We further analyze the models to explore the potential causes of the AAIW bias. In particular, we focus on the potential roles of the surface climate bias, the subsurface biases through the inter-basin exchange and from the North Atlantic.
The role of surface climate bias
We first examine if the AAIW bias is caused by the bias in the surface climate in the coupled models. AAIW density surface outcrops near Antarctic polar front zone (APFZ). Sverdrup et al. (1942) firstly proposed that AAIW is formed by circumpolar subduction or along-isopycnal transport along the APFZ, supported by the circumpolar presence of AAIW. If this is true in model, AAIW bias may be tracked back to the surface climate bias far south. However, the model net precipitation shows a wet bias near the APFZ, as seen in the ensemble mean in Fig. 5, which causes a fresh APFZ in most models (Fig. 3). For many models, the fresh bias dominates almost the entire South Atlantic and the maximum fresh bias is located in the tropical South Atlantic, because of the southward shift in the Atlantic ITCZ in the model, which is a part of the tropical bias in climate models (Liu et al. 2014; Harrison et al. 2014). The fresh bias in the mid- and high latitude, is distinguished from the tropical bias, reflecting the excessive rainfall in the Southern Ocean storm track. The freshening surface water between 30°S and 40°S (Fig. 5d) in South Atlantic will subduct as the South Atlantic Subtropical Mode Water (SASTMW) and ventilate the thermocline water northward, leading a fresh thermocline bias at 34°S (Liu et al. 2015). Important here is that this fresh surface and thermocline bias in the upper 500 m across the South Atlantic and Southern Ocean is in contrast to the overall salty bias at the intermediate depth in the South Atlantic. Therefore, the AAIW bias cannot be caused by the Southern Ocean surface climate bias.
To further detect the role of the common atmospheric deficiency and the resultant surface climate bias on setting the subsurface and intermediate water bias, we analyze the results from CORE-II experiments (Griffies et al. 2009; Danabasoglu et al. 2014). In CORE-II, the atmospheric deficiency and therefore the ocean surface climate bias is absent. However, the vertical salinity (temperature) distribution of Atlantic in CORE-II (Fig. 6) shows that the intermediate salty (warm) bias still exists although the surface bias is diminished. This further confirms that the AAIW bias is not caused mainly by the biases in surface salinity and temperature. This point is also consistent with the IPCC AR4 results (Liu et al. 2014) which shows the same fresh bias in the AAIW regardless of the surface flux adjustment in the model.
Moreover, the AAIW bias cannot even be related to the ocean upper 300 m. We compare a series of PB experiments with no PB experiment (control, CIAF) under IAF forcing using POP2. The CIAF shows a consistent salinity and temperature bias pattern (Fig. 8b) similar to CORE-II results (Fig. 6) and its vertical salinity profile along 34°S resembles the CMIP5 multi-model mean in intermediate depth (Fig. 1, POP2). In our PB experiments, global ocean salinity and temperature from surface to 100, 200 and 300 m were restored (with a restoring time of 90 days) towards their respective monthly climatology in the observation, but with no substantial improvement in intermediate bias (not shown), which implies it is not the surface and mixed layer (one less than 300 m) oceanic process that causes AAIW bias.
The role of inter-basin exchange
The bias in the Atlantic AAIW can be affected locally within the Atlantic sector, it can also be caused remotely by inter-basin exchanges with the upstream Pacific in the Drake Passage (McCartney 1977; Talley 1996) and the downstream Indian Ocean through the leakage of the Agulhas Current (e.g. Rintoul 1991; Gordon et al. 1992; Weijer et al. 1999; Beal et al. 2011). McCartney (1977) suggested that AAIW is primarily a by-product of SAMW, formed in a deep convective mixed layer in the southeast Pacific Ocean off southern Chile, feeding into the Pacific Ocean via the Drake Passage to the southwest Atlantic. Therefore the bias in modeling the mixed layer in southeast Pacific could lead to the bias of South Atlantic AAIW. Here, we find that the mixed layer in the southeast Pacific in CMIP5 models is indeed much shallower than the observation (not shown). This is consistent with Danabasoglu et al. (2012) which indicated that the winter mixed layer around the ACC region seems to be shallower than the observation in both the coupled model CCSM4 and ocean alone model POP2. Gordon et al. (1992) also reported a significant interoceanic exchange of the intermediate water between the South Atlantic and Indian oceans through the South Atlantic Current, Agulhas Retroflection and Benguela Current. Furthermore, several studies revealed that the overestimation of the inter-basin exchange could cause too salty South Atlantic and too fresh Indian Ocean due to the lack of spatial resolution to properly represent the inertial dynamics of the Agulhas retroflection and ring shedding process (Weijer and van Sebille 2014).
