existing hydrological models (such as traditional TOPMODEL and SWAT models): insufficient accuracy of multi-source data (satellite precipitation, reanalysis runoff) to directly support model input; failure to consider the impact of glacial meltwater and permafrost freeze-thaw cycles on runoff generation, leading to large simulation errors; and a disconnect between hydrological and hydrodynamic models, hindering accurate simulation of flood confluence processes and inundation extent. Summary of the Invention

[0003] Purpose of the invention: The purpose of this invention is to provide a method for flood simulation and forecasting in high-altitude and cold mountainous areas based on hydrological and hydrodynamic coupling; it can solve the problems of insufficient accuracy of multi-source data in existing technologies, failure to consider the impact of glacial meltwater and permafrost freeze-thaw on runoff generation, and the disconnect between hydrological models and hydrodynamic models.

[0004] Technical Solution: To solve the above-mentioned technical problems, according to one aspect of the present invention, more specifically, a method for flood simulation and forecasting in high-altitude cold mountainous areas based on hydrological-hydrodynamic coupling, the method specifically includes the following steps:

[0005] S1. Collection and preprocessing of multi-source basic data: Collect multi-type and multi-scale basic data for the target high-altitude and cold mountainous watershed;

[0006] S2. Deep learning correction and fusion of satellite precipitation data: To address the problem of local overestimation and underestimation in satellite precipitation data, deep learning algorithms are used for hourly-scale correction and fusion.

[0007] S3. Correction of GloFAS reanalysis runoff data and initial calibration of BTOP model parameters: To address the scarcity of measured runoff data in high-altitude and cold mountainous areas, the random forest algorithm is used to correct GloFAS reanalysis runoff data and apply it to the initial calibration of BTOP hydrological model parameters.

[0008] S4. Optimization of initial conditions for soil moisture content: Constructing the correlation between soil moisture content in the BTOP model and soil moisture content in ERA5-Land;

[0009] S5. Improved BTOP hydrological runoff model construction: In view of the impact of glacial meltwater and permafrost freeze-thaw on runoff in high-altitude and cold mountainous areas, an improved BTOP runoff model is formed by coupling the snowmelt and ice melt runoff sub-model and the soil freeze-thaw runoff sub-model on the basis of the optimized BTOP model.

[0010] S6. Construction of Hydrological-Hydrodynamic Coupled Model and Confluence Simulation: A distributed hydrological model and a two-dimensional hydrodynamic model are coupled to achieve accurate simulation of the flood confluence process and inundation range;

[0011] S7. Model Validation and Flood Forecasting Application: Ensure model reliability through multi-dimensional validation, and build a multi-time-effect flood forecasting system based on the model.

[0012] Furthermore, in step S1, the multi-source basic data includes:

[0013] Satellite remote sensing data: near real-time satellite precipitation data, MODIS land cover data, TPPG2017 glacier survey catalog data;

[0014] Further analysis of data: GloFAS daily runoff data, ERA5-Land stratified soil moisture data;

[0015] Ground observation data: hourly precipitation data from meteorological stations within the basin, measured runoff data from hydrological stations, and soil freezing threshold temperature observation data;

[0016] Geospatial data: watershed digital elevation model, land use type map, soil type map;

[0017] The above data were formatted, spatiotemporally matched, and precipitation and runoff anomalies were removed using the 3σ criterion to form a standardized dataset.

[0018] Furthermore, in step S2, the specific method for deep learning correction and fusion of satellite precipitation data is as follows: using hourly precipitation data from ground meteorological stations as the true value, an LSTM correction model of satellite precipitation data and measured precipitation data is constructed. The training process uses the Adam optimizer, with the loss function being the mean square error. After correcting the three types of satellite precipitation data respectively, a weighted fusion method is used to generate fused precipitation data with an hourly scale and a 1km resolution. The applicability of the fused precipitation data is verified using classification evaluation indicators and quantitative evaluation indicators.

[0019] Furthermore, in step S3, the specific steps for GloFAS reanalysis runoff data correction and BTOP model parameter initial calibration are as follows:

[0020] S31. Perform trend analysis and error assessment on the raw GloFAS daily runoff data to identify data deviation characteristics;

[0021] S32. Using the measured daily runoff data from a limited number of hydrological stations within the basin as the dependent variable, and the GloFAS raw runoff data, DEM slope, and land use type as independent variables, a random forest correction model was constructed and corrected at the time scale, spatial scale, and spatiotemporal coupling scale, respectively.

[0022] S33. Use the corrected GloFAS runoff data as observations and input them into the BTOP model. Use the SCE-UA optimization algorithm for initial parameter calibration. The objective function is to maximize NSE. After initial calibration, the model NSE should not be lower than the target value.

