The provision of seasonal outlooks with and without skill masks

Abstract:

Many meteorological agencies around the world provide seasonal climate outlooks for their region or the globe as a whole. The majority of these outlooks are provided for the entire domain of interest, but the skill of these outlooks is variable in space and time, lead-time dependent, variable dependent, and modulated by climate modes of variability. Skill maps are provided in addition to outlooks by many agencies to inform stakeholders of model skill and allow for informed decision-making based on past model performance. Often the skill of seasonal outlooks is marginal, so this begs the question of whether these outlooks should only be provided where and when there is demonstrable skill, or if lack of skill should be communicated in other ways. Here, we use hindcast simulations from the Australian model, ACCESS-S1, to explore the effect of skill masks on seasonal outlooks of precipitation in Australia. We use different skill measures and thresholds to illustrate the effects of applying skill masks to S2S outlooks and, by employing a cross-validation method, show that there can be benefits to employing a skill masking approach. We also explore the effects of skill masking outlooks during El Niño and La Niña conditions and at times of extreme rainfall. There are advantages and disadvantages to the use of skill masking that we discuss as well as some important considerations when applying skill masks to operational outlooks. We hope that our study will spur further consideration and analysis of how seasonal outlooks are provided and communicated to the public and specific stakeholders.