Human occupation and ecosystem change on Upolu (Samoa) during the Holocene

To track the peopling of the South Pacific and assess their impact on terrestrial and aquatic ecosystems.


| INTRODUC TI ON
Tracking the initial spread of humans around the globe allows an assessment of how humans have shaped ecosystems and landscapes.
Identifying the first arrival of humans into a landscape, however, is often challenging because populations were likely small, transient and often did not produce materials that are frequently preserved (Gosden, 1993;Graves & Addison, 1995). Measures for assessing the arrival and impact of humans in ecosystems can be divided into two categories: (a) direct measures, which seek to obtain physical evidence of past human activity and (b) indirect measures, which infer past human activity from secondary observations. Aside from the discovery of sites or artefacts representing a past material culture, one of the most effective direct measures for tracking past human activity has been found to be the presence, and abundance, of ancient (fossil) charcoal found in soils or sediments (following Whitlock & Larsen, 2001).
The use of charcoal to track past human activity has been shown to be particularly effective in tropical settings where natural fire is limited due to either a lack of ignition source or flammability, that is, the appearance of fire is dependent on the arrival of humans (Argiriadis et al., 2018;Burney et al., 2004;Gosling et al., 2017;Huebert & Allen, 2016).
The South Pacific Islands (Oceania) are believed to contain some of last ecosystems on Earth to be colonized by humans (Figure 1). Indirect evidence from linguistic and genomic analysis suggest that the dispersal of Austronesian-speaking peoples into Oceania began c. 5,000 calibrated years before present (cal BP) when it had reached as far as the Solomon Islands (Gray, Drummond, & Greenhill, 2009). Subsequently, people from the Bismarck Archipelago dispersed east c. 3,000 years ago and reached the easternmost islands of Polynesia within the last 1,000 years (Gray et al., 2009;Matisoo-Smith, 2015). These dates are supported by direct archaeological evidence that suggests the homonid occupation of New Guinea began during the late Pleistocene (c. 50,000 years ago; Summerhayes, Field, Shaw, & Gaffney, 2017), but that people did not reach western Polynesia until c. 2,850 cal BP (Burley, Edinborough, Weisler, & Zhao, 2015) or the easternmost islands, such as Rapa Nui (Easter Island), until c. 900-800 cal BP (Wilmshurst, Hunt, Lipo, & Anderson, 2011). The timing of human dispersal across the South Pacific in the Holocene has been linked to heightened periods of El Niño activity that elevated drought frequency and likely pushed peoples to migrate eastwards (Anderson, Gagan, & Shulmeister, 2007).
The impact of the dispersal of human populations across the South Pacific, and the interplay between humans and climatic factors, in modifying ecosystems is the subject of ongoing debate (Kirch, 2017;Nunn, 2007;Spriggs, 2014). It is likely that ecosystems on islands in the South Pacific were impacted by removal, and addition, of resources by the first peoples (Anderson, 1952;Flenley & King, 1984;Gosden, 1993;Prebble et al., 2019;Prebble & Wilmshurst, 2009).
Many South Pacific island ecosystems are also known to be sensitive to changes in moisture availability (Mueller-Dombois & Fosberg, 1998), such as those resulting from changes related to the El Niño Southern Oscillation (ENSO) (Hassall, 2017;Zhang, Leduc, & Sachs, 2014). Similarly, short-lived abiotic events such as volcanic or tsunami events have been shown to have influenced the ecosystems on small islands through burial (Spriggs, 2014) or arrival of new species via rafting (García-Olivares et al., 2017). Unravelling the different drivers of ecosystem change has become important for understanding the dynamics of natural ecosystems and for discovering the specific impacts resulting from the arrival of humans. Here we present a new c. 10,500 year record of fire and ecosystem change from Upolu (Samoa) and assess the relative importance of humans, volcanic activity and climate in shaping terrestrial and aquatic ecosystems.

