Accuracy and Bias in Global Ocean Reanalyses

Before global coupled atmosphere/ocean reanalyses become practical it is necessary to identify and correct bias in the individual system components. Here we focus on temperature bias in the ocean. Bias in ocean reanalysis temperature enters through errors in initial conditions, surface fluxes, the corrections used to account for observation bias, error statistics, and model physics and numerics. Here we examine the impact of these error sources through a dozen reanalysis experiments all using a single ocean reanalysis system (SODA3). Initial condition errors do not affect reanalyses spanning recent decades.

Surprisingly, our experiments also confirm that errors in surface flux are also fairly easy to correct and have restricted impact on the evolving ocean state. In the attached figure we show 0/300m temperature as function of time for the Pacific Ocean for the 27 year period 1980-2016. The different colored curves identify an ocean reanalysis forced with five different atmospheric reanalysis forcing data sets (MERRA-2, ERA-I, JRA-55, CORE2, DFS5.2). The gradual warming of the Pacific by 0.2C is evident in rise of all of these time series, as is its rough correspondence to the warming of Pacific SST (grey curve). The insensitivity of the warming estimates to the specific details of the atmospheric forcing is evident in the close similarity of the individual colored curves.

Instead, the significant differences among the current generation of ocean reanalyses seem to result from differing treatments of observation bias correction, error statistics, and model physics and numerics. To explore the first of these we test the impact of time dependent biases in sea surface temperature and bathythermograph profile observations. Among the consequences of both the bias and the geographic limitations of the bathythermograph observation sampling is an apparent increase in the temperature of the upper ocean in the early 2000s coincident with the appearance of large numbers of profiling float records of temperature and salinity. Incidentally, this bathythermograph bias and its interaction with subsequent data from the profiling float observations seem to have led to an overestimation of the size of the climate hiatus of the late-1990s.

We conclude this presentation with a comparison of three ‘next-generation’ ocean reanalyses: ORAS5, ECCO4r3, and SODA3.