Isotope tracing based method and system for flood model parameter calibration

Abstract

The application provides a flood model parameter calibration method and system based on isotope tracing, wherein the method comprises the following steps: determining a parameter set and a prior distribution in a flood model which need to be analyzed, the parameter set comprising parameters related to runoff and parameters related to isotope transmission in the flood model; generating a group of parameter samples according to the prior distribution of the parameters; evaluating the sensitivity of each parameter in each group of parameter samples by analyzing the relationship between the parameter samples and corresponding target function values; and screening out sensitive parameters according to the sensitivity evaluation results. The flood model parameter calibration method based on isotope tracing provided by the application solves the problem of the same effect of different parameters of the flood model, that is, greatly reduces the same effect of different parameters of the model (that is, the parameter range when the model evaluation index reaches above a certain threshold), and greatly improves the simulation effect of the flood model in the verification period.

Description

Technical Field

[0001] This invention relates to the field of flood simulation and forecasting technology, and in particular to a method and system for calibrating flood model parameters based on isotope tracing. Background Technology

[0002] When a model is applied in a scenario, parameter calibration is crucial, as it determines the model's effectiveness in that scenario. Parameter calibration for flood models is key to accurately simulating and predicting floods. In recent years, many studies have applied advanced technologies such as deep learning and genetic algorithms to parameter calibration. However, since the calibration target is still runoff, it cannot fundamentally solve the pain points of traditional parameter calibration methods. Problems such as overfitting and different parameters having the same effect often exist, resulting in good performance during calibration but poor performance during validation and prediction. Summary of the Invention

[0003] This invention provides a method for calibrating flood model parameters based on isotope tracing, the method comprising: Determine the set of parameters and prior distributions to be analyzed in the flood model, the set of parameters including parameters related to runoff and parameters related to isotope transport in the flood model; A set of parameter samples is randomly generated according to the prior distribution of the parameters; The sensitivity of each parameter in each set of parameter samples is evaluated by analyzing the relationship between the parameter samples and the corresponding objective function values. Sensitive parameters were selected based on the sensitivity assessment results; Optionally, the step of evaluating the sensitivity of each parameter in each set of parameter samples by analyzing the relationship between the parameter samples and the corresponding objective function values ​​includes: After each sampling, the flood model is run using a randomly generated set of parameter samples to obtain the simulated runoff process and isotope transport process; Calculate the objective function value corresponding to each parameter sample in each set of parameter samples, wherein the objective function value includes the NSE of runoff and the KGE of isotopes; Calculate the correlation coefficient between each parameter in each set of parameter samples and the objective function value, and determine the sensitivity of the parameter based on the correlation coefficient.

[0004] Optionally, determining the sensitivity of the parameter based on the correlation coefficient includes: For NSE of runoff, parameters with an absolute value of correlation coefficient greater than a first predetermined threshold are selected; For KGE of isotopes, parameters with an absolute value of correlation coefficient greater than a second predetermined threshold are selected; Combine all selected parameters for the two objective functions to form the sensitive parameters.

[0005] Optionally, the step of evaluating the sensitivity of each parameter in each set of parameter samples by analyzing the relationship between the parameter samples and the corresponding objective function values ​​includes: After each sampling, the flood model is run using a randomly generated set of parameter samples to obtain the simulated runoff process and isotope transport process; Calculate the objective function value corresponding to each parameter sample in each set of parameter samples, wherein the objective function value includes the NSE of runoff and the KGE of isotopes; Establish a regression model of the objective function with respect to the parameters, and determine the sensitivity of the parameters by the significance and magnitude of the regression coefficients.

[0006] Optionally, establishing a regression model of the objective function with respect to each parameter in each set of parameter samples includes: first standardizing the parameters and response, and then performing multiple linear regression.

[0007] Optionally, the step of evaluating the sensitivity of each parameter in each set of parameter samples by analyzing the relationship between the parameter samples and the corresponding objective function values ​​includes: After each sampling, the flood model is run using a randomly generated set of parameter samples to obtain the simulated runoff process and isotope transport process; Calculate the objective function value corresponding to each parameter sample in each set of parameter samples, wherein the objective function value includes the NSE of runoff and the KGE of isotopes; Calculate the first-order and higher-order sensitivity indices of the objective function value, and determine the sensitivity of the parameters based on the first-order and higher-order sensitivity indices.

