Environmental Engineering Reference
In-Depth Information
is statistical downscaling, which “trains” a relationship between current cli-
mate (precipitation, surface air temperature, and other variables) produced
by a climate model and observations, in such a way as to correct for climate
model biases and spatial and temporal scale mismatches. The same rela-
tionship is then applied to future climate simulations. For both current and
future climate, the downscaled climate model output is used to force a river
basin hydrology model. Examples of this approach are the study by Hayhoe
et al. (2007) of climate change impacts on the northeastern United States,
the Maurer et al. (2007) study of California water resources, and Christensen
and Lettenmaier's (2007) study of Colorado River water resources. The third
approach is dynamical downscaling, in which a regional climate model
(RCM) is nested within a global model to produce more spatially resolved
climate model output. Although dynamical downscaling may be preferred
on theoretical grounds (because it is based on physical rather than statistical
relationships; its application in practice has been problematic because the
computational burden is high, hence the number of RCM runs that can be
performed for different global models is small. Furthermore, practical con-
straints, like the fact that global model output sufficient to provide requisite
variables at the RCM domain boundary often are not archived, or are not
archived at sufficient vertical levels in the atmosphere and time resolution
to meet RCM needs. Finally, Wood et al. (2004) showed that even after
dynamical downscaling, application of statistical post-processing, similar to
that used in statistical downscaling methods, was necessary to remove RCM
bias prior to using RCM output to force a hydrological model.
The second pathway is direct analysis of global climate model runoff
predictions. This approach usually has not been favored for river basin
studies, because of the scale mismatch between the spatial resolution of the
global models (typically several degrees latitude by longitude) and the river
basin scale. However, for analysis of large river basins, Milly et al. (2005)
(and replots of the Milly et al. [2005] results at the scale of the U.S. hydro-
logic regions in the USCCSP Synthesis and Assessment Product 4.3 [USCCSP,
2008b]) have found this approach to be useful, especially because it avoids
the issues associated with downscaling and inconsistencies in fluxes at the
river basin scale. Seager et al. (2007), in an analysis of projected changes in
runoff over the U.S. Southwest, also analyzed GCM runoff directly.
Despite the fact that direct analysis of GCM runoff has not been widely
performed, we use this approach here for two reasons. First, post processing
of global model output by statistical downscaling and hydrological modeling
results in a mismatch in the surface water balance between the hydrologic
model that produces streamflow simulations and the global model. Second,
