Weak Southern Hemispheric monsoons during the Last Interglacial period

Due to different orbital configurations, high northern latitude boreal summer insolation was higher during the Last Interglacial period (LIG; 129–116 thousand years before present, ka) than during the preindustrial period (PI), while high southern latitude austral summer insolation was lower. The climatic response to these changes is studied here with focus on the southern hemispheric monsoons, by performing an equilibrium experiment of the LIG at 127 ka with the Australian Earth System Model, ACCESS-ESM1.5, as part of the Paleoclimate Model Intercomparison Project 4 (PMIP4). In our simulation, mean 5 surface air temperature increases by 6.5 ◦C over land during boreal summer between 40◦ N and 60◦ N in the LIG compared to PI, leading to a northward shift of the Inter-Tropical Convergence Zone (ITCZ) and a strengthening of the North African and Indian monsoons. Despite 0.4 ◦C cooler conditions in austral summer in the Southern Hemisphere (0–90◦ S), annual mean air temperatures are 1.2 ◦C higher at southern mid-to-high latitudes (40◦ S–80◦ S). These differences in temperature are coincident with a large-scale reorganisation of the atmospheric circulation. The ITCZ shifts southward in the Atlantic and Indian sectors 10 during the LIG austral summer compared to PI, leading to increased precipitation over the southern tropical oceans. However, the decline in Southern Hemisphere insolation during austral summer induces a significant cooling over land, which in turn weakens the land-sea temperature contrast, leading to an overall reduction (−20 %) in monsoonal precipitation over the Southern Hemisphere’s continental regions. The intensity and areal extent of the Australian, South American and South African monsoons are consistently reduced. This is associated with greater pressure and subsidence over land due to a strengthening of 15 the southern hemispheric Hadley cell during austral summer.


Precipitation change
Simulated annual mean precipitation anomalies are shown in Figure 3a, and compared to a recent compilation of LIG precipitation reconstructions based on a range of proxy records (including pollen, speleothems, landscape features, loess and sediment composition) (Scussolini et al., 2019). In the NH the model is in agreement with 64 out of 109 proxy records (59 %) (where the model and proxy data show the same sign of change, or where the change in the simulations is <100 mm yr −1 and the 160 corresponding proxy suggests no change). Although the agreement is not compellingly strong, it is worth pointing out that the majority of disagreements arise in central Europe where reconstructions are abundant and mostly display wetter conditions, whereas the simulation shows a slight decrease in precipitation. In contrast, there is a good model-data agreement in northern Africa, the Middle East, Asia and North America.
As seen in Figure 3a, the largest changes in annual mean precipitation are simulated in the tropics where both the rainfall 165 mean and variability are higher. Annual precipitation increases over land in the northern tropics, particularly over North Africa (the Sahel, >+300 %), South Asia (India, +100 %), and Central America (Mexico, +40 %). Coinciding with increasing precipitation, the significant cooling (∼ −3 • C; Fig. 3b) over the Sahel and India is associated with stronger evaporation over land during boreal summer (not shown). All ocean regions in the northern tropics are simulated to be wetter, except for the South China Sea and Philippine Sea. In general there is good agreement with proxy records in the NH.

170
In contrast, the model generally simulates drier conditions over the southern tropics during the LIG (Fig. 3a), particularly over South America (−20 %), South Africa (−40 %) and northern Australia (−40 %). This decrease in precipitation mostly occurs in DJF, during the Southern Hemispheric monsoon season (Fig. 4a). In addition, this precipitation reduction is consistent with an overall increase in mean sea-level pressure over the SH land (Fig. 4b). Therefore, the simulation suggests a major change in precipitation in the tropics, associated with a shift in the Inter-Tropical Convergence Zone (ITCZ) and a weakening 175 of the convergence zones in the SH, which will be discussed in the next section.

