Storm characteristics and extreme sub-daily precipitation statistics over CONUS

The analysis of short-duration precipitation extremes is of foremost importance as heavy precipitation is directly related to many hazards, e.g. flash floods, landslides and crop damage. Here, we adopt an extreme value framework based on the concept of ordinary events, defined as independent realizations of the process of interest. In particular, we aim at investigating the link between the characteristics of ordinary storms (e.g. seasonality, average duration, autocorrelation) and the statistics of the emerging extremes at sub-daily durations (1-24 h). We used the Global Sub-Daily Rainfall (GSDR) dataset, which provides quality controlled hourly precipitation data from rain gauges over the Contiguous United States (CONUS). 

First, we tested the hypothesis that a Weibull distribution can describe the tail of ordinary events and independently reproduce the annual maxima. Then, we quantified the portion of ordinary events, termed tail hereinafter, which share the statistical properties with annual maxima. Analysis of the storm characteristics show shorter average duration storms (< 12h) in the central portion of CONUS, between latitudes 90ºW and 105ºW. Seasonality analysis showed predominance of summer events in all central and eastern areas, with exception to a region encompassing the northwestern areas of the southern US states, which are dominated by spring events. On the western coast, winter events dominate the tail of the distribution of the ordinary events. The majority of these events happened in the afternoon (12PM to 6PM) or night (6PM to 12AM). The parameters describing our extreme value distribution revealed insightful features. The scale parameter of the Weibull distribution describing the tail followed the local climatology, with higher values over the southeast of CONUS (region characterized for high intensity precipitation), and small values over the northwest. The shape parameter indicates heavier-tailed distributions on the north and central regions of the US, as opposed to the majority of stations CONUS-wide. On average, the number of events per year is larger in the east (50 to 100 events per year) when compared to the west (0 to 50 events per year) . 

Further analyses include investigating the influence of storm properties in the parameters of our extreme value distribution. This link, if proven significant, can be used to establish predictors for extreme precipitation statistics that stem from characteristics of ordinary storm events.