Simulated and observed variability in oceantemperature and heat contentK. M. AchutaRao*†, M. Ishii‡, B. D. Santer*, P. J. Gleckler*, K. E. Taylor*, T. P. Barnett§, D. W. Pierce§, R. J. Stouffer¶,and T. M. L. Wigley储*Program for Climate Model Diagnosis and Intercomparison, Lawrence Livermore National Laboratory, Livermore, CA 94550;‡Frontier Research Center forGlobal Change, Japan Agency for Marine–Earth Science and Technology, Yokohama 236-0001, Japan;§Climate Research Division, Scripps Institution ofOceanography, La Jolla, CA 92037;¶National Oceanic and Atmospheric Administration/Geophysical Fluid Dynamics Laboratory, Princeton, NJ 08542;and储National Center for Atmospheric Research, Boulder, CO 80307Edited by Carl Wunsch, Massachusetts Institute of Technology, Cambridge, MA, and approved May 16, 2007 (received for review December 20, 2006)Observations show both a pronounced increase in ocean heatcontent (OHC) over the second half of the 20th century andsubstantial OHC variability on interannual-to-decadal time scales.Although climate models are able to simulate overall changes inOHC, they are generally thought to underestimate the amplitudeof OHC variability. Using simulations of 20th century climateperformed with 13 numerical models, we demonstrate that theapparent discrepancy between modeled and observed variability islargely explained by accounting for changes in observationalcoverage and instrumentation and by including the effects ofvolcanic eruptions. Our work does not support the recent claimthat the 0- to 700-m layer of the global ocean experienced asubstantial OHC decrease over the 2003 to 2005 time period. Weshow that the 2003–2005 cooling is largely an artifact of a system-atic change in the observing system, with the deployment of Argofloats reducing a warm bias in the original observing system.climate 兩 models 兩 observations 兩 ocean heat contentObservations suggest that the world’s oceans were responsi-ble for most of the heat content increase in the earth’sclimate system between 1955 and 1998 (1). This increase isembedded in substantial variability on interannual-to-decadaltime scales. State-of-the-art climate models have been able toreplicate both the overall increase in ocean heat c ontent (OHC)during this period and its horizontal and vertical structure (2–7).Such detection and attribution studies have identified a largeanthropogen ic c omponent in the observed changes and find thatthe ‘‘noise’’ of natural climate variability is an inadequateex planation for these changes.The credibility of these results is strongly dependent on thereliabilit y of natural variability estimates, particularly on themultidecadal time scales against which a slowly evolving anthro-pogen ic signal must be discerned. This low-frequency noiseinfor mation cannot be obtained from the relatively short (45- to50-year) observational record and is typically estimated f rommodel ‘‘undisturbed earth’’ experiments (‘‘control runs’’), whichassume no changes in greenhouse gases or other external forc-ings (8). Several studies have reported that models may signif-icantly underestimate the observed OHC variability (3, 9, 10),raising concerns about the reliability of detection and attributionfindings (11, 12).A lthough observational estimates of OHC change given in the2005 World Ocean Atlas (WOA-2005) (1) are based on millionsof individual temperature measurements, these measurementsare unevenly distributed in space and time. Until recently, manyportions of the global ocean were poorly sampled. To reconstructthe true (but unknown) four-dimensional structure of globalocean temperature and OHC changes, it is necessary to ‘‘infill’’missing data. This has been done using either statistical ap-proaches (1, 3, 13, 14) or physically based ocean models (15).Because there is no unique solution to the infilling problem,and in view of concerns that previously applied statisticalinfilling approaches may alter ocean temperature variability (7,16), it is preferable to restrict comparisons of modeled andobserved variabilit y to the actually observed portions of theocean and, hence, to volume-averaged ocean temperature ratherthan OHC. This type of ‘‘model subsampling’’ strategy has beenused in recent detection and attribution work (5, 6).A prev ious study (16) employed model results f rom controlr uns and an idealized climate change experiment (17) to inves-tigate the impact of incomplete space- and time-varying obser-vational data coverage on simulated estimates of ocean temper-ature variability. Results were reported f rom eight differentatmosphere/ocean general circulation models. Subsampling spa-tially complete model data with the observational data coveragemask amplified the temporal variability of ocean temperatures.In the control runs, the variability estimated from subsampleddat a was below variability levels in the subsampled observations.In the idealized experiment with 1%/year atmospheric CO2increases, however, the simulated variability of subsampled dat awas consistently larger than observed, primarily because of theunrealistically large CO2forcing (compared with the estimatedobserved forcing).To evaluate the ability of models to simulate the observedamplitude of ocean temperature variability, it is therefore im-port ant to analyze model experiments that employ realisticestimates of historical forcings and to account for observationalc overage and instrumentation changes. We consider all threeissues here and address uncertainties in both model results andin the observations themselves.Model and Observational DataWe examine a suite of recently completed climate model sim-ulations carried out in support of the Fourth Assessment Reportof the Intergovernmental Panel on Climate Change. Unlike theidealized experiments used in ref. 16, the simulations of 20th-century climate change [designated ‘‘20c3m’’ runs in the WorldClimate Research Program’s Coupled Model Interc omparisonProject Phase 3 (WCRP CMIP3) data archive] include estimatedAuthor contributions: K.M.A., B.D.S., K.E.T., T.P.B., R.J.S., and T.M.L.W. designed research;K.M.A., B.D.S., P.J.G., K.E.T., and D.W.P. performed research; M.I. contributed new re-agents/analytic tools; K.M.A., M.I., P.J.G., and D.W.P. analyzed data; and K.M.A., B.D.S.,P.J.G., K.E.T., T.P.B., D.W.P., R.J.S., and T.M.L.W. wrote the paper.The authors declare no conflict of interest.This article is a