The Indian Ocean Dipole (IOD) Mode Direction

Argo Float Data – Indian Ocean

Data availability: Jan 2001 - till date A Legacy Active No. of Floats: 745 Total No. of Floats: 1982 of Indian Floats: 284

T and S Profiles: ~ 2,20,000 TOGA Data formats: ASCII, NetCDF, Plots

New In-sights Deployment of different types of floats Indian Ocean dipole and monsoon • Generation and termination of IOD mechanism • Summer cooling in the Arabian Sea • Impact of IOD on SSS revealed from Argo • Role of subsurface Temp anomalies in the SW Indian Ocean during 2006 dipole event • Monsoon interactions and drought over India in relation to Indian ocean sub-surface features Indian Ocean Observing System (IndOOS)

Multi-platform Multi-national Multi-institutional Long-term Ocean Observation Network

Model used : MOM 4 Global Ocean analysis (GFDL) Domain: Global Resolution: 50 km zonal and 25 km meridional, 40 vertical levels. Atmospheric forcing: Fluxes from Global Assimilation Forecast System (GFS)- T254L64 run at NCMRWF. Data assimilation scheme: 3D VAR -GODAS (NCEP) Parameters assimilated: Temperature and salinity profiles from Argo, XBT and moorings Relaxation: OISST-V2 [Reynolds, 2007] Outputs: Temperature, To provide an accurate estimate of the ocean state, which is used Salinity, SSH, and to initialize coupled model for the seasonal monsoon forecast and Currents to understand the variability of ocean in different time scales letters to nature resulting relaxation of the SSH ®elds would force eastward-propa- ...... gating and downwelling Kelvin waves which would deepen the eastern mixed layer and return the system to a normal con®gura- Adipolemodeinthe tion. These changes may be seen in the spring and early summer of tropical Indian Ocean 1998 in Fig. 1. The results presented above suggest that the Indian Ocean N. H. Saji*, B. N. Goswami², P. N. Vinayachandran* & T. Yamagata*³ exhibits strong coupled ocean±atmosphere±land interactions that are self-maintaining, and are capable of producing signi®cant * Institute for Global Change Research, SEAVANS N 7F, 1-2-1 Shibaura, perturbations to the annual cycle, at least during the 1997±1998 Minato-ku, Tokyo 105 6791, Japan period. Arguably, the evolution of the perturbation is independent ² Center for Atmospheric and Oceanic Sciences, Indian Institute of Science, of ENSO. We note that other hypotheses3,4 have been suggested for Bangalore 560 012, India observed interannual variability in the Indian Ocean. These theories ³ Department of Earth and Planetary Physics, Graduate School of Science, rely on the Indian Ocean responding locally to either weaker or The University of Tokyo, Tokyo 113 0033, Japan stronger monsoon winds which, through changes in upwelling and ...... mixing, introduce a biennial component to the system. But the For the tropical Paci®c and Atlantic oceans, internal modes of weakness of these theories is the maintenance of the upper-ocean variability that lead to climatic oscillations have been recog- anomalies from year to year. Our hypothesis adds a coupled dynamical nized1,2, but in the Indian Ocean region a similar ocean±atmos- component that has a longer timescale than the thermodynamics of phere interaction causing interannual climate variability has not the mixed layer, and which may form a link from one monsoon yet been found3. Here we report an analysis of observational data season to the next. Thus we suggest that the Indian Ocean may not over the past 40 years, showing a dipole mode in the Indian Ocean: be a passive player in climate variability on seasonal to interannual a pattern of internal variability with anomalously low sea surface timescales, but may enact a very active and independent role. M temperatures off Sumatra and high sea surface temperatures in the western Indian Ocean, with accompanying wind and precipi- Received 6 July 1998; accepted 9 July 1999. letters to nature tation anomalies. The spatio-temporal links between sea surface 1. Nicholls, N. Air-sea interaction and the quasi-biennial oscillation. Mon. Weath. Rev. 106, 1505±1508 temperatures and winds reveal a strong coupling through the (1983)...... resulting relaxation of the SSH ®elds would force eastward-propa-precipitation ®eld and ocean dynamics. This air±sea interaction 2. Nicholls, N. All-India summer monsoon rainfallgating and sea and surface downwelling temperature Kelvin around waves northern which would deepen the Adipolemodeinthe Australia and Indonesia. J. Clim. 8, 1463±1467eastern (1995). mixed layer and return the system to a normalprocess con®gura- is unique and inherent in the Indian Ocean, and is shown tion. These changes may be seen in the spring and early summer of 3. Meehl, G. A. Coupled ocean-atmosphere-land processes and south Asian monsoon variability. Science tropical Indian Ocean 1998 in Fig. 1. to be independent of the El NinÄo/Southern Oscillation. The 265, 263±267 (1994). The results presented above suggest that the Indian Ocean N. H. Saji*, B. N. Goswami², P. N. Vinayachandran* & T. Yamagata*³ 4. Meehl, G. A. The south Asian monsoon and theexhibits tropospheric strong biennial coupled oscillation. ocean±atmosphere±landJ. Clim. 10, 1921± interactionsdiscovery that of this dipole mode that accounts for about 12% of are self-maintaining, and are capable of producing signi®cant 1943 (1997). the sea surface* Institute temperature for Global Change variability Research, SEAVANS in N the 7F, 1-2-1 Indian Shibaura, OceanÐand, perturbations to the annual cycle, at least during the 1997±1998 Minato-ku, Tokyo 105 6791, Japan 5. Webster, P. J. & Palmer, T. N. The past and futureperiod. of El Arguably, NinÄo. Nature the390, evolution562±564 of (1997). the perturbation is independent ² Center for Atmospheric and Oceanic Sciences, Indian Institute of Science, 3,4 in its active years, also causes severe rainfall in eastern Africa and 6. Reynolds, R. & Marisco, D. An improved real-timeof ENSO. global sea We surface note that temperature other hypotheses analysis. J. Clim.have6, been suggested for Bangalore 560 012, India observed interannual variability in the Indian Ocean. Thesedroughts theories in³ Department IndonesiaÐbrightens of Earth and Planetary Physics, the Graduate prospects School of Science, for a long-term 114±119 (1993). rely on the Indian Ocean responding locally to either weaker or The University of Tokyo, Tokyo 113 0033, Japan 7. Kalnay, E. et al. The NCEP/NCAR 40-year reanalysisstronger project. monsoonBull. Am. winds Meteorol. which, Soc. through77, 437±471 changes in upwellingforecast and of...... rainfall anomalies in the affected countries...... mixing, introduce a biennial component to the system. But the For the tropical Paci®c and Atlantic oceans, internal modes of (1996). weakness of these theories is the maintenance of the upper-oceanThe catastrophicvariability that rains lead of to climatic 1961 in oscillations tropical have eastern been recog- Africa and 8. Hendricks, J. R., Leben, R. R., Born, G. H. & Koblinsky,anomalies C. from J. Empirical year to year. orthogonal Our hypothesis function analysis adds a coupledsubsequent dynamical nized abrupt1,2, but discharge in the Indian of Ocean the White region a Nile similar are ocean±atmos- now known4±6 to of global TOPEX/POSEIDON altimeter data andcomponent implications that for has detection a longer of timescale global sea than level the rise. thermodynamics of phere interaction causing interannual climate variability has not the mixed layer, and which may form a link from onebe monsoon part ofyet an been anomalous found3. Here we climate report an state analysis over of observational the tropical data Indian J. Geophys. Res. 101, 14131±14145 (1996). season to the next. Thus we suggest that the Indian Ocean may not over the past 40 years, showing a dipole mode in the Indian Ocean: 9. Yu, L. & Rienecker, M. M. Mechanisms for the Indianbe a passive Ocean warmingplayer in during climate the variability 1997±1998 on El seasonal NinÄo. toOcean. interannual A dipolea pattern structure of internal characterized variability with anomalously the sea surface low sea surface temperature Geophys. Res. Lett. 26, 735±738 (1999). timescales, but may enact a very active and independent role. M temperatures off Sumatra and high sea surface temperatures in (SST) anomalythe western during Indian this Ocean, event: with warmer accompanying than wind usual and precipi- SSTs occurred Received 6 July 1998; accepted 9 July 1999. 10. Barnett, T. P. Interaction of the monsoon and Paci®c Ocean trade wind systems at interannual time over large partstation anomalies. of the western The spatio-temporal basin, while links between SSTs off sea surfaceSumatra were 1. Nicholls, N. Air-sea interaction and the quasi-biennial oscillation. Mon. Weath. Rev. 106, 1505±1508 temperatures and winds reveal a strong coupling through the scales. Part I: the equatorial zone. Mon. Weath. Rev.(1983).111, 756±773 (1983). precipitation ®eld and ocean dynamics. This air±sea interaction 11. Arkin, P. & Meisner, B. The relationship between2. Nicholls,large-scale N. All-India convective summer monsoon rainfall rainfall and cloud and sea surface cover temperature over aroundcooler northern than usual. Rainfall increased over tropical eastern Africa and Australia and Indonesia. J. Clim. 8, 1463±1467 (1995). process is unique and inherent in the Indian Ocean, and is shown the western hemisphere during 1982±1984. Mon.3. Meehl, Weath. G. A. Rev. Coupled115, ocean-atmosphere-land51±74 (1987). processes and south Asian monsoonthe variability. westernScience to Indian be independent Ocean, of while the El over NinÄo/Southern the Indonesian Oscillation. archipelagoThe 265, 263±267 (1994). discovery of this dipole mode that accounts for about 12% of 12. Webster, P. J. Response of the tropical atmosphere4. Meehl, to local G. A. steady The south forcing. Asian monsoonMon. and Weath. the tropospheric Rev. 100, biennial518± oscillation. J. Clim. 10, 1921± 1943 (1997). it decreased,the resulting sea surface temperature in severe variability drought. in the Equatorial Indian OceanÐand, surface winds, 541 (1972). 5. Webster, P. J. & Palmer, T. N. The past and future of El NinÄo. Nature 390, 562±564which (1997). in a normalin its active summer years, also season causes severe blow rainfall towards in eastern the Africa east, and weakened 6. Reynolds, R. & Marisco, D. An improved real-time global sea surface temperature analysis. J. Clim. 6, droughts in IndonesiaÐbrightens the prospects for a long-term 13. Gill, A. E. Some simple solutions for heat-induced114±119 tropical (1993). circulation. Q. J. R. Meteorol. Soc. 106, forecast of rainfall anomalies in the affected countries. 447±462 (1981). 7. Kalnay, E. et al. The NCEP/NCAR 40-year reanalysis project. Bull. Am. Meteorol.and Soc. 77, reversed437±471The Indian Ocean Dipole (IOD) Mode direction. There was no El NinÄo in the Paci®c, while (1996). The catastrophic rains of 1961 in tropical eastern Africa and 14. Near Real-Time Analysis of the Ocean and Atmosphere8. Hendricks,Fig J. T R., 29, Leben, p. 36 R. R., (Climate Born, G. H. Diagnostics & Koblinsky, C. J.Bull. Empirical NO. orthogonalIndia function analysis experiencedDMI = subsequentΔSST between Tropical W. Indian Ocean theabrupt highest discharge of summer the White Nile monsoon are now known rainfall4±6 to in the of global TOPEX/POSEIDON altimeter data and implications for detection of global sea level rise. be part of an anomalous climate state over the tropical Indian 97/11, Climate Diagnostics Center, National CenterJ. Geophys. of Environmental Res. 101, 14131±14145 Prediction, (1996). NOAA, Washing- past 150 yearsOcean. (ref. A dipole 5). structure By examining characterized thelong-term sea surface temperature data sets of SST ton DC, 1997). 9. Yu, L. & Rienecker, M. M. Mechanisms for the Indian Ocean warming during the 1997±1998 El NinÄ(50-70E, 10S-10N)) and E. Indian Ocean (90-110E, o. Geophys. Res. Lett. 26, 735±738 (1999). (SST) anomaly during this event: warmer than usual SSTs occurred 15. Gill, A. Atmosphere-Ocean Dynamics (Academic,10. Barnett, London, T. P. 1982).Interaction of the monsoon and Paci®c Ocean trade wind systems at interannual time over large parts of the western basin, while SSTs off Sumatra were scales. Part I: the equatorial zone. Mon. Weath. Rev. 111, 756±773 (1983). 20S-0) cooler than usual. Rainfall increased over tropical eastern Africa and 16. Webster, P. J. et al. Monsoons: processes, predictability11. Arkin, P. and & Meisner, the prospectsB. The relationship for prediction.between large-scaleJ. Geophys. convective rainfall and cloud cover over the western hemisphere during 1982±1984. Mon. Weath. Rev. 115, 51±74 (1987). the western Indian Ocean, while over the Indonesian archipelago Res. 103, 14451±14510 (1998). 12. Webster, P. J. Response of the tropical atmosphere to local steady forcing. Mon. Weath. Rev. 100, 518± it decreased, resulting in severe drought. Equatorial surface winds, 17. Nicholson, S. E. & Kim, J. The relationship of the El-Nin541 (1972).Äo Southern Oscillation to African rainfall. Int. 4 which in a normal summer season blow towards the east, weakened 13. Gill, A. E. Some simple solutions for heat-induced tropical circulation. Q. J. R. Meteorol. Soc. 106, DMI Nino3 447±462 (1981). and reversed direction. There was no El NinÄo in the Paci®c, while J. Climatol. 17, 117±135 (1997). σDMI=0.3°C σNino=0.8°C 14. Near Real-Time Analysis of the Ocean and Atmosphere Fig T 29, p. 36 (Climate Diagnostics Bull. NO. India experienced the highest summer monsoon rainfall in the 18. Nicholson, S. E. An analysis of the ENSO signal in97/11, the Climate tropical Diagnostics Atlantic Center, and National western Center Indian of Environmental Oceans. Prediction, NOAA, Washing-2 past 150 years (ref. 5). By examining long-term data sets of SST ton DC, 1997). Int. J. Climatol. 17, 345±375 (1997). 15. Gill, A. Atmosphere-Ocean Dynamics (Academic, London, 1982). 16. Webster, P. J. et al. Monsoons: processes, predictability and the prospects for prediction. J. Geophys. 19. Reverdin, G., Cadet, D. & Gutzler, D. Interannual displacementsRes. 103, 14451±14510 of (1998). convection and surface circulation 0 17. Nicholson, S. E. & Kim, J. The relationship of the El-NinÄo Southern Oscillation to African rainfall. Int. 4 over the equatorial Indian Ocean. Q. J. R. Meteorol. Soc. 122, 43±67 (1986). DMI Nino3 J. Climatol. 17, 117±135 (1997). σDMI=0.3°C σNino=0.8°C 20. Kapala, A., Born, K. & Flohn, H. in Proc. Int. Conf.18. Nicholson, on Monsoon S. E. An Variability analysis of the and ENSO Prediction signal in the(ed. tropical Newson, Atlantic and western Indian Oceans. 2 Int. J. Climatol. 17, 345±375 (1997). –2 R.) 119±126 (Tech. Doc. 619, World Meteorological Organization, Geneva, Switzerland, 1994). 19. Reverdin, G., Cadet, D. & Gutzler, D. Interannual displacements of convection and surface circulation Ueq0 over the equatorial Indian Ocean. Q. J. R. Meteorol. Soc. 122, 43±67 (1986). Normalized anomaly 21. Rao, K. G. & Goswami, B. N. Interannual variations of the sea-surface temperature over the Arabian σUeq=0.96 ms–1 20. Kapala, A., Born, K. & Flohn, H. in Proc. Int. Conf. on Monsoon Variability and Prediction (ed.– Newson, 4 –2 R.) 119±126 (Tech. Doc. 619, World Meteorological Organization, Geneva, Switzerland, 1994). Ueq

