Using the Ocean to Identify Forecasts of Opportunity for Decadal Prediction

Recent efforts in decadal predictability have highlighted the capability of modern numerical modeling systems to accurately predict surface temperature and precipitation on decadal timescales. A majority of the memory within the climate system on decadal timescales is contained within the oceans and land surface, and so these predictability advancements have been hypothesized to be the result of more realistic ocean and land surface models. The ocean in particular is characterized by numerous modes of decadal variability that evolve both independently and in tandem with other synchronous modes. Because of this continuous evolution, it is possible that particular regimes of oceanic variability offer greater decadal predictability than others – periods of which are known as forecasts of opportunity.

We show that nonlinear modes of sea-surface temperature (SST) variability offer forecasts of opportunity for continental surface temperature and precipitation on decadal timescales within fully coupled climate models. In particular, we identify the evolutions of nonlinear SST patterns that are associated with the dominant modes of decadal surface temperature and precipitation variability across the globe using a physically interpretable neural network. To ensure these modes are not statistical constructs, we then constrain the SST of a simplified climate model to the identified SST patterns and test the modeled response. The identified modes of SST variability lend statistically significant predictability on continental scales within the climate simulations, particularly over the North American and European sectors.