I used CRU TS3.22 for an estimate of the high end of the Elusive Global Surface Temperature. Berkeley Earth Surface Temperature (BEST) is a global land product that includes an estimate of the actual honest to goodness surface temperature with an amazing +/- 0.06 C of accuracy. That is pretty hard to beat doncha know. Well, I am not all that convinced that BEST actually represents a useful surface temperature. It is a nice product, but "useful" surface temperature is an energy thing.
If I cut out the Antarctic by limiting the comparison to 55 south, the actually temperatures show that the oceans are in the drivers seat. Best which long range interpolate as needed, that is a feature of kriging, track SST extremely closely while CRUTS3.22 provides more weight to local temperature. One isn't particularly better than the other, they have two different functions. Since the average temperature of the Antarctic is around -35 C degrees, you have the large difference in southern hemisphere land temperature.
Well here is the energy issue. Antarctica has an average elevation of over 2000 meters. It should be about 2x9.8 = 19.6 degrees colder than the surface due the dry adiabatic lapse rate. If you are estimating the southern hemisphere's "surface" energy do you use actual temperature of "potential" temperature? If you are in charge of the agricultural department in Paraguay, do you use BEST or CRUTS3.22? You have to chose the one that best fits your intended purpose.
On the whole BEST is a great product but it doesn't fill every bill. Its uncertainty is a "global" thing and its purpose was to double check other "global" products. Even though BEST has regional and local "products" you need to be a bit "skeptical" before blindly passing judgement based on BEST all by itself.