Evaluation of NARCCAP Simulations against Site Observations within a Lake Modified Zone

The recently released regional climate model (RCM) simulations from the North American Regional Climate Change Adaptation Project (NARCCAP) have garnered considerable interest from the climate impacts community, including as input for climate change impact, adaptation and vulnerability assessments. This paper evaluates the performance of the 50-km resolution NARCCAP multi-model simulations for the NCEP reanalysis-driven hindcast period (1979-2004) and the Global Climate Model (GCM)-driven control period (1968-2000) for three locations within the lake modified zones of the Great lakes region of the United States. Our analysis focused on model skill in simulating daily surface climate variables (i.e., daily precipitation and mean, maximum and minimum temperature). When viewed across the six RCMs and the three locations, between-model biases were larger although differences in model agreement were observed between locations, particularly for daily precipitation. Cumulative frequency distributions indicated that most of the RCMs underestimated observed precipitation at the study locations, although overestimation at high percentiles was seen for some models and locations. Even after adjusting for light precipitation events, all the RCMs produced too many wet days. Although all RCMs reproduced well the annual cycle and interannual variability of the temperature variables, they poorly simulated temperature extremes at both the low and high percentiles. For the NCEP-driven hindcasts, the temporal accuracy (i.e., co-occurrence between model and observations) of temperature extremes (maximum temperatures >32°C), freeze events (minimum temperatures <0°C) and hard freeze events (minimum temperatures <-2°C) were also relatively low (less than 20 percent) for all models. As expected, deviations compared to observations in temperature and precipitation mean values, frequency distributions, and extremes were considerably greater for the GCM-driven control period simulations. Post-processing methods to adjust for model biases should be considered prior to the use of the RCM simulations for impact assessments, as model inaccuracies can substantially affect the outputs of any “downstream” impact models employed in climate assessments.