[0023] Furthermore, in step S4, the specific method for optimizing the initial soil moisture content conditions is as follows: extract the key soil moisture content variables of the BTOP model and the soil moisture content data of the ERA5-Land4 layer, perform multi-dimensional correlation analysis, screen variable combinations with correlation coefficients not lower than the threshold, construct the mapping relationship of the screened variable combinations using traditional curve fitting functions and LSTM networks respectively, take the ERA5-Land soil moisture content as input and the BTOP soil moisture content as output, select the mapping model with higher NSE through cross-validation, and replace the initial soil moisture content of the BTOP model with the soil moisture content calculated by the optimal mapping model to complete the optimization of the initial conditions of the model.

[0024] Furthermore, in step S5, when improving the BTOP hydrological runoff model construction, TPPG2017 glacier data and MODIS snow cover data are used as initial conditions. The degree-day factor method is used to calculate glacier / snow cover meltwater. The temperature threshold method is used to classify precipitation patterns, and meltwater and precipitation are combined into the total runoff input to construct a sub-model for snowmelt and icemelt runoff. An LSTM freeze-thaw runoff model is developed using Python. The input layer is the daily average temperature and precipitation, and the output layer is the runoff in the permafrost region. When the daily average temperature is greater than the soil freezing threshold, the runoff is calculated according to the original BTOP full-storage mechanism. When the daily average temperature is less than or equal to the soil freezing threshold, the runoff is calculated according to the infiltration runoff mechanism, directly generating surface runoff and constructing a soil freeze-thaw runoff sub-model. The above two sub-models are embedded into the BTOP runoff module using Python and FORTRAN hybrid programming technology, and the model is verified through multiple historical flood events.

[0025] Furthermore, step S6 specifically includes the following steps:

[0026] S61. Using the improved BTOP hydrological runoff model, the total runoff of the watershed is calculated at a 1km grid scale. The runoff from each grid to the river channel is calculated using the BTOP built-in runoff module, and the flow process of the target section of the river channel is output.

[0027] S62. A two-dimensional hydrodynamic model was developed using FORTRAN language. The river channels in the basin were discretized into single-line one-dimensional channels, and a two-dimensional grid was retained for the inundation risk zone.

[0028] S63. The cross-sectional flow rate of the river channel output by the improved BTOP model is used as the upstream boundary condition of the two-dimensional hydrodynamic model and loaded onto the corresponding river node. The two-dimensional diffusion wave equation is solved by the finite volume method to simulate the evolution process of the flood in the river channel and the water level and velocity distribution in the inundation area. The coupled model is verified by the inundation range of historical floods.

[0029] Furthermore, step S7 specifically includes the following steps:

[0030] S71. Verify the accuracy of the sub-model using field survey data, and verify the accuracy of the coupled model's flow simulation and inundation simulation using historical flood events;

[0031] S72. Using fused precipitation data as input, construct short-term, medium-term, and long-term flood forecasting modules based on the improved BTOP-hydrodynamic coupling model;

[0032] S73. By combining the "mountain torrent gene bank" with the flood depth and flow velocity data output by the model, high-risk areas within the basin are identified, and forecast results and risk zoning maps are generated to support the issuance of flood warnings.

[0033] Beneficial effects: By collecting four core types of data—satellite remote sensing, reanalysis, ground observation, and geospatial data—it covers all elements required for flood simulation in high-altitude and cold mountainous areas, including precipitation, runoff, topography, and underlying surface. This avoids the one-sidedness of traditional modeling caused by the lack of data types. The 3σ criterion is used to remove outliers in precipitation and runoff, and the data format and spatiotemporal scale are unified to form a standardized dataset. This avoids interference from outliers, format incompatibility, and spatiotemporal mismatches on subsequent model inputs, laying a reliable data foundation for the entire modeling process.

[0034] To address the shortcomings of satellite precipitation data such as GSMAP, IMERG, and PERSIANN in terms of "local overestimation / underestimation," an hourly-scale correction is performed using an LSTM model, followed by weighted fusion to generate high-precision precipitation data. This produces hourly-scale precipitation data with a 1km resolution, which is suitable for simulating "small-scale heavy precipitation" in complex terrains of high-altitude and cold mountainous areas. This solves the problem that traditional low-resolution precipitation data cannot characterize localized rainstorms, providing accurate precipitation input for subsequent runoff simulation.

[0035] To address the challenges of limited hydrological stations and insufficient measured runoff data in high-altitude mountainous areas, a random forest algorithm is used to calibrate GloFAS reanalysis runoff data at temporal / spatial / spatiotemporal scales. This calibrated data can then be used to replace measured data for model calibration. The SCE-UA optimization algorithm is employed to perform initial BTOP model calibration using the calibrated GloFAS data as observations. This avoids model bias caused by the traditional approach of setting parameters based on experience when no measured data is available, and provides reliable initial parameters for subsequent model optimization.

