Model climate and scenario integrations
For 20C, we show some hydrographic sections and stations in the model. Figure 2a and b show the 1860–2000 average potential temperature and salinity at WOCE-CLIVAR section A17. The large-scale water masses like NADW, Antarctic Bottom Water, or Antarctic Intermediate Water are represented reasonably, although local differences of their properties from observations can be considerable. Also shown is the meridional overturning stream function for the same period (Fig. 2c), indicating a maximum of 22 Sv (1 Sv = 106 m3/s) at 45° N and a standard deviation of 1 Sv in respect to annual mean values for the years 1860–2000. The Atlantic outflow is 17 Sv at 30° S, and the ocean heat transport is 1.15 PW at 20° N. Additional aspects of the model performance are discussed elsewhere (van Oldenborgh et al. 2005; Jungclaus et al. 2006a). Figure 3 shows simulated and observed potential temperatures at OWS S. The figure indicates a general agreement between observations and model but a too warm intermediate to deep ocean.
Simulated annual mean global surface temperature for different ensemble members are shown in Fig. 4a, together with the corresponding observations by Jones et al. (2001). Note that temperature curves are normalized with respect to the period 1961–1990. The ensemble of twentieth century integrations agrees well with the observations in terms of variability and the warming trend over the last decades of the twentieth century. Note that the continuation of the 20C experiments into the twenty-first century with emissions fixed at year 2000 values indicate a very moderate warming for the twenty-first century (Fig. 4a).
The time evolution of the Atlantic maximum meridional overturning stream function is shown in Fig. 4b (annual mean values). The seasonal MOC strength has a standard deviation of appr. 3 Sv in our simulation, but in the following, we concentrate on the annual mean values. The meridional overturning for A1B shows a weakening of 5 Sv from year 2000 to 2100 due to warming and freshening of high-latitude surface water. We find, though the large-scale MOC structure seen in Fig. 2c is not affected, considerable changes in the deep water formation regions and their relative role (not shown). It is interesting to note that the run 20C shows a very moderate MOC reduction for the twenty-first century, but an in-depth discussion is not within the scope of this paper. For our further analyses, we concentrate on the SRES A1B scenario (IPCC 2007), which is a scenario with a moderate climate response (Fig. 4a) and with an overturning response similar to that of A2 and B1 in the twenty-first century (cf. Fig. 4b).
The simulated WOCE-CLIVAR hydrographic sections, as well as temperature and salinity profiles at the OWS sites, are regressed against the time series of annual mean MOC strength for the period 1860–2000. We take the MOC maximum rather than the value at a fixed latitude (results related to 30° N give similar results). Since the simulated annual mean MOC has a standard deviation of 1 Sv, the normalized overturning index (MOI) and anomalous overturning strength (in units of sievert) can be associated with each other. The regression patterns are used to obtain a MOI signature pattern. The temperatures to MOI regressions for the period 1860–2000 are calculated but only shown for OWS S (Fig. 5a) and OWS E (Fig. 6a). The later analysis will show that the information from the other sites cannot be used to estimate MOC trends.
The temperature regression at OWS S has three minima at 500, 1,500, and 2,500 m, indicating cold conditions when there is a strong MOC (Fig. 5a). The pattern is close to the baroclinic first mode as described by Cunningham and Alderson (2007). The changes in the 300–800 m are related to the thermoclince, the second peak between 900 to 1750 m is related to intermediate water, 1,750 to 2,500 m is related to upper North Atlantic Deep Water, and below that is lower Atlantic Deep Water. A decrease in MOC is related to temperature increase (Fig. 5a), which is dominated by isopycnal heave. The temperature trend analysis (Figs. 5b, 6b) for the period 2000–2100, when the MOC weakens, shows a warming in the upper 2,000 m. The temperature variability is relatively small at mid-depth (700–2,200 m), which could be used for an optimized signal-to-noise analysis (not shown). In the upper 1,000 m, temperature variability is highly dominated by interannual-to-decadal variability (not shown).
The regression of temperature θ and salinity S to MOC strength is evaluated at the cross-section A17 for the period 1860–2000 (Fig. 7a,c). A strong MOC is associated with a relatively warm salty surface ocean between 50° and 80° N, a relatively warm salty downward MOC branch between 45° and 60° N in 3,000-m depth, and a moderately cool fresh layer at 100 to 2,500-m depth in the tropical and subtropical Atlantic Ocean. The latter temperature is due to inflow of cold intermediate and deep water masses from the north. The same argument is seen for the mid-latitude OWS at these depths.
