A rainfall-runoff simulation method for flood data cleaning based on dynamic thresholds

Abstract

This invention discloses a rainfall-runoff simulation method for cleaning flood data based on dynamic thresholds. The method involves collecting multi-year hourly-scale measured rainfall and cross-sectional flow data from a target watershed; segmenting the target watershed's continuous rainfall-runoff data sequence into flood events based on the morphological characteristics of the flow process lines, extracting rainfall-runoff data corresponding to each flood event; identifying and removing abnormal flood events with unreasonable magnitude relationships in the rainfall-runoff data through preliminary rainfall screening, rainfall-runoff total volume comparison screening, and refined screening based on runoff coefficient truncation analysis; removing flood event data with abnormal temporal sequences by screening key time points during the flood process; and using the cleaned flood data as input to a hydrological model for normal simulation. This method standardizes the flood data cleaning process, preemptively removing unreasonable flood event data, ensuring clean input data, and thus improving flood forecast accuracy.

Description

Technical Field

[0001] The invention belongs to a flood model data processing method, and in particular relates to a rainfall runoff simulation method for cleaning flood data based on a dynamic threshold. Background Art

[0002] Flood forecasting, as a primary non-engineering measure for reducing flood risk, is crucial for rationally predicting flood formation, characterizing flood risks, and scientifically preparing for flood preparedness. Flood formation is influenced by multiple factors, including basin topography, underlying vegetation, and human activities. The high temporal and spatial variability and nonlinearity of runoff also pose significant challenges to flood forecast accuracy. Improving flood forecast accuracy has become a key issue in the fields of hydrometeorology and disaster prevention.

[0003] Improving flood forecast accuracy currently focuses on three key areas: input data, forecast model structure, and parameter calibration. Based on the actual operation of forecast projects and the experience of project managers, abnormal or erroneous flood data in samples often significantly restricts forecast effectiveness. Irrational data input cannot be fundamentally corrected through real-time process correction technology and may even affect the prediction results of other correct flood events and the direction of error correction.

[0004] Error correction techniques can be used to perform real-time corrections on both rainfall and runoff data inputs. Multi-source data fusion is an effective means of improving the performance of hydrological forecast simulations. Current correction methods primarily rely on the error relationship between the predicted results and the measured runoff during or after the forecast, attempting to attribute the error contribution in the hope of improving the next forecast. While these methods are effective, most require complex mathematical and statistical calculations, resulting in cumbersome and inefficient processes. Beyond multi-source data fusion, few methods have been developed to enhance forecast accuracy by using technical means to improve data quality at the initial stage of model input. Summary of the Invention

[0005] The purpose of the present invention is to propose a rainfall runoff simulation method for cleaning flood data based on dynamic thresholds, to create a scientific and feasible flood data cleaning method, to standardize the flood data cleaning process, and to eliminate unreasonable flood data in advance, thereby solving the above technical problems.

[0006] In order to solve the above technical problems, the present invention is implemented through the following solutions:

[0007] The rainfall runoff simulation method for cleaning flood data based on dynamic thresholds includes the following steps:

[0008] Step S1: Data collection and compilation: Collect multi-year hourly rainfall and cross-sectional runoff data for the target basin.

[0009] Step S2: Flood Event Segmentation: Based on the morphological characteristics of the hydrograph, the continuous rainfall and runoff data sequence of the target watershed is segmented into flood events. The rainfall and runoff data corresponding to each flood event are extracted to form independent flood event data sequences. The morphological characteristics of the hydrograph (such as the rising point, flood peak, and water retreat points) can reflect the occurrence, development, and retreat of floods. By segmenting flood events, accurate rainfall and runoff information for each flood can be obtained, providing a foundation for subsequent cleaning and simulation.

[0010] Step S3: Magnitude cleaning and correction: Through preliminary rainfall screening, rainfall-runoff total volume comparison screening, and fine screening based on runoff coefficient truncation analysis, abnormal flood events with unreasonable magnitude relationships in rainfall-runoff data are identified and eliminated.

[0011] Step S4: Process cleaning and correction: By screening the key time points (rising point, peak time) in the flood process, the flood event data with abnormal time sequence after screening in step S3 are eliminated; ensuring that the flood process conforms to the natural hydrological laws and providing data with reasonable physical meaning for subsequent simulations.

