Are the more extreme seasonal climate conditions easier to predict?

Relatively extreme climate events could have much profound impacts on both human society and the natural environment. Although providing useful forecast skill for anomalous seasonal climate is still a challenge to many operational centers, it is of interests to learn whether the relative rarity of a seasonal event would affect forecast skill systematically. One can speculate that the out of ordinary condition would be even harder to capture than simple above climatic mean condition. But the rare events could be triggered by much stronger forcing and therefore might be easier for the climate model to simulate.

Using the data archive from EU DEMETER multi-model ensemble dynamical seasonal forecast system, we investigated this problem by changing the threshold of seasonal prediction categories for probabilistic forecast. Forecast skill verification is based on the cross-validated hindcast runs from 1980 to 2001. There are seven coupled global circulation models (developed by ECMWF, LODYC, CNRM, CERFACS, INGV, MPI, UKMO) used in the DEMETER forecast system. There are 9 ensemble members in each model run. For the probabilistic forecast, the forecast skills are evaluated using relative operating characteristics (ROC) score, reliability and sharpness diagrams, Brier skill score. Both the equitable threat score (ETS) and extreme dependence score (EDS) are used to investigate the deterministic forecast skill from normal to relatively extreme conditions. The dependence of hindcast skill scores on the range of extreme categories of seasonal precipitation will be discussed. The sensitivity of such dependence on annual cycle will also be explored.