Debris flow model parameter sensitivity analysis and parameter determination method

Abstract

The invention belongs to the technical field of debris flow model parameter optimization, and particularly relates to a debris flow model parameter sensitivity analysis and parameter determination method, which comprises the following steps: performing hydrological analysis by using GDAL, NumPy and SciPy libraries in Python according to collected digital elevation data of a research area; using a GDAL library in the Python to carry out preprocessing on the collected DEM data of the research area; parameters needing sensitivity analysis in the debris flow model are determined as clear water peak flow # imgabs0 #, a gully bed roughness coefficient # imgabs1 #, a gully bed longitudinal slope # imgabs2 #, a friction coefficient # imgabs3 # and a turbulence coefficient # imgabs4 #; according to the method, key parameters in the Massflow model are analyzed and identified through Morris, parameter values are calibrated in combination with a GLUE method, the model simulation precision is remarkably improved, the method is applied to Massflow, debris flow process simulation is conducted, and scientific support is provided for disaster prevention and control.

Description

Technical Field

[0001] The present invention belongs to the technical field of debris flow model parameter optimization, and specifically relates to a method for sensitivity analysis and parameter determination of debris flow model parameters. Background Art

[0002] In recent years, using numerical models to simulate and predict debris flows, especially the Massflow model, has become an important means to study the occurrence, flow, and deposition processes of debris flows. The Massflow model provides a scientific basis for disaster prevention and control planning by accurately simulating the dynamic process of debris flows. However, in the application of the model, the randomness of parameter selection and adjustment still affects the accuracy of simulation results.

[0003] Chinese invention patent CN117057508A discloses a method, device, terminal, and medium for identifying the risk of mountain flood and debris flow disasters. The method includes obtaining basic environmental data within a target area, processing the basic environmental data using the Hec-GeoHMS model, importing it into the HEC-HMS hydrological model for hydrological process simulation, and obtaining the final parameters of sub-basins and river channels; calculating the stability coefficient of the covered slope, determining key indicators that can reflect water source sensitivity, and constructing a mountain flood disaster risk index system; the analysis and determination accuracy of the analysis and determination method for the parameters of this debris flow model is low, and the analysis and determination quality is poor.

[0004] In the above-mentioned technology, common methods for determining model parameters include the empirical method and the manual trial-and-error method.

[0005] The empirical method usually uses parameters from similar studies for prediction, but it is easy to ignore the differences in regional environments and geological conditions, thereby reducing the accuracy of the results.

[0006] The manual trial-and-error method relies on the experience of researchers to optimize the model by gradually adjusting parameters. Although it can be adjusted according to local conditions, it takes a long time and has low efficiency.

[0007] Therefore, in order to improve the applicability and accuracy of the Massflow model in debris flow simulation, combining the Morris screening method and the GLUE method for sensitivity analysis and parameter calibration, thereby effectively reducing the uncertainty in the parameter optimization process, has become an effective way to improve the model accuracy. Summary of the Invention

[0008] The present invention provides a method for sensitivity analysis and parameter determination of debris flow model parameters to solve the problems that the empirical method in the above is easy to ignore the differences in regional environments and geological conditions, thereby reducing the accuracy of the results, while the manual trial-and-error method takes a long time and has low efficiency.

[0009] To achieve the above object, the present invention provides the following technical solutions: A method for sensitivity analysis and parameter determination of a debris flow model, comprising the following steps: Step S1: According to the collected digital elevation data of the research area, use the GDAL library, NumPy library, and SciPy library in Python for hydrological analysis; Step S2: Use the GDAL library in the Python to preprocess the DEM data of the collected research area; Step S3: Determine that the parameters to be subjected to sensitivity analysis in the debris flow model are the peak discharge of clear water , the roughness coefficient of the gully bed , the longitudinal slope of the gully bed , the friction coefficient , the turbulence coefficient ; Step S4: Calculate the results of the Morris screening method analysis under three sampling methods (the initial value of parameter manual calibration, the median initial value of the parameter value range, and the initial value of manual calibration with different perturbation ranges); Step S5: In the Python environment, use the Latin Hypercube Sampling (LHS) method to randomly sample the parameter set to enhance the representativeness and randomness of parameter selection; Step S6: After sampling, use the Python to batch replace the input parameters of the Massflow model, call the execution file to start the simulation, use the GLUE method to statistically analyze the simulation results, gradually narrow the value range of the parameters by calculating the relationship between the effective parameters and the model determination coefficient R², and perform four iterations to ensure the accuracy of the parameters; Step S7: Use the parameter group corresponding to the maximum R² value obtained by the fourth GLUE method in Step 6 to perform the simulation, and verify the model by calculating the relative error and the accuracy factor to ensure the parameter rationality and the model accuracy; Step S8: Based on the verified optimal parameter set in Step 7, use the Massflow to analyze the debris flow activity range and impact in the research area under different rainfall frequencies (P = 20%, P = 2%, P = 1%, and P = 0.5%).

