It is demonstrated that current level models, both with and without assimilation, generate too shallow an overturning in the North Atlantic (just like their predecessors) because they do not reproduce the descent of plumes of cold water from the Greenland and Norwegian Seas. Consequently, the prediction of decadal change from one such model reported by C. Wunsch and P. Heimbach is queried.
In a recent paper, Wunsch and Heimbach (2006, hereafter WH) have examined the results from a global ocean 1° model at 26°N in the Atlantic and found decadal changes in the meridional overturning circulation there. The model was constrained by ocean data over a 12-yr period, namely climatological hydrography, temperature and salinity soundings, and sea surface height measurements, while surface fluxes were obtained from the National Centers for Environmental Prediction– National Center for Atmospheric Research (NCEP–NCAR) reanalysis project. The adjoint method was utilized to harmonize model and observations. An unconstrained model run from a modified initial state employing modified surface forcing then furnished the principal results of the paper. At 26°N in the decade 1993–2004, the northward transport above 1200 m weakened slightly while the core of southward transport near 2000 m strengthened, as did the abyssal northward flow. Overall, the results are reminiscent of the observations reported by Bryden et al. (2005a), although there are significant differences. Observations and models agree that the upper (above 1200 m) northward flow weakened and the upper component of southward transport strengthened. In the lower component of the southward transport (below 3000 m) the observed change is large but absent in the model. Indeed, this component is almost missing from the model, and consequently we are writing this note to express concerns we have about the credibility of the model results.
It is widely agreed that the earliest ocean models of the North Atlantic generated a meridional overturning circulation that was too shallow (see, e.g., Willebrand et al. 2001). Unfortunately, the problem has persisted to this day. Figure 1 shows the transport per meter as a function of depth on 26°N for the average of five hydrographic sections and three models. The observations are for the years 1957, 1981, 1992, 1998, and 2004 and have been treated in the manner described by Bryden et al. (2005a) and averaged. The model outputs are from WH, OCCAM, and Smith et al. (2000). The latter two models were added to illustrate a common problem for level models. The WH results are averaged for the years 1993–2004 and the OCCAM results, which are from a 1/12° version (Lee et al. 2007) of an earlier global 1/4° forward model (Marsh et al. 2005), are averaged for the years 1991–98. They show no trend over this 8-yr period, similar in that respect to the 1/4° model. The Smith et al. results are from a 1/10° parallel ocean program (POP) model of the North Atlantic only, averaged for the years 1991–94. The Smith et al. results were selected because, out of almost all forward models known to the authors, they showed realistic behavior of the North Atlantic Current south and east of the Grand Banks; but they, too, like both OCCAM versions and WH, failed to show the strong observed southward penetration below 3000 m.
Readers may well question the “observations” because the values shown are the averages over the 6800-km section and are a combination of measurement and inference. Simply, the basinwide interior baroclinic transport, calculated from adjacent CTD station pairs, is added to the measured Ekman transport and direct measurements of the western boundary current. The first step is the estimation of the baroclinic circulation by selecting suitable reference velocity levels. Across the western boundary, long-term current meter measurements suggest that a level of 1000 dbar is suitable, dividing the northward-flowing Antilles Current from the deep western boundary current (DWBC) below (Bryden et al. 2005b). Offshore, where the distribution of dissolved oxygen marks the edge of the western boundary, the reference level is set to be 3200 dbar, ensuring a weak circulation in the deep eastern basin (Saunders 1982; Lavin et al. 1988). Finally, any imbalance in the net transport is then set to zero by distributing it uniformly across the section. The last step is the crucial one; the argument is made that if the bottom depth is uniform, then the location of the inferred imbalance does not influence the result when averaged over the section. But the section, even outside of the western boundary current region, is not of uniform depth; there is the mid-Atlantic Ridge and eastern boundary. Could the mass imbalance (which is found to imply southward flow for each of the five sections) be concentrated in these regions so as to enhance the flow down to 3000 m and weaken or eliminate the southward flow between 3000 and 5000 m? On the mid-Atlantic Ridge, almost no topography reaches shallower than 3000 m on 26°N, so the missing flow could not occur there. On the eastern continental slope, if all the flow was concentrated there, a large southward current system [10–20 Sv (1 Sv ≡ 106 m3 s−1)] would be implied. But there is no observational evidence for this whatsoever. Further evidence for the magnitude of the transport below 3000 m is found in the work of Ganachaud (2003), who employed global hydrographic data collected during the World Ocean Circulation Experiment (WOCE). He estimated layer transports using the method of inverse analysis, and on the Atlantic 24°N section (A5) his values are shown in Table 1 (column 2). In column 1 of the table are the estimates given in Bryden et al. (2005a), and these are closely mirrored by the Ganachaud values. Other columns in Table 1 present transports from the three models of Fig. 1, again showing how severely they underestimate the flow below 3000 m.
