Pacific decadal oscillation hindcasts relevantto near-term climate predictionTakashi Mochizukia,1, Masayoshi Ishiia,b, Masahide Kimotoc, Yoshimitsu Chikamotoc, Masahiro Watanabec,Toru Nozawad, Takashi T. Sakamotoa, Hideo Shiogamad, Toshiyuki Awajia,e, Nozomi Sugiuraa, Takahiro Toyodaa,Sayaka Yasunakac, Hiroaki Tatebea, and Masato Moric

aJapan Agency for Marine–Earth Science and Technology, 3173-25 Showa-machi, Kanazawa-ku, Yokohama, Kanagawa 236-0001, Japan; bMeteorologicalResearch Institute, Japan Meteorological Agency, 1-1 Nagamine, Tsukuba, Ibaraki 305-0052, Japan; cCenter for Climate System Research, University ofTokyo, 5-1-5 Kashiwanoha, Kashiwa, Chiba 277-8568, Japan; dNational Institute for Environmental Studies, 16-2 Onogawa, Tsukuba, Ibaraki 305-8506,Japan; and eDepartment of Geophysics, Kyoto University, Kitashirakawa Oiwake-cho, Sakyo-ku, Kyoto 606–8502, Japan

Edited by John M. Wallace, University of Washington, Seattle, WA, and approved December 17, 2009 (received for review June 11, 2009)

Decadal-scale climate variations over the Pacific Ocean and itssurroundings are strongly related to the so-called Pacific decadaloscillation (PDO) which is coherent with wintertime climate overNorth America and Asian monsoon, and have important impactson marine ecosystems and fisheries. In a near-term climate predic-tion covering the period up to 2030, we require knowledge of thefuture state of internal variations in the climate system such as thePDO as well as the global warming signal. We perform sets ofensemble hindcast and forecast experiments using a coupled atmo-sphere-ocean climate model to examine the predictability of inter-nal variations on decadal timescales, in addition to the response toexternal forcing due to changes in concentrations of greenhousegases and aerosols, volcanic activity, and solar cycle variations.Our results highlight that an initialization of the upper-ocean stateusing historical observations is effective for successful hindcasts ofthe PDO and has a great impact on future predictions. Ensemblehindcasts for the 20th century demonstrate a predictive skill inthe upper-ocean temperature over almost a decade, particularlyaround the Kuroshio-Oyashio extension (KOE) and subtropicaloceanic frontal regions where the PDO signals are observed stron-gest. A negative tendency of the predicted PDO phase in the com-ing decade will enhance the rising trend in surface air-temperature(SAT) over east Asia and over the KOE region, and suppress it alongthe west coasts of North and South America and over the equato-rial Pacific. This suppression will contribute to a slowing down ofthe global-mean SAT rise.

climate change ∣ data assimilation ∣ decadal prediction ∣decadal variability ∣ global warming

A near-term climate prediction covering the period up to 2030is a major issue to be addressed in the next assessment report

of the Intergovernmental Panel on Climate Change (1, 2). Tomake the political decisions required to solve the socioeconomicproblems arising from climate change over the coming decades,we need to take into account the large-scale climate changesassociated with internal climate variability as well as the globalwarming signals (i.e., the response to external forcing due tochanges in concentrations of greenhouse gases and aerosols,volcanic activity, and solar cycle variations) (3–6). A globallyaveraged surface-air-temperature (SAT) forecast up to 2030depends little on specific socioeconomic scenarios or models usedin centennial climate projection experiments (7, 8). On decadaltimescales, SAT changes due to internal climate variability arecomparable to those associated with global warming in magni-tude (9). The predictability of internal climate variations is cen-tral to validating our skills in predicting the near-term climatevariations.

Prediction of internal decadal variability in the climate systemrepresents one of the newest and toughest challenges. It is onlyrecently that near-term climate projection experiments have beencarried out focusing on internal decadal variations (10–12).