| Geographical setting
The Samoan archipelago lies within the western Pacific Ocean and comprises 10 islands (13-14 o S, 170-173 o W). Just two islands, Savai'i and Upolu, make up 96% of the landmass (total 2,934 km 2 ; Figure 1). The Samoan islands were formed though volcanic activity during at least the last c. 1 million years as a consequence of the movement of the Pacific Plate (Hawkins, 1976;Kear, 1967;Kear & Wood, 1959). The geology of Samoa predominantly comprises basaltic cones that have been heavily eroded during the Pliocene and early Pleistocene, and they are now buried by late Pleistocene lava flow deposits (Stearns, 1944). The last volcanic eruption in Samoa occurred on Savai'i in AD 1911 and on Upolu at some unknown time during the Holocene (Venzke, 2013). The majority of the modern Samoan flora is of Malaysian origin and can be divided into four groups broadly defined by altitude (Table 1).
The climate of the Samoan archipelago is controlled by the interplay between the South Pacific Convergence Zone (SPCZ), the Trade Winds and the Inter-Tropical Convergence Zone (ITCZ). Mean monthly minimum (23°C-24°C) and maximum (29°C-30°C) temperature are settlers on the vegetation seems to have been minimal; however, a subsequent opening up of the landscape is suggested through the gradual increase in ferns. The absence of any significant change in the aquatic community associated with, or after, the arrival of people on the islands suggests that humans rarely visited the lake. We suggest that on Upolu a simple model of decreasing human impact away from coastal areas is applicable.
algae, aquatic, charcoal, cyanobacteria, fire, Pacific, pollen, Polynesia, terrestrial, vegetation F I G U R E 1 Climatic and geographical context of study region. (a) Distribution of precipitation across the globe, precipitation to the Samoan archipelago is influenced by changes in the South Pacific Convergence Zone (SPCZ) and the Inter-Tropical Convergence Zone (ITCZ) data from GPCPV2.2 (1979GPCPV2.2 ( -2014. (b) Samoan archipelago (inset) and Upolu Island (yellow box), position of Lake Lanoto'o study site indicated by red box. (c) Topographic map of the area around Lake Lanoto'o, position of core site indicated by red dot. (d) Photograph of Lake Lanoto'o taken from the rim of the volcanic crater, position of coring platform on lake indicated by red arrow. (e) Bathymetric map (upper panel) with depth indicated in meters below lake surface, and cross section of Lake Lanoto'o (lower panel), cross section in lower panel follows the line from i to ii, position of core site indicated by red dot [Colour figure can be viewed at wileyonlinelibrary.com] relatively constant throughout the year, but precipitation can vary from > 400 mm (January) to < 150 mm (July). The climate of Upolu is classified as tropical, with a mean annual temperature of 26.6°C and mean annual precipitation of c. 2,800 mm (Lagomauitumua, Seuseu, & Faasaoina, 2011), with the south-east trade wind belt delivering slightly more rain to the southern region. No distinct dry season exists, but during November and April it is generally warmer and wetter with a greater frequency of tropical storms and hurricanes (Mueller-Dombois & Fosberg, 1998). Strong year-to-year climatic variation has been observed to occur related to the movement of the SPCZ during different ENSO phases, that is, the La Niña phase delivers more wet season rainfall, while the El Niño phase leads to relatively drier conditions (Lagomauitumua et al., 2011). Sedimentary archives of past climate on Samoa have been reconstructed over the Holocene and show abrupt changes in precipitation proxies interpreted to indicate regional changes in the position and intensity of the SPCZ (Hassall, 2017).
Patterns of social organization are not thought to have greatly altered through the period of human occupation of Samoa (Wallin & Martinsson-Wallin, 2007); however, a four-step sequence of settlement patterns has been suggested (Green, 2002;Morrison et al., 2015): they have not, to date, been directly associated with external factors; however, the subsequent arrival of Europeans on the islands is seen as a major driver of change (Wallin & Martinsson-Wallin, 2007).

