An intercomparison of climate model bias correction methods across Australia

The National Partnership for Climate Projections (NPCP) was established as a collaborative effort of the Australian climate projections community to develop a consistent approach to deliver future climate information. As bias correction of climate model outputs is important for many applications, a NPCP bias correction intercomparison project was initiated. The first phase of the intercomparison aimed to support the production of national-scale climate projections by the Australian Climate Service. It focused on five methods – Equidistant Cumulative Density Function matching (ECDFm), Quantile Matching for Extremes (QME), Quantile Delta Change (QDC), N-Dimensional Multi-Variate Bias Correction (MBCn) and Multivariate Recursive Nesting Bias Correction (MRNBC) – and applied them to daily timescale Coordinated Regional Climate Downscaling Experiment (CORDEX) data produced by NPCP partner organisations. Each method was assessed over a calibration period and also via cross-validation on several metrics relating to the temperature and precipitation climatology, variability, distribution, extremes and trends. The best-performing bias correction methods were QME and MRNBC. The ECDFm method also performed well on most metrics, but under certain circumstances it could dramatically increase the model bias. The QDC method is a delta change method (i.e. it perturbs the observations rather than correcting model data) and compared very favourably to the four bias correction methods. The QME, MRNBC and QDC methods were subsequently used by the Australian Climate Service to produce climate projections datasets for Australia.