Frequency Analysis of Extreme Sub-Daily Precipitation under Stationary and Non-Stationary Conditions across Two Contrasting Hydroclimatic Environments

Abstract

Abstract. Observed sub-daily precipitation intensities from contrasting hydroclimatic environments in the USA are used to evaluate temporal trends and to develop Intensity-Duration-Frequency (IDF) curves under stationary and non-stationary climatic conditions. Analyses are based on observations from two United States Department of Agriculture (USDA)-10 Agricultural Research Service (ARS) experimental watersheds located in a semi-arid and a temperate environment. We use an Annual Maximum Series (AMS) and a Partial Duration Series (PDS) approach to identify temporal trends in maximum intensities for durations ranging from 5-to 1440-minutes. A Bayesian approach with Monte Carlo techniques is used to incorporate the effect of non-stationary climatic assumptions in the IDF curves. The results show increasing trends in observed AMS sub-daily intensities in both watersheds whereas trends in the PDS observations are mostly positive in the 15 semi-arid site and a mix of positive and negative in the temperate site. Stationary climate assumptions lead to much lower estimated sub-daily intensities than those under non-stationary assumptions with larger absolute differences found for shorter durations and smaller return periods. The risk of failure (R) of a hydraulic structure is increased for non-stationary effects over those of stationary effects, with absolute differences of 25% for a 100-year return period (T) and a project life (n) of 100 years. The study highlights the importance of considering non-stationarity, due to natural variability or to climate change, in 20 storm design.

Can Regional Climate Modeling Capture the Observed Changes in Spatial Organization of Extreme Storms at Higher Temperatures?

“…Intensifying short‐duration rainfall extremes in a warming climate have been confirmed in numerous studies based on both observations and climate model projections (Beck et al, ; Demaria et al, ; Leahy & Kiely, ). This is of great concern to society as it can lead to an increased risk of flash flooding (Westra et al, ).…”

Response to Reviewer Comments

“…The method is flexible and robust and avoids any modification of the original dependence structure of the data which constitutes one of the strengths of the method. P4L21-26: T = 1/P and T = 1/rP are not definitions but expected values obtained under specific model assumptions (see, Fernández and Salas, 1999;Douglas et al 2002;Serinaldi 2015;Volpi et al 2015;Salvadori et al 2016) Thank you pointing this out. We will replace the equations in the manuscript with Expected values T=E(x)=.…”

“…For each Monte Carlo realization (40000), the time-variant location parameter (in the case of the GEV) linearly varies in time ((t) = 1t + 0 values). The model parameters are used to estimate nonstationary precipitation intensities for different exceedance probabilities (AghaKouchak et al, 2013;.…”

Sub-Daily Rainfall Intensity Extremes: Evaluating Suitable Indices at Australian Arid and Wet Tropical Observing Sites

“…Most cells had lifetimes of <1 h, and showed highest intensities at about 0.3 of the event duration (i.e.,~15-20 min into the lifetime of the rainfall event). In light of this kind of behaviour, in a growing number of studies, sub-hourly data are used to characterise rainfall extremes [51][52][53][54][57][58][59][60][61]. In a recent study, for instance, [62] used 10 min data (as well as these data aggregated further to yield 30 min and 60 min totals) in a study of extreme rainfall in Sicily.…”

“…In many studies, particular percentiles of the distribution of intensities (here symbolised 'Q', for quantile), derived using various ADs, are used as indices of extreme intensity. These include Q90, Q95, Q98, Q99, and Q99.9 [57,59,62,[67][68][69][70]. These percentiles, most commonly Q95, are also frequently adopted as a threshold to be used in generating partial-series data using peak-over-threshold (PoT) methods.…”