| Study site: Lake Lanoto'o (Upolu island)
Lake Lanoto'o (171°50′W, 13°54′S) is a 0.11-km 2 volcanic crater lake at the centre of Upolu island c. 760 m above sea level (a.s.l.) with a catchment area of 0.23 km 2 . The Salani volcanic deposits that define the region are thought to have formed between c. 310,000 and 64,000 years ago (Kear & Wood, 1959). Today the lake is c. 400 m diameter with a maximum depth of 17.5 m (Figure 1), in the surface layers lake water is pH 7, with a temperature of c. 27°C, oxygen saturation of 105% and a conductivity of 15 μs/cm (measured in September 2014). At c. 10-m depth, a thermocline results in a relatively abrupt change to cooler (23°C) anoxic conditions (dissolved Oxygen 10%), pH 4 and increased conductivity 21 μs/cm (Hassall, 2017). Lake Lanoto'o is boarded by steep slopes that reach up to 790 and 770 m a.s.l. to the east and west, respectively. The soils that surround the lake today contain highly weathered red lateritic clay with low silica but high TiO 2 content (Wright, 1963) and are vegetated with dense montane forest (Olson et al., 2001;Whistler, 1992).

| Core recovery
A sequence of overlapping cores was obtained from the deepest region of Lake Lanoto'o in September 2014 ( Figure 1). A UWITEC gravity-type corer was used to recover the upper 60 cm. A cam-modified TA B L E 1 Summary of composition of major vegetation groups on Samoa (Olson et al., 2001)  piston corer was used to retrieve overlapping sediment cores to a depth of 302 cm below the mud-water interface (Colinvaux, Oliveira, & Moreno, 1999). All cores were stored intact in airtight tubes and kept in cold storage (+4°C) at the University of Southampton (UK).
Subsamples were subsequently extracted from the cores for radiocarbon dating and palaeoecological analysis.

| Core analysis
The cores were correlated on the basis of sedimentary characteristics determined by loss-on-ignition (LOI), magnetic susceptibility and Itrax core scanning data ( Figure 2). True core depths (cm below mud-water interface) were then re-calculated to form a single composite depth model that was, in turn, used for age versus depth modelling.
Loss-on-ignition analysis was applied to contiguous 1-cm 3 subsamples (following Lamb, 2004). Whole cores were measured for volume-specific low-frequency magnetic susceptibility using a Bartington Instruments MS2K sensor at contiguous 1-cm intervals (following Dearing, 1994). Geochemical analysis was undertaken using an Itrax core scanner (Cox Analytical Systems, Gothenburg, Sweden; following Croudace, Rindby, & Rothwell, 2006) and %C and %N (used to calculate C/N) using an elemental analyser. A molybdenum tube (30 kV, 30 mA) was used to scan each core at 500-µm resolution and at 200 µm for the surface gravity core. The analytical dwell time was set at 30 s.

| Age versus depth model
Radionuclide dating was used to generate and age model for the Lake Lanoto'o sediments. For the surface gravity core, the activity of 137 Cs and 210 Pb was measured using Canberra well-type HPGe gammaray spectrometers (Canberra UK Ltd., Didcot, UK-now Mirion Technologies). The gamma ray spectra were acquired for 100,000 s for each contiguous 0.5 cm samples and processed using Radiocarbon ( 14 C) measurements were obtained from 18 bulk sediment samples using the acid-alkali-acid pre-treatment method at SUERC (Table 2). All dates were reported in conventional 14 C years before AD 1950, with analytical confidence expressed at the ± 1σ interval. Radiocarbon dates were calibrated to calendar ages using the SHCal13 curve for southern hemisphere (Hogg et al., 2013).
All dates were used to create an age versus depth model for Lake Lanoto'o in BACON 2.2 Bayesian modelling software (Blaauw & Christen, 2011; Figure 2). While the majority of the dates are stratigraphically consistent throughout the sequence, BACON highlights two samples (LAN14-2-2 11-12 cm and LAN14-1-3 17-18 cm) that are outliers, the former being older than expected and the latter being younger than expected.

| Palaeoecological analysis
Two different sets of subsamples were processed from the Lake Lanoto'o sediment core to extract different types of palaeoecological information: (a) microscopic remains, to reconstruct regional fire history (microcharcoal), vegetation history (pollen), lake status (algae and cyanobacteria) and volcanic eruptions (cryptotephra) and (b) macroscopic remains, to reconstruct local fire history (macrocharcoal) and local vegetation (plant remains). Algae and cyanobacteria were identified through comparison with reference material at the University of Amsterdam and published atlases (Canter-Lund & Lund, 1995;van Geel, 1976van Geel, , 1978Medeanic, 2006;Poulíčková, Žižka, Hašler, & Benada, 2007). Algae and cyanobacteria were counted until sums of > 300 had been achieved.
Cryptotephra layers were identified following subsampling of contiguous 5-cm long sections throughout the entire core sequence.
Cryptotephras were extracted by sieving to recover the 125-25 µm fraction, and density separation using sodium polytungstate; cleaning float of 1.95 g/cm 3 , extraction float of 2.5 g/cm 3 (following Blockley et al., 2005). The supernatant of the extraction float was mounted on glass slides in Canada Balsam and scanned for the presence of glass shards under an optical microscope fitted with cross-polarizing filters. The numbers of shards were then counted and concentrations per gram of sample (dry weight) calculated. Where tephra shards were identified the sample resolution was refined to 1 cm.