Keenlyside et al. (11) have suggested that, in the coming decade,decadal-scale weakening of the Atlantic meridional overturningcirculation (AMOC) can work to suppress the globally averagedSATrise due to anthropogenic forcing. Here, we have consideredthat the Pacific decadal oscillation (PDO) (13) can also have largeimpacts on decadal-scale projections of the pan-Pacific climateand decadal-scale modulations of the global warming trend.The PDO is a dominant internal oscillation in the climate systemwith phase shifts developing on decadal timescales. After the endof the 1970s, for example, corresponding to the positive phase ofthe PDO, the upper-ocean temperatures were higher than thosein the preceding decades along the west coasts of North andSouth America and over the equatorial Pacific, while they werelower over the western and central North Pacific (14, 15). ThePDO has recently attracted much attention not only in climatestudies but also in impact assessments in the areas of agriculture,water resources, and fisheries (13, 16, 17). Successful predictionof the state of the PDO over timescales of years to one decade hasconsiderable societal benefits (18).

A near-term climate prediction requires not only good perfor-mance from the models in simulating major climate processes butalso realistic estimates of initial climate states using observationaldata, while physics of internal variations itself and the stochasticforcing can also contribute to limiting predictability. In particular,the climate prediction community faces a major difficulty inobtaining good atmospheric and oceanic initial conditions thatare compatible with both the model and the observations. Evenin recent studies using coupled atmosphere-ocean general circu-lation models (GCMs), the El Ninõ southern oscillation, which isa dominant internal fluctuation on interannual timescales, seemsonly to be predictable up to at most 1 year in advance (19, 20). Inthe present paper, we examine the impacts of the initialization onpredicting internal decadal variations and validate the hindcastskill of our coupled GCM for the PDO.

The coupled atmosphere-ocean GCM adopted here is version3.2 of the Model for Interdisciplinary Research On Climate(MIROC) in medium resolution. Using MIROC, Nozawa et al.(2005) (21) and Shiogama et al. (2007) (6) performed 10-memberensemble simulations during 1850–2000 using the historical dataof natural and anthropogenic forcing toward reproducing the20th Century Climate in Coupled Models (20C3M) and

Author contributions: T.M., M.I., and M.K. designed research; T.M. performed research;T.M., M.I., M.K., Y.C., M.W., T.N., T.T.S., H.S., T.A., N.S., T.T., S.Y., H.T., and M.M. contributednew reagents/analytic tools; T.M. analyzed data; T.M., M.I., M.K., Y.C., and M.W. wrotethe paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.1To whom correspondence should be addressed. E-mail: motizuki@jamstec.go.jp.

This article contains supporting information online at www.pnas.org/cgi/content/full/0906531107/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.0906531107 PNAS ∣ February 2, 2010 ∣ vol. 107 ∣ no. 5 ∣ 1833–1837

GEO

PHYS

ICS

Dow

nloa

ded

by g

uest

on

Nov

embe

r 20

, 202

0

during 2001–2030 with the A1B-type emissions scenario in theSpecial Report on Emissions Scenarios (22). We define these10-member ensembles as a reference field in the present paper(referred to as NoAS). These forcing data have been used in anumber of global warming studies (6, 10, 11) and we similarlyuse them as a boundary condition in all experiments in the pre-sent paper. Our initialization is done on the basis of a data assim-ilation approach together with an ensemble technique, usingMIROC and the objective analyses of sea surface temperature(SST), subsurface ocean temperature, and salinity (23–25). Usingthe ensemble snapshots of the assimilation data as initialconditions, we perform seven sets of 14.5-year-long, 10-memberensemble hindcast experiments every 5 years between July 1, 1960and December 31, 2004 (referred to as HCST). In a similar man-ner, we also perform an ensemble forecast experiment startingfrom July 1, 2005 up to December 31, 2030 (referred to as FCST).

