We employ the Australian Community Climate and Earth System Simulator-Earth System Model version 1.5 (ACCESS-ESM1.5), a comprehensive Earth system model developed for the Coupled Model Intercomparison Project Phase 6 (CMIP6) . The atmospheric model is the UK Met Office Unified Model version 7.3 (UM7.3) with a horizontal resolution of 1.875° × 1.25° and 38 vertical levels. The land surface is represented by the Community Atmosphere Biosphere Land Exchange (CABLE) model version 2.4 , which includes biogeochemical processes through the CASA-CNP module . The ocean component consists of the NOAA/GFDL Modular Ocean Model version 5 (MOM5) , featuring a horizontal resolution of 1° × 1°, with the meridional spacing refined to 0.4° in the Southern Ocean and 0.33° near the equator, and 50 vertical levels. Furthermore, sea ice processes are simulated using the Los Alamos National Laboratory Consortium Model for Sea Ice Development (LANL CICE) version 4.1 . The coupling between the atmosphere, ocean and sea ice components is achieved through the Ocean Atmosphere Sea Ice Soil–Model Coupling Toolkit (OASIS-MCT) . The ocean carbon cycle is modeled using the World Ocean Model of Biogeochemistry and Trophic dynamics (WOMBAT) model for ocean biogeochemistry which includes a nutrient-phytoplankton-zooplankton-detritus (NPZD) scheme .

This model has previously been used to simulate the Last Interglacial (lig127k) time slice (127 ka) as part of the Tier 1 Paleoclimate Model Intercomparison Project 4 (PMIP4)-CMIP6 experiments , MIS9e (∼ 336–321 ka; ), and the mid-Holocene (6 ka; ).

As a reference for our paleoclimate simulations, we use the 1000-year pre-industrial (PI) control simulation (piControl) based on the year 1850 CE by . We analyse the average of the last 100 years of this simulation (Fig. , black line). Orbital parameters and GHG concentrations for the PI configuration are provided in Table .

To investigate the climate around 49 ka, the boundary conditions are changed in a step-wise manner using the ACCESS-ESM1.5 model. The first experiment, 49ka-co, simulates a baseline climate with orbital parameters and GHG concentrations being set to 49 ka values (Table ). 49ka-co is run for 618 years (Fig. , cyan line).

The model uses prescribed Plant Functional Types (PFTs) that remain fixed during each individual simulation. We modified the PFT distributions in selected simulations to reflect the altered boundary conditions. Therefore, in a second step, 49ka-alb, we modify the land surface properties by incorporating only the ice-sheet extent (no topography yet) from corresponding to 52.5 ka. We adopted the maximal scenario with an ice-capped Hudson Bay (Fig. d). This time period was chosen because the simulated ice sheet at 52.5 ka most closely matches the sea-level estimates for 48–52 ka, based on global seawater δ18O stack records . The NH ice sheets at this time period include two separate ice sheets over the North American continent – Laurentide in the east and Cordilleran in the west – as well as the Greenland Ice Sheet and Fennoscandian and Eurasian ice sheets (Fig. d). Their extent and height are smaller than during the LGM (not shown), which featured a single, extensive LIS over North America and continuous ice sheet coverage from Eurasia to Scandinavia, fully covering the Barents Sea. The Antarctic Ice Sheet (AIS) at 52.5 ka also has a smaller extent compared to that of the LGM, especially over the Ross and Weddell Seas. At 52.5 ka, the LIS was up to 3 km thick, while the AIS was up to 3.6 km thick. In the present day, permanent ice cover in the NH is restricted to Greenland (Fig. c).

In the 49ka-alb experiment, additional modifications are introduced at different stages (Fig. , orange line). The vegetation cover is modified to reflect ice coverage by removing vegetation in ice-covered regions and converting needle leaf evergreen forests in areas between the Cordilleran and LIS to bare soil (Fig. a, b). These modifications align with the vegetation reconstruction from for 52 ka. Additionally, C₃ crops are removed, and replaced by the next two most dominant vegetation types in the corresponding grid cell. For example, in India, C₃ crop values are replaced by the broadleaf deciduous type and C₄ grass. In Europe and eastern North America, they are replaced by the broadleaf deciduous type and C₃ grass, while they are replaced by the broadleaf deciduous type in China. 49ka-alb (Fig. , orange line) starts from year 297 of 49ka-co, and was run for a total of 272 years with the above mentioned albedo and vegetation changes (Table ). As can be seen in Fig. , 49ka-alb is not in complete equilibrium. To this 49ka-alb experiment, we add a gradual increase in surface salinity from year 547 (Fig. , magenta line), to account for the 44 m sea level drop relative to PI, based on the prescribed ice sheet, achieving a global average increase of +0.33 psu from PI levels. At year 729, the Bering Strait is closed.