[0012] Step S5: Simulation and Effect Evaluation: The cleaned flood data is used as input to the hydrological model for simulation, and the simulation results are evaluated using evaluation indicators. This simulation can predict the occurrence and development of floods, providing a basis for decision-making in flood prevention and disaster reduction. Using accuracy evaluation indicators to evaluate simulation results verifies the quality of data cleaning and the applicability of the model, providing a reference for further improving the model and data processing methods.

[0013] For further optimization, step S1 requires that the collected data be temporally continuous and meet consistency requirements. For data series affected by human activities, runoff restoration calculations are performed and compiled. Human activities (such as water conservancy project construction and water resource development and utilization) can alter the natural runoff process in a watershed. Runoff restoration calculations can eliminate these influences, ensuring that the data better reflects the natural hydrological characteristics of the watershed.

[0014] The Thiessen polygon method is used to calculate the target basin's surface rainfall, generating a continuous rainfall-runoff data sequence. This method more reasonably distributes rainfall data from various rain gauges within the basin to the entire basin, resulting in more accurate surface rainfall data. The Thiessen polygon method is currently available and will not be further described.

[0015] During the collection and compilation process, understanding the basin profile helps to understand the basin's hydrological characteristics and possible influencing factors. Checking the consistency of the data can ensure data quality. Large basins, due to their large area and complex impacts of human activities, require runoff restoration calculations to ensure data reliability.

[0016] Further optimization, the step S3 specifically includes:

[0017] Step S3.1: Preliminary Rainfall Screening: Determine whether rainfall occurred between the time series of each flood event. If not, the event is deemed to have failed rainfall levels and is removed. If rainfall did occur, proceed to step S3.2 for further level cleaning. Precipitation is a key factor in flood formation. If no rainfall occurred between flood time series, this indicates data anomalies or failure to meet the basic flood formation conditions, and is therefore removed.

[0018] Step S3.2: Comparison and screening of total rainfall and runoff volume: By comparing the total rainfall and runoff volume, further screening out flood events with unreasonable rainfall-runoff relationship. Specifically including:

[0019] Step S3.2.1: Calculate the total rainfall P for each event 总 :

[0020] ;

[0021] in, n is the number of time periods, that is, the number of time periods into which a rainfall process is divided; is the time interval, p i Indicates the i The average rainfall intensity during a period of time.

[0022] Step S3.2.2: Calculate the total runoff R for each flood event 总 ;

[0023] ;

[0024] in, Q t express t Cross-sectional flow at a given moment.

[0025] Step S3.2.3: Calculate the net runoff R for each flood event net :

[0026] R net =R 总 -R b -R pre ;

[0027] ;

[0028] ;

[0029] Among them, Q0 is the initial water flow of the corresponding session, Q b is the base flow, Q pis the early stage water flow; λ is the water retreat coefficient, t 0 is the time corresponding to the flood rising point of this incident, t e is the time corresponding to the end point of the previous water retreat, t f is the time corresponding to the end point of the flood; R b represents base flow runoff; R pre Indicates the previous runoff volume.

[0030] Step S3.2.4: Introduce the sgn() function to perform R net and P 总 Compare,

[0031] ;

[0032] δ=R net -P 总 ;

[0033] The corresponding flood events with an sgn(δ) value of 1 are retained, and the remaining flood event data are discarded, and then the next step of magnitude cleaning is entered.

[0034] Step S3.3: Based on the runoff coefficient truncation analysis, the magnitude cleaning is further performed to remove the events with unreasonable rainfall-runoff magnitude relationship to improve the data quality. Specifically, it includes:

[0035] Step S3.3.1: Calculate the flood runoff coefficient: Use the formula C = R net / P 总 Calculate the runoff coefficients of all floods filtered out in step S3.2; then sort the runoff coefficients in ascending order to obtain an ordered array {C j | 1≤j≤N}, where N is the total number of flood events screened in step S3.2.

[0036] Step S3.3.2: Set the percentage ρ 1% and ρ 2%, and then calculate the characteristic quantile m k , and the corresponding runoff coefficient sequence quantile G ρk% ;

[0037] m k =( N +1)× ρ k% ;

[0038] ;

[0039] Where K = 1, 2; Indicates floor operation.

[0040] Step S3.3.3: Determine reasonable value boundaries: Keep C j ∈[G ρ 1% , G ρ 2% ] corresponding flood event data, eliminating the cases where C is less than G ρ 1% and greater than G ρ 2% The magnitude cleanup and correction of abnormal flood events is now complete. By calculating quantiles and defining reasonable value boundaries, the normal range of runoff coefficient can be determined and outliers can be excluded.