ResultsWhen examining the global-mean SAT variations (Fig. 1), on de-cadal timescales, the observed values lie within 1 SD of the HCSTand/or NoAS ensembles throughout virtually all of the predictionperiods. The temporal variations of the globally averaged SATarelargely controlled by the external forcing prescribed in all experi-ments in the same manner. The linear trends during 1961–1990display a negligible difference between the observation(0.085 °C∕10 yr) and the NoAS data (0.080 °C∕10 yr). On theother hand, internal variations can contribute to the decadal-scale modulations such as the late-1970’s change in climate(26), which is defined here as the 5-year running mean during1979–1986 minus that during 1971–1978. The magnitude of thislate-1970’s change in the observations (0.149 °C) is substantiallylarger than that in the NoAS data (0.091 °C), when the SD of the10 ensembles (0.078 °C) is considered. In the NoAS experiment,because the individual ensemble member should arbitrarilyrepresent states of internal oscillations due to the absence ofan initialization based on observational data, realistic phasechanges of internal decadal variations are simulated by only afew ensemble members rather than an ensemble mean. It isnot easy to identify a significant improvement due to initializationin the temporal evolution of the global-mean SAT (Fig. 1),whereas some differences between the NoAS and HCST/FCSTdata are noticeable. For example, internal variations may workto make the globally averaged SAT in the FCST-derived ensemblemean slightly lower than that in the NoAS data (Fig. 1). In fact,such a slowing down of the globally averaged SATrise can be clos-

ely related to a significantly negative tendency of the predictedPDO phase as suggested below.

Our major interest is directed toward the predictability ofinternal decadal variations in the climate system, particularlythe PDO. To identify and isolate internal variations from a forcedvariation governed by the external forcing more effectively, weanalyze the HCST-derived and observed deviations from anexternally forced variation (hereafter called internal compo-nents). Here, we define the externally forced variation as a lead-ing mode obtained from a signal-to-noise maximizing empiricalorthogonal function (EOF) analysis (27, 28) using the NoAS dataduring 1958–2017. Figs. 2A and B indicate that the internaldecadal-scale variations of SST and vertically averaged oceantemperature over the upper 300 m (VAT300) are predictablealmost a decade in advance in specific areas. Around the KOE(particularly its northern part and downstream), subtropical

Fig. 1. Hindcasted, forecasted, and observed time series of globallyaveraged SAT anomaly. Globally averaged anomalies of the annual-meanSAT (°C) relative to the averages during 1961–1990. Red, blue, and green linesrepresent the observation (Hadley Centre-Climate Research Unit gridded sur-face temperature dataset, HadCRUT3v) and the ensemble-mean values of theHCST/FCST and NoAS data, respectively. Blue and green shades represent theranges of 1 SD of the 10 ensembles in the HCST/FCST and NoAS data, respec-tively. Plotted values for the HCST/FCST data are the results of hindcast andforecast experiments starting from 1960, 1975, 1990, and 2005.

Fig. 2. Skills in hindcasting internal decadal variations in terms of anomalycorrelation coefficients at each grid point and area-averaged errors. (A)Areas where the 5-year-mean SST deviations from the externally forced var-iation are significantly hindcasted at 90% confidence levels in terms of theanomaly correlation coefficients at forecast years 2, 4, 6, and 8, respectively.Significance tests are conducted on the basis of a bootstrap approach using2,000 sets of resampled ensemble-mean values (SI Text). (B) The same as (A),except for the VAT300 deviations. Two blue rectangles represent the KOE andSTF regions defined in the present paper. (C) RMSEs of the 5-year-meanVAT300 deviations projected onto the leading EOF (i.e., Fig. 3A) (broken lines)and those averaged over the KOE (solid lines) and STF (dashed-dotted lines)regions, respectively. Thick blue, green, and black lines represent the errors inthe ensemble-mean states of the HCST and NoAS data and the persistenceprediction (for the PDO time series only), respectively. Thin green lines repre-sent values indicating that the RMSEs in the HCST data are significantlysmaller than those in the NoAS data at 60%, 70%, and 80% confidence limits,respectively (SI Text). Red lines represent the observed SDs.

