The Simulation of Daily Temperature Time Series from GCM Output. Part II: Sensitivity Analysis of an Empirical Transfer Function Methodology

Abstract

Empirical transfer functions have been proposed as a means for ?downscaling? simulations from general circulation models (GCMs) to the local scale. However, subjective decisions made during the development of these functions may influence the ensuing climate scenarios. This research evaluated the sensitivity of a selected empirical transfer function methodology to 1) the definition of the seasons for which separate specification equations are derived, 2) adjustments for known departures of the GCM simulations of the predictor variables from observations, 3) the length of the calibration period, 4) the choice of function form, and 5) the choice of predictor variables. A modified version of the Climatological Projection by Model Statistics method was employed to generate control (1 ? CO2) and perturbed (2 ? CO2) scenarios of daily maximum and minimum temperature for two locations with diverse climates (Alcantarilla, Spain, and Eau Claire, Michigan). The GCM simulations used in the scenario development were from the Canadian Climate Centre second-generation model (CCC GCMII). Variations in the downscaling methodology were found to have a statistically significant impact on the 2 ? CO2 climate scenarios, even though the 1 ? CO2 scenarios for the different transfer function approaches were often similar. The daily temperature scenarios for Alcantarilla and Eau Claire were most sensitive to the decision to adjust for deficiencies in the GCM simulations, the choice of predictor variables, and the seasonal definitions used to derive the functions (i.e., fixed seasons, floating seasons, or no seasons). The scenarios were less sensitive to the choice of function form (i.e., linear versus nonlinear) and to an increase in the length of the calibration period. The results of Part I, which identified significant departures of the CCC GCMII simulations of two candidate predictor variables from observations, together with those presented here in Part II, 1) illustrate the importance of detailed comparisons of observed and GCM 1 ? CO2 series of candidate predictor variables as an initial step in impact analysis, 2) demonstrate that decisions made when developing the transfer functions can have a substantial influence on the 2 ? CO2 scenarios and their interpretation, 3) highlight the uncertainty in the appropriate criteria for evaluating transfer function approaches, and 4) suggest that automation of empirical transfer function methodologies is inappropriate because of differences in the performance of transfer functions between sites and because of spatial differences in the GCM?s ability to adequately simulate the predictor variables used in the functions.