[0010] Preferably, the specific steps of S1 are: Step S11: Use the tool in the GDAL library to fill the depressions in the digital elevation model (DEM), fill the depression points in the raster and remove the peaks, and unify the water flow convergence direction; Step S12: Compare the elevation of each raster with the elevations of its surrounding 8 rasters through the NumPy and SciPy to determine the water flow direction; Step S13: Calculate the number of upstream confluence grids for each grid using the flow accumulation algorithm of NumPy, and finally form a river network.

[0011] Preferably, the specific steps of S2 are as follows: S21: Convert the DEM data into an ASCII format file recognizable by the Massflow software through the GDAL library and import it into the Massflow to complete the model construction; S22: The precipitation data of the study area collected and the debris flow initiation points determined in step S1 will be used as the input of the Massflow model. Set the running time and simulation duration of the model to ensure that the precipitation impact is accurately reflected during the simulation. During the simulation, select the Voellmy model to simulate the movement and deposition process of the debris flow.

[0012] Preferably, the specific steps of S4 are as follows: S41: In the Python environment, perform perturbation sampling with a 2% step size to obtain 14, 18, and 27 groups of parameter samples respectively; S42: Through the Python, call the input file interface of the Massflow model, batch replace different groups of parameter values, and start the model for simulation; S43: Extract the objective functions (such as average flow and maximum flow) from the simulation results, and then substitute them into the formula to calculate the sensitivity index S of each parameter.

[0013] Preferably, the calculation formula for the sensitivity index S of the parameter is: , In the formula, S is the sensitivity index of the parameter; and are respectively the output values of the th and th runs of the model; is the initial calibration result; and are respectively the percentage changes in the th and th perturbed parameters relative to the calibration result; n is the total number of simulations.

[0014] Preferably, for the determination of the sensitivity index S: When |S| ≤ 0.05, the parameter is considered insensitive; when 0.05 < |S| < 0.2, the parameter is considered to have medium sensitivity; when 0.2 < |S| < 1.0, the parameter is considered sensitive; when |S| ≥ 1.0, the parameter is considered very sensitive.

[0015] Preferably, the specific steps of S5 are as follows: S51: Uniformly divide the value range of each parameter into several small intervals; S52: Generate a random value according to the uniform distribution within each small interval; S53: Randomly shuffle the finally generated n parameter samples to form a new set of parameter samples; S54: Generate a random distribution value according to the inverse function of the probability distribution as the input for the next step.

[0016] Preferably, the calculation formula of the coefficient of determination R² is: , Where: is the observed value at the th moment; is the simulated value at the th moment; is the average value of the observation results; After the first iteration, count the feasible parameter sets that meet R²>0.65, and draw the probability distribution and cumulative probability density graph of the parameters to determine the range of the next step parameters; After the second iteration, screen the parameter sets that meet R²>0.7; After the third iteration, screen the parameter sets that meet R²>0.75; After the fourth iteration, screen the parameter sets that meet R²>0.8.

[0017] Preferably, the calculation formula of the relative error is: , In the formula: is the relative error, is the area of the accumulation hazard range obtained from actual measurement and remote sensing image interpretation, is the area of the accumulation hazard obtained from the numerical simulation results.

[0018] Preferably, the calculation formula of the accuracy factor Accuracy is: , In the formula, Accuracy represents the accuracy factor of the numerical simulation; represents the overlapping area of the actual debris flow accumulation hazard range interpreted from high-precision images and the debris flow accumulation hazard range after numerical simulation; represents the debris flow accumulation hazard range interpreted from remote sensing images; represents the debris flow accumulation hazard range after numerical simulation in the study area.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: This method identifies the key parameters in the Massflow model through Morris analysis, calibrates the parameter values by combining the GLUE method, significantly improves the model simulation accuracy, and is further applied to Massflow for debris flow process simulation, providing scientific support for disaster prevention and control. This method takes the Massflow model as the core, combines the Morris method and GLUE analysis technology, and proposes a parameter optimization method for debris flow disaster simulation. Through the improved Morris method, combined with three sampling strategies and dual evaluation criteria, the sensitivity index S is used to identify the key parameters and rank the influence degree of the parameters on the model. This method uses the Latin hypercube sampling method (LHS) to generate parameter sets, iteratively optimizes the parameter values four times by the GLUE method, and combines the R² value to evaluate the results, gradually refining the parameter range. The parameter optimization method for the Massflow model developed by this method makes up for the deficiencies in parameter setting of this model, significantly improves its simulation accuracy and applicability, and provides a scientific basis and technical support for debris flow disaster prevention and control. Description of the Drawings

