Method and device for quantifying parameter uncertainty and probabilistic prediction of distributed hydrological model

Abstract

The present application belongs to the technical field of hydrological forecasting and mountain torrent disaster early warning, and relates to a distributed hydrological model parameter uncertainty quantification and probability forecasting method and device, comprising: identifying sensitive parameters which have significant influence on model output, and then calculating statistical acceptable set and corresponding statistical likelihood weight through Monte Carlo sampling and likelihood function; based on basin runoff, giving expected interval of excess seepage runoff proportion under different underlying surface categories and antecedent moisture state combinations, thereby constructing runoff mechanism prior discriminant matrix; inversely calculating simulated excess seepage runoff proportion for each parameter combination in the statistical acceptable set, comparing with the expected interval to obtain deviation and calculating mechanism consistency score, so as to screen parameter combinations; fusing the statistical likelihood weight and the mechanism consistency score to obtain comprehensive weight, and generating flood probability forecasting interval by using the comprehensive weight; and according to proportion of mechanism incompatible parameter combinations, carrying out parameter prior interval review feedback.

Description

Technical Field

[0001] This invention belongs to the field of hydrological forecasting and flash flood disaster early warning technology, and relates to a method and device for quantifying the uncertainty of distributed hydrological model parameters and predicting probabilistic forecasts. Background Technology

[0002] The parameters of distributed hydrological models often lack direct observation methods and need to be calibrated through measured flood events. The resulting parameter uncertainty is a significant factor limiting the accuracy and reliability of flood forecasts. Currently, the most widely used method for quantifying parameter uncertainty is the statistical framework represented by Generalized Likelihood Uncertainty Estimation (GLUE). This method uses Monte Carlo methods to randomly sample a large number of parameters within the prior interval. A certain likelihood function, such as the Nash efficiency coefficient (NSE), is used to evaluate the goodness of fit between the simulation results and the measured process for each set of parameters. Parameter combinations with likelihood values ​​reaching a preset threshold are included in the "behavioral parameter set," and the probability distribution or confidence interval of the forecast results is obtained by weighting the likelihood values.

[0003] The above statistical framework has the following shortcomings in practice:

[0004] First, the selection of behavioral parameter sets by methods such as GLUE relies entirely on the overall goodness-of-fit index at the output end, which is a "black box" test and does not test whether the parameter values ​​themselves conform to the physical laws of watershed runoff generation. Because hydrological models generally exhibit the phenomenon of equifinality, that is, multiple sets of parameter values ​​have significant differences but can all achieve a similar fitting accuracy between the simulated output and the measured process, the behavioral parameter set may contain some statistically acceptable parameter combinations that are unreasonable in internal runoff generation mechanisms.

[0005] Second, once the aforementioned parameter combinations are weighted according to likelihood values ​​and used in the generation of the probability forecast interval, it will unduly expand the uncertainty interval of the forecast and dilute the weight that parameter combinations that truly conform to the physical mechanism should have, thereby reducing the physical reliability of the probability forecast results. This problem is particularly prominent in the scenario of flash flood disaster early warning in mountainous small watersheds: Firstly, the lead time is usually only 0.5 to 3 hours, and the forecast system is highly dependent on the ability to extrapolate out-of-sample (combinations of rainfall intensity and previous water content not covered by historical calibration events) rainstorm scenarios. The risk of distortion of parameter combinations with unreasonable mechanisms under extrapolation conditions is significantly higher than that under conventional conditions; secondly, there is a lack of measured data in mountainous small watersheds, and the number of flood events that can be used for calibration and cross-validation is limited. It is difficult to fully constrain the parameter space by relying solely on statistical likelihood functions.

