A nonstationary index-flood technique for estimating extreme quantiles for annual maximum streamflow

Abstract

Summary The magnitude and timing of peak streamflow events may be affected by land-use changes along with climate change, thus leading to nonstationarity in the records. Temporal trend, along with change-points, in peak flow records can affect the accuracy of quantile estimates; therefore, these issues should not be disregarded. Commonly used techniques for pooled flood frequency analysis do not account for nonstationarity found in the data recorded for members of a region. To overcome this shortcoming, the objective of this research is to introduce a trend centered pooling approach for regionalization in which pooling groups are created based on the form of trend found in the at-site data. The approach involves the formation of regions comprised entirely of sites exhibiting either statistically significant increasing or decreasing trends. Regional parameter estimates are determined using a maximum likelihood approach, which is carried out with the assumption of second-order nonstationarity. The technique was applied to four homogenous regions all located in differing hydroclimatological Canadian regions. The uncertainty of quantile estimates calculated through the implementation of this technique was established using a balanced regional vector resampling approach. The results indicate that there is less uncertainty in quantile estimates found through the application of the trend centered pooling approach when compared to a regional stationary analysis of the same regions. The potential for overestimation/underestimation of design quantiles in the presence of significant regional nonstationarity (i.e. decreasing/increasing trends) was elucidated.