Our final experiment, referred to as 49ka-full (Fig. , blue line, Table ), starts at year 824, when the topography is adjusted to reflect conditions at 52.5 ka, based on , including an increase in ice sheet height and changes to the land-sea mask, such as the closure of the Bering Strait (done in the previous step), Hudson Bay, the Sahul and Sunda shelves, and the Tasman Gateway (Fig. ). Nearest-neighbour interpolation is applied to determine vegetation types on newly exposed land. The river runoff is also adjusted at year 1258, to reflect the 52.5 ka topography. In total, 49ka-full is run for 760 years, with 325 years incorporating all the boundary conditions. In the Results section, we compare the averages of the last 100 years of 49ka-co, the last 50 years of 49ka-alb, and the last 100 years of the 49ka-full experiment (Table , Fig. , darker colors). Although the deep ocean has not fully equilibrated in the 49ka-co and 49ka-alb simulations, our comparison with the 49ka-full simulation focuses primarily on atmospheric circulation responses, which adjust to changes in boundary conditions much more rapidly than the ocean. Surface and atmospheric variables exhibit no systematic drift over the analysis period, supporting the robustness of the diagnostics presented here.

The ITCZ position is determined using the precipitation centroid index , which identifies the latitude that evenly divides the tropical precipitation amount between 20° S and 20° N. This approach effectively represents the center of mass of tropical rainfall distribution. The strength of the northern Hadley cell is determined by the maximum value of the atmospheric mass streamfunction, ψ, within the 400–600 hPa pressure range, while the strength of the southern Hadley cell is determined by its minimum value over the same vertical pressure range (Eq. 1). 1ψ=2πRcos⁡ϕg∫0pvdp where, R is the radius of the earth, ϕ is latitude, g is gravitational acceleration, v is the meridional wind and p is pressure. ψ is expressed in Sverdrups (Sv), where 1 Sv = 10⁹ kg s⁻¹.

We first describe the 49ka-full experiment in comparison to PI. At 49 ka, the higher obliquity (24.43°) compared to PI (23.45°) results in slightly increased summer insolation in both hemispheres (Fig. ). Despite the higher obliquity, the glacial simulation (49ka-full) exhibits substantial oceanic and atmospheric cooling due to lower GHG concentrations and extensive ice sheet expansion at this time (Fig. ). Global annual mean surface air temperatures (SATs) at 49 ka are 2.7 °C lower than PI (Table  and Fig. a), with the NH cooling by 4.9 °C and the SH by 2 °C (not listed). The most pronounced cooling is simulated in ice sheet regions and northern high latitudes. Seasonal variations are evident in the NH (Table ), where temperatures drop by 6.4 °C in winter (DJF) and 2.6 °C in summer (JJA). Over SH, DJF (austral summer) SATs are 1.7 °C lower, while JJA (austral winter) SATs are 2.4 °C lower (Table ).