[0041] Further optimization, ρ 1% is 5%, ρ 2% is 95%.

[0042] Further optimization, the step S4 specifically includes:

[0043] Step S4.1: Screening of flood starting points: Identify the time point t0 corresponding to each flood starting point and the rainfall start time t p , compare their order, if t0 <t p , then the flood is regarded as an abnormal flood and eliminated; if t0≥t p , then enter the next step of the cleaning process. By screening the rising point time, abnormal flood events with illogical rising point times are eliminated, ensuring the rationality of the flood data in terms of chronological order, improving the quality and reliability of the data, and providing a more accurate data foundation for subsequent simulation and analysis.

[0044] Step S4.2: Peak time screening: Identify the peak time t of each flood Qmax , peak time of rain t Pmax , compare their order, if t Qmax <t Pmax , then the flood is regarded as an abnormal flood and is eliminated; if t Qmax ≥t Pmax , then the flood will be retained, and the flood data cleaning is completed.

[0045] Generally, flood peaks should occur after rainfall peaks. If a flood peak occurs earlier than the rainfall peak, it does not conform to the natural process of rainfall runoff and is therefore excluded. By screening for peak times, we eliminate abnormal flood events whose peak times do not conform to natural logical relationships, ensuring the rationality of flood data in terms of peak times, improving data quality and reliability, and providing a more accurate data foundation for subsequent simulations and analysis.

[0046] Further optimization, in step S5, the simulation effect is evaluated using the flood peak relative error Rep, the peak appearance time relative error Ret, the flood volume relative error Rew and the Nash efficiency coefficient NSE as indicators.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] 1. The present invention forms a dual screening mechanism through magnitude cleaning and process cleaning, which not only focuses on the magnitude relationship between rainfall and runoff, but also ensures the rationality of the time sequence of the flood process, standardizes the flood data cleaning process, eliminates unreasonable flood event data in advance, ensures the cleanliness of the input data, and thus improves the accuracy of flood forecasting.

[0049] 2. The method described in the present invention does not perform complex and arduous error attribution or numerical contribution rate calculation. The method is simple, the physical mechanism is clear, and it does not rely on subsequent real-time simulation (forecast) feedback. It is independent, stable, and has strong anti-interference capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flow chart of the rainfall-runoff simulation method for cleaning flood data based on dynamic thresholds according to the present invention;

[0051] Figure 2 This is the distribution map of flood runoff coefficients for the screened events;

[0052] Figure 3 This is a schematic diagram of the flood process on April 13, 2020;

[0053] Figure 4 This is a schematic diagram of the flood process No. 2021072108. DETAILED DESCRIPTION

[0054] The technical solution of the present invention is described in detail below with reference to the embodiments, but the protection scope of the present invention is not limited to the embodiments.

[0055] This example selects a small- to medium-sized river basin in a key flood-prone area in southwest China for case study, and uses the Xin'anjiang model (the Xin'anjiang model is a hydrological model proposed by Professor Zhao Renjun's team at East China University of Water Resources (now Hohai University)) to simulate the rainfall-runoff process.

[0056] In this embodiment, if Figure 1 As shown in FIG, a rainfall runoff simulation method for cleaning flood data based on a dynamic threshold includes the following steps:

[0057] Step S1: First, a hydrological survey is conducted on the small and medium-sized watershed. Through investigation and data review, it is determined that there have been no new water conservancy projects or human activities in the watershed since 2015. Therefore, the collected continuous rainfall and runoff data series meet the representativeness, reliability and consistency review and can be used for the subsequent hydrological simulation.

[0058] In this embodiment, the Thiessen polygon method is used to compile rainfall data, and finally a continuous time series of data from June 2017 to October 2022 is obtained.

[0059] Step S2: A total of 48 flood events were selected from the continuous time series data based on flood characteristics, and the corresponding rainfall and runoff flow data were extracted. Each flood was then generated into a separate Excel file containing three columns of data: time, rainfall, and runoff. The floods were named after their starting time.