[0008] Optionally, the radial basis function (RBF) can be used as an alternative model to find candidate optimal parameters, and the sensitive parameters can be iteratively optimized with the sum of runoff NSE and isotope KGE as the objective function.

[0009] Optionally, the iterative optimization process for the sensitive parameters includes: The RBF model is trained using the parameter-target values ​​of the sample, establishing the mapping relationship between the parameters and the target function, and setting the convergence condition for training. Find candidate optimal parameters on the RBF model, and retain the top 5% of parameter combinations with the highest target values ​​among the candidate points as the selected parameters; The selected parameters are input into the flood model for parallel calculation to obtain accurate NSE and KGE values; The RBF model is updated based on the accurate flood model NSE and KGE values, and the process proceeds to the next iteration. After iteration, Pareto fronts for runoff and isotope targets are constructed based on the individual values ​​of runoff NSE and isotope KGE during the iteration, and parameters in the top 10 percentile of the comprehensive index of the flood model are selected as effective parameter combinations.

[0010] This invention provides a flood model parameter calibration system based on isotope tracing, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the aforementioned flood model parameter calibration method based on isotope tracing.

[0011] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the aforementioned method for calibrating flood model parameters based on isotope tracing.

[0012] The isotope tracing-based flood model parameter calibration method provided by this invention solves the problem of heterogeneous parameter equivalence in flood models, that is, it greatly reduces the heterogeneous parameter equivalence of the model (i.e., the parameter range when the model evaluation index reaches a certain threshold), and at the same time greatly improves the simulation effect of the flood model during the validation period. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0014] Figure 1 This is a schematic diagram of the flood model constructed in this invention.

[0015] Figure 2 This is a flowchart illustrating the method for calibrating flood model parameters based on isotope tracing provided by the present invention.

[0016] Figure 3 A flowchart illustrating the process of using Monte Carlo sampling.

[0017] Figure 4 This is a flowchart illustrating the optimization process for sensitive parameters. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0019] The THREW-T model is a flood model that incorporates isotopes, based on our architecture. For example... Figure 1 As shown, this model is constructed based on topographic data (from high-precision topographic data such as GTOPO30 from the USGS), meteorological data (from global reanalysis datasets such as ERA5, including elements such as temperature, precipitation, wind speed, and radiation)), underlying surface data (from USGS global land cover data (including land use type, vegetation cover, soil type, etc.)), and precipitation isotope data (from measured station data or the Global Isotope Observation Network (GNIP)). Figure 1 As shown, the parameters related to the snowmelt process include three parameters: snowmelt day factor, ice melt day factor, and snowmelt temperature threshold for snow and glacier topography; the parameters related to surface runoff include two parameters: average water storage capacity and shape coefficient for calculating saturated runoff area; the parameters related to groundwater runoff include two parameters: exponential coefficient and linear coefficient; and the parameters related to river confluence include two Muskinggan coefficients. The symbols and definitions of the parameters in the specific model are shown in the table below.

[0020]

[0021] This invention provides a method for calibrating flood model parameters based on isotope tracing, such as... Figure 2 As shown, the method includes: S21. Determine the set of parameters to be analyzed and the prior distribution in the flood model, wherein the set of parameters includes parameters related to runoff and parameters related to isotope transport in the flood model; S22. Randomly generate a set of parameter samples according to the prior distribution of the parameters; S23. By analyzing the relationship between the parameter samples and the corresponding objective function values, the sensitivity of each parameter in each set of parameter samples is evaluated; S24. Based on the sensitivity assessment results, select the sensitive parameters; In the parameter calibration of flood models based on isotope tracing, the purpose of using the Monte Carlo method for parameter sensitivity analysis is to screen out sensitive parameters that have a significant impact on runoff and isotope simulation results from a large number of model parameters. This allows for a focus on these sensitive parameters in subsequent parameter calibration, improving calibration efficiency. The specific implementation process is explained below.

