Divergence in climate model projections of Arctic Atlantification

The Arctic Ocean is strongly stratified by salinity in the uppermost layers. This stratification is a key attribute of the region as it acts as an effective barrier for the vertical exchanges of Atlantic Water heat, nutrients, and CO2 between intermediate depths and the surface of the Eurasian and Amerasian basins (EB and AB). Observations show that from 1970 to 2017, the stratification in the AB has strengthened, whereas, in parts of the EB, the stratification has weakened. The strengthening in the AB is linked to freshening and deepening of the halocline. In the EB, the weakened stratification is associated with salinification and shoaling of the halocline (Atlantification). Simulations from a suite of CMIP6 models project that, under a strong greenhouse-gas forcing scenario (ssp585), the overall surface freshening and warming continue in both basins, but there is a divergence in hydrographic trends in certain regions. Within the AB, there is agreement among the models that the upper layers will become more stratified. However, within the EB, models diverge regarding future stratification. This is due to different balances between trends at the surface and trends at depth, related to Fram Strait fluxes. The divergence affects projections of future state of Arctic sea ice, as models with the strongest Atlantification project the strongest decline in sea ice volume in the EB. From these simulations, one could conclude that Atlantificaton will not spread eastward into the AB; however, we need to improve models to simulate tendencies in a more delicately stratified EB correctly. 14


31
Much of the present-day central Arctic Ocean is a so-called beta ocean -it is strongly stratified by 32 salinity, unlike subtropical seas where the upper layers are stratified by temperature (Nansen 1902; 33 Carmack 2007). Over the last few decades, the Arctic region has experienced surface warming at 34 more than twice the global rate (Cohen et al. 2020;IPCC 2021), and an intensive loss of Arctic sea Ocean, which includes changes in both the physical, geochemical, and biological components. 80 The hydrographic changes related to Atlantification and Pacification are expressed regionally 81 and have opposite effects on stratification ( Fig. 1  increased river runoff (Haine 2020). The freshwater flux due to melting sea ice has been a large 108 contributor to the recent freshening, but is likely to decrease into the future, and become relatively 109 5 small by the second half of the 21st century, as less ice is available to melt (Shu et al. 2018 (Nummelin et al. 2016) have examined the potential effects of increased river runoff, and 112 they find that the Arctic stratification will increase and that the freshwater has a larger effect than 113 elevated wind-driven mixing (Davis et al. 2016). However, these studies do not consider other 114 freshwater sources, the regional aspect, or the opposing effects of Atlantification. For example, 115 using a single climate model (HiGEM), Lique et al. (2018) showed that under an extreme global 116 warming scenario, the stratification in this model is strongly enhanced in the AB but reduced in 117 the EB.

118
It is well known that climate models experience crucial biases in simulated Arctic hydrography.

143
This manuscript is structured as follows: We start by describing the observational and model 144 data used in this study and present a new diagnostic used to evaluate integral changes in Arctic

145
Ocean stratification (Section 2). We then compare observed and simulated stratification in recent 146 decades (Section 3a) before we investigate the future trends (Section 3b and 3c) and finally discuss 147 the mechanisms responsible for these changes (Section 3d) and the impacts on sea ice (Section 3e). 148 We focus particularly on the role of advective contra local processes and finish with a summary of 149 our findings and a discussion on the broader implications of our work (Section 4). and halocline stability, and has been made available through the Arctic Data Center (?, , reference 157 to appear latest during copy-editing). The temporal and spatial coverage for the data used in 158 this study is shown in Fig. A1. Unfortunately, historical observations of the Arctic Ocean are 159 generally sparse and have limited spatial coverage. Especially in the 1990s, data coverage is 160 not good, and in general, there have been few winter campaigns in the central basins. However, 161 autonomous Ice-Tethered Profilers (ITP), crewed ice-drift stations, and some ship-based campaigns 162 ensure a relatively good seasonal coverage in later decades (Fig. A2). The bulk of historical data 163 was gathered to construct the climatological atlases of the Arctic Ocean by Gorshkov (1980), 164 Treshnikov (1985), and Timokhov and Tanis (1997). Before 1980 most observations used Nansen  (Table 1). 184 We evaluated the last 45 years of the historical run, i.e., January 1970 -December 2014, and the  potential temperature "thetao", and sea ice concentration "siconc" and thickness "sivol/sithick" or 203 sea ice mass "simass" (Table 1) The primary objective of this paper is to quantify trends in stratification. Traditionally, strat-209 ification has been quantified using the Brunt-Väisälä buoyancy frequency 2 = −( / 0 ) / , 210 where is potential density, 0 is a reference density, and is the gravitational acceleration. This ). Traditionally, the definition of AW is based on temperature, salinity, or density values.