Global mean ocean temperature is 1 °C lower, and global annual mean sea surface temperature (SST) (Fig. b) is reduced by 1.2 °C (Table ). The simulated SATs and SSTs anomalies show a good agreement with proxy records indicating a qualitative estimate of the changes between Greenland Interstadial (GI 13; 49ka-50ka) and PI climate (Fig. a,b). Higher SSTs in the Labrador region are associated with a stronger AMOC in 49ka-full, with a transport rate of 31 Sv (about 10 Sv greater than the PI value). This strengthening is attributed to topographical changes and the presence of ice sheets, which enhance westerly winds over Baffin Bay and weaken them over the Labrador Sea (Fig. , middle panels). These wind changes modify the North Atlantic gyres, leading to an overall strengthening while shifting the separation between the subpolar and polar gyres southward and orienting the subpolar gyre in a more zonal direction (Fig. , bottom panels). A strengthening of the cyclonic circulation in the Labrador Sea is also simulated, which increases the westward transport of warm water from south of Iceland to the Labrador Sea and further strengthens the AMOC (reflected by the increase in mixed layer depth in this region in 49ka-full compared to PI, Fig. , top panels). The stronger AMOC also influences sea ice extent. During boreal winter, the sea ice edge at 49 ka expands in the North Pacific and Nordic Sea sectors but retreats in the Labrador Sea due to the influx of warm water (Fig. b, orange dashed line compared to black dashed line). In boreal summer, the 49 ka sea ice edge is similar to PI across most regions, except in the Nordic seas, where the sea ice edge extends further south than in PI by 5 to 6° latitude (Fig. b, orange solid line compared to black solid line). In the SH, austral summer sea ice retreats in the South Atlantic as stronger North Atlantic Deep Water (NADW) formation leads to a deep-ocean warming of about 0.3 °C (not shown), and this warm water gets upwelled in the South Atlantic (Fig. b, orange dashed line compared to black dashed line). However, the austral winter sea ice edge remains similar to PI in the South Atlantic while expanding in the Indian and Pacific sectors of the Southern Ocean (Fig. b, orange solid line compared to black solid line).

Drier conditions are simulated at 49 ka compared to PI over NH high-latitude land regions, particularly over the LIS, due to colder conditions and reduced atmospheric moisture availability (Fig. c). However, precipitation increases in the North Atlantic and Eurasia between 30° N and 60° N, as well as over India and East Asia. In contrast, most land areas in the SH display decreased precipitation, except over northern Australia, parts of Southeast Africa, and over eastern Brazil.

At 49 ka (experiment 49ka-full), the presence of large ice sheets over North America significantly alters the atmospheric circulation, particularly in the North Atlantic region. In DJF, the westerlies over the North Atlantic strengthen due to the strong meridional temperature gradient and shift by ∼ 6° southward compared to PI (Figs. a, c, and a), with a more zonal flow replacing the northeast-southwest tilt of the wind field (Fig. ). This pattern is evident both at the surface and at the jet stream level in the upper troposphere (Fig. ). Geopotential height (GH) differences (Fig. e) indicate that the LIS induces a high-pressure system (not shown) over North America, initiating a Rossby wave response. This Rossby wave causes a southward shift of the Icelandic low and a weakening of the Azores High, contributing to the zonal orientation of the westerlies (Figs. c, , and ). The southward shift and more zonal orientation of the jet alter moisture-bearing storm tracks, which in PI were tilted southwest-northeast, to a more southerly pathway at 49 ka. This reorganization of pressure systems (not shown), with a negative GH anomaly extending from the North Atlantic to eastern Eurasia (Fig. e), alters precipitation patterns, enhancing precipitation over the North Atlantic and western Europe (Fig. g). However, colder conditions over Eurasian landmasses lead to drier conditions inland in DJF, despite negative GH anomalies (Fig. a, e, g).

The westerlies are also shifted southward in JJA (Figs. d and b), with the flow remaining more zonal in 49 ka than during PI (Fig. , right panels), coinciding with a positive GH anomaly over the Labrador Sea region while a negative GH anomaly forms over Europe (Fig. f). These GH anomalies enhance the westerly flow and contribute to moisture transport convergence (not shown) toward the Eurasian landmass, increasing precipitation along the westerlies in the North Atlantic Ocean and over parts of Eurasia (Fig. d, f, h). In addition, minor cooling over Eurasia (Fig. b) further contributes to the increased precipitation. Precipitation also increases north of 60° N in Asia (Fig. h) due to higher specific humidity (not shown) resulting from warmer conditions and the development of a negative GH anomaly (Fig. b, f). The warming in this region arises from the topographical changes corresponding to 49 ka, where ocean points have been converted to land. Since land has a lower heat capacity than the ocean, it warms more rapidly in summer, amplifying the temperature-driven moisture increase. Conversely, north of 60° N in Europe precipitation decreases. This is due to the colder north-easterly winds and the presence of a blocking high over the Labrador-Greenland sector, which suppresses rainfall in these areas.

