Abstract
For climate change impact analyses, local scenarios of surface variables at the daily scales are frequently required. Empirical transfer functions are a widely used technique to generate scenarios from GCM data at these scales. For successful downscaling, the impact analyst should take into account certain considerations. First, it must be demonstrated that the GCM simulations of the required variable are unrealistic and therefore that downscaling is required. Second, it must be shown that the GCM simulations of the selected predictor variables are realistic. Where errors occur, attempts must be made to compensate for their effect on the transfer function–generated predictions or, where this is not possible, the effect on the transfer function–generated climate series must be understood. Third, the changes in the predictors between the control and perturbed simulation must be examined in the light of the implications for the change in the predicted variable. Finally, the effect of decisions made during the development of the transfer functions on the final result should be explored. This study, presented in two parts, addresses these considerations with respect to the development of local scenarios for daily maximum (TMAX) and minimum (TMIN) temperature for two sites, one in North America (Eau Claire, Michigan) and one in Europe (Alcantarilla, Spain).
Part I confirms for a selected GCM that simulations of daily TMAX and TMIN, whether taken from the nearest land grid point, or obtained by interpolation to the site location, are inadequate. Differences between the GCM 1 × CO2 and observed temperature series arise because of a 0°C threshold in the model data. At both sites, variability is suppressed during periods affected by the threshold. The thresholds persist into the perturbed simulation, affecting not only GCM-predicted 2 × CO2 temperatures but also, because the duration and timing of the threshold effect changes in the perturbed simulation, the magnitude and seasonal distribution of the 2 × CO2 –1 × CO2 GCM differences.
Comparison of modeled and observed 500-hPa geopotential height (Z500) and sea level pressure (SLP) shows that, although systematic errors of the type associated with the 0°C threshold in the temperature data are absent, significant errors do occur in certain seasons at both sites. For example, SLP is poorly modeled at Alcantarilla, where the control and observed means differ significantly in every season. The worst results at both sites are in summer. These results will affect the performance of the transfer functions when initialized with model data. Whereas little change is found to occur in SLP at either site between the 1 × CO2 and 2 × CO2 simulation, there is a noticeable increase in Z500. Other things being equal, therefore, the temperature changes predicted by the transfer functions are likely to be greatest when Z500 contributes the most to the explained variances.
In Part II, a range of transfer functions are developed from the free atmosphere variables and validated, using observations. The performance of these transfer functions when initialized with model data is evaluated in the light of the findings in Part I. The sensitivity of the perturbed climate scenarios to a range of user decisions is explored.
Corresponding author address: Jean P. Palutikof, Climatic Research Unit, University of East Anglia, Norwich NR4 7TJ, United Kingdom.
Email: j.palutikof@uea.ac.uk
