Impact of Bathythermograph Temperature Bias Models on an Ocean Reanalysis

Historical bathythermograph datasets are known to be biased, and there have been several efforts to model this bias. Three different correction models of temperature bias in the historical bathythermograph dataset are compared here: the steady model of Hanawa et al. and the time-dependent models of Levitus et al. and Wijffels et al. The impact of these different models is examined in the context of global analysis experiments using the Simple Ocean Data Assimilation system. The results show that the two time-dependent bias models significantly reduce warm bias in global heat content, notably in the 10 years starting in the early 1970s and again in the early 1990s. Overall, the Levitus et al. model has its greatest impact near the surface and the Wijffels et al. model has its greatest impact at subtropical thermocline depths. Examination of the vertical structure of temperature error shows that at thermocline depths the Wijffels et al. model overcompensates, leading to a slight cool bias, while at shallow levels the same model causes a slight warm bias in the central and eastern subtropics and at thermocline depths on the equator in the Pacific Ocean as a result of reduced vertical entrainment. The results also show that the bias-correction models may alter the representation of interannual variability. During the 1997/98 El Niño and the subsequent La Niña the Levitus et al. model, which has its main impact at shallow depths, reduces the 50-m temperature anomalies in the eastern equatorial Pacific by 10%–20% and strengthens the zonal currents by up to 50%. The Wijffels et al. correction, which has its main impact at deeper levels, has much less effect on the oceanic expression of ENSO.