Exploring sources of error in decadal predictions of the Pacific surface and sub-surface ocean

A recent 40-member large ensemble of decadal predictions using the Community Earth System Model (CESM-DPLE) shows good levels of skill in many fields in the ocean and in the atmosphere. However, some regions such as the tropical Pacific show very low skill for the surface ocean over the first few lead years. Here, we develop a Linear Inverse Model (LIM) to generate decadal forecasts for the surface and sub-surface ocean and compare its decadal forecast skill against that from the CESM-DPLE. The LIM was constructed using global observed seasonally-averaged anomalies over the period of 1958-2017. To be consistent with CESM-DPLE forecasts, LIM forecasts were initialized every November and integrated forward from one season out to 10 years.

The results show that LIM forecast skill is comparable to and sometimes better than that from the CESM-DPLE ensemble mean over many regions around the globe. In particular, LIM skill is better in the Eastern Subtropical Pacific for lead times from 1 to 5 years. These results indicate that the evolution of the system in those areas may not be not fully driven by unpredictable dynamics and that there may be some room for improvement in the CESM decadal forecasts, given that a low-dimensional linear model is able to generate better skill than the fully-coupled nonlinear model. Therefore, the low skill in the CESM may be related to the misrepresentation of processes in the ocean within and among those regions that exhibit low skill. We investigate possible sources of errors by comparing the LIM feedback matrix obtained from observations with that obtained from the CESM. This matrix show time-scale interactions between components, that in our case represent different subregions of the Pacific Ocean. Preliminary results show that the oceanic feedback matrix from CESM differs in several regions from that obtained with observations. These may indicate sources of error in CESM and therefore in its decadal prediction skill that merit focused attention in future development work.