Drift correction of decadal experiments
Phase Five of the Coupled Model Intercomparison Project (CMIP5) includes a new set of experiments, called the ‘decadal’ or ‘near-term’ predictions, to bridge the gap between weather forecasting systems and climate projections. These experiments are initialized with the observations and provide predictions over 10 or 30-years periods, while also accounting for external forcings. Climate models, being imperfect replicas of the real world, have equilibrium states that differ from that of the observed climatology. Thus, when a model is pulled away from its mean state on initializing with the observations (as is the case for decadal predictions), it tends to revert back to its preferred biased state over a period of time. The resulting biases are referred to as ‘drift’, and some sort of drift-correction must be applied prior to using the decadal forecasts. This talk would focus on a brief overview of the issue of drift and the different drift correction methods in place for decadal predictions. Alongside, an instance of sampling biases on the estimate of drift and drift correction would be presented followed by a systematic intercomparison of the different drift correction methods and their implications on multi-model averaging.
Figure: Choudhury D., Gupta A. S., Sharma A., Mehrotra R. & Sivakumar B. (2017). An assessment of drift correction alternatives for CMIP5 decadal predictions. Journal of Geophysical Research: Atmospheres, 122. https://doi.org/10.1002/2017JD026900