Ocean dominated expansion and contraction of the late Quaternary tropical rainbelt

The latitude of the tropical rainbelt oscillates seasonally but has also varied on millennial time-scales in response to changes in the seasonal distribution of insolation due to Earth’s orbital configuration, as well as climate change initiated at high latitudes. Interpretations of palaeoclimate proxy archives often suggest hemispherically coherent variations, some proposing meridional shifts in global rainbelt position and the ‘global monsoon’, while others propose interhemispherically symmetric expansion and contraction. Here, we use a unique set of climate model simulations of the last glacial cycle (120 kyr), that compares well against a compilation of precipitation proxy data, to demonstrate that while asymmetric extratropical forcings (icesheets, freshwater hosing) generally produce meridional shifts in the zonal mean rainbelt, orbital variations produce expansion/contractions in terms of the global zonal mean. This is primarily a dynamic response of the rainbelt over the oceans to regional interhemispheric temperature gradients, which is opposite to the largely local thermodynamic terrestrial response to insolation. The mode of rainbelt variation is regionally variable, depending on surface type (land or ocean) and surrounding continental configuration. This makes interpretation of precipitation-proxy records as large-scale rainbelt movement challenging, requiring regional or global data syntheses.


Impact of Heinrich events vs. seasonal insolation
The model results suggest that the oceanic tropical rainbelt seasonal range responds to changes in insolation seasonality (primarily controlled by orbital precession) by expanding and contracting whereas the terrestrial response changes through meridional shifts in the rainbelt. At the northern limit in boreal summer, the land and ocean regions respond in concert, whereas the southern rainbelt limit in austral summer produces opposing ocean and land responses (see main text for details). In contrast, northern high latitude forcing via freshwater hosing (in order to approximate a Heinrich event) produces a homogeneous southward shift in the seasonal range over both land and ocean.
Concentrating on the southern hemisphere around the Atlantic and Africa, when austral summer insolation is lower local land temperatures decrease more than sea surface temperatures (SSTs), decreasing the land-sea temperature contrast ( Fig. S6). The outcome is reduced monsoon circulation and lower moisture advected from ocean to land. In addition, there is less convective activity, lower relative humidity (Fig. S6), and lower rainfall over much of southern Africa. One can contrast this response to that of Heinrich event forcing (Fig. 2b). In austral summer, even though the land-sea temperature contrast is similarly reduced in southern Africa, it is because South Atlantic SSTs have increased (Fig. S6), following the bipolar seesaw mechanism. This results in greater moisture transport onto the continent, increasing humidity and rainfall over southern Africa, even though the monsoonal circulation is reduced. Consequently, the terrestrial rainbelt appears to shift southward in response to Heinrich event forcing.
The southern Atlantic responds to broader interhemispheric changes, which shift both the northern and southern limit of the Atlantic rainbelt southwards due to the latitudinally asymmetric extratropical forcing. The model rainbelt northern limit produces larger southerly shifts than the southern limit in response to the high northern latitude forcing, as can be seen in Fig. 2b.
The model patterns observed over the Atlantic and surrounding continents extend into the Indian Ocean and Easternmost Pacific the response is less clear in the Pacific as a whole. Likewise, other climate models demonstrate the strongest and clearest southerly shift of the tropical rainbelt over the Atlantic due to similar high northern latitude freshwater input, while there is a greater variation in response outside the Atlantic region 74 .

Comparison of HadCM3 with palaeodata
The multi-millennial trends in precipitation from HadCM3 over the last glacial cycle have been compared with palaeodata (Table S1) time series that have been previously interpreted as representing changes to precipitation or hydroclimate in a broader sense. As shown in Fig. 2c of the main text we calculated the timing of maximum precipitation during the Holocene from the palaeodata records and for each grid cell of the model for comparison of the local phasing of precipitation variation. We found that at 81% of the points the model and data timing of maximum precipitation match to within 2kyr. Where possible we chose records that cover more than one precessional cycle and include the majority of the Holocene time period. In most of the locations the models and data demonstrate not only similar phasing of precipitation, but also similar longer-term changes in the magnitude and rate of variations over the glacial cycle (e.g. Fig. S3h). There are locations where there is disagreement and one example of this is over East Africa (Fig. S3b). Here the models and data show opposing trends. Model-data comparison over this region is discussed further in [25]. The differences partly relate to the low resolution of the model, which impacts the location of atmospheric boundaries over the high topography of the rift valley. In such cases it may be the case that climate patterns are produced that are shifted in their location from where they would occur in reality.
Similarly, over the Gulf of Guinea there is a mismatch in the timing of maximum precipitation during the Holocene (Fig. 2c), where in the data the maximum occurs in the early Holocene, whereas in the model it occurs in the late Holocene. Looking at the full available time series (Fig. S3k), it becomes clear that there is a better match in the phasing during the glacial period, and that the trends in model and data only really diverge during the Holocene 25 . This relates to the impact of the ice-sheet forcing in the model, which pushes the Atlantic ITCZ south during the glacial and leads to this particular location sitting in the northern part of the rainbelt that is in phase with northern hemisphere summer insolation, whereas in the Holocene (with less expansive ice-sheets) it sits in the central part of the rainbelt that is in antiphase with northern summer insolation in the model (which doesn't match with the data). Overall, locations that are close to geographical boundaries of physical phenomena (relating to topographic or atmospheric circulation boundaries) are more likely to display poorer model-data comparisons. However, in general we demonstrate that there is a high level of model-data agreement in multi-millennial variations in hydroclimate over the last glacial cycle.