The role of individual boundary conditions becomes evident in the 49ka-co experiment, which includes only orbital parameters and GHGs. During DJF, strong cooling in the 49ka-co experiment reduces moisture availability, leading to widespread drying over NH landmasses (Fig. a, c). The westerlies are enhanced and slightly shift southward (∼ 1°) (Figs. a and a). This shift is accompanied by a low-pressure anomaly over the North Atlantic and a high-pressure system over Europe, which enhances subsidence and suppresses precipitation over Europe (Fig. b, c). This pattern is consistent with the overall drying trend in the NH, except for a small increase in precipitation over the North Atlantic Ocean (Fig. c), likely due to storm activity in this region. In 49ka-alb, changing the vegetation and albedo has little impact compared to 49ka-co, except over northern North America (Fig. d, f), where cooling is enhanced by 8 °C and drying increases by 25 % due to the albedo change over the LIS. The westerlies intensify further compared to 49ka-co, but their position shifts back to the PI configuration (Fig. a). The GH response features a low-pressure anomaly over the Hudson Bay and Labrador Sea, which is consistent with some increased precipitation in this region (Fig. e, f).

In JJA, the 49ka-co experiment shows no change in westerly strength and position over the North Atlantic (Figs. b, b, and b) and modest temperature and little significant precipitation changes over the NH landmass (Fig. a, c). In contrast, the 49ka-alb experiment exhibits pronounced drying over the Laurentide region and northern Europe, driven by intense cooling from albedo changes and persistent high pressure over these regions (Fig. d–f). While the westerlies strengthen in 49ka-alb, their position remains unchanged (Figs. b and b). This intensification leads to slightly wetter conditions inland over Eurasia in the 49ka-alb scenario.

In summary, only the full implementation of the ice sheet, including its topography, can induce a significant planetary wave response in the atmosphere, with southward-shifted and intensified North Atlantic westerlies (Figs. a, , , and ). This response results in substantial drying over the northern high-latitude landmasses, including the Laurentide region during DJF, along with increased milder and wetter conditions over Eurasia during JJA.

In DJF, the presence of the LIS forces a southward shift of the Ferrell cell. This displacement further leads to a southward migration of the NH Hadley cell's southern edge, shifting from 7.5° S in PI to 11° S in 49ka-full (Fig. a, see zero contour), and results in a globally averaged ITCZ shift of approximately 1.5° southward in DJF (Figs. g and a). Additionally, the NH Hadley cell strengthens by 20 Sv relative to PI (refer to Sect. 2.2 for diagnostics). The southward shift of the NH Hadley cell and ITCZ during DJF is also captured in 49ka-co (Fig. c), where the NH Hadley cell strengthens by +11 Sv relative to PI. The shift in the ITCZ is driven by peak summer insolation in the SH due to higher obliquity at 49 ka. However, the magnitude of change in 49ka-co is smaller than in 49ka-full (Figs. a and a), with an ITCZ shift of 0.6°. Implementing albedo changes in 49ka-alb enhances the Laurentide cooling, further strengthening the NH Hadley cell (by +8 Sv), albeit without additional displacement (9.5° S in 49ka-co vs. 9.6° S in 49ka-alb) (Fig. c, e). The minor shift in the NH Hadley cell and ITCZ position explains why lower latitude precipitation changes remain insignificant between 49ka-alb and 49ka-co (Fig. c, f). However, the inclusion of topography in 49ka-full results in a slight additional strengthening of the NH Hadley cell and a more substantial southward ITCZ shift (by 1.5° relative to 49ka-alb), due to the stronger temperature gradient between the two hemispheres (Fig. a, Table ).

During JJA, the southern edge of the SH Hadley cell shifts northward by 2° in 49ka-full, moving from 32° S in PI to 30° S, along with a northward shift of the Southern Ferrel cell (Fig. b, see zero contour). This shift causes a slight northward migration of the global ITCZ (Fig. h), while the SH Hadley cell strengthens by 35 Sv. In 49ka-co, a slight warming over the North Atlantic and stronger cooling in the SH shifts the SH Hadley cell's southern edge to 31° S, accompanied by a strengthening of +22 Sv (Figs. d and b). The changes in 49ka-alb do not alter the SH Hadley cell's latitudinal position significantly but weaken its strength to PI levels (Figs. f and b), due to intensified Laurentide cooling. As a result, precipitation changes in the lower latitudes remain small (Fig. c, f). Temperature contrasts between hemispheres are reduced by ice sheet topography (Table , comparison of JJA NH minus SH SAT differences between 49ka-full, 49ka-alb and 49ka-co), which induces localized warming in Siberia and strong Antarctic cooling, and by changes in the land-sea mask with the replacement of ocean grid cells by land. This reduced thermal gradient with warming in the NH leads to a slight northward ITCZ shift, further strengthening the SH Hadley cell and shifting it an additional 1° northward, toward the warmer hemisphere, relative to 49ka-alb.