1834 ∣ www.pnas.org/cgi/doi/10.1073/pnas.0906531107 Mochizuki et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 20

, 202

0

oceanic frontal (STF) and eastern off-equatorial regions of theextratropical North Pacific (see two rectangles in Fig. 2B forthe KOE and STF regions), for example, the hindcastedVAT300 deviations are significantly correlated with the observa-tion for several years (Fig. 2B). For the ensemble-mean VAT300sin the KOE and STF regions, the rms errors (RMSEs) of theHCST data are significantly smaller than those of the NoAS dataduring about the initial 6 and 4 years (Fig. 2C), respectively, whilesignificance levels are limited to 70%. The SST in the midlatitudealso represents a similar predictive skill (Fig. 2A). As indicated bySmith et al. (2007) (10) and Keenlyside et al. (2008) (11), inaddition, the high latitudes of the North Atlantic exhibit longpredictability in both SST and VAT300.

These regional but significant predictive skills in the upper-ocean temperature suggest that the initialization enables us topredict the PDO on decadal timescales, because the KOE andSTF regions are the centers of action of the PDO (Fig. 3A)(29). The temporal evolution of the observed PDO, which is de-fined here by the projection onto the leading EOF of internalcomponents (i.e., Fig. 3A), represents a decadal variation suchas the late-1970’s change with interannual fluctuations (red linesin Fig. 3B). On decadal timescales (e.g., in 5-year-mean fields),the area-averaged VAT300 variations in both the KOE and STFregions are observed significantly correlating with the PDO time

series at 90% confidence levels. Overall, the hindcasted time se-ries of the PDO (i.e., the projection of the hindcasted VAT300deviations) are in fair agreement with the observations, particu-larly in an initial few years (Fig. 3B). Fluctuations of the NoAS-derived projections would represent noise probably due to smallsize of the ensembles. When compared to the NoAS experiments,the RMSEs of the projected VAT300 deviations suggests that thePDO phases are predictable almost 6 years in advance at 70%confidence levels (Fig. 2C). Errors in a persistence prediction(i.e., keeping the initial conditions unchanged) grow quite rapidlyto a level of the NoAS data (Fig. 2C). Over the coming decade,the projected VAT300 deviation of the FCST data indicates thatthe positive phase of the PDO on July 2005 rapidly shifts to asignificantly negative phase within several years (Fig. 3B, Top).The observational data for recent years (2005–2008) also displaythis rapid phase change of the PDO. Although the time seriesplotted in Fig. 3B contains both interannual and decadal compo-nents, this agreement suggests that the FCST data can represent apredictive skill at least during the initial 3 years, which is consis-tent with the skills statistically found in the HCST data.

DiscussionThe predictability of the PDO phase can contribute to that ofglobal climate changes on decadal timescales (e.g., global-meanSAT changes) as implied in Fig. 1. In fact, when examining theindividual ensemble members of the HCST data between July1975 and December 1989 together with the assimilation datafor the preceding initialization up to June 1975, a large global-mean SATrise is always simulated with a positive PDO tendencyin the 1980s. The global-mean SAT rising rate (see Fig. 1) andchanges in the PDO tendency (see Fig. 3B) exhibit a significantrelationship in the averages during 1979–1986 minus those during1971–1978; the correlation coefficient is 0.791. This suggests thatthe external forcing and the PDO interplay to realize the late-1970’s change in climate, and that the initialization has a greatimpact on the future prediction associated with global climatechange as well as the PDO. Fig. 3B suggests that the PDOmay also contribute to the decadal-scale climate changes inrecent years. The FCST data reveal that the internal decadal var-iations may suppress the globally averaged SATrise in the comingdecade (Fig. 1). The SAT deviations over the initial 6 years(Fig. 4A and B) indicate that the slowing down of global warmingin the FCST data is largely due to the tropical low SATcomprisinga PDO-like pattern (e.g., Fig. 3A), in agreement with the obser-vation over the initial 3 years (Fig. 4C). The rising trend in SAT isenhanced over east Asia and over the KOE region, while it is sup-pressed along the west coasts of North and South America andover the equatorial Pacific. The PDO-like patterns in the SATdifference shown in Fig. 4 Left are closely related to a significantlynegative phase of the predicted PDO after 2007 (Fig. 3B).