[0020] Figure 1 is the flowchart of the method for analyzing the parameter sensitivity and determining the parameters of the debris flow model of the present invention; Figure 2 is the schematic diagram of the application example study area of the present invention; Figure 3 is the Morris sensitivity analysis result diagram of five parameters of the present invention under different sampling methods; Figure 4 is the parameter of the present invention when the rainfall intensity is P = 20%, P = 2%, P = 0.5% , , , Morris sensitivity analysis result diagram under different sampling methods; Figure 5 is the probability distribution and cumulative probability distribution diagram between the effective parameter values and the determination coefficient R² of the present invention based on the GLUE method when the rainfall frequency is P = 5%; Figure 6 is the change and distribution characteristic diagram of the R² value in four iterations of the present invention based on the GLUE method; Figure 7 is the schematic diagram of the calculation accuracy factor when the present invention conducts model verification; Figure 8 is the optimal parameter group based on sensitivity analysis and parameter calibration of the present invention. Using the Massflow model, the activity range and influence diagram of debris flow when the rainfall frequency is P = 20% in Huangyanggou are analyzed; Figure 9is the optimal parameter group based on sensitivity analysis and parameter calibration of the present invention. The Massflow model is used to analyze the activity range and influence diagram of debris flow in Huangyanggou when the rainfall frequency is P = 2%; Figure 10 is the optimal parameter group based on sensitivity analysis and parameter calibration of the present invention. The Massflow model is used to analyze the activity range and influence diagram of debris flow in Huangyanggou when the rainfall frequency is P = 1%; Figure 11 is the optimal parameter group based on sensitivity analysis and parameter calibration of the present invention. The Massflow model is used to analyze the activity range and influence diagram of debris flow in Huangyanggou when the rainfall frequency is P = 0.5%. Specific embodiments

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0022] Debris flow is a common geological disaster in mountainous areas. With the increase of extreme climate events and the influence of human activities, the occurrence frequency of debris flow has increased significantly, causing serious impacts on social economy and infrastructure. Traditional debris flow research methods analyze factors such as water source, material source and flow conditions, and conduct statistical evaluation and dynamic analysis in combination with regional survey data. Although these methods have high credibility, they are often time-consuming and cumbersome to implement.

[0023] The present invention selects Huangyanggou as the research area. Huangyanggou is located in the Longxi River Basin of Longchi Town, Dujiangyan. This basin belongs to the mid-subtropical humid monsoon climate zone. Before the "5.12" earthquake in 2008, no debris flow had ever occurred in Huangyanggou. After the "5.12" earthquake, the exposed rock masses on both sides of the gully slid down along the weak surface along the slope due to the influence of seismic movement, resulting in landslides. A large number of landslide and collapse loose accumulation bodies are distributed in the gully and at the foot of the slope. After a long period of heavy rainfall on August 13, 2010, two debris flows occurred in Huangyanggou successively. The heavy rainfall met the water source required for the initiation of debris flow, and at the same time, a large amount of sufficient material sources were accumulated on both sides of the gully. The surface runoff formed by the rain erosion carried the material sources in the gully and flowed down along the slope to form debris flow.

[0024] The Huangyanggou watershed has an area of 0.88 km², a longitudinal gradient of the gully bed of 521‰, an altitude ranging from 750 m to 940 m, medium vegetation coverage, steep slopes, most of which are greater than 45°. The rainfall data selected are the rainfall at 1-hour intervals from August 12th to August 14th, 2010 obtained from the Longchi meteorological station. The elevation data are the 12.5 m DEM collected by the ALOS satellite. The data are processed in each step by combining GIS software and programming software.

[0025] Now in combination with Figures 1 to 11 as shown, introduce the Massflow debris flow simulation method based on Morris analysis and GLUE parameter calibration, including the following steps: Step 1: According to the collected digital elevation data of the study area, use the GDAL library, NumPy library, and SciPy library in Python for hydrological analysis.

[0026] The content of hydrological analysis includes: depression filling, flow direction calculation, flow rate calculation, and river network generation.

[0027] The specific steps of S1 are: Step S11: Use the tools in the GDAL library to perform depression filling on the digital elevation model (DEM), fill the depression points in the raster and remove the peaks, unify the water flow convergence direction, and improve the accuracy of the analysis.

[0028] The GDAL library is an open-source library for reading and writing raster geospatial data formats. In hydrological analysis, the GDAL library can be used to process remote sensing images and extract water body information. For example, by calculating the normalized difference water index (NDWI), the water body area can be identified and water body information can be extracted, providing basic data for subsequent hydrological analysis.