Sea and the Indian Monsoon: A new perspective. Mon. Weath. Rev. 116, 558±568 (1988). Normalized anomaly 21. Rao, K. G. & Goswami, B. N. Interannual variations of the sea-surface temperature over the Arabian σUeq=0.96 ms–1 1960– 19654 1970 1975 1980 1985 1990 1995 Sea and the Indian Monsoon: A new perspective. Mon. Weath. Rev. 116, 558±568 (1988). 22. Shukla, J. & Mooley, D. A. Empirical prediction of the summer monsoon over India. Mon. Weath. Rev. 1960 1965 1970 1975 1980 1985 1990 1995 22. Shukla, J. & Mooley, D. A. Empirical prediction of the summer monsoon over India. Mon. Weath. Rev. Time (years) 115, 695±703 (1987). 115, 695±703 (1987). Time (years)

Supplementary information is available on Nature's World-Wide Web site (http:// Figure 1 Dipole mode and El NinÄo events since 1958. Plotted in blue, the dipole mode www.nature.com) or as hard copy from the London editorial of®ce of Nature. Supplementary information is available on Nature's World-Wide Web site (http:// Figure 1 Dipoleindex mode (DMI) exhibits and El a pattern NinÄo of events evolution since distinctly 1958. different Plotted from that of in the blue, El NinÄo, the which dipole mode www.nature.com) or as hard copy from the London editorial of®ce of Nature. is represented by the Nino3 sea surface temperature (SST) anomalies (black line). On the Acknowledgements index (DMI) exhibitsother hand, a pattern equatorial of zonal evolution wind anomalies distinctlyUeq (plotted different in red) coevolves from that with of the the DMI. El All NinÄo, which This work was supported by the Of®ce of Global Programs, NOAA, and the NSF (P.J.W., the three time series have been normalized by their respective standard deviations. We A.M.M., J.P.L.), and by NASA (R.R.L.). is represented byhave the removed Nino3 variability sea surface with periods temperature of 7 years or longer, (SST) based anomalies on harmonic (black analysis, line). On the from all the data sets used in this analysis. In addition, we have smoothed the time series Acknowledgements Correspondence and requests for materials should be addressed to P.J.W.other hand, equatorial zonal wind anomalies Ueq (plotted in red) coevolves with the DMI. All (e-mail: [email protected]). using a 5-month running mean. This work was supported by the Of®ce of Global Programs, NOAA, and the NSF (P.J.W., the three time series have been normalized by their respective standardSaji et al., Nature, 1999 deviations. We © 1999 Macmillan Magazines Ltd A.M.M., J.P.L.), and by NASA (R.R.L.). 360 have removed variability with periodsNATURE of 7| VOL years 401 | or23 SEPTEMBER longer, based 1999 | www.nature.com on harmonic analysis, Correspondence and requests for materials should be addressed to P.J.W. from all the data sets used in this analysis. In addition, we have smoothed the time series (e-mail: [email protected]). using a 5-month running mean.

