Characterization of climate model errors over North America at climate and NWP timescales using the NARCCAP RCM and Transpose-AMIP experiments

Abstract:

The NARCCAP ensemble of regional climate model (RCM) simulations over North America is aimed at providing a range of regional-scale climate change projections for use in impacts analyses. Previous studies have shown that the performance of the NARCCAP RCMs and their projections – of seasonal precipitation for instance – could be very different and at times much larger than those obtained from their driving GCMs. We explore the idea of using seamless assessment approach to diagnose model errors at various spatial and temporal scales and identify models that provide plausible realizations of historical climates. This is achieved by carrying out regional assessment of model performance over North America by combining information from the NARCCAP RCMs, their drivers (NCEP2, CMIP3 GCMs), and the Transpose-AMIP (TAMIP, NWP-style) experiments. An assessment of the performance of the CMIP3-GCMs used in dynamical downscaling indicates that although the RCMs add value to the GCM results, a large fraction of the RCM errors are inherited from the driving GCM. A seasonal dry bias over central and southern US and a wet bias over the Rockies are some of the errors that are not only seen in the NARCCAP RCMs and the driving GCMs, but are also common to many current generation of climate models (CMIP5). We demonstrate that some of these errors in climate simulations such as an underestimation of precipitation in the southern US develop within the first five days of the integration when the climate model is run in weather forecast mode, i.e., the TAMIP experiments. The design of the TAMIP experiments facilitates linking these systematic errors to model physics. We explore if such an assessment across spatial and temporal scales can be used to gain process-level understanding of model errors and if it has the potential to help constrain future projections by identifying models with credible simulations of historical climates.