Drainage basin rainfall-runoff response relation determination method based on multi-factor influence quantitative analysis
By using a multi-factor quantitative analysis method, a basin rainfall-runoff response model was constructed, which solved the uncertainty problem caused by the interaction of factors in basin flood forecasting, and achieved more accurate flood forecasting and group forecasting, thus improving the applicability and stability of the model.
Patent Information
- Application Number
- CN202511492478.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-20
- Publication Date
- 2026-01-23
AI Technical Summary
Existing watershed flood forecasting models struggle to achieve accurate forecasts when faced with the influence of multi-scale coupled factors. This is especially true in large watersheds where rainfall is unevenly distributed in time and space, the underlying surface is complex, and water conservancy projects cause interference, resulting in high uncertainty in forecast results. There is a lack of systematic quantitative analysis methods for the interaction of multiple factors.
A multi-factor impact quantitative analysis method was adopted. By constructing a systematic analysis framework of key influencing factors of rainfall-runoff, representative factor indicators were selected. By combining Spearman correlation coefficient analysis and stepwise regression analysis, significant factors were screened and a multiple linear regression model was constructed. The decision tree binning algorithm was used to classify flood events into groups and quantify the rainfall-runoff response relationship.
It improves the accuracy and reliability of flood forecasting models, enables scientific grouping and differentiated forecasting of flood events, and enhances the applicability and stability of models under different scenarios.
Smart Images

Figure CN121389058A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of hydrological forecasting and water resources management, and relates to a method for determining the basin rainfall-runoff response relationship based on quantitative analysis of the influence of multiple factors. It aims to guide the establishment of a refined basin flood forecasting model and improve its accuracy and reliability. Background Technology
[0002] Watershed flood forecasting is a crucial technological tool for water resource management. Essentially, it uses mathematical models to simulate the rainfall-runoff response process within a watershed, providing key information for decisions regarding flood control, disaster reduction, and flood resource utilization. However, actual forecasts are influenced by the multi-scale coupling of various natural factors and human activities, leading to uncertainties in the forecast results. This problem is particularly pronounced in large watersheds: uneven spatial and temporal distribution of rainfall, complex underlying surface conditions, and interference from human-induced water storage and regulation measures such as hydraulic engineering all contribute to the uncertainty.
[0003] Rainfall, as the main input data for forecasting models, plays a decisive role in the accuracy of flood forecasting. Uncertainties in rainfall magnitude, temporal distribution, and spatial distribution during precipitation processes significantly affect watershed runoff mechanisms, thereby increasing the uncertainty of flood forecasting. Xu Jing et al. [Xu Jing, Ren Liliang, Liu Xiaofan, et al. Research on the propagation impact of rainfall uncertainty based on fuzzy set theory [J]. Advances in Water Science, 2009, 20(03): 422-427.] used fuzzy set theory and divided the watershed into sub-watersheds to quantitatively analyze the impact of rainfall uncertainty, such as rainfall magnitude and spatiotemporal distribution, on runoff processes in different sub-watersheds. Qiu Chao et al. [Qiu Chao. Research and application of fuzzy clustering analysis in hydrological forecasting [D]. Zhejiang University, 2007.] argued that varying rainfall intensity with different rainfall processes will exacerbate the randomness and fuzziness of water source division, thereby affecting the uncertainty of runoff generation and confluence mechanisms.
[0004] The underlying surface is the physical carrier of precipitation-runoff transformation, and its spatial heterogeneity and dynamic changes significantly affect runoff generation and collection processes. Different land use types and vegetation cover rates have different canopy interception capacities and soil infiltration characteristics. The spatiotemporal variations of these underlying surface factors (such as seasonal variations and urbanization expansion) directly determine the efficiency and form of the precipitation-runoff transformation process. Therefore, accurately understanding underlying surface conditions is crucial for improving the accuracy of flood forecasting.
[0005] Furthermore, the reservoirs, dams, and other water conservancy projects widely constructed in modern river basins play a crucial regulatory role in flood control, water supply, and power generation. However, they also fundamentally alter the natural runoff generation and confluence patterns and spatiotemporal characteristics of the basin, posing a significant challenge to the structure and parameter applicability of traditional hydrological models calibrated based on natural basin conditions. The impacts of water conservancy projects on flood runoff generation and confluence mainly include: increasing soil moisture before flood season and in the lead-up to floods, thus increasing runoff generation; impounding some floodwater at the beginning of the flood season; and potentially increasing flood volume and peak value due to dam collapse during severe rainstorms. Assani (Assani A.A., Landry R., Daigle J., et al. Reservoirs Effects on the Interannual Variability of Winter and Spring Stream flow in the St-Maurice River Watershed (Quebec, Canada). Water resources management [J]. 2011, 25: 3661-3675.) confirmed that the impact of water conservancy projects on runoff is real and exists independently of the impact of climate change.