[0036] By using multi-dimensional correlation analysis, the optimal variable combination of soil moisture content in the BTOP model and stratified soil moisture content in ERA5-Land was selected. Then, by comparing and selecting the mapping model with higher NSE through "curve fitting + LSTM network", the calculation results were used to replace the initial soil moisture content of BTOP to solve the problem of "initial soil moisture content being assigned based on experience" in traditional models. This makes the initial parameters of the model more consistent with the actual hydrological conditions of the watershed, avoids the transmission of initial condition errors to subsequent runoff generation and confluence simulations, and improves the reliability of the overall simulation results.

[0037] Coupled with the "snowmelt-ice sub-model", the glacier / snowmelt water volume is accurately calculated, solving the problem that traditional models ignore the contribution of meltwater. Coupled with the "soil freeze-thaw sub-model", the conversion between "full-storage runoff" and "excessive infiltration runoff" in permafrost areas is dynamically simulated, which is consistent with the actual situation of high-altitude permafrost. The two sub-models are embedded into BTOP through mixed programming of Python and FORTRAN. After verification by multiple historical floods, the error rate is effectively reduced.

[0038] The improved BTOP model calculates the runoff flow at a 1km grid scale and outputs the flow process at the target cross-section of the river channel, ensuring the accuracy of the runoff data. The two-dimensional hydrodynamic model (developed with FORTRAN) discretizes the river channel into one dimension and retains the two-dimensional grid of the inundation zone. The diffusion wave equation is solved by the finite volume method to accurately simulate the water level and velocity distribution in the inundation zone. This solves the defect of traditional hydrological models that "can only output flow and cannot characterize the inundation range", providing comprehensive data of "flow + spatial distribution" for flood risk zoning and evacuation route planning.

[0039] Based on the coupled model, a three-level forecasting module for short-term, medium-term and long-term forecasts is constructed. The sub-models are verified by field survey data and the coupled model is verified by historical floods to ensure the reliability of forecast results and improve the practicality of disaster prevention. Combined with the "mountain torrent gene bank" to identify high-risk areas, it assists local departments in quickly initiating emergency response and allocating flood control materials, directly supporting the entire process of flood "forecasting, early warning, rehearsal and contingency planning", and effectively reducing casualties and property losses. Attached Figure Description

[0040] Figure 1 is a schematic diagram of the method flow. Detailed Implementation

[0041] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] Example 1: Step 1, Multi-source basic data collection and preprocessing

[0043] Data type:

[0044] Satellite remote sensing data: ① Near real-time satellite precipitation: GSMAP (JAXA website, 0.1° resolution), IMERG (NASA website, 0.1° resolution), PERSIANN (UCI website, 0.25° resolution); ② MODIS land cover data (MOD13Q1, 1km resolution, NASA); ③ TPPG 2017 Glacier Survey Catalog (International Centre for Integrated Mountain Development).

[0045] Further analysis of the data: ①GloFAS daily runoff data (ECMWF official website, 0.1° resolution); ②ERA5-Land stratified soil moisture data (CopernicusClimateDataStore, 0.1° resolution, 4 layers: 0-7cm, 7-28cm, 28-100cm, 100-289cm).

[0046] Ground monitoring data: ① Meteorological stations: Lhasa station (hourly precipitation in July 2022, accuracy 0.1 mm), Damxung station (hourly precipitation during the same period); ② Hydrological stations: Lhasa hydrological station (daily runoff in July 2022, accuracy 1 m³ / s); ③ Soil freezing threshold: measured at Damxung permafrost observation station -2.0℃.

[0047] Geospatial data: ① Digital Elevation Model (DEM, 30m resolution, Geospatial Data Cloud); ② 2020 Land Use Type Map (1km resolution, Resource and Environment Data Center, Chinese Academy of Sciences); ③ Soil Type Map (1:1,000,000, Ministry of Natural Resources).

[0048] Preprocessing steps:

[0049] Standardized format: Convert all data to NetCDF format (for easy model access);

[0050] Spatiotemporal matching: Spatial resolution was resampled to 1km (using bilinear interpolation), and the time scale was unified to hour (precipitation) / day (runoff, soil moisture content);

[0051] Outlier removal: Using the 3σ criterion, data with hourly precipitation >50mm (extreme anomaly) at Lhasa station and daily runoff >2000m³ / s (exceeding historical extremes) at Lhasa hydrological station were removed, resulting in a standardized dataset.

[0052] The second step involves deep learning correction and fusion of satellite precipitation data.

[0053] LSTM correction model construction

[0054] True data: Hourly precipitation data from Lhasa and Damxung stations from January to June 2022 (4320 records in total, 70% for training and 30% for validation).

[0055] Model structure: Input layer (3 neurons, corresponding to GSMAP, IMERG, and PERSIANN precipitation) → Hidden layer (2 layers, 64 neurons per layer, ReLU activation function) → Output layer (1 neuron, corrected hourly precipitation);

[0056] Training parameters: Adam optimizer (learning rate 0.001), loss function is mean squared error (MSE), number of iterations 100, batch size 32.

