The Australian Bureau of Meteorology has commissioned a project to provide high-resolution projections of water availability to the Australian Community. The project will require providing projections of temperature, rainfall, surface winds and solar radiation at approximately 5 km resolution to drive the Australian Water Resources Assessment Landscape (AWRA-L) hydrological model. The output of Global Climate Models (GCMs) provide projections at roughly 150 km resolution. To bridge the scale gap between the GCM output and that required to drive AWRA-L the Bureau is implementing three statistical downscaling techniques. The methods comprise a trend preserving univariate approach (ISIMIP2b; Hempel et al., 2013); a multivariate nested bias correction method (MRNBC; Johnson and Sharma, 2012; Mehrotra and Sharma, 2016); and a quantile matching for extremes (QME) method (Dowdy, 2019).

The techniques differ in their philosophy and methodology based on their intended application. The ISIMIP2b method preserves the climate change signal and corrects for the over-representation of drizzle days inherent in GCMs (e.g. Stephens et al., 2010); the MRNBC corrects the quantiles across all intervariable correlations on daily, monthly and yearly timescales; the QME method has a novel method of accounting for distribution extremes. The ISIMIP2b and QME methods apply the bias correction at the scale of the GCM whereas the MRNBC applies it at the GCM scale and uses spatial disaggregation to resolve to the observation scale; bias correction methods at observational scales have been criticised for imposing unrealistic physics on GCM output.

Together this provides an opportunity to examine the contribution to projection uncertainty from a suite of commonly applied bias correction techniques. In this presentation we will give an overview of these methods, evaluate their output, present the differences in the climate change signal among them and relate the impact of the methods to application of the outputs to climate risk modelling.