Precipitation predictors for downscaling: observed and general circulation model relationships

Abstract

Because of the coarse resolution of general circulation models (GCM), ‘downscaling’ techniques have emerged as a means of relating meso-scale atmospheric variables to grid- and sub-grid-scale surface variables. This study investigates these relationships. As a precursor, inter-variable correlations were investigated within a suite of 15 potential downscaling predictor variables on a daily time-scale for six regions in the conterminous USA, and observed correlations were compared with those based on the HadCM2 coupled ocean/atmosphere GCM. A comparison was then made of observed and model correlations between daily precipitation occurrence (a time series of zeroes and ones) and wet-day amounts and the 15 predictors. These two analyses provided new insights into model performance and provide results that are central to the choice of predictor variables in downscaling of daily precipitation. Also determined were the spatial character of relationships between observed daily precipitation and both mean sea-level pressure (mslp) and atmospheric moisture and daily precipitation for selected regions. The question of whether the same relationships are replicated by HadCM2 was also examined. This allowed the assessment of the spatial consistency of key predictor–predictand relationships in observed and HadCM2 data. Finally, the temporal stability of these relationships in the GCM was examined. Little difference between results for 1980–1999 and 2080–2099 was observed. For correlations between predictor variables, observed and model results were generally similar, providing strong evidence of the overall physical realism of the model. For correlations with precipitation, the results are less satisfactory. For example, model precipitation is more strongly dependent on surface divergence and specific humidity than observed precipitation, while the latter has a stronger link to 500 hPa divergence than is evident in the model. These results suggest possible deficiencies in the model precipitation process, and may indicate that the model overestimates future changes in precipitation. Correlation field patterns for mslp versus precipitation are remarkably similar for observed data and HadCM2 output. Differences in the correlation fields for specific humidity are more noticeable, especially in summer. In many cases, maximum correlations between precipitation and mslp occurred away from the grid box; whereas correlations with specific humidity were largest when the data were propinquitous. This suggests that the choice of predictor variable and the corresponding predictor domain, in terms of location and spatial extent, are critical factors affecting the realism and stability of downscaled precipitation scenarios. Copyright © 2000 Royal Meteorological Society