and surface winds we found several occurrences of these patterns. evolution of a dipole mode event using composite analysis. We These observations motivated this study. However, here we use the analysed the recent six extreme events of 1961, 1967, 1972, 1982, GISST2.3b data set7 (1958±98), the NCEP surface winds reanalysis8 1994 and 1997 to present the life cycle of a typical dipole mode event (1958±98) and the Xie±Arkin rainfall analyses9 (1979±98) to in Fig. 2. Cool SST anomalies ®rst appear in the vicinity of the present our ®ndings, as the essential signatures of this phenomenon Lombok strait by May±June, accompanied by moderate south- are unchanged in the various data sets we have examined. easterly wind anomalies in the southeastern tropical Indian Ocean. Basin-scale anomalies of uniform polarity cover3 the tropical In the following months, the cold anomalies intensify and appear to Indian Ocean basin during El NinÄo/Southern Oscillation (ENSO) migrate towards the Equator along the Indonesian coastline, while events. Using empirical orthogonal function (EOF) analysis we see the western tropical Indian Ocean begins to warmletters up. Zonal wind to nature that this pattern (EOF1) explains about 30% of the total variation of anomalies along the Equator and alongshore wind anomalies off and surface winds weanomalous found several Indian Ocean occurrences SSTs. Next, of the these dipole patterns. mode (EOF2) evolutionSumatra intensify of a together dipole with mode the SST event dipole. using A dramatically composite analysis. We These observations motivatedexplains about this 12%. study. Characteristic However, to the here dipole we mode use is a the reversal analysedrapid peaking the of recent these features six extreme occurs in October, events following of 1961, by a 1967, 1972, 1982, GISST2.3b data set7 (1958±98),in sign of SST the anomaly NCEP across surface the basin. winds This reversal reanalysis is so striking,8 1994rapid and demise. 1997 The compositeto present time the series life shown cycle in Fig. of a 3 typicalillustrates dipole mode event 9 (1958±98) and the Xie±Arkinthat the dipole mode rainfall may be analysesidenti®ed by a simple(1979±98) index time to series inthis Fig. sudden 2. Coolchange of SST the anomalies DMI and Ueq more ®rst clearly. appear A strong in the vicinity of the present our ®ndings, aswhich the describes essential the signatures difference in SST of anomaly this phenomenon between the tropical Lombokfeature in the strait time series by May±June, is the tight coupling accompanied between the zonal by moderate south- are unchanged in the various data sets we have examined. easterly wind anomalies in the southeastern tropical Indian Ocean. Basin-scale anomalieswestern of Indian uniform Ocean (50polarity8 E±708 E, cover 108 S±103 8theN) and tropical the tropical Inwind the anomaly following and the months, dipole intensity. the cold A biennial anomalies tendencyÐin intensify and appear to Indian Ocean basin duringsouth-eastern El Nin IndianÄo/Southern Ocean (908 E±110 Oscillation8 E, 108 S±Equator). (ENSO) The migratethis case a towards tendency for the a dipole Equator mode event along to be the preceded Indonesian by an coastline, while events. Using empiricalstrong orthogonal correlation function (.0.7) between (EOF) this index, analysis referred we to see as the theevent western of the opposite tropical polarityÐis Indian evident. Ocean Note begins that the to biennial warm up. Zonal wind 10 that this pattern (EOF1)dipole explains mode index about (DMI), 30% and theof timethe seriestotal associated variation with of EOF2 anomaliestendency of U alongeq is already the well Equator known . Figure and 3 also alongshore indicates that wind a anomalies off anomalous Indian Oceanindicates SSTs. the accuracy Next, of the the DMI dipole in representing mode the (EOF2) dipole mode Sumatrapile-up of intensify warm water together off the Sumatran with coast, the together SST dipole. with A dramatically explains about 12%. Characteristic to the dipole mode is a reversal rapid peaking of these features occurs in October, following by a in sign of SST anomalyin across SST. the basin. This reversal is so striking, rapidwesterly demise. wind anomalies The composite over the central-eastern time series shownequatorial in Fig. 3 illustrates 11,12 that the dipole mode mayThe be dipole identi®ed mode event by is a independent simple index of the ENSO time in series the Paci®c thisIndian sudden Ocean change, signals the of occurrence the DMI of a dipole and modeUeq eventmore in clearly. A strong which describes the differenceOcean. Todemonstrate in SST anomaly this, we plot between SSTanomalies the representative tropical of featurethe following in the year. time series is the tight coupling between the zonal western Indian Oceanthe (50 central8 E±70 and8 easternE, 10 equatorial8 S±108 Paci®cN) and (from the the so-called tropical Nino3 windBasin-wide anomaly data for and rainfall the is availabledipole only intensity. since 1979. A Analyses biennial tendencyÐin south-eastern Indianregion) Ocean against (90 the8 E±110 DMI time8 E, series 108 inS±Equator). Fig. 1. Note the signi®cant The thisof the case rainfall a tendency data (or proxy for data a sets dipole such as mode outgoing event longwave to be preceded by an strong correlation (.0.7) between this index, referred to as the event of the opposite polarityÐis evident. Note that the biennial dipole mode events of 1961, 1967 and 1994 coinciding with no radiation) reveal that during a dipole mode event rainfall10 decreases dipole mode index (DMI), and the time series associated with EOF2 tendency of Ueq is already well known . Figure 3 also indicates that a indicates the accuracyENSO, of the a La DMI NinÄa and in representinga weak El NinÄo, respectively. the dipole There mode are years in pile-upover the oceanic of warm tropical water convergence off zone the (OTCZ) Sumatran and increases coast, together with in SST. which dipole mode events coincide with strong ENSO events as in westerlyover the western wind tropical anomalies Indian Ocean over (Fig. 4). the This central-eastern pattern of equatorial The dipole mode event1972 is or independent 1997. We note here of that the the ENSO correlation in the between Paci®c the DMI Indianrainfall is Ocean dynamically11,12, consistent signals with the the occurrence convergence/divergence of a dipole mode event in Ocean. Todemonstrateand this, Nino3 we SST plot anomaly SSTanomalies time series is weak representative (,0.35). of theshifts following associated with year. the wind ®eld in the composite maps. Thus the central and eastern equatorial Paci®c (from the so-called Nino3 Basin-wide data for rainfall is available only since 1979. Analyses region) against the DMIDuring time dipole series mode in Fig. events, 1. the Note surface the wind signi®cant ®eld over the ofwhat the emerges rainfall here data is strong (or proxy empirical data evidence sets for such coupling as outgoing longwave dipole mode events oftropical 1961, Indian 1967 Ocean and experiences 1994 coinciding large changes, especially with no in its radiation)between the reveal SST and that the wind during ®eld a through dipole the mode precipitation event rainfall decreases ENSO, a La NinÄa and azonal weak (east-to-west) El NinÄo, respectively. component over There the Equator. are years Maximum in over®eld. the Based oceanic on knowledge tropical about the convergence climate system zonein the Indian (OTCZ) and increases which dipole mode eventschanges coincide in the zonal with wind strong occur over ENSO the equatorial events central as in and overOcean the and western from our own tropical studies Indian with a variety Ocean of modelling (Fig. 4). This pattern of 1972 or 1997. We noteeastern here Indian that Ocean, the correlation where we noted between a correlation the maximum DMI of rainfalldevices, we is presentdynamically the following consistent model for withhow these the ®elds convergence/divergence are and Nino3 SST anomaly time: series is weak (,0.35). shifts associated with the wind ®eld in the composite maps. Thus During dipole mode. j0 events,6j with the the DMI. surface By plotting wind the area-averaged ®eld over equatorial the whatconnected emerges to each other. here is strong empirical evidence for coupling tropical Indian Oceanzonal experiences wind anomalies large (Ueq changes,) over this region especially (708 E±90 in8 E, its 58 S± betweenIn a normal the year SST southeast and trade the winds wind converge ®eld into through the South the precipitation zonal (east-to-west)5 component8 N), in Fig. 1, we over demonstrate the that Equator. the intensity Maximum of the SST dipole ®eld.Equatorial Based Trough on13 associated knowledge with the about high-rainfall the climate (.10 mm d system-1) in the Indian changes in the zonal windmode and occur the strength over of the the zonal equatorial wind anomaly central over the and Equator OceanOTCZ. When, and during from a dipole our mode own event, studies SSTs off with Sumatra a begin variety of modelling eastern Indian Ocean,are where strongly we dependent noted on a each correlation other. maximum of devices,cooling, convection we present weakens the at following the OTCZ and model the consequent for how these ®elds are . j0:6j with the DMI. By plotting the area-averaged equatorial connected to each other. Seasonal phase locking is an important characteristic of the DMI surface pressure modi®cation (not shown) makes the southeast zonal wind anomalies (Ueq) over this region (708 E±908 E, 58 S± In a normal year southeast trade winds converge into the South 58 N), in Fig. 1, we demonstratetime series. Thus that signi®cant the intensity anomalies of appear the around SST dipole June, inten- Equatorialtrade winds extend Trough and13 convergeassociated further downstream. with the high-rainfall This altered (.10 mm d-1) mode and the strengthsify of in the the zonal following wind months anomaly and peak inover October. the BecauseEquator of this OTCZ.large-scale When, wind ®eld during enhances a dipole convergence mode and moisture event, supplySSTs off Sumatra begin are strongly dependentsystematic on each seasonality other. of phase it is meaningful to demonstrate the cooling,at the extended convection downstream weakens end of the at trade the winds, OTCZ thereby and the consequent Seasonal phase locking is an important characteristic of the DMI surface pressure modi®cation (not shown) makes the southeast time series. Thus signi®cant anomalies appear around June, inten- trade winds extend and converge further downstream. This altered sify in the following months and peak in October. Because of this large-scale wind ®eld enhances convergence and moisture supply systematic seasonality of phase it is meaningful to demonstrate the at the extended downstream end of the trade winds, thereby a May-Jun b Jul-Aug

a May-Jun b Jul-Aug 5 ms–1 5 ms–1

5 ms–1 5 ms–1

c Sep-Oct d Nov-Dec c Sep-Oct d Nov-Dec

5 ms–1 5 ms–1 5 ms–1 5 ms–1

Figure 2 A composite dipole mode event. a±d, Evolution of composite SST and surface analysed anomalies were estimated by the two-tailed t-test. Anomalies of SSTs and winds A composite dipole mode event. ± , Evolution of composite SST and surface analysed anomalies were estimated by the two-tailed t-test. Anomalies of SSTs and winds wind anomalies from May±JuneFigure (a 2) to Nov±Dec (d). The statisticala d signi®cance of the exceeding 90% signi®cance are indicated by shading and bold arrows, respectively. wind anomalies from May±June (a) to Nov±Dec (d). The statistical signi®cance of the exceeding 90% signi®cance are indicated by shading and bold arrows, respectively. © 1999 Macmillan Magazines Ltd NATURE | VOL 401 | 23 SEPTEMBER 1999 | www.nature.com 361 © 1999 Macmillan Magazines Ltd NATURE | VOL 401 | 23 SEPTEMBER 1999 | www.nature.com 361 letters to nature

encouraging precipitation northwest of the normal position of the in the Paci®c and Atlantic oceans24, and presumably would have OTCZ. These abnormally extended trade winds also interrupt the attained a similar state had it not been for the rapid demise of the normal heat supply to the coast off Sumatra. In a normal summer instability after the boreal autumn. Our preliminary understanding monsoon season, the westerly winds along the Equator accumulate indicates that it is the changes in the state of the climate system, warmer water along this coast through downwelling equatorial and brought about by the seasonal monsoonal reversals, that are res- coastally trapped Kelvin waves11,12,14. The strongest manifestation of ponsible for the demise of the dipole mode event. The main factor this process, occurring twice a year during monsoon transitions, in¯uencing the instability is the cooler-than-normal SST off Sumatra. is known as the Yoshida±Wyrtki Jet11. This process counters the Because the mass and heat transport from the west, achieved cooling tendencies by evaporation, coastal upwelling and oceanic through equatorial ocean dynamics, play an important role in the heat advection brought about by the strong alongshore winds off heat balance of this region, its ¯uctuations affect the SST. However, this coast. In a year with a dipole mode event, abnormally extended after the boreal autumn, until the next spring, weakening winds trade winds with an easterly component along the Equator, by both along the Equator and along the coast diminish the importance of preventing the intrusion of the equatorial current15,16, allow the ocean dynamics in the regulation of SST. Consequently, ¯uctuations cooling processes to dominate off Indonesia. This cooling is further of the mass and heat transport caused by equatorial dynamics play a enhanced as entrainment processes associated with the coastal lesser role in SST ¯uctuations. On the other hand, the increased winds become more effective due to the shallower thermocline17,18. insolation (as a result of the seasonal movement of the Sun) and The shallowing thermocline is manifested in the observed lowering reduced evaporation (due to decreasing windspeeds) dominate the of the sea level16±20. Hydrographic observations in this region also changes in SST in this region during this season. In this context, it is favour the above model of the coupling between surface winds and likely that the higher-than-normal insolation, due to the reduction SST through ocean dynamics21. The warming in the west is initiated or absence of cloudiness in the tropical southeastern Indian Ocean, as a consequence of this chain of cause and effect taking place in the acting on the thin mixed layer, returns the system back to normality eastern half. The shifting trade winds, by increasing convergence by removing the cool SST anomaly. in the west, lead to reduced wind speed and reduced evaporation18,20 As the dipole mode is strongly dependent on the state of the which aid in warming up the SST. Entrainment is inhibited as system set up by the monsoonal circulation, it is expected that increased rainfall raises the stability of surface waters through the variability of the monsoons would signi®cantly affect this mode. reduced salinity18. The thermocline anomalously deepens owing We note that the dipole mode shares certain common features, such to reduced eastward transport11,22 resulting from the altered trade as the biennial tendency25, with the monsoonal variability. Also, winds. The warmer SST increases the precipitation anomaly and easterly anomalies along the Equator as well as reduced convection consequently the wind anomaly to its east, thus introducing a in the OTCZ are well-known features of strong monsoons26. Never- positive feedback mechanism. The see-saw of the thermocline theless, the statistical correlation of the DMI with precipitation over implied in the above discussion of the dipole mode mechanism is the Asian monsoon regime does not yield a signi®cant relationship supported by the see-saw in both sea level16±20 and annual mean (Fig. 4). Therefore the relationship of the DMI to the Indian subsurface temperature anomalies23 (not shown). monsoon variability is still not clear. It is clear, however, that the Here we have described a chain of events, which once initiated dipole mode has important implications for climate variability in keeps the SST cooler in the tropical southeastern Indian Ocean, other regions surrounding the Indian Ocean. The shift of the warmer in the tropical western Indian Ocean, and the southeast convergence zone, which leads to ¯oods in east Africa and drought trade winds in the eastern half stronger than normal throughout the in Indonesia (Fig. 4), manifest the weakening or reversal of the zonal boreal summer and autumn. At this stage, the ocean±atmosphere (Walker) circulation across the Indian Ocean. Besides causing this system in the Indian Ocean approaches the con®guration of those very discernible local climate variability across the basin, the zonal

3.5 Figure 3 Strong coupling of SST dipole intensity to Ueq. Shown is the coevolution of 3.0 Ueq –1 intensity of the dipole mode (DMI, black bars) and strength of equatorial zonal winds 2.5 σUeq=0.96 ms Ueq 2.0 anomalies (Ueq, grey bars) from the year before, to the year following a typical dipole mode σUeq=0.96 ms–1 1.5 event. Bars indicating signi®cant anomalies (estimated by a two-tailed t-test) exceeding a 1.0 0.5 90% con®dence level are marked with spots. 0.0 -0.5 -1.0 -1.5 -2.0 Normalized anomaly -2.5 The year before During a DM event After a DM event -3.0 a DM event -3.5 -4.0 Jan Apr Jul Oct Jan Apr Jul Oct Jan Apr Jul Oct

Figure 4 Rainfall shifts northwest of the OTCZ during dipole mode events. The map correlates the DMI and rainfall to illustrate these shifts. The areas within the white curve exceed the 90% level of con®dence for non-zero correlation (using a two-tailed t-test).

