Performance of an empirical bias-correction of a high-resolution climate dataset

We describe the method and performance of a bias-correction applied to high-resolution (10km) simulations from a stretched-grid Regional Climate Model (RCM) over Tasmania, Australia. The bias-correction is a quantile mapping of empirical cumulative frequency distributions. Corrections are applied at a daily time step to five variables: rainfall, potential evaporation (PE), solar radiation, maximum temperature and minimum temperature. Corrections are calculated independently for each season. We show that quantile mapping of empirical distributions can be highly effective in correcting biases in RCM outputs. Cross-validation shows biases are effectively reduced across the range of cumulative frequency distributions, with few exceptions. The bias-correction is not as effective at correcting biases for values at or near zero (e.g. in rainfall simulations), although even here the bias-correction improves biases evident in the uncorrected simulations. In addition, the bias-correction improves frequency characteristics of variables such as the number of rain days. We use a detrending technique to apply the bias-correction to 140-year time series of RCM variables. We show that the bias-correction effectively preserves long-term changes (e.g. to the mean and variance) to variables projected by the uncorrected RCM simulations. Correlations between key variables are also largely preserved, thus the bias-corrected outputs reflect the dynamics of the underlying RCM. However, the bias-corrected simulations still exhibit some of the deficiencies of the RCM simulations, e.g. the tendency to underestimate the magnitude and duration of large, multi-day rain events, and the tendency to underestimate the duration of dry spells. The bias-corrected simulations for six downscaled GCMs for the A2 and B1 emissions scenarios are available to researchers from http://www.tpac.org.au.