Hydrological impact of Middle Miocene Antarctic ice-free areas coupled to deep ocean temperatures

Oxygen isotopes from ocean sediments (δ18O) used to reconstruct past continental ice volumes additionally record deep water temperatures (DWTs). Traditionally, these are assumed to be coupled (ice-volume changes cause DWT changes). However, δ18O records during peak Middle Miocene warmth (~16–15 million years ago) document large rapid fluctuations (~1–1.5‰) difficult to explain as huge Antarctic ice sheet (AIS) volume changes. Here, using climate modelling and data comparisons, we show DWTs are coupled to AIS spatial extent, not volume, because Antarctic albedo changes modify the hydrological cycle, affecting Antarctic deep water production regions. We suggest the Middle Miocene AIS had retreated substantially from previous Oligocene maxima. The residual ice sheet varied spatially more rapidly on orbital timescales than previously thought, enabling large DWT swings (up to 4 °C). When Middle Miocene warmth terminated (~13 million years ago) and a continent-scale AIS had stabilized, further ice-volume changes were predominantly in height rather than extent, with little impact on DWT. Our findings imply a shift in ocean sensitivity to ice-sheet changes occurs when AIS retreat exposes previously ice-covered land; associated feedbacks could reduce the Earth system’s ability to maintain a large AIS. This demonstrates ice-sheet changes should be characterized not only by ice volume but also by spatial extent. Middle Miocene deep ocean temperatures were linked to Antarctic ice-sheet extent, not volume, due to distinct vegetation–climate feedbacks, according to coupled atmosphere–ocean–vegetation general circulation modelling.


Differing ice growth-DWT relationships
Some key observations from the Middle Miocene Antarctic cryosphere still require explanation. There is a long-held assumption that continental ice volume is inherently coupled to deep water temperatures (DWTs) because expanding ice sheets are assumed to cool high-latitude regions of deep convection 1,3,23 . We would therefore expect both the MCO and MMCT to be associated with DWT changes, yet the δ 18 O c record, combined with independent temperature reconstructions ( Fig. 1), reveals some challenging observations. During the MCO, both δ 18 O c and DWT were highly variable (70% of δ 18 O c variability is attributed to changes in DWT 24 ). During the MMCT, δ 18 O c was highly variable but DWT variations reduced in amplitude. During the MMCT glaciation, δ 18 O c was highly variable but DWT variations were small (70% of the δ 18 O c variability is attributed to changes in ice volume 24,25 ). After the MMCT, δ 18 O c and DWT were variable but less variable than during the MCO.
Oxygen isotopes from ocean sediments (δ 18 O) used to reconstruct past continental ice volumes additionally record deep water temperatures (DWTs). Traditionally, these are assumed to be coupled (ice-volume changes cause DWT changes). However, δ 18 O records during peak Middle Miocene warmth (~16-15 million years ago) document large rapid fluctuations (~1-1.5‰) difficult to explain as huge Antarctic ice sheet (AIS) volume changes. Here, using climate modelling and data comparisons, we show DWTs are coupled to AIS spatial extent, not volume, because Antarctic albedo changes modify the hydrological cycle, affecting Antarctic deep water production regions. We suggest the Middle Miocene AIS had retreated substantially from previous Oligocene maxima. The residual ice sheet varied spatially more rapidly on orbital timescales than previously thought, enabling large DWT swings (up to 4 °C). When Middle Miocene warmth terminated (~13 million years ago) and a continent-scale AIS had stabilized, further ice-volume changes were predominantly in height rather than extent, with little impact on DWT. Our findings imply a shift in ocean sensitivity to ice-sheet changes occurs when AIS retreat exposes previously ice-covered land; associated feedbacks could reduce the Earth system's ability to maintain a large AIS. This demonstrates ice-sheet changes should be characterized not only by ice volume but also by spatial extent.  49 ). Shading: MCO, yellow; MMCT, blue and grey; major ice-growth event (MMCT, blue). Vertical lines are indicative of the typical maximum δ 18 O c /DWT amplitudes during the MCO (4), MMCT before the major ice-growth event (3), MMCT major ice-growth event (2) and post-MMCT (1). Since the data cover different times and resolutions, these lines are not coincident in time for panels a-c.