[0006] In summary, existing methods for quantifying parameter uncertainty generally lack mechanistic verification methods, and an effective mutual constraint relationship has not been established between statistical screening results and the physical laws of watershed runoff generation. Summary of the Invention

[0007] To address the aforementioned technical problems in the existing technology, this invention proposes a method and apparatus for quantifying the uncertainty of parameters in a distributed hydrological model and for probabilistic prediction. The specific technical solution is as follows:

[0008] A method for quantifying the uncertainty of parameters in a distributed hydrological model and for probabilistic prediction includes the following steps:

[0009] S1. Identify sensitive parameters that have a significant impact on the model output;

[0010] S2. Perform Monte Carlo sampling and likelihood function calculation on the sensitive parameters to obtain the statistically acceptable set and the corresponding statistical likelihood weights;

[0011] S3. Based on watershed runoff generation, obtain the expected range of the proportion of excess infiltration runoff under different combinations of underlying surface type and previous water content state, and thus construct a priori discrimination matrix for runoff generation mechanism;

[0012] S4. For each parameter combination in the statistically acceptable set, back-calculate the simulated over-permeability flow ratio, compare it with the expected range to obtain the deviation degree, and calculate the consistency score to screen parameter combinations.

[0013] S5. The statistical likelihood weights and the mechanism consistency scores are combined to obtain a comprehensive weight, and a flood probability forecast interval is generated.

[0014] S6. Based on the proportion of incompatible parameter combinations obtained from S4, generate parameter prior interval verification feedback.

[0015] Further, S1 specifically involves selecting five parameters as parameters to be calibrated: evapotranspiration capacity conversion factor, surface soil free water storage capacity, interflow recession coefficient, infiltration capacity parameter, and excess infiltration runoff area ratio parameter. A Monte Carlo method is used to randomly sample within the prior intervals of each parameter to generate a parameter combination sample set, which is then substituted into the distributed hydrological model to obtain the output indicators of total runoff, peak flow, surface runoff, interflow, and groundwater runoff. The standardized regression coefficient method is used to calculate the correlation strength between each parameter and each output indicator, and a subset of sensitive parameters that significantly affect the model output is identified based on set criteria.

[0016] Further, S2 specifically involves: for sensitive parameters, conducting no less than 5,000 to 10,000 independent Monte Carlo samplings within the prior distribution range of the sensitive parameters, and substituting each group into the distributed hydrological model to simulate historical flood events in the target watershed; using the Nash efficiency coefficient NSE as the likelihood function, setting an acceptable threshold NSE0, including parameter combinations with NSE≥NSE0 into the statistically acceptable set, and determining the statistical likelihood weight of each parameter combination based on the proportion of the likelihood function value of each parameter combination in the statistically acceptable set to the sum of the likelihood function values ​​of all parameter combinations in the set.

[0017] Furthermore, step S3 specifically includes the following sub-steps:

[0018] S31. Underlying surface classification: Based on the watershed lithology, soil type, and topography, the target watershed and its similar watersheds are divided into several underlying surface categories;

[0019] S32. Obtain runoff control characteristics: For each type of underlying surface, obtain one or more control variables that can characterize the relative runoff generation capacity of that type of underlying surface;

[0020] S33. Accumulate runoff generation mechanism samples: By statistically analyzing historical measured flood events or by performing forward modeling of historical or synthetic rainfall events using physical mechanism models, obtain samples of the proportion of excess infiltration runoff corresponding to different underlying surface categories under different anterior soil moisture content states.

[0021] S34. Statistical expectation intervals by state grouping: Divide the previous soil moisture content into several state intervals. For each combination of underlying surface category and moisture content state, statistically analyze the distribution characteristics of the excess infiltration runoff ratio sample to obtain the expected interval of excess infiltration runoff ratio corresponding to the combination.

[0022] S35. Summarize and form a discrimination matrix: Organize each underlying surface category and each combination of water content states and their corresponding expected intervals into a two-dimensional or multi-dimensional lookup table, that is, form a priori discrimination matrix for runoff generation mechanism;

[0023] S36. Dynamic Update: As new historical data or forward modeling samples are added, the expected range is updated in a rolling manner.