Keenlyside et al. (2008) (11) have suggested that global warm-ing, particularly over the North Atlantic region, may slow downdue to an internal change of the AMOC in the coming decade.Our results suggest that the PDO also plays a major role in mod-ulating the global warming trend on decadal timescales. Both thePDO and the AMOC have considerable impact on the pan-Pacific and pan-Atlantic climates and can influence a globallyaveraged state. Over the high latitudes of the North Atlanticand Europe, our FCST experiment simulates a high SAT consis-tent with recent observations (Fig. 4). Although our focus in thepresent paper is on the PDO, physical processes influencing theAtlantic climate should be examined in more detail. Develop-ment of the initialization system by advancing the assimilationand ensemble generation techniques and by assimilating addi-tional observational data may enable us to further improvethe estimates of initial conditions. In addition, improvement ofthe model components may raise the level of performancein simulating major climate processes. A higher-resolution

Fig. 3. Hindcasted, forecasted, and observed time series of the leadingmode of internal decadal variations in upper-ocean temperature over theNorth Pacific. (A) Spatial pattern of the leading EOF of the North PacificVAT300 deviations from the externally forced variation in the NoAS data dur-ing 1958–2017. (B) The projections onto the leading EOF [i.e., (A)] for annual-mean VAT300 deviations in the HCST/FCST (blue lines) and NoAS data (greenlines) and in the observation (broken red lines), respectively. Solid red linesare 5-year-runningmean values in the observation. Blue shades represent theranges of 1 SD of the 10 ensembles of the projected deviations in the HCST/FCST data. Plotted values are normalized using the SD of the observations.

Mochizuki et al. PNAS ∣ February 2, 2010 ∣ vol. 107 ∣ no. 5 ∣ 1835

GEO

PHYS

ICS

Dow

nloa

ded

by g

uest

on

Nov

embe

r 20

, 202

0

climate model should provide us with more detailed regionalinformation.

MethodsModel Description.We perform data assimilation and prediction experimentsusing the objective analyses and a coupled atmosphere-oceanmodel, MIROC.MIROC has been developed at the Center for Climate System Research ofthe University of Tokyo, National Institute for Environmental Studies, andFrontier Research Center for Global Change of the Japan Agency for Ma-rine–Earth Science and Technology. The resolution of the atmospheric com-ponent is horizontally the same as the commonly used T42 spectral modeland has 20 levels on a vertical σ-coordinate. The resolution of the oceancomponent is 1.4° in longitude and 0.56–1.4° in latitude (finer around theequator) and has 44 vertical levels.

Assimilation Procedure for Initialization. Objective analyses of ocean tempera-ture and salinity have been conducted beforehand on a monthly basis(23–25). We assimilate anomalies of the objective analyses relative to theaverages during 1961–1990, because correction of model climatology by dataassimilation sometimes leads to a climate drift that contaminates temporalevolutions of the internal variability of interest in predicted fields. By an in-cremental analysis update method (30), the model temperature and salinityin the upper 700 m depth are forced to approach the observations whichcontain the dominant signals of the PDO. We perform 10-member assimila-tion runs with different initial conditions on January 1, 1945 that are derivedfrom the 10-ensemble snapshots of the NoAS data.

Estimates of analysis errors are also compiled in the objective analysisdataset (24), which enable us to evaluate statistical reliability of the analysesat each grid point. Using the analysis errors, we control the magnitudes ofanalysis increments (i.e., forcing by the data assimilation) in such a way thatthese become relatively large at a grid point where an objective analysis isavailable with high reliability. A model-to-observation ratio of analysis errorsis also taken into consideration to determine the magnitudes of analysisincrements. In addition, we introduce a constraint that ensures no change

in global averages of both heat and salinity in the assimilation. When apply-ing this conservation constraint for global averages, the density of oceanwater is assumed to be invariant at each grid point during the assimilationinterval (i.e., 1 day). Impacts of sea-ice variations on a near-term climateprediction have not been assessed so far, and an inconsistency in sea-ice coverbetween the model and the observations usually results in unfavorabletemperature and salinity variations over the polar regions in data assimila-tion experiments. Within our present framework of data assimilation, theobjective analyses are not assimilated at a grid point where sea ice existsin the model. In addition, at a grid point located away from the modeledsea ice within a 10° distance in latitude, the analysis increment is dampedin proportion to the distance of the nearest sea ice. The data assimilationprocedure described here has allowed us to prevent the model climate fromunrealistic drift in subsequent prediction experiments.

