What is Downscaling?

Because the scale of a Global Circulation Model (GCM) output is coarse -- a grid cell from a typical GCM model run being 2.5 x 2.5 degrees or roughly 250 kilometers square -- it is difficult to use GCM output directly in most regional and local-scale environmental modeling. Rather, it is necessary to downscale the GCM output: that is, to add information to the model so that it has a finer spatial resolution. There are two general strategies for downscaling. The first is termed dynamical downscaling. In this approach, physical meteorological processes are directly modeled at a finer spatial scale, incorporating the effects of regional topography and land cover, typical scales being 1/2 to 1/16th of a degree. Because dynamical downscaling is computationally very intensive, it is not often used for climate projections, especially for those that are multi-decadal in time scale or incorporate multiple models. Rather, most downscaled climate projections take statistical approaches.

Several different types of statistical approaches are commonly used. The first method, called a delta or change factor approach, simply adds the predicted change for an entire GCM cell to a fine-scale grid of a baseline current climate variable. The second method, termed weather typing or constructed analogues, takes a GCM prediction, consults a databank of historical weather patterns, and statistically combines a small set of patterns similar to the GCM prediction to construct a fine-scale map of the climate variable. The third method, bias-corrected spatial disaggregation (BCSD), relates quantiles of the GCM predictions to historical patterns to produce daily time series to construct the downscaled grid.

The California Basin Characterization Model uses several stages of downscaling to produce climate projection maps at a 270-meter resolution. First, the 2 degree GCM output is reduced to a 12 kilometer grid using a constructed analogues method. In the second step, the 12-kilometer output is interpolated down to a 4-kilometer scale using inverse distance squared weighting. Then a 4 km resolution historical climate map (the PRISM climate model) is used to develop bias correction statistics which are used to refine the 4 km scale climate projections. Finally, gradient inverse squared distance interpolation is used to generate a climate projection surface at a 270 meter scale.

Some resources pertinent to downscaling include:

• Micheli et al. 2011, Downscaling future climate projections to the watershed scale: a North San Francisco Bay Estuary case study
• Mauer and Hidalgo 2008, Utility of daily vs. monthly large-scale climate data: an intercomparison of two statistical downscaling methods
• Hidalgo et al. 2008, Downscaling with Constructed Analogues: Daily Precipitation and Temperature Fields Over the United States
• Flint and Flint 2012, Downscaling future climate scenarios to fine scales for hydrologic and ecological modeling and analysis
• Daniels et al. 2012, Climate Projections FAQ