Regional Climate Models

Global models, typically having horizontal resolution of about 250 km, do not allow sufficient detail to determine climatic information for decision making in regions of complex orography such as California (Figure 15.1). Increasing resolution by a factor of 5 (Figure 15.1) produces a substantial improvement in representing orographic features of the California landscape, but even this resolution is insufficient for some purposes. Determining the appropriate resolution for this application will require careful evaluation of the impact of resolution changes on representing specific climate elements in specific circumstances.

RCMs represent the prevalent current approach to dynamical downscaling (U.S. CCSP, 2003). Following the pioneering work of F. Giorgi and colleagues (see Giorgi and Mearns, 1991, 1999, for reviews), regional climate modeling

ax=250km ax=50 km ax=10 km

Figure 15.1 Illustration of terrain heights (top) and land use (bottom) at three different horizontal grid resolutions (Ax).

ax=250km ax=50 km ax=10 km

Figure 15.1 Illustration of terrain heights (top) and land use (bottom) at three different horizontal grid resolutions (Ax).

has emerged as a valuable approach in creating climate information at resolutions higher than are available from global models (Intergovernmental Panel on Climate Change [IPCC], 2001, chapter 10, "Regional Climate Information — Evaluation and Projections"). RCMs driven by lateral boundary conditions supplied by reanalysis data show considerable skill compared to global models in representing the spatial structure of temperature and precipitation in areas of complex orography (Leung et al., 2003). The higher resolution of RCMs also allows simulation of fine-scale dynamics (atmospheric jets, drainage flows, frontal structure, and regional convergence and divergence patterns) and temporal changes associated with smaller horizontal scales (temporal variability in wind speed, precipitation frequency, precipitation intensity, and cloudiness).

Regional models use results of global models as lateral boundary conditions and simulate (by use of essentially the same procedures as global models) scenario climates that are dynamically consistent with surface and external radiative forcing and lateral boundary conditions provided by the global model. Geographic features are represented with more detail in regional models as can be seen in Figure 15.1. Intercom-parison studies of results of several regional models driven at lateral boundaries by the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis (NNR) (Kalnay et al., 1996) have helped reveal the strengths and weaknesses of this approach to downscaling global climate information (Takle et al., 1999; Anderson et al., 2003).

The agriculturally intensive central United States is unique in the country in that summertime mesoscale convec-tive precipitation (Wallace and Hobbs, 1977) is dependent on nocturnal water vapor flux convergence (Anderson et al., 2003). Neither the NNR (Higgins et al., 1997) nor global climate models (Ghan et al., 1995) capture this essential mechanism. Finer grid spacing is needed to resolve the fine-scale dynamical processes that lead to timing, location, intensity, and amounts of precipitation (Anderson et al., 2003). Most, but not all, regional models are able to capture the nocturnal maximum in hourly precipitation in this region (Anderson et al., 2003), which is an indication that nocturnal moisture convergence at the outflow of the low-level jet is being simulated. For this reason, we expect that regional climate models can offer better representation of climate factors of critical importance to agriculture compared to low-resolution climate models.

The recently issued (24 July 2003) Strategic Plan for the CCSP (2003) calls for more regional climate modeling through Objective 16: "Accelerate the development of scientifically based predictive models to provide regional and fine-scale climate and climate impact information relevant for scientific research and decision support applications."

15.2.3 Statistical Methods

The computational demands of regional climate modeling have stimulated the search for more efficient methods of obtaining site-specific scenarios of future climates. Statistical downscaling (IPCC, 2001, chapter 10) assumes that regional climate is determined by two factors: the large-scale climatic state and local factors. Features of the large-scale climate and known local factors such as elevation are input to a statistical procedure, usually linear regression, that has been adjusted to yield the local climate. Many forms of statistical downscal-ing have been applied, such as simple regression and sophisticated artificial neural networks. The procedure requires good input data, both for predictors and the predictands, in order to arrive at an acceptably accurate statistical model.

Wilby and Wigley (1997) provide a review of statistical downscaling, and give a variety of predictors and predictands for applications to climate. The success of statistical down-scaling is highly location dependent, but Wilby and Wigley (2000) confirmed results of previous research that showed strong correlation for winter precipitation and circulation predictors and precipitation for regions near oceanic sources of moisture. Kidson and Thompson (1998) assert that empirical downscaling methods provide skill comparable to that of dynamical modeling for application to the current climate. In a direct comparison of statistical downscaling to regional climate modeling (Wilby et al., 1999), we found advantages for both methods with neither method clearly superior in our hydrological application.

Statistical procedures tend to have low computational demand, which makes statistical downscaling attractive. However, some cautions should be recognized. When down-scaling future climate, it must be assumed that statistical relations based on observed climate of the past will hold in the future. The statistical relations likely are strongly limited to those regions with sufficient data for calibration and validation. The adherence of statistical relations to physical laws may be obscure, limiting insight into climatic processes. These features complicate attempts to conduct systematic appraisals of statistical downscaling.

Was this article helpful?

0 0

Post a comment