[0006] The three factors mentioned above—precipitation, underlying surface, and water conservancy projects—do not act independently, but rather influence the rainfall-runoff response through a "multi-scale spatiotemporal coupling" mechanism. For example, localized heavy rainfall superimposed on reservoir pre-release scheduling in urbanized areas may lead to extreme flood events where traditional models completely fail. Current research often focuses on single factors and lacks systematic quantitative analysis methods for the interactions of multiple factors. This makes it difficult for flood forecasting models to simultaneously consider the complex coupling effects of natural processes and human regulation, ultimately limiting the accuracy and reliability of the models.
[0007] In summary, a key prerequisite for constructing a refined watershed flood forecasting model lies in deeply analyzing and quantifying the specific contributions and interaction mechanisms of multiple factors, such as precipitation characteristics, underlying surface conditions, and hydraulic engineering interventions, on the rainfall-runoff response process. Only through quantitative analysis of the influence of multiple factors can we clarify the main influencing factors of rainfall-runoff response in different scenarios, thereby guiding the optimization of model parameters, structural improvements, and uncertainty reduction. Summary of the Invention
[0008] To address the challenges in improving the accuracy of watershed flood forecasting, including difficulties in forecast model calibration and the complexity of the physical mechanisms of rainfall-runoff response and the difficulty in quantifying the uncertainties caused by the interaction of multiple factors, this invention provides a method for determining the rainfall-runoff response relationship in watersheds based on quantitative analysis of multiple factors. This method selects and quantitatively calculates the factors influencing the rainfall-runoff response relationship, analyzes the degree of influence and interactions of these factors, and classifies significant influencing factors into hierarchical categories, providing a theoretical basis and criteria for the construction and optimization of accurate flood forecasting models. This method, based on quantitative analysis of multiple factors, determines the rainfall-runoff response relationship and is applicable to the quantitative identification of rainfall-runoff response mechanisms and the exploration of flood forecasting mechanisms under complex watershed conditions. The method clarifies the analytical framework for key influencing factors of rainfall-runoff and verifies the rationality and application potential of this invention using an actual flood process in the Minjiang River basin as a case study.
[0009] To achieve the above objectives, the technical solution of the present invention is as follows:
[0010] A method for determining the rainfall-runoff response relationship in a watershed based on multi-factor quantitative analysis is proposed. This method utilizes a systematic framework for analyzing key influencing factors of rainfall and runoff, starting from three core dimensions: rainfall characteristics, underlying surface conditions, and hydraulic engineering intervention. First, representative factor indicators (e.g., rainfall intensity, reservoir storage and discharge) are selected based on the relevant characteristics of typical flood events in the watershed. Then, Spearman correlation coefficient analysis and stepwise regression analysis are combined to assess the statistical correlation strength (correlation strength) between each factor and the runoff generation coefficient. Factors with strong correlations are selected, and a multiple linear regression model is constructed to determine the interaction between factors. Finally, a decision tree binning algorithm is used to classify significant indicators, classify flood events into groups, and evaluate the grouping effect through inter-group statistical tests to ensure significant differences in runoff generation mechanisms between groups. The method for determining the rainfall-runoff response relationship in a watershed includes the following steps:
[0011] Step 1: Analysis of typical flood events' rainfall and flood characteristics.
[0012] Based on typical flood events, basin flood forecasting is carried out. Based on basin hydrological data, typical flood events are divided into categories, and then their rainfall and flood characteristics are analyzed to obtain the basin hydrological characteristics.
[0013] Step 1.1: Classify typical flood events based on watershed hydrological data.
[0014] First, the observation data of each rain gauge and flow station in the basin are organized according to a unified time step (such as 1 day or 6 hours). Then, independent flood events are identified based on the flow process line, the start and end times of each flood are determined, and a reasonable peak flow threshold is set in combination with the basin conditions. Then, representative flood events that exceed the peak flow threshold are screened out by the over-quantitative method. Finally, the selected flood events are uniformly numbered to obtain typical flood event sequences.
[0015] Step 1.2: Calculate rainfall characteristic indicators.
[0016] For the typical flood sequences obtained in step 1.1, rainfall characteristic indicators are calculated for each flood, including the time of peak rainfall, rainfall duration, peak rainfall intensity, and total rainfall.
[0017] Step 1.3: Calculate flood characteristic indicators.
[0018] Meanwhile, for the same typical flood sequence, flood characteristic indicators are calculated for each flood, including peak flow, total flood volume, peak time and flood duration; based on the receding process, receding curves are plotted to estimate the basin base flow, and the base flow needs to be deducted from the calculation of the total flood volume.
[0019] Step 2: Calculation of the runoff coefficient and key factor indicators.
[0020] The runoff coefficient represents the transformation relationship between rainfall and runoff. First, the runoff coefficient of typical flood sequences is calculated and its characteristics are analyzed. Then, the key factors affecting the runoff coefficient are identified and calculated.