[0024] Further, S4 specifically involves: using each set of parameter combinations in the statistically acceptable set to drive the distributed hydrological model to re-run the runoff calculation for each flood in the target watershed, back-calculating the simulated excess runoff ratio corresponding to that set of parameter combinations, then comparing it with the expected interval to calculate the deviation, and defining a method for obtaining the mechanism consistency score based on the deviation; finally, using the mechanism consistency score, according to a preset lower limit, parameter combinations below the preset lower limit are judged as mechanism incompatible, thereby filtering parameter combinations.

[0025] Furthermore, the mechanism consistency score is obtained through a monotonically decreasing function of deviation; the combination of screening parameters is used to exclude from the statistically acceptable set, or to retain but assign low weights.

[0026] Furthermore, the fusion of comprehensive weights in S5 specifically involves: a power-weighted product of statistical likelihood weights and mechanism consistency scores, followed by normalization, where the power of the mechanism consistency score is an adjustable mechanism constraint strength adjustment coefficient.

[0027] Furthermore, the specific method for generating the flood probability forecast interval in S5 is as follows: using comprehensive weights, the simulated flow process corresponding to the parameter combinations retained after screening is weighted, and the quantiles of the flow are calculated for each time period, expressed as:

[0028] ,

[0029] In the formula, Indicates the overall weight. Let i be the simulated flow rate corresponding to the i-th parameter combination at time t; Indicates weighted The weighted quantile function is calculated by sorting the simulated flow rates corresponding to each group of parameters according to their numerical values, accumulating them according to their weights, and finding the flow rate value that makes the accumulated weight exactly reach p. When p is taken as 5%, 50%, and 95% respectively, The weighted median of the flow process is used as the central estimate of the forecast. and These constitute the lower and upper bounds of the 90% confidence interval, respectively, and the narrowed flood probability forecast interval is generated for each time period accordingly.

[0030] Furthermore, S6 specifically involves: calculating the proportion of parameter combinations that were determined to be mechanistically incompatible in S4 for each watershed and each event to the statistically acceptable set; when the proportion does not exceed a preset threshold, the prior interval of the parameters for that watershed is determined to be reasonable and does not require adjustment; when the proportion exceeds the preset threshold, it is determined that the existing prior interval settings for infiltration-related parameters in that watershed may have a systematic deviation, and a review prompt is automatically generated to guide the subsequent correction of the prior intervals of related parameters.

[0031] A distributed hydrological model parameter uncertainty quantification and probability forecasting device includes one or more processors for implementing the distributed hydrological model parameter uncertainty quantification and probability forecasting method.

[0032] Beneficial effects: Based on statistical likelihood screening, this invention further introduces the physical laws of runoff generation mechanism as a verification constraint, and performs secondary screening and differentiated weight adjustment on the behavioral parameter set to narrow the unreasonable uncertainty interval and improve the physical reliability and practical value of flood probability forecasting in small mountain watersheds. Attached Figure Description

[0033] Figure 1 This is a flowchart of a distributed hydrological model parameter uncertainty quantification and probability forecasting method according to the present invention;

[0034] Figure 2 This is a schematic diagram comparing the flood probability forecast intervals of traditional GLUE statistical likelihood screening with the statistical and mechanistic dual screening of this invention.

[0035] Figure 3 This is a schematic diagram of the structure of a distributed hydrological model parameter uncertainty quantification and probability forecasting device according to an embodiment of the present invention. Detailed Implementation

[0036] To make the objectives, technical solutions, and technical effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0037] like Figure 1 As shown in the figure, a method for quantifying the uncertainty of parameters in a distributed hydrological model and predicting probabilistic forecasts according to an embodiment of the present invention includes the following steps:

[0038] S1, Initial screening based on parameter sensitivity.

[0039] The distributed hydrological model used in this embodiment is an improved hybrid runoff model based on the Xin'anjiang model's runoff generation and confluence framework, which divides the watershed runoff into two parts: overflow runoff and infiltration runoff. The area within the watershed with infiltration capacity lower than the rainfall intensity generates infiltration runoff, while the remaining area generates runoff according to the Xin'anjiang model's overflow runoff mechanism. The area ratio and infiltration capacity of the two are controlled by the infiltration runoff area ratio parameter and the infiltration capacity parameter, respectively. The evapotranspiration conversion factor, surface soil free water storage capacity, and interflow recession coefficient are conventional runoff generation and confluence parameters of the Xin'anjiang model, and their meanings are consistent with those of the conventional Xin'anjiang model.

