SST Errors Growth in the Tropical Pacific Ocean from High Resolution Coupled Model

Significant SST errors in the tropical oceans are common in coupled ocean-atmosphere general circulation models. These systematic errors can develop quickly (fast errors) or may result from model drift (saturated errors). The saturated errors can easily reach larger magnitudes than the desired climate signals, calling the prediction into question. One hypothesis holds that the errors may decrease if a model’s resolution is increased and more of the sub-grid-scale processes become resolvable, such as mixing and stratification. In this study, a set of high resolution (1/10o oceanic and 1/2o atmospheric) January 1st hindcasts of twelve-month integrations initialized by real-time oceanic and atmospheric reanalyses (CFSR) from 1982 to 2003 are examined. Our goal is quantify the processes responsible for the development of SST errors in the Community Climate System Model (CCSM4). The SST fast errors are higher and positive (~+0.1oC) to the north of equator. On the other hand, south of the equator the errors are smaller and negative (~-0.05oC). The largest errors occur at the continental boundaries and in the equatorial region. The SST saturated errors are spatially more ubiquitous and larger (~+-1oC), with positive equatorial errors compensated by negative errors at midlatitudes (between 20o and 30o). Five subareas were selected according to the largest errors’ variances. Three of them correspond to El Niño Index regions (1+2, 3, 4) and the other two are offshore regions of Guatemala/El Salvador and Chile/Peru. The cause of the fast errors in the tropical Pacific are primarily oceanographic. Nevertheless, saturated errors have different signature in each subarea. The region near Central America has a strong contribution from wind-induced upwelling, while that in the southeast Pacific has a strong atmospheric contribution from cloud-affected radiation. In sum these reveal a sensitivity to physical processes similar to that inferred in low-resolution models. The talk will also compare similar error calculations from a parallel set of forecasts from a lower resolution version of this model.