Human settlement of East Polynesia earlier, incremental, and coincident with prolonged South Pacific drought.

The timing of human colonization of East Polynesia, a vast area lying between Hawai'i, Rapa Nui, and New Zealand, is much debated and the underlying causes of this great migration have been enigmatic. Our study generates evidence for human dispersal into eastern Polynesia from islands to the west from around AD 900 and contemporaneous paleoclimate data from the likely source region. Lake cores from Atiu, Southern Cook Islands (SCIs) register evidence of pig and/or human occupation on a virgin landscape at this time, followed by changes in lake carbon around AD 1000 and significant anthropogenic disturbance from c. AD 1100. The broader paleoclimate context of these early voyages of exploration are derived from the Atiu lake core and complemented by additional lake cores from Samoa (directly west) and Vanuatu (southwest) and published hydroclimate proxies from the Society Islands (northeast) and Kiribati (north). Algal lipid and leaf wax biomarkers allow for comparisons of changing hydroclimate conditions across the region before, during, and after human arrival in the SCIs. The evidence indicates a prolonged drought in the likely western source region for these colonists, lasting c. 200 to 400 y, contemporaneous with the phasing of human dispersal into the Pacific. We propose that drying climate, coupled with documented social pressures and societal developments, instigated initial eastward exploration, resulting in SCI landfall(s) and return voyaging, with colonization a century or two later. This incremental settlement process likely involved the accumulation of critical maritime knowledge over several generations.


Core Extraction Methods and Description
To capture the full sediment sequence we deployed a UWITEC gravity-type corer with core catcher and a barrel length of 60 cm; this allowed us to capture the sediment-water interface and most recent sediments. Sediment loss was avoided with the core catcher that seals the bottom of the core in situ before withdrawal through the water column. For subsequent sediment depths a 120 cm-long Livingstone-type corer (Geocore) was employed with both 5 cm and 3 cm core diameters to retrieve overlapping cores. All cores were kept intact and stored in an airtight tube during transport and when placed in cold storage (+4 °C). The cores were split longitudinally and underwent low-frequency magnetic susceptibility analysis and micro X-ray fluorescence (μXRF) analysis before being sub-sampled at contiguous 1 cm intervals.
A sequence of overlapping cores was obtained for Lake Te Roto to an overall depth of 783 cm. The sequence is laminated gyttja throughout, alternating between light grey and black lamina. (Fig S1). In Lake Lanoto'o 1 , a sequence of overlapping cores was obtained to an overall depth of 302 cm. There is some visual stratigraphic change throughout the sequence, primarily alternating between two units of very dark brown gyttja and strong brown silicilastic sediment. In Lake Emoatul, a sequence of overlapping cores was taken from the deepest part of the lake (7.1m) giving an overall sediment depth of 3.40m. Visual stratigraphy showed laminations of alternating dark brown lake gyttja and darker brown-black organic rich layers. The lower 10 cm of the core were grey clay, which prevented further penetration and sampling. All cores were correlated using LOI, magnetic and Itrax data, with core depths then re-mapped onto a single depth model that has been used for age/depth modelling ( Figure S2).