Comparison of HadCM3 with other climate models
There is a lack of simulations with other full complexity climate models that cover a whole glacial cycle. The Palaeoclimate Model Intercomparison Project 3 (PMIP3) provides the best opportunity for inter-model comparison of multi-millennial variation in the tropical rainbelt.
There are several key regions where the majority of climate model simulations contributed to the PMIP3 suite show similar changes in tropical precipitation. In the mid-Holocene (MH; 6kyr BP) compared to the pre-industrial (PI; 0 kyr BP) all PMIP3 models demonstrate northward enhancement of precipitation over N Africa (Fig. S4) as does HadCM3 (Fig. S14a). This is also seen through calculations of the northern limit of the tropical rainbelt, where this is further north at the MH than PI for HadCM3 ( Fig. 5) and the PMIP models 21,25 . Similarly, there is consistency in the movement of the southern limit of the rainbelt over Africa. Here, the majority of models (10 of 13 models) suggest the rainbelt southern limit to be further north in the MH than PI [ Fig.   3b in [25]). The overall result is that there is a high level of model agreement that the African tropical rainbelt underwent a meridional shift south in its seasonal range between the MH and PI. The intermodel variation over central and Eastern parts of Africa is much higher.
The majority of PMIP3 models demonstrate expansion/contraction in the Atlantic tropical rainbelt seasonal range (Fig. S4) as HadCM3 does (Fig. S14a). This manifests as lower annual mean precipitation at the equator in the MH than PI and higher MH-PI precipitation to the north and south of this. The annual mean variation belies the variation in any one month/season, in which the Atlantic rainbelt shifts north/south (Fig. 3). Similarly, the boreal summer (JJAS) Atlantic tropical rainbelt in the last glacial maximum (LGM) shifts southwards in most PMIP3 models, but spatial patterns of anomalies are less consistent in boreal winter, leading to larger intermodel variation in the LGM-PI anomalies than MH-PI ( Fig. S15 and Fig. S14b).
Regions where there is less consistency between models provide vital challenges to the palaeodata community to focus new data collection efforts and for overall synthesis, and for the palaeoclimate modelling community to understand the reasons for such divergences and for model optimization exercises. The degree of intermodel variation depends on the driving forces introduced. For example, the MH-PI precipitation anomalies over North Africa are highly consistent between models, whereas the LGM-PI response in the same region is one of the least consistent even in terms of the sign of the anomaly. Conversely, the Indo-Pacific response displays low model agreement in the MH, but a greater agreement at the LGM. Similarly, the Eastern Pacific precipitation response at the MH is more variable than the southward shift that occurs at the LGM.

