Reduced Space Optimal Estimation has been applied to the global sea surface temperature (SST) record MOHSST5 (ATLAS7) from the U.K. Meteorological Office to produce 136 years of analyzed global SST anomalies (with regards to normals of 1951-1980) , where data gaps are removed and sampling errors are diminished. The results of this analysis are available as a dataset, including an interactive viewer and downloadable data files. This analysis is described in the paper

Kaplan, A., M. Cane, Y. Kushnir, A. Clement, M. Blumenthal, and B. Rajagopalan,
Analyses of global sea surface temperature 1856-1991, *Journal of
Geophysical Research*, **103**, 18,567-18,589, 1998

The method itself is described in

Kaplan, A., Y. Kushnir, M. Cane, and M. Blumenthal, Reduced space optimal
analysis for historical datasets: 136 years of Atlantic sea surface
temperatures, *Journal of Geophysical Research*, **102**,
27,835--27,860, 1997.

The dataset contains results of all 5 stages of the analysis:
projection, Optimal Interpolation (OI), Kalman Filter (KF) forecast,
KF analysis, and Optimal Smoother (OS). Each stage gives the estimate
of the field and corresponding theoretical error estimate. The
projection gives the poorest results: it creates large spurious anomalies in
the areas where no observations were available. The OI regularizes
the solution and suppress such anomalies. The KF forecast is what Markov
model (MM) of time transitions predicts a month ahead, which is a weak
field with a possibly wrong structure. The KF analysis is much better: it
is an improvement over the OI in that it uses MM to provide
propagation of information forward in time. The Optimal
Smoother (OS) provides further improvement by using MM in both
directions in time. ** The use of OS is recommended for all practical applications **.

In the process of analysis we use information on number of observations for every gridbox. Since it is not a part of MOHSST5, we took its 1x1 version from MOHSST6, averaged it on 5x5 with capping it by maximum of 60 monthly observations per 1x1 box. The resulting coverage fields are put into this dataset.

The input data (the MOHSST5 version of the GOSTA dataset) from the U.K. Meteorological Office on which the analysis was based can be obtained from the U.K. Hadley Centre or from a link here.

The 1951-1980 normals with regards to which the anomalies were
computed can be accessed here. These
fields need to be added to the analyzed (e.g. OS) anomalies in order
to produce the corresponding analyzed values of the **total** SSTs.

*This documentation has last been updated on:* 27 April, 2011