Demonstrating a Preference for Dynamical Downscaling over Statistical Downscaling for Extreme Rainfall Projections
On this page:
Estimates of potential changes in future extreme rainfall events support communities’ flood adaptation efforts and are increasingly desired for stormwater and transportation infrastructure redesign. Statistical downscaling (SD) is a favored method to improve spatial and temporal distributions of rainfall from global climate models (GCMs) because of its relatively low computational requirements, high spatial resolution, and multiple-member ensemble availability. However, the observational training data used to develop the statistical relationships in SD imposes the stationarity assumption on the projected dataset which is a limitation when considering extreme rainfall events. Dynamical downscaling (DD) uses a dynamical atmospheric model driven by boundary conditions from a GCM, thus it is not constrained by the stationarity assumption and can better resolve rainfall extremes at a spatial resolution of the dynamically downscaled simulation.
Here we select overlapping GCMs that were downscaled using multiple methods to explore uncertainty in rainfall extremes magnitude toward the development of future rainfall Intensity- Duration- Frequency (IDF) curves to support stormwater infrastructure design. This study compares output from statistical downscaling methods that apply different gridded rainfall observations for training (LOCA and MACAv2-LIVNEH at 6-km, MACAv2-METDATA at 4km), dynamical downscaling experiments from CORDEX (25-km), and our dynamical downscaled experiment using the Weather Research and Forecasting (WRF) model at 36-km. This study focuses on percent changes in the daily extreme rainfall by the end of the 21st century for the RCP8.5 scenario. Preliminary analyses reveal significant differences in rainfall extremes magnitude and spatial distribution of the changes between the various downscaling methods. Additionally, we show that multi-model, averaged, datasets obscure variability of the ensemble and subdues regional inconsistencies. This abstract does not reflect the views and policies of the U.S. Environmental Protection Agency.