© 1999 Macmillan Magazines Ltd IOD affects rainfall over North and East Africa, the Maritime Continent, 362 NATURE | VOL 401 | 23 SEPTEMBER 1999 | www.nature.com East Asia, and other regions of the tropics via the Walker Circulation teleconnections

Climate model simulations suggested Sahel drying trend may be attributable to anthropogenic forcing (GHG and aerosols)

Fig. 1. Observed (CRU) 5-year running mean (July–August–September) Sahel rainfall, normalized by its mean value over 1901–2000 (black line), historical CM2 Fig. 1. Observed (CRU)ensemble 5-year mean running normalized mean so that (July–August–September)its mean value is unity over the same Sahel time interval rainfall, (thick normalized light blue line), byand its the mean historical value realization over that 1901–2000 most resembles (black line), historical CM2 ensemble mean normalizedthe observations so that in the its period mean 1950–2000 value is (thick unity redover line). The the gray same areatime represents intervalϮ1 standard (thick deviation light blue within line), the ensemble. and the The historical future scenarios realization are B1 that most resembles the observations in the(green), period A1B (blue), 1950–2000 and A2 (red). (thick Therered are two line). linesThe for each gray scenario, area one represents from CM2.0 andϮ1 another standard from deviationCM2.1. within the ensemble. The future scenarios are B1 (green), A1B (blue), and A2 (red). There are two lines for each scenario, one from CM2.0 and another fromHeld et al. PnAS, 2005 CM2.1. anthropogenic aerosols on clouds as well as the direct effect of Three integrations of CM2.0 and five of CM2.1 for the period anthropogenic aerosolsthe aerosols on on clouds the scattering as well and absorptionas the direct of radiation. effect The of 1860–2000Three are available,integrations forced byof estimates CM2.0 of and changes five in of well CM2.1 for the period the aerosols on theindirect scattering effect is thought and to absorption increase the magnitude of radiation. of the aerosol The mixed1860–2000 greenhouse gases,are available, tropospheric sulfateforced and by carbonaceous estimates of changes in well indirect effect is thoughtcooling substantially. to increase A significant the magnitude response of of the the Sahel aerosol to the aerosols,mixed volcanic greenhouse aerosols, ozone, gases, solar tropospheric irradiance, and land sulfate use. and carbonaceous cooling substantially.large aerosol A significant cooling in this response model is observed. of the Sahel to the Singleaerosols, realizations volcanic of the IPCC aerosols, future scenarios ozone, designated solar irradiance, B1, and land use. A1B, and A2 are also available for each model (23); by 2100, large aerosol coolingIn contrast, in this increasing model greenhouse is observed. gases, to the extent that Single realizations of the IPCC future scenarios designated B1, they influence interhemispheric SST gradients at all, are ex- these scenarios reach CO2 values of Ϸ550, 720, and 860 ppm, In contrast, increasing greenhouse gases, to the extent that respectively.A1B, and A2 are also available for each model (23); by 2100, pected to warm the Northern hemisphere more quickly than the these scenarios reach CO values of Ϸ550, 720, and 860 ppm, they influence interhemisphericSouthern hemisphere. If SST the effect gradients on Sahel at rainfall all, is are through ex- The ensemble mean and the standard2 deviation of the eight respectively. pected to warm thethe Northern interhemispheric hemisphere SST gradient, more then quickly this effect than should the 20th century simulations are plotted against the observations in Southern hemisphere.increase If Sahel the rainfall. effect It on is Sahel sometimes rainfall suggested is through that the Fig. 1,The along ensemblewith the single mean realization and that the provides standard the best deviation fit of the eight the interhemisphericrecovery SST of the Sahel gradient, rainfall in then the last thistwo decades effect could should be the to the20th observed century series simulations over the last half are of the plotted 20th century. against Just the observations in increase Sahel rainfall.first sign of the It emergence is sometimes of this greenhouse suggested signal (17). that If this the as forFig. the 1, observations, along with we the average single the simulated realization rainfall that over provides the best fit recovery of the Sahelsignal dominatesrainfall in the the 21st last century, two we decades should, by could this argument, be the July–August–Septemberto the observed and series over over the region the (10°-20°N, last half 20°W– of the 20th century. Just first sign of the emergenceexpect further of moistening this greenhouse in the Sahel. Several signal models (17). simulate If this 40°E),as for and usethe a observations, 5-year running mean we before average displaying the simulated the rainfall over results. The observations and model results are both normalized signal dominatessuch in the a trend 21st (18, century, 19). On thewe other should, hand, by the this results argument, in ref. 10 July–August–September and over the region (10°-20°N, 20°W– suggest that the Sahel dries in some models in response to by their40°E), mean and over the use century a 5-year (the mean running rainfall in meanCM2 in this before displaying the expect further moistening in the Sahel. Several models simulate region is 20% larger than observed). We also show the six such a trend (18,uniform 19). On warming the otherof SSTs, hand, with the the implication results that in ref. further 10 results. The observations and model results are both normalized drying in the 21st century is a possibility. individualby their 21st mean century over simulations the century generated by (the the two mean models rainfall in CM2 in this suggest that the Sahel dries in some models in response to for the three scenarios. region is 20% larger than observed). We also show the six uniform warmingResults of SSTs, with the implication that further The ensemble mean simulation shows a drying trend over the secondindividual half of the 21st 20th century, century with simulations a hint of recovery generated near the by the two models drying in the 21stWe century present results is a here possibility. for Sahel rainfall from a climate model present.for the The three model’s scenarios. mean drying trend over 1950–2000 is recently developed at the Geophysical Fluid Dynamics Labora- Results 14%͞50The years. ensemble Based on 500-year mean control simulation simulations shows with both a drying trend over the tory of the National Oceanic and Atmospheric Administration. second half of the 20th century, with a hint of recovery near the We present results here for Sahel rainfall from a climate model models, we estimate that the width of the 95% confidence We refer to two versions of this model (designated CM2.0 and intervalpresent. for an eight-member The model’s ensemble mean mean drying of 50-year trend trends over 1950–2000 is recently developedCM2.1). at the For aGeophysical description of both Fluid versions, Dynamics and an analysis Labora- of the is Ϯ14%3–4%.͞50 Therefore, years. this Based ensemble on mean 500-year trend is control primarily the simulations with both tory of the Nationalsimulation Oceanic of surface and temperature Atmospheric trends in Administration.the 20th century that response of the model to varying external forcing rather than these models generate, see refs. 20 and 21. In middle and high models, we estimate that the width of the 95% confidence We refer to two versions of this model (designated CM2.0 and internalinterval variability. for an eight-member ensemble mean of 50-year trends CM2.1). For a descriptionlatitudes, CM2.1 of both produces versions, better positioned and an analysis wind fields of and the If we consider CM2.0 and CM2.1 separately, we find that the smaller biases in surface temperature, but aspects of the tropical is Ϯ 3–4%. Therefore, this ensemble mean trend is primarily the simulation of surface temperature trends in the 20th century that CM2.0response ensemble of mean the moistens model the toSahel varying over the first external 50 years forcing rather than these models generate,circulation, see including refs. the 20 simulated and 21. El In Nin middle˜o-Southern and Oscilla- high of the 20th century, as in the observations, whereas the CM2.1 tion (ENSO) variability, are superior in CM2.0 (20, 22). Both internal variability. latitudes, CM2.1 produces better positioned wind fields and ensemble mean dries more continuously over the century. A models have been used to generate multiple realizations of 20th largerIf ensemble we consider would be CM2.0 needed to and determine CM2.1 whether separately, this we find that the smaller biases in surfaceand 21st century temperature, climate following but aspects the protocol of the set tropical by the CM2.0 ensemble mean moistens the Sahel over the first 50 years circulation, including the simulated El Nin˜o-Southern Oscilla- distinction between CM2.0 and CM2.1 is significant or due to Intergovernmental Panel on Climate Change (IPCC) for its samplingof the of 20th internal century, variability. as The in observed the observations, drying trend is whereas the CM2.1 tion (ENSO) variability,upcoming Fourth are superior Assessment (AR4).in CM2.0 The two (20, models 22). produce Both 39%ensemble over 1950–2000 mean and 14% dries over morethe full century, continuously as compared over the century. A models have beensimilar used responses to generate in the multiple Sahel. Neither realizations model includes of 20th the to thelarger corresponding ensemble figures would averaged be over needed the full CM2 to ensem- determine whether this and 21st centuryindirect climate effects following of aerosols on the clouds, protocol and neither set includes by the bledistinction of 14%͞50 years between and 13% CM2.0͞100 years. and The CM2.1 latter figure is significant or due to Intergovernmentalinteractive Panel vegetation on Climate or interactive Change dust aerosol. (IPCC) The primary for its happenssampling to be similar of internal to that in the variability. observations, The but it observed is clear drying trend is upcoming Fourthdifference Assessment between (AR4). the two models The is two in the models numerical produce advection from39% Fig. over 1 that 1950–2000 the observed and trend 14% is very over sensitive the full to the century, as compared similar responsesscheme in the in the Sahel. atmosphere, Neither but there model are also significant includes differ- the endpointsto the of corresponding the interval chosen. figures averaged over the full CM2 ensem- indirect effectsences of aerosols in the treatment on of clouds, frozen soil and and subgrid neither scale includes mixing in Asble seen of in Fig. 14% 1, the͞50 realization years that and best 13% fits the͞100 observations years. The latter figure interactive vegetationthe ocean. or We interactive refer to the set dust of both aerosol. models The as CM2. primary (onehappens of the CM2.0 to be integrations) similar happens to that to in generate the observations, droughts but it is clear difference between the two models is in the numerical advection from Fig. 1 that the observed trend is very sensitive to the scheme in the atmosphere,17892 ͉ www.pnas.org but͞cgi there͞doi͞10.1073 are͞pnas.0509057102 also significant differ- endpoints of the interval chosen. Held et al. ences in the treatment of frozen soil and subgrid scale mixing in As seen in Fig. 1, the realization that best fits the observations the ocean. We refer to the set of both models as CM2. (one of the CM2.0 integrations) happens to generate droughts

17892 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.0509057102 Held et al. INAUGURAL ARTICLE Climate models significantly underestimate Sahel drought INAUGURAL ARTICLE

Fig. 2. Observed and modeled rainfall trends. (Left) The linear trend from 1950 to 2000 in the observed (CRU) July–August–September rainfall over land, in mm͞month per 50 years. Blue areas correspond to a trend toward wetter conditions, and brown areas toward a drier climate. (Center) The linear trend for the eight-member ensembleObs mean of CM2 but plotted over both landCoupled SST and ocean. (Right) Linear trend for an ensemble meanPrescribed SST of 10 simulations with the atmospheric͞land component of CM2.0 running over observed sea surface boundary conditions.

