Hadley Centre Evidence for the Atlantic Multidecadal Oscillation

Hadley Centre Evidence for the Atlantic Multidecadal Oscillation as an internal climate mode from coupled GCM simulations Jeff Knight Hadley Centre, Exeter, UK 4 th International CLIVAR Climate of the 20 th Century Workshop, Hadley Centre, Exeter, UK Wednesday, 14 th March 2007 © Crown copyright 2004 Page 1

Hadley Centre AMO in observations © Crown copyright 2004 Page 2

AMO in observations Hadley Centre Mean North Atlantic SST § ‘AMO index’ § Low-pass (> 13. 3 y) filtered detrended Had. ISST § Larger than trend or interannual Surface temperature anomaly § Regression 1870 -1999 § Had. CRUTv blended SST/air temp § 90% confidence interval accounting for autocorrelation Palaeoclimate – AMO back to C 16 -17 th? § Tree rings (Gray et al. , 2004) § Multiproxy (Delworth and Mann, 2000) Models show some THC-SST links § e. g. Delworth and Mann, 2000 Is the AMO long-lived/periodic? Forced or internal? © Crown copyright 2004 Page 3

Hadley Centre AR 4 20 th Century Forcings Coupled Ensembles © Crown copyright 2004 Page 4

AR 4 Ensembles Hadley Centre Single model North Atlantic mean SST • Example of an AR 4 20 c 3 m ensemble with few members • High inter-member variability leads to very broad uncertainty in the ensemble mean • Not easy to distinguish the observations from the possible model estimates of the forced response Grey = annual means for 3 ensemble members Red = ensemble mean Black = 90% limits of estimated ens mean Blue = Observed SST from Had. SST 2 All data relative to 1900 -99 average © Crown copyright 2004 Page 5

AR 4 Ensembles Hadley Centre Multi-model ensemble • Super-ensemble (34) using data from 11 models with natural + anthro forcings and available SST • Narrower uncertainty on ensemble mean • Range is now a function of both internal climate variability and model and forcing differences • Represents a ‘best estimate’ of the forced response Red = super-ensemble mean (34 members) Black = 90% limits of estimated ens mean Blue = Observed SST from Had. SST 2 All data relative to 1900 -99 average © Crown copyright 2004 • Atlantic SST is inconsistent with the forced response for much of the last 150 years Page 6

AR 4 Ensembles Hadley Centre North Atlantic SST trends Linear change = Trend (K/year) x Period (year) OBS minus AR 4 OBS • Obs show clear multidecadal trend oscillations AR 4 AVG • Model trends are present but relatively weak • Difference therefore resembles obs • Obs trends almost always significantly different from the forced signal © Crown copyright 2004 Page 7

AR 4 Ensembles Hadley Centre A better AMO index? Removing a model-based estimate of the historical forced response as an improvement on • linear detrending Black = 90% limits of estimated ens mean Blue = Observed SST from Had. SST 2 All data relative to 1900 -99 average © Crown copyright 2004 • subtracting ‘background’ estimates based on global mean temperature Page 8

AR 4 Ensembles Hadley Centre Implications The inconsistency between observed North Atlantic SST and the ensemble estimate of the forced response suggests several possibilities: • The AMO is an internal mode • Models are inadequate to represent the effects of known forcings on climate* • The forcings used are incorrect or incomplete* *In the latter 2 cases, the errors would have to be specific to the Atlantic as the models perform well for the global mean © Crown copyright 2004 Page 9

Hadley Centre Had. CM 3 Control Simulation © Crown copyright 2004 Page 10

Control Simulation Hadley Centre 1400 Year Coupled Model Representation of the AMO 70 -180 Year band Observed AMO Pattern 0° 60° § Similar pattern and time scale to observed AMO fluctuations. 120° § Similar magnitude – North Atlantic low frequency (>45 year) standard deviation is 0. 10 K, 0. 14 K in observations. 180° § Observed AMO likely to be long -lived climate mode. © Crown copyright 2004 Page 11