[0060] Step S3: Level cleaning and calibration, specifically including:

[0061] Step S3.1: Preliminary rainfall screening: Carry out flood magnitude cleaning. Check whether there is rainfall in the extracted floods and calculate the total rainfall P for each flood. 总 It was found that the total rainfall of the two floods on June 18, 2017 and May 24, 2019 was 0, which was determined to be a false rainfall-runoff relationship and was removed. The remaining 46 reasonable flood events, after this cleaning, entered the next step of magnitude cleaning.

[0062] Step S3.2: Comparison and screening of total rainfall runoff: First, calculate the total flood runoff R for each rainfall event. 总 , according to formula R net =R 总 -R b -R pre Remove base flow and any remaining flow from the previous flood.

[0063] From the long time series, we find the period of drought and no rain, and get the base flow value of 2.35m 3 / s. In step S3.1, the total rainfall for all events was calculated. The SGN value of the difference between the total rainfall and the total runoff for each event was calculated. It was found that the SGN values ​​of four floods were less than or equal to 0. These were determined to have incorrect rainfall-runoff relationships and were removed. The remaining 42 floods were reasonable, and the next step of magnitude cleaning was performed.

[0064] Step S3.3: Use the runoff coefficient to remove events with unreasonable rainfall-runoff magnitude relationships. Since all flood runoff generating and confluence underlying surfaces are uniform, the runoff coefficient should not have too large a difference in frequency distribution. Specifically, it includes:

[0065] Step S3.3.1: Calculate the flood runoff coefficient: Use the formula C=Rnet / P 总 Calculate the runoff coefficients of all 42 screened floods. Since the names are too long, the flood numbers are used instead of the flood names. The runoff coefficient distribution is as follows: Figure 2 shown.

[0066] Step S3.3.2: Quantile calculation: Sort the runoff coefficients in ascending order and calculate the corresponding characteristic quantiles at 5% and 95% respectively. m k , and the corresponding runoff coefficient sequence quantile G ρk% .

[0067] m k =( N +1)× ρ k% ;

[0068] ;

[0069] Calculate the flood sequence characteristic quantile G 5% =2.353, G 95% =5.287, K=1,2.

[0070] Step S3.3.3: Determine the reasonable value boundary. Use the truncation method to determine the reasonable value boundary:

[0071] ;

[0072] A total of 19 flood events that did not meet the boundary constraints were identified and eliminated, and a total of 23 reasonable flood events were output, and the magnitude cleaning and correction was completed.

[0073] Step S4: Process cleaning and correction, by screening the key time points in the flood process, eliminating the flood event data with abnormal time sequence after the screening in step S3. Specifically including:

[0074] Step S4.1: Screening of flood starting points: Preliminary screening of flood starting points by traversing 23 floods, identifying the time point t0 corresponding to each flood starting point and the rainfall start time t p , compare their order, if t0 <t p , then the flood is regarded as an abnormal flood and eliminated; if t0≥t p , then enter the next step of cleaning.

[0075] In this embodiment, a total of 1 flood was eliminated during the cleaning process, and the remaining 22 floods passed the cleaning. Figure 3As shown in the figure: Flood observation No. 2020070413 is a single-peak flood. The green dot marked in the figure shows that the flood peak starting point t0 is 18:00:00 on July 5, 2020. The corresponding rainfall period is 0. The rainfall starting point t p It is 06:00:00 on July 6, 2020. The flood rising point is before the rainfall occurs. The rainfall-runoff relationship is wrong, so it is removed.

[0076] Step S4.2: Peak time screening: Identify the peak time t of each flood Qmax , peak time of rain t Pmax , compare their order, if t Qmax <t Pmax , then the flood is regarded as an abnormal flood and is eliminated; if t Qmax ≥t Pmax , then the flood will be retained, and the flood data cleaning is completed.

[0077] In this embodiment, this cleaning process removes a total of 5 floods. Take flood No. 2021072108 as an example. Figure 4 As shown in the figure, it is clear that the peak of the rain event occurred at 14:00:00 on July 22, 2021, while the peak of the flood event occurred at 05:00:00 on July 22, 2021. The flood peak occurred before the rain peak, which is inconsistent with the runoff generation and runoff mechanism. The rainfall-runoff relationship is unreasonable, so it is removed. At this point, the flood event cleaning is complete, and a total of 17 high-quality flood events have been output.

[0078] Step S5: The Sanshuiyuan Xin'anjiang model is selected for hydrological simulation. The flood peak relative error Rep, peak time relative error Ret, flood volume relative error Rew and Nash efficiency coefficient NSE are selected as accuracy assessment indicators. The 48 floods before dynamic threshold data cleaning and the 17 floods after cleaning are used as input for hydrological process simulation and accuracy assessment.

