tions of processes, and multiple model experiments to explore different assumptions and model specifications; and (2) improvements in theoretical and mechanistic understanding of the climate system and the processes being modeled, which in turn are tied to basic research and improvements in observational capabilities. Today, continued improvements in computational power, scientific understanding, and supporting observations are still the primary factors driving improvements in climate models—or stated conversely, even if the evolution of future climate forcing were known exactly, limits in computer power, observational data, and scientific understanding of the climate system would still constrain the ability of models to produce perfect predictions of future climate (Shapiro et al., in press).
For example, the typical horizontal grid spacing of a state-of-the-art global climate model is on the order of 60 miles (100 km), but climatically relevant features such as clouds, topography, and land cover often vary at a scales of a half-mile or less. These subgridscale features and processes must be parameterized—approximated using numerical techniques that specify the large-scale influence of small-scale processes—or upscaled through statistical or “nested model” approaches that extend representative small-scale simulations to larger spatial scales. As a result of these approximations and other factors (described below), global climate models generally only provide consistent and reliable simulations of temperature, precipitation, and other relevant climate variables at continental to global scales.
The lack of regionally specific climate information from global climate models poses a major challenge, because many climate-related decisions, especially those related to adaptation, demand information on regional to local scales. A variety of downscaling approaches have been developed to obtain this regional information. One widely used approach is statistical downscaling, wherein empirical relationships between past observations of local- and regional-scale climate variations are used to translate large-scale projections from global climate models to smaller space scales and shorter time scales. Alternatively, finer-scale regional models can be “nested” within coarser-resolution global models to simulate regional climate changes (e.g., Hay et al., 2002; Leung et al., 2003; UCAR, 2007). A related approach is linking models currently used to predict weather and seasonal to interannual climate variations with those that predict climate change on decadal to centennial time scales (this is sometimes called “seamless prediction”).
In general, downscaling techniques are not as well developed or understood as global