Control Simulation Hadley Centre 1400 Year Had. CM 3 control simulation § Maximum overturning streamfunction at 30°N § Persistent band of variability between 70 -120 years § Compares with observed period of ~65 years (instrumental) and 40 -130 years (palaeo – Gray et al. 2004). © Crown copyright 2004 Page 12

Hadley Centre Large-scale temperatures © Crown copyright 2004 Page 13

Control Simulation Hadley Centre THC-Mean temperature cross-correlations Northern Hemisphere Southern Hemisphere 0. 09°C Sv-1 (0. 55) 0. 01°C Sv-1 (0. 13) Global 0. 05°C Sv-1 (0. 59) Suggests potential predictability of climate for several decades into the future © Crown copyright 2004 Page 14

Hadley Centre Mechanism © Crown copyright 2004 Page 15

Mechanism Hadley Centre Density anomalies related to the THC § Regress 0 -800 m averaged density onto THC § At THC peak, high densities in the mid-latitude and sub-polar ocean § Low densities in sub-tropical ocean § Density anomalies at 60°N mostly result from the contribution of salinity anomalies, rather than thermal anomalies. © Crown copyright 2004 From Vellinga and Wu (2004) Page 16

Mechanism Hadley Centre Coupled ocean-atmosphere interactions § Precipitation change associated with an ITCZ shift caused by SST anomalies supplies the tropical fresh water flux forcing § Coupled mechanism involving a delayed oceanic salinity feedback. © Crown copyright 2004 From Vellinga and Wu (2004) Page 17

Hadley Centre Climate Impacts © Crown copyright 2004 Page 18

Climate Impacts Hadley Centre North East Brazil Rainfall § NE Brazil has large multidecadal wet season (MAM) rainfall variability § Simulated ITCZ shifts north and away when N Atlantic warm (AMO+) drier NE Brazil § Simulated rainfall changes similar in size to observations © Crown copyright 2004 Page 19

Climate Impacts Hadley Centre Sahel Rainfall § African Sahel has large multidecadal rainfall variability § JJA simulated ITCZ shifts north when N Atlantic warm (AMO+) wetter Sahel § Simulated changes about one-third of those observed. § Compare ITCZ shifts with Caribbean palaeo salinity variations (Schmidt et al. , 2004). © Crown copyright 2004 Page 20

Climate Impacts Hadley Centre North Atlantic-European circulation response to the AMO • No winter NAO DJF MAM Simulated MSLP regression with AMO index JJA signal at any lead/lag • Anomalies typically smaller than observed multidecadal NAO change Simulated precipitation regression with AMO index • Summer/Autumn signal in Europe SON Broadest signal in summer and autumn © Crown copyright 2004 • Little sign of US summer signal (Sutton & Hodson, 2005) Page 21

Climate Impacts Hadley Centre Goldenberg et al. (2001) claim a link between the frequency of major Atlantic hurricane formation and AMO variations in North Atlantic SST. Suggest AMO affects vertical shear in the hurricane formation region via circulation changes Major Hurricanes Atlantic Hurricanes – the observed relationship 1944 1998 Emanuel (2005) suggests a more direct link between SST and the integrated intensity of storms. © Crown copyright 2004 Page 22

Climate Impacts Hadley Centre Atlantic Hurricanes – obs model comparisons NCEP/NCAR reanalysis 200 -850 h. Pa shear Had. CM 3 decadal AMO-shear correlation August-October (ASO) (1951 -60)-(1971 -80) Goldenberg main development area highlighted Had. CM 3 AMO index (red), versus mean Goldenberg area shear (black) Correlation of simulated main development area shear with SST Model supports an AMO relationship with hurricane development shear, but also shows an IPO relationship. AMO and IPO are uncorrelated (0. 06). © Crown copyright 2004 Page 23

