Glacial forcing of central Indonesian hydroclimate since 60,000 y B.P.

Sulawesi lies at the center of the humid, unstable air mass overlying the IPWP, and its climate responds strongly to large-scale changes in regional atmospheric circulation and sea surface temperature associated with the Australian–Indonesian summer monsoon (AISM). The Lake Towuti basin receives ∼2,700 mm of precipitation annually and is surrounded by dense closed-canopy rainforest (12). This region experiences a wet season from December to May when the intertropical convergence zone (ITCZ) migrates southward over Indonesia (13). During this season strong northerly flow associated with the AISM, warm sea surface temperatures, and strong local convective activity (14) maintain regional precipitation at >250 mm/mo. Precipitation falls below 200 mm/mo from July to October, when much of southern and central Indonesia experiences a dry season (13). During this time, the ITCZ is displaced northward, and cool sea surface temperatures and strong southeasterly flow associated with the east Asian summer monsoon suppress regional convective activity (14).

Our paleoclimate record is based on high-resolution core scanning and organic geochemical analyses to reconstruct changes in surface runoff and terrestrial vegetation (Materials and Methods). We performed X-ray fluorescence (XRF) elemental scanning augmented by discrete analyses of elemental concentrations to correct for water content effects to determine sedimentary Ti content (Figs. S2 and S3), which is used to reconstruct climate-driven changes in clastic inputs to sedimentary basins (15). Ti provides a relatively simple and redox-insensitive proxy for the integrated processes of rainfall, erosion, and fluvial discharge from Towuti’s ∼1,500 km² catchment.

We also measured the carbon-isotopic composition of long-chain, even-numbered n-alkanoic acids (δ¹³C_wax), a main component of plant epicuticular waxes (SI Materials and Methods and Fig. S4). The δ¹³C_wax is primarily used to distinguish between plants using C₃ and C₄ photosynthetic pathways (16) because C₄ plants use a CO₂-concentrating mechanism that improves their photorespiration and water-use efficiency relative to C₃ plants (17). The δ¹³C of epicuticular waxes from C₄ plants, which are mainly tropical and warm season grasses, typically range from −14‰ to −26‰, and waxes from C₃ plants range between about −29‰ and −38‰ (18–20). Precipitation is the dominant control on the distribution of C₃ and C₄ plants through much of the tropics, such that δ¹³C_wax has been widely used to reconstruct past changes in tropical hydroclimate (21, 22).

In modern plants, δ¹³C_wax varies not only due to changes in plants’ photosynthetic pathways, but also due to the influence of edaphic factors (water availability, temperature) (23) and plant biome structure (closed-canopy vs. open-canopy forest; ref. 24). Wetter conditions permit more efficient leaf–gas exchange and stronger fractionation against ¹³CO₂, and carbon recycling under closed-canopy forest can cause carbon-isotopic depletion of CO₂ (25). Recent global surveys of the δ¹³C of vegetation indicate these processes can cause ∼6‰ depletion in the δ¹³C of leaf matter in tropical rainforests relative to more xeric C₃ ecosystems, and ∼4‰ depletion in tropical rainforests relative to drier, tropical deciduous forests (23). Thus, we interpret more depleted δ¹³C_wax to represent C₃ forests growing in wet conditions; whereas, enriched δ¹³C_wax reflects a drier climate and increasing C₄ grasses.

Vegetation changes on glacial–interglacial timescales, including changes in C₃ and C₄ plant communities, are controlled not just by precipitation, but also temperature, atmospheric CO₂ concentrations, soil moisture, and other environmental factors (26). Similarly, sedimentary Ti concentrations are affected by changes in lake levels, soil chemical weathering, and other processes that affect clastic sediment deposition and erosion. Despite these differences, sedimentary Ti concentrations and δ¹³C_wax in Lake Towuti exhibit coherent orbital-scale variations over much of the past 60 ky B.P. (SI Materials and Methods and Fig. S5), demonstrating that the glacial–interglacial variability in these proxies is mostly driven by changing precipitation.