[0021] Step 2.1: Calculate and analyze the runoff generation coefficient of the flood field in the watershed.
[0022] Based on the total rainfall and total flood volume obtained in steps 1.2 and 1.3, the runoff generation coefficient of typical flood sequences is calculated for each flood event according to formula (1). The runoff generation coefficient ranges from 0 to 1, and its value reflects the proportion of rainfall converted into runoff. The extreme values, standard deviations, and mean values of all flood runoff generation coefficients are further statistically analyzed to determine their variation range and distribution characteristics. When the range of runoff generation coefficients is greater than 0.5 or the standard deviation is greater than the mean, it indicates that the runoff generation coefficient is highly discrete and the uncertainty of the basin flood forecast is relatively large. In addition, the runoff generation model can be identified based on the range of runoff generation coefficients. For example, the runoff generation coefficient is mostly greater than 0.40 in areas dominated by runoff generation due to full storage, while it is mostly less than 0.20 in areas with excessive infiltration, providing a basis for subsequent model selection.
[0023]
[0024] In the formula, C is the runoff coefficient; R is the runoff depth (mm); and P is the rainfall depth (mm).
[0025] Step 2.2: Identify and quantify the key factors affecting the runoff coefficient.
[0026] Based on the distribution characteristics of the runoff generation coefficient in step 2.1, the complexity of the rainfall-runoff response relationship in the watershed is determined. According to the results of step 2.1, if the runoff generation coefficient has a high degree of dispersion, the rainfall-runoff response relationship in the watershed is considered complex, and the interaction of multiple factors should be considered; if the runoff generation coefficient has a low degree of dispersion, the rainfall-runoff response relationship in the watershed is considered simple, and only the influence of a single factor needs to be considered. Therefore, based on the hydrological runoff generation theory, the following five main factors affecting the rainfall-runoff response relationship are considered as key factors affecting the runoff generation coefficient: anterior rainfall, distance from the rainfall center, average rainfall intensity, peak rainfall time skewness, and reservoir storage-discharge ratio. The five key factor indicators are obtained as follows:
[0027] The preceding rainfall is characterized by the cumulative effect of recent rainfall, which reflects the soil moisture level in the basin in the early stage. The calculation is based on the commonly used traditional hydrological methods, as shown in formula (2). The distance between the rainfall centers is calculated by quantifying the spatial distribution centroid of the average rainfall in the basin and the spatial distance between the basin outlet section. The centroid identification is achieved using ArcGIS spatial analysis tools. The average rainfall intensity is calculated based on the ratio of total rainfall to rainfall duration, as shown in formula (3). The rainfall peak time skewness is an indicator designed by this invention to characterize the asymmetry of the rainfall peak time location, and is calculated according to formula (4). The reservoir storage-discharge ratio is calculated according to formula (5), which reflects the flood storage and regulation capacity of water conservancy projects during the period. The calculation is based on representative reservoirs in the basin that have strong control, large regulation capacity and are close to the downstream section.
[0028]
[0029] In the formula, The preceding rainfall amount (mm) for this period; The preceding period's rainfall (mm) represents the initial impact rainfall; K is the soil moisture content decline coefficient, which can be determined based on watershed characteristics. I represents the rainfall amount in the previous period (mm); I represents the average rainfall intensity (mm / min). This represents the cumulative change in rainfall over the period (mm). tp is the observation time step (min); S is the rainfall peak time skewness; tp is the duration from the start of rainfall to the appearance of the rainfall peak (h); Tp is the total duration of rainfall (h). The reservoir storage-to-discharge ratio is represented by Q, and the storage-to-discharge volume is represented by m³ / s. 3 W represents the total inflow of water (m³). 3 ).
[0030] Step 3: Analysis of the Indicator's Impact Effect
[0031] The impact of the runoff generation coefficient sequence obtained in step 2.1 and the key factor indicators calculated in step 2.2 on the runoff generation coefficient is systematically analyzed. First, Spearman rank correlation analysis is used to identify key influencing indicators that have a monotonic correlation with the runoff generation coefficient. Then, multifactor stepwise regression analysis is used to reveal the interaction among key influencing indicators in complex watersheds and their joint explanatory power for the runoff generation coefficient.
[0032] Step 3.1, Spearman rank correlation analysis.
[0033] Based on the key factor indicators and runoff coefficients of each flood event obtained in step 2, the Spearman correlation coefficient is calculated. The Spearman correlation coefficient is a non-parametric rank correlation measure used to assess the strength and direction of monotonic relationships between variables: the sign represents positive or negative correlation, the absolute value reflects the degree of influence, and the p-value is used to determine statistical significance. The correlations are ranked, and key factor indicators with p-values less than 0.05 are selected, resulting in n key influence indicators (1 ≤ n ≤ 5). The rainfall-runoff response relationship of a single watershed is then determined, and influence analysis is conducted based on these key factor indicators. The p-value is the probability of observing the current sample data or more extreme data under the null hypothesis, obtained by calculating the test statistic and comparing it with its theoretical distribution.