[0029] Step S12: Compare the elevation of each raster with the elevations of its surrounding 8 rasters through NumPy and SciPy to determine the water flow direction.

[0030] NumPy is the core library for scientific computing in Python, providing powerful multi-dimensional array objects and rich mathematical functions. In hydrological analysis, NumPy can be used to process and analyze raster data, such as calculating water body indices, generating mask images, etc.; by setting thresholds, the pixels in the NDWI image can be classified as water bodies and non-water bodies, thus generating a water body mask image; SciPy is a scientific computing library built on top of NumPy, providing a wide range of numerical calculation methods and data structures. In hydrological analysis, SciPy can be used for more complex numerical analysis, such as interpolation, optimization, and signal processing. For example, the interpolation function of SciPy can be used to fill in the missing values in the data; through interpolation, a more complete water body distribution map can be generated, providing more accurate data for subsequent hydrological analysis.

[0031] Step S13: Use the flow accumulation algorithm of NumPy to calculate the number of upstream confluence grids for each grid, finally form a river network, and based on the generated river network results and combined with the consulted data, determine the starting point location of the debris flow.

[0032] Step 2: Use the GDAL library in Python to preprocess the DEM data of the collected study area.

[0033] The specific steps of S2 are as follows: S21: Convert the DEM data into an ASCII format file recognizable by the Massflow software through the GDAL library and import it into Massflow to complete the model construction.

[0034] The Massflow software is a high-performance surface disaster dynamic process simulation software, mainly used to simulate the dynamic evolution processes of geological disasters such as landslides, debris flows, debris flows, flash floods, avalanches, and barrier lakes, as well as their disaster chains. In addition, it is also applicable to numerical simulation work of a series of disaster problems such as mountainous watershed hydrological calculations, tailings pond dam failures, and fluid-structure coupling. By using this software, the spatio-temporal evolution process of geological disasters can be revealed in real time, providing theoretical and technical support for quantitative risk assessment of geological disasters, infrastructure and urban planning layout, and emergency disaster reduction and relief strategy formulation.

[0035] S22: The collected precipitation data of the study area and the debris flow starting point determined in Step 1 will be used as the input of the Massflow model. Set the running time and simulation duration of the model to ensure that the precipitation impact is accurately reflected during the simulation. During the simulation, select the Voellmy model to simulate the movement and deposition process of the debris flow. The Voellmy model is a fluid friction model used to describe the movement of debris flows (such as avalanches, debris flows, etc.) and is widely used in the numerical simulation of geological disasters. Among them, the Voellmy model divides the frictional resistance into two parts: the dry friction term: the dry friction coefficient (μ) related to the normal stress, similar to the Coulomb friction law; the turbulent term: the viscosity / turbulence coefficient (ξ) related to the square of the velocity, used to describe the turbulent characteristics of the fluid.

[0036] Step 3: Consult the literature and relevant data to determine that the parameters that need to be subjected to sensitivity analysis in the debris flow model are the peak clear water flow rate ( ), the channel bed roughness coefficient ( ), the channel bed longitudinal slope ( ), the friction coefficient ( ), the turbulence coefficient ( ), etc.

[0037] The specific parameter value ranges are shown in Table 1

[0038] Step 4: Calculate the results of the Morris screening method analysis under three sampling methods (the initial value of parameter manual calibration, the median initial value of the parameter value range, and the initial value of manual calibration with different perturbation ranges).

[0039] The specific steps of S4 are as follows: S41: In the Python environment, perform perturbation sampling with a 2% step size to obtain 14, 18, and 27 groups of parameter samples respectively.

[0040] S42: Through Python, call the input file interface of the Massflow model, batch replace different groups of parameter values, and start the model for simulation.

[0041] S43: Extract the objective functions (such as average flow and maximum flow) from the simulation results, and then substitute them into the formula to calculate the sensitivity index S of each parameter.

[0042] The Morris screening method is a global sensitivity analysis method, mainly used to evaluate the influence degree of model input parameters on the output results, especially suitable for the situation where the number of input parameters is large and the model calculation cost is high; the Morris method randomly perturbs each input parameter within its value range in the way of "changing one parameter at a time", calculates the change amount of the output result, and through statistical analysis of the parameter values of multiple random samplings, calculates the average value (μ) and standard deviation (σ) of each parameter, so as to evaluate the sensitivity of the parameter. μ: represents the average influence degree of the parameter on the output, and the larger the value, the more sensitive the parameter; σ: represents the strength of the interaction between parameters, and the larger the value, the more complex the interaction between parameters.