ITCZ changes
The ITCZ position corresponds to the latitudinal band of maximum precipitation. Previous studies have demonstrated that the position of the ITCZ aligns with the zero energy flux equator, i.e. where the atmospheric meridional energy flux divergence vanishes (Schneider et al., 2014;Ceppi et al., 2013). In the present climate, this occurs north of the equator, thus placing 180 the ITCZ in the NH. This is essentially due to an interhemispheric energy imbalance, caused primarily by the northward transport by the Atlantic Meridional Overturning Circulation (AMOC). The net northward oceanic heat transport to the NH is compensated by a southward atmospheric heat transport via a northward-displaced Hadley cell and the positioning of the ITCZ north of the Equator. Therefore, one would expect that an interhemispheric temperature gradient would drive a cross-equatorial atmospheric energy flux, which in turn determines the position of the ITCZ towards the warmer hemisphere (e.g., Schneider As seen in Figure 5h, zonally averaged precipitation in JJA displays a slight northward shift of the ITCZ at the LIG compared to PI. Although this northward ITCZ shift is seen in all ocean basins, there are differences in the magnitude of the peak precipitation. The peak is 9 % stronger in the Pacific sector (130 • E to 70 • W), but 39 % weaker in the Atlantic sector. In the Indian sector (0 to 130 • E), even though the ITCZ location is less defined, precipitation is higher between 5 • S and 20 • N at the 190 LIG. In short, the ITCZ shifts northward in the Pacific and Atlantic sectors in JJA compared to PI. This consistent northward shift during JJA can be explained by the higher NH summer temperatures, which accentuate the interhemispheric temperature gradient.
The globally averaged precipitation in DJF suggests a contraction of the ITCZ, with an increase in strength (by 10 %) at 5 • S, while precipitation at 5 • N decreases by 23 % (Fig. 5d). However, the DJF precipitation response across basins varies 195 significantly. In the Indian sector, there is a southward shift (∼3 • ) and slight strengthening of the ITCZ. In contrast, in the Pacific sector, which displays a double ITCZ in DJF in both PI and LIG, a weakening and northward shift (∼1 • to 4 • ) of the precipitation peaks at 10 • S and 7 • N are simulated. Similarly, in the Atlantic sector, a ∼25 % weakening of DJF precipitation at 5 • N during the LIG is simulated, while there is a strengthening of the ITCZ peaks at 5 • S (Figs. 4a, 5c). The simulation thus displays a southward ITCZ shift over the Atlantic and Indian Oceans, but a northward shift in the Pacific Ocean. These 200 differences in longitudinal responses are associated with changes in large-scale atmospheric circulation. Since low latitude precipitation is associated with monsoon systems, we will now look into more detail at the precipitation changes occurring in each of the monsoon regions.

Precipitation changes in monsoon regions
Monsoon domains are defined as regions in which the monsoon season precipitation is greater than 2.5 mm day −1 compared 205 to dry season (Wang et al., 2011). In the NH, the May-to-September (general monsoon season) precipitation should therefore be 2.5 mm day −1 higher than the November-to-March (dry season) precipitation to qualify as a monsoon region (Wang et al., 2011). The reverse holds for the SH: precipitation during November to March is required to be 2.5 mm day −1 higher than May to September. As shown in Figure 6a, the model simulates a general extension of the monsoon domains in the northern tropics, also associated with increased precipitation rates, whereas there is a contraction of the monsoon domains and reduced 210 precipitation in the southern tropics.
The North African monsoon domain expands significantly into the Sahara region. This expansion is associated with enhanced southwesterly winds. Indian monsoon also strengthens with monsoon domain covering most of India and stronger onshore winds from the Arabian Sea and convergence inland (Fig. 6a). On the other hand, a contraction of the monsoon domain over the Philippine Sea and South China Sea is simulated and attributed to a northeastern wind anomaly associated with trade winds 215 strengthening in JJA.
As evident from Figure 6b, there is a marked difference in DJF precipitation anomalies over the SH land and ocean regions during LIG relative to PI. To investigate this difference further, we separate the simulated tropical precipitation anomalies over land and ocean (Fig. 7). Precipitation over land is generally higher in the northern tropics, with an increase of over 60 % in total precipitated water for the North African monsoon (NAF) (Fig. 7). The South Asian Monsoon (SA) also displays a 220 large increase in precipitated water (>60 %), mainly due to a large increase in areal-extent (+>55 %), as the increase in areaaveraged precipitation is small (∼5 %). The North American monsoon (NAM) is slightly weakened, with a slight decrease in While the reduced insolation in the SH during DJF would tend to reduce precipitation over the SH, changes in the global atmospheric circulation produce a southward shift of the ITCZ in the Indian and Atlantic Oceans (Sect. 3.2.1, Figs. 4a, 5a, and 5c) that counteract the insolation changes and lead to higher precipitation over those tropical ocean regions (Fig. 7).

260
In addition, the land-ocean contrast helps maintain this configuration via changes in zonal pressure gradient between land and adjacent oceans, creating local Walker-type circulation anomalies (with anomalous ascending motion over oceans and compensatory subsidence over land).
A strengthening of the Hadley circulation in the SH is simulated in DJF (Fig. 8). It contributes to the simulated drier conditions in the subtropics at ∼30 • S due to greater subsidence, as seen from the increase of surface pressures over land. The 265 southern boundary of the Hadley cell also experiences a slight northward shift, from ∼31 • S to ∼28 • S, which pushes regions of high pressure in the subtropics to the north (Fig. 4b). This favours an anomalous subsidence and weaker convection over the SH convergence zones thus reducing monsoonal precipitation.

