C13B-0449:
Patterns in Polar Climate Change: How Well Do GCMs Capture Melt-Season Variability?

Monday, 15 December 2014
David B Reusch, New Mexico Institute of Mining and Technology, Department of Earth and Environmental Science, Socorro, NM, United States, David P Schneider, National Center for Atmospheric Research, Boulder, CO, United States, Christopher Charles Karmosky, University of Tennessee Martin, Department of Agriculture, Geosciences and Natural Resources, Martin, TN, United States and Derrick Julius Lampkin, Univesity of Maryland, College Park, United States
Abstract:
GCM skill in the climate change-sensitive polar regions is a key factor in using them to predict future changes in ice sheet/shelf surface melting. Knowing why a GCM is wrong (or right) versus observed climate is also key. We address these topics by using a neural network-based tool, self-organizing maps (SOMs), to evaluate the variability and skill of modern GCM melt-season surface climates in Antarctica (January) and Greenland (July).

Three scenarios test GCMs in contemporary polar settings: how well do models capture variability in a reference dataset (the ERA-Interim reanalysis, ERAI); locating GCMs in a “universal” data space covering all models plus ERAI; and how well do GCMs reproduce the patterns associated with extremes such as the July 2012 Greenland ice sheet melt event. In each case, SOMs abstract complex datasets (e.g., daily near-surface air temperature) into a relatively small set of generalized patterns representing dataset variability. We study 19 models from the CMIP5 archive that provide the sub-daily data needed for regional-scale modeling plus the recent 30-member CESM1.0-CAM5-BGC Large Ensemble.

A SOM of ERAI (1981-2000) provides both a summary of variability in the reanalysis and a mapping of each GCM within the ERAI data space. How well GCMs reproduce ERAI pattern frequencies reflects GCM skill. With frequency differences indicating over/underoccurence of patterns, insights to mechanisms behind model skill are possible (albeit complex to develop).

By analyzing ERAI and CMIP5 models together, the “universal” SOM (1981-2000) summarizes the variability of multiple realizations of climate and helps place each GCM in a common context. For example, four main pattern groups can summarize Antarctic temperature: warm- and cold-everywhere and ocean/continent warm/cold dipoles. Warm GCMs are found in both the warm-everywhere group (MRI-CGCM3) and the warm continent/cold ocean group (BNU-ESM).

ERAI SOMs also provide context to test GCM skill versus extreme events. Analysis of 35 years (1979-2013) of Greenland temperature indicates that SOM patterns associated with the July 2012 ice sheet-wide melt event are present only since 2006. GCM skill in reproducing such statistics will inform our analyses of GCM-based predictions of future melting.