| Macroscopic remains: Macrocharcoal, plant remains and sediment composition
In all, 86 subsamples of 1 cm 3 were prepared for macroscopic analysis through bleaching with 3% H 2 O 2 on a hotplate at 150°C for 15 min, and sieving at 160 µm. Identification of macrocharcoal, plant remains and sedimentary material was done simultaneously using a Leica MZ16 stereo microscope at 0.71-11.5× magnification.
Macrocharcoal was identified based upon colour (black), shape (angularity) and breakage pattern when pressure was applied (splintering), following Whitlock and Larsen (2001). Macrocharcoal particles were counted and digitally photographed using a Fuji X-M1 camera.
The images of macrocharcoal particles were analysed in ImageJ to calculate area (Rasband, 2008). A threshold of ≥ 5% of the maximum TA B L E 2 Radiocarbon ages obtained from Lake Lanoto'o, presented as 14 C years before AD 1950. The δ 13 C is expressed as relative to the VPDB (δ 13 C VPDB ) from Hassall (2017

| Statistical analysis
Detrended correspondence analysis (DCA) was used to characterize the major trends in the pollen and spore, and algae and cyanobacteria datasets. Percentage abundance data for all taxa were used in both cases. The DCA analysis was performed in R 3.5.1 (R Core Team, 2018) using the package Vegan 2.5-2 (Oksanen et al., 2017).

| Sediment characteristics
The Lake Lanoto'o sediment core was found to comprise two distinct sediment types: (a) orange-coloured (Munsell 7.5YR 5/6) sediments with high titanium (Ti/inc), high magnetic susceptibility (K), high C/N ratio and low organic carbon, and (b) dark brown coloured (Munsell 7.5YR 2.5/5) sediments with low titanium (Ti/inc), low magnetic susceptibility (K), low C/N ratio and high organic carbon (Figure 3).
At one point in the core (181-180 cm, c. 6,837 cal BP), rounded silicate sand (> 160 µm) was found within the sediment in an obliquely slanting layer that extended over 7.3 cm.

| Microcharcoal
Prior to the first archaeological evidence of humans on Upolu (c.

| Algae and cyanobacteria
Algae and cyanobacteria were found in high concentrations (mean 1,500,000 per cm 3 ) throughout the Lake Lanoto'o sediment core, except for the period between c. 6500 and 5300 cal BP when concentrations drop dramatically (< 250,000 per cm 3 ) ( Figure 5).
During periods of high concentration, the algae and cyanobacteria assemblage is comprised predominantly of Botryococcus (11%-50%). During the period of low concentration, the assemblage mainly comprises four types of algae/cyanobacteria that have not been taxonomically assigned: Type 128A (up to 25%), Type 128B (up to 54%) and Type A (up to 21%; van Geel, 1976van Geel, , 1978. There is no change observed in the algae and cyanobacteria community coincident with the first archaeological evidence of humans on Upolu.

| Tephra
Cryptotephra particles were found in three samples from the Lake  Figure 2). Within these samples between 70 and 10,000, tephra particles per gram were found.

| Macrocharcoal
Macrocharcoal is absent from the Lake Lanoto'o sediments prior to the earliest archaeological record of humans on the island around c.
2,800 years ago (Figure 3). Within the last c. 2,800 years, macrocharcoal fragments were found in nine of the 32 (28%) samples analysed.
Macrocharcoal particle counts within the last c. 2,800 years ranged from 0 to 9, with the maximum area of charcoal recorded occurring at c. 1,200 cal BP (0.86 mm 2 ); all but one of the occurrences of macrocharcoal exceeded 5% of the maximum area value (0.04 mm 2 ).

| Plant remains
Three types of macroscopic plant remains were identified within the sediments from Lake Lanoto'o. One of these was identified as fern sporangia, whereas the other two were identified as seeds from an unknown plant (Appendix 1). The fern sporangia and Type-2 plant remain from Lake Lanoto'o were present intermittently throughout the entire record. The Type-1 plant remains first appears in the Lake Lanoto'o record c. 4,200 cal BP.

