Improving the reliability of seasonal climate forecasts through empirical downscaling and multi-model considerations; presentation

Abstract

Investigation into the predictability of seasonal climate extremes such as droughts and flood seasons provide insight into the limits of predictability of the ocean-land-atmosphere system. However, expressions on what the future may hold always embody degrees of uncertainty, often expressed as a probabilistic outcome. Since seasonal prediction is inherently probabilistic in nature they are judged (i.e. verified) probabilistically through attributes including reliability, resolution, discrimination and sharpness. We present seasonal prediction verification for the equatorial Pacific Ocean (where El Niño and La Niña events occur) sea-surface temperatures. The verification is done over a recent multi-decadal period for which hindcasts (re-forecasts) have been generated by a statistical model and by state-of-the-art fully coupled ocean-atmosphere general circulation models. Since forecast users generally require well-calibrated probability forecasts we employ a model output statistics approach to improve on raw coupled model forecasts, and further enhance the forecasts by considering a range of possible methods for combining the coupled models' output in order to provide the most informative forecasts of future observables.