[0034] Step 3.2, multivariate stepwise regression analysis.
[0035] Using the inflow coefficient as the dependent variable and the n key influencing indicators selected in step 3.1 as independent variables, a multiple linear regression model is constructed using a stepwise regression method. This process is based on a significance level of p≤0.05, where the p-value refers to the probability of observing the current sample data or more extreme data if the null hypothesis is true. By iteratively adding or removing independent variables, combinations of independent variables that significantly contribute to the inflow coefficient are automatically selected, and the final multiple linear regression model is established.
[0036] The goodness of fit of the multiple linear regression model is measured by adjusted R², where adjusted R² is the adjusted coefficient of determination. The closer its value is to 1, the stronger the joint explanatory power of the selected variables on the runoff coefficient. The standardized regression coefficients of each independent variable can be used to compare the magnitude of their influence. If the adjusted R² increases significantly after including multiple variables, it indicates that there is a complementary explanatory effect among these indicators, which jointly affect the runoff mechanism.
[0037] For example, when n=3, there are three key influencing indicators: a, b, and c. Starting from an empty model, the key influencing indicators are sorted according to their correlation magnitude obtained in step 3.1. First, the indicator with the highest correlation, a, is added, followed by indicators b and c in sequence. Models containing a and b, and models containing a, b, and c are constructed respectively. If the adjusted R² of the model containing only a and b is greater than the adjusted R² of the model containing a, b, and c, it indicates that the model fit decreases after adding c. Therefore, indicator c is not significant, while indicators a and b are significant influencing indicators and interact to affect the inflow coefficient.
[0038] Step 3.3, test the multiple linear regression model.
[0039] The accuracy of the multiple linear regression model established in step 3.2 was verified using the Nash-Sutcliffe efficiency coefficient. The Nash-Sutcliffe efficiency coefficient is a core indicator for evaluating the performance of hydrological models, and its calculation formula is as follows:
[0040]
[0041] In the formula, NSE is the Nash-Sutcliffe efficiency coefficient; This is the measured flow generation coefficient value; The generated flow coefficient value simulated by the model; This represents the average value of the measured flow generation coefficient.
[0042] NSE is used to quantitatively assess the reliability of the regression model's accuracy in simulating the runoff coefficient process: the closer its value is to 1, the better the model's simulation capability and the higher its efficiency. For complex physical processes like rainfall-runoff response, which are influenced by multiple factors, when NSE ≤ 0, the prediction results are unsatisfactory. This method can comprehensively evaluate the model's performance in matching the overall trend and peak values, and is a key basis for verifying the effectiveness of the established rainfall-runoff response relationship model. If the number of unsatisfactory results with NSE ≤ 0 exceeds half of the typical flood events, the model's validation effect is poor, and the process returns to step 3.1 to screen key influencing indicators; if more than half of the typical flood events have NSE > 0, the model's validation effect is good, and the results can be used subsequently.
[0043] Step 4: Grouping typical flood events based on key impact indicators
[0044] Step 4.1: Classify the indicators according to the optimal segmentation threshold.
[0045] If the multiple linear regression model is found to be effective after testing, the significant impact indicators identified by the final multiple linear regression model in step 3.2 are used as the grouping criteria, and a supervised binning algorithm based on decision trees is employed for indicator classification. This supervised binning algorithm uses the inflow coefficient as the target variable and the significant impact indicators as features to automatically find the optimal splitting threshold that maximizes the difference in inflow coefficients between groups (such as the F-statistic), thereby achieving objective classification of each indicator.
[0046] Step 4.2: Group typical flood events.
[0047] Based on the optimal segmentation threshold obtained in step 4.1, each significant impact indicator is divided into two levels: high and low. Then, each level is combined to divide all flood events into several homogeneous groups.
[0048] After grouping, statistical tests were used to evaluate the grouping effect: one-way ANOVA was used to test whether the difference in the mean of the inflow coefficient between different groups was statistically significant, i.e., a p-value ≤ 0.05 indicates that the test is significant. Here, the p-value refers to the probability of observing the current sample data or more extreme data when the null hypothesis is true; at the same time, the consistency within groups was verified by calculating and comparing the variances between groups.
[0049] If the differences between groups are significant and the dispersion within groups is small, then the grouping is effective. Different groups represent significantly different runoff generation mechanisms, which can provide a reliable basis for subsequent grouping modeling or differentiated forecasting.
[0050] The beneficial effects of this invention are as follows:
[0051] (1) This invention reveals the mechanism of multi-factor coupling. The method comprehensively quantifies the impact of rainfall characteristics, dynamic changes in the underlying surface, and interventions in water conservancy projects, and through stepwise regression analysis of their interactions, it more realistically reflects the runoff generation and confluence mechanism under the combined effects of human activities and natural conditions.
