Impact of Diurnal Warming on Assimilation of Satellite Observations of Sea Surface Temperature.

摘要

Sea surface temperature (SST) varies on a range of temporal scales according to variations in insolation, advection, and mixing. A prominent diurnal signal can frequently be identified in the SST of midlatitude to tropical regions, particularly under conditions of high insolation and low wind speed. Case studies in the Gulf of Mexico and Mediterranean Sea are used to examine the impact of such variations on assimilative SST analyses and forecasts. The scenarios provide infrared observations from polar-orbiting or geostationary satellites to an assimilative ocean model using a 24-hour update cycle. SST innovations are determined relative to the prior 24-hour SST forecast or using a first guess at the appropriate time (FGAT) approach which matches each observation to its corresponding time-varying forecast. It was anticipated that the FGAT would have its largest impact in the Gulf of Mexico summer, when the occurrence of the relatively large diurnal cycle maximum is nearly in phase with the nowcast. In contrast, FGAT was anticipated to have relatively little impact in the Mediterranean summer, where the diurnal maximum and nowcast are 90 out of phase. The impact of FGAT in the fall-spring seasons would be more affected by the skill in forecasts of the non-diurnal trend, as the diurnal signal is smaller in these seasons. FGAT is found to have its largest benefit in reduction in the mean error of the SST forecasts; its impact on standard deviation is mixed. It is also found to have larger impact in the cases assimilating observations from geostationary satellites, which give a broad sample of SST over all times of the day. Observations from the polar orbiter come at a sun-synchronous 10:00 AM or PM, sampling near the midpoints of the diurnal variation. The effectiveness of FGAT is dependent on model forecast skill and effective only if the model is able to adequately predict diurnal or other dominant variations between analysis times.