[0043] Among them, the calculation formula of the sensitivity index S of the parameter is: , In the formula, S is the parameter sensitivity index; and are respectively the th and th running output values of the model; is the initial calibration result; and are respectively the th and th perturbation parameter change percentages relative to the calibration result; n is the total number of simulations.

[0044] Determination of the sensitivity index S: When the value of |S| ≤ 0.05, the parameter is considered insensitive; when 0.05 < |S| < 0.2, the parameter is considered to have medium sensitivity; when 0.2 < |S| < 1.0, the parameter is considered sensitive; when |S| ≥ 1.0, the parameter is considered very sensitive. According to the calculation and comparison of the sensitivity index S, the accuracy and reliability of the data can be obtained quickly.

[0045] Step 5: In the Python environment, use the Latin Hypercube Sampling (LHS) method to randomly sample the parameter set to enhance the representativeness and randomness of parameter selection.

[0046] Latin Hypercube Sampling (LHS) is a multi-dimensional statistical sampling method widely used in fields such as uncertainty analysis, Monte Carlo simulation, and computer experiment design. Its basic principle is: Stratified sampling: Divide the value range of each parameter into several intervals with equal probability, and randomly select a sample point from each interval; Random combination: Randomly combine the sample points of different parameters to form a multi-dimensional sample space; Uniform coverage: Through stratified sampling, ensure that the value range of each parameter is evenly covered, thereby improving the representativeness of the sample.

[0047] The specific steps of S5 are as follows: S51: Evenly divide the value range of each parameter into several small intervals.

[0048] S52: In each small interval, generate a random value according to the uniform distribution.

[0049] S53: Randomly shuffle the finally generated n parameter samples to form a new parameter sample set.

[0050] S54: Generate a random distribution value according to the inverse function of the probability distribution as the input for the next step.

[0051] Step 6: After completing the sampling, use Python to batch replace the input parameters of the Massflow model and call the execution file to start the simulation. Subsequently, use the GLUE method to perform statistical analysis on the simulation results. By calculating the relationship between the effective parameters and the model determination coefficient R², gradually narrow the value range of the parameters and perform four iterations to ensure the accuracy of the parameters.

[0052] The GLUE method is a technique for model uncertainty assessment and parameter estimation, widely used in the fields of hydrology, environmental science, and engineering. The GLUE method is an uncertainty analysis framework based on Bayesian ideas, used to evaluate the uncertainty of model parameters and prediction results. It is achieved through the following steps: Parameter sampling: Randomly draw a large number of parameter combinations from the prior distribution of parameters; Model run: Run the model using each set of parameters to generate model outputs; Likelihood evaluation: Calculate the likelihood value of each parameter combination according to the fitting degree between the model output and the observed data; Weight assignment: Weight the parameter combinations according to the likelihood values to generate the posterior distribution; Uncertainty analysis: Based on the posterior distribution, evaluate the uncertainty of model predictions.

[0053] Among them, the calculation formula for the coefficient of determination R² is: , Where: is the observed value at the moment; is the simulated value at the moment; is the average value of the observed results.

[0054] After normalization, it is considered that there is no correlation between the two when the coefficient of determination R² value is less than 0. The closer the coefficient of determination R² value is to 1, the better the agreement between the measured value and the simulated value.

[0055] The iterative process is as follows: After the first iteration, count the feasible parameter sets that meet R²>0.65, and draw the probability distribution and cumulative probability density diagrams of the parameters to determine the range of the next-step parameters.

[0056] After the second iteration, screen the parameter sets that meet R²>0.7.

[0057] After the third iteration, screen the parameter sets that meet R²>0.75.

[0058] After the fourth iteration, screen the parameter sets that meet R²>0.8.

[0059] After each iteration, update the parameter range and draw the probability distribution and cumulative probability density diagrams, gradually narrowing the parameter range.

[0060] Step 7: Use the parameter group corresponding to the maximum R² value obtained by the fourth GLUE method in Step 6 for simulation, and verify the model through the relative error and calculation accuracy factor methods to ensure the parameter rationality and model accuracy.

[0061] The relative error method is used for verification by comparing the area of the accumulation hazard range obtained from the actual measurement and remote sensing image interpretation after the "8.13" debris flow outbreak with the hazard area obtained from the numerical simulation results.

[0062] Among them, the relative error is calculated by the formula: , In the formula: is the relative error, is the area of the accumulation hazard range obtained from actual measurement and remote sensing image interpretation, is the area of the accumulation hazard obtained from the numerical simulation results.

[0063] The accuracy factor method uses three types of data, namely, the accumulation hazard range of the "8.13" rainstorm-induced Huangyanggou debris flow obtained from numerical simulation, the actual debris flow accumulation hazard range interpreted from high-precision images, and the hazard range where the simulation results coincide with the image interpretation results, for geometric operations.

