Predictability of Multiyear Trends of the Pacific Decadal Oscillation in an MPI‐ESM Hindcast Ensemble

While hindcast skill for the Pacific Decadal Oscillation (PDO) has so far been limited to a few years, we present hindcast skill for PDO trends up to 10 years ahead. Our analysis is based on an initialized hindcast ensemble with the global Max Planck Institute Earth System Model (MPI‐ESM). As in previous studies, we find hindcast skill limited to a few years, when we first construct a lead‐year time series, from which we second calculate the PDO. We find similar hindcast skill when we first calculate the PDO for each start year and second construct a lead‐year time series. However, we find hindcast skill considerably increased, when we first calculate the PDO for each start year, second estimate multiyear trends, and third construct a lead‐year time series. Our results suggest hindcast skill for the low‐frequency variability of the PDO, which holds important implications for predictability analyses of other modes of long‐term climate variability. Plain Language Summary The dominant multiyear variability pattern of the sea surface temperature in the Pacific is described as the Pacific Decadal Oscillation (PDO). So far, skillful predictions of the PDO are only available 2–3 years ahead. In our study, we show that PDO trends might be skillfully predictable for up to 10 years ahead. We achieve this increase in predictive skill by both focusing on the prediction of multiyear trends rather than the exact state and modifying the processing of the analyzed time series. We suggest that this shift in perspective might turn to be useful also for further decadal prediction analyses of multiyear variability. Key Points PDO variability in both observations and hindcast ensemble can be robustly estimated by multiyear trends For smoothed multiyear PDO trends we find significant hindcast skill in the hindcast ensemble PDO hindcast skill in MPI‐ESM is limited to a few lead years using the conventional lead‐year method