Interpreting δ 18 O c is complicated because both temperature and the ambient seawater isotopic composition (δ 18 O sw ) are recorded. δ 18 O sw itself depends on global continental ice volume, the isotopic composition of this ice and localized salinity effects 26 . Paired independent reconstructions can isolate the temperature signal, and analysis of spatially distributed δ 18 O c records can reduce the salinity component. However, for the MMCT glaciation there remains the observation of a large ice increase but little DWT cooling, raising the question of why, if there is a strong coupling between ice volume and DWT as assumed 1,3,23 , did DWT vary so much less during the MMCT when ice-sheet growth was most rapid? Here we present new climate model results assessing the impact of ice-sheet size on DWT across the MCO and MMCT. Our results confirm the findings of a previous modelling study that DWT is insensitive to ice-sheet growth at the MMCT 27 . While the previous study explains the MMCT ice volume-DWT decoupling in terms of strong feedbacks in the coupled atmosphere-ocean-sea ice system 27 , our study provides further mechanistic understanding of the differing degrees of Middle Miocene ice volume-DWT coupling by proposing a key role for the hydrological cycle. We here advance our understanding of the paradigm by highlighting the important role not only of ice-sheet volume but also of spatial ice-sheet coverage in determining the DWT response to glaciation during the MMCT and the MCO.
We use a fully coupled atmosphere-ocean-vegetation general circulation model, HadCM3LB-M2.1aE 28 , configured with Middle Miocene palaeogeography (see Methods and Fig. 2). For our initial assessment using preindustrial CO 2 concentrations, we find that AIS expansion from ice-free (ICE FREE ) to the 22 m SLE regional-scale ice-sheet configuration (ICE PART 22m) and from ICE PART 22m to the 55 m SLE continental ice-sheet configuration (ICE FULL 55m) reduces DWT by ~0.5 °C for each step; thus, here, ice growth and DWT are coupled (Fig. 3a). However, AIS expansion from ICE FULL 55m to the 90 m SLE continental-scale ice-sheet configuration (ICE FULL 90m) does not cause further deep ocean cooling (by contrast, a slight temperature increase is seen); thus, here, ice growth and DWT are decoupled (ice-volume changes do not affect DWT).
We propose that coupling between ice-sheet volume and DWT occurs only until the ice sheet reaches the coast because the icealbedo feedback mechanism and vegetation-climate interactions invoke additional feedback processes identified here. To demonstrate it is ice-sheet spatial extent (rather than height/volume) that is coupled to DWT, we carry out a non-realistic sensitivity study imposing AIS configurations spanning extreme endmembers from ice-free to ice covered but keep ice volume constant. We assume the ice sheet is of 'skin thickness' (no effective change in elevation as compared with the ice-free state, nominally 1 m SLE when fully ice covered; 'ICE FULL 1m') and vary the ice extent longitudinally, latitudinally and topographically. We use preindustrial CO 2 concentrations throughout and conduct an additional high-CO 2 sensitivity test (~850 ppm; Fig. 4). Combining our results, we find a strong relationship between ice-free extent and DWT, with no evidence of nonlinearity (Fig. 3b).

The mechanism linking ice cover to DWTs
Our modelling results suggest summer 'ice-free' Antarctica (ICE FREE , Fig. 3, column 1) would be warm and wet because the land-sea thermal contrast drives monsoon winds, which transport moisture into the Antarctic continental interior from the Southern Ocean ( Fig. 5a-c). This moisture falls over the relatively warm continent as rain, not snow, during the summer months ( Fig. 5b) and over much of the continent during the winter months for the two highest CO 2 scenarios. Summer Antarctic temperatures and precipitation are similar to proxy reconstructions for a vegetated Antarctica 4,21,22,29-31 . A comprehensive model-data comparison (Supplementary Note) indicates peak CO 2 would need to be >850 ppm for a complete overlap with proxy reconstructions, in agreement with recent MCO reconstructions 32 . ICE FREE also results in the warmest, freshest deep ocean of all the simulations (Fig. 5d-e). Surface runoff from the active hydrologic cycle, being less saline and thus less dense than the seawater it drains into, forms a polar halocline at the surface. This halocline reduces ventilation of the deep ocean (Fig. 5d), weakening overturning. In our simulations, deep water is in all cases produced primarily in the Southern Ocean; thus, DWTs are determined by southern sinking regions. Antarctic Bottom Water (AABW) production never ceases completely in the model for any scenario (Supplementary Discussion B).
