Skill of U.S. Coastal SST Forecasts from the North American Multimodel Ensemble

Variability in the ocean state, especially the sea surface temperature (SST), is known to strongly influence marine ecosystems. As a first step in the process of ecological prediction, we explored SST forecasts in large marine ecosystems (LMEs), from the coupled climate models in the North American Multi-Model Ensemble (NMME, Kirtman et al. 2014, BAMS). Forecasts were evaluated at monthly time scales out to one year. Skill was assesed based on deterministic measures, including anomaly corrleation and rms error and the Brier skill score, a measure of the probabilistic forecast skill, i.e. what chance would a predicted SST anomaly be above (upper tercile) or below (lower tercile) average. As in most regions, the ensemble mean monthly SST predictions have skill and it is greater than those from most individual models, especially for probabilty forecasts. The ensemble provides a better estimate of the full range of forecast values than any individual model, thereby correcting for the systematic over-confidence (under-dispersion) of predictions from an individual model.

We explored the forecasts for the California Current System (CCS) in more detail, using the Canadian forecast model (CanCM4), perhaps the most skillful NMME model in the CCS. We evalauted several mechanisms that could drive SST predictability in the CCS: skill mainly arouse due to ENSO teleconnections to the extratropics and persistence of SST anomalies.