[0022] 1. Determine the model parameters and their prior distributions. First, determine the set of parameters to be analyzed in the flood model. Parameters related to runoff in the flood model include: soil saturated hydraulic conductivity, surface depression water storage capacity, soil layer thickness, porosity, and rainfall interception coefficient. For isotope simulations, this includes parameters related to isotope transport, such as mixing coefficient and fractionation coefficient during evaporation.

[0023] Then, a reasonable prior distribution (such as uniform distribution, normal distribution, etc.) is set for each parameter in each set of parameter samples. Usually, in the absence of prior knowledge, a uniform distribution is used, and its range is determined based on physical meaning or experience.

[0024] 2. Define the objective function (or sensitivity metric). This paper defines indices to evaluate the fit between flood model outputs (runoff and isotopes) and observed data. Currently, commonly used indices for runoff include Nash-Sutcliffe Efficiency (NSE) and Root Mean Squared Error (RMSE). For isotopes, commonly used indices include the root mean square error of isotope values ​​and time-dynamic fitting indices for isotope signals. This invention considers both runoff and isotopes, requiring the construction of a multi-objective function or an algorithm that calculates and combines runoff and isotope indices separately.

[0025] 3. Monte Carlo sampling like Figure 3 As shown, the parameter sampling process includes the following steps: S31. Randomly generate a set of parameter samples according to the prior distribution of the parameters. The number of samplings needs to be large enough (e.g., 10,000 times) to ensure statistical stability.

[0026] S32. For each sampling, a flood model is run using a randomly generated set of parameter samples to obtain simulated runoff and isotope transport processes.

[0027] S33. Calculate the objective function value (e.g., NSE for runoff and KGE for isotopes) for each parameter sample in each set of parameter samples.

[0028] 4. Sensitivity Analysis Based on Monte Carlo sampling, the sensitivity of each parameter in each set of parameter samples is evaluated by analyzing the relationship between the parameter samples and the corresponding objective function values. At least three methods can be used, which are described below.

[0029] 1) Correlation coefficient method: This method calculates the correlation coefficient (such as Pearson correlation coefficient or Spearman rank correlation coefficient) between each parameter in each set of parameter samples and the objective function value. The larger the absolute value of the correlation coefficient, the stronger the sensitivity. In this method, the sensitivity of each parameter in each set of parameter samples is evaluated by analyzing the relationship between the parameter samples and the corresponding objective function values. This includes: after each sampling, running a flood model using a randomly generated set of parameter samples to obtain simulated runoff and isotope transport processes; calculating the objective function value corresponding to each parameter sample in each set of parameter samples, where the objective function value includes the NSE of runoff and the KGE of isotopes; calculating the correlation coefficient between each parameter in each set of parameter samples and the objective function value, and determining the sensitivity of the parameter based on the correlation coefficient.

[0030] As one specific implementation method, a parameter-response matrix can be constructed first, and then the Spearman rank correlation coefficient can be calculated. The specific process is as follows: from scipy.stats import spearmanr # Constructing the parameter-response matrix param_matrix = np.array([r['params'] for r in results]) obj_values ​​= np.array([r['obj_value'] for r in results]) # Calculate the Spearman rank correlation coefficient sensitivity_indices = [ ] for i in range(n_params): rho, p_value = spearmanr(param_matrix[:, i], obj_values) sensitivity_indices.append({ 'param_index': i, 'rho': abs(rho), # Take the absolute value 'p_value': p_value }) In this embodiment, determining the sensitivity of a parameter based on the correlation coefficient includes: for NSE of runoff, selecting a parameter whose absolute value of the correlation coefficient is greater than a first predetermined threshold; for KGE of isotopes, selecting a parameter whose absolute value of the correlation coefficient is greater than a second predetermined threshold; and merging all selected parameters for the two objective functions to form a sensitive parameter.

