Redating of the To’aga site (Ofu Island, Manu’a) and a revised chronology for the Lapita to Polynesian Plainware transition in Tonga and Samoa

Radiocarbon dating Pacific archaeological sites is fraught with difficulties. Often situated in coastal beach ridges or sand dunes, sites exhibit horizontal and vertical disturbances, while datable materials such as wood charcoal are typically highly degraded, or derived from old trees or drift wood and bone collagen rarely survives in the tropical conditions. Shell, therefore, is the most logical material for dating Pacific sites since it is resistant to alteration, can be sampled to ensure only the last few seasons of growth are represented and is often closely tied to human economic activities. However, shell radiocarbon (14C) dating has been plagued by interpretive problems largely due to our limited knowledge of the 14C cycle in near shore marine and estuarine environments. Consequently, shell dates are typically ignored in regional chronometric evaluations and often avoided for dating altogether. Recent advances in our understanding of the source of shell 14C content as well as the development of the first South Pacific Gyre model of changing marine 14C over time, combined with Bayesian statistical modelling, have now provided us with insight into the value of these shell radiocarbon dates, enabling a revision of the age of the To’aga site on Ofu Island, an early occupation site associated with the initial Polynesian Plainware period in Samoa, the earliest use of which is now dated to between 2782 and 2667 cal BP.


Introduction
The Lapita cultural complex, of which dentate-stamped pottery is a defining component, is argued to have begun in the Bismarck Archipelago around 3350 years ago. From there it spread into previously uninhabited areas of Remote Oceania (Fig 1), reaching its eastern-most extent in the Tongan and Samoan archipelagos around 2900-2750 BP [1][2] [3]. The timing of first landfall on each island group by these Lapita explorers has been the subject of a number of chronological evaluations (e.g., [4][5] [6] [7][8] [9][10] [11]) but several parts of this story remain controversial.
Of note are debates over the timing of the transition from dentate-stamped Lapita ceramics to undecorated ceramics (termed Polynesian Plainware [PPW]) -a significant change in material culture that is considered to mark the onset of Ancestral Polynesian society [12]. There is no doubt that Lapita colonizers reached the island of 'Upolu in Western Samoa, where the submerged coastal site of Mulifanua (SU-17-1) contains classic dentate-stamped Lapita ceramics 3 of local manufacture [13] [14]. The exact date of this settlement remains somewhat problematic because the material was obtained by dredging during the construction of a ferry terminal [15].
Radiocarbon dates (here reported at 68% probability) obtained on material collected from the dredging spoils produced a date on turtle bone and shell of 2880-2750 cal BP [16]. Comparably early charcoal dates have also been obtained from the sites of 'Aoa (AS-21-5) on Tutuila Island, and To'aga (AS-13-1) on Ofu Island in the Manu'a group. Dates from these sites suggested that occupation either pre-dated, or was contemporaneous with, Mulifanua [17][18] [19], but neither 'Aoa or To'aga produced ceramics with dentate stamped decoration. Instead, red-and orangeslipped thin, fine tempered plainware was recovered from the deepest units of both sites. These ceramics were considered to represent a distinct marker horizon between the earliest layers and those that contained thicker, coarse-tempered pottery which became dominant after 2400 cal BP ( [20] pg 91).
Rieth and Hunt [21] evaluated all available 14 C dates from the Samoan archipelago. Their analysis "challenged the validity of the earliest dates from 'Aoa and To'aga" because of their low precision (standard errors in excess of ±100 years). This left only shell dates from To'aga for consideration, suggesting a ~300 year separation between the ceramics from Mulifanua and the earliest occupations at To'aga (2500-2400 cal BP), and two other early sites at Utumea (Tutuila Island; 2500-2100 cal BP; AS-  and Jane's Camp (Upolu; 2300-2000 cal BP; SU-18-1, SU-F1-1). Addison and Morrison ( [22] pg 363) subsequently concluded that the limitations of current radiocarbon technology and calibration methods, combined with the problems of stratification and mixing of sandy Pacific coastal sites, meant that further refinement of the absolute chronology for early Samoa was unlikely.
4 This post-2500 BP "re-colonization" date was questioned by Clark et al. [23] who presented new data from three PPW sites on the island of Ofu: Va'oto (AS-13-13), Coconut Grove (AS-  and Ofu Village (AS- . Combining these results in a single-phase Bayesian model populated by a combination of short-lived Cocos nucifera endocarp charcoal and highly precise U/Th coral dates, Clark et al ( [23] pg 272) concluded that initial settlement of Ofu Island occurred between 2717 and 2663 cal BP. This date overlapped with modelled dates for the end of Lapita from sites on Tongatapu, Ha'apai, and Vava'u (2703-2683 cal BP) where a chronological progression from Lapita to PPW sites had been identified [24], suggesting settlement of Ofu soon after the loss of Lapita ceramics in Tonga. The question of whether the Lapita-PPW transition in Samoa represented two discrete settlement events, or a single transitional event, remained unresolved.
While new data using both 14 C and new dating techniques (U/Th), combined with reevaluation of existing data, provide great promise for resolving many chronological issues in the Pacific, the avoidance of key dating materials and the use of single-phase Bayesian evaluations unconstrained by stratigraphy or other form of independent dating control, effectively leave us with an imprecise chronology that is smeared over many decades. Ultimately, we are still left with an inability to truly resolve important chronological questions: 1. When was the earliest occupation? 2. How fast did people spread? 3. Was settlement continuous? 4. From what direction did settlement spread?