Hadley Centre Conclusions © Crown copyright 2004 Page 24

Conclusions Hadley Centre § The AMO is inconsistent with an estimate of the response of Atlantic SST to natural and anthropogenic forcings from the AR 4 models § Either the AMO is internal or the models or their forcings are wrong § This analysis shows an increasing AMO in recent decades § A 1400 year Had. CM 3 control simulation suggests the AMO is a longlived coupled mode of climate variability associated with modern-day variations in the strength of the THC § Diagnosis of the simulated mechanism reveals a delayed salinity feedback via displacements of ITCZ rainfall caused by THC-related temperature anomalies § The simulation confirms AMO links with a range of important regional climate phenomena such as NE Brazil and Sahel rainfall, Atlantic Hurricane formation and European circulation. © Crown copyright 2004 Page 25

Hadley Centre Questions & Answers © Crown copyright 2004 Page 26

Hadley Centre Climate Impacts © Crown copyright 2004 Page 27

Hadley Centre Reconstruction and Forecast of the THC © Crown copyright 2004 Page 28

Reconstruction and forecast of the THC Hadley Centre Decadal Northern North Atlantic SST as a statistical predictor Use Had. CM 3 simulation to make a statistical model between SST-THC Use SST from Had. ISST dataset to reconstruct running decadal THC 1870 -2002 © Crown copyright 2004 Page 29

Reconstruction and forecast of the THC Hadley Centre THC Predictability § Look for points in the control simulation where the THC index rises through present day (decade 1997 -2003) reconstructed value (0. 63 Sv) § Track the subsequent THC evolution for each of these ‘analogues’ for 6 decades. § Use these to represent the next ~35 years (observed period shorter than in model). § Natural downturn in THC in next decade, to levels of 1960 s before 2030 (on average -0. 70 Sv) © Crown copyright 2004 Page 30

Motivation Hadley Centre Large scale SST patterns (after Folland et al. , 1999) Had. ISST Low-pass (> 13. 3 y) EOFs § 1911 -2002 § 40ºS - 70ºN § Projections 1870 -2002 © Crown copyright 2004 Page 31

Control Simulation Hadley Centre Coupled Model Representation of the AMO 70 -180 Year band 25 -125 Year band 0° 60° 120° 180° © Crown copyright 2004 Page 32

AR 4 Ensembles Hadley Centre Temperature (°C) North Atlantic mean temperature § North Atlantic (0°-80°W, 10°-70°N) § Annual mean SST § 4 member ensemble with Had. CM 3 (black) § Solar+Volc+Anthro. Stott et al. (2000) § Observed SST data from Had. ISST (blue) Trend (°C decade-1) Year Centre year of 30 -year trend © Crown copyright 2004 § Anomalies difficult without bias § 30 year trends § Uncertainty in ensemble mean trend § 90% limits (shaded) § Inconsistent (1900 -1930) to (1925 -1955) § Also (1945 -1975) to (1965 -1995) § Uncertainty still large with 4 members Page 33

Mechanism Hadley Centre Salinity leading density anomalies § 0 -800 m salinity contribution to density regressed onto zonal mean density at 60°N § First signs of positive salinity anomalies in subtropics 6 decades (half a period) before a THC peak © Crown copyright 2004 Page 34

Mechanism Hadley Centre Salinity budget analyses § 0 -800 m mean salinity driven density tendencies regressed onto the THC § In tropics (0 -35°N) density increases ~ 6 decades before the peak THC, induced by surface flux forcing and removed by transport § In mid-latitudes (35 -48°N) density increases ~ 4 decades before, caused by transport and removed by surface flux forcing § Sub-polar (48 -65°N) density increases ~ 2 decades before by transport © Crown copyright 2004 Page 35

Mechanism Hadley Centre Transport time scale § 100 year run with a unit tracer at the surface between 0 -15°N § Follow tracer concentrations averaged over 0 -800 m § Slow buildup in sub-polar ocean © Crown copyright 2004 Page 36