[0031] 2) Regression analysis method, which involves establishing a regression model of the objective function with respect to the parameters (such as linear regression), and judging the sensitivity by the significance (such as p-value) and magnitude of the regression coefficients. In this method, the sensitivity of each parameter in each set of parameter samples is evaluated by analyzing the relationship between the parameter samples and the corresponding objective function values. This includes: after each sampling, running a flood model using a randomly generated set of parameter samples to obtain simulated runoff and isotope transport processes; calculating the objective function value corresponding to each parameter sample in each set of parameter samples, where the objective function value includes the NSE of runoff and the KGE of isotopes; establishing a regression model of the objective function with respect to the parameters, and judging the sensitivity of the parameters by the significance and magnitude of the regression coefficients.

[0032] As one specific implementation method, the parameters and response are first standardized, and then multiple linear regression is performed. The specific process is as follows: from sklearn.linear_model import LinearRegression # Standardized parameters and response X = (param_matrix - np.mean(param_matrix, axis=0)) / np.std(param_matrix, axis=0) y = (obj_values ​​- np.mean(obj_values)) / np.std(obj_values) # Multiple linear regression model = LinearRegression().fit(X, y) src_coeffs = np.abs(model.coef_) # Take the absolute value 3) Variance-based methods, such as the Sobol index method, can calculate first-order and higher-order sensitivity indices. In this method, the sensitivity of each parameter in each set of parameter samples is evaluated by analyzing the relationship between parameter samples and their corresponding objective function values. This includes: after each sampling, running a flood model using a randomly generated set of parameter samples to obtain simulated runoff and isotope transport processes; calculating the objective function value corresponding to each parameter sample in each set of parameter samples, where the objective function value includes the NSE of runoff and the KGE of isotopes; calculating the first-order and higher-order sensitivity indices of the objective function values, and determining the sensitivity of the parameters based on these indices. However, this method involves a large amount of computation. The specific process is as follows: from SALib.sample import saltelli from SALib.analyze import sobol problem = { 'num_vars': n_params, 'names': ['Param1', ...], 'bounds': [[min1,max1], ...] } # Generate Sobol sequence samples sobol_samples = saltelli.sample(problem, 1024) # ...Run the model to calculate the objective function... # Calculate the first-order sensitivity index Si = sobol.analyze(problem, y_values, calc_second_order=False) Considering computational efficiency, correlation coefficient method or regression analysis method is usually used for preliminary screening.

[0033] 5. Filter sensitive parameters Based on the results of the sensitivity analysis, sensitive parameters were selected for each objective function (runoff and isotope). For the runoff objective function, parameters with an absolute correlation coefficient greater than a certain threshold (e.g., 0.2) were selected; similarly, parameters with an absolute correlation coefficient greater than the threshold were selected for the isotope objective function. All sensitive parameters of the two objective functions were merged to form a sensitive parameter set. This sensitive parameter set was then deduplicated to generate the final sensitive parameter set.

[0034] As a specific implementation, an actual flood model needs to be invoked to run the objective function, returning the NSE from the runoff simulation and the RMSE from the isotope simulation. We want the NSE to be as high as possible, and the RMSE to be as low as possible. To uniformly represent how increasing the parameters improves the objective function using a positive correlation, we can take the negative sign of the RMSE and then analyze the negative RMSE.

[0035] # Runoff Target: Nash Efficiency Coefficient NSE = 1 - np.sum((Q_sim - Q_obs) 2) / np.sum((Q_obs - np.mean(Q_obs)) 2) # Isotope Target: Weighted Root Mean Square Error wRMSE_iso = np.sqrt(np.mean((Iso_sim - Iso_obs) 2)) iso_weight # Overall Goal (The smaller the better) return (1 - NSE) + wRMSE_iso 6. Visual Analysis Plot a scatter plot of the parameters versus the objective function value to visually display the impact of the parameters on the objective function. Alternatively, plot a bar chart of the correlation coefficients to facilitate comparison of the sensitivity of different parameters.

[0036] 7. Parameter Calibration The selected sensitive parameters are used for subsequent parameter optimization (such as using optimization algorithms for automatic calibration), while the non-sensitive parameters are fixed to the default value or average value.

[0037] Through the above process, sensitive parameters that have a significant impact on both runoff and isotope simulations can be effectively identified, providing a basis for parameter selection for subsequent high-dimensional parameter calibration (such as PSO and SCE-UA algorithms), thereby improving calibration efficiency and model reliability.