217
However, since we expect these properties to be biased in the models, we instead chose to define 218 the AW core as the depth of the temperature maximum below 100 m. When we further refer to 219 AW properties, we thus refer to the properties at the depth of the AW core. According to Heuzé where ℎ is the depth of the lower boundary of the halocline and ℎ is the potential density 232 at that lower boundary of the halocline.

233
In observations, the lower boundary of the halocline is usually determined using a density ratio such density ratio is defined as where is the thermal expansion coefficient, is the haline contraction coefficient, Θ is the to properly define the halocline using the same criteria as in the observations. Manually deriving 244 model-specific definitions is not ideal either, as the biases might vary over time. We, therefore, 245 find that the uncertainty of properly defining the "correct" halocline in CMIP6 models based on 246 Equation (1) is too high and have chosen to investigate Arctic stratification in CMIP6 models using 247 an indicator whose definition is less dependent on defining a halocline. 248 We therefore propose a new indicator of stratification strength, Δ ( ). First, we define the 249 potential energy of the water column following Tailleux (2009) as: where is a chosen depth level. We then look at the difference in potential energy between the 251 simulated stratified water column and a fully mixed water column, which reflects the energy needed 252 to fully mix the water column from the surface to a given depth: Here, ( ) is the potential energy of a completely mixed water column with a mean 254 temperature and salinity down to depth . Δ ( ) thus represents the potential energy energy 255 stored in stratification, and as long as is well below the typical halocline depth, APE and Δ 256 should capture similar changes and be equally good indicators of stratification strength. However, 257 Δ ( ) is preferred in models as its definition is independent of temperature and salinity gradients.

258
Throughout the paper we will refer to Δ as stratification strength or potential energy stored in 259 stratification. A comparison of APE and Δ is given in Fig. A3. We use = 300 m (well below  The freshwater input from river runoff is expected to continue to increase, but due to the prevailing 355 wind patterns in the region, most of this will accumulate in the Beaufort Gyre region and not stay

384
We quantify and summarize the historical and future trends for each region in Fig. 6. The   assume that most changes at the surface are driven by local processes (e.g., sea ice melt/growth, Arctic (below 200 m). This is partly attributed to an absence of ventilation, and as a result, the 417 properties of the Arctic AW layer are closely linked to the inflows. 418 We start by detailing the evolution of AW core temperature and salinity in the four different 423 regions. As expected, with continued global warming, the AW temperature is projected to increase 424 in all regions by all models (Fig. 7). Thick lines in Fig. 7 represent the multimodel mean  Table 2. We note that AW core properties are calculated based 428 on each model's AW core depth (details in Section 2c), which varies substantially from model to Water core temperature (a) and Atlantic Water core salinity (b) from the CMIP6 models listed in Table 1 we speculate that the Arctic freshening is partly remotely driven.

463
It is important to remember that it is not only the water mass properties but also the depth and   (Fig. A5). In the EB, on the other hand, many models project a freshening, but some 478 project a surface salinification (Fig. A5). Some of the models that project a surface salinification We compare two models, GFDL-CM4 and NorESM2-LM, which project distinctly opposite 519 changes in stratification in the EB (Fig. 6). In Fig. 9 we present the temporal development of  for all models are shown in Fig. 12.

564
In the EB, most models agree on a negative density trend below 200 m, but above they diverge.

570
Here we also see large discrepancies in how quickly the density trends increase or decrease with 571 depth, thus the extent of the water column that is changed. Again, this is related to the SML depth, 572 which varies and changes differently over time (Fig. 12). In the Beaufort Gyre region, the models 573 have a very similar shape, but already in the Chukchi Sea, we see that models start to diverge, with  and this is likely related to the mean sea ice state of the models or other important processes. 616 For example, MPI-ESM1-2-HR has a very weak decline in sea ice volume compared to its strong 617 degree of Atlantification in the Eastern EB, but since it finished the historical run with a low sea ice 618 thickness compared to the other models (not shown), it simply cannot have a large volume trend. 619 For reference we have therefore provided a table of mean sea ice volume at the beginning and in the 620 middle of the ssp585 scenario (Table A2)