[0057] Verification of correction effect

[0058] The corrected Nash efficiency coefficients (NSE) of the three types of satellite precipitation data are: IMERG 0.81, GSMAP 0.77, and PERSIAN N 0.73, all of which are more than 30% higher than the original data (NSE 0.52-0.65).

[0059] Weighted fusion

[0060] The weights were determined based on the NSE during the validation period: IMERG (0.4), GSMAP (0.35), and PERSIANN (0.25). After fusion, precipitation data with a resolution of 1 km and an hourly scale were generated. The NSE during the validation period reached 0.82, and the root mean square error (RMSE) was reduced to 4.2 mm / h.

[0061] The third step is to correct the runoff data using GloFAS reanalysis and perform initial calibration of the BTOP model parameters.

[0062] 1. Original GloFAS Error Analysis

[0063] Compared with the measured runoff at the Lhasa hydrological station in 2020-2021, the original GloFAS daily runoff data was systematically 15.3% lower, and the flood peak time was delayed by 2-3 days, requiring multi-scale correction.

[0064] 2. Random Forest Correction Model

[0065] Independent variables: original GloFAS daily runoff, DEM slope (mean value of 1km grid), land use type (grassland / bare land / cultivated land code).

[0066] Dependent variable: Daily runoff measured at Lhasa Hydrological Station in 2020-2021 (730 runoffs in total, 80% training, 20% validation);

[0067] Model parameters: number of decision trees 100, maximum tree depth 10, 5-fold cross-validation, corrected time scale NSE 0.78, spatial scale (sub-basin) NSE 0.75, spatiotemporal coupling scale NSE 0.80.

[0068] 3. Initial calibration of the BTOP model

[0069] Model parameter range: soil saturated hydraulic conductivity (1-10 mm / h), soil porosity (0.3-0.5), surface roughness (0.02-0.05);

[0070] Optimization algorithm: SCE-UA (Hybrid Complex Evolutionary Algorithm), objective function is to maximize NSE;

[0071] Initial calibration results: The model's NSE reached 0.68, and the peak error was 12.5%, which meets the basic requirements for subsequent optimization.

[0072] Step 4: Optimization of initial soil moisture conditions

[0073] Correlation analysis of variables

[0074] Key variables of the BTOP model (root zone soil moisture content Srz, unsaturated zone soil moisture content Suz, and saturated soil water deficit SD) were extracted and compared with the soil moisture content of the ERA5-Land4 layer. The Pearson correlation coefficient was calculated.

[0075] The correlation coefficient between Srz and ERA5-Land layer 2 (7-28cm) is 0.72;

[0076] The correlation coefficient between Suz and ERA5-Land layer 3 (28-100cm) was 0.68;

[0077] The above two sets of variable combinations are selected for constructing the mapping relationship.

[0078] Mapping Model Comparison

[0079] Traditional curve fitting: using a power function (y=ax^b), Srz mapping R² 0.65, Suz mapping R² 0.62;

[0080] LSTM network: Input layer (4 neurons, ERA5-Land4 layer water content) → Hidden layer (1 layer with 32 neurons) → Output layer (2 neurons, Srz and Suz), after cross-validation (8:2), NSE is 0.76 (Srz) and 0.73 (Suz).

[0081] The LSTM model was chosen as the optimal mapping scheme.

[0082] Initial condition replacement

[0083] Replacing the default initial values ​​of the BTOP model with Srz and Suz calculated by LSTM, the model's NSE improved to 0.71 after recalibration, and the initial condition error was reduced by 28%.

[0084] Step 5: Improve the construction of the BTOP hydrological runoff model

[0085] Snowmelt and ice melt flow generation model

[0086] Initial conditions: TPPG2017 glacier data (to determine the spatial distribution of glaciers), MODIS snow cover data (MOD10A1, 8-day composite, 1km resolution).

[0087] Day-of-day factor: Based on glacial meltwater observations at Damxung Station from 2019 to 2021, the day-of-day meltwater factor was determined to be 6.5 mm / (°C). d) Snowmelt per day factor: 3.2 mm / (°C) d);

[0088] Precipitation patterns are classified as follows: 0℃ temperature threshold (measured at Damxung station), temperatures ≥0℃ are considered precipitation, and <0℃ are considered snowfall. Meltwater and precipitation are combined to form the total runoff input.

[0089] Soil freeze-thaw flow generation model

[0090] Model development: The LSTM model is built using Python, with an input layer (daily average temperature and daily precipitation) → hidden layers (2 layers with 48 neurons) → output layer (permafrost runoff).