[0064] Among them, the calculation formula of the accuracy factor Accuracy is: , In the formula, Accuracy represents the accuracy of the numerical simulation; represents the overlapping area between the actual debris flow accumulation hazard range interpreted from high-precision images and the accumulation hazard range after debris flow numerical simulation; represents the debris flow accumulation hazard range interpreted from remote sensing images; represents the accumulation hazard range after debris flow numerical simulation in the study area.

[0065] Step 8: Based on the verified optimal parameter set in Step 7, use Massflow to analyze the debris flow activity range and impact in the study area under different rainfall frequencies (P = 20%, P = 2%, P = 1%, and P = 0.5%).

[0066] In this embodiment, Figure 3 shows the sensitivity analysis results of the Morris screening method for five parameters under different sampling methods, revealing the influence and importance of different parameters in mud depth prediction.

[0067] Under different objective functions, the sensitivity results of the parameters are also different. When the objective function is the average mud depth, the order of the influence degree of the parameters on the model is ; when the objective function is the maximum mud depth, the order of the influence degree of the parameters on the model is .

[0068] Numerically, except that the friction coefficient is 0.042 when the sampling method is the same range of rated values for the maximum mud depth objective function, the sensitivity indices of other parameters are all above 0.2.

[0069] Regardless of whether the objective function is the average mud depth or the maximum mud depth, The impact on the model is the most significant. When the objective function is the maximum mud depth and the sampling method is the same range of median values, has a sensitivity index of 0.5, and the sensitivity indices in other cases are all above 0.8.

[0070] Regardless of whether the objective function is the average mud depth or the maximum mud depth, has the weakest impact on the model simulation results, and regardless of which sampling method is used, the fluctuations are very small, belonging to secondary factors. Therefore, other parameters are focused on determining during the simulation process.

[0071] The clear water flow rate is particularly important for the model simulation results, and it is necessary to explore the influence degree of other parameters on the model simulation results.

[0072] Figure 4 Shows the sensitivity analysis results of parameters , , , under different sampling methods when the rainfall intensity is P = 20%, P = 2%, P = 0.5%. When the objective function is the average mud depth ( Figure 4 a, c, e), generally speaking, the sensitivity of different parameter factors to the average mud depth is in the following order: , which is consistent with the results of the sensitivity analysis of all previous parameters.

[0073] And as the rainfall intensity increases, the sensitivity indices of parameter and parameter are much larger than those of parameter ξ and parameter . Especially when the rainfall intensity P = 0.5%, the average sensitivity indices of parameter and parameter are 0.49 and 0.446 respectively, indicating that these two parameters play a dominant role and are the main factors when considering the average mud depth.

[0074] While the average sensitivity indices of parameter and parameter are only 0.289 and 0.242, with a large difference, so they are secondary factors.

[0075] When the objective function is the maximum mud depth ( Figure 4 b, d, f), the stability of the sensitivity index ranking of the parameters is poor.

[0076] At P = 20%, the influence degree of the friction coefficient on the model is the largest among all parameters, and the sensitivity index is close to 0.6, indicating that when the rainfall intensity is small and the maximum mud depth is considered, the value of parameter needs to be focused on.

[0077] With the increase in rainfall intensity (P = 2%, P = 0.5%), the sensitivity of the influence of different parameter factors on the maximum mud depth tends to be consistent again, in the order of: , which is also consistent with the previous results.

[0078] Figure 5 shows the probability distribution and cumulative probability distribution between the effective parameter values and the determination coefficient under the condition of a rainfall frequency of once in 20 years.

[0079] Statistical analysis was performed on the feasible parameter groups whose first iteration results satisfied the probability objective function ( > 0.65). The probability distribution and cumulative probability distribution of the four parameters are as shown in Figure 5 (a).

[0080] Overall, the differences between the posterior distributions and prior distributions of the four parameters are relatively obvious. Under the condition that the cumulative probability greater than zero approaches 0.9, the probability density of parameter reaches its peak between 17 and 19, and the range shrinks from (14 - 26) to (14 - 23.5). The probability distribution of parameter

[0081] is relatively concentrated, mainly distributed in the range of 100 to 300, and the range shrinks from (100 - 400) to (100 - 348). The probability distribution of parameter

[0082] shows a left-skewed trend, and the parameter range shrinks from (0.1 - 0.3) to (0.11 - 0.25). The probability distribution of parameter

[0083] is mainly concentrated in the middle, and the parameter range shrinks from (0.3 - 0.8) to (0.32 - 0.7). The probability distribution of parameter

[0084] Latin hypercube sampling was performed on the parameter range after the first reduction, and the above steps were repeated. Statistical analysis was performed on the feasible parameter groups whose second iteration results satisfied the probability objective function ( > 0.7), third iteration results satisfied the probability objective function ( > 0.75), and fourth iteration results satisfied the probability objective function ( > 0.8). The statistical results are as shown in Figure 5 (b)(c)(d), gradually narrowing the value range.

