Weddell Sea Control of Ocean Temperature Variability on the Western Antarctic Peninsula

Recent ice loss on the western Antarctic Peninsula has been driven by warming ocean waters on the continental shelf. However, due to the short observational record, our understanding of the dynamics and variability in this region remains poor. High‐resolution ocean model simulations show that the temperature variability along the western Antarctic Peninsula is controlled by the rate of dense water formation in the Weddell Sea. Passive tracer advection reveals connectivity between the Weddell Sea and the coastline of the western Antarctic Peninsula and Bellingshausen Sea. During multi‐year periods of weak Weddell dense water formation, dense overflow transport in the Weddell Sea decreases, while the transport of cold water around the tip of the Antarctic Peninsula strengthens, driving a temperature decrease of 0.4°C along the western Antarctic Peninsula. This mechanism implies that western Antarctic Peninsula coastal ocean temperature may cool in the future if Weddell Dense Shelf Water production slows down.

. Anomalous increases in heat content on the shelf have also been linked to local storm forcing driving large warm water intrusions across the shelf break (Martinson et al., 2008). Lastly, intrusions of cold Weddell-sourced waters from Bransfield Strait (see Figure 1a for place names), southward along the coast of the western Antarctic Peninsula, have been identified to occur during negative Southern Annular Mode years (Wang et al., 2022).
Variability may also occur on longer (decadal) timescales, but little is known about the potential mechanisms for such possible low-frequency variability in this region. Here, we identify a new mechanism of decadal-scale temperature variability, linked to a remote driver in the Weddell Sea.

Materials and Methods
We use ACCESS-OM2-01 (Kiss et al., 2020), a global ocean-sea ice model, with 0.1° horizontal resolution and 75 vertical z* levels. The model does not include explicitly simulated tides or ice shelf cavities. The model is forced with prescribed JRA55-do atmospheric forcing (Tsujino et al., 2018). ACCESS-OM2-01 has a good representation of observed water masses around Antarctica (Moorman et al., 2020), shelf/slope processes including Dense Shelf Water (DSW) formation and overflows into the abyss Solodoch et al., 2022) and the Antarctic Slope Current (Huneke et al., 2022). The simulated ocean temperature has a similar distribution of features compared with instrumented seal observations (Treasure et al., 2018) along the western Antarctic Peninsula and in the Bellingshausen Sea (Figures 1a and 1b; Text S2 in Supporting Information S1). The main features captured by the model include the warm intrusions of Circumpolar Deep Water with temperature >1.5°C along the shelf break, cold Weddell Sea waters at the northern tip of the peninsula and cool waters in a narrow coastal pathway along the western Antarctic Peninsula shelf and Bellingshausen Sea. The observed temperature is warmer across most of the shelf than the model ( Figure S1a in Supporting Information S1), possibly because eddy heat fluxes across the shelf break are not completely resolved in the model (Graham et al., 2016). Transects and a map of the simulated coastal current are shown in Figures S2 and S4a in Supporting Information S1.
We use four simulations in this study: two baseline simulations and two freshwater perturbations. The two baseline simulations are: (a) an interannual-forced ("IAF") simulation spanning 1958-2018, and (b) a repeat year forced control run ("CONTROL"), which is forced by the repeat year May 1990 to April 1991, chosen as a neutral year (Stewart et al., 2020). The CONTROL is spun up for 250 years before the analysis period. Further model configuration details are given in Supporting Information S1. Two perturbation simulations are branched off from the CONTROL: DSW+, in which the freshwater input in the south-west Weddell Sea is halved, and DSW−, in which the freshwater input is doubled. The region of freshwater input change is west of 30°W and south of 71°S (along the cyan/orange lines in Figures 2b and 2c). The magnitude of these perturbations are realistic in a future context: meltwater input from Antarctica is expected to approximately double by 2,050 under high emissions (Golledge et al., 2019;Li et al., 2023).

