Hybrid SST Retrieval Algorithm in thermal IR: Combining Regression and Radiative Transfer Model Approaches

Retrieval of sea surface temperature (SST) from top-of-the atmosphere thermal IR radiances is a well established remote sensing technique. The most commonly used approach to SST retrieval is regression, derived from matchups of sensor brightness temperatures (BT) with in situ SST. The major limitations of the regression method are biases in the retrieved SST, which depend upon observational conditions and may easily reach several tenths of a degree Kelvin. Recent developments of fast Radiative Transfer Models (RTMs), along with analysis SST (e.g., Reynolds, OSTIA etc.) and Numerical Weather Prediction (e.g., NCEP, ECMWF) atmospheric fields have given rise to SST algorithms aimed at more comprehensive accounting for variations in local atmospheric transmission. The SST retrieval problem can be now posed in the incremental formulation, i.e., as restoring the difference between SST and the first-guess field from the difference between observed and simulated BT. In particular, the Optimal Estimation (OE) technique was recently applied to SST retrieval from 11 and 12 μm channels of MetOp AVHRR and MSG2 SEVIRI. It has been found however that OE underestimates spatial and temporal SST variations due to stabilizing the solution of the ill-posed set of RTM equations by forcing it to the low-space and time resolution first guess SST field. This presentation describes an alternative incremental algorithm – Hybrid (i.e., mixing regression and RTM). The Hybrid algorithm uses first guess SSTs and BTs to establish regression between deviations of in situ SST and observed BT from the corresponding first guesses, thus avoiding inversion of the set of RTM equations. The Hybrid SST algorithm, along with conventional regression and OE algorithm, has been implemented within the Advanced Clear-Sky Processor for Oceans at NESDIS. Based on results of sensitivity studies of all three algorithms using data of polar-orbiting AVHRR and geostationary SEVIRI sensors, we show that the hybrid algorithm provides a reasonable tradeoff between the pure regression and pure RTM inversion. Compared with conventional regression, the Hybrid algorithm minimizes local SST biases, in particular, at extreme values of view zenith angle and water vapor content in the atmosphere. Compared with RTM inversion, the Hybrid algorithm provides more objective weighting of observations and a priori information, thus avoiding underestimation of SST variations.