IPCC Fourth Assessment Report: Climate Change 2007
Climate Change 2007: Working Group I: The Physical Science Basis

11.5.2 Skill of Models in Simulating Present Climate

Individual AOGCMs in the MMD vary in their ability to reproduce the observed patterns of pressure, surface air temperature and precipitation over North America (Chapter 8). The ensemble mean of MMD models reproduces very well the annual-mean mean sea level pressure distribution (Section 8.4). The maximum error is of the order of ±2 hPa, with the simulated Aleutian Low pressure extending somewhat too far north, probably due to the inability of coarse-resolution models to adequately resolve the high topography of the Rocky Mountains that blocks incoming cyclones in the Gulf of Alaska. Conversely, the pressure trough over the Labrador Sea is not deep enough. The depth of the thermal low pressure over the southwest region in summer is somewhat excessive.

The MMD models simulate successfully the overall pattern of surface air temperature over North America, with reduced biases compared to those reported in the TAR. Ensemble-mean regional mean bias ranges from –4.5°C to 1.9°C for the 25th to 75th percentile range, and medians vary from –2.4°C to +0.4°C depending on region and season (Supplementary Material Table S11.1). The ensemble mean of MMD models reproduces the overall distribution of annual mean precipitation (Supplementary Material Table S11.1), but almost all models overestimate precipitation for western and northern regions. The ensemble-mean regional mean precipitation bias medians vary from –16% to +93% depending on region and season. The ensemble-mean precipitation is excessive on the windward side of major mountain ranges, with the excess reaching 1 to 2 mm day–1 over high terrain in the west of the continent.

Regional Climate Models are quite successful in reproducing the overall climate of North America when driven by reanalyses. Over a 10° × 10° Southern Plains region, an ensemble of six RCMs in the North American Regional Climate Change Assessment Program (NARCCAP; Mearns et al., 2005) had 76% of all monthly temperature biases within ±2°C and 82% of all monthly precipitation biases within ±50%, based on preliminary results for a single year. RCM simulations over North America exhibit rather high sensitivity to parameters such as domain size (e.g., Juang and Hong, 2001; Pan et al., 2001; Vannitsem and Chomé, 2005) and the intensity of large-scale nudging (providing large-scale information to the interior of the model domain, see e.g., von Storch et al., 2000; Miguez-Macho et al., 2004) if used. In general, RCMs are more skilful at reproducing cold-season temperature and precipitation (e.g., Pan et al., 2001; Han and Roads, 2004; Plummer et al., 2006) because the warm-season climate is more controlled by mesoscale and convective-scale precipitation events, which are harder to simulate (Giorgi et al., 2001a; Leung et al., 2003; Liang et al., 2004; Jiao and Caya, 2006). On the other hand, Gutowski et al. (2004) find that spatial patterns of monthly precipitation for the USA, when viewed as a whole rather than broken into individual regions, are better simulated in summer than winter. Several studies point to the large sensitivity of RCMs to parametrization of moist convection, including the vertical transport of moisture from the boundary layer (Chaboureau et al., 2004; Jiao and Caya, 2006) and entrainment mixing between convective plumes and the local environment (Derbyshire et al., 2004). In a study of the simulation of the 1993 summer flood in the central USA by 13 RCMs, Anderson et al. (2003) find that all models produced a precipitation maximum that represented the flood, but most underestimated it to some degree, and 10 out of 13 of the models succeeded in reproducing the observed nocturnal maxima of precipitation. Leung et al. (2003) examined the 95th percentile of daily precipitation and find generally good agreement across many areas of the western USA.

A survey of recently published RCM current-climate simulations driven with AOGCMs data reveals that biases in surface air temperature and precipitation are two to three times larger than the simulations driven with reanalyses. The sensitivity of simulated surface air temperature to changing lateral boundary conditions from reanalyses to AOGCMs appears to be high in winter and low in summer (Han and Roads, 2004; Plummer et al., 2006). Most RCM simulations to date for North America have been made for time slices that are too short to properly sample natural variability. Some RCMs have employed less than optimal formulations, such as outdated parametrizations (e.g., bucket land surface scheme), too few levels in the vertical (e.g., 14) or a too-low uppermost computational level (e.g., 100 hPa).