F I G U R E 4
Summary diagram for pollen, spore and macroscopic remains from Lake Lanoto'o (Uplou, Samoa). Pollen and spore taxa shown represent all taxa occurring at > 5% in at least one sample. For images of macroscopic remains see Appendix 1. Notation follows Figure 3

| Fire and the human occupation of Upolu
Macro-and microcharcoal are commonly interpreted as evidence of burning in the local and regional environment (Clark & Royall, 1996). The definition of what is considered 'local' versus 'regional' is depended on the site-specific setting and conditions, important factors include topography and prevailing wind direction (Whitlock & Larsen, 2001). In the case of Lake Lanoto'o, the macrocharcoal signal could be derived from two sources: (a) burning within the small catchment (0.23 km 2 ) immediately around the lake and (b) burning on the coast c. 9 km to the south; fires > 0.02 km 2 can generate convective columns > 1,000 m (Palmer & Northcutt, 1975) that are capable of transporting macrocharcoal c. 10 km (Clark, 1988). It seems, however, unlikely that macrocharcoal material could reach Lake Lanoto'o from beyond the shores of Upolu as the main neighbouring island is c. 50 km northwest. Microcharcoal, however, is known to travel 10s kms from its source (Clark, 1988), so could have reached Lake Lanoto'o from fires anywhere on Upolu or on a nearby island.
The first appearance of macrocharcoal in the Lake Lanoto'o sediments occurs at c. 2,700 cal BP, and is (within analytical uncertainty) within the dates from the earliest archaeological evidence for human occupation on Upolu (Figure 3). The close agreement between the macrocharcoal and the archaeological evidence suggests that humans were necessary to provide a source of ignition and/or sustain fire on Samoa, and mirrors similar patterns seen on Vanuatu, New Caledonia and Fiji (Hope, Stevenson, & Southern, 2009). The abundance of macrocharcoal found within the Upolu sediments is low (below 10 particles per cm 3 ) when compared with abundances found in sediments known to be associated with large-scale landscape clearance (above > 200 particles per cm 3 ) on Mauritius . The relatively low abundance of macrocharcoal on Upolu suggests that either fire activity within the lake catchment was very limited, or that larger scale fires elsewhere on the island contributed material via long-distance transport. Both interpretations fit with the idea that initial colonizers focused on coastal areas (Cochrane et al., 2016).
Archaeological evidence indicates that settlements had appeared throughout the Samoan archipelago by c. 2,100 cal BP (Rieth & Hunt, 2008 (Figure 3). The continued presence of charcoal within the Lake Lanoto'o sediments confirms that humans continued to be present on the island during the Samoan Dark Ages, and corresponds with preliminary evidence of increased inland settlement at this time (Morrison et al., 2015).
The idea of more widespread past human populations on Upolu is supported by recent analysis of LiDAR data which shows a wider distribution of human made structures across the island than was previously thought (Jackmond et al., 2018).
The microcharcoal record follows the major trend of the macrocharcoal with persistent occurrence only after the first archaeology, that is, microcharcoal is above background levels in all samples after 2900-2700 cal BP (Figure 3). However, the microcharcoal record suggests that fire was present on Upolu, and/or nearby islands, prior to the start of the archaeological record (background levels are 4,000 cal BP, respectively, which are attributed to lightening strikes during drier El Niño conditions (Hope et al., 2009). This and the absence of archaeological or sedimentary evidence to support early human populations on intervening islands currently favours the climatic hypothesis; however, alternatively this peak may just represent a chance fire event caused by volcanic activity or a lightening strike somewhere in the region.