Ti concentrations and δ¹³C_wax show strong glacial–interglacial variability over the past 60 ky B.P. (Fig. 2). Low Ti concentrations indicate reduced precipitation at ∼60 ky B.P., during late marine isotope stage (MIS) 4, followed by an abrupt rise in Ti at ∼58 ky B.P. reflecting a shift toward wetter conditions. At the base of our record, δ¹³C_wax averages −32.4‰, and exhibits an abrupt ∼4‰ depletion at 58 ky B.P., synchronous with the rise in Ti. This ¹³C depletion could result either from a reduction in the abundance of C₄ grasses, or from an increase in precipitation that altered the structure of C₃ forests in Lake Towuti’s catchment, confirming an abrupt onset of wet conditions at the MIS 4/3 boundary. Ti concentrations are high and δ¹³C_wax averages −37.8‰ between ∼58 and 39 ky B.P., recording a wet climate and closed-canopy rainforest in central Sulawesi during much of MIS3 (SI Materials and Methods). Ti concentrations fall, and δ¹³C_wax becomes somewhat more enriched, beginning ∼39 ky B.P., followed by an abrupt transition at ∼33 ky B.P. to relatively dry conditions. The δ¹³C_wax averages −25.0‰ between 30 and 18 ky B.P., indicating substantial C₄ grass expansion during the LGM.

Ti concentrations are low during the LGM and reach a minimum at 16 ± 0.4 ky B.P., coincident with Heinrich event 1 (H1; Fig. 3) (27), providing additional evidence of the significance of H1 to tropical hydroclimate (28). Heinrich events 3 and 4 are also associated with declines in Ti concentrations, and aridity at Lake Towuti ∼60 ky B.P. could be correlated to H6, one of the strongest Heinrich events in the North Atlantic (29). However, we see no clear evidence for H2 or H5, nor do we see evidence for the Younger Dryas, which appears to have had a limited impact over much of Indonesia (8). The expression of all six Heinrich events of the past 60 ky B.P. in speleothem δ¹⁸O from Borneo (11), to the north of our site, coupled with the variable expression of Heinrich events in our record from Sulawesi, could reflect latitudinal gradients in the regional sensitivity of Indo-Pacific hydrology to millennial-scale forcing from the North Atlantic. This is perhaps related to variations in the strength and/or underlying mechanisms of the Heinrich events themselves and their ability to perturb regional rainfall.

Ti concentrations abruptly increase at 14.8 ± 0.38 ky B.P., synchronous with the abrupt onset of wetter conditions recorded in speleothem δ¹⁸O records from China (30) (Fig. 3A). The δ¹³C_wax reaches its Holocene average of −37.1‰ at 11 ± 0.28 ky B.P., recording the development of closed-canopy tropical rainforest that persists today. Ti concentrations are also high in the early Holocene, but fall during the mid-Holocene. The δ¹³C_wax exhibits a muted but simultaneous mid-Holocene enrichment of ∼1.5‰, indicating that the mid-Holocene reduction in precipitation implied by sedimentary Ti was insufficient to strongly perturb the region’s rainforests under Holocene boundary conditions. Both proxies indicate a return to moist conditions in the late Holocene, similar to records from southern Indonesia (11, 29) and likely related to intensification of the austral summer monsoon (11).

