Improved Global Net Surface Heat Flux

Surface heat flux estimates from widely used atmospheric reanalyses differ locally by 10–30 W m−2 even in time mean. Here a method is presented to help identify the errors causing these differences and to reduce these errors by exploiting hydrographic observations and the resulting temperature increments produced by an ocean reanalysis. The method is applied to improve the climatological monthly net surface heat fluxes from three atmospheric reanalyses: MERRA‐2, ERA‐Interim, and JRA‐55, during an 8 year test period 2007–2014. The results show that the time mean error, as evaluated by consistency with the ocean heat budget, is reduced to less than ±5 W m−2 over much of the subtropical and midlatitude ocean. For the global ocean, after all the corrections have been made, the 8 year mean global net surface heat imbalance has been reduced to 3.4 W m−2. A method is also presented to quantify the uncertainty in the heat flux estimates by repeating the procedure with many different atmospheric reanalyses and then examining the resulting spread in estimates. This reevaluation of net surface flux reveals, among other results, that the Southern Ocean is a source of heat to the atmosphere. Plain Language Summary Heat exchange between the atmosphere and ocean is the primary way the ocean gains or loses heat, and thus is a key indicator of climate variability and change. Current estimates of surface heat exchange, also known as surface heat flux, are primarily based on meteorological observations and models and are uncertain. This article proposes a method to improve these estimates by exploiting additional information contained in ocean observations in an algorithm known as data assimilation. Tests of this method are shown to reduce the uncertainty of surface heat flux by up to a factor of five. Among the unexpected results is a recognition that the Southern Ocean is losing heat, rather than gaining it from the atmosphere. Key Points Current atmospheric reanalysis heat flux estimates differ by 10–30 W m^–2 We identify the errors and reduce them to