Improving Projections of Regional Climate Change with Hybrid Downscaling

 

Projections of climate change made with global climate models (GCMs) are often too coarse to be used directly in impacts models or by stakeholders.  Thus, GCM output is often downscaled to higher resolution with either statistical or dynamical techniques.  Statistical techniques are frequently used because their low computational cost makes them easy to apply to a large number of GCMs, scenarios, and time slices.  However, they may miss important local variations in the climate change signal, such as the narrow bands of enhanced warming (+3 °C) due to snow-albedo feedback in the mountainous western U.S.  Dynamical downscaling with a regional climate model can capture these variations, as underlying physical processes are explicitly simulated.  However, the hefty computational cost of dynamical downscaling means that it is typically applied only to 1-2 GCMs, for a single future time slice, and a single emissions scenario.  A new approach, hybrid dynamical-statistical downscaling, uses statistical methods to extend dynamically downscaled results to multiple GCMs, time slices, and scenarios.  Here we will explore how the method works, and see it used to generate projections of temperature and snow cover for California’s Sierra Nevada mountain range.