Core Chronometric Analyses
To build precise chronologies for upper core sediments (top 0.5 m), 210 Pb was measured (via its granddaughter 209 Po) using double aqua regia acid leaching of sediment, 209 Po spiking, and autodeposition onto silver discs to determine 210 Pb activities using alpha spectrometry 2 . A chronology was generated from excess 210 Pb activity data for each core by applying the constant flux:constant sedimentation (CF:CS) model.
The age-depth models are constrained by 210 Pb, 137 Cs and 14 C AMS ages and in Vanuatu, a single dated tephra. The 14 C ages were calibrated using the SHCal13 atmospheric curve 3 . BACON, the software used for Bayesian age-depth modelling, negates the effects of outlying dates because the ages are modelled using a student-t distribution with wide tails 4 . Given the greater confidence generated by 210 Pb, 137 Cs , tephra and 14 C SLM sample dates, these ages were assigned more narrow Gaussian error distributions reflecting their greater reliability. Thus, in the final Bayesian models, age measurements are constrained by the more robust short-lived materials. All ages throughout the article refer to modelled ages based on the resulting age:depth model. The model is provided in the data link at the start of this SI.
The Lake Emoatul (Vanuatu) age model is based on seven AMS 14 C analyses on SLM, specifically terrestrial monocot leaf fragments and in one case a seed, from throughout the core and 33 210 Pb analyses from the upper 0.5 m. A distinctive tephra geochemically linked to the large, local Kuwae eruption of c. AD 1457 provides an additional chronological constraint. Final age model uncertainties around the proposed period of East Polynesia colonization (c. AD 900-1200) were ±85 yrs (Table S1, Fig. S2). The age model for the Lake Lanoto'o (Samoa) core is derived from 14 AMS 14 C analyses on bulk sediments, three on short lived leaf fragments (SLM), one on an unidentified species of wood, along with 21 210 Pb dates from the upper 0.5 m of the core. The resulting calibrated age model provides 2σ uncertainties of ±59 yrs around the presumed period of migration into East Polynesia (Table S2, Fig. S2). An absolute age model for Lake Te Roto (southern Cook Islands) is provided by 24 210 Pb dates (from top 0.5 m) and 20 AMS 14 C dates from throughout the core (see Table S3, Figs. S1 and S2).The final age model uncertainties associated with the period of likely anthropogenic disturbance are ±140 yrs. In the Te Roto core, terrestrial monocot leaf fragments were preferentially sought for 14 C dating, but were not always available, requiring some bulk sediment analyses (see Table S3, Figure 2). Results on bulk sediment are typically older than those on equivalent-depth SLM, by between 30 to 284 radiocarbon years. Given the finely laminated nature of the sediments, we suggest this does not reflect sedimentary disturbances, but rather the inclusion of older carbon in bulk sediment samples as is typical when dating bulk materials. In light of this, while all dates were included in the final Bayesian model 3 , higher weights were given to SLM results (Fig. S2).

Archaeological Settlement Chronologies
Archaeological research on Aitu has historically focused on the island's surface architecture and caves. In the 1960s, Roger Duff 5 and Michael Trotter 6 recorded 32 sites, including marae (temple sites), caves with human burials, and habitation areas. Steadman 7 explored some 20 caves on Atiu in an effort to document the island's indigenous avifauna-most of these showed little sign of human habitation. Steadman speculated that human arrival on Aitu was probably contemporaneous with that on Mangaia Island, noting that the makatea on the former was somewhat less rugged. Kurashina, Stevenson and Sinoto 8 mapped and stabilised some of the island's marae.
In 1987 Allen and Steadman tested a habitation terrace (designated Site 33) on the margins of a swamp to the north of Lake Te Roto that had been used for taro cultivation in the past. A single charcoal sample from the lowest occupation layer returned a conventional 14 C age of 280 ± 80 (Beta-27438). Calibrating to c. AD 1460-1950 (2σ; median AD 1662), the result suggests use of this swampy area pene-contemporaneous with the late prehistoric spikes in burning and soil erosion recorded in the Lake Te Roto core. This occupation layer was overlain by a thick clay deposit that included a considerable number of land snails, suggesting removal of native forest or shrub at this time.
Although no early settlement sites have been formally identified on Atiu, Walter 9,10 excavated a 14 th century village complex on nearby Ma 'uke, an island in regular contact with Atiu at the time of European arrival. Walter 10 observed that Atiu is a comparatively high and well-watered island. Its lake, extensive swampland suitable for wetland taro cultivation, and patches of good volcanic soil might have made it a more attractive island for early settlement relative to nearby Ma'uke and Mitiaro.
Most recently, bio-archaeological research was carried at the Rimu Rau burial cave by Angela Clark and colleagues 11 . This sizable limestone cave complex, along with two others, are in close proximity to Lake Te Roto. Clark and team identified more than 600 individuals, interments traditionally associated with a major battle between rival tribes. Although Clark et al. 11 attempt to make a case that the interments date to early settlement period, none of the cave materials were dated and distinctive artefacts that might have suggested relative ages were lacking. Given the oral traditions, it seems more likely that the Rimu Rau interments are largely, if not solely, late prehistoric in age.
To compare the results from Lake Te Roto with current archaeological chronologies, we recalibrated the extant corpus of 14 C analyses from the earliest archaeological contexts for the southern Cook Islands of Mangaia and Aitutaki ( Fig S3). This dataset includes samples of shortlived plant material and bones of the introduced Polynesian rat (Rattus exulans) recovered from Zones SZ1B and SZ2 at the Mangaian site of MAN-44 12,13 . The Aitutaki samples (mostly coconut endocarp) come from the earliest contexts at two mainland sites located on Aitutaki's west coast (Ureia/AIT-10 and Hosea/AIT-50), and the earliest occupation in the Moturakau Rockshelter (MR-1) located on a small offshore islet 14,15 . The 14 C results were calibrated using OxCal ver. 4.3 16 and the SHCal13 atmospheric curve 3 .