[0085] The range of parameter shrinks from (14 - 23.5) to (17.59 - 21.34).

[0086] Parameter The range is reduced from (100 - 348) to (173 - 269).

[0087] Parameter The range is reduced from (0.11 - 0.25) to (0.171 - 0.222).

[0088] Parameter The value range is reduced from (0.32, 0.7) to (0.485, 0.625).

[0089] Figure 6 Shows The variation and distribution characteristics of the value in four iterations.

[0090] Generally speaking The value is distributed between 0.40 - 0.97. As the number of iterations increases, the median of the box plot gradually rises.

[0091] The median of the value rises from 0.65 in the first time to 0.85 in the fourth time. The largest increase appears in the second time, increasing from 0.65 to 0.75, with an increase of 15%.

[0092] For the four iterations The value distribution ranges are between 0.40 - 0.82, 0.52 - 0.87, 0.63 - 0.86 and 0.65 - 0.97 respectively.

[0093] The maximum value appears in the fourth iteration, and the maximum value is 0.97.

[0094] Figure 7 It is a schematic diagram of the calculation accuracy factor. Geometric operations are performed on three types of data: the accumulation hazard range of the Huangyanggou debris flow induced by the "8.13" rainstorm obtained by numerical simulation, the actual accumulation hazard range of the debris flow interpreted by high-precision images, and the hazard range where the simulation results coincide with the image interpretation results, to obtain the accuracy factor of the debris flow numerical simulation using this model.

[0095] The accuracy factor can accurately reflect the results of numerical simulation. See the following formula for details

[0096] In the formula, Accuracy represents the accuracy of numerical simulation; represents the overlapping area between the actual accumulation hazard range of the debris flow interpreted by high-precision images and the accumulation hazard range after debris flow numerical simulation; represents the accumulation hazard range of the debris flow interpreted by remote sensing images; It represents the accumulation hazard range after the numerical simulation of debris flow in the study area. The value range of Accuracy is between 0 and 1. The closer this value is to 1, the more accurate the final numerical simulation result is.

[0097] The results are shown in Table 2

[0098] The simulation accuracy reaches 76.22%, indicating that the parameter group determined by four iterations based on the GLUE method can accurately simulate debris flow disasters.

[0099] Figures 8 to 11 It is the optimal parameter group based on the above sensitivity analysis and parameter calibration. Using the Massflow model, the activity range and impact of debris flow in Huangyanggou are analyzed when the rainfall frequencies are P = 20%, P = 2%, P = 2% and P = 0.5% respectively.

[0100] The above results prove that the present invention based on the Morris sensitivity analysis method can help identify the parameters that have a greater impact on the debris flow simulation results. Based on the GLUE method, the parameter values are iterated four times to gradually narrow the parameter value range, filling the gap in parameter optimization of the debris flow numerical model Massflow. It not only improves the accuracy and applicability of the model, but also provides a more scientific decision-making basis for the prevention and control of debris flow disasters.

[0101] The implementation of this method will effectively promote the research and management of debris flow disasters and provide important references for researchers and policymakers in related fields.

[0102] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0103] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A debris flow model parameter sensitivity analysis and parameter determination method, characterized in that: The following steps are involved: Step S1: Based on the collected digital elevation data of the study area, hydrological analysis is performed using the GDAL library, NumPy library, and SciPy library in Python; Step S2: preprocessing the collected DEM data of the study area using the GDAL library in Python; Step S3: Determine the parameter in the debris flow model that needs to be subjected to sensitivity analysis as the peak flow rate of clean water ( )、Groove bed roughness coefficient( )、Ditch bed longitudinal slope( )、Friction coefficient( )、turbulence coefficient( ); Step S4: Calculate the results of Morris screening analysis under three sampling modes (artificial calibration initial value of the parameter, initial value of the median of the parameter value range, and artificial calibration initial value of different disturbance ranges); Step S5: In the Python environment, a Latin hypercube sampling (LHS) method is used to randomly sample the parameter set to enhance the representativeness and randomness of parameter selection; Step S6: After the sampling is completed, the input parameters of the Massflow model are replaced in batches using the Python, and the execution file is called to start the simulation. The simulation results are statistically analyzed using the GLUE method. The range of the parameter values ​​is gradually narrowed by calculating the relationship between the effective parameters and the model determination coefficient R², and four iterations are performed to ensure the accuracy of the parameters. Step S7: Use the parameter group corresponding to the maximum R² value obtained by the fourth GLUE method in step 6 to perform simulation, and verify the model by calculating the relative error and precision factor to ensure the rationality of the parameters and the accuracy of the model; Step S8: Based on the verified optimal parameter set in step 7, the Massflow is used to analyze the range and impact of debris flow activities in the study area under different rainfall frequencies (P = 20%, P = 2%, P = 1% and P = 0.5%).