| The impact of humans on the ecosystems of Upolu
Lake Lanoto'o has a diameter of c. 400 m suggesting that around half of the pollen captured within the sediments likely came from further than 100 m from the lake (termed 'regional' following Jacobson & Bradshaw, 1981). The prevailing winds from the south-east likely transport pollen upslope and consequently bias, and expand, the range of the signal to vegetation found on the southern side of the island. The pollen data indicate that Moraceae/Urticaeae, Myrtaceae, Weinmannia and ferns were the major components of the terrestrial vegetation for most the last c. 10,500 years ( Figure 4). Significant changes are observed in all of these taxa before, and after, the first archaeological evidence of humans on Upolu indicating a naturally dynamic vegetation mosaic.
The certain arrival of humans on Upolu c. 2900-2700 cal BP, as indicated by the archaeological and the macrocharcoal record ( Figure 3), is coincident with changes in the relative abundance of terrestrial taxa commonly found in montane and lowland forests and a climatic drying (Hassall, 2017). Following the initial replacement of Myrtaceae with Moraceae/Urticaceae, the subsequent gradual increase in Cyatheaceae pollen is the most notable change. The gradual increase in the abundance of Cyatheaceae interestingly mirrors the pattern observed in the Galapagos (Restrepo et al., 2012) and Mo'orea (Stevenson, Benson, Athens, Kahn, & Kirch, 2017), and the increase in fern species more generally found on Rimitara (Austral Islands) (Prebble & Wilmshurst, 2009) and on Atiu, Mangaia and Rarotonga in the Cook Islands (Fujiki et al., 2014;Kirch & Ellison, 1994;Parkes, 1997). The Cyatheaceae and fern signals on these islands have been interpreted as indicative of an opening up of the landscape by humans and a similar explanation seems likely on Upolu.
The gradual nature of the suggested landscape opening on Upolu indicates an incremental human impact that mirrors trajectories of change observed on Fiji (Roos, Field, & Dudgeon, 2016), but is in contrast to abrupt impacts recorded in Rapa Nui (Flenley & King, 1984;Rull, 2016). It is likely that, in all situations, the ratio between the human population size and the amount of available land and coastal areas plays an important role on the degree of impact. Smaller islands are therefore likely to be more vulnerable to rapid landscape transformation by the arrival of humans. The incremental nature of terrestrial ecosystem change on Upolu suggests that the initial human populations may have been quite small relative to the island size, mirroring inferences drawn from data related to the size of the language phylogenies (Gray et al., 2009) and the low number of archaeological sites found during this early period (Cochrane et al., 2016;Leach & Green, 1989). Initial human impact on Hawaiian Islands exhibits a similar spatial pattern with rapid changes seen to occur shortly after occupation in coastal areas (Athens, Ward, & Wickler, 1992) while remote inland areas remain relatively unaffected (Selling, 1948).
The major component of the algal community within Lake Lanoto'o for most of the last c. 10,500 years is Botryococcus ( Figure 5). There is no significant change in the algal or cyanobacteria community composition or concentration at, or following the first archaeological evidence of humans on Upolu (c. 2,800 cal BP). The absence of any significant change in the algae/cyanobacteria of Lake Lanoto'o during the period of human occupation suggests that human activity at, or near, the lake did not impact water quality or aquatic ecosystems. The absence of any discernible impact on the aquatic community is commensurate with the interpretation from the charcoal record of a low level, or absence, of humans at Lake Lanoto'o.
Our data support the view that the impact of the arrival of humans on islands is strongly controlled by the local scale and conditions of each island (Kennett, Anderson, Prebble, Conte, & Southon, 2006;Prebble, 2006