Our most significant findings are the existence of a very wet climate during much of MIS3 and the Holocene, interrupted by abrupt transitions to and from a dry LGM. No prior record from Sulawesi contains sediments covering the LGM, let alone the past 60 ky. However, sediment cores from the shoreline of Lake Tondano, North Sulawesi (Fig. 1), suggest a wet climate ca. 35 ky B.P., a moist early Holocene, and a discontinuity from ∼31 to ∼13 ky B.P. interpreted to reflect low lake levels (31), a history very similar to that of Lake Towuti. Even more discontinuous palynological records from the Wanda peatland, near Lake Towuti, suggest rainforest predominated during late MIS3 and the Holocene and grasslands expanded during the late LGM (12), supporting our interpretation of δ¹³C_wax. Shorter but high-resolution reconstructions from marine sediments offshore from western Sulawesi suggest a moistening trend from the late glacial into the Holocene (32). Thus, our data appear compatible with existing reconstructions from Sulawesi, and suggests a strong response of regional climate to glacial–interglacial forcings over the past 60 ky B.P.

In contrast, recently published speleothem δ¹⁸O records from northern Borneo exhibit relatively little glacial–interglacial variability (Fig. 3), and instead suggest that boreal fall insolation at the equator, and hence precessional forcing, plays a dominant role in Indo-Pacific precipitation and convection (10) (Fig. 3B). These differences could suggest significant heterogeneity in LGM climate over Indonesia and the climate responses of these regions to remote and local forcings. Although the precise mechanisms that link changes in October insolation to Borneo δ¹⁸O on orbital timescales are not known (10), in the present day seasonal and interannual precipitation variability in northern Borneo and our site are not strongly correlated. Borneo experiences little to no precipitation seasonality, and the austral winter precipitation minimum in central Sulawesi is characteristic of much of central and southern Indonesia (13). Moreover, variability in the δ¹⁸O of precipitation at Borneo is most strongly correlated to precipitation variations over the oceans north of Borneo and Indonesia, and weakly or not correlated to precipitation variability in central and southern Indonesia (33). However, as there is widespread evidence for aridity during the LGM in Indonesia (7, 8), it is also possible that the δ¹⁸O–precipitation amount relationship was weakened during the LGM, perhaps due to the effects of competing processes such as changes in precipitation sources, moisture trajectories, and upstream convective effects (10). Indeed, precessional forcing is least evident in Borneo δ¹⁸O during MIS2, which could reflect the effects of strong glacial forcings.

Our record alone cannot reconcile these differences, yet the wet climate we observe during much of MIS3 and the Holocene, interrupted by abrupt transitions to and from a dry LGM, has important implications for the mechanisms controlling climate variability in this region. This history suggests that the climate of Sulawesi is highly sensitive to glacial–interglacial forcings, and that the region may experience nonlinear, threshold responses to changes in global climate boundary conditions. Climate modeling studies indicate that IPWP hydrology could respond to a variety of glacial–interglacial forcings including regional processes, particularly exposure of the Sunda Shelf during sea level minima (34), as well as global forcings, including changes in global temperature and atmospheric greenhouse gas concentrations (GHGs) and remote forcing from ice sheet albedo and topography (5, 6). All of these forcings reached their minima or maxima during the LGM, potentially triggering threshold responses in regional hydroclimate. Because of their short nature, most previous proxy reconstructions have only evaluated the impacts of these forcings across the last glacial termination, when large and nearly synchronous changes in shelf exposure, temperature and GHGs, and ice volume complicate the attribution of precipitation changes. Our long, continuous record from Lake Towuti sheds light on the impacts of these forcings across a broader range of changing boundary conditions of the past 60 ky B.P.

Previous work suggests that the exposure of the Sunda Shelf could play an important role in transmitting glacial–interglacial signals from the high latitude to the IPWP (30, 35). In many climate model simulations, the exposed land cools strongly relative to the ocean, and, together with reductions in atmospheric latent heating and humidity due to the replacement of ocean with land, weakens atmospheric convection and induces widespread drying over Indonesia (34). Although Sunda Shelf exposure during sea level lowstands could contribute to the dry conditions we observe on Sulawesi during the LGM, the history of shelf exposure is very different from the pattern of climate variability we observe at Lake Towuti. In particular, the shelf had nearly the same exposed area during the LGM and MIS3 (Fig. 2C; 2.3 vs. 1.9 × 10⁶ km², respectively), yet we observe large differences in Sulawesi hydrology between these stages. Thus, shelf exposure alone did not trigger the threshold responses in rainfall that we observe at Towuti. Rather, to explain our observations this mechanism would necessitate large threshold responses in regional rainfall to small changes in the area of exposed land. This possibility has not been investigated in climate models. Moreover, although Sulawesi is not connected to the Sunda shelf and its climate might be less sensitive to shelf exposure, the Borneo δ¹⁸O record also shows little influence of shelf exposure despite the location of that site on Sundaland (10). Taken together, these data suggest shelf exposure alone cannot impose a dry climate across this region.