Co-Variability With the Weddell Sea
The ocean temperature simulated in the ACCESS-OM2-01 IAF simulation, at 206 m depth on the western Antarctic Peninsula continental shelf, warms at an average rate of 0.17°C/decade over the period 1963-2018 (see Figure S1b in Supporting Information S1 for the non-detrended temperature time series). However, overlaid on the long-term warming, there is large decadal variability (detrended blue line in Figure 1c). The detrended time series has a range of 1.0°C between the warmest and coldest years (equivalent to 60 years of the modeled trend), with clear fluctuations on a decadal time scale. Note that we choose a depth of 206 m here, because the temperature variability is largest. Although the western Antarctic Peninsula region has been relatively well observed since the early 1990s (Martinson et al., 2008), the time scale of the variability we find in the IAF simulation (∼20 years) is approximately equal to the length of the observational record. It is therefore difficult to identify decadal variability in the existing observational record, given the additional long-term warming trend over this time.
Somewhat unexpectedly, the simulated ocean temperature variability along the western Antarctic Peninsula is tightly correlated with the variability in DSW formation in the Weddell Sea, averaged over the preceding 4 years (orange line in Figure 1c, see Text S3 in Supporting Information S1 for DSW formation definition). There is no significant trend in Weddell Sea DSW formation. The correlation coefficient between the detrended western Antarctic Peninsula shelf temperature at 206 m depth and the Weddell Sea DSW formation averaged over the preceding 4 years is r = 0.80 (significant at the 95% level, see Text S4 in Supporting Information S1 for statistical analysis methods). The correlation peaks at 0 years lag, noting however that the Weddell Sea DSW formation has been averaged over the preceding 4 years relative to the western Antarctic Peninsula shelf temperature variability. With no rolling average applied to the Weddell Sea dense water formation, the correlation peaks at r = 0.59 (significant at the 99% level) with the Weddell Sea DSW formation leading the western Antarctic Peninsula ocean temperature by 2 years. This reduces to r = 0.35 with no lag.
Other studies suggest that Weddell Sea dense water formation variability, as observed and simulated by this model, may be controlled by local wind, which in turn is linked to larger scale modes of variability such as the Southern Annular Mode and the Amundsen Sea Low (Hattermann et al., 2021;Schmidt et al., 2023). In this paper, we do not focus on the mechanisms driving the decadal variability in dense water formation, and instead focus on the dynamical link between variability in dense water formation and the temperature on the western Antarctic Peninsula.
The correlation between Weddell Sea DSW formation and western Antarctic Peninsula ocean temperature on its own does not imply causation, and could result from independent responses to the variability in regional atmospheric forcing. However, the time lag in the correlation suggests that there may be a dynamical connection whereby changes in the Weddell Sea drive subsequent changes on the western Antarctic Peninsula shelf via an advective pathway. To investigate this possibility of a dynamical connection between the two regions, in the following section we perturb the rate of DSW formation in the Weddell Sea to quantify the impact on the flow of shelf waters westward around the tip of the Antarctic Peninsula.

Connectivity Between the Weddell Sea and West Antarctica
There is speculation that there may be westward connectivity via the Antarctic Coastal Current between the north-west Weddell Sea and the central western Antarctic Peninsula and beyond (Heywood et al., 2004). The westward flowing Antarctic Coastal Current has been observed on the eastern side of the peninsula in the northwest Weddell Sea (Heywood et al., 2004;Thompson et al., 2009). The Antarctic Coastal Current has also been observed at multiple locations in West Antarctica, including at the tip of the Antarctic Peninsula (heading westward from the Weddell Sea into Bransfield Strait (Heywood et al., 2004)), at several discrete locations along the western Antarctic Peninsula (Moffat et al., 2008;Savidge & Amft, 2009), and along a continuous coastal pathway in the Bellingshausen Sea (Schubert et al., 2021). However, the continuity of the coastal current from the east side to the west side of the peninsula, and along the northern part of the western Antarctic Peninsula remains poorly constrained (Moffat & Meredith, 2018). In a recent review, Moffat and Meredith (2018) suggested that the Antarctic Coastal Current observed in West Antarctica may originate in the central western Antarctic Peninsula near Anvers Island (∼64°S), implying that there may be no or limited connectivity around the tip of the Antarctic Peninsula beyond Bransfield Strait.