| Non-human drivers of ecosystem change on Upolu
Humans seem to have played no role in modifying ecosystems on Upolu prior to c. 2,800 cal BP. However, since significant changes are recorded in the palaeoecological record, it is therefore necessary to look for alternative explanations ( Figure 6). Independent sedimentological evidence from Lake Lanoto'o provides some insights into the likely drivers of these changes (Figure 3).
Three volcanic events have been identified through cryptotephras within the Lake Lanoto'o sediments, but no discernable response is found in either the terrestrial or aquatic communities (Figures 4 and 5). Similar sized lakes in tropical settings have been shown to suffer community turnover in response to the deposition of volcanic tephra (Matthews-Bird et al., 2017). The absence of a response on Upolu likely reflects a comparatively low input of volcanic material. The highest concentration of cryptotephra was found between AD 1869 and 1888 (81-62 cal BP) and could represent a precursor to the documented eruption on Savai'i in AD 1911 (Venzke, 2013).
The presence of siliceous sand in the Lake Lanoto'o sediments at c.
6,700 cal BP is soon after a rapid decline in the abundance of algal and cyanobacteria found within the Lake Lanoto'o sediments ( Figure 5).
This section of the core is also subject to the greatest uncertainty F I G U R E 6 Composite diagram characterizing abiotic and biotic environmental change on Upolu (Samoa) over the last c. 10,500 years. Notation follows Figure 3 within the age versus depth model (Figure 2). The presence of sand and reversed radiocarbon dates may indicate that a significant perturbation of the sediment occurred, such as a landslide, which altered the water quality and caused the turnover and decline in the algal and cyanobacteria community. Following this rapid decline, the abundance of algae and cyanobacteria remained low within the Lake Lanoto'o record for c. 1,200 years. The recovery of the algal and cyanobacteria community at c. 5,300 cal BP is coincident with an increase in the concentration of titanium and magnetic susceptibility within the lake sediments (Figure 3), which represents increased in-wash of soil from the catchment as a result of a prolonged wetter period (Hassall, 2017), and likely resulted in a freshening of the lake.
The vegetation communities on Pacific islands have been shown to fluctuate in response to variations in past climate regime.
On Tahiti, drier and cooler climates during the late Pleistocene (c. 45,000-41,000 years ago) saw an increased dominance of grasses, ferns and sedges (Prebble et al., 2016), while droughts associated with the Medieval Climate Anomaly (AD 800-1250/1150-700 cal BP) resulted in a loss of moisture demanding Cyathea ferns (Restrepo et al., 2012). In the Lake Lanoto'o core, precipitation has been interpreted as being high between c. 5,500 and 2,500 years ago (Hassall, 2017; Ti/inc curve Figure 3). The elevated precipitation through this period does not appear to have resulted in any significant change in the vegetation community ( Figure 6). This suggests that while island ecosystems of the South Pacific are vulnerable to drought events, elevated precipitation on the levels experienced during the Holocene were not sufficient to drive species turnover or effect a major change in the balance of the community composition.

| CON CLUS IONS
Despite low abundance, the macrocharcoal record from Lake Lanoto'o detects the first human occupation of Upolu confirmed by direct archaeological evidence. The persistence of a strong charcoal signal through the Samoan Dark Ages suggests that people persisted on the island through this period. The close correlation between charcoal and human occupation of Upolu lends further weight to the argument that the investigation of ancient charcoal preserved in the sedimentary records is an effective method for detecting human occupation of ecosystems that do not naturally carry fire, even when the human populations are small.
Comparison of terrestrial and aquatic ecosystem change with the archaeological record and fire history of Upolu suggests that the arrival of humans on the island did not have a major impact on ecosystems in the central, higher elevation areas of the island. Our data support a simple model of decreasing human impact away from coastal areas towards high mountain environments with a gradual, but continual, process of opening up of the landscape; this model is in contrast to rapid, island wide, impacts identified on some other small islands. The inferred spatial gradient in human impact suggests that remote areas, such as Lake Lanoto'o, were buffered against human-induced ecosystem change and are thus important sites of understanding native biodiversity.
The long-term ecological dynamics observed in the aquatic community of Lake Lanoto'o seems to be driven by internal (sedimentological) and external (climatological) factors, rather than by human factors. In contrast the terrestrial ecosystems seem to be unresponsive to increased precipitation during the middle Holocene. The absence of major abrupt changes in the terrestrial ecosystem composition, and the likely natural drives of changes seen in the aquatic ecosystems, suggests that the early occupation of Upolu by humans did not have a significant ecological impact close to Lake Lanoto'o.

ACK N OWLED G EM ENTS
Permission for this investigation was granted by the Ministry of Natural for their helpful and positive comments that improved the manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data available from the Dryad Data Repository: https ://doi. org/10.5061/dryad.51c59 zw4j (Gosling et al., 2019).