Application of a 21-year Radar Rainfall Record to Assess Changing Probability of Extreme Rainfall in the Mid-Atlantic Region

ABSTRACT ONLY - The prospective impact of climate change on extreme rainfall and flood frequency has led to increasing interest in updating rainfall intensity-duration frequency relationships such as those accessed from the NOAA Atlas 14 Point Precipitation Frequency database. The need for improved understanding is highlighted by events such as the 2016 and 2018 Ellicott City floods, both of which have been described as exceeding a 1000-year recurrence interval for rainfall durations of 2 to 3 hours, and both of which resulted in estimated discharge peaks that exceed the envelope curve for watersheds of comparable size in the mid-Atlantic region.
In this presentation we report analyses of a two-decade radar rainfall data set with spatial resolution of 1 km2 and temporal resolution of 15 minutes over a domain covering 11,640 km2 in the Baltimore-Washington metropolitan area. We developed bias-corrected radar rainfall fields using observations from the Sterling, Virginia WSR-88D radar and a network of rain gages. The data set covers the period between 2000 and 2020, focusing on the warm season (April - September). The period spans the transition from conventional reflectivity-based estimates of rainfall to estimates based on polarimetric fields. An overlap period from 2012-2015 is used to assess the impacts of changing measurement technologies on extreme rainfall estimates.
We use a peaks-over-threshold approach grounded in the Gumbel distribution to examine rainfall extremes at 15, 30, 60, 120 and 180 minutes for each pixel in the domain. We map the spatial distribution of estimated rainfall depths corresponding to 2, 10, and 50-year recurrence intervals for durations between 15 minutes and 3 hours, and estimate the slope and magnitude of the 21-year change over the study domain. We observe an increasing trend in the rainfall totals associated with specific combinations of duration and recurrence interval as well as increasing frequency of rainfall accumulations exceeding threshold values. We also observe complex spatial patterns in rainfall amounts and trends over time. Analysis of annual maximum streamflow records and frequency of discharge peaks over threshold from watersheds within the study domain reveal trends that mimic the trends derived from corresponding rainfall estimates over the same time period.