[0052] (2) This invention improves the accuracy and reliability of flood forecasting models. By combining Spearman correlation analysis with multifactor stepwise regression, the key indicators that have the most significant impact on the runoff coefficient are objectively selected from many potential factors, overcoming the limitations of subjective selection of explanatory variables in traditional methods, and constructing a more accurate quantitative response model.
[0053] (3) This invention realizes the scientific grouping and differentiated forecasting of flood events. Based on the decision tree algorithm, key impact indicators are classified into groups with significantly different runoff generation mechanisms, which provides a basis for constructing categorized forecasting schemes and parameter calibration, and enhances the applicability and stability of the model under different scenarios. Attached Figure Description
[0054] Figure 1This is a schematic diagram of the process for accurate quantitative analysis of the multi-factor coupling effects based on the rainfall-runoff response relationship in this invention;
[0055] Figure 2 This is a unit interval division map of the Minjiang River Basin in an example application of the present invention.
[0056] Figure 3 This is a diagram showing the segmentation effect of the decision tree regression method in an example application of the present invention; Figure 3 (a) in the figure represents the effect of early-stage rainfall segmentation; Figure 3 (b) in the figure represents the effect of segmenting the average rainfall intensity.
[0057] Figure 4 This is a binning diagram of the flow generation coefficient distribution in an example application of the present invention. Detailed Implementation
[0058] The present invention will be further illustrated below with specific examples.
[0059] This invention proposes a multi-factor coupled influence quantification method based on the rainfall-runoff response relationship. The Minjiang River basin covers a vast area of approximately 180,000 km². 2 The underlying surface conditions are complex, and the region is significantly affected by the regulation and storage effects of the two large reservoirs, Zipingpu and Pubugou (total storage capacity 6.444 billion m³). The region experiences strong temporal and spatial variability in rainfall and frequent heavy rainstorms, leading to an extremely complex rainfall-runoff response mechanism. Therefore, to verify the applicability of this method, the middle and lower reaches of the Minjiang River Basin were selected as a research example. Based on rainfall-flood process data and reservoir operation data from the flood season (May-September) of 2015-2021, the key influencing factors of the basin's rainfall-runoff response relationship were analyzed. The main steps are as follows:
[0060] Step 1: Analysis of typical flood events' rainfall and flood characteristics
[0061] Based on typical flood events, basin flood forecasting is carried out. Based on basin hydrological data, typical flood events are divided into categories, and then their rainfall and flood characteristics are analyzed to obtain the basin hydrological characteristics.
[0062] Step 1.1: Classify typical flood events in the basin based on hydrological data. Collect ERA5 reanalysis rainfall, evaporation data, flow data from high-speed stations at the outlet section, and inflow and outflow processes of the two control reservoirs, Zipingpu and Pubugou, from 2015 to 2021 in the Minjiang River basin. Statistical data are collected using a 6-hour minimum rainfall interval, and then the peak flow exceeding the 7000m³ threshold is screened using the excess flow method. 3 The representative flood events of / s were divided into 34 typical flood events, with the flood name numbered according to the flood start time.
[0063] Step 1.2: Organize the flood data and calculate the typical rainfall characteristics of the floods in the middle and lower reaches. The calculation results are shown in Table 1. The rainfall duration was 2-5 days, the average peak rainfall intensity was 48.73 mm, and the average total rainfall was 194.24 mm.
[0064] Table 1 Rainfall characteristic indicators
[0065]
[0066] Step 1.3: Organize the flood data and calculate the flood characteristic indicators of typical flood events in the middle and lower reaches. The calculation results are shown in Table 2. The calculated base flow is 832 m³ / h. 3 / s, with a receding coefficient of 0.0152. The flood volume is relatively large, and major flood events with peak flows exceeding 15,000 m³ / s are prone to occur. The largest was the flood peak on August 16, 2020, which reached 36,800 m³ / s, with an average total flood volume of 216,400 m³.
[0067] Table 2 Flood Characteristic Indicators
[0068]
[0069] Step 2: Calculation of the runoff coefficient and key factor indicators
[0070] Step 2.1: Based on the total rainfall and flood volume obtained in Steps 1.2 and 1.3, the runoff generation coefficient of typical flood events is calculated for each event using Formula 1. The results show that the average runoff generation coefficient in the middle and lower reaches is 0.56, ranging from 0.25 to 0.79, with a range of 0.54 and a standard deviation of 0.14, which is 29% of the mean of 0.49, indicating a high degree of data dispersion. Finally, based on the fact that the runoff generation coefficient is mostly greater than 0.40, and considering that the basin is located in the humid region of Sichuan, it is believed that the runoff generation in the region is mainly characterized by full-storage runoff, and a full-storage runoff model should be considered.
