The value of interactive ozone in predicting extreme stratospheric ozone-loss events in the Arctic: insights from targeted WACCM hindcasts

Extreme stratospheric ozone-loss events, such as the Arctic spring of 2020, can emerge spontaneously in free-running chemistry–climate models during winters and springs with a strong and persistent polar vortex. While ozone is known to potentially affect stratospheric variability, its specific role in the subseasonal-to-seasonal (S2S) predictability of the stratosphere remains unclear. The first step consists of assessing its role in the prediction of internally generated extreme depletion events.

Here we analyse a suite of targeted hindcast experiments for several extreme ozone-loss winters identified in a 200-year free-running WACCM integration. Hindcasts are initialized from January to April to examine how predictability evolves through the winter season, and paired experiments compare configurations with fully interactive ozone to those in which the radiative transfer scheme uses a prescribed climatological ozone distribution.

Preliminary results show that early-winter vortex conditions are not a good predictor of the occurrence of extreme depletion events in late winter-spring, highlighting the limited predictability of the polar vortex, even under the strong vortex conditions that are conducive to ozone depletion in spring. For the most pronounced ozone-loss case, differences between interactive and prescribed ozone ensembles indicate that ozone–radiative feedbacks can, under certain conditions, support the persistence of a strong, cold vortex into late winter and spring, thereby maintaining the dynamical environment in which severe ozone depletion can occur. At the same time, the impact of interactive ozone on S2S skill varies with initialization date and event characteristics.

These findings provide first insights into how chemistry–dynamics coupling affects the predictability of extreme stratospheric states and point to the value of interactive ozone schemes in S2S prediction systems.