[0091] Runoff generation mechanism switching: If the average daily temperature is > -2.0℃ (soil freezing threshold), runoff generation is calculated based on the original full storage capacity of BTOP; if it is ≤ -2.0℃, runoff generation is calculated based on infiltration excess (precipitation directly forms surface runoff).

[0092] Module coupling: Python (sub-model) and FORTRAN (BTOP core) are used in mixed programming, and sub-model embedding is achieved through dynamic link libraries (DLLs).

[0093] Model Validation

[0094] The improved BTOP model was validated using the Lhasa River flood in September 2018 (measured peak flow of 980 m³ / s). The NSE of the improved model was 0.75, which is 21% lower than the original model (NSE 0.61).

[0095] Step 6: Construction of hydrological-hydrodynamic coupled model and confluence simulation

[0096] 1. Hydrological model runoff calculation

[0097] The improved BTOP model calculates the total runoff at a 1km grid scale. It uses the topographic index confluence module built into BTOP to calculate the confluence flow from each grid to the Lhasa hydrological station section and outputs the flood flow process in July 2022 (simulated peak flow 1250 m³ / s, measured 1280 m³ / s, error 2.3%).

[0098] 2. Construction of a two-dimensional hydrodynamic model

[0099] Model development: Written in FORTRAN, based on two-dimensional diffuse wave equations, using unstructured mesh;

[0100] Grid discretization: The main stream of the Lhasa River (from its source to the Lhasa Hydrological Station) is discretized into a single-line one-dimensional river channel (grid length 500m), while the flood risk area (surrounding Lhasa city and Duilongdeqing District) retains a two-dimensional grid (500m×500m).

[0101] Parameter settings: Manning roughness coefficient (0.03 for river channels, 0.05 for grasslands, and 0.04 for cultivated land), time step 60s (to meet CFL conditions).

[0102] 3. Model Coupling and Simulation

[0103] Boundary conditions: The Lhasa hydrological station cross-sectional flow rate output by the improved BTOP model is used as the upstream boundary of the two-dimensional hydrodynamic model, and the downstream boundary is the Lhasa River estuary tidal level (measured, average tidal level in July 2022 is 3590.2m).

[0104] Numerical solution: The finite volume method is used to solve the two-dimensional diffusion wave equation to simulate the evolution of flood in the river channel and the distribution of water level and velocity in the inundation area;

[0105] Verification: The flooding range interpreted by Sentinel-2 satellite (July 17, 2022, 10m resolution) showed a simulation overlap rate of 86% and a maximum flooding level error of 0.25m.

[0106] Step 7: Model Validation and Flood Forecasting Application

[0107] 1. Multi-dimensional verification

[0108] Sub-model validation: The meltwater volume error of the snow accumulation sub-model was 10%, and the flow rate error of the freeze-thaw sub-model was 8%.

[0109] Coupled model verification: The flow simulation NSE is 0.83 and the flooding range overlap rate is 86%, both of which meet the accuracy requirements for flood simulation in high-altitude and cold mountainous areas.

[0110] 2. Construction of multi-time-leader forecast module

[0111] Short-term forecast (0-12h): Input fused precipitation data (hourly forecasts for the next 12 hours), model calculation time 45 minutes, forecast peak flood error 4.8%;

[0112] Medium-range forecast (12-72h): Input ECMWF numerical forecast precipitation (updated every 12 hours), forecast flood peak error 7.5%;

[0113] Long-term forecast (72-168h): Input precipitation from China Meteorological Administration's T639 model (updated every 24 hours), forecast flood peak error is 11.2%.

[0114] 3. Identification and early warning of high-risk areas

[0115] Flash flood gene bank: integrating the topographic (slope > 25°) and hydrological (flood peak > 800 m³ / s) characteristics of 12 flash flood events in the Lhasa River Basin from 2015 to 2021;

[0116] Risk identification: High-risk areas (southeast of Lhasa city and north of Duilongdeqing District) are identified by using flood depth (>1.5m) and flow velocity (>2.0m / s) data output by the coupled model.

[0117] Warning issuance: The forecast results and risk zoning maps are generated and issued through the Tibet Autonomous Region Flood Control Command Platform. The time from data input to warning output is ≤1.5 hours.

[0118] Example 2: Step 1, Multi-source basic data collection and preprocessing

[0119] Satellite remote sensing data: ① Satellite precipitation: GSMAP (JAXA, 0.1°), IMERG (NASA, 0.1°), PERSIANN (UCI, 0.25°); ② MODIS land cover (MOD13Q1, 1km, NASA); ③ TPPG 2017 Glacier Inventory (International Centre for Integrated Mountain Development).

[0120] Further analysis of the data: ①GloFAS daily runoff (ECMWF, 0.1°); ②ERA5-Land soil moisture content (Copernicus, 0.1°, 4 layers).