Discussion
At the LIG the insolation reaching Earth was different from PI, with higher insolation in JJA in the NH and lower insolation At low to mid southern latitudes, simulated surface air temperature anomalies in the ACCESS-ESM are also very similar to the PMIP4 multi-model mean with ∼3 • C warming over land in JJA, and ∼2 • C cooling over land in DJF. However, the model simulates much warmer conditions over the Southern Ocean all year round, with values higher than the multi-model mean.
This is concurrent with a large decrease in Southern Ocean sea-ice at the LIG. While the simulated warming over the Southern 280 Ocean throughout the year is at the higher end of the PMIP4 lig127k multi-model mean, the warming over Antarctica is at the lower end (Otto-Bliesner et al., 2020).
The simulated spatial distribution of annual precipitation anomalies (Fig. 3a) agrees well with the PMIP4 multi-model mean (Otto-Bliesner et al., 2020), with drier conditions over SH land regions including North Australia, South Africa and South America, while wetter conditions are simulated at ∼10 • S in the Atlantic Ocean, the Indian Ocean and the western 285 equatorial Pacific Ocean. However, while our simulation suggests wetter conditions over the equatorial Pacific at 10 • N due to a strengthening of the ITCZ in this basin in JJA (Fig. 5f), this does not occur in the multi-model mean.
Reduced precipitation over land and increased precipitation over the ocean in DJF in low to mid-southern latitudes is also evident in the PMIP4 multi-model mean, indicating that the ACCESS-ESM1.5 lig127k simulation provides a good representation of the LIG features as simulated by coupled climate models. Despite the strong signal emanating from the lig127k simulations, 290 9 https://doi.org/10.5194/cp-2020-149 Preprint. Discussion started: 25 November 2020 c Author(s) 2020. CC BY 4.0 License. these precipitation anomalies are not necessarily evident from SH paleoproxy records, highlighting the need for additional hydrological records from low-and mid-southern latitudes. Furthermore, the lig127k is an equilibrium simulation, which does not take into account potential meltwater discharges from the Antarctic and Greenland ice-sheets, and their associated impact on deep water formation and therefore variability in ocean and atmospheric circulation (e.g. Hayes et al., 2014;Rohling et al., 2019;Tzedakis et al., 2018). The time evolution across the LIG of the hydrological proxy records should thus be looked at in Eurasia and North America in regions with less summer rainfall during the LIG (Fig. S3). In contrast, increased precipitation over the Sahel and surrounding regions (Fig. 6a) corresponds to elevated LAI. This affects evapotranspiration, as an increase in LAI enhances canopy evapotranspiration, while reducing soil evaporation, and ultimately also alters the hydrological cycle.
It has been previously shown that due to changes in albedo and moisture availability, changes in vegetation might amplify the changes in precipitation (e.g. Messori et al., 2019). For example, a large increase in precipitation over the Sahel would 310 lead to a greening of the Sahara (Hopcroft et al., 2017;Larrasoaña et al., 2013;Osborne et al., 2008;Pausata et al., 2020).
The greening would tend to increase the sensible and latent heat fluxes into the atmosphere, and produce a cyclonic circulation anomaly over the Sahara, whose anomalous westerly flow would transport more moisture from the neighbouring Atlantic Ocean into the region (Messori et al., 2019;Patricola and Cook, 2007;Rachmayani et al., 2015). Additional simulations with interactive vegetation, or with prescribed changes in vegetation cover, should thus be performed to quantify the coupled effect 315 of precipitation and vegetation changes.
Due to the importance of the southern hemispheric monsoon systems for water availability, such reduced precipitation in the southern tropics in a past warmer world are concerning. However, since these negative precipitation anomalies mostly result from the strong cooling over land, such hydrological changes are not currently expected over the coming centuries under the RCP/SSP (Representative Concentration Pathway / Shared Socioeconomic Pathways) scenarios where future climate changes 320 are due to greenhouse gas forcing and not to changes in orbital parameters. CMIP6 future projections indeed generally simulate no significant changes in the Australian, South African and South American monsoons by 2100 under all SSP scenarios, while drier conditions are simulated over these regions in JJA (Cook et al., 2020). In Australia, northern rainfall in DJFMAM is more constrained in CMIP6 than in CMIP5 and it is projected to experience a ∼2 % increase by the year 2090 under high-emission