The SH westerlies also exhibit distinct responses across the simulations. In DJF, they weaken slightly in 49ka-co and 49ka-alb, accompanied by a small northward shift (Fig. c). In contrast, in 49ka-full, they remain in the same position as in PI but strengthen due to the increased height and extent of the AIS. During JJA, the SH westerlies weaken across all 49 ka simulations compared to PI, with the most pronounced weakening occurring in the experiment where albedo is changed (Fig. d). However, in 49ka-full, they shift northward by 2.5°, a displacement linked to the northward shift of the SH Hadley cell in 49ka-full (Figs. d and b).

The implementation of 49 ka boundary conditions leads to a notable increase in Australian rainfall during the monsoon season (Figs. g and a). This section focuses on the large-scale processes driving this regional rainfall response. While a detailed analysis of monsoon changes in other regions lies beyond the scope of this study, our results underscore the influence of NH ice sheets on hemispheric-scale atmospheric dynamics.

The presence of NH and larger AISs induces substantial atmospheric circulation changes by modifying the Hadley and Ferrel cells and influencing the westerlies in both hemispheres. These shifts propagate through the atmosphere, altering surface pressure systems and triggering a hemispheric-scale response.

During DJF, the full boundary conditions in 49ka-full force a southward shift of the Hadley and Ferrel cells (Fig. a), enhancing subsidence over East Asia around 40° N and leading to drier conditions (Fig. e, g). This southward displacement of the NH Hadley cell extends across hemispheres, impacting Australian rainfall by shifting convergence southwards along with the ITCZ (Fig. a). These changes explain the wetter conditions in northern Australia and drier conditions over the Maritime Continent (Fig. g). A comparable but weaker response is simulated in the 49ka-co experiment (Fig. c). In contrast, changes in albedo do not impact Australian precipitation significantly (Fig. f). However, modifying the topography induces wetter conditions in northern Australia driven by shifts in the Hadley cell and ITCZ, while southern Australia experiences drier conditions (Fig. h, i).

During JJA, in response to the 49 ka boundary conditions, most of Australia experiences drying (Fig. h), primarily due to the establishment of blocking high-pressure systems south of the continent (Fig. f). These high-pressure systems reinforce subsidence and restrict moisture transport from the Southern Ocean into the Australian region (not shown). However, the drying response is not significant over central Australia and remains less pronounced along the eastern and western edges of the continent. The southwestern part of Australia becomes wetter, while the southeastern region experiences drying, due to the northward displacement of the westerlies, resembling a negative winter Southern Annular Mode (SAM). Significant drying also occurs across northern Australia and much of the Maritime Continent south of the equator.

When only orbital parameters and GHGs are modified, the SH Hadley cell shifts northward and strengthens. However, in this scenario, the weakening of the SH westerlies and the southward displacement of subtropical high-pressure systems result in increased rainfall over southern Australia (Fig. b, c). Introducing albedo changes alone has little impact on the Australian climate (Fig. d–f). However, when ice sheet topography is incorporated, the northward shift in SH westerlies linked with the strengthening and equatorward displacement of the SH Hadley cell causes enhanced subsidence and thus drier conditions over Australia, coinciding with the northward ITCZ migration (Figs. i and d). Compared to the PI climate, this drying effect is somewhat mitigated, as the westerlies – despite their northward shift – remain weaker overall.

Proxy records suggest significant AMOC variability during the MIS 3 period, with stronger AMOC during the onset of interstadials , and weak AMOC during stadials. The processes driving this AMOC variability are still debated . Model simulations of MIS 3 climates show a range of AMOC responses depending on the prescribed boundary conditions, particularly the configuration of Northern Hemisphere ice sheets. For example, CCSM3 simulations at 35 and 44 ka show AMOC reductions of approximately 40 % and 50 % , respectively, while a University of Victoria (UVic)-ESM simulation at ∼ 40 ka produces a ∼ 33 % reduction relative to PI . In contrast, other simulations show little change or a modest strengthening of the AMOC. For example, simulated a ∼ 7 % strengthening at 38 ka using CCSM3, while reported a ∼ 13 % strengthening at 38 ka using NorESM. Similarly, simulated a 22 % increase in AMOC strength during MIS 3 compared to PI (estimated from ) using the LOVECLIM model.