Fig. 2. Observed and modeled rainfall trends. (Left) The linear trend from 1950 to 2000 in the observed (CRU) July–August–September rainfall over land, in comparable in timing and intensity to those observed in the The spatial pattern of the precipitation trends from 1950 to 1970s andmm͞ 1980s,month per but 50 noneyears. Blue of the areas runs correspond captures to a trendthe strength toward wetter of conditions,2000 is and shown brown in areas Fig. toward 2 for the a drier CRU climate. observations (Center) The linear(Left)andthe trend for the the observedeight-member maximum ensemble in mean rainfall of CM2 in but the plotted 1950s. over From both land the and ocean.eight-member (Right) Linear ensemble trend for anmean ensemble with mean the of full 10 simulationsset of forcings with the perspective of these time series in isolation, we conclude that (Center). The model captures the moistening near the Gulf of CM2’satmospheric 20th century͞land simulations component of are CM2.0 marginally running over consistent observed sea with surface boundaryGuinea conditions. as well as the basic pattern of the drying but with the the observations, with the realization that most resembles the center of mass of the drying shifted northwest of that observed. observed time series being the driest member of the ensemble in As evident from Fig. 1, the amplitude of the ensemble mean the 1980s. The model supports the hypothesis that the drought drying is weaker than that observed. The spatial pattern in the of thecomparable later part of in the timing century and was intensity partly to related those observedto a forced in therealizationThe spatial shownpattern in Fig. 1 of that the best precipitation fits the observed trends from time 1950 series to drying trend and partly a consequence of internal variability that is similar in structure but Ϸ25% larger in amplitude than the SCIENCES

1970s and 1980s, but none of the runs captures the strength of 2000 is shown in Fig. 2 for the CRU observations (Left)andthe ENVIRONMENTAL happened to reinforce this trend. pattern in Fig. 2b (data not shown). Thethe decline observed in Sahel maximum rainfall in in rainfallthe second in the half 1950s. of the From 20th the Becauseeight-member variations ensemble in SSTs aremean thought with to the underlie full set the of changes forcings century in the model’s ensemble mean is the result of anthro- in Sahel rainfall, it is important to test a model by running it over pogenicperspective forcing. An of ensemble these time of series three in CM2.1 isolation, runs we with conclude natural thatthe( observedCenter). The sea model surface captures as athe boundary moistening condition. near the Lu Gulf and of forcingCM2’s (solar plus20th volcanic century simulationsonly) produces are an marginally insignificant consistent trend. withDelworthGuinea (12) as well have as the analyzed basic pattern Sahel ofrainfall the drying in a but set with of 10 the Avarietyofintegrationsexaminingtheresponsetodifferent simulations for the period 1950–2000, using the atmospheric componentsthe observations, of the anthropogenic with the realization forcing are that underway. most resembles Results thecomponentcenter of of mass CM2.0, of the in drying which shifted only thenorthwest SST and of that sea observed. ice are to dateobserved indicate time that series aerosols being and the thedriest well member mixed of greenhouse the ensemble inmodifiedAs evident after from observations, Fig. 1, the with amplitude no changes of the in ensemble greenhouse mean gases are of comparable importance for the drying trend, but the gases, aerosols, solar irradiance, volcanoes, or land use. Given existingthe ensemble 1980s. The size model does supports not permit the hypothesis a more quantitative that the droughtthe observeddrying is weaker SSTs, the than model that comesobserved. much The closer spatial than pattern the freely in the estimateof of the their later relative part of importance. the century The was aerosolpartly related results to are a forcedin evolvingrealization coupled shown model in Fig. to capturing 1 that best the fits full the range observed of the time rainfall series qualitative agreement with ref. 14. Because we only model the reduction from the 1950s to the 1980s (Fig. 3). It also captures SCIENCES direct effectdrying of trend aerosols and partly on radiative a consequence fluxes ofand internal not the variability indirect thatsomeis of similar the year-to-year in structure variability but Ϸ25% in larger recent in years amplitude but with than some the effects on clouds, we may be underestimating the size of the disconcerting departures from observations in earlier periods, ENVIRONMENTAL aerosolhappened forcing. The to reinforce fact that this increasing trend. greenhouse gases also suchpattern as the in very Fig. wet 2b (data prediction not shown). for 1971. (In a freely running tends to dryThe the decline Sahel in in Sahel CM2 rainfall runs counter in the to second the conventional half of the 20thcoupledBecause model, variations there is in no SSTs reason are thought to expect to underlie correlation the changes with wisdom that this effect should be to increase precipitation by observations on a yearly time scale, because the model’s internal warmingcentury the Northern in the model’s hemisphere ensemble more mean rapidly is the than result the South- of anthro-variabilityin Sahel of rainfall, SSTs it will is important be uncorrelated to test a model with by the running observed it over ern hemisphere.pogenic forcing. An ensemble of three CM2.1 runs with naturalinternallythe observed generated sea variations; surface as however, a boundary when condition. one runs Lu anand Although our eight historical CM2 runs include estimates of atmospheric model over the observed SSTs, one can hope to historicalforcing changes (solar in plus the volcanic properties only) of producesthe land surface,an insignificant we have trend.simulateDelworth some (12) of the have year-to-year analyzed variability Sahel rainfall in tropical in a rainfall.) set of 10 yet to performAvarietyofintegrationsexaminingtheresponsetodifferent coupled model experiments in which the effects Thesimulations ensemble mean for the linear period trend 1950–2000, map for rainfall using over the 1950–2000 atmospheric of these changes are isolated. However, we have performed from these integrations is shown in the right panel of Fig. 2. The preliminarycomponents tests in of a the simpler anthropogenic model in forcing which are the underway. ocean com- Resultsatmospherecomponent͞land of component CM2.0, in which of this only model the SSTrun over and sea observed ice are ponent is replaced by a motionless slab with uniform heat SSTs captures the spatial pattern of the rainfall trend with capacityto and date specified indicate air–sea that aerosols heat flux and adjustments the well mixed that greenhouse mimic considerablemodified fidelity.after observations, with no changes in greenhouse the effectsgases of are oceanic of comparable heat transport, importance in for which the drying we study trend, the but the Togases, put the aerosols, Sahel drying solar irradiance, trend in the volcanoes, context of or the land trends use. Givenover response to the change in land surface properties imposed in the the entire African continent, we include in Fig. 4 plots identical historicalexisting runs. ensemble The results size indicate does not that permit the response a more to quantitative these to thosethe observed in Fig. 2, SSTs, except the for model the annual comes much mean closer precipitation. than the freely The changesestimate in land of surfacetheir relative properties importance. is modest The aerosol and primarily results are inannualevolving mean coupled simultaneously model to captures capturing the the rainy full range season of the in differ- rainfall confined in our model to the West African coastal forests rather ent regions at different times of year. The drying trend in the than thequalitative Sahel. agreement with ref. 14. Because we only model theSahelreduction over the from last thehalf 1950s of the to 20th the 1980s century (Fig. has 3). been It also accompa- captures direct effect of aerosols on radiative fluxes and not the indirect some of the year-to-year variability in recent years but with some Held et aleffects. on clouds, we may be underestimating the size of the disconcertingPNAS departures͉ December from 13, 2005 observations͉ vol. 102 in͉ earlierno. 50 ͉ periods,17893 aerosol forcing. The fact that increasing greenhouse gases also such as the very wet prediction for 1971. (In a freely running tends to dry the Sahel in CM2 runs counter to the conventional coupled model, there is no reason to expect correlation with wisdom that this effect should be to increase precipitation by observations on a yearly time scale, because the model’s internal warming the Northern hemisphere more rapidly than the South- variability of SSTs will be uncorrelated with the observed ern hemisphere. internally generated variations; however, when one runs an Although our eight historical CM2 runs include estimates of atmospheric model over the observed SSTs, one can hope to historical changes in the properties of the land surface, we have simulate some of the year-to-year variability in tropical rainfall.) yet to perform coupled model experiments in which the effects The ensemble mean linear trend map for rainfall over 1950–2000 of these changes are isolated. However, we have performed from these integrations is shown in the right panel of Fig. 2. The preliminary tests in a simpler model in which the ocean com- atmosphere͞land component of this model run over observed ponent is replaced by a motionless slab with uniform heat SSTs captures the spatial pattern of the rainfall trend with capacity and specified air–sea heat flux adjustments that mimic considerable fidelity. the effects of oceanic heat transport, in which we study the To put the Sahel drying trend in the context of the trends over response to the change in land surface properties imposed in the the entire African continent, we include in Fig. 4 plots identical historical runs. The results indicate that the response to these to those in Fig. 2, except for the annual mean precipitation. The changes in land surface properties is modest and primarily annual mean simultaneously captures the rainy season in differ- confined in our model to the West African coastal forests rather ent regions at different times of year. The drying trend in the than the Sahel. Sahel over the last half of the 20th century has been accompa-

Held et al. PNAS ͉ December 13, 2005 ͉ vol. 102 ͉ no. 50 ͉ 17893

Atlantic Multi-Decadal Oscillation NAO index (station based) = Winter (December through March) index of the NAO based on the difference of normalized sea level pressure (SLP) between Lisbon, Portugal and Stykkisholmur/Reykjavik, Iceland since 1864. NAO index (EOF based) = The principal component (PC) time series of the leading EOF of seasonal (December through March) SLP anomalies over the Atlantic sector (20-80N, 90W-40E) (Hurrell 1995).

A plausible mechanism

Warmer Arctic -> Decrease baroclinicity-> Jetstream rainfall moves southward-> Less precipitation over N. Atlantic -> Less fresh water flux increase salinity-> Denser water sinks -> Increase THC -> Increase poleward heat transport-> Shift jetstream poleward -> Fresh water flux increases -> Reduce salinity-> Reduce THC -> less poleward heat transport -> Colder Arctic…

Cycles repeat in interannual, decadal to multi-decadal time scales

A Pacific Interdecadal Climate Oscillation with Impacts on Salmon Production*

Nathan J. Mantua,+ Steven R. Hare,# Yuan Zhang,+ John M. Wallace,+ and Robert C. Francis@

ABSTRACT

Evidence gleaned from the instrumental record of climate data identifies a robust, recurring pattern of ocean–atmosphere climate variability centered over the midlatitude North Pacific basin. Over the past century, the amplitude of this climate pattern has varied irregularly at interannual-to-interdecadal timescales. There is evidence of reversals in the prevailing polarity of the oscillation occurring around 1925, 1947, and 1977; the last two reversals correspond to dramatic shifts in salmon production regimes in the North Pacific Ocean. This climate pattern also affects coastal sea and continental surface air temperatures, as well as streamflow in major west coast river systems, from Alaska to California.