[0040] The parameters to be calibrated for the above-mentioned distributed hydrological model include the evapotranspiration capacity conversion factor K, the surface soil free water storage capacity SM, the interflow recession factor CI, and the infiltration capacity parameter f. m The parameters such as the excess infiltration runoff area ratio (imf) and the spatial distribution index of infiltration capacity (γ) were used. A Monte Carlo method was employed to randomly sample parameters within their prior intervals to generate a parameter combination sample set. This sample set was then substituted into the distributed hydrological model and run group by group to obtain output indicators such as total runoff, peak flow, surface runoff, soil runoff, and groundwater runoff. The standardized regression coefficient (SRC) method was used to calculate the correlation strength between each parameter and each output indicator. A |SRC|≥0.2 was used as the criterion to identify a subset of sensitive parameters that significantly affect the model output, which were then used as parameters to be analyzed in subsequent steps.

[0041] S2, First-level screening: Statistical likelihood screening.

[0042] For the sensitive parameters identified by S1, a large sample size, such as no less than 5,000 to 10,000 independent Monte Carlo samplings, is performed within the prior distribution range of the sensitive parameters. Each sample is then substituted into a distributed hydrological model to simulate historical flood events in the target watershed. The Nash efficiency coefficient (NSE) is used as the likelihood function, and the calculation formula is as follows:

[0043] ,

[0044] In the formula , These are the simulated value and the measured value at time i, respectively. This is the average of the measured values;

[0045] A preset acceptable threshold NSE0 is set. Parameter combinations with NSE≥NSE0 are included in the statistically acceptable set Θstat, and the statistical likelihood weight of each parameter combination is calculated using the following formula:

[0046] ,

[0047] In the formula To statistically analyze the first digit of the acceptable set Θstat Group parameter combinations.

[0048] S3. Construct the prior discrimination matrix for the flow generation mechanism, which includes the following sub-steps:

[0049] S31. Underlying Surface Classification: Based on the characteristics of the underlying surface, such as watershed lithology, soil type, and topography, the target watershed and its similar watersheds are divided into several underlying surface categories.

[0050] S32. Obtaining runoff generation control characteristics: For each underlying surface category, obtain one or more control variables that can characterize the relative runoff generation capacity of that type of underlying surface, such as ratio-type indicators that reflect the relative relationship between the infiltration capacity of the underlying surface and the rainfall intensity of the storm, or other equivalent physical and statistical characteristics. The specific method of obtaining these characteristics is not limited.

[0051] S33. Accumulate samples of runoff generation mechanisms: By statistically analyzing historical measured flood events, or by performing forward modeling of historical or synthetic rainfall events using physical mechanism models such as infiltration models and runoff generation models, samples of the proportion of excess infiltration runoff corresponding to different underlying surface categories under different anterior soil moisture contents are obtained.

[0052] S34. Grouping and Statistically Determining the Expected Interval by State: Divide the previous soil moisture content into several state intervals, such as using a certain critical threshold to distinguish between slightly wet and slightly dry states. For each combination of "underlying surface type × moisture content state", statistically analyze the distribution characteristics of its excess infiltration runoff proportion sample, such as the mean, quantiles, or range of values, to obtain the expected interval of excess infiltration runoff proportion corresponding to that combination. The upper and lower limits of the interval are determined based on the statistical results of historical events in the basin.

[0053] S35. Summarize and form a discrimination matrix: Organize each underlying surface category and each combination of water content states and their corresponding expected intervals into a two-dimensional or multi-dimensional lookup table, which forms the prior discrimination matrix of runoff generation mechanism, for subsequent S4 steps to look up and call.

[0054] S36. Dynamic Update: As new historical sessions or forward modeling samples accumulate, the expected interval can be updated in a rolling manner to improve the timeliness of the discrimination matrix.

