Determination of Vertical Thermal Structure from Sea Surface Temperature

A recently developed parametric model by P. C. Chu et al. is used in this paper for determining subsurface thermal structure from satellite sea surface temperature observations. Based on a layered structure of temperature fields (mixed layer, thermocline, and lower layers), the parametric model transforms a vertical profile into several parameters: sea surface temperature (SST), mixed layer depth (MLD), thermocline bottom depth (TBD), thermocline temperature gradient (TTG), and deep layer stratification (DLS). These parameters vary on different timescales: SST and MLD on scales of minutes to hours, TBD and TTG on months to seasons, and DLS on an even longer timescale. If the long timescale parameters such as TBD, TTD, and DLS are known (or given by climatological values), the degree of freedom of a vertical profile fitted by the model reduces to one: SST. When SST is observed, one may invert MLD, and, in turn, the vertical temperature profile with the known long timescale parameters: TBD, TTG, and DLS. The U.S. Navy’s Master Oceanographic Observation Data Set (MOODS) for the South China Sea in May 1932–94 (10 153 profiles) was used for the study. Among them, there are 40 data points collocating and coappearing (same week) with the weekly daytime NASA multichannel SST data in 1986–94. The 40 MOODS profiles were treated as a test dataset. The MOODS dataset excluding the test data is the training dataset, consisting of 10 113 profiles. The training dataset was processed into a dataset consisting of SST, MLD, TBD, TTG, and DLS using the parametric model. SST from the test dataset was used for the inversion based on the known information on TBD, TTG, and DLS. The 40 inverted profiles agreed quite well with the corresponding observed profiles. The rms error is 0.728C, and the correlation between the inverted and observed profiles is 0.79. This is much better than the simple method of estimating subsurface temperature anomaly from SST anomaly by correlating the two in the training dataset. The possibility of using this method globally is also discussed.