The simulated AMOC in our 49ka-full simulation is strong (31 Sv) with a 47 % increase relative to PI. This strengthening is driven by changes in the LIS, which enhances westerly winds over Baffin Bay while weakening them over the Labrador Sea, intensifying the cyclonic circulation in the Labrador Sea and facilitating the transport of warm, saline waters to the region. This process is also associated with sea ice retreat in polar regions, which further reinforces AMOC strength. These findings align with earlier modelling studies emphasizing the role of topography and ocean–atmosphere interactions in AMOC enhancement . For example, demonstrated that the presence of an extensive ice sheet during the LGM enhanced AMOC by altering atmospheric circulation patterns. However, their simulations also showed that the cooling effects from lower GHG concentrations and increased albedo suppressed AMOC strength. In our 49ka-full experiment, these cooling effects appear to be offset by the relatively high insolation state at 49 ka, allowing topographical impacts to dominate.

The simulated AMOC in 49ka-full is stronger than that reported in most previous MIS 3 simulations. However, the spread among MIS 3 simulations highlights the sensitivity of AMOC strength to the prescribed LIS configuration. Previous studies focusing on LGM climate have shown that changes in North Atlantic westerlies induced by the presence of the LIS can strengthen the AMOC through enhanced wind-driven circulation and deep-water formation . Consistent with this mechanism, the stronger AMOC in our 49ka-full simulation is associated with LIS-driven changes in atmospheric circulation and enhanced westerlies over the North Atlantic. Conversely, changes in LIS height have also been shown to strongly affect AMOC stability and can trigger transitions between different overturning states , indicating that different ice-sheet reconstructions can produce markedly different AMOC responses. Since the LIS evolved substantially throughout MIS 3 and remains poorly constrained , alternative but equally plausible ice-sheet configurations may yield weaker AMOC states than simulated here. This sensitivity is consistent with evidence for large AMOC variability during MIS 3, including rapid transitions and possible overshoots following Heinrich stadials. We therefore interpret the strong AMOC in 49ka-full as a response to the specific ice-sheet configuration prescribed in our experiments rather than as a robust feature of all MIS 3 climates.

Evidence based on foraminifer assemblages and mineral grains in ice-rafted debris also suggests a northward retreat of sea ice and higher temperatures in the Labrador Sea between ∼ 56 and 49 ka, potentially driven by the insolation maxima . After 49 ka, sea ice variability increased, with near-perennial sea ice cover during most stadials and seasonal ice-free conditions during interstadials. However, it remains uncertain whether interstadials reflect a mean state or are primarily linked to Dansgaard–Oeschger (D-O) variability.

The simulated changes in the North Atlantic westerlies and associated precipitation patterns during DJF highlight the critical role of ice sheet topography in driving atmospheric circulation (Fig. a). At 49 ka, the surface westerlies shift southward and adopt a more zonal flow (Fig. ), aligning with previous studies , which emphasize the influence of glacial ice sheet topography on jet stream reorganization. Our results further expand on this, demonstrating that the high-pressure system induced by the LIS initiates a Rossby wave response, shifting the Icelandic low southward and weakening the Azores High. These circulation changes lead to reduced precipitation over NH landmasses during DJF.

During JJA, the changes in westerlies and pressure result in increased rainfall over Eurasia (Fig. b), with the jet stream extending further inland. Additionally, this rainfall increase is supported by milder land temperatures and enhanced moisture availability (not shown). Our findings on seasonal precipitation changes are consistent with those of , as the ice sheet configuration used in our study closely resembles their MIS 4 ice sheet setup. However, while their study reports a fully zonal orientation of the jet stream only during the LGM, our results demonstrate that this change can also occur with a MIS 3 ice sheet configuration, suggesting a dynamic response to glacial ice sheets earlier in the glacial period.

The presence of ice sheet topography leads to a more pronounced southward shift of the NH Hadley cell during DJF and a stronger northward shift of the SH Hadley cell during JJA, compared to our simulations that include only GHG concentrations, orbital parameters, and an ice mask at 49 ka (Fig. a, b). This enhanced Hadley cell shift in 49ka-full drives a more substantial southward displacement of the ITCZ during DJF (Fig. a). Consequently, the northern tropics experience drier conditions, while precipitation increases in the southern tropics, consistent with the mechanisms proposed by .