September 1915 (Pacific Fisherman 1915) The [May, June and July chinook] catch this year is one of the lowest in the history of the Never before have the Bristol Bay [Alaska] Columbia [Washington and Oregon]. salmon packers returned to port after the season’s operations so early. August/September 1972 (National Fisherman 1972)

The spring [chinook salmon] fishing season on Bristol Bay [Alaska] salmon run a disaster. the Columbia River [Washington and Oregon] closed at noon on August 25, and proved to be Gillnetters in the Lower Columbia [Washing- one of the best for some years. ton and Oregon] received an unexpected bo- nus when the largest run of spring chinook 1939 Yearbook (Pacific Fisherman 1939) since counting began in 1938 entered the river. The Bristol Bay [Alaska] Red [sockeye salmon] run was regarded as the greatest in history. 1995 Yearbook (Pacific Fishing 1995)

Alaska set a new record for its salmon harvest in 1994, breaking the record set the year before. *JISAO Contribution Number 379. +Joint Institute for the Study of the Atmosphere and Oceans, University of Washington, Seattle, Washington. Columbia [Washington and Oregon] spring #International Pacific Halibut Commission, University of Wash- chinook fishery shut down; west coast troll ington, Seattle, Washington. coho fishing banned. @Fisheries Research Institute, University of Washington, Seattle, Washington. Corresponding author address: Nathan Mantua, Joint Institute for 1. Introduction the Study of the Atmosphere and Oceans, University of Washing- ton, Box 354235, Seattle, WA 98195-4235. E-mail: [email protected]. Pacific salmon production has a rich history of In final form 6 January 1997. confounding expectations. For much of the past

Bulletin of the American Meteorological Society 1069 PDO index = Leading PC of EOF of N. Pacific monthly SST, poleward of 20N

Mantua et al 2002

Climatic impacts (+ve phase) of PDO

• Warm phase of PDO, strongest signal in Nov- April - Warmer Alaska, northwest N. America - Cooler eastern China, Korea, Japan, Kamchatka, southeastern US - Drier Nov-April in Korea, Japan, Russian far east; wetter Gulf of Alaska, Pacific Northwest, southwestern US

The Arctic Oscillation (AO), or Northern Hemisphere Annular Mode (NAM)

A Global Warming Hiatus (1998-2015) ?

A “Pause” in Global Warming?

Karl et al., 2015: Possible artifacts of data biases in the recent global surface warming hiatus. Science

Rajaratnam, et al, 2015, Debunking the climate hiatus. Nat. Geosci,

Trenberth and Fasulo, 2013, Globel warming and Natural Variability, Future Corrected NOAA temperature record Earth

Earth’sFuture 10.1002/2013EF000165

Global mean surface temperature 12-month running mean 0.7

0.6

0.5

Earth’sFuture 10.1002/2013EF000165 0.4More frequent, but weaker ENSO cycles since 2000? °C Global mean surface temperature 0.3 12-month running mean 0.7

0.2 0.6

0.5 0.1 0.4 El Niño °C 0.3 La Niña 0.0 0.2 events

-0.1 0.1 1970 1975 1980 1985 1990 1995 2000El 2005Niño 2010 2015 La Niña 0.0 events -0.1 Niño 3.4 2 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

Niño 3.4 2 0 0

–2 Base 1950-79 –2SST anomaly °C Base 1950-79 SST anomaly °C 1970 1980 1990 2000 2010 1970 1980 1990 2000 2010 Figure 6. The NOAA global mean 12 month running mean surface temperatures are given relative to 1901–2000 along with a linear trend fit. Marked on the graph are the El Nino˜ (buff)andLaNina˜ (sky blue) periods as defined by NOAA’s ONI, based on the Nino˜ 3.4 FigureSST 6. anomalies,The NOAA as given global in the mean lower 12 panel month relative running to a base mean period surface of 1950–1979. temperatures are given relative to 1901–2000 along with a linear trend fit. Marked on the graph are the El Nino˜ (buff)andLaNina˜ (sky blue) periods as defined by NOAA’s ONI, based on the Nino˜ 3.4 the AMO. However, these results depend on the fidelity of models and the forcings used, and the latter are SST anomalies,not well known, as given especially in the lower for panel aerosols. relative to a base period of 1950–1979.

The NAO index (Figure 7) depicts NAO (DJFM) 4 the AMO. 4 However, these results depend on the ﬁdelity of modelsthe strength and the of theforcings westerliesTrenberth and Fasulo, 2013 used, from and the latter are 2 2 the North Atlantic into Europe and

not well 0 known, especially for aerosols. 0 correlates well with temperatures in Eurasia and inversely with those over –2 –2 Greenland, as well as precipitation as –4 –4 anorth-southdipoleoverEurope:wetThe NAO index (Figure 7) depicts 4 NAO (DJFM) 4 1970 1980 1990 2000 2010 in the norththe strength and dry in of the the south westerlies in the from positive phase. The winter (December AMO 2 through2 the March) North station-based Atlantic into index Europe and 0.2 0.2 of the NAO [Hurrell, 1995] is based on 0 correlates well with temperatures in 0 0.0 0.0 the difference of normalized sea-level °C hPa pressureEurasia (SLP) between and inversely Lisbon, Portu- with those over –2 –0.2 –0.2 –2gal, and Stykkisholmur/Reykjavik, Ice- land, sinceGreenland, 1864 in hectopascal as well as (hPa); precipitation as –4 1970 1980 1990 2000 2010 –4http://climatedataguide.ucar.edu/guidanorth-southdipoleoverEurope:wet ance/hurrell-north-atlantic-oscillation- Figure 7. Time series of values of the NAO in northern winter (DJFM) and nao-index-station-based. The NAO is annual1970 mean AMO 1980 along with a low-pass 1990 (13-term) decadal 2000 filter used 2010in in the north and dry in the south in the IPCC [Trenberth et al., 2007]. For AMO the units are K and for NAO the units important in the northern extratrop- are hPa. ics in winterpositive [Hurrell phase.,1996]whereit The winter (December AMO accountedthrough for 31% March) of the 20station-based∘N–90∘N index 0.2 0.2 of the NAO [Hurrell, 1995] is based on TRENBERTH AND FASULLO © 2013 The Authors. 24 0.0 0.0 the difference of normalized sea-level °C hPa pressure (SLP) between Lisbon, Portu- –0.2 –0.2 gal, and Stykkisholmur/Reykjavik, Ice- land, since 1864 in hectopascal (hPa); 1970 1980 1990 2000 2010 http://climatedataguide.ucar.edu/guid ance/hurrell-north-atlantic-oscillation- Figure 7. Time series of values of the NAO in northern winter (DJFM) and nao-index-station-based. The NAO is annual mean AMO along with a low-pass (13-term) decadal filter used in IPCC [Trenberth et al., 2007]. For AMO the units are K and for NAO the units important in the northern extratrop- are hPa. ics in winter [Hurrell,1996]whereit accounted for 31% of the 20∘N–90∘N

surface temperature variance for 1935–1994 for DJFM, but subsequently NAO has not gone hand-in-hand with global temperature and there is no significant correlation overall. surface temperature variance for 1935–1994 for DJFM, but subsequently NAO has not gone hand-in-hand Thewith AMO global is a measure temperature of SSTs and in the there North is no Atlantic, significant north correlation of the equator, overall. relative to the global mean [Tren- berth and Shea,2006].Therecentpost-1970variationsintheAMOandNAO(Figure7)showindeedthat The AMO is a measure of SSTs in the North Atlantic, north of the equator, relative to the global mean [Tren- variability is quite large. Note that in terms of global mean temperature, the scale on Figure 7 would be berth and Shea,2006].Therecentpost-1970variationsintheAMOandNAO(Figure7)showindeedthat reduced by the ratio of the area of the North Atlantic to the global area, which is 7.3%. variability is quite large. Note that in terms of global mean temperature, the scale on Figure 7 would be Thereduced PDO has by been the identifiedratio of the with area changes of the North in SLP Atlantic over the to North the global Pacific area, [Trenberth which is and 7.3%. Hurrell,1994]. Often it is defined by using SSTs in the Pacific [Mantua et al., 1997] using 110∘Eto100∘W, 20∘Nto70∘N The PDO has been identified with changes in SLP over the North Pacific [Trenberth and Hurrell,1994]. as a core region, with the global mean SSTs removed, to compute the first empirical orthogonal function Often it is defined by using SSTs in the Pacific [Mantua et al., 1997] using 110∘Eto100∘W, 20∘Nto70∘N (EOF) pattern and associated time series, and then regress the time series with SSTs over the entire globe as a core region, with the global mean SSTs removed, to compute the first empirical orthogonal function (Figure 8). This is a new analysis (courtesy of Adam Phillips; cf. Deser et al.[2004])andtheEOFaccountsfor (EOF) pattern and associated time series, and then regress the time series with SSTs over the entire globe 25% of the monthly anomaly variance of SST for the period 1900 to May 2013, using the HADISST dataset. (Figure 8). This is a new analysis (courtesy of Adam Phillips; cf. Deser et al.[2004])andtheEOFaccountsfor Chen et al.[2008]provideanalternativederivationofPacificdecadalvariabilitythatshowshowrobust 25% of the monthly anomaly variance of SST for the period 1900 to May 2013, using the HADISST dataset. it is to different approaches. They also note how similar many aspects of the pattern are to ENSO but Chen et al.[2008]provideanalternativederivationofPacificdecadalvariabilitythatshowshowrobust that the PDO does not account for changes in global mean surface temperature owing to large regional it is to different approaches. They also note how similar many aspects of the pattern are to ENSO but cancellations. that the PDO does not account for changes in global mean surface temperature owing to large regional cancellations. Pacific Decadal Oscillation Core domain 25% 110°E-100°W Pacific Decadal Oscillation variance 20°-70°NCore domain 25% 110°E-100°W variance 20°-70°N

2 4

0 2

–2 0

PDO index normalized Base period 1900-2013 –4 –2 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010

PDO index normalized Base period 1900-2013 –4 Figure 8. The Pacific Decadal Oscillation based on an EOF analysis of SST anomalies with the global mean removed from 1900 to 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 May 2013 in the 20∘N–70∘Nand110∘E–100∘W region of the North Pacific, which explains 25% of the variance. The principal component time series, given below in normalized units, is regressed on global SSTs to give the map above. The black curve is a 61 Figure 8. The Pacific Decadal Oscillation based on an EOF analysis of SST anomalies with the global mean removed from 1900 to month running average. May 2013 in the 20∘N–70∘Nand110∘E–100∘W region of the North Pacific, which explains 25% of the variance. The principal component time series, given below in normalized units, is regressed on global SSTs to give the map above. The black curve is a 61 month running average. TRENBERTH AND FASULLO © 2013 The Authors. 25

