Regional Variability in the Performance of Annual Maxima vs. Peaks-Over-Threshold Methods for Predicting Frequent Floods

Much geomorphic research on rivers focuses on the role of frequent floods (e.g., with return periods between one and two years), which have been shown in many regions to simultaneously perform sufficient geomorphic work as well as occur often enough, so that they tend to determine the shape of the channel. As compared to the Annual Maxima (AM) method, using the Peaks-Over-Threshold (POT) method for flood frequency analysis (also known as the Partial Duration method) allows for the inclusion of a larger number of peak values from the series of past flow observations, resulting in a better estimation of the probabilistic model. As it only considers events above a threshold, POT also decreases the likelihood of incorporating smaller events, when relatively dry years occurred within the period of observations. In the AM method, such smaller events could potentially come from a different population and have an inordinate influence on the predicted floods, introducing variability. Because of all these reasons, the POT approach should result in sounder statistical analyses when predicting frequent floods, with relatively short average recurrence intervals (ARIs). However, much geomorphic research into channel-forming floods has traditionally used AM instead of POT, presumably because it is much easier to obtain annual maxima data and perform frequency analyses when there are as many data as years in the record, while there is subjectivity in choosing an adequate threshold for POT analyses.

In this work, we study the variability in peak flow estimates for frequent (return period < 3 years) events, using both AM and POT, over multiple regions in the US. The objectives are: i) to search for homogeneous hydrological regions where the relation between the two methods is similar, and ii) to study the stability of predictions obtained with the two approaches, when considering different record lengths. The former objective aims at exploring how external factors related to the geographical location and the characteristics of the basin affect discrepancies in the results achieved by the two methods, such as the climate and the size of the basin. The former affects the magnitude and average number of other flooding events, neglected by AM, that occur every year besides the annual maximum. The latter influences the extent to which different types of rainfall events, with different spatial coverages, can involve the watershed. This insight might lay some groundwork for introducing “correction coefficients” for AM-based predictions of relatively frequent floods, depending on the characteristics of the study area. The latter objective is intended to test the stability of the statistical model and check whether POT leads to less variable predictions than AM.

Special care is adopted in two crucial aspects that may introduce bias in the analysis: i) the choice of the case-study gaging stations, in order to minimize any human impact on the studied flow time series, and ii) the methodology for selecting the flood threshold in the POT method, aimed at avoiding subjective decisions.