[0071] Step 2.2: Identify and quantify the key factors affecting the runoff generation coefficient. Based on the distribution characteristics of the runoff generation coefficient in Step 2.1, it is determined that the rainfall-runoff response relationship in the watershed is highly complex, requiring consideration of the interactive effects of multiple factors. The characteristic indexes of typical flood events in the interval are calculated according to Formula 2-5: anterior rainfall, peak skewness, average rainfall intensity, rainfall spatial moment, and reservoir storage-to-discharge ratio. The calculation results of the runoff generation coefficient and key factor indexes are shown in Table 3.
[0072] Table 3 Calculation results of runoff coefficient and main factor indicators
[0073]
[0074] Step 3: Analysis of the Indicator's Impact Effect
[0075] The impact of the runoff generation coefficient sequence obtained in step 2.1 and the key factor indicators calculated in step 2.2 on the runoff generation coefficient is systematically analyzed. First, Spearman rank correlation analysis is used to identify key influencing indicators that have a monotonic correlation with the runoff generation coefficient. Then, multifactor stepwise regression analysis is used to reveal the interaction among key influencing indicators in complex watersheds and their joint explanatory power for the runoff generation coefficient.
[0076] Step 3.1: Calculate the Spearman correlation coefficients between the main factor indicators and the runoff generation coefficient. The results are shown in Table 3. Statistical analysis shows that the anterior rainfall amount and average rainfall intensity are significantly positively correlated, with correlation coefficients of 0.54 and 0.50, respectively, and p-values less than 0.05, indicating that these two indicators are key influencing factors on the runoff generation coefficient. Higher soil moisture content leads to stronger runoff generation capacity; greater rainfall intensity facilitates surface runoff formation. However, the correlations between peak rainfall skewness, reservoir storage-to-discharge ratio, and rainfall spatial moment and the runoff generation coefficient are not significant, with p-values greater than 0.05. Therefore, in the middle and lower reaches, the runoff generation process is mainly dominated by two factors: "anterior watershed moisture level" and "local rainfall intensity."
[0077] Table 4. Correlation results between characteristic indicators and Spearman's flow coefficient
[0078]
[0079] Step 3.2: A stepwise regression method was used to construct a model to further analyze the combined interaction effects of multiple factors. Based on the aforementioned correlation analysis results, anterior rainfall (Pa) and average rainfall intensity (I) were selected as independent variables. First, a univariate linear regression model was established by introducing only anterior rainfall, and then a bivariate linear regression model was established by introducing average rainfall intensity. The regression equations, adjusted R², and significance of the two models are shown in Table 4. The change in adjusted R² shows that the explanatory power of the model for the runoff coefficient significantly improved after introducing average rainfall intensity, increasing from 0.271 to 0.379, indicating that the bivariate model has better statistical reliability and was therefore selected as the final model.
[0080] Table 5 Results of Stepwise Regression
[0081]
[0082] Step 3.3: Accuracy verification of the multiple linear regression model. Based on the simulated runoff coefficient and the actual runoff coefficient output by the model, the NSE for typical flood events was calculated. The results are shown in Table 5. Among them, the NSE of 19 typical flood events was greater than 0.5, and the NSE of 6 events was less than 0, indicating that the model has a good simulation effect.
[0083] Table 6. Accuracy Test Results of the Multiple Linear Regression Model
[0084]
[0085] Step 4: Grouping typical flood events based on key impact indicators
[0086] Step 4.1: The optimal segmentation threshold (the point that maximizes the effect difference between the high and low groups) for the previous influences on rainfall and average rainfall intensity is detected using decision tree regression. Only one segmentation is performed, classifying the indicators. After classification, each sample is labeled as either high or low. The segmentation results are shown in Table 6. Figure 3 As shown.
[0087] Table 7 Optimal splitting thresholds for decision tree regression
[0088]
[0089] Step 4.2: Divide the runoff generation coefficient data into 4 groups and test the grading effect. Based on the grading results of previous impact rainfall and average rainfall intensity, all flood samples were divided into 4 groups. The grouping statistics are shown in Table 7. Figure 4 One-way ANOVA was used to examine the differences in mean runoff coefficients among different groups, yielding a p-value of 6.27e-08. At the 0.05 significance level, a significant difference existed between the group means. The different groups exhibited a clear gradient in runoff coefficients: the group with both "low" antecedent rainfall and "low" average rainfall intensity (Group 1) had the lowest mean runoff coefficient (0.338), while the group with both "high" was the highest (Group 4) (0.645), showing a trend of increasing runoff coefficients with increasing influencing factor intensity. The variance within each group was small (between 0.00325 and 0.00960), indicating low dispersion and high consistency of runoff coefficients within the groups. The sample distribution was relatively uniform, with sample sizes ranging from 6 to 10, demonstrating a certain degree of statistical representativeness. In conclusion, this grouping method effectively constrains the range of runoff coefficient values for each group, and the grouping results are reasonable and usable.