Deniers of climate change often cherry-pick points on time series and seize on the El Nino˜ warm year of 1998 as the start of the hiatus in global mean temperature rise (Figure 6). This turns out, arguably, to have been the transition time from a positive to a negative phase of the PDO. The monthly time series (Figure 8) readily reveals the multidecadal regimes of the PDO (given by the black line) with positive phases from 1923 to 1942 and 1976 to 1998, and negative phases from 1943 to 1976 and after 1999. While naturally emphasizing the North Pacific, the pattern covers the entire Pacific with a somewhat ENSO-like pattern but one that is broader in the tropics [Chen et al., 2008]. If we now examine the hiatus period of 1999–2012 and compare it to the time when global warming really took off from 1976 to 1998 (Figure 9), the negative PDO pattern emerges very strongly throughout the Pacific although warming prevails in the Atlantic and Indian Oceans and on land. In other words, it is the central and eastern Pacific more than anywhere else that has not warmed in the past decade or so. In spite of some cold European winters, Europe does not standout in Figure 9 and instead is a warm region. The AMO is positive (Figure 7) and is revealed in Figure 9 to be part of a wider warming. One approach to estimating ocean heat content (OHC) changes is by combining the available observations (surface, ocean, and from space) with an ocean model to produce a dynamically consistent ocean analysis. The new ORAS-4 ocean reanalysis from ECMWF has revealed very distinctive climate signatures that are realistic in magnitude and duration in terms of changes in OHC [Balmaseda et al., 2013] (Trenberth et al., submitted manuscript, 2013). Figure 10 shows the five ensemble members of the ORAS-4 ocean reanalysis OHC for 0–700m and full-depth ocean and reveals the increased heating below 700m depth of 0.21 W m−2 globally after 2000. The orange bars show the times of the El Chichón and Pinatubo vol-

Earth’sFuturecanic eruptions when sharp drops occurred in OHC that quantitatively match10.1002/2013EF000165 estimates of TOA radiative changes (such as in Pinatubo) [Trenberth and Dai,2007],asdemonstratedinanewanalysisbyTrenberth etDeniers al. (submitted of climate manuscript, change often 2013). cherry-pick ORAS-4 points also reveals on time a majorseries and cooling seize of on the the tropical El Nino˜ Pacific warm yearOcean of in 1998 as the start of the hiatus in global mean temperature rise (Figure 6). This turns out, arguably, to have associationbeen the transition with the time 1997–1998 from a positive El Nino˜ to event. a negative Following phase this, of the the PDO. ocean The warmed monthly at time a startling series (Figure rate of 8) over readily reveals−2 the multidecadal regimes of the PDO (given by−2 the black line) with positive phases from 1.21923 W tom 1942from and the 1976 2000s to 1998, for the and global negative ocean phases (or 0.84 from W m1943for to the1976 global and after area), 1999. and While the overall naturally heating emphasizing the North Pacific,−2 the pattern covers the entire Pacific with a somewhat ENSO-like pattern isbut estimated one that to is broader be 0.91 Win mthe tropicsglobally [Chen when et almelting., 2008]. sea ice and other components are included as well [IfBalmaseda we now examine et al., 2013] the hiatus (Trenberth period et of al., 1999–2012 submitted and manuscript, compare it 2013). to the More time than when 30% global of the warming heat was really took off from 1976 to 1998 (Figure 9), the negative PDO pattern emerges very strongly throughout depositedthe Pacific into although the ocean warming below prevails 700 m in in the an Atlantic unprecedented and Indian fashion Oceans in and the on post land. 2000 In otherrecord words, from ORAS-4 it is the central and eastern Pacific more than anywhere else that has not warmed in the past decade or so. In andspite was of some identified cold European mainly with winters, changes Europe in the does tropical not standout and subtropical in Figure winds 9 and ininstead the Pacific. is a warm region. The AMO is positive (Figure 7) and is revealed in Figure 9 to be part of a wider warming. FigureOne approach 11 shows to the estimating regime changesocean heat for content 1999–2012 (OHC) versus changes 1979–1998 is by combining from the the ERA-I available reanalysis observa- for SLP tions (surface, ocean, and from space) with an ocean model to produce a dynamically consistent ocean andanalysis. surface The winds. new ORAS-4 Reanalysis ocean winds reanalysis and surface from ECMWF fluxes, has bias revealed corrected, very were distinctive used to climate drive the signatures ocean in that are realistic in magnitude and duration in terms of changes in OHC [Balmaseda et al., 2013] (Trenberth et al., submitted manuscript, 2013). Figure 10 shows the five ensemble members of the ORAS-4 ocean reanalysis OHC for 0–700m and full-depth ocean and reveals the increased heating below 700m depth of 0.21 W m−2 globally after 2000. The orange bars show the times of the El Chichón and Pinatubo volcanic eruptions when sharpAnnual drops mean occurred surface in OHC that temperatures quantitatively match estimates of TOA radiative changes (such as in Pinatubo) [Trenberth[1999-2012] and Dai − [1976-1998],2007],asdemonstratedinanewanalysisbyTrenberth et al. (submitted manuscript, 2013). ORAS-4 also reveals a major cooling of the tropical Pacific Ocean in association with the 1997–1998 El Nino˜ event. Following this, the ocean warmed at a startling rate of over 1.2 WGISS m−2 from the 2000s for the global ocean (or 0.84 W m−2 for the global area), and the overall heating is estimated to be 0.91 W m−2 globally when melting sea ice and other components are included as well [Balmaseda et al., 2013] (Trenberth et al., submitted manuscript, 2013). More than 30% of the heat was deposited into the ocean below 700 m in an unprecedented fashion in the post 2000 record from ORAS-4 and was identified mainly with changes in the tropical and subtropical winds in the Pacific. Figure 11 shows the regime changes for 1999–2012 versus 1979–1998 from the ERA-I reanalysis for SLP andPDO confounded the GW signal during 1998-2012 surface winds. Reanalysis winds and surface fluxes, bias corrected, were used to drive the ocean in

Annual mean surface temperatures [1999-2012] − [1976-1998]

GISS

Figure 9. Mean annual surface temperature diﬀerences from GISS for 1999–2012 and 1976–1998 in ∘C, with zonal means at right for ocean (blue), land (red), and zonal mean (black). Figure 9. Mean annual surface temperature diﬀerences from GISS for 1999–2012 and 1976–1998 in ∘C, with zonal means at right TRENBERTH AND FASULLO for ocean (blue), land (red), and zonal mean (black).© 2013 The Authors. 26

TRENBERTH AND FASULLO © 2013 The Authors. 26 RESEARCHEquatorial Pacific SST (~8% of global ocean area) LETTER play outsized role in “warming hiatus” a Figure 1 | Observed and simulated 0.7 global temperature trends. Annual- mean time series based on 0.6 observations, HIST and POGA-H 0.5 (a) and on POGA-C (b). Anomalies are deviations from the 1980–1999 0.4 averages, except for HIST, for which the reference is the 1980–1999 0.3 Pinatubo average of POGA-H. SAT anomalies 0.2 over the restoring region are plotted 0.1 El Chichon in b, with the axis on the right. Major volcanic eruptions are indicated in 0 Observations a. c, Trends of seasonal global Agung –0.1 temperature for 2002–2012 in observations and POGA-H. Shading –0.2 represents 95% confidence interval of –0.3 ensemble means. Bars on the right of a show the ranges of ensemble Observed, HIST and POGA-H

annual-mean global average (ºC) –0.4 2002–2012 average spreads of the 2002–2012 averages. POGA-H 95th percentile –0.5 Ensemble mean –0.6 Observations HIST 99th percentile –0.7 POGA-H HIST –0.8 b 0.2 0.6

0.1 5-yr running mean 0.3 Restoring region SAT 0 0

–0.1 –0.3

–0.2 –0.6

global average (ºC) –0.3 –0.9 POGA-C annual-mean –0.4 –1.2 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010

0.1 Kosaka and Xie, 2013, Nature 0 36 –0.1 POGA-H –0.2 Observations –0.3 seasonal mean (ºC per 11 years) Observed and POGA-H 2002–2012 DJF JFM FMA MAM AMJ MJJ JJA JAS ASO SON OND NDJ

Pacific is similar in winter and in summer (Extended Data Fig. 4a), increase in the chance of an extratropical cooling in winter is partly stationary/transient eddies, which are the dominant mechanism for because the tropical influence is stronger in winter than in summer. meridional heat transport14, are stronger in winter than summer. As a We examined regional climate change associated with the hiatus. result, the tropical cooling effect on the extratropics is most pronounced Although models project a slowdown of the Walker circulation in in winter (the seasonality of the temperature trend in the Southern Hemis- global warming15, the Pacific Walker cell intensified during the past phere extratropics is weak). The tropical influence on the Northern decade (Fig. 2c). POGA-H captures this circulation change, forced by Hemisphere extratropics is weak during the summer, allowing the the SST cooling across the tropical Pacific (Fig. 2d). As in interannual radiative forcing to continue the warming trend during the recent ENSO, the tropical Pacific cooling excites global teleconnections in decade (Extended Data Fig. 4b). December, January and February (DJF; the season is denoted by the This seasonal contrast is evident also in HIST. For 1970–2040, a first letters of the months). SST changes in POGA-H are in broad period when the ensemble-mean global temperature shows a steady agreement with observations over the Indian, South Atlantic and increase in HIST, the probability density function for the 11-year trend Pacific oceans outside the restoring domain (Fig. 2a, b). The model is similar in winter and in summer for tropical temperatures, with reproduces the weakening of the Aleutian low as the response of the means both around 0.25 uC (Extended Data Fig. 4c). The probability Pacific–North American pattern to tropical Pacific cooling11 (Fig. 2c, d). density function is much broader for winter than for summer for Northern As a result, the SAT change over North America is well reproduced, Hemisphere extratropical temperatures (Extended Data Fig. 4d). The including a pronounced cooling in the northwest of the continent. The chance of the 11-year temperature change falling below –0.3 uCis8% model fails to simulate the SAT and sea-level pressure (SLP) changes for winter but only 0.7% for summer in the Northern Hemisphere over Eurasia, suggesting that they are due to internal variability unre- extratropics (around 4% in the tropics for both seasons). The 11-fold lated to tropical forcing (Extended Data Fig. 5a and c).

20 22 Earth’sFuture ORAS4 OHC 10 J 10.1002/2013EF000165 OHC continued to increase. No “GW Hiatus ” 15

20 ORAS4 OHC 1022 J 10 15 Joules

10 22 5 10 Joules

22 5 10 0 Upper 700m 0 TotalUpper Depth 700m -5 Total Depth -5 19801980 1985 1985 1990 1990 1995 1995 2000 2000 2005 2005 2010 2010

Figure 10. FigureThe ﬁve 10. ensembleThe ﬁve ensemble members members of the of ORAS-4 the ORAS-4 ocean ocean reanalysis reanalysis OHC for for 0–700 0–700 m andm and full-depth full-depth ocean ocean are shown, are whereshown, where 22 −2 they have been aligned for 1980 to 1985, in 10 J.22 The increased heating below 700 m of about 0.2− W2 m globally is revealed after about 2000.they The have orange been bars aligned show for 1980 thetimes to 1985, of in the 10 ElJ.Chichón The increased and heating Pinatubo below volcanic 700 m of eruptions. about 0.2 W m globally is revealed after about 2000. The orange bars show the times of the El Chichón and Pinatubo volcanic eruptions.