Alternatively, Sulawesi hydrology could be highly sensitive to changes in global temperature, GHGs, and/or remote forcing by the ice sheets. Lower GHGs reduce precipitation rates in convective regions such as the IPWP through their control on global temperatures and atmospheric water vapor concentrations (4, 35, 36), yet climate model experiments suggest that average tropical precipitation declined by only ∼10% during the LGM (35). Although our data cannot precisely quantify past precipitation changes, the shift from rainforest to more open terrestrial ecosystems almost certainly requires a >10% reduction in regional precipitation in light of the global distribution of C₄ grasslands. The IPWP could have experienced a stronger GHG-driven reduction in precipitation than the tropical mean due to regional temperature and humidity changes; however, regional sea surface temperature records indicate similar temperatures during MIS3 and MIS2 (Fig. 2A) when Towuti experienced large changes. Thus, although GHG forcing may have contributed to drying at Lake Towuti, it is unlikely to be the primary forcing of the pattern of hydrologic change observed between MIS3 and the Holocene.

On the other hand, the drying observed at Lake Towuti from late MIS3 to the LGM closely tracks changes in global ice volume (Fig. 2D; ref. 9), suggesting a strong remote forcing by the Northern Hemisphere ice sheets during the LGM. Indeed, glacial–interglacial variations in proxy data from Lake Towuti are most strongly correlated to changes in ice volume/sea level among these glacial–interglacial forcings (Table S3). Climate modeling studies indicate that both albedo and orographic forcing by the LGM ice sheet can alter tropical Pacific climate. Albedo forcing can alter the interhemispheric thermal gradient, shifting the ITCZ southward (37), and topographic forcing can alter westerly flow over the North Pacific, intensify northern trade wind circulation, and also force the ITCZ to migrate southward (6). The change in tropical winds and moisture convergence shift the locus of western tropical Pacific precipitation southward from equatorial sites such as Lake Towuti.

Although threshold responses of tropical Pacific precipitation to ice sheet forcing have not been explicitly investigated in modeling studies, the Pacific climate has been shown to be extremely sensitive to differences in ice sheet topography. Simulations of the LGM climate with the Community Climate System Model 3 with differing ice sheet topography can produce very different patterns of atmospheric circulation and temperature over the northern Pacific as well as differences in tropical Pacific SST gradients (38). Given this sensitivity, we suggest that the growth of the Laurentide Ice Sheet during late MIS3 and MIS2 could have triggered a threshold response of the Pacific ITCZ, causing the abrupt changes we observe in Lake Towuti’s hydrology. Northwesterly flow from the east Asian winter monsoon, which is also sensitive to remote forcing from the northern ice sheets, strengthened during late MIS3 and the LGM (39), and could also contribute to a southward shift in the locus of precipitation in the region, suppressing rainfall during Sulawesi’s rainy season. Although much of northern Australia was dry during the LGM (8), speleothem records from Flores do suggest a southward shift of the ITCZ during the LGM (40), as does a wet MIS2 climate observed at Papua New Guinea (41). Existing data are too sparse to fully document the spatial patterns of IPWP climate during the last 60 ky B.P., and it is certainly possible that the threshold behavior we observe in Sulawesi could result from nonlinear interactions between multiple glacial–interglacial forcings. However, these regional patterns suggest that dynamical changes in the atmospheric circulation, potentially driven by northern high-latitude ice sheet forcing, exert a strong influence on orbital-scale changes in western tropical Pacific precipitation, as suggested by climate modeling experiments (4). Future modeling experiments could better assess thresholds and nonlinear interactions among these forcings in the western Pacific.

