Climate Dynamics Manuscript

Uncertainty in 21st Century Projections of the Atlantic Meridional Overturning Circulation in CMIP3 and CMIP5 models

A. Reintges 1 (corresponding author), T. Martin 1, M. Latif 1,2, N. S. Keenlyside 3,4

1GEOMAR Helmholtz Centre for Ocean Research Kiel, Düsternbrooker Weg 20, 24105 Kiel, Germany

2Kiel University, Christian-Albrechts-Platz 4, 24118 Kiel, Germany

3Geophysical Institute and Bjerknes Centre for Climate Research, University of Bergen, Allégaten 70, 5020 Bergen, Norway

4Nansen Environmental and Remote Sensing Center, Thormøhlens Gate 47, 5006Bergen, Norway

e-mail:

telephone: +49 431 600-4007

fax: +49 431 600-4052

Supplementary equations:

The inter-model correlations listed in tables 3 and 4 are computed following these equations:

.

XA and XB are the variables that are to be correlated; s denotes the simulation (historical, RCP4.5) and t denotes the time period (1970-2000, 2070-2100).

Supplementary Figures:

Fig. S1 Year-to-year differences in the annual AMOC index at 30°N in the CMIP5 ensemble (a) during the period 1970-2000 of the historical scenario. (b) during the period 2070-2100 of theRCP4.5 and the RCP8.5 scenario.

Fig. S2 Mean AMOC index at 30°N during the reference period 1970-2000 and its corresponding depthfor (a) CMIP3and (b) CMIP5.Each symbol represents one model; the line connects the symbols for the 20C3M (historical) run averaged over 1970-2000 with the SRES A1B (RCP4.5) run averaged over 2070-2100. The depth at which the AMOC index is taken varies from model to model. This model-dependence is much larger in CMIP3 than in CMIP5. There is a weak tendency for models with a stronger overturning to have the streamfuntion’s maximum (AMOC index) depth at deeper levels.

Fig. S3Vertical profiles of the meridional overturning streamfunction at 30°N during 1970-2000 and 2070-2100 averaged over all models of(a) CMIP3 (mean of the scenarios A2, A1B, B1) and (b) CMIP5mean of the scenarios RCP4.5 and RCP8.5. The standard deviation between the ensemble members are shown by the bars at selected depths. The North Atlantic Deep Water (NADW) cell in the models is shallower than in observations (depth range from surface to zero-crossing of the stream function in the figure). Nevertheless, the overall structure seems reasonable and the depth of the maximum in the overturning at 30°N is close to 17.5 Sv observed at 26°N.

Fig. S4Number of vertical ocean layers of each CMIP5 model we used versus the subpolar gyre (SPG) index (a) mean during 1970-2000 and (b) change by 2070-2100. The vertical resolution of the ocean model component seems to be crucial for the mean state and the mean change of the subpolar gyre index. Models with higher resolution tend to have a stronger gyre and simulate a stronger weakening during the 21st century.

Fig. S5AMOC uncertainty results with a polynomial order of 2. (a) Mean signal (decadal means of the polynomial fit averaged over all models), (b) uncertainties, (c) signal-to-noise ratio based on the 90%-confidence level.

Fig. S6Same as Fig. S5 but with a polynomial order of 3.

Fig. S7Same as Fig. S5 but with a polynomial order of 4.

Fig. S8Same as Fig. S5 but with a polynomial order of 5.

Fig. S9 Sources of the uncertainties in three CMIP5 models (CanESM2, CCSM4, and MPI-ESM-LR) projecting the AMOC at 30°N until 2100. The scenarios RP4.5 and RCP8.5 are used. Data are averaged to decadal means. (a)-(c) based on the method described in our ‘Statistical method’ section; (d)-(f) based on the method presented in Yip et al. (2011). The AMOC time series is presented as decadal means of the polynomial fit (a) and as decadal means of the raw time series (d). Individual uncertainties are shown in (b) and (e). The signal-to-noise ratio based on the 90%-confidence limit is shown in (c) and (f).

Fig. S10Decadal residuals of the AMOC index at 30°N in five realizations with differing initial condition; (a) for the CMIP3 model CGCM3.1(T47) in the A1B scenario;(b) for the CMIP3 model MRI-CGC;2.3.2ain the A1B scenario;(c) for the CMIP5 model CanESM2 in the RCP8.5 scenario;(d) for the CMIP5 model CCSM4 in the RCP8.5 scenario. Red lines depict the residuals to the polynomial fit (see ε in our equation 1). Blue lines depict the residuals to the run-average of the raw time series.

Fig. S11The AMOC index at 30°N is shown (as anomalies referenced to 1970-2000). The thin blue lines indicate the CMIP5 time series of the RCP4.5 scenario and the thin red lines those of the RCP8.5 scenario. The thick solid black line is the mean signal as defined in equation 8. The thick dashed black line is the mean signal computed from the raw time series instead of the polynomial fit.

Fig. S12Sources of the uncertainties in projections of the AMOC until 2100. This figure differs from Fig. 2 because only models are used that simulate an AMOC strength at 26°N that are within the range 17.5 ±2.5 Sv during the reference period (1970-2000). (a-c): CMIP3 (SRES A1B, A2 and B1). (d-f) CMIP5 (RCP4.5 and RCP8.5). (a) and (d): AMOC long-term changes of the individual models at 30°N; the 10-year running mean is presented (the climate mean of the reference period 1970-2000 has been removed). (b) and (e): individual absolute uncertainties of the AMOC projections (Sv2) at 30°N. (c) and (f): signal-to-noise ratio based on the 90%-confidence limit for the AMOC changes at 30°N (red) and 48°N (blue)

1