ORAS-4 during the assimilation to result in the OHC fields in Figure 12. Figure 11a reveals the very strong changes toward higher pressures over the cool central and eastern Pacific especially in the subtropics and the muchORAS-4 stronger during the than assimilation normal to tradewinds result in the by OHC more fields than in Figure 1 m 12. s−1 Figurein the 11a vicinity reveals of the the very equator strong from 15∘Nto15changes∘S, 150 toward∘Eto150 higher∘W, pressures and in theover subtropics the cool central farther and eastern east (160 Pacific∘Wto110 especially∘W). in the The subtropics SST pattern of change isand reflected the much in stronger the OHC than changes normal downtradewinds to 700 by more m (Figure than 1 12)m s− signifying1 in the vicinity the of extra the equator heat storage from in the tropical western∘ Pacific∘ and∘ deeper∘ thermocline, but with much cooler∘ conditions∘ throughout the eastern Pacific from15 Nto15 30∘Nto30S, 150∘S.Eto150 VariabilityW, and in in the the surface subtropics wind farther field east is independently (160 Wto110 W). corroborated The SST pattern by of changes in sea levelchange based is reflected on both in the the altimetry OHC changes and down gauge to 700 records m (Figure as 12)the signifying easterly the anomalous extra heat windsstorage have in the driven a“pilingup”ofwaterinthewesternPacificOcean.Becauseofthisetropical western Pacific and deeper thermocline, but with much coolerff conditionsect, some throughout regions in the the eastern western Pacific have experienced sea-level rise at three times the rate of the global ocean in recent decades. The length ofPacific the gauge from 30 record∘Nto30 provides∘S. Variability an extended in the surface record wind overfield is which independently this regional corroborated increase bychanges can be linked to the PDO [inMerrifield sea level based et al., on 2012]. both the altimetry and gauge records as the easterly anomalous winds have driven Figure 11ba“pilingup”ofwaterinthewesternPacificOcean.Becauseofthise presents the northern polar view of the same changes inffect, Figure some 11a regions to highlight in the western the rela- tionships of the apparent wavelike structure extending northward from the Pacific, across the pole into Europe. ThisPacific aspect have experienced is likely better sea-level seen rise in at the three upper times troposphere the rate of the as global a quasi-stationary ocean in recent decades. Rossby The wave [Ineson andlength Scaife of the,2009],anaspecttobepursuedelsewhere.Nonetheless,itisverysuggestiveofa gauge record provides an extended record over which this regional increase can be linked to relationshipthe PDOwith [Merrifield the NAO et in al., its 2012]. negative phase. This also highlights the influence of the changes in the Pacific with the high latitudes of both hemispheres, the extension to the North Atlantic in the Northern HemisphereFigure and 11b to presents the Southern the northern Oceans polar in view the of Southern the same changes Hemisphere in Figure (Figure 11a to 11a), highlight where the rela- the wave structure relates to changes in Antarctic sea ice. tionships of the apparent wavelike structure extending northward from the Pacific, across the pole into These kinds of changes have been independently simulated in hiatus periods in warming scenarios of the 21st centuryEurope. in the This CCSM4 aspect is [Meehl likely better et al., seen 2011, in the 2013] upper in troposphere association as also a quasi-stationary with negative Rossby PDO wave (or IPO) periods and more[Ineson frequent and Scaife La Ni,2009],anaspecttobepursuedelsewhere.Nonetheless,itisverysuggestiveofana˜ events. They are identified with a stronger wind-driven overturning in the Pacific, withrelationship upwelling with near the NAO the in equator its negative and phase. subsiding This also waters highlights in the the subtropics, influence of leading the changes to the in the buildup in heat off the equator in the western and central Pacific. Pacific with the high latitudes of both hemispheres, the extension to the North Atlantic in the Northern So what about the cold northern winters in the 2000s that have been associated with the strong negative phaseHemisphere of the NAO? and to In the Figure Southern 7 the Oceans NAO in reveals the Southern some Hemisphere low-frequency (Figure variability 11a), where that the appearswave to be in phase withstructure the relates PDO variationsto changes in (Figures Antarctic 11b sea ice.and 13). Given the global nature of the atmosphere it is not surprising that links between the Pacific and Atlantic Oceans form at times, but these modes are not inherentlyThese coupled. kinds of Low-frequency changes have been variability independently in NAO simulated and links in hiatus to ENSO periods are in discussed warming scenarios by Ineson of the and Scaife [2009] who21st note century the in important the CCSM4 role[Meehl of et the al., global 2011, 2013] teleconnection in association also pathway with negative from the PDO Pacific (or IPO) region periods via the stratosphere. Moreover, small effects from the Sun in the ultraviolet from the lower stratosphere can be amplifiedand [Ineson more etfrequent al., 2011]. La Nina Together,˜ events. They the are PDO, identified AMO, withand a NAO stronger account wind-driven for a lot overturning of the regional in the and seasonal climatePacific, changes with upwelling going near on. the While equator these and are subsiding the predominant waters in the natural subtropics, modes leading of to variability, the buildup it in is quite heat off the equator in the western and central Pacific. TRENBERTH AND FASULLO © 2013 The Authors. 27 So what about the cold northern winters in the 2000s that have been associated with the strong negative phase of the NAO? In Figure 7 the NAO reveals some low-frequency variability that appears to be in phase with the PDO variations (Figures 11b and 13). Given the global nature of the atmosphere it is not surprising that links between the Pacific and Atlantic Oceans form at times, but these modes are not inherently coupled. Low-frequency variability in NAO and links to ENSO are discussed by Ineson and Scaife [2009] who note the important role of the global teleconnection pathway from the Pacific region via the stratosphere. Moreover, small effects from the Sun in the ultraviolet from the lower stratosphere can be amplified [Ineson et al., 2011]. Together, the PDO, AMO, and NAO account for a lot of the regional and seasonal climate changes going on. While these are the predominant natural modes of variability, it is quite

TRENBERTH AND FASULLO © 2013 The Authors. 27 Summary q Nature variability (interannual-interdecadal) confounds global warming (long-term trend). q Given strong interannual-to-multi-decadal variability, there is no reason to expect global warming rate estimate for any given period of time to be constant q Pauses, or global cooling (e.g. volcanic eruptions) may occur q Surface air temperature is only one expression of GW, subject to natural fluctuations q Ocean heat content (a more reliable measure of GW) continues to increase during the “Hiatus” q GW can modulate natural variability. Can we really separate GW with natural variability??? Supplementary Material The Atlantic Multi-decadal Oscillation (AMO): Quasi-periodic (50-70 years) oscillation of temperature of SST of the Atlantic Ocean [0-60N, 75 -10W) related to the thermohaline circulation (THC) of the ocean

During AMO positive phase:

- Warmer central and northern Europe - Northward shift of jestream over N. Atlantic - Increased equatorial trade winds (low level easterlies) over tropical Atlantic - More Atlantic hurricanes Earth’sFuture 10.1002/2013EF000165

Global mean surface temperature 12-month running mean 0.7

0.6

0.5

0.4 Earth’sFuture°C 10.1002/2013EF000165 0.3

0.2 Global mean surface temperature 12-month running mean 0.7 0.1 0.6 El Niño La Niña 0.0 0.5 events 0.4-0.1 °C 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 0.3 Niño 3.4 0.2 2

0.1 0 El Niño La Niña 0.0 events –2 Base 1950-79 -0.1SST anomaly °C 19701970 1975 1980 19851980 1990 1995 1990 2000 2005 2010 2000 2015 2010 Niño 3.4 Figure 6. The2 NOAA global mean 12 month running mean surface temperatures are given relative to 1901–2000 along with a linear trend fit. Marked on the graph are the El Nino˜ (buff)andLaNina˜ (sky blue) periods as defined by NOAA’s ONI, based on the Nino˜ 3.4 SST anomalies,0 as given in the lower panel relative to a base period of 1950–1979.

–2 Base 1950-79 the AMO.SST anomaly °C However, these results depend on the fidelity of models and the forcings used, and the latter are 1970 1980 1990 2000 2010 not well known, especially for aerosols. Figure 6. The NOAA global mean 12 month running mean surface temperatures are given relative to 1901–2000 along with a linear trend fit. Marked on the graph are the El Nino˜ (buff)andLaNina˜ (sky blue) periods as defined by NOAA’s ONI, based on the Nino˜ 3.4 SST anomalies, as given in the lower panel relative to a base period of 1950–1979. The NAO index (Figure 7) depicts NAO (DJFM) 4 4 the strength of the westerlies from the AMO. However, these results depend on the fidelity of models and the forcings used, and the latter are Declining NAO, increasing AMO (warming North Atlantic) 2 2 not well known, especially for aerosols. the North Atlantic into Europe and 0 0 correlates well with temperatures in The NAO index (Figure 7) depicts 4 NAO (DJFM) 4 Eurasia and inversely with those over –2 the strength–2 of the westerlies from 2 2 the North Atlantic intoGreenland, Europe and as well as precipitation as –4 –4 0 0 correlates well with temperaturesanorth-southdipoleoverEurope:wet in Eurasia and inversely with those over –2 1970 1980 1990 2000–2 2010 in the north and dry in the south in the Greenland, as well as precipitation as –4 positive phase. The winter (December AMO –4 anorth-southdipoleoverEurope:wet through March) station-based index 0.21970 1980 1990 2000 2010 in the north0.2 and dry in the south in the positive phase. The winterof the (December NAO [Hurrell, 1995] is based on AMO through March) station-based index 0.2 0.0 0.2 0.0 the difference of normalized sea-level °C hPa of the NAO [Hurrell, 1995] is based on pressure (SLP) between Lisbon, Portu- 0.0 0.0 the difference of normalized sea-level °C hPa –0.2 pressure (SLP)–0.2 betweengal, Lisbon, and Stykkisholmur/Reykjavik, Portu- Ice- –0.2 –0.2 gal, and Stykkisholmur/Reykjavik,land, since Ice- 1864 in hectopascal (hPa); land, since 1864 in hectopascal (hPa); 1970 1980 1990 2000 2010 http://climatedataguide.ucar.edu/guid 1970 1980 1990 2000 2010 http://climatedataguide.ucar.edu/guid ance/hurrell-north-atlantic-oscillation-ance/hurrell-north-atlantic-oscillation- FigureFigure 7. 7.TimeTime series series of values of of values the NAO of in the northern NAO winter in northern (DJFM) and winter (DJFM)nao-index-station-based. and nao-index-station-based. The NAO is The NAO is annualannual mean mean AMO AMO along along with a low-pass with a (13-term) low-pass decadal (13-term) filter used decadal in filter used in IPCC [Trenberth et al., 2007]. For AMO the units are K and for NAO the units important in the northern extratrop- IPCC [Trenberth et al., 2007]. For AMO the units are K and for NAO the units important in the northern extratrop- are hPa. ics in winter [Hurrell,1996]whereit are hPa. accounted for 31% ofics the in 20 winter∘N–90∘N [Hurrell,1996]whereit accounted for 31% of the 20∘N–90∘N TRENBERTH AND FASULLO © 2013 The Authors. 24