Downscaled Projections of Extreme Rainfall in New York State

Future changes in the frequency and magnitude of extreme precipitation have profound implications for urban and rural development, public infrastructure, watershed management, agriculture, and human health. In consideration of these socioeconomic issues, the Northeast Regional Climate Center (NRCC) is partnering with the New York State Energy Research and Development Authority (NYSERDA) to compare various methods of downscaling global climate model (GCM) output and create extreme rainfall projections that can be incorporated into climate change adaptation planning. Primary objectives of this study include: 1) evaluation of downscaling method–climate model combinations to assess their ability to replicate present-day rainfall extremes, 2) downscaling of projected rainfall extremes for future time periods, 3) quantification of methodological and climate model uncertainties, and 4) outreach and development of web-based tools to make results accessible to users. At this time, we have completed the first of several candidate downscaling procedures, which involves bias correcting dynamically downscaled GCM output.

Our extreme rainfall projections are based on data from two primary sources: 1) daily precipitation observations from 157 first-order and cooperative observer stations in New York and surrounding areas of adjacent states and Canada, and 2) gridded daily precipitation estimates from regional climate models (RCMs) run at 50-km resolution and driven by atmosphere–ocean general circulation models (AOGCMs). The seven RCM simulations are obtained from the North American Regional Climate Change Assessment Program (NARCCAP) and include output for base (1970–1999) and future (2040–2069; A2 emissions scenario) periods. We began our analysis by constructing partial duration series (PDS) of 1-day extreme rainfall events at each station and the nearest RCM grid box for the base period. Recurrence interval rainfall amounts at each station and RCM grid box were then estimated using two statistical fitting approaches. The station-based Beta-P approach employs a maximum likelihood distribution and assumes that the PDS frequency distributions at all stations are statistically independent. The regionalized L-moments approach employs a generalized extreme value distribution and groups stations together based on similarities in the shape and scale parameters of their PDS frequency distributions. In order to convert average precipitation over a grid box to point values of precipitation, we applied areal reduction factors (ARFs) to the RCM recurrence interval rainfall amounts. Next, we used quantile–quantile mapping to calculate the remaining base period biases between station and ARF-adjusted RCM recurrence interval rainfall amounts. Future projections of extreme rainfall at each station were obtained by applying these bias correction factors to the future period ARF-adjusted RCM recurrence interval rainfall amounts.