Micro-and Macro-charcoal Analyses
Micro-and macro-charcoal have been shown to reflect regional and local burning, with the latter frequently interpreted in Pacific island sediment archives as an indicator of the arrival of humans 17 . Samples were analyzed in the Te Roto (Atiu) core for both microscopic (<125 μm) and macroscopic (>125 μm) charcoal, following standard protocols 18,19 . Micro-charcoal samples were mounted on slides using silicon oil and counted under a high-power microscope at 600x magnification. Macro-charcoal samples were placed into a grooved Perspex sorting tray (Bogorov tray) with a groove 5 mm deep and 5 mm wide and counted under a stereo microscope. Samples were initially scanned at 20x magnification with charcoal identification confirmed at 40x magnification. A minimum count sum of 500 items (i.e., the sum of microscopic charcoal particles and Lycopodium exotic markers) per sample was used. For macroscopic charcoal (>125 μm), all fragments present within the sample were counted. Concentrations (particles/cm 3 ) were multiplied by sediment accumulation rate (cm/yr -1 ) to obtain charcoal accumulation rates (CHAR; particles cm -2 yr -1 ).

Faecal Sterol Analyses
Faecal sterols have been used to address questions relating to prehistoric animal husbandry, human palaeodemography, and the arrival of humans in virgin landscapes 20,21,22 . However, archaeological applications have been largely limited to temperate environments 21 ; the present analysis is the first application to tropical Polynesia, a region where pigs, dogs, and rats are nonnative, having been introduced by human colonists. Faecal sterols often enter lake environments via runoff and thus typically inform on human activities within the wider lake catchment 22 . Faecal Sterols (including faecal sterols (5β-stanols)) have low water solubility and are mainly adsorbed to particulate organic matter 20 ; consequently they are not prone to leaching 23 .
For the Te Roto lake core, 9 samples obtained between 426 cm (3695 cal BP) and 144 cm (c. AD 1404) sediment depth were analyzed for faecal sterol ratio values representing pig and human excrement', of which 7 were in the period c. AD 0-1404 (Fig 2). All samples earlier than c. AD 883 (800-966 2σ) were below the sterol ratios indicative of the presence of pigs or humans 20 . Analyses were conducted following standard protocols 24 . Briefly, 10 l of androstanol (0.1 mg/ml) was added as an internal standard to each sample of approximately 1 g of dried, homogenized sediment. Lipid compounds were extracted with solvents (DCM:MeOH, 3:1) using Microwave Assisted Extraction 25 , saponified and separated into neutral and acid fractions using aminopropyl SPE columns. The neutral fraction of each sample was then separated using silica gel column chromatography to isolate the sterol fraction. The sterol fraction was trimethylsilylated using 30 l N,O Bis(trimethylsilyl)trifluoroacetamide (BSTFA)/ trimethylcholorosilane (TMCS) (99:1 v/v) and heated at 70 °C overnight. Excess BSTFA-TMCS was removed by drying gently under nitrogen. Samples were dissolved in 50-100 l of ethyl acetate prior to gas chromatography-flame ionization detection (GC-FID) and gas chromatography-mass spectrometry (GC-MS) analysis. GC-MS analyses were performed on an Agilent 7890B GC injector (280 °C) linked to an Agilent 5977B MSD (electron voltage 70eV, source temperature 230°C, quad temperature 150°C multiplier voltage 1200V, interface temperature 310°C) in full scan mode (50-600 amu/sec). Separation was performed on an Agilent fused silica capillary column (HP-5, 60 m, 0.25 mm ID, 0.25 um df), with Helium as a carrier gas. The sample (1l) was injected in splitless mode (1 min splitless time). Sterol derivatives were analyzed using the following temperature program: 50 °C (held for 2 min) to 200 °C at 10°C min -1 then to 300 °C at 4 °C min -1 and held for 20 min. GC-MS peaks were identified through comparisons with known mass spectra (NIST08 20 ) and standards where possible. Analytes were quantified based on internal standards.