[0090] Table 8 Flood Event Grouping Results
[0091]
[0092] The above analysis results demonstrate that the method proposed in this invention, based on the actual hydrological and meteorological conditions of the Minjiang River Basin, objectively screens and accurately identifies antecedent rainfall and average rainfall intensity as key indicators regulating the rainfall-runoff response relationship in the basin by combining Spearman correlation analysis with multi-factor stepwise regression. By establishing optimal segmentation thresholds for these two indicators, reasonable grouping of flood events is achieved, effectively revealing the differences in runoff generation mechanisms under different hydrological conditions. This grouping framework provides a basis for constructing categorized forecasting schemes and parameter calibration, enhancing the applicability and stability of the model under different scenarios.
[0093] The embodiments described above represent only one specific implementation of the present invention and should not be considered as limiting the scope of the patent rights of the present invention. It should be emphasized that those skilled in the art can still make various modifications and improvements without departing from the basic principles of the present invention, and these are all included within the scope of the patent rights of the present invention.
Claims
1. A method for determining a rainfall-runoff response relationship of a watershed based on quantitative analysis of multi-factor influence, characterized in that, The method for determining the rainfall-runoff response relationship of a basin comprises the following steps: Step 1: Analysis of rainfall and flood characteristics of typical flood events According to the typical flood events, the basin flood forecasting is carried out, the typical flood events are divided based on the basin hydrological data, and then the rainfall and flood characteristics are analyzed to obtain the basin hydrological characteristics; Step 2: Calculation of runoff coefficient and main factor index The runoff coefficient is used to represent the conversion relationship between rainfall and runoff. First, the runoff coefficient of the typical flood event sequence is calculated and analyzed, and then the key factor index affecting the runoff coefficient is identified and calculated; Step 3: Analysis of the influence effect of the index The influence effect of the key factor index on the runoff coefficient is analyzed. First, the key influence index having a monotonic correlation with the runoff coefficient is identified by using Spearman rank correlation analysis, and then the interaction between the key influence indexes in the complex basin and the joint explanation ability of the key influence indexes on the runoff coefficient are revealed by using multi-factor stepwise regression analysis; Step 4: Grouping of typical flood events based on the key influence index 2.The method of determining the rainfall-runoff response relationship of a watershed based on multi-factor influence quantitative analysis according to claim 1, characterized in that, The step 1 specifically comprises the following steps: Step 1.1: Dividing typical flood events according to basin hydrological data First, the observation data of each rainfall station and flow station in the basin are summarized according to a unified time step; Then, the independent flood events are identified according to the flow process line, the start and end time of each flood is determined, the flood peak flow threshold is set according to the basin condition, and the representative flood events exceeding the flood peak flow threshold are selected by using the super-quantitative method; finally, the selected flood events are numbered uniformly to obtain the sequence of typical flood events; Step 1.2: Calculation of rainfall characteristic index The rainfall characteristic index is calculated for each typical flood event, including the rain peak occurrence time, rainfall duration, rain peak intensity and total rainfall; Step 1.3: Calculation of flood characteristic index The flood characteristic index is calculated for each typical flood event, including the flood peak flow, total flood volume, peak occurrence time and flood duration; the recession curve is drawn based on the recession process, the base flow of the basin is calculated, and the total flood volume is calculated by deducting the base flow.
3. The method of claim 2, wherein the method is characterized by, The step 2 specifically comprises the following steps: Step 2.1: Calculation and analysis of the runoff coefficient of the basin flood event The runoff coefficient of the typical flood event sequence is calculated according to formula (1) based on the total rainfall and total flood volume obtained in steps 1.2 and 1.3; the extreme value, standard deviation and mean value of all flood runoff coefficients are calculated and analyzed to analyze the variation range and distribution characteristics; when the extreme difference of the runoff coefficient is greater than 0.5 or the standard deviation is greater than the mean value, it indicates that the uncertainty of the basin flood forecasting is large; ; In the formula, C is the runoff coefficient, R is the runoff depth, and P is the rainfall depth; Step 2.2: Identification and quantification of key factor index affecting runoff coefficient The key factor index affecting the runoff coefficient is identified and quantified. Based on the distribution characteristics of the runoff yield coefficient in step 2.1, the complexity of the rainfall-runoff response relationship of the basin is judged; The following five main factor indexes influencing the rainfall-runoff response relationship are considered as the key factors influencing the runoff yield coefficient: the previous influencing rainfall, the rainfall center distance, the average rainfall intensity, the rain peak time skewness, and the reservoir storage-discharge ratio.
4. The method of claim 3, wherein the method is characterized by, In step 2.1, the runoff yield coefficient ranges from 0 to 1, reflecting the proportion of rainfall converted into runoff; the runoff yield coefficient range can also be used to assist in identifying the runoff mode.
