. CMIP5/CMIP6 model-analog seasonal forecast skill: a metric for model evaluation of ENSO dynamics

Abstract
Recently, Ding et al showed that tropical Indo-Pacific SST forecast skill (both deterministic and probabilistic) from the operational models of the North America Multi-Model Ensemble (NMME) can be matched or even exceeded with a simple "model-analog" method, applied to the corresponding long control runs from each NMME model. In this method, forecast initialization is made directly on a model's own attractor, using an analog approach where model states close to the observed initial state are drawn from a library obtained from prior uninitialized CGCM simulations. The subsequent evolution of those model-analogs yields a forecast ensemble, without additional model integration. This suggests seasonal forecasts can be made using almost all CMIP models. We have made retrospective tropical Indo-Pacific forecasts for the 1961-2015 period from 28 CMIP5 models (and available CMIP6 models), using model-analogs. All show skill for sea surface temperature (SST) forecasts, but there is a considerable range among the models, with even greater skill differences for precipitation. Model-analog forecasts from the ten "best" CMIP5 models have skill for SST and precipitation comparable to that of both the NMME model-analog forecast ensemble and (since 1982) traditional assimilation-initialized bias-corrected NMME hindcasts. ENSO forecast skill has no trend over the 55-yr period. Its decadal variations appear to be random, although skill does improve during epochs of increased ENSO activity. Including the CMIP5-projected effects of external radiative forcings improves the tropical SST skill of the model-analog forecasts, but not within the ENSO region. Finally, we demonstrate that the 55-yr tropical Pacific ocean hindcast skill, plotted on a Taylor diagram for leads of 1-12 months, is also a robust metric of how each CMIP5/6 model's attractor corresponds to nature's attractor, since it directly measures the fidelity of each model's ENSO dynamics (rather than static measures of ENSO) to the observed evolution of all ENSO events.