Predicting Pacific Decadal Variability Using State Space Reconstruction

Pacific Decadal Variability (PDV) is a key driver of global climate with important consequences for hydroclimate on several continents, marine ecosystems, and the rate of global mean surface temperature increase under anthropogenic greenhouse gas forcing. For these reasons, the ability to predict changes in PDV phase would be highly valuable. Earlier work has shown that the PDV in the Zebiak-Cane model has some potential predictability. However, the small number of PDV phase shifts that have occurred within the period of the observational record and the lack of consensus on the mechanisms that produce these phase shifts pose considerable challenges to understanding the nature of this potential predictability and applying it to make predictions of this phenomenon.

In order to overcome these challenges, we borrow methods from theoretical ecology and data mining that allow us to reconstruct the full state space of the tropical Pacific system from a single index. We perform this reconstruction using the NINO3 index from a 100,000-year unforced run of the Zebiak-Cane model. This reconstruction of the state space allows us to identify regime-like behavior of the tropical Pacific system in this model on decadal timescales that corresponds to the PDV. We find that the transitions into negative PDV states occur in a localized region of the state space, allowing for high-skill prediction. We then examine the characteristics of this region of the state space in terms of physical variables such as sea surface temperatures, thermocline depth, and wind stress in the tropical Pacific, and draw a comparison of this state of the system in the model with the state of the system at the beginning of known transitions in the observations with the ultimate aim of developing a method to detect PDV phase shifts in the real world as they occur.