2. The method for debris flow model parameter sensitivity analysis and parameter determination according to claim 1, characterized in that: The specific steps of S1 are: Step S11: Use the tools in the GDAL library to perform depression filling processing on the digital elevation model (DEM), fill the depressions in the grid and remove the peaks, and unify the direction of water flow convergence; Step S12: using NumPy and SciPy to compare the elevation of each grid with the eight surrounding grids to determine the direction of water flow; Step S13: Calculate the number of upstream confluence grids of each grid using the flow accumulation algorithm of NumPy, and finally form a river network.

3. The method for debris flow model parameter sensitivity analysis and parameter determination according to claim 1, characterized in that: The specific steps of S2 are: S21: converting the DEM data into an ASCII format file recognizable by the Massflow software through the GDAL library and importing the file into the Massflow to complete the model construction; S22: The collected precipitation data of the study area and the debris flow starting point determined in step S1 will be used as the input of the Massflow model. The running time and simulation duration of the model are set to ensure that the impact of precipitation is accurately reflected during the simulation process. During the simulation process, the Voellmy model is selected to simulate the movement and accumulation process of the debris flow.

4. The method for debris flow model parameter sensitivity analysis and parameter determination according to claim 1, characterized in that: The specific steps of S4 are: S41: In the Python environment, a 2% step size is used for perturbation sampling to obtain 14, 18, and 27 groups of parameter samples respectively; S42: calling the input file interface of the Massflow model through the Python, replacing parameter values ​​of different groups in batches, and starting the model for simulation; S43: Extract the objective function (such as average flow and maximum flow) in the simulation results, and then substitute it into the formula to calculate the sensitivity index S of each parameter.

5. The method for debris flow model parameter sensitivity analysis and parameter determination according to claim 4 is characterized in that: The calculation formula of the sensitivity index S of the parameter is: ， Where S is the sensitivity index of the parameter; and The model Second and Output value of the run; is the initial calibration result; and Respectively, they are relative to the calibration results. Second and The percentage of perturbation parameter change; n is the total number of simulations.

6. The method for debris flow model parameter sensitivity analysis and parameter determination according to claim 5, characterized in that: For the determination of the sensitivity index S: When the |S| value is ≤0.05, the parameter is considered to be insensitive; when 0.05<|S|<0.2, the parameter is considered to have moderate sensitivity; when 0.2<|S|<1.0, the parameter is considered to be sensitive; and when |S|≥1.0, the parameter is considered to be very sensitive.

7. The method for debris flow model parameter sensitivity analysis and parameter determination according to claim 1, characterized in that: The specific steps of S5 are: S51: evenly divide the value range of each parameter into several small intervals; S52: In each small interval, a random value is generated according to a uniform distribution; S53: Randomly shuffle the n parameter samples finally generated to form a new parameter sample set; S54: Generate a random distribution value according to the inverse function of the probability distribution as input for the next step.

8. The method for debris flow model parameter sensitivity analysis and parameter determination according to claim 1, characterized in that: The calculation formula of the determination coefficient R² is: ， in: For the Observation value at time; For the Simulation value at the moment; is the average value of the observed results; After the first iteration, the feasible parameter sets that meet R²>0.65 are counted, and the probability distribution and cumulative probability density diagram of the parameters are plotted to determine the range of the next parameter; After the second iteration, the parameter set that meets R²>0.7 is screened; After the third iteration, the parameter set that meets R²>0.75 is screened; After the fourth iteration, the parameter set that meets R²>0.8 is screened.

9. The method for debris flow model parameter sensitivity analysis and parameter determination according to claim 1, characterized in that: The relative error The calculation formula is: ， Where: is the relative error, The area of ​​accumulation danger obtained by actual measurement and remote sensing image interpretation. is the accumulation hazard area obtained from the numerical simulation results.

10. The method for debris flow model parameter sensitivity analysis and parameter determination according to claim 1, characterized in that: The calculation formula of the accuracy factor is: ， Where Accuracy represents the precision factor of numerical simulation; Represents the overlapping area of ​​the actual debris flow accumulation danger range interpreted by high-precision images and the accumulation danger range after debris flow numerical simulation; Represents the debris flow accumulation hazard range interpreted from remote sensing images; Represents the accumulation hazard range after numerical simulation of debris flow in the study area.

Images