Plant Lipid Analyses
In all three lakes we reconstructed hydroclimate using hydrogen isotope ratios of biomarker proxies. In Emoatul (Vanuatu) and Lanoto'o (Samoa) we analyzed the algal lipid biomarker dinosterol according to the methods detailed in 26 . In Lake Te Roto (Atiu), the periodic lake-ocean connections precluded use of  2 Hdinosterol since hydrogen isotope ratios in algal lipids vary with salinity 27 and changes in this lake's salinity may not be correlated with changes in hydroclimate, also variable contributions from different species of marine dinoflagellates and freshwater lake dinoflagellates could complicate hydroclimate interpretation. Instead we used hydrogen isotopes in terrestrial leaf wax lipids 28,29 . δ 2 HDinosterol (Lake Emoatul and Lanoto'o) Sediment subsamples (1 cm-thick) were removed from split cores or from field-sectioned material. Dinosterol was extracted, identified, quantified, and purified using 2-step column chromatography and HPLC following procedures detailed in 26,30 . Prior to HPLC, samples were acetylated using acetic anhydride with a known hydrogen isotopic composition (-123.8 ± 8.2‰).  2 Hdinosterol values were measured via gas-chromatography isotope-ratio mass spectrometry (GC-IRMS) using instrument conditions outlined in 26 . The H3 + factor 31 was measured prior to every sequence and was 1.74 ± 0.02 ppm nA -1 during the 2.5 months that 6 sequences were run for Vanuatu and was 1.98 ± 0.33 ppm nA -1 during the 2 months that 4 sequences were run for Samoa. External standards of known hydrogen isotopic composition (Dr. Arndt Schimmelmann, Indiana University, http://mypage.iu.edu/~aschimme/compounds.html) were injected throughout each run. Any peak areas less than 11 Vs were disregarded to avoid size dependent  2 H effects 32 and isotopic compositions were evaluated in the Isodat 2.0 software relative to calibrated H2 reference gas.  2 Hdinosterol values were corrected using the regression of known versus Isodatreported n-alkane standard  2 H values.  2 Hdinosterol values were then corrected for hydrogen added during acetylation by a mass balance calculation as in 33 . For Vanuatu 40 samples were injected on average 3.4 times and had a pooled analytical uncertainty of 4.0‰, for Samoa 20 samples were injected on average 4.1 times and had a pooled analytical uncertainty of 7.2‰. Using the  2 Hdinosterol values produced for Lake Emoatul (Vanuatu) and Lake Lanoto'o (Samoa), precipitation was calculated with the  2 Hdinosterol -GPCP core top calibration 26 using equation (1): where Pp is the paleoprecipitation rate (mm/day),  2 Hdinosterol is a down core measurement of sedimentary dinosterol, b is the intercept of the regression (211 ± 15), and m is the slope of the regression (12.1 ± 2.6). Uncertainties were calculated using a Monte Carlo approach with 100,000 iterations with normally distributed errors from analytical uncertainty in  2 Hdinosterol measurements plus calibration error in the slope (±2.6) and intercept (±15).

