Well, the pper is <a href="http://onlinelibrary.wiley.com/doi/10.1002/qj.2297/full" rel="nofollow">here</a>. Sec 5.2 is on SST. Here is the full text:
<i>"SSTs are better modelled by ordinary kriging. When the calculation is performed with SSTs only, the range of the variogram is 915 km (compared with 830 km with land–ocean data) and the optimum cross-validation results are obtained for s = 0.0 (i.e. ordinary kriging) or s = 0.2; this is consistent with the poor correlation between SST and satellite temperatures (section 3). However, the kriging results are only marginally better than the null reconstruction.
Given the difference in the optimum value of s, is it reasonable to use a single approach for land and ocean data? For short extrapolation ranges (e.g. one cell in latitude or 550 km), the difference between kriging and the hybrid method is small and at midlatitudes the unobserved regions in the SST data tend to be small and isolated; thus the choice of infilling method makes little difference.
The only large contiguous unobserved regions in the SST data are in the Arctic and Southern Oceans. These regions are also characterized by seasonal or perennial sea ice, which must be considered separately."</i>