Because climate model results generally are not sufficiently accurate (in terms of absolute values) at regional scales to be used directly (Mearns et al., 1997), mean differences between the control (or current climate) run and the future climate run usually are calculated and then combined with some baseline observed climate data set (IPCC, 1994). Conventionally, differences (future climate minus control) are used for temperature variables, and ratios (future climate/control) are used for other variables such as precipitation, solar radiation, relative humidity, and windspeed. Most impact applications consider one or more fixed time horizon(s) in the future (e.g., the 2020s, the 2050s, and the 2080s have been chosen as 30-year time windows for storing change fields in the IPCC-DDC). Some other applications may require time-dependent information on changes, such as vegetation succession models that simulate transient changes in plant composition (e.g., VEMAP members, 1995).
One of the major problems in applying GCM projections to regional impact assessments is the coarse spatial scale of the gridded estimateson the order of hundreds of kilometersin relation to many of the exposure units being studied (often at one or two orders of magnitude finer resolution). Concern about this issue is raised in Chapters 4 and 5. Several solutions have been adopted to obtain finer resolution information.
Conventionally, regional "detail" in climate scenarios has been incorporated by appending changes in climate from the nearest coarse-scale GCM grid box to the study area (observation point or region) (e.g., Rosenzweig and Parry, 1994) or by interpolating from GCM grid box resolution to a higher resolution grid or point location (Leemans and van den Born, 1994; Harrison and Butterfield, 1996).
Three major methods have been developed to produce higher resolution climate scenarios at the sub-GCM grid scale: regional climate modeling (Giorgi and Mearns, 1991, 1999; McGregor, 1997), statistical downscaling (von Storch et al., 1993; Rummukainen, 1997; Wilby and Wigley, 1997), and variable- and high-resolution GCM experiments (Fox-Rabinovitz et al., 1997). All three methods are presented in Table 3-4 and discussed in detail in TAR WGI Chapter 10, but we briefly review here the first two, since they have been most commonly applied to impact assessments. Both methods are dependent on large-scale circulation variables from GCMs. Large-scale circulation refers to the general behavior of the atmosphere at large (i.e., continental) scales.
The basic strategy with regional models is to rely on the GCM to reproduce the large-scale circulation of the atmosphere and to use the regional model, run at a higher resolution, to simulate sub-GCM scale regional distributions of climate. In numerous experiments with regional models driven by control and doubled CO2 output from GCMs for regions throughout the world, the spatial pattern of changed climateparticularly changes in precipitationsimulated by the regional model departs from the more general pattern over the same region simulated by the GCM (TAR WGI Chapter 10).
Statistical methods are much less computationally demanding than dynamic methods; they offer an opportunity to produce ensembles of high-resolution climate scenarios (for reviews, see von Storch, 1995; Wilby and Wigley, 1997). However, these techniques rely on the (questionable) assumption that observed statistical relationships will continue to be valid under future radiative forcingthat is, they are time-invariant (Wilby, 1997).
Although regional modeling and statistical techniques have been available for at least a decadetheir developers claiming use in impact assessments as one of their important applicationsit is only recently that they have actually provided scenarios for impact assessments (Mearns et al., 1998, 1999, 2001; Sælthun et al., 1998; Hay et al., 1999; Brown et al., 2000; Whetton et al., 2001). Mearns et al. (1999, 2001) demonstrate that a high-resolution scenario results in agricultural impacts that differ from those produced with a coarser resolution GCM scenario (discussed in Chapter 5). Hay et al. (1999) found differences in runoff calculations, based on a GCM-scenario and a statistically downscaled scenario.
Other reports in this collection