In ICE FULL 1m, ICE FULL 55m and ICE FULL 90m (Fig. 5, columns 3-5), cold surface temperatures near the ice sheet and the large increase in albedo cause localized radiative cooling of the air column and a reduction in vapour holding capacity. The land-sea thermal contrast reduces (Fig. 5a), and the summer monsoon system ceases to operate. Katabatic winds form as the cold dense air flows away from the elevated areas towards the coast (Fig. 5c). The interaction between the winds and sea ice is complex and depends on background CO 2 (Fig. 5b and Supplementary Discussion A). Reduced precipitation (Fig. 5b) and subsequent runoff reduce   Fig. 2 for details). CO 2 is 280 ppm in all cases unless specified and a modern orbit assumed; refer to Methods for more information. ocean stratification (Fig. 5d), permitting the cold surface waters to sink more freely from the continental shelf into the abyss (Fig. 5e). Increased AABW production invigorates ocean ventilation.
Empirical studies show a clear relationship between ice-sheet volume and spatial extent 33 , implying ice-sheet thickness is limited by spatial extent. Therefore, to grow vertically, an ice sheet must also    grow spatially (Supplementary Discussion C). After the ice sheet reaches the ocean, additional growth is necessarily predominantly vertical. Although a thickening ice sheet is accompanied by further cooling and drying of the air, this does not markedly affect runoff because precipitation has already been reduced to a low level and is falling as snow, not rain. Consequently, the surface ocean salinity does not change much, and hence, neither does deep water production, ocean ventilation or, crucially, DWT. This explains the ice volume-DWT decoupling among ICE FULL 1m, ICE FULL 55m and ICE FULL 90m. The global mean DWTs begin to rise slowly as ice volume (height) increases (in the absence of CO 2 changes) because the higher topography reduces the amount of summer low clouds around the Antarctic coastline by 10-15% (not shown), allowing more solar radiation to reach the surface and reducing sea ice, which locally causes greater absorption of solar radiation into the ocean. Our model has a fairly linear response to both a gradually increasing and a decreasing ice-sheet extent. However, we note in a dynamic ice-sheet model, ice melt in the decreasing ice-sheet scenario would result in additional surface runoff that would probably impact AABW production, at least temporarily, as demonstrated in studies of the modern AIS 34 .

Sensitivity to atmospheric Co 2 and orbit
Our model does not have an interactive AIS, so the response of the ice sheet to CO 2 forcing is not included, and our study is limited to a single model with mid-range CO 2 sensitivity 35 . However, our results show that atmospheric CO 2 has a much smaller impact on the hydrological regime than ice-sheet configuration (Supplementary Discussion A). CO 2 impacts sea-ice extent and sea surface temperatures, which in turn affect the deeper layers via vertical mixing, in a process that is complex and nonlinear (Supplementary Discussion A).
For the MCO, the most-recent CO 2 record suggests an average range of concentrations between 630 and 470 ppm 32 . Using the relationship between CO 2 and DWT calculated from our simulations, we infer a consequent ~0.8 °C mean temperature change in the 2-to 3-km-deep layer in the Southern Hemisphere (Fig. 6), which is about 80% of the ~1.0 °C impact from increasing ice extent (Fig. 3, ICE FULL 1m-ICE FREE ) in the same layer. This provides a picture of the average DWT changes. The site-specific temperature changes (of 2-4 °C, Fig. 1), will depend also on local dynamics. At Site 761, our simulations estimate a contribution of 0.5 °C from CO 2 variations compared with a 0.9-1.9 °C contribution from ice-extent changes, and at Site 1171, the contribution from CO 2 is between 0.5 and 0.6 °C compared with 1.0-1.5 °C from ice extent (Supplementary Discussion D,E). Our results show that CO 2 changes alone cannot explain the observed DWT range at the MCO, and moreover, for both the mean layer and the specific sites, our model suggests that ice extent had a larger impact on DWT than did CO 2 . For the MMCT glaciation, the most-recent CO 2 reconstructions show at most a 170 ppm reduction from ~570 to 400 ppm 32 , for which we infer from our ice-covered model simulations a temperature drop of 0.5-0.8 °C at the two sites (Supplementary Discussion E). This is consistent with the reconstructions (Fig. 1) if we assume Antarctica was ice covered before the MMCT glaciation (that is, little DWT change occurred as a result of increasing ice-sheet extent). In the absence of ice-sheet changes, we find a minimal effect of orbital configuration on DWT (Supplementary Discussion F).