Previous modeling studies have simulated a coastal current that originates near the tip of the Antarctic Peninsula and flows continuously to the Amundsen Sea, forced by a combination of local winds and buoyancy forcing (Holland et al., 2010;Wang et al., 2022). Modeling studies have also shown limited evidence that coastal connectivity westward from the Weddell Sea may be dynamically important (Moorman et al., 2020;Wang et al., 2022). However, the extent of connectivity remains an open question, which we address here using simulated passive tracers released in the Weddell Sea. Specifically, passive tracer is released at the surface in the dense water formation region in the south-west Weddell Sea continuously during the 12 yr CONTROL simulation with repeat year forcing (hatched area in Figure 2a; see Methods and Supporting Information S1). A fraction of this tracer is entrained into DSW and is exported northwards into the abyssal ocean, while a clear signal of tracer is also advected westwards to the Amundsen Sea in a continuous upper-ocean pathway along the coast of West Antarctica ( Figure 2a). This connectivity around the tip of the Antarctic peninsula is consistent with a recent Lagrangian analysis of along-shelf connectivity in ACCESS-OM2-01 (Dawson et al., 2023).
To investigate whether variability in the Weddell Sea dense water formation rate dynamically drives changes in the westward connectivity, we next perturb the dense water formation in the Weddell Sea. Two perturbation simulations are branched off from the CONTROL repeat year forced simulation. In the "DSW+" simulation, the freshwater input in the south-west Weddell Sea is halved along the orange line in Figure 2c. In the "DSW−" simulation, the freshwater input is doubled along the same stretch of coastline (cyan line in Figure 2b). The CONTROL freshwater input in this region is 0.01 Sv. Figure S3 in Supporting Information S1 shows the response of the surface water mass transformation in the perturbation simulations, which roughly spans the DSW formation variability seen in the IAF simulation (Figure 1c). Figure 2 shows the passive tracer at two different ocean depths: offshore of the 1,000 m isobath (thick black line) passive tracer is shown at the ocean bottom to reveal changes in DSW export, while inshore of the 1,000 m isobath passive tracer is depth averaged to highlight the Antarctic Coastal Current pathway. As expected, in the DSW− simulation, the offshore export of passive tracer at the ocean bottom decreases (Figure 2b), while in the DSW+ simulation, this export is enhanced (Figure 2c). The advection of passive tracer from the Weddell Sea toward the Amundsen Sea in West Antarctica also responds to the DSW perturbations, with an inverse relationship between the amount of passive tracer exported offshore in bottom waters and the amount of tracer flowing around the Antarctic Peninsula on the continental shelf toward the Amundsen Sea. In the DSW− simulation, there is weaker bottom water transport northward along the continental slope (between the 1,000 and 3,000 m isobaths) in the Weddell Sea, in conjunction with increased transport of Weddell Sea tracer along the coast of West Antarctica (Figure 2b). The reverse situation is seen in the DSW+ simulation: bottom water transport increases at the expense of passive tracer transport around the Antarctic Peninsula toward the Amundsen Sea (Figure 2c).