[0121] Ground observation data: ① Meteorological stations: Bayi Station (hourly precipitation in August 2020), Gongbujiangda Station (hourly precipitation during the same period); ② Hydrological stations: Bayi Hydrological Station (daily runoff in August 2020); ③ Soil freezing threshold: -2.5℃ measured at Gongbujiangda Frozen Soil Station.

[0122] Geospatial data: ①DEM (30m, geospatial data cloud); ②2020 land use map (1km, Resource Center of Chinese Academy of Sciences); ③Soil type map (1:1,000,000, Ministry of Natural Resources).

[0123] Preprocessing operations

[0124] Standardized format: Converted to NetCDF format for easy model usage;

[0125] Spatiotemporal matching: spatial resampling to 1km (bilinear interpolation), and time uniformity to hours (precipitation) / day (runoff);

[0126] Outlier removal: The 3σ criterion was used to remove data with hourly precipitation > 60 mm (extreme rainstorm) and daily runoff > 1500 m³ / s (exceeding historical extreme values) at Bayi Station, forming a standardized dataset.

[0127] The second step involves deep learning correction and fusion of satellite precipitation data.

[0128] LSTM calibration model

[0129] True data: Hourly precipitation at Bayi and Gongbujiangda stations from January to July 2019 (5040 data points in total, 70% training, 30% validation).

[0130] Model structure: Input layer (3 neurons, three types of satellite precipitation) → Hidden layer (2 layers, 48 ​​neurons, ReLU activation) → Output layer (1 neuron, corrected precipitation).

[0131] Training parameters: Adam optimizer (learning rate 0.0015), MSE loss function, 120 iterations, batch size 64.

[0132] Correction effect

[0133] The corrected satellite precipitation NSE values ​​are: IMERG 0.83, GSMAP 0.79, PERSIAN N 0.75, representing a 32% improvement over the original data (NSE 0.55-0.67).

[0134] Weighted fusion

[0135] Weights: IMERG (0.45), GSMAP (0.3), PERSIANN (0.25), after fusion, the precipitation data NSE is 0.85 and RMSE is 3.8 mm / h (1 km, hourly scale).

[0136] The third step is to correct the runoff data using GloFAS reanalysis and perform initial calibration of the BTOP model parameters.

[0137] Compared with the measured runoff at the Bayi Hydrological Station in 2018-2019, the original GloFAS was systematically higher by 12.1%, and the flood peak was delayed by 1-2 days.

[0138] Random Forest Correction

[0139] Variables: Independent variables (GloFAS runoff, DEM slope, land use), dependent variable (measured runoff at Bayi station, 730 data points, 80% training data).

[0140] Model parameters: Decision tree 120, maximum depth 12, 5-fold cross-validation;

[0141] Correction results: NSE 0.80 for time scale, NSE 0.76 for spatial scale, and NSE 0.82 for spatiotemporal scale.

[0142] BTOP initial rate determination

[0143] Parameter range: soil saturated hydraulic conductivity (1.5-12 mm / h), porosity (0.32-0.52), surface roughness (0.025-0.055);

[0144] After SCE-UA optimization, the model has an NSE of 0.71 and a peak error of 10.8%.

[0145] Step 4: Optimization of initial soil moisture conditions

[0146] Correlation analysis

[0147] The correlation between BTOP variables and ERA5-Land: Srz has a correlation coefficient of 0.75 with layer 2 (7-28cm), and Suz has a correlation coefficient of 0.70 with layer 3 (28-100cm).

[0148] Mapping Model Comparison

[0149] Curve fitting (power function): SrzR²0.68, SuzR²0.65;

[0150] LSTM model: Input (water content of 4 layers of ERA5-Land) → Output (Srz, Suz), cross-validation NSE 0.79 (Srz), 0.76 (Suz);

[0151] Choose the LSTM mapping model.

[0152] Initial condition replacement

[0153] After the replacement, the NSE of the BTOP model improved to 0.74, and the initial error decreased by 31%.

[0154] Step 5: Improve the construction of the BTOP hydrological runoff model

[0155] Accumulated snow melting ice model

[0156] Initial conditions: TPPG2017 glacier data, MOD10A1 snow cover data;

[0157] Day-to-day factor: Based on observations at the Gongbujiangda station, ice melt rate is 7.0 mm / (°C). d) Snow melting 3.5mm / (℃) d);

[0158] Precipitation pattern: 0℃ threshold, meltwater + rainfall is the runoff input.

[0159] Soil freeze-thaw sub-model

[0160] Python-LSTM model: Input (daily temperature, daily precipitation) → Output (permafrost runoff), 2 hidden layers with 56 neurons;

[0161] Production flow switching: Full production flow is achieved when the daily temperature is > -2.5℃ (freezing threshold), and excessive production flow occurs when the temperature is ≤ -2.5℃;

[0162] Coupling method: Python-FORTRAN hybrid programming, BTOP is embedded through API interface.