[0079] According to GB / T22482-2008, "Specifications for Hydrological Information Forecasting," the permissible error for flood peak forecasts is 20% of the measured flood peak flow rate. The permissible error for peak time is 30% of the time interval between the forecast start time and the measured flood peak time. If the permissible error is less than 3 hours or one calculation period, then 3 hours or one calculation period is used as the permissible error. Since the calculated permissible error is generally less than 3 hours, in this embodiment, 3 hours is used as the permissible error for peak time. For runoff forecasts, the permissible error is 20% of the measured value, and the pass rates for each item are calculated separately.

[0080] Based on the above analysis, the flood data before and after data cleaning were input into the model for simulation. The results were then output and the accuracy was assessed. The accuracy assessment tables for the Xin'an River model before flood cleaning are shown in Tables 1, 2, 3, and 4, and the accuracy assessment tables for the Xin'an River model after flood cleaning are shown in Tables 5, 6, 7, and 8.

[0081] Table 1 Evaluation of the simulation accuracy of the Xin'anjiang model flood peak flow before cleaning

[0082]

[0083] Table 2 Evaluation of the simulation accuracy of the peak time of the Xin'anjiang model before cleaning

[0084]

[0085] Table 3 Evaluation of the accuracy of flood simulation of the Xinanjiang model before cleaning

[0086]

[0087] Table 4 Evaluation of the certainty coefficient of the Xinanjiang model before cleaning

[0088]

[0089] Table 5 Evaluation of flood peak simulation accuracy of the Xin'anjiang model after cleaning

[0090]

[0091] Table 6 Evaluation of the simulation accuracy of the peak time of the Xin'anjiang model after cleaning

[0092]

[0093] Table 7 Evaluation of flood simulation accuracy of the Xin'anjiang model after cleaning

[0094]

[0095] Table 8 Evaluation of the deterministic coefficient of the Xinanjiang model after cleaning

[0096]

[0097] Tables 1-4 show that rainfall-runoff simulations were conducted for 48 floods before the flood cleansing phase. The pass rate for peak runoff simulations was only 39.58%, with the largest simulation error occurring in flood 2017-08-2408, reaching a relative error of 315.2%. The pass rate for peak time simulations was only 29.17%, with the largest error occurring in flood 2017-08-03-03, which saw the peak occur 41 hours earlier. The pass rate for flood volume simulations was 35.42%, with the largest relative error occurring in flood 2017-08-2408, reaching -173.9%. Overall, the simulations for the pre-cleaning phases were poor, failing to meet the minimum Class C accuracy requirement of the standard and therefore lacking reference value. Only three floods, 2020-07-2400, 2019-09-03-00, and 2020-07-10-15, achieved good simulation accuracy, reaching Class A accuracy. The remaining floods failed and exhibited significant errors.

[0098] Tables 5-8 show that for the 17 flood events cleaned using the method described in this invention, the pass rate for peak runoff simulation increased to 88.24%, with the largest error occurring at 15:00 September 2018, and a relative error of -46.08%. The pass rate for peak timing increased to 52.94%, with the largest error occurring at 16:00 October 2019, a 12-hour delay. The pass rate for flood volume increased to 94.12%, with the only unsatisfactory result occurring at 15:00 September 2018, with a relative error of 35.1%. Overall, the simulation results for each event were good, while the peak timing simulation results were poor. Overall, the simulation achieved a Class B accuracy or higher, significantly improving the accuracy of the simulations before cleaning.