Seasonality in the northern limit of the South American rainbelt
The most northern position of the rainbelt over part of South America (280-300°E) during boreal summer displays interesting differences between the ORB-ONLY and ALL experiments. This is the region around the Cariaco Basin, where a run-off proxy record was interpreted as precipitation that was in phase with local summer (e.g. July) insolation, and thus that the ITCZ shifts were also in phase with local insolation. Our model results suggest that the northernmost ITCZ position of the ITCZ is, in general, in phase with local insolation and precipitation amount for ORB-ONLY, but in antiphase with local insolation and precipitation amount when ALL forcings are included (Fig. 5d top; Fig. S11). In the model the cause of this is a change in the timing of the most northward position of the rainbelt.
In the ORB-ONLY experiments the most northerly position of the rainbelt over S America occurs in September/October (Fig. S16, top). In this season the northerly position of the ITCZ varies in phase with local insolation. In the ALL experiment the most northerly extent occurs in July/August. During this season the ITCZ varies in anti-phase with local insolation, moving south when boreal insolation is high.
In the ALL experiment the October rainbelt is positioned much further south than in ORB-ONLY (Fig. S16, bottom). This is because the position of the rainbelt in October in the ALL experiment varies more in response to the ice sheet forcing than it does to insolation. When the Laurentide Ice Sheet expands the seasonal snow cover extends much further south and the resultant increase in the local latitudinal temperature gradient shifts the ITCZ south in Sept/Oct.
The reason for the anti-phasing of the location of the July/August rainbelt with insolation, which is seen in both experiments, is the configuration of the land: most local land is to the south of the northernmost part of the July/August ITCZ. When local summer insolation is high, the land-sea temperature contrast increases and the atmospheric circulation intensifies, however, because the land is to the south of the ITCZ the rain belt moves south, apparently away from the warmer hemisphere, but in fact towards the locally warmed land. This mechanism occurs during July and August in the model in both the ORB-ONLY and ALL experiments. Therefore, when there is extensive Northern Hemisphere ice cover and the maximum northward extent to the ITCZ occurs in July/August, we see an anti-phase relationship with insolation. When there is no ice and the maximum northward extent occurs in September/October we see an in phase relationship. This has implications for the interpretation of proxies.
Regardless of whether there is ice present or not, the amount of rainfall in Northern S America shows an in phase relationship with insolation, increasing when the boraeal summer insolation increases ( Fig S11). However, as we have just described, when ice is present the northward extent of the ITCZ can move south in response to boreal summer insolation. If, therefore, a proxy is sensitive to the maximum extent of the ITCZ, for example a record that lies at the very edge of the ITCZ's range, it cannot necessarily be interpreted in the context of changes in the total amount of rainfall in the ITCZ. Conversely, a proxy that is sensitive to the mean amount of rainfall, for example a record in the core of the ITCZ, should not be interpreted in the context of shifts in the ITCZ Whether or not the modelled response and the actual climate's response to insolation are the same in the exact detail, this result demonstrates the potential for over-or mis-interpreting hydroclimate shifts from single proxy records.

Southern Atlantic rainbelt and obliquity
A notable feature over the marine southern rainbelt limit ( Fig. 1c; bottom) is that while the ORB-ONLY experiment displays precessionally influenced 21-kyr cyclicity, the ALL experiment equatorward movement of the rainbelt is reduced every other precession cycle (at ~96 kyr and ~44 kyr BP), when obliquity is also high (Fig. S12).
This response is most prominent in the southern Atlantic (Fig. 3).
During these times, when both obliquity is high and expanded ice-sheets are present, there is a change in the seasonality of the rainbelt movement such that the maximum southern latitude occurs in April rather than in February. (Fig. S9d and Fig   S13). The latitudinal pattern of austral summer mean rainfall at these times ( Fig. S9a; 44 kyr) is more similar to that of time slices with low summer insolation (Fig. S9a, 104kyr) than high insolation (Fig. S9a, 22kyr, 70kyr). The influence of ice-sheets on the interhemispheric temperature gradients forces the zone of convergence and uplift to move south (Fig. 1c) resulting in greater influence from high southern latitudes. When obliquity is high there is larger south Atlantic warming at high southern latitudes vs. mid southern latitudes in austral summer (Fig. S9). Sensible and latent heat flux increase more at extratropical southern latitudes and shifts the zone of convergence further south. Cloud cover and rainfall occur further south and earlier in the austral summer (Fig. S9).
There are low correlations with global interhemispheric radiative balance or global temperature gradients, particularly when ice-sheet forcing is introduced. However, the movement of the rainbelt correlates most highly with the regional tropical interhemispheric temperature gradient ( Fig. S9b and e), influenced by both land and ocean, but with increased oceanic influence when ice-sheets are included (Fig. S9e) because of the overall southward shift in the ITCZ, where there is a lower proportion of land. The findings suggest that while the terrestrial rainbelt responds to local changes in insolation and other forcings the oceanic rainbelt responds to regional interhemispheric gradients.  Table S1. Details of the palaeorecords used in the synthesis map of Fig. 1c     [Created using NCL ( https://www.ncl.ucar.edu) version 6.2.1]  The two right hand columns correlate the southern Atlantic rainbelt/ITCZ limit with the tropical (0-30°) interhemispheric sea surface temperature gradient and the wider Atlantic region surface temperature gradient (between 50°W and 0°E). The global mean rainbelt southern limit is highly influenced by the Atlantic, and although when only orbital forcing drives the model this leads to a strong influence of the IRB, the non-linear impacts of ice-sheets and their influence on the sensitivity of the rainbelt position to obliquity results in much lower correlation of global IRB/ITG with the ALL southern limit. Only the Atlantic tropical ITG (including local land influence) produces high correlations with the Atlantic rainbelt southern limit for both the ORB-ONLY and ALL simulations.  Fig. S13 Monthly mean latitude of the southern Atlantic rainbelt limit over the last glacial cycle for austral summer (January to April) from the (a) ORB-ONLY experiment, and (b) ALL experiment. The comparison shows how at times when the rainbelt is closest to the equator, the most southerly latitude occurs in April, whereas when the rainbelt is farthest south the most southerly extent occurs in February. In the ALL experiment the change in seasonality of the southernmost extension of the rainbelt at times when obliquity is high (~44 kyr and ~ 96 kyr) is clear.