From thin and vulnerable to thick and established
We introduce the hydrological cycle as a crucial mechanism mediating the link between the DWT and the ice spatial extent (rather than absolute volume), thus explaining the different degrees of coupling between ice-sheet changes and DWT during the MMCT and MCO.
Our new results lead us to propose that DWT varied by up to 4 °C during the MCO because the spatial extent of ice and vegetation rapidly altered. Taken together with existing δ 18 O c , temperature, vegetation and CO 2 reconstructions, this implies the AIS had retreated substantially during the MCO, when average CO 2 concentrations were probably 470-630 ppm, reaching 780-1,100 ppm at times 32 . Previous work clearly demonstrates the dynamic behaviour of a small AIS when driven by CO 2 changes combined with orbital forcing 7 . How far exactly the ice sheet retreated during these warmest intervals, however, is unknown. Ice-sheet modelling suggests a retreat exposing 60-70% of the Antarctic land surface is consistent with the palaeorecord 13 . Other work concludes a retreat even greater than this 36,37 , perhaps even ice-free 11 . The evidence for vegetation, including trees, growing on the continent throughout the MCO 22,29 implies both warm and wet conditions, and it is suggested the moisture supply derived from the Southern Ocean 29 . To achieve this, our results indicate a greater reduction of ice is needed than the ICE PART 22m scenario ice-sheet extent because the Wilkes Land winds are directed landward in ICE FREE but seaward for ICE PART 22m (Fig. 3c). We suggest these monsoon moisture-carrying winds induced by spatial ice retreat could provide an explanation for major ice advance onto the continental shelf in the Ross Sea 38,39 during the MCO occurring at the same time as open water and woody vegetation in the Wilkes Land 22 (Supplementary Discussion G).
We further infer DWT varied so much less during the MMCT when the AIS volume was growing rapidly because it had already extended to cover most of the continent before the major ice-growth event, in agreement with previous findings 27 . Thus, the ice sheet subsequently increased mainly in thickness, not area, and so DWTs were largely unaffected because, without the additional ice-albedo feedback, changes to the hydrological cycle were much smaller. Post-MMCT (label 1 in Fig. 1), both δ 18 O c and DWT are variable, but less so than during the MCO. The exact degree of coupling, and its mechanism, needs to be explored in a model set-up that includes marine-based ice sheets and ice shelves, not included in this study. However, the physical limits on seawater temperatures (-1.8 °C) will set the lower boundary on possible temperature changes as climate cools.
Interpretation of our results leads us to support a highly dynamic MCO AIS, and state, alongside CO 2 , it was changes in ice-sheet area and proximity to the coast, not volume, that were of key importance for global DWTs. This fundamentally changes the way we should characterize ice-sheet changes and how we must view the long-term δ 18 O c records spanning greenhouse-icehouse transitions.   In the absence of independent temperature proxies, it must not be assumed that DWTs scale with ice-volume changes. While we do not propose the MCO Antarctica was ever completely ice-free, our results demonstrate any spatial retreat of the AIS can increase precipitation, causing associated warming of the deep ocean-changes perhaps having the ability both to accelerate ice melt of ice shelves and glaciers through hydrofracturing from increased precipitation falling into crevasses 40,41 and to accelerate ice melt of marine-based subglacial basins 34,41 . Although the temperature changes resulting from changing ice-sheet extent are similar to those resulting from CO 2 changes, our study does not include feedbacks to the carbon cycle or to the ice sheet itself, and therefore the significance of our results could be greater than indicated here. Our non-realistic sensitivity studies using only a skin thickness of ice demonstrate the importance of both surface albedo and roughness for a hydrologic control on DWT evolution. It is therefore possible that our mechanism could operate in areas even without complete ice loss if these two factors change markedly, for example, in regions of debris-covered glaciers, rock glaciers, vegetation-covered rock glaciers and 'glacier mice' , which all increase in the context of retreating ice glaciers [42][43][44][45] , in regions of accumulating dark particles (dust and soot) 46 and in regions of glacier algae, which bloom in supraglacial meltwater 47 .

online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41561-021-00745-w.