5. The method of claim 3, wherein the method is characterized by, In step 2.2, the five key factor indexes are obtained as follows: The previous influencing rainfall is calculated by quantifying the cumulative effect of recent rainfall through formula (2); the rainfall center distance is calculated by quantifying the spatial distance between the spatial distribution centroid of the average rainfall of the basin and the outlet section of the basin, and the centroid is identified based on the ArcGIS spatial analysis tool; the average rainfall intensity is calculated according to the ratio of the total rainfall to the rainfall duration, as shown in formula (3); the rain peak time skewness is an index designed to describe the asymmetry of the rain peak time position, calculated according to formula (4); and the reservoir storage-discharge ratio is calculated according to formula (5) by a representative reservoir; ; wherein is the antecedent rainfall for the previous time period; is the antecedent rainfall for the previous time period; K is a soil moisture decay coefficient determined by the characteristics of the watershed; is the rainfall for the previous time period; I is the average rainfall intensity; is the accumulated rainfall variation within the time period; is the observation time step; S is the time skewness of the rain peak; tp is the duration from the start of rainfall to the occurrence of the rain peak; Tp is the total duration of rainfall; is the reservoir storage ratio; Q is the storage and discharge amount; and W is the total amount of water into the reservoir.
6. The method of claim 5, wherein the method is characterized by, Step 3 is specifically: Step 3.1, Spearman rank correlation analysis; Based on the key factor indexes and the runoff yield coefficient of each flood obtained in step 2, the Spearman correlation coefficient is calculated, the statistical significance is judged, the correlation size is sorted, and the key factor indexes with a p value less than 0.05 are selected to obtain n key influencing indexes, 1≤n≤5; the p value refers to the probability of observing the current sample data or more extreme data under the assumption that the original hypothesis is true; Step 3.2, multiple factor stepwise regression analysis; A multiple linear regression model is constructed by using the stepwise regression method with the runoff yield coefficient as the dependent variable and the n key influencing indexes selected in step 3.1 as the independent variables; this process is based on the standard of a significance level p≤0.05, where the p value refers to the probability of observing the current sample data or more extreme data under the assumption that the original hypothesis is true, and through iterative addition or removal of independent variables, a combination of independent variables that significantly contributes to the runoff yield coefficient is automatically selected, and a final multiple linear regression model is established; Step 3.3, test the multiple linear regression model; The Nash-Sutcliffe efficiency coefficient was used The accuracy of the multiple linear regression model established in step 3.2 was tested.
7. The method of claim 6, wherein the method is characterized by, The goodness of fit of the multiple linear regression model in step 3.2 is measured by the adjusted R², where the adjusted R² is the adjusted determination coefficient; the standardized regression coefficients of each independent variable are used to compare the size of their influence; if the adjusted R² increases after multiple variables are included, it indicates that there is a complementary explanatory effect among these indexes, which jointly affect the runoff mechanism.
8. The method of claim 7, wherein the method is characterized by, In step 3.3, The calculation formula is: ; where NSE is the Nash-Sutcliffe efficiency coefficient; is the observed value of the coefficient of runoff; is the simulated value of the coefficient of runoff; is the average value of the coefficient of runoff; When NSE≤0, the prediction result is not satisfactory; if the number of unsatisfactory results with NSE≤0 is more than half of the number of typical flood cases, the multiple linear regression model test effect is not good, and the key influencing indexes are reselected in step 3.1; if the NSE of more than half of the typical flood cases is greater than 0, the multiple linear regression model test effect is good, and the results can be used subsequently.
9. The method of claim 7, wherein the method is characterized by, Step 4 is specifically: Step 4.1, index classification according to the optimal segmentation threshold; If the multiple linear regression model is valid after the test, the significant influence indicators identified by the final multiple linear regression model in step 3.2 are taken as the grouping basis, and a supervised binning algorithm based on a decision tree is used for index classification; Step 4.2, grouping of typical flood events; According to the optimal segmentation threshold obtained in step 4.1, each significant influence indicator is divided into high and low levels, and then each level is combined to divide all flood events into several homogeneous groups; After grouping, the statistical test method is used to evaluate the grouping effect; If the difference between groups is significant and the dispersion within groups is small, it is proved that the grouping is effective, and different groups represent significantly different runoff generation mechanisms, which provides a basis for subsequent grouping modeling or differentiated prediction.
10. The method of claim 7, wherein the method is characterized by, In step 4: In step 4.1, the supervised binning algorithm takes the runoff coefficient as the target variable and the significant influence indicators as the characteristics to automatically find the optimal segmentation threshold that maximizes the difference between the runoff coefficients of different groups, thereby achieving objective classification of each indicator; The evaluation of the grouping effect in step 4.2 is as follows: the one-way analysis of variance is used to test whether the difference between the mean values of the runoff coefficients of different groups is statistically significant, that is, p value ≤ 0.05 indicates that the test has statistical significance, where p value refers to the probability of observing the current sample data or more extreme data under the assumption that the null hypothesis is true; at the same time, by calculating and comparing the inter-group variance, the consistency within the group is verified.