[0055] S4. Second screening: Mechanism consistency verification.

[0056] For each parameter combination in the statistically acceptable set Θstat obtained from S2 Using this set of parameters, the distributed hydrological model is rerun to calculate the runoff generation for each flood in the target watershed, and the simulated excess runoff generation ratio corresponding to this parameter combination is calculated in reverse:

[0057] ,

[0058] In the formula , These represent the excess permeability flow rate and the full storage flow rate, respectively.

[0059] Based on the lithological classification of the watershed to which this event belongs and the current state of the real-time soil moisture content (Pa), the corresponding expected interval is found in the discrimination matrix constructed in S3. ,calculate The degree of deviation relative to the expected range :

[0060] when When falling within the interval ;

[0061] when When the value exceeds the interval, the proportional distance relative to the interval boundary value is taken as the normalized deviation, specifically calculated piecewise using the following formula:

[0062] ,when ,

[0063] ,when ,

[0064] , ,

[0065] Based on this, a mechanism consistency score is defined, which is a monotonically decreasing function of the deviation:

[0066]

[0067] For deviation penalty coefficient; The range of values ​​is ; The larger the value, the further the flow generation mechanism corresponding to that set of parameters deviates from the prior discriminative expectation. The smaller; Below the preset lower limit The parameter combinations are considered to be incompatible in mechanism:

[0068]

[0069] It can be selectively removed from the set of behavioral parameters, or retained but given a very low weight.

[0070] S5, Dual-weighted fusion and probabilistic forecast generation.

[0071] The statistical likelihood weights obtained from S2 Mechanism consistency score obtained with S4 By merging the results, a comprehensive weight is obtained:

[0072] ,

[0073] In the formula For the mechanism constraint strength adjustment coefficient, when It degenerates into the traditional GLUE framework without introducing a mechanism consistency check:

[0074] ,

[0075] Increasing the value of the mechanism consistency enhances its impact on the final weight.

[0076] Using comprehensive weights For the simulated flow processes corresponding to the parameter combinations retained after double screening, the quantiles of the flow are calculated for each time period:

[0077] ,

[0078] In the formula Let i be the simulated flow rate corresponding to the i-th parameter combination at time t; Indicates weighted The weighted quantile function is calculated by sorting the simulated flow rates corresponding to each group of parameters according to their numerical values, accumulating them according to their weights, and finding the flow rate value that makes the accumulated weight exactly reach p. When p is taken as 5%, 50%, and 95% respectively, The weighted median of the flow process is used as the central estimate of the forecast. and These form the lower and upper bounds of the 90% confidence interval, respectively. Based on this, a narrowed flood probability forecast interval is generated for each time period, for use in flash flood early warning decision-making. (For reference) Figure 2 As shown, a comparison is made between the flood probability forecast interval of the traditional GLUE method that only uses statistical likelihood screening and the method of statistical and mechanistic dual screening of the present invention.

[0079] S6, Closed-loop diagnostic feedback.

[0080] The proportion of parameter combinations deemed mechanistically incompatible in S4 for each watershed and each event is statistically analyzed within the statistically acceptable set Θstat. If this proportion does not exceed a preset threshold of 30%, the prior interval of the parameters for that watershed is considered basically reasonable and requires no adjustment. If this proportion exceeds the preset threshold of 30%, the existing IMF and F parameters for that watershed are determined to be... m There may be systematic deviations in the prior interval settings of infiltration-related parameters. Automatic verification prompts are generated to guide the subsequent correction of the prior intervals of related parameters, forming a closed loop of "parameter uncertainty analysis - mechanism consistency verification - parameter prior interval correction".

[0081] In summary, this invention introduces a mechanism consistency verification step based on the runoff mechanism discrimination matrix on the basis of traditional statistical likelihood screening. It performs dual screening and weight fusion of statistical and mechanistic parameters, thereby eliminating or reducing the weight of statistically acceptable but physically unreasonable parameter combinations, narrowing the uncertainty range of flood probability forecasting, and providing an automatically executable diagnosis and feedback mechanism for the rationality of parameter calibration results.

