In this study we develop methods for dynamically downscaling output from six general circulation models (GCMs) for two emissions scenarios using a variable-resolution atmospheric climate model. The use of multiple GCMs and emissions scenarios gives an estimate of model range in projected changes to the mean climate across the region. By modeling the atmosphere at a very fine scale, the simulations capture processes that are important to regional weather and climate at length scales that are subgrid scale for the host GCM. We find that with a multistaged process of increased resolution and the application of bias adjustment methods, the ability of the simulation to reproduce observed conditions improves, with greater than 95% of the spatial variance explained for temperature and about 90% for rainfall. Furthermore, downscaling leads to a significant improvement for the temporal distribution of variables commonly used in applied analyses, reproducing seasonal variability in line with observations. This seasonal signal is not evident in the GCMs. This multistaged approach allows progressive improvement in the skill of the simulations in order to resolve key processes over the region with quantifiable improvements in the correlations with observations.