Leaf Wax Biomarkers (Lake Te Roto)
The hydrogen isotopic composition of plant leaf waxes (δ 2 H lw) including the long-chain n-alkyl compounds that comprise the waxes, is largely controlled by the hydrogen isotopic composition of a plant's source water δ 2 H 35,28 . n-alkanoic acids are straight-chain hydrocarbons, with the main chain length, carbon number distributions and isotopic composition dependent on the source organism 36 . Since C3 and C4 plants fractionate meteoric water differently leading to changes in biomarker δ 2 H values, we used higher chain length (C26) terrestrial-sourced δ 13 C biomarker records to determine hydrogen isotope ratios so as to avoid salinity affects potentially found in lake-based carbon sources 37,36 .
Sub-samples were taken from cores and freeze-dried. For δ 2 H determinations each sample was first dissolved in a solution of hexane which contained sacrificial compounds ethyldecanoate and pentadecane at ca. 0.3 mg/cm 3 . Compound-specific δ 2 H determinations were performed using a ThermoScientific Trace 2000 gas chromatograph coupled to a ThermoScientific Delta V via a GCIsolink and Conflow IV interface. All analyses were conducted in duplicate. The instrument performance was evaluated and a H3 + factor calculated on a daily basis using H2 reference gas (stable and <2 ppm/mV) 31 . Data were initially calibrated to two H2 reference peaks injected directly into the ion source, before being normalized using the equation of a line from a plot of measured versus known δ 2 H values for a standard suite of 15 n-alkanes (C16-C30; Mixture B3, Arndt Schimmelmann, University of Indiana) which was injected prior to every two sample runs. Peak heights <50mV identified on the GC-IRMS were below the cutoff and not considered to avoid size-dependent δ 2 H effects 38 . Instrument error was typically less than 5‰, calculated using the same n-alkane standard. As the δ 2 H for the esterification agent was not obtained, a correction for the addition of three Hs added during methylation to the C26 n-acid δ 2 H value has not been undertaken for this site. As the trends will be the same, a qualitative interpretation of changes in the δ 2 H values is sufficient to determine paleoenvironmental and paleoclimatic changes. δ 13 CTOC plot within generalised C3 terrestrial plant values, indicating that changes in δ 2 HC26 are not a consequence of changes between C3 and C4 plants 29 . Moreover, δ 2 HC26 values track changes in Ti/inc and XLF magnetics (both indicative of terrigenous inwash) such that more negative (less negative) δ 2 HC26 are associated with higher (lower) Ti/inc and XLF (Fig S4). Thus we interpret changes in δ 2 HC26 with qualitative changes in precipitation amount in Atiu for the period of record.

Regional Climate Proxies
For regional climate proxies we used published data from lake or swamp sites across the Pacific (Table S4). We converted all proxy data records for the past 2,000 years into normalized z-scores and averaged the values for two time periods 39 : AD 900 to 1150, the period of arrival and establishment in the southern Cook Islands (based on results from this study) and the Society Islands (see main text), and AD 1150 to 1300 the period in which much of the remainder of East Polynesia was colonized. We plotted the average z-scores in Figure 4 using graduated proportional circles in ARCMAP 9.2, color-coded to indicate wet and dry. We assume the values reflect relative magnitude of changes in wetness. S1. Stratigraphy of the main Te Roto core (T1-3). The first two columns, optical and x-ray images of core T3-1, show the alternating light:dark laminations that indicate no disturbance or evidence of bioturbation. Coloured line graphs show the associated Titanium and Silica μXRF geochemistry. The position of AMS radiocarbon samples taken from this core are identified together with sample type, uncalibrated radiocarbon age and radiocarbon laboratory number (see Table S3 for further details). The additional radiocarbon sample from sediment depth 202.5 cm is not shown as it was sampled from a different core (T2-1). All macrofossils other than the one seed are fragments of monocot leaf fragments, tentatively identified as terrestrial grasses, but the species were undeterminable (see Table S3). Blue arrows denote the locations of the major changes in proxies as shown in Figure 2 of the main text.  Overlap with the Lake Te Roto lake sequence is indicated by the colored bars that show Aitu core dates for discovery (green), colonization (yellow), and settlement (purple). The archaeological records, with one exception, post-date Atiu Island discovery as identified by the faecal sterol records of the present study (see main text for detailed discussion).

Fig. S4.
Comparison between δ 2 HC26 leaf wax (precipitation proxy) and Ti/inc (runoff proxy). δ 2 HC26 axes inverted so drier conditions are less negative. Dry phase c. AD 850-950 corresponds with lower Ti/inc and least negative δ 2 HC26 in the last 2,000 years. Peaks in Ti/Inc. and δ 2 HC26 also correspond indicating wetter conditions. We conclude that, qualitatively, the Atiu biomarker record based on compound specific terrestrial plant leaf wax hydrogen reflects changes in precipitation amount.    45