During JJA, LIS topography induces warming over Siberia, in agreement with previous studies , which attribute this warming to large-scale atmospheric circulation changes driven by pressure differences linked to LIS height. In contrast, AIS topography enhances cooling over Antarctica during austral winter. This interhemispheric contrast, i.e. relatively warmer conditions in the NH and colder conditions in the SH, pushes the ITCZ northward and strengthens the SH Hadley cell. Additionally, the SH Hadley cell northward shift is associated with the northward displacement of the SH westerlies (Fig. b), consistent with , who demonstrated that interhemispheric teleconnections link ITCZ migration with changes in Hadley cell strength and SH jet positioning. Likewise, NH cooling during DJF drives a southward shift of the North Atlantic westerlies, influencing the Hadley cell strength and position, and pushing the ITCZ closer to the equator.

These circulation shifts generate regional impacts. Increased subsidence over East Asia creates a high-pressure anomaly, leading to drier conditions, while the southward-shifted ITCZ enhances convergence over Australia, increasing rainfall. The larger shift in the ITCZ and the NH Hadley cell coincides with the maximum increase in Australian rainfall (Fig. a).

The weakening of the SH westerlies during both DJF and JJA under a high-obliquity climate (49ka-co and 49ka-alb) compared to a low-obliquity climate (PI) aligns with findings from . However, the increased height and extent of the AIS counteract this weakening during DJF, leading to SH westerly intensification in 49ka-full. Additionally, our results indicate that NH westerlies strengthen with albedo changes relative to orbital parameter and GHG adjustments, whereas SH westerlies weaken under the same conditions (Fig. ). Introducing topographical changes induces an equatorward shift in both NH and SH westerlies during their respective winter seasons (Fig. a, d).

Overall, our study reveals nonlinear interactions between large-scale atmospheric circulation, surface westerlies, and precipitation patterns under different combinations of boundary conditions (Fig. ). Orbital forcing, GHGs, albedo, and ice sheet topography collectively shape these interactions in a nonlinear manner.

Proxy records from marine sediments, lake sediments, and river systems show that Australia experienced spatially variable climates during MIS 3, with greater-than-modern mean water availability peaking between ∼ 49 and ∼ 40 ka . Wet conditions, especially in central and northern Australia between ∼ 50 and ∼ 45 ka, are linked to an intensified Australian monsoon driven by the summer insolation peak. Our model results align with these observations, showing increased annual precipitation over much of Australia at 49 ka, with particularly intensified monsoonal rainfall in the northern regions (Figs. c and g). The increase in monsoonal rainfall during austral summer is consistent with a southward displacement of the ITCZ due to the intensified NH Hadley cell, driven by increased SH insolation in response to high obliquity. However, our results also reveal that the southward shift of the ITCZ and Hadley cell occurs in the 49ka-full simulation (which includes ice sheet topography) as well as in the 49ka-co and 49ka-alb simulations (which lack ice sheet topography). Despite this, the intensified Australian monsoon rainfall is more pronounced in the 49ka-full experiment, as the presence of ice sheets induces larger shifts of the Hadley cell and ITCZ.

Our simulations generally show an insignificant response in southern Australian annual rainfall at 49 ka, although austral winter exhibits some drier conditions. These results are consistent with those of , in which speleothem growth rates from southwest and southeast Australia indicate drier-than-modern conditions around 49 ka. In contrast, paleo records from suggest increased rainfall in southern Australia. The wetter signal in that study is interpreted as a response to changes in the SH westerlies, involving either strengthening or an equatorward shift. While the wind response is consistent with the equatorward shift simulated in the 49ka-full experiment, the rainfall response is not reproduced, indicating that southern Australian precipitation is not solely controlled by latitudinal shifts of the westerlies in ACCESS-ESM1.5. Instead, rainfall anomalies appear to be more strongly influenced by persistent subtropical high-pressure anomalies and associated subsidence over the Australian continent, which suppress moisture transport convergence despite the shift in westerlies. This may indicate limitations in the simulated coupling between storm-track variability and regional precipitation under glacial boundary conditions. However, the contrasting interpretations from available proxy records also suggest that hydroclimate conditions in southern Australia during MIS 3 remain uncertain.