An improvement of analog model strategy for more reliable local climate change scenarios

Downscaled results derived using a linear regression model are compared with corresponding analysis based on an analog model, and the effect of systematic biases in climate models is examined. Here, a correction of the biases in the climate model is achieved using a common principal component analysis basis and by adjusting the part of the principal components corresponding to the control period. The results suggests that the downscaled results have a distribution more similar to the observations if the systematic biases are corrected for. The analog model can utilise weighted as well as unweighted principal components as input, and the effect of this choice was ex-amined. The results suggest that the weighted principal components yield more realistic results than the unweighted ones. Analog models are by definition incapable of making extrapolations outside the range of observed values whereas a linear model is well-suited for extrapolation. A combined approach involves superimposing a linear trend from the regression-based model onto the results of the analog model. It is theoretically possible for the combined method to make projections with a realistic level of variance as well as higher values than in the calibration data sample. A comparison between the linear, analog, and the combined strategies suggest that the linear model not always give the strongest trend, but also that the combined method may shift the analog-derived distribution towards higher values.