[0082] Corresponding to the aforementioned embodiment of a distributed hydrological model parameter uncertainty quantification and probability forecasting method, the present invention also provides an embodiment of a distributed hydrological model parameter uncertainty quantification and probability forecasting device.

[0083] See Figure 3 The present invention provides a distributed hydrological model parameter uncertainty quantification and probability forecasting device, which includes one or more processors for implementing a distributed hydrological model parameter uncertainty quantification and probability forecasting method in the above embodiments.

[0084] An embodiment of the distributed hydrological model parameter uncertainty quantification and probability forecasting device of the present invention can be applied to any device with data processing capabilities, such as a computer. The device embodiment can be implemented through software, hardware, or a combination of both. Taking software implementation as an example, as a logical device, it is formed by the processor of any data processing device loading the corresponding computer program instructions from non-volatile memory into memory for execution. From a hardware perspective, such as... Figure 3 The diagram shown is a hardware structure diagram of any data processing-capable device, including the distributed hydrological model parameter uncertainty quantification and probability forecasting device of the present invention. (Except for...) Figure 3 In addition to the processor, memory, network interface, and non-volatile memory shown, any data processing device in the embodiment may also include other hardware depending on the actual function of the data processing device, which will not be described in detail here.

[0085] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0086] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the present invention according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0087] This invention also provides a computer-readable storage medium storing a program thereon, which, when executed by a processor, implements a distributed hydrological model parameter uncertainty quantification and probability forecasting method as described in the above embodiments.

[0088] The computer-readable storage medium can be an internal storage unit of any data processing device described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.

[0089] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any way. Although the implementation process of the present invention has been described in detail above, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for quantifying the uncertainty of parameters in a distributed hydrological model and for probabilistic forecasting, characterized in that, Includes the following steps: S1. Identify the sensitive parameters that affect the model output; S2. Perform Monte Carlo sampling and likelihood function calculation on the sensitive parameters to obtain the statistically acceptable set and the corresponding statistical likelihood weights; S3. Based on watershed runoff generation, obtain the expected range of the proportion of excess infiltration runoff under different combinations of underlying surface type and previous water content state, and thus construct a priori discrimination matrix for runoff generation mechanism; S4. For each parameter combination in the statistically acceptable set, back-calculate the simulated over-permeability flow ratio, compare it with the expected range to obtain the deviation degree, and calculate the consistency score to screen parameter combinations. S5. The statistical likelihood weights and the mechanism consistency scores are combined to obtain a comprehensive weight, and a flood probability forecast interval is generated. S6. Based on the proportion of incompatible parameter combinations obtained from S4, generate parameter prior interval verification feedback.

2. The method as described in claim 1, characterized in that, Specifically, S1 involves selecting five parameters as parameters to be calibrated: evapotranspiration capacity conversion factor, surface soil free water storage capacity, interflow recession coefficient, infiltration capacity parameter, and excess infiltration runoff area ratio parameter. A Monte Carlo method is used to randomly sample within the prior intervals of each parameter to generate a parameter combination sample set, which is then substituted into the distributed hydrological model to obtain the output indicators of total runoff, peak flow, surface runoff, interflow, and groundwater runoff. The standardized regression coefficient method is used to calculate the correlation strength between each parameter and each output indicator, and a subset of sensitive parameters that significantly affect the model output is identified based on set criteria.

3. The method as described in claim 1, characterized in that, S2 specifically involves: for sensitive parameters, conducting no fewer than 5,000 to 10,000 independent Monte Carlo samplings within the prior distribution range of the sensitive parameters, and substituting each sampling into a distributed hydrological model to simulate historical flood events in the target watershed; using the Nash efficiency coefficient NSE as the likelihood function, setting an acceptable threshold NSE0, including parameter combinations with NSE≥NSE0 into the statistically acceptable set, and determining the statistical likelihood weight of each parameter combination based on the proportion of the likelihood function value of each parameter combination in the statistically acceptable set to the sum of the likelihood function values ​​of all parameter combinations in the set.