The first arrival of passive tracer from the south-west Weddell Sea to the western Antarctic Peninsula region occurs just before 2 years after release, and continues to increase over the 10 yr simulation (Figure 2d). Passive tracer concentration in the western Antarctic Peninsula region increases by 59% in the DSW− simulation, and decreases by 45% in the DSW+ simulation, averaged over years 2-4. Note that we do not expect an exactly symmetric response between the two perturbations, because the DSW formation response (and therefore the connectivity response) to the freshwater input change is not necessarily linear. The volume transport in the Antarctic Coastal Current along both the western and eastern Antarctic Peninsula also increases in the DSW− simulation, concurrent with the connectivity increase around the tip of the Antarctic Peninsula ( Figure S4 in Supporting Information S1). At 67°S on the western Antarctic Peninsula, the transport of the coastal current increases by ∼20%-25% after the second year of the DSW− simulation, and decreases by ∼15%-20% in the DSW+ simulation. Thus, modifying DSW formation rates in the perturbation simulations acts to directly modify both passive tracer and volume transport around the Antarctic Peninsula. . Freshwater forcing is altered in the perturbation simulations along the cyan/orange lines along the coastal margin. In (a-c), thick and thin black contours show the 1,000and 3,000 m isobaths respectively. Inshore of the 1,000 m isobath, passive tracer is depth averaged, while offshore of the 1,000 m isobath, passive tracer is shown at the bottom of the ocean to highlight DSW export changes. (d) Time series of depth averaged passive tracer, averaged over the western Antarctic Peninsula shelf (blue region shown in Figure 1d). A 12 months rolling mean has been applied to remove the seasonal cycle.

Forced Response of Western Antarctic Peninsula Ocean Temperature and Salinity
The western Antarctic Peninsula ocean cools as a result of the increased connectivity from the Weddell Sea in the DSW− simulation (Figure 3a). The cooling signal is concentrated along the coast and is aligned with the Antarctic Coastal Current pathway, consistent with the passive tracer transport shown in Figure 2. A composite average of the 20 coldest years of the IAF simulation shows a similar spatial distribution of cooling (Figure 3b). In the model, the coastal waters on the western Antarctic Peninsula are influenced by two distinct source waters: (a) warm and salty Circumpolar Deep Water that intrudes locally across the shelf break and (b) cold and fresh Weddell Sea waters that are transported along the coastal current pathway. When DSW formation is decreased, the transport from the Weddell Sea increases, resulting in an increased influence of cold and fresh Weddell Sea waters on the western Antarctic Peninsula shelf (dashed lines in Figure 3c), relative to the influence of the warm and salty Circumpolar Deep Water intrusions. The temperature and salinity response is roughly symmetric for an increase or decrease in DSW formation (Figure 3c, Figure S5 in Supporting Information S1). These results are consistent with a recent modeling study (Wang et al., 2022) that showed coastal intrusions of cold Weddell Sea waters controlled the temperature variability of the northern part (∼64-65°S) of the western Antarctic Peninsula during 2008-2009. The cooling at the coast is maximum at a depth of ∼180 m in the central and northern parts of the western Antarctic Peninsula (i.e., north of 68°S), and the cooling extends down to a depth of ∼400 m at the coast (Figure 3d). In the Bellingshausen Sea, the maximum cooling occurs slightly deeper at ∼230-300 m (not shown). Although the passive tracer advection continues into the Amundsen Sea (Figure 2a), the influence of Weddell Sea waters on the temperature in the Amundsen Sea is weak relative to other local forcing mechanisms. Beneath the mixed layer, the three-dimensional spatial structure of the temperature anomaly is aligned with the spatial structure of the passive tracer anomaly (green contours in Figure 3d). This is consistent with the hypothesis that an advective mechanism, and not a locally driven vertical shift of the stratification, controls the thermal response along the western Antarctic Peninsula sector. While the temperature anomaly is relatively shallow, it has been previously shown that the basal melt rates of ice shelves on the western Antarctic Peninsula are sensitive to upper ocean and coastal processes (Cook et al., 2016;Padman et al., 2012).
In the DSW-simulation, the density surfaces along the western Antarctic Peninsula and Bellingshausen Sea deepen at the coast relative to the CONTROL simulation due to the freshening of coastal waters (Figure 3d). The change in stratification is consistent with the increased transport of the Antarctic Coastal Current, which advects more cold, fresh and less dense Weddell Sea waters along the western Antarctic Peninsula coast. A largely symmetric response occurs in the DSW+ simulation ( Figure S5 in Supporting Information S1).

