A Robust Depth-Duration-Frequency Model for Analysis of Extreme Precipitation Events, with Application to Past and Projected Future Climates in Ireland

A simple but robust depth-duration-frequency (DDF) model is presented to reveal the asymptotic characteristics of extreme but short-lived (sub-daily) precipitation events that satisfy a peak over threshold (POT) size criterion. Our objective is to reliably estimate the return periods for events of a given intensity (as measured by rainfall depth and duration).

For each depth threshold and duration period (ranging from 15 minutes to 24 hours), the number of qualifying POT events is simply counted over multi-year periods, whether from observations or model output, at each location separately. The distribution of events as a function of their size above the threshold is modelled by a generalized Pareto distribution (GPD), following standard extreme value theory. Those exceedance distributions are shown, to a good approximation, to be independent of location within Ireland. This justifies the aggregation of exceedances from multiple locations, which is a key feature of the model. Aggregation acts as a data multiplier, enabling more reliable estimation of GPD fits and return periods.

The model is applied to intense precipitation observations spanning 30–64 years at 23 stations in Ireland. Three-hourly output from an ensemble of CMIP5 global climate simulations, downscaled to high-resolution over Ireland, were also used to compute both historical and projected future intense event return periods under two different emission scenarios. 

Future numbers of events per time-period are projected to increase by 20-80%, depending on event threshold and duration, location, emission scenario and time-period. Return periods are projected to shorten by factors of 2 or more for the most intense events, as illustrated by return period maps for events of any given size.

Return period uncertainty is quantified mainly by the spread among the different CMIP5 models.  For any given model, however, robustness is demonstrated by the convergence of the empirical exceedance distributions as more stations (or grid-points) are aggregated, which then leads naturally to convergence of the GPD fits.