4. The method as described in claim 1, characterized in that, S3 specifically includes the following sub-steps: S31. Underlying surface classification: Based on the watershed lithology, soil type, and topography, the target watershed and its similar watersheds are divided into several underlying surface categories; S32. Obtain runoff control characteristics: For each type of underlying surface, obtain one or more control variables that can characterize the relative runoff generation capacity of that type of underlying surface; S33. Accumulate runoff generation mechanism samples: By statistically analyzing historical measured flood events or by performing forward modeling of historical or synthetic rainfall events using physical mechanism models, obtain samples of the proportion of excess infiltration runoff corresponding to different underlying surface categories under different anterior soil moisture content states. S34. Statistical expectation intervals by state grouping: Divide the previous soil moisture content into several state intervals. For each combination of underlying surface category and moisture content state, statistically analyze the distribution characteristics of the excess infiltration runoff ratio sample to obtain the expected interval of excess infiltration runoff ratio corresponding to the combination. S35. Summarize and form a discrimination matrix: Organize each underlying surface category and each combination of water content states and their corresponding expected intervals into a two-dimensional or multi-dimensional lookup table, that is, form a priori discrimination matrix for runoff generation mechanism; S36. Dynamic Update: As new historical data or forward modeling samples are added, the expected range is updated in a rolling manner.

5. The method as described in claim 1, characterized in that, S4 specifically involves: using each parameter combination in the statistically acceptable set to drive the distributed hydrological model to re-run the runoff calculation for each flood in the target watershed, back-calculating the simulated excess runoff ratio corresponding to that parameter combination, then comparing it with the expected range to calculate the deviation, and defining a method for obtaining the mechanism consistency score based on the deviation; finally, using the mechanism consistency score, according to a preset lower limit, parameter combinations below the preset lower limit are judged as mechanism incompatible, thereby filtering parameter combinations.

6. The method as described in claim 5, characterized in that, The consistency score of the mechanism is obtained by a monotonically decreasing function of the deviation; the combination of screening parameters is to remove from the statistically acceptable set or retain them but assign them low weight.

7. The method as described in claim 1, characterized in that, The fusion of comprehensive weights in S5 is specifically as follows: the statistical likelihood weight and the mechanism consistency score are weighted by a power and then normalized, wherein the power of the mechanism consistency score is an adjustable mechanism constraint strength adjustment coefficient.

8. The method as described in claim 1, characterized in that, The specific method for generating the flood probability forecast interval in S5 is as follows: Using a comprehensive weighting, the simulated flow process corresponding to the retained parameter combinations after screening is weighted, and the quantiles of the flow are calculated for each time period. The expression is: ， In the formula, Indicates the overall weight. Let i be the simulated flow rate corresponding to the i-th parameter combination at time t; Indicates weighted The weighted quantile function is calculated by sorting the simulated flow rates corresponding to each group of parameters according to their numerical values, accumulating them according to their weights, and finding the flow rate value that makes the accumulated weight exactly reach p. When p is taken as 5%, 50%, and 95% respectively, The weighted median of the flow process is used as the central estimate of the forecast. and These constitute the lower and upper bounds of the 90% confidence interval, respectively, and the narrowed flood probability forecast interval is generated for each time period accordingly.

9. The method as described in claim 1, characterized in that, Specifically, S6 involves: calculating the proportion of parameter combinations that were determined to be mechanistically incompatible in S4 for each watershed and each event to the statistically acceptable set. If the proportion does not exceed a preset threshold, the prior interval of the parameters for that watershed is deemed reasonable and does not require adjustment. If the proportion exceeds the preset threshold, it is determined that the existing prior interval settings for infiltration-related parameters in that watershed may have systematic deviations, and a review prompt is automatically generated to guide subsequent corrections to the prior intervals of related parameters.

10. A distributed hydrological model parameter uncertainty quantification and probability forecasting device, characterized in that, It includes one or more processors for implementing the distributed hydrological model parameter uncertainty quantification and probabilistic forecasting method as described in any one of claims 1 to 9.

Images