Comparison of retrospective analyses of the global ocean heat content

In this study, we compare seven retrospective analyses of basin- to global- scale upper ocean temperature. The analyses span a minimum of 10 years duri ng the 50-year period since World War II. Three of the analyses (WOA-94, WH ITE, BMRC) are based on objective analysis and thus, do not rely on a numer ical forecast model. The remaining four (NCEP, WAJSOWICZ, ROSATI, SODA) are based on data assimilation in which the numerical forecast is provided by some form of the Geophysical Fluid Dynamics Laboratory Modular Ocean Model driven by historical winds. The comparison presented here is limited to hea t content in the upper 250 m, information that is available for all analyse s. The results are presented in three frequency bands: seasonal, interannua l (periods of 1-5 years), and decadal (periods of 5-25 years). At seasonal frequencies, all of the analyses are quite similar. Otherwise, the differen ces among analyses are limited to the regions of the western boundary curre nts, and some regions in the Southern Hemisphere. At interannual frequencie s, significant differences appear between the objective analyses and the da ta assimilation analyses. Along the equator in the Pacific, where variabili ty is dominated by El Nino, the objective analyses have somewhat noisier fi elds, as well as reduced variance prior to 1980 due to lack of observations . Still, the correlation among analyses generally exceeds 80% in this regio n. Along the equator in the Atlantic, the correlation is lower (30-60%) alt hough inspection of the time series shows that the same biennial progressio n of warm and cool events appears in all analyses since 1980. In the midlat itude Pacific agreement among objective analyses and data assimilation anal yses is good. The analysis of Rosati et al. [Rosati, A., Gudgel, R., Miyako da, K., 1995. Decadal analysis produced from an ocean assimilation system. Mon. Weather Rev., 123, 2, 206.] differs somewhat from the others apparentl y because in this analysis, the forecast model is weighted more heavily rel ative to the observations. The analysis of Levitus et al. [Levitus. S., Boy er, T.P., Antonov, J., 1994. Interannual variability of upper ocean thermal structure. World Ocean Atlas, 1994, Vol. 5. Natl. Env. Satell. Data and In t. Serv., Natl. Oceanic and Atmos. Admin. Atlas series, Washington, DC, 176 pp.] has a much different spatial distribution of variability in the inter annual band than the others. Partly, this results from the yearly time-aver aging of this analysis. Three of the monthly analyses extend over 20 years and thus are useful for examining decadal variations. Comparison of these a nalyses shows in common a slow progression of warm water westward and then eastward along the equator in the tropical Pacific that is linked to the de cadal fluctuations of El Nino. (C) 1999 Elsevier Science B.V. All rights re served.