| 15 Oct 2025
Study on Critical Rainfall for Flash Flood Disasters in Small Watersheds of the Three Gorges Reservoir Area: A Case Study of Futian Small Watershed in Wushan County of Chongqing
Abstract. Taking the Futian Small Watershed in Wushan, within the Three Gorges Reservoir area, as the research object, this study utilized hourly rainfall data from 2010 to 2023 collected at the Futian Small Watershed and nearby rainfall stations, historical disaster information on mountain flood disaster processes, digital elevation models, land use data, and other relevant information. Statistical analysis methods such as the single-station critical rainfall method, regional critical rainfall method, probability distribution method, and the hydrodynamic model FloodArea, were employed to simulate and calculate the critical rainfall amounts leading to disasters. Results indicate that the trends of critical rainfall amounts leading to disasters calculated by various methods are generally consistent. However, at different time scales, the critical rainfall amounts calculated by different methods exhibit variations. The FloodArea simulation yields the smallest critical rainfall amounts for 1-hour, 2-hour, 24-hour durations; the single-station critical rainfall method provides the smallest values for 5-hour, 6-hour, and 12-hour durations; the regional critical rainfall method gives the smallest results for 3-hour, 4-hour durations. Statistical methods can swiftly and efficiently establish critical rainfall amounts leading to disasters at different time scales, the FloodArea model can more precisely depict the precipitation-runoff processes of mountain flood disasters. Therefore, by integrating statistical methods with hydrological model simulations to leverage their respective strengths, we can more accurately determine the critical rainfall amounts leading to mountain flood disasters.
The manuscript assesses the estimation of critical precipitation thresholds related to mountain flood disasters using four different methods in a watershed within the Three Gorges Reservoir area. The authors argue that integrating statistical methods with hydrological model simulations enables a more accurate determination of critical rainfall volumes that can trigger flood-related disasters.
The topic is of high relevance, particularly given the growing importance of flood early warning systems and disaster risk management. However, despite the importance of the subject, the manuscript currently lacks sufficient clarity in the description of data and methods, which prevents its publication in its present form.
In Section 2.2, the authors present a list of historical disasters, including descriptive information about each event. It is not clear whether this list was used solely for event identification and selection or whether it also served as validation for the model results. For example, in lines 197–199, the authors state that model results are in good agreement with actual conditions, yet the manuscript does not explicitly describe the evaluation methods or performance metrics used to support this claim.
Regarding the modeling approach, the study employs a CN-based hydrological model. While the Curve Number (CN) method is widely used, it is an empirical approach; thus, a discussion on the accuracy and representativeness of the estimated parameters for the local characteristics of the case study area is necessary. Furthermore, the manuscript notes that CN values depend on antecedent soil moisture conditions. However, in the development of the mathematical models for estimating inundation depth and cumulative areal relationships (Figure 5 and subsequent equations), it remains unclear how antecedent humidity conditions were accounted for in the regression analyses. Additionally, the models are presented as quadratic regression curves, while linear correlation coefficients are reported (lines 238–240), which introduces some inconsistency that should be clarified.
Section 3.2, which describes the statistical approaches, would benefit from a clearer explanation of the data structures used for each method. For example, in the Single Station approach, were annual maximum precipitation series derived for each gauge station? Similarly, how were the data organized for the Probability Distribution Method? In the Regional approach, it appears that only precipitation data from disaster events were used, which should be explicitly stated.
Regarding the Single Station approach, it is also unclear whether using a spatially averaged precipitation series (lines 294–296) still qualifies as a “single station” analysis or effectively represents a regionalized dataset.
Finally, in the Probability Distribution Method, the authors assess the goodness of fit of several theoretical distribution functions. Table 5 suggests that this comparison was based on the magnitude of the Kolmogorov–Smirnov (KS) statistic. However, the KS statistic is a relative measure dependent on the tested reference distribution; therefore, D-values from different distributions are not directly comparable. More appropriate model comparison criteria, such as the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), Chi-Square test, or Anderson–Darling test, should be used for a more robust and statistically valid evaluation of goodness of fit.