Definition of the PDO.We apply an EOF analysis to the deviations of the NoAS-derived 10-ensemble members from the externally forced variation during1958–2017, which is defined using a signal-to-noise maximizing EOF techni-que (27, 28). The leading EOF should define the spatial pattern of the domi-nant internal variation that is not driven by the external forcing. As a whole,the leading EOF for the North Pacific (15°S-70°N, 15°E-90°W) SST, which is agood proxy for the SATover the ocean, represents major characteristics of thespatial pattern of the observed PDO in a reasonable manner. When projectedonto the leading EOF, the observed SST anomaly relative to the averages dur-ing 1961–1990 displays decadal variations similar to the so-called PDO index(13). The spatial pattern of the leading EOF for VAT300 is quite similar to thatfor SSTand is characterized by strong signals over the KOE and STF regions ofthe midlatitude North Pacific (29). These leading EOFs obtained from the en-semble simulations are in good agreement with the observed counterparts.

ACKNOWLEDGMENTS. The authors thank Dr. J. P. Matthews for his carefulreading and editing of the manuscript. This work is supported by the Japa-nese Ministry of Education, Culture, Sports, Science, and Technology, throughthe Innovative Program of Climate Change Projection for the 21st Century.

1. Cox PM, Stephenson D (2007) A changing climate for prediction. Science,317:207–208.

2. Hibbard KA, Meehl GA, Cox PM, Friedlingstein P (2007) A strategy for climate change

stabilization experiments. EOS, 88:217–219.

Fig. 4. Impacts of initialization on the predicted SAT in the coming decade. The FCST-derived ensemble-mean SAT deviations (°C) from the externally forcedvariation during (A) 2006–2008 and (B) 2009–2011, respectively. Shaded regions are the significant areas at 90% confidence levels (SI Text). (C) The same as (A),except for the observed deviation. Note that the observational dataset contains missing values, particularly over the high latitudes.

1836 ∣ www.pnas.org/cgi/doi/10.1073/pnas.0906531107 Mochizuki et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 20

, 202

0

3. Murphy JM, et al. (2004) Quantification of modelling uncertainties in a large ensembleof climate change simulations. Nature, 430:768–772.

4. Barnett DN, et al. (2006) Quantifying uncertainty in changes in extreme eventfrequency in response to doubled CO2 using a large ensemble of GCM simulations.Clim Dynam, 26:489–511.

5. Clark R, Brown SJ, Murphy JM (2006) Modelling Northern Hemisphere summer heatextreme changes and their uncertainties using a physics ensemble of climate sensitivityexperiments. J Climate, 19:4418–4435.

6. Shiogama H, Nozawa T, Emori S (2007) Robustness of climate change signals in nearterm predictions up to the year 2030: Changes in the frequency of temperatureextremes. Geophys Res Lett, 34:L12714 doi:10.1029/2007GL029318.

7. Stott PA, Kettleborough JA (2002) Origins and estimates of uncertainty in predictionsof twenty first century temperature rise. Nature, 416:723–726.

8. Meehl GA, et al. (2007) Climate Change 2007: The Scientific Basis. Contribution ofWorking Group I to the Fourth Assessment Report of the Intergovernmental Panelon Climate Change, eds JT Houghton et al. (Cambridge Univ Press, New York), pp525–582.

9. Collins M, Allen MR (2002) Assessing the relative roles of initial and boundary condi-tions in interannual to decadal climate predictability. J Climate, 15:3104–3109.

10. Smith DM, et al. (2007) Improved surface temperature prediction for the comingdecade from a global climate model. Science, 317:796–799.

11. Keenlyside NS, Latif M, Jungclaus J, Kornblueh L, Roeckner E (2008) Advancingdedadal-scale climate prediction in the North Atlantic sector. Nature, 453:84–88.

