A Combined Dynamical/Statistical Technique for Regionalizing an Ensemble of Climate Change Signals

In this study we demonstrate a novel technique for creating high-resolution regional climate projections from an ensemble of global climate models. This technique is applied to 18 global models from the CMIP5 archive to calculate an ensemble of projections for surface air temperature change between 1981-2001 and 2041-2061 over Southern California. Creating an ensemble of projections allows for an estimation of uncertainty due to differences in global models. Ideally, each global model would be dynamically downscaled but this is infeasible due to the high computational cost. Instead, dynamical downscaling is performed on a single global model (CCSM4) and a statistical model is then used to approximate how the dynamically derived warming pattern would change if a different global model were used. The statistical model scales the high-resolution spatial patterns using large-scale parameters sampled from the global model. By applying the statistical model to each global model we obtain an ensemble of high-resolution climate projections. Our combined dynamical/statistical downscaling technique provides an advantage over other methods by leveraging the physically consistent, high-resolution output of dynamical downscaling with the computational economy of statistical downscaling.