Discussion and Conclusions
The simulations presented here suggest that multi-year variations in the formation rate of DSW in the Weddell Sea directly alter the coastal ocean temperature along the continental shelf of the western Antarctic Peninsula and Bellingshausen Sea, via the advection mechanism depicted in the schematic in Figure 4. In the years following periods of strong Weddell dense water production (left panel), transport between the Weddell Sea and the western Antarctic Peninsula shelf decreases, resulting in coastal warming due to the increased influence of local warm Circumpolar Deep Water intrusions, relative to the influence of cold and fresh Weddell Sea waters. Conversely, following years of weak Weddell Sea dense water production (right panel), westward transport around the tip of the Antarctic Peninsula increases, which advects more cold and fresh waters along the Antarctic Coastal Current pathway.
The near-coastal wind forcing at the north-west tip of the peninsula may also play a dynamical role in controlling temperature variability on the western Antarctic Peninsula (Wang et al., 2022). We find that when the winds are anomalously north-eastward at the tip of the peninsula in the IAF simulation, the temperature on the western Antarctic Peninsula is warmer ( Figure S6 in Supporting Information S1) and the coastal current has anomalously low transport (not shown), consistent with the mechanism of Wang et al. (2022). Thus it is possible that in the IAF simulation, the wind forcing at the tip of the peninsula and the changes in Weddell DSW formation are working in concert to drive the temperature variability on the western Antarctic Peninsula. However, our perturbation simulations, which have no interannual variation in wind stress, clearly show that changes in the Weddell DSW formation rate can drive large temperature variability on the western Antarctic Peninsula, even in the absence of any interannual change in wind stress. The connectivity we find between the Weddell Sea and the western Antarctic Peninsula coastal margins has implications for accurate modeling of the Antarctic continental shelf. Inadequate model representation of DSW formation leads to model biases around the Antarctic continental shelf (Purich & England, 2021) and in the abyssal ocean (Heuzé et al., 2013). The mechanism we report here implies that inaccurate simulation of dense water formation will result in model biases downstream (westward) along on the continental shelf. In particular, models with dense water formation that is too weak in the Weddell Sea may have a cold bias along the coast of the western Antarctic Peninsula, due to the enhanced along-shelf advection of cold Weddell Sea waters. Indeed, it is likely that the model used in this study has too weak dense water overflows in the Weddell Sea due to inadequate resolution, and this may be the cause of the cold bias at the northern tip of the Antarctic Peninsula (Figures 1a and 1b).
A caveat of our results is that the model does not have ice shelf cavities and the meltwater input is inserted into the upper ocean. It is unclear whether resolving ice shelf cavities and inserting meltwater at depth would significantly alter the coastal current dynamics and connectivity between the Weddell Sea and the western Antarctic Peninsula (see Figure 10 in Mathiot et al. (2017)). Nevertheless, we believe it is still useful to understand the mechanism we describe here, because it is likely to also occur in other models. We note that the mechanism occurs in our simulations both due to meltwater-driven changes in DSW formation (in the DSW+, DSW− simulations, Figure 3) and due to atmospheric (likely wind) driven changes in DSW formation (IAF simulation, Figure 1c).
Finally our results have implications for future ice shelf melt along the western Antarctic Peninsula and in the Bellingshausen Sea. Ice shelves in these sectors have shallower grounding lines and ice drafts than those elsewhere in Antarctica (Adusumilli et al., 2018;Fretwell et al., 2013), and as a result have been shown to be sensitive to upper ocean and coastal processes (Cook et al., 2016;Padman et al., 2012). In the future, dense water production in the Weddell Sea may slow down due to the additional freshwater input from melting ice around the continent (Hellmer et al., 2017;Li et al., 2023;Moorman et al., 2020). The behavior identified from our model simulations suggest that a decrease in upstream DSW formation may result in cooling along the western Antarctic Peninsula and Bellingshausen Sea coastal margins. Such ocean cooling would provide a transient negative feedback to ice shelf melt, thereby slowing the sea level rise contribution arising from ice melt in these sectors.