[0038] Using radial basis functions (RBF) as an alternative model to find candidate optimal parameters can accelerate the model development process. Parameter optimization is performed using the sum of runoff NSE and isotopic KGE (Kling-Gupta Efficiency) as the objective function. A symmetric Latin hypercube design is employed to randomly generate parameter combinations, with 50-100 samples uniformly covering the parameter space being selected for optimal results. KGE comprehensively evaluates the model's isotopic performance in terms of mean, standard deviation, and correlation. An iterative parameter optimization process is then conducted based on the RBF alternative model, the objective function, and the parameter samples. Figure 4 As shown, the iterative optimization process of the sensitive parameters includes: S41. Training the RBF model using the parameter-target values ​​of the sample, establishing the mapping relationship between parameters and the objective function, and setting the convergence condition for training; S42. Finding candidate optimal parameters on the RBF model, and retaining the top 5% parameter combinations with the highest target values ​​among the candidate points as the selected parameters; S43. Inputting the selected parameters into the flood model for parallel computation to obtain accurate NSE and KGE values; S45. Updating the RBF model based on the accurate flood model NSE and KGE values, and entering the next iteration; S46. After the iteration is completed, constructing the Pareto front of the runoff and isotope targets based on the individual values ​​of runoff NSE and isotope KGE during the iteration, and selecting the parameters with the top 10% quantiles of the comprehensive index of the flood model as effective parameter combinations. This invention is not limited to the top 5% parameter combinations with the highest target values ​​among the candidate points, nor is it limited to the parameters with the top 10% quantiles of the comprehensive index of the flood model. The 5% and 10% are preferred values ​​in practice, and the specific values ​​can be flexibly adjusted according to the actual situation. A detailed explanation follows.

[0039] The RBF model is trained using the parameter-target values ​​of the sample, establishing a mapping relationship between parameters and the objective function. The convergence condition is set to stop when the target improvement is less than 1% after 10 consecutive iterations. Candidate optimal parameters are found on the RBF model using a dynamic weighted particle swarm optimization algorithm or the EI (expected improvement) criterion. The top 5% of parameter combinations with the highest target values ​​among the candidate points are retained as the selected parameters. The selected parameters are input for parallel computation to obtain accurate NSE and KGE values. The RBF model is updated based on the accurate flood model NSE and KGE values, and the next iteration begins.

[0040] After iteration, Pareto fronts for runoff and isotope KGE targets are constructed based on the individual values ​​of runoff NSE and KGE during the iterations. Parameters with the top 10% quantiles of the comprehensive index of the flood model are selected as effective parameter combinations. The specific selection process is as follows: All runoff NSE and isotope implementation methods in all iterations are recorded. The comprehensive index is the comprehensive score, which is calculated as 0.6 × NSE + 0.4 × KGE, where 0.6 and 0.4 are the weights of NSE and KGE, respectively, and these weights are adjustable. The comprehensive scores are sorted, and the top individual values ​​of KGE are selected. Then, the NSGA-II algorithm is used to perform a non-dominated sorting of the two targets. Based on the sorting results, a two-dimensional Pareto front plot of NSE-KGE is drawn, and all solution sets distributed on the front are selected. The comprehensive index is calculated based on all solution sets, and the weights of NSE and KGE in all solution sets are adjustable. A parameter combination with a 10% quantile (e.g., selecting the top 100 out of 1000 sets) is selected as an effective parameter.

[0041] The validity of the effective parameters can be verified by Monte Carlo sampling on historical data to check their stability.

[0042] The flood model is run based on effective parameters to assess model uncertainty; the model is driven by publicly available or other sources of future weather forecast data (such as ECMWF weather forecast data or short-term forecast results from meteorological departments) to predict future runoff.

[0043] This invention provides a flood model parameter calibration system based on isotope tracing, the system comprising: The parameter set prior distribution module is used to determine the parameter set and prior distribution to be analyzed in the flood model. The parameter set includes parameters related to runoff and parameters related to isotope transport in the flood model. The parameter sample generation module is used to randomly generate a set of parameter samples according to the prior distribution of the parameters. The parameter sensitivity assessment module is used to assess the sensitivity of each parameter in each set of parameter samples by analyzing the relationship between the parameter samples and the corresponding objective function values. The sensitive parameter filtering module is used to filter out sensitive parameters based on the sensitivity assessment results.