12. Pohlmann H, Jungclaus JH, Köhl A, Stammer D, Marotzke J (2009) Initializing decadalclimate predictions with the GECCO oceanic synthesis: Effects on the North Atlantic.J Climate, 22:3926–3938.

13. Mantua NJ, Hare SR, Zhang Y, Wallace JM, Francis RC (1997) A Pacific interdecadalclimate oscillation with impacts on salmon production. B Am Meteorol Soc,78:1069–1079.

14. Graham N (1994) Decadal-scale climate variability in the tropical and North Pacificduring the 1970s and 1980s observations and model results. Clim Dynam, 10:135–162.

15. Trenberth K, Hurrell JW (1994) Decadal atmosphere-ocean variations in the Pacific.Clim Dynam, 9:303–319.

16. Miller AJ, Schneider N (2000) Interdecadal climate regime dynamics in the North PacificOcean: Theories, observations and ecosystem impacts. Prog Oceanogr, 47:355–379.

17. Miller AJ, Chai F, Chiba S, Moisan JR, Neilson DJ (2004) Decadal-scale climate andecosystem interactions in the North Pacific ocean. J Oceanogr, 60:163–188.

18. Seager R, et al. (2004) Earth’s Climate: TheOcean-Atmosphere Interaction, eds CWang,SP Xie, and JA Carton (American Geophysical Union, Washington, DC), pp 105–120.

19. Stockdale TN, Anderson DLT, Alves JOS, Balmaseda MA (1998) Global seasonal rainfallforecasts using a coupled ocean-atmosphere model. Nature, 392:370–373.

20. Palmer TN, et al. (2004) Development of a European multimodel ensemble system forseasonal-to-interannual prediction (DEMETER). B Am Meteorol Soc, 85:853–872.

21. Nozawa T, Nagashima T, Shiogama H, Crooks SA (2005) Detecting natural influenceon surface air temperature change in the early twentieth century. Geophys Res Lett,32:L20719 doi:10.1029/2005GL023540.

22. Nakicenovic N, et al. (2000) Intergovernmental Panel on Climate Change: EmissionsScenarios. A Special Report of Working Group III of the Intergovernmental Panelon Climate Change, eds N Nakicenovic et al. (Cambridge Univ Press, New York) p 599.

23. Ishii M, Shouji A, Sugimoto S, Matsumoto T (2005) Objective analyses of SST andmarine meteorological variables for the 20th century using ICOADS and the KobeCollection. Int J Climatol, 25:865–879.

24. Ishii M, Kimoto M, Kachi M (2003) Historical ocean subsurface temperature analysiswith error estimates. Mon Weather Rev, 131:51–73.

25. Ishii M, Kimoto M, Sakamoto K, Iwasaki S (2006) Steric sea level changes estimatedfrom historical ocean subsurface temperature and salinity analyses. J Oceanogr,62:155–170.

26. Meehl GA, Hu A, Santer BD (2009) The mid-1970s climate shift in the Pacific and therelative roles of forced versus inherent decadal variability. J Climate 780–792.

27. Venzke S, Allen MR, Sutton RT, Rowell DP (1999) The atmospheric response over theNorth Atlantic to decadal changes in sea surface temperature. J Climate,12:2562–2584.

28. Ting M, Kushnir Y, Seager R, Li C (2009) Forced and internal twentieth-century SSTtrends in the North Atlantic. J Climate, 22:1469–1481.

29. Nakamura H, Lin G, Yamagata T (1997) Decadal climate variability in the North Pacificduring the recent decades. B Am Meteorol Soc, 78:2215–2225.

30. Bloom SC, Takacs L, da Silva AM, Ledvina D (1996) Data assimilation using IncrementalAnalysis Updates. Mon Weather Rev, 124:1256–1271.

Mochizuki et al. PNAS ∣ February 2, 2010 ∣ vol. 107 ∣ no. 5 ∣ 1837

GEO

PHYS

ICS

Dow

nloa

ded

by g

uest

on

Nov

embe

r 20

, 202

0