Claims

1. A rainfall runoff simulation method for cleaning flood data based on dynamic thresholds, characterized in that: The steps include: Step S1: Data collection and compilation: Collect multi-year hourly rainfall and cross-sectional runoff data in the target basin; Step S2: Flood event segmentation: Based on the morphological characteristics of the flow process line, the continuous rainfall and runoff data sequence of the target basin is segmented into flood events, and the rainfall and runoff data corresponding to each flood event are extracted to form an independent flood event data sequence; Step S3: Magnitude cleaning and correction: Through preliminary rainfall screening, rainfall-runoff volume comparison screening, and fine screening based on runoff coefficient truncation analysis, abnormal flood events with unreasonable magnitude relationships in rainfall-runoff data are identified and eliminated; specifically, the following steps are performed: Step S3.1: Preliminary rainfall screening: Determine whether there is rainfall in the time series of each flood event. If not, the rainfall level is determined to be unqualified and is eliminated. If rainfall is present, proceed to step S3.2 for the next level cleaning. Step S3.2: Comparison and screening of total rainfall runoff, including: Step S3.2.1: Calculate the total rainfall P for each event 总 : ； in, n is the number of time periods, that is, the number of time periods into which a rainfall process is divided; is the time interval, p i Indicates the i Average rainfall intensity during a period of time; Step S3.2.2: Calculate the total runoff R for each flood event 总 ; ； in, Q t express t The cross-sectional flow at the time, t 0 is the time corresponding to the flood rising point of this incident, t f The time corresponding to the end point of the flood; Step S3.2.3: Calculate the net runoff R for each flood event net : R net =R 总 -R b -R pre ; ； ； Among them, Q0 is the initial water flow of the corresponding session, Q b is the base flow, Q p is the early stage water flow; λ is the water retreat coefficient, t 0 is the time corresponding to the flood rising point of this event, t e is the time corresponding to the end point of the previous water retreat, t f The time corresponding to the end point of the flood; R b represents baseflow runoff; R pre represents the runoff volume of the previous period; Step S3.2.4: Introduce the sgn() function to perform R net and P 总 Compare, ； δ=R net -P 总 ; The corresponding flood events with sgn(δ) value of 1 are retained, and the rest of the flood event data are discarded, and then the next step of magnitude cleaning is carried out; Step S3.3: Magnitude cleaning based on runoff coefficient truncation analysis, specifically including: Step S3.3.1: Calculate the flood runoff coefficient: Use the formula C = R net / P 总 Calculate the runoff coefficients of all floods filtered out in step S3.2; then sort the runoff coefficients in ascending order to obtain an ordered array {C j | 1≤j≤N}, where N is the total number of flood events screened in step S3.2; Step S3.3.2: Set the percentage ρ 1% and ρ 2%, and then calculate the characteristic quantile m k , and the corresponding runoff coefficient sequence quantile G ρk% ; m k =( N +1)× ρ k% ； ； Where K = 1, 2; Indicates floor operation; Step S3.3.3: Determine reasonable value boundaries: Keep C j ∈[G ρ 1% , G ρ 2% ] corresponding flood event data, eliminating the cases where C is less than G ρ 1% and greater than G ρ 2% The abnormal flood frequency data has been cleaned and corrected at this level; Step S4: Process cleaning and correction: By screening the key time points in the flood process, the flood event data with abnormal time sequence after the screening in step S3 are eliminated; Step S5: Simulation and effect evaluation: The cleaned flood data is used as the input of the hydrological model for simulation, and the simulation effect is evaluated using evaluation indicators.

2. The rainfall runoff simulation method for cleaning flood data based on dynamic threshold according to claim 1 is characterized in that: In step S1, the collected data is required to be continuous in time and meet consistency requirements; for data sequences affected by human activities, runoff restoration calculations are performed and compiled; the Thiessen polygon method is used to calculate the target basin surface rainfall to form a continuous rainfall runoff data sequence for the target basin.

3. The rainfall runoff simulation method for cleaning flood data based on dynamic threshold according to claim 2 is characterized in that: ρ 1% is 5%, ρ 2% is 95%.

4. The rainfall runoff simulation method for cleaning flood data based on dynamic thresholds according to claim 3 is characterized in that: The step S4 specifically includes: Step S4.1: Screening of flood starting points: Identify the time point t0 corresponding to each flood starting point and the rainfall start time t p , compare their order, if t0 <t p , then the flood is regarded as an abnormal flood and eliminated; if t0≥t p , then enter the next step of cleaning; Step S4.2: Peak time screening: Identify the peak time t of each flood Qmax , peak time of rain t Pmax , compare their order, if t Qmax <t Pmax , then the flood is regarded as an abnormal flood and is eliminated; if t Qmax ≥t Pmax , then the flood will be retained, and the flood data cleaning is completed.

5. The rainfall runoff simulation method for cleaning flood data based on dynamic threshold according to claim 4 is characterized in that: In step S5, the simulation effect is evaluated using the relative error of flood peak Rep, the relative error of peak appearance time Ret, the relative error of flood volume Rew and the Nash efficiency coefficient NSE as indicators.

Images