[0044] As one implementation, the parameter sensitivity evaluation module includes: The flood model running submodule is used to run the flood model with a randomly generated set of parameter samples after each sampling to obtain the simulated runoff process and isotope transport process; The objective function value calculation submodule is used to calculate the objective function value corresponding to each parameter sample in each group of parameter samples. The objective function value includes the Nash efficiency coefficient (NSE) of runoff and the root mean square error (RMSE) of isotopes. The correlation coefficient calculation submodule is used to calculate the correlation coefficient between each parameter in each set of parameter samples and the target function value, and to determine the sensitivity of the parameter based on the correlation coefficient.

[0045] The correlation coefficient calculation submodule selects parameters whose absolute correlation coefficient value is greater than a first predetermined threshold for NSE of runoff and parameters whose absolute correlation coefficient value is greater than a second predetermined threshold for KGE of isotopes. It then combines all selected parameters for the two objective functions to form sensitive parameters.

[0046] In a second implementation, the parameter sensitivity assessment module includes: a flood model running submodule, used to run the flood model using a randomly generated set of parameter samples after each sampling to obtain simulated runoff and isotope transport processes; an objective function value calculation submodule, used to calculate the objective function value corresponding to each parameter sample in each set of parameter samples, the objective function value including the NSE of runoff and the KGE of isotopes; and a regression coefficient calculation submodule, used to establish a regression model of the objective function with respect to the parameters, and to determine the sensitivity of the parameters by the significance and magnitude of the regression coefficients.

[0047] The regression coefficient calculation submodule first standardizes the parameters and response, and then performs multiple linear regression.

[0048] As a third implementation, the parameter sensitivity assessment module includes: a flood model running submodule, used to run a flood model using a randomly generated set of parameter samples after each sampling to obtain simulated runoff and isotope transport processes; an objective function value calculation submodule, used to calculate the objective function value corresponding to each parameter sample in each set of parameter samples, the objective function value including the NSE of runoff and the KGE of isotopes; and a first-order and higher-order sensitivity index calculation submodule, used to calculate the first-order and higher-order sensitivity indices of the objective function value, and determine the sensitivity of the parameters based on the first-order and higher-order sensitivity indices.

[0049] The flood model parameter calibration system based on isotope tracing also includes a sensitive parameter optimization module, which uses radial basis function (RBF) as an alternative model to find candidate optimal parameters and performs iterative optimization of sensitive parameters with the sum of runoff NSE and isotope KGE as the objective function.

[0050] The sensitive parameter optimization module performs the following process: It trains the RBF model using the parameter-target values ​​of the sample, establishes the mapping relationship between parameters and the objective function, and sets the convergence condition for training; it searches for candidate optimal parameters on the RBF model, retaining the top 5% parameter combinations with the highest target values ​​among the candidate points as selected parameters; it inputs the selected parameters into the flood model for parallel computation to obtain accurate NSE and KGE values; it updates the RBF model based on the accurate flood model NSE and KGE values, and enters the next iteration; after iteration, it constructs the Pareto front for runoff and isotope targets based on the individual values ​​of runoff NSE and isotope KGE during iteration, and selects the parameters with the top 10% quantiles of the comprehensive index of the flood model as effective parameter combinations.

[0051] The present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the aforementioned method for calibrating flood model parameters based on isotope tracing.

[0052] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the aforementioned method for calibrating flood model parameters based on isotope tracing.

[0053] The isotope tracing-based flood model parameter calibration method provided by this invention solves the problem of heterogeneous parameter equivalence in flood models, that is, it greatly reduces the heterogeneous parameter equivalence of the model (i.e., the parameter range when the model evaluation index reaches a certain threshold), and at the same time greatly improves the simulation effect of the flood model during the validation period.

[0054] 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.

