Influence of threshold selection in modeling peaks over threshold based nonstationary extreme rainfall series

Recent studies report that the extreme rainfall characteristics in most parts of the globe exhibit temporal non-stationarity. Therefore, modeling the nonstationary behavior of extreme rainfall for different water resources applications is vital. When modeling non-stationarity in extreme rainfall series, previous studies consider a single threshold value in the peaks over threshold (POT) approach to extract extreme rainfall series. However, extreme rainfall series extracted with different threshold values may have a different degree of non-stationarity. Consequently, it is essential to understand the effect of threshold selection in modeling peaks over threshold based nonstationary extreme rainfall series. This study aims at quantifying the threshold uncertainty (i.e., uncertainty in extreme rainfall return levels due to the choice of the threshold) in modeling peaks over threshold based nonstationary extreme rainfall series using the Generalized Pareto Distribution (GPD). To study the threshold uncertainty, extreme rainfall series over India from the India Meteorological Department's high-resolution gridded (0.25 degrees Longitude x 0.25 degrees Latitude) daily rainfall dataset is used. For modeling non-stationarity in extreme rainfall series, different indices representing four physical processes, namely, global warming, El Nino-Southern Oscillation (ENSO), Indian Ocean Dipole (IOD) and local temperature anomaly are linked with the scale parameter of the GPD. Uncertainties in extreme rainfall return levels calculated over India indicate that the uncertainty created due to the choice of threshold is 54% higher under the nonstationary condition when compared to the stationary condition.