Climate change in the Northeast United States: An analysis of the NARCCAP multimodel simulations

Abstract

Based on the outputs from multiple regional climate models (RCMs) forced by global climate models (GCMs) in phase II of the North American Regional Climate Change Assessment Program, current simulations and future projections of surface air temperature and precipitation over the Northeast U.S. region are analyzed. We address two questions in this study: How do the combined biases from the GCM-RCM pairs compare to that from their driving GCMs? Are the future responses projected by these GCM-RCM pairs robust and consistent with that predicted by their driving GCMs? The GCM-RCM pair may produce decreased combined biases compared to the inherent biases in the driving GCM, but the possibility of compensating errors suggest that this does not necessarily imply improved representations of current climate. An RCM simulation may also impart biases which exacerbate those from the driving GCM, leading to a large combined error. The substantial underestimates of winter and summer temperature produced by Geophysical Fluid Dynamics Laboratory (GFDL) Experimental Climate Prediction-2 and GFDL Regional Climate Model version 3 are partly attributed to moderate cold biases from their driving GFDL, whereas dynamically downscaling Hadley Centre Climate Model version 3 with Hadley Centre Regional Climate Model leads to mitigated combined biases. The “probability density functions” of temperature and precipitation are estimated for each individual model pair to illustrate the temporal and spatial distributions of these two climate variables. Under the Special Report on Emissions Scenarios A2 emission scenario, the ensemble averaged changes in winter and summer temperature for the mid-21st century (2041–2070) vary between 2.5 to 3.2°C across the subregions. These warming signals are consistent and statistically significant across the model pairs and far exceed the estimated natural variability. The projected future changes in precipitation indicate generally wetter winters and drier summers, but the magnitudes, directions, and spatial distributions of precipitation changes are model-dependent. Moreover, the ensemble average summer precipitation changes (0.6 to −7.9% as estimated by percentage of present-day values) fall within the estimated range of natural variability. Different changes in moisture flux convergence-divergence appear to contribute to the disagreement of the precipitation responses between some GCM-RCM pairs and their driving GCMs. The “reliability ensemble averaging” procedure is also applied and provides a complement to a simpler averaging method to estimate the average and uncertainty range of the simulated climate changes.