[0055] 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 method for calibrating flood model parameters based on isotope tracing, characterized in that, include: Determine the set of parameters to be analyzed in the flood model and the prior distribution of the parameters in the parameter set, the parameter set including parameters related to runoff and parameters related to isotope transport in the flood model; A set of parameter samples is randomly generated according to the prior distribution of the parameters; The runoff and isotopic processes were obtained by running the THREW-T model, and the correlation coefficient between the objective function values ​​of the Nash efficiency coefficient (NSE) of runoff and the Kling-Gupta efficiency coefficient (KGE) of isotopes was calculated to evaluate the sensitivity of each parameter in each set of parameter samples. The THREW-T model was constructed based on topographic data, meteorological data, underlying surface data, and precipitation isotope data. Sensitive parameters are selected based on the sensitivity evaluation results of each parameter in each set of parameter samples.

2. The method according to claim 1, characterized in that, The evaluation of the sensitivity of each parameter in each set of parameter samples includes: After each sampling, the flood model is run using a randomly generated set of parameter samples to obtain the simulated runoff process and isotope transport process; Calculate the objective function value corresponding to each set of parameter samples. The objective function value includes the Nash efficiency coefficient (NSE) of runoff and the Kling-Gupta efficiency coefficient (KGE) of isotopes. Calculate the correlation coefficient between each parameter in each set of parameter samples and the objective function value, and determine the sensitivity of the parameter based on the correlation coefficient.

3. The method according to claim 2, characterized in that, Determining the sensitivity of the parameters based on the correlation coefficient includes: For NSE of runoff, parameters with an absolute value of correlation coefficient greater than a first predetermined threshold are selected; For KGE of isotopes, parameters with an absolute value of correlation coefficient greater than a second predetermined threshold are selected; All selected parameters for NSE and KGE for runoff are combined to form sensitive parameters.

4. The method according to claim 1, characterized in that, The evaluation of the sensitivity of each parameter in each set of parameter samples includes: After each sampling, the flood model is run using a randomly generated set of parameter samples to obtain the simulated runoff process and isotope transport process; Calculate the objective function value corresponding to each set of parameter samples, wherein the objective function value includes the NSE of runoff and the KGE of isotopes; Establish a regression model of the objective function with respect to each parameter in each set of parameter samples, and determine the sensitivity of the parameters by the significance and magnitude of the regression coefficients.

5. The method according to claim 4, characterized in that: The process of establishing a regression model of the objective function with respect to the parameters includes: first, standardizing the parameters and response, and then performing multiple linear regression.

6. The method according to claim 1, characterized in that: Evaluate the sensitivity of each parameter in each set of parameter samples, including: After each sampling, the flood model is run using a randomly generated set of parameter samples to obtain the simulated runoff process and isotope transport process; Calculate the objective function value corresponding to each parameter sample in each set of parameter samples, wherein the objective function value includes the NSE of runoff and the KGE of isotopes; Calculate the first-order and higher-order sensitivity indices of the objective function value, and determine the sensitivity of the parameters based on the first-order and higher-order sensitivity indices.

7. The method according to claim 1, characterized in that: Radial basis function (RBF) is used as an alternative model to find candidate optimal parameters, and sensitive parameters are iteratively optimized with the sum of runoff NSE and isotopic KGE as the objective function.

8. The method according to claim 7, characterized in that, The iterative optimization process for the sensitive parameters includes: The RBF model is trained using the parameter-target values ​​of the sample, establishing the mapping relationship between the parameters and the target function, and setting the convergence condition for training. Find candidate optimal parameters on the RBF model, and retain the top 5% of parameter combinations with the highest target values ​​among the candidate points as the selected parameters; The selected parameters are input into the flood model for parallel calculation to obtain accurate NSE and KGE values; The RBF model is updated based on the accurate flood model NSE and KGE values, and the process proceeds to the next iteration. After iteration, Pareto fronts for runoff and isotope targets are constructed based on the individual values ​​of runoff NSE and isotope KGE during the iteration, and parameters in the top 10 percentile of the comprehensive index of the flood model are selected as effective parameter combinations.

9. 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 program, it implements the flood model parameter calibration method based on isotope tracing as described in any one of claims 1 to 8.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the flood model parameter calibration method based on isotope tracing as described in any one of claims 1 to 8.

Images