Precipitation extremes on multiple time scales -- Bartlett-Lewis Rectangular Pulse Model and Intensity-Duration-Frequency curves

Abstract

For several hydrological modelling tasks, precipitation time-series with a high (i.e. sub- daily) resolution are indispensable. This data is, however, not always available and thus model simu- lations are used to compensate. A canonical class of stochastic models for sub-daily precipitation are Poisson-cluster processes, with the Bartlett-Lewis rectangular pulse model (BLRPM) as a prominent representative. The BLRPM has been shown to well reproduce certain characteristics found in ob- servations. Our focus is on intensity-duration-frequency relationship (IDF), which are of particular interest in risk assessment. Based on a high resolution precipitation time-series (5-min) from Berlin- Dahlem, BLRPM parameters are estimated and IDF curves are obtained on the one hand directly from the observations and on the other hand from BLRPM simulations. Comparing the resulting IDF curves suggests that the BLRPM is able to reproduce main features of IDF statistics across sev- eral durations but cannot capture singular events (here an event of magnitude 5 times larger than the second larges event). Here, IDF curves are estimated based on a parametric model for the duration dependence of the scale parameter in the General Extreme Value distribution; this allows to obtain a consistent set of curves over all durations. We use the BLRPM to investigate the validity of this approach based on simulated long time series.