Downscaling CMIP6 Models with Localized Constructed Analogs (LOCA) Version 2: an Improved Representation of Precipitation Extremes
Abstract
Localized Constructed Analogs (LOCA) is a statistical downscaling technique that produces a finer spatial resolution representation (6 km here) of Global Climate Model (GCM) data using a library of fine resolution historical observations, while still preserving the original climate change signal from the GCM at large regional scales. LOCA was originally applied to RCP4.5 and RCP8.5 for 31 models in the CMIP5 era. An updated version of LOCA, version 2, has now been applied to 27 GCMs from the CMIP6 archive, covering the historical period (1950-2014) and 2015-2099 under SSPs 245, 370, and 585, with up to 10 ensemble members per model. The domain covers the CONUS and extends into central Mexico and southern Canada, with daily time resolution of Tmin, Tmax, and precipitation (P). Enhancements in LOCA2 improve the depiction of daily precipitation extremes by using a new precipitation training data set that better represents daily precipitation extremes than the training data used in the original LOCA product. The other improvement is an ensemble bias correction approach that better preserves extreme events in models that have more than one ensemble member. These improvements are critical to how informative LOCA2 is for flood events and even for annually-averaged runoff, since surface runoff efficiency is nonlinear function of the precipitation distribution.
After LOCA2 downscaling, 5-year return values of daily P are projected to increase 10, 14, and 16% by the end of this century in the multi-model ensemble average for SSPs 245, 370, and 585, respectively, averaged across the CONUS exclusive of Arizona and New Mexico, where changes are smaller. Changes are greatest in winter (DJF), averaging 14, 19, and 23% for the three SSPs, but near zero in summer. At a 50-yr return value the increases are modestly greater in winter, spring, and autumn, and considerably greater in summer, which shows increases of about 7% across SSPs. Remarkably, in most locations in the CONUS, the precipitation distribution shows an increasing number of days/year that fall in the (historically defined) 99.9th percentile of wet days, with a doubling of the number of very wet days for SSP370 by 2100. These improvements in LOCA2 result in daily precipitation extremes that are more severe than found in LOCA-CMIP5, even for a similar level of anthropogenic emissions.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.A53B..01P