To investigate the role of inter-basin exchange, here in our sensitivity experiments, ocean temperature and salinity are restored in two regions from the surface to 5000 m depth: the downstream region of 20°–50°S and 20°–100°E, and the upstream region of 40°–70°S and 50°–110°W. The former includes part of Indian Ocean and Agulhas Current system while the latter includes the southeastern Pacific and Drake Passage. As such, the inter-basin areas are restored towards the observation completely. In the following we refer to the experiment as PB_SIOD.
The PB_SIOD result shows that the inter-basin exchange bias plays a minor role in the South Atlantic AAIW bias. There is a slight decrease in the salinity of intermediate water and a significant increase in the salinity of the thermocline water in South Atlantic (Fig. 8c). Both the salt intermediate bias around 40°S and fresh thermocline bias from 30°S–50°S of the South Atlantic in the upper 500 m (SA500m, short for this fresh bias region) in CIAF is corrected mainly by the southeastern Pacific Ocean through Drake Passage while the Indian ocean has no significant contribution, as demonstrated in further PB experiment for upstream region and downstream region, respectively (not shown). The bias in SA500m may contribute to the AMOC stability bias, a point to be returned later.
The role of North Atlantic
Since the salty bias in the AAIW appears to intensify all the way to the intermediate depth of the subtropical North Atlantic, one may also speculate that the bias may also be contributed by the North Atlantic formation of intermediate water. In observation, the AAIW salinity minimum terminates in the North Atlantic between 20°–25°N (Fig. 7a), where it meets the Mediterranean Water in the eastern Atlantic (Talley 1996). But in models, the confluence of these two watermasses locates more south, implying an enhanced Mediterranean Water penetration (Figs. 2, 7b). The multi-model mean bias across Atlantic in late winter shows that both intermediate salty and warm bias have their maxima located in the North Atlantic ~15°N where the salty bias and warm bias compensate each other, leading to the near zero deviation in potential density (Fig. 7c, d). The bias calculation is also applied to each single model, in which the bias distribution is similar to the multi-model mean results (Fig. 3). With the depth getting deeper and deeper, the bias reaches further and further southward (not shown). It seems that the salty and warm bias originates from the eastern subtropical North Atlantic, especially near the Strait of Gibraltar and spread to the south. Moreover, the combination of fresh bias in the Mediterranean Sea with salty bias in the Atlantic could indicate too much exchange across Gibraltar Strait.
Consistent with the CMIP5 results, CIAF shows corresponding salty and warm bias maxima in northeastern Atlantic outside Gibraltar Strait. It should be mentioned here that there exists another bias maxima locating in tropical Atlantic upper 300 m which may due to the surface forcing error, but it has little to do with intermediate water bias as demonstrated in surface PB experiments described in Sect. 4.1. To clarify the role of North Atlantic, we carry out another PB experiment similar to PB_SIOD except that the restoring region is located from 20° to 40°N and from 30° to 10°W, namely PB_Gib hereafter, such that the water from the Mediterranean Sea is restored towards the observation. Surprisingly, there seems to be no relation between the North and South Atlantic salty bias in POP2. Figure 8d plots the difference between PB_Gib and CIAF. The freshening intermediate water in PB_Gib is confined locally in the subtropical North Atlantic, with no clearly south of equator. This muted effect of North Atlantic may be caused by the dominant northward flow in the AAIW across the equator in the subsurface North Brazil Current (not shown), which prevents the southward penetration of North Atlantic intermediate water. Therefore, the salty and warm bias maximum in the northeastern Atlantic outside Gibraltar Strait is more likely a regional phenomenon caused by excessive exchange between Mediterranean Sea and Atlantic resulting in the salty subtropical North Atlantic, with little impact on the bias in the South Atlantic AAIW.