Our data also highlight the importance of glacial–interglacial changes in hydroclimate in structuring the vegetation on Sulawesi, one of the most biologically complex and diverse regions on Earth. Glacial–interglacial sea level variations are thought to play a dominant role in the phytogeography of this region through the expansion of land surface areas available for terrestrial ecosystems during Sunda Shelf exposure (42, 43). Indeed, modeling studies of Indonesian biomes suggest that wet evergreen lowland forests are currently in a refugial stage due to the flooding of the Shelf and geographic constraints on plant distributions (42). This stands in contrast to regions in which glacial climate change drove ecological bottlenecks. Our δ¹³C_wax data challenge this prediction, and instead shows that Sulawesi rainforest contracted substantially in a relatively arid glacial climate. This would suggest that glacial–interglacial variations in both climate and geography play an important role in structuring Southeast Asian evergreen forests, which may migrate between climate refugia during glacials and geographic refugia during interglacials.

Records of surface runoff and vegetation from Lake Towuti highlight the strong sensitivity of Sulawesi hydrology and ecosystems to glacial–interglacial climate forcing and particularly forcing from the high-latitude ice sheets. The similarity of regional climate during the Holocene and MIS3, and the abrupt transitions into and out of MIS2, implies that central Indonesian hydrology can exhibit nonlinear, threshold responses to glacial climate forcing, tipping rapidly between wet interglacial and dry glacial climates. Although existing records are inadequate to fully evaluate the history of Indo-Pacific hydrology at glacial–interglacial timescales, such behavior, if widespread, could amplify climate change and ice sheet growth during glaciations through atmospheric water vapor feedbacks and help to synchronize abrupt temperature changes during glacial transitions.

Sediment cores were recovered from Lake Towuti using a modified Kullenberg piston corer with core-site selection guided by a high-resolution seismic stratigraphy. Sediment core IDLE-TOW10-9B (abbreviated TOW-9) was recovered from 154 m water depth and is 1,156 cm long. The age model for core TOW-9 is based on 23 radiocarbon dates yielding average sedimentation rates of 19 cm ky⁻¹ (SI Materials and Methods). We carried out high-resolution elemental analysis using an ITRAX XRF core scanner equipped with a Cr tube. Selected samples were also prepared for elemental analysis using flux fusion and measured on an inductively coupled plasma atomic emission spectrometer (FF-ICP-AES). Ti counts measured by XRF were corrected for water content variations and other matrix effects and show a strong correlation to Ti concentrations measured by FF-ICP-AES (SI Materials and Methods).

We also measured the carbon isotopic composition (δ¹³C) of long-chain n-alkanoic acids (C₂₆, C₂₈, and C₃₀), components of terrestrial leaf waxes. Lipids were extracted by accelerated solvent extraction (Dionex) with CH₂:Cl₂:MeOH (9:1). Alkanoic acids were purified from the resulting extract, methylated, and the fatty acid methyl esters were analyzed by gas chromatography–isotope ratio mass spectrometry (SI Materials and Methods). All δ¹³C_wax data are corrected for the isotopic composition of the methyl group added during methylation. The pooled SE of samples measured in triplicate or greater is 0.47‰. See SI Materials and Methods for detailed analytical procedures.

We thank Gerald Tamuntuan, Satrio Wicaksono, and PT Vale Indonesia for field assistance; and David Murray, Joe Orchardo, and Rafael Tarozo for laboratory assistance. This material is based upon work supported by the National Science Foundation under Grant 0902845 (to J.M.R.). Research permits for this work were granted by the Indonesian Ministry of Research and Technology (RISTEK).