Observations of southward transport below 3000 m are also found in direct measurements in the DWBC. At 26°N, the water mass involved is generally termed Lower North Atlantic Deep Water (LNADW), but farther north it is termed Denmark Strait Overflow Water and Iceland Scotland Overflow Water, indicating the primary sources. The measured transport below 3000 m is 5 Sv in a southeast direction out of the Labrador Sea at 53°N (Fischer et al. 2004), 5 Sv in a southward direction east of the Grand Banks at 43°N (Schott et al. 2004), and 10 Sv directed southward on this section at 26°N (Bryden et al. 2005b). All of these measurements are multiyear averages. We use the OCCAM 1/12° version for comparison, and the corresponding results are <0.1 Sv northwest, <0.1 Sv northward, and 3.8 Sv southward. Clearly, the deepest flow is entirely missing or too weak at these three locations, whereas at shallower levels between 1000 and 3000 m the cold water outflow is adequately represented. We believe these results are broadly typical of level models because they fail to deal realistically with the northern overflows, the flow through the Denmark Strait, and that across the Iceland–Scotland Ridge, and they mix away the overflowing cold water at far too shallow a depth. In the OCCAM 1/12° model, a transport of approximately 3.0 Sv passes annually through the Denmark Strait (1.5 Sv each above and below 300 m). By the latitude of the tip of Greenland (Cape Farewell) the overflow is found only between the surface and 600 m instead of on the bottom (3000 m). In the Irminger Sea, at depths below 1500 m, the water actually warms from the climatological input, and this suggests that the Iceland–Scotland Overflow Water also does not reach the correct depth. So here the following question is raised: If the modeled northern overflows do not penetrate deep enough, do they then influence (contaminate?) the Upper North Atlantic Deep Water (UNADW), which arrives at 24°N in the depth interval 1–3 km?
With the high vertical and spatial resolution of OCCAM, 66 levels and 9 km in both directions in the overflow region, our hopes for improved representation of the overflows have been severely dashed. Between the Denmark Strait sill of the model and 25 km farther southwest where the depth increased by 250 m (so the downstream slope is 10−2), the potential temperature of the flow increased from 0–1° to 3–4°C. The results of Winton et al. (1998) and Legg et al. (2006) show us why this might be so. The former authors show that for a level model to accurately reproduce a 100 m thick overflow down a slope of 10−2 requires a resolution of 30–50 m in the vertical and 3–5 km in the horizontal. OCCAM falls short of the spatial resolution somewhat, but the severity of its failure is perhaps surprising. However, both the Winton and Legg results are for a two-dimensional slope in which the overflow descends down the slope. Strait topography is, of course, three-dimensional and perhaps better represented by a two- dimensional slope across which the overflow descends.
Some will argue that it is unnecessary to have perfect modeling of the meridional circulation to obtain significant results with the best current models. This conclusion is plausible (but unproven) for changes at shallow depth but seems unlikely to us for changes even as shallow as 1000 m, since at this depth the model overflows may unrealistically influence water-mass properties. In a model with a realistic deep transport component, additional decadal trends can occur, which are missing in the WH model. These trends might have either sign, augmenting, compensating, or reducing any trends at shallower levels.
To summarize: our ground for questioning the credibility of the WH 1° model to simulate decadal climate change in the North Atlantic resides in the inability of the model to reproduce the formation of the northern waters in a realistic manner—possibly producing too much upper NADW and certainly too little lower NADW—and thereby an imperfect meridional circulation.
Corresponding author address: Stuart A. Cunningham, National Oceanographic Centre, University of Southampton, Waterfront Campus, European Way, Southampton SO14 3ZH, United Kingdom. Email: firstname.lastname@example.org