Optimal Estimation Approach to Sea Surface Temperature Retrieval from Satellite IR Radiometers and its Dependence on the a Priori

The Optimal Estimation (OE) approach gained popularity in the satellite sea surface temperature (SST) retrieval community with the promise of improving accuracy over the traditional non-linear SST (NLSST) retrieval algorithms, especially overcoming their problem with regional biases.

The NLSST algorithms rely on coefficients that are obtained by least-squares regression to in situ data (mostly drifting buoy measurements) and therefore produce good estimates of SST in conditions representative of those use to derive these coefficients. When the sea or atmospheric conditions depart form the average state, the NLSST estimates can be quite erroneous. One way of reducing the NLSST errors is to derive multiple sets of coefficients specific to different seasons and geographical regions. In the OE approach a prior knowledge of the state of the ocean and the atmosphere at the time and place each measurement is used to derive the SST so the OESST estimates should in principle be bias free.

However, the OE process can depend rather strongly on the a priori information. If for example the a priori SST field used is biased the OE will produce biased OESST estimates. Since typically a good a priori is needed to produce good OESST the question arises how much the OESST estimates are an improvement on the a priori itself. We try to answer this question with an example of OESST retrievals for the MODIS instrument and the data in the MODIS match-up database.