I'm really not a fan of ERSST, for a variety of reasons. It's constantly being updated w/ a bunch of shady homogenizations and does not adequately utilize satellite observations.
ERSSTv3 used satellite data, however after several attempts to correct a residual cold bias in the data that emanated from cloud contamination of infrared based data from AVHRR for example (which is used in the NESDIS SST product,) most weren't sure how to correct this. As I note below from Huang et al's paper that was published last October, ERSSTv4 has taken a few more strides to integrate satellite data into the dataset, as opposed to ERSSTv3b. Of course about 75% of the historical record has absolutely no satellite data in it, which of course means the length of record of the satellite derived sets is incredibly short, w/ most datasets only containing enough data for at the very most 1 30-yr base period (1981-2010), if at all. The satellite data will likely become more important in the coming decades to the ONI calculations. as new 30 year base periods (1986-2015, 1991-2020, 1996-2025, 2001-2030) become available in 2016, 2021, 2026, & 2031 respectively, and I plan on making the tri-monthly data from some of the satellite derived sets available in my combined ONI reconstruction. I've finished the calculations for the first 3 datasets, and even at first glance, HADCRUT4 is exceptionally noisy before the mid-late portions of the 20th century. The combination of SSTs & Near Surface Air temps accounts for 2 of the 6 variables used in the MEI, and are the 2 strongest components of the MEI, showing correlations 0.96 & 0.95 to the index itself according to Wolter & Timlin. (1998) Eventually, I may try my hand at integrating SLP into this index, but cloudiness, meridional, & zonal winds will likely remain left out since they weren't reliably accounted for (especially the former) until the satellite era.
ERSSTv4 uses 130 EOT derived from in situ & satellite based SSTs in OISSTv2 as opposed to CODAS that was implemented in v3b, most notably, this correction allows for a realistic depiction of the 1877-78 Super El Nino, which was intentionally dampened in ERSSTv3b (along w/ all the pre 1880 data, hence, even though it's length of record is equivalent to v4, I've excluded it from this ONI analysis until 1895 when a 30 year base period w/ "reliable" data finally becomes accessible.
ERSST made the following improvements to v3b to produce the newer v4 dataset.
SST data now is infused with a more comprehensive ICOADS v2.5 dataset (1875-2007), which has more ship observations, across all timescales, especially during the 1880s & has been infused with more buoy data from 1970-1995 (w/ >50% more buoy data during the 1980s)
NCEP GTS (Global Telecommunication Systems) is used in SSTs for the post 2007 period
Ice data is derived from HADISST (1870-2010) & NCEP instead of UKMO (1870-1980) & GFSC (1981-2004)
The 130 EOTs continue to be derived from OISSTv2, however v4 now uses OISSTv2 thru 2011 instead of 2005. & the EOT criterion have been lowered to 0.1 from 0.2. This allows for ERSSTv4 to present a more realistic depiction of 19th century ENSO events, which in this case, allows for the 1877-78 & 1888-89 Super El Ninos to be the 2 strongest El Ninos in the historical record, 1997-98 has been knocked down to 3rd place overall in ERSSTv4.
SST STD for Quality Control in ERSSTv4 is now based on the in situ & satellite data from OISSTv2 during the 1982-2011 period instead of COADS (1950-1979).
The SST Anomaly calculations now actually use the actual in situ locations of observations as opposed to grid boxes
Nearby anomaly filling is used to fill in the low frequency gaps in observations.
Nighttime Marine Air Temp (NMAT) adjustment now uses HadNMAT2 vs IOCADS v2.4. HadNMAT2 includes more strict QC & accounts for the bias in generally increasing ship height over the observational record (that's related to increasing ship size) that may have lead to a cool bias in on-deck air temp observations.
A Lowess filter coefficient is used for bias adjustment smoothing vs linear interpolation.
Ship-Buoy SST adjustment was finally applied in ERSSTv4, which added 0.12C to buoy observations.
The weighting function in ERSST has also been changed in v4 to allow grid boxes w/ greater observational coverage &/or lower potential random errors to take precedent over others, in ERSSTv3b all grid boxes were given equal weight into the final SST product regardless of their obs. coverage/errors.