The section A17 temperature trend analysis (Fig. 7b) shows a general warming, as expected from the increasing atmosphere greenhouse gases. The warming trend at intermediate depths, which is accompanied by a cooling in the downward branch of the North Atlantic Deep Water, is associated with Atlantic MOC weakening during this period. The A17 salinity trend analysis (Fig. 7d) is determined by enhanced evaporation in the subtropics in conjunction with high-latitude freshening and projects not directly onto the MOC trend.
Regression and trend analysis for the zonal section A05 are shown in Fig. 8. During times of high MOC, an anomalous zonal temperature and salinity gradient in the upper 1,000 m is detected: cooling and freshening in the west with a warming and increase in salinity in the east. The east–west asymmetry in salinity is also consistent with the anomalous surface salinity (not shown). The anomalous zonal hydrography is directly linked to the overturning circulation. The trend pattern (Fig. 8b,d) stems from an anomalous increase in the east relative to the west.
Constructing the overturning trends
In the model, the hydrographic evolution for θ and S is converted to the MOI using the regression patterns r shown in Figs. 5a, 6a, 8a–c, and 9a,b. To give an example for OWS E, the potential temperature evolution θ(z,t) is multiplied for each level with r
−1(z) and summed over all z levels to get the evolution in the anomalous MOI. This procedure is called construction of the MOI through the hydrographic evolution (in the model).
Since r is based on period 1860–2000, it is not surprising that the derived meridional overturning indices (the colored lines) are similar to the simulated MOI (shown as the black line in Fig. 10). A comparison of the simulated MOI with the MOI as derived from stations and sections shows that some stations and sections can be used to predict the MOC trend for the period 2000–2100. For the OWS stations, the OWS E and OWS S regression of potential temperature provide reasonable approximations to construct the MOC trend (Fig. 10a), whereas the salinity-based MOI construction at all OWS sites do not project onto MOC trends (Fig. 10b). For the WOCE-CLIVAR sections, the temperature at the A17 section can be used as a fingerprint to construct MOC trends (Fig. 10c). In contrast to the meridional salinity structure along A17, the zonal subtropical section A05 in salinity is associated with the MOC dynamics (Fig. 10d). The strong trend in subtropical salinity (Fig. 8d) is consistent with the horizontal salinity distribution. For all trends of the constructed MOI, we refer to Fig. 9. In this figure, we see that some of the stations even do not show the right sign (a MOC weakening). This shows that this information cannot be used for the (modeled) construction of MOC trend.
In order to quantify the potential of the different sections to predict the MOC trend for the period 2000 to 2100 AD, we plot the error between the constructed and the simulated MOI trend (Fig. 11). The value of one would mean that the constructed MOI trend overestimates the overturning decline by 1 Sv/100 years. Furthermore, the standard deviation ratio between simulated MOI and constructed MOI is displayed. A value less than 1 mean that the standard deviation of the constructed MOI is smaller than the simulated MOI.
The analysis reveals that the temperatures at OWS E and OWS S, as well as temperatures at section A17, provide useful indirect measure of large-scale changes of deep circulation, with a standard deviation even smaller than the simulated MOC. The reason for the smaller standard deviation could be that the MOC is affected by the atmospheric noise (e.g. the North Atlantic Oscillation), whereas the subsurface quantities see already a reduced noise integrated by oceanic processes (e.g., through vertical mixing, advection). The atmospheric forcing can contribute to decadal MOC variations via Labrador Sea Water (Haak 2004) and mask longer-term MOC trends (Delworth and Dixon 2000; Vellinga and Wood 2004).
Via the described regression technique, observed temperature information from OWS E and OWS S (Curry 2002) are used to derive a 1900–2100 MOI time series through projection of hydrographic changes onto the model-derived MOI fingerprints of Figs. 5a and 6a. Figure 12 indicates small non-significant trends in the observational-based MOI for the last 80 years, which is also in line with the model simulation (Fig. 2b). Similarly, the simulation-based MOI at OWS E and OWS S show almost no MOC trend for the last 100 years but a decrease of about 5 Sv over the next 100 years. For the last decades, there is a moderate warming at mid-depth possibly associated with a small MOC decrease (Fig. 12). From the modeled hydrographic data, we estimate the time at which the uncertainty cross the two standard deviations-threshold is reached as the year 2070.