[0163] verify

[0164] The results were verified using the June 2019 flood (measured peak flow of 820 m³ / s). The improved BTOP runoff NSE was 0.78, while the original model had an NSE of 0.63, resulting in a 22% reduction in error.

[0165] Step 6: Construction of hydrological-hydrodynamic coupled model and confluence simulation

[0166] Hydrological runoff calculation

[0167] The BTOP algorithm was improved to calculate runoff generation on a 1km grid, which was then channeled to the Bayi Hydrological Station to output the flood flow process in August 2020 (simulated peak flow rate of 1030 m³ / s, actual peak flow rate of 1050 m³ / s, with an error of 1.9%).

[0168] Two-dimensional hydrodynamic model

[0169] Developed with FORTRAN, two-dimensional diffused wave equations, unstructured mesh;

[0170] Discretization method: one-dimensional river channel (500m grid) of the main stream of Niyang River (above Bayi Hydrological Station), and two-dimensional grid (500m×500m) of the flooded area (Bayi Town and Linzhi Town);

[0171] Parameters: Manning roughness (channel 0.028, forest 0.055, grassland 0.045), time step 60s.

[0172] Coupled simulation

[0173] Boundary: Upstream (flow rate at Bayi Hydrological Station), downstream (water level at the mouth of the Niyang River, 3012.5m);

[0174] Solution: Finite volume method, simulating a maximum water level of 3.8m (Bayi Town) and a flow velocity of 2.1m / s;

[0175] Verification: The area flooded by Sentinel-2 remote sensing (August 7, 2020) has an overlap rate of 88%.

[0176] Step 7: Model Validation and Flood Forecasting Application

[0177] Verification results

[0178] Sub-model error: meltwater volume 8%, freeze-thaw flow rate 7%;

[0179] Coupled model: flow rate NSE 0.85, flooding overlap rate 88%.

[0180] Forecast module

[0181] Short-term (0-12h): Error 4.5%, time 50 minutes;

[0182] Mid-term (12-72h): Error 7.2%;

[0183] Long-term (72-168h): Error 10.8%.

[0184] Risk warning

[0185] Flash flood gene pool: characteristics of eight flash flood events from 2016 to 2021;

[0186] High-risk areas: northeast of Bayi Town and western Linzhi Town (flood depth > 1.5m, flow velocity > 2.0m / s).

[0187] Warning output: The warning was sent to the Linzhi City Flood Control Office within 2 hours to assist in flood avoidance in August 2020, and more than 1,200 people were evacuated.

[0188] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for flood simulation and forecasting in high-altitude cold mountainous areas based on hydrological-hydrodynamic coupling, characterized in that, The method specifically includes the following steps: S1, Multi-source basic data collection and preprocessing: Collecting multi-type and multi-scale basic data for the target high-altitude mountainous watershed; S2, Deep learning correction and fusion of satellite precipitation data: Addressing the local overestimation and underestimation issues in satellite precipitation data, using deep learning algorithms for hourly-scale correction and fusion; S3, GloFAS reanalysis runoff data correction and initial calibration of BTOP model parameters: Addressing the scarcity of measured runoff data in high-altitude mountainous areas, using the random forest algorithm to correct GloFAS reanalysis runoff data and applying it to the initial calibration of BTOP hydrological model parameters; S4, Optimization of initial soil moisture conditions: Constructing the soil moisture content for the BTOP model. The correlation with ERA5-Land soil moisture content; S5, Improved BTOP hydrological runoff model construction: Addressing the impact of glacial meltwater and permafrost freeze-thaw on runoff in high-altitude mountainous areas, an improved BTOP runoff model is formed by coupling a snowmelt / ice meltwater runoff sub-model and a soil freeze-thaw runoff sub-model based on the optimized BTOP model; S6, Hydrological-hydrodynamic coupling model construction and confluence simulation: A distributed hydrological model and a two-dimensional hydrodynamic model are coupled to achieve accurate simulation of flood confluence processes and inundation extent; S7, Model validation and flood forecasting application: Multi-dimensional validation ensures model reliability, and a multi-time-effect flood forecasting system is built based on the model; In step S1, multi-source base... The basic data includes: satellite remote sensing data: near-real-time satellite precipitation data, MODIS land cover data, TPPG2017 glacier survey catalog data; reanalysis data: GloFAS daily runoff data, ERA5-Land stratified soil moisture data; ground observation data: hourly precipitation data from meteorological stations within the basin, measured runoff data from hydrological stations, and soil freezing threshold temperature observation data; geospatial data: basin digital elevation model, land use type map, and soil type map; the above data are formatted, spatiotemporally matched, and precipitation and runoff anomalies are removed using the 3σ criterion to form a standardized dataset; step S6 specifically includes the following steps: S61, using the improved BT... The OP hydrological runoff model calculates the total runoff of the watershed at a 1km grid scale. It uses the BTOP built-in confluence module to calculate the confluence flow from each grid to the river channel and outputs the flow process at the target cross-section of the river channel. S62: A two-dimensional hydrodynamic model is developed using FORTRAN language. The river channel within the watershed is discretized into a single-line one-dimensional channel, while the two-dimensional grid is retained for the inundation risk zone. S63: The cross-sectional flow of the river channel output by the improved BTOP model is used as the upstream boundary condition of the two-dimensional hydrodynamic model and loaded onto the corresponding river node. The two-dimensional diffusion wave equation is solved using the finite volume method to simulate the evolution of the flood in the river channel and the water level and velocity distribution in the inundation zone. The coupled model is verified using the inundation range of historical floods.

2. The flood simulation and forecasting method for high-altitude cold mountainous areas based on hydrodynamic coupling as described in claim 1, characterized in that: In step S2, the specific method for deep learning correction and fusion of satellite precipitation data is as follows: using hourly precipitation data from ground meteorological stations as the true value, an LSTM correction model of satellite precipitation data and measured precipitation data is constructed. The training process uses the Adam optimizer, with the mean square error as the loss function. After correcting the three types of satellite precipitation data respectively, a weighted fusion method is used to generate fused precipitation data with an hourly scale and a 1km resolution. The applicability of the fused precipitation data is verified using classification evaluation indicators and quantitative evaluation indicators.

3. The flood simulation and forecasting method for high-altitude cold mountainous areas based on hydrodynamic coupling as described in claim 1, characterized in that: In step S3, the specific steps for GloFAS reanalysis runoff data correction and BTOP model parameter initial calibration are as follows: S31, perform trend analysis and error assessment on the original GloFAS daily runoff data to identify data bias characteristics; S32, using the measured daily runoff data from a limited number of hydrological stations within the watershed as the dependent variable, and the original GloFAS runoff data, DEM slope, and land use type as independent variables, construct a random forest correction model and perform corrections at the time scale, spatial scale, and spatiotemporal coupling scale; S33, use the corrected GloFAS runoff data as observations, input them into the BTOP model, and use the SCE-UA optimization algorithm for parameter initial calibration. The objective function is to maximize NSE, and the model NSE after initial calibration must not be lower than the target value.

4. The flood simulation and forecasting method for high-altitude cold mountainous areas based on hydrodynamic coupling as described in claim 1, characterized in that: In step S4, the specific method for optimizing the initial soil moisture content conditions is as follows: extract the key soil moisture content variables of the BTOP model and the soil moisture content data of the ERA5-Land4 layer, perform multi-dimensional correlation analysis, screen variable combinations with correlation coefficients not lower than the threshold, construct the mapping relationship of the screened variable combinations using traditional curve fitting functions and LSTM networks respectively, take the ERA5-Land soil moisture content as input and the BTOP soil moisture content as output, select the mapping model with higher NSE through cross-validation, and replace the initial soil moisture content of the BTOP model with the soil moisture content calculated by the optimal mapping model to complete the optimization of the initial conditions of the model.

5. The flood simulation and forecasting method for high-altitude cold mountainous areas based on hydrodynamic coupling as described in claim 1, characterized in that: In step S5, when constructing the improved BTOP hydrological runoff model, TPPG2017 glacier data and MODIS snow cover data are used as initial conditions, and the degree-day factor method is used to calculate the glacier / snow cover meltwater. The temperature threshold method was used to classify precipitation patterns. Meltwater and precipitation were combined into the total runoff input to construct a runoff sub-model for snowmelt and icemelt. An LSTM freeze-thaw runoff model was developed using Python. The input layer is the daily average temperature and precipitation, and the output layer is the runoff in the permafrost region. When the daily average temperature is greater than the soil freezing threshold, the runoff generation mechanism is calculated according to the original BTOP full storage mechanism; when the daily average temperature is less than or equal to the soil freezing threshold, the runoff generation mechanism is calculated according to the infiltration excess mechanism, and surface runoff is directly generated. A soil freeze-thaw runoff generation sub-model is constructed. The above two sub-models are embedded into the BTOP runoff generation module using Python and FORTRAN hybrid programming technology. The model is verified through multiple historical flood events.

6. The flood simulation and forecasting method for high-altitude cold mountainous areas based on hydrodynamic coupling as described in claim 1, characterized in that: Step S7 specifically includes the following steps: S71, verifying the accuracy of the sub-model using field survey data, and verifying the accuracy of the flow simulation and inundation simulation of the coupled model using historical flood events; S72, using fused precipitation data as input, constructing short-term, medium-term, and long-term flood forecasting modules based on the improved BTOP-hydrodynamic coupled model; S73, combining the "mountain torrent gene bank", identifying high-risk areas within the basin through the flood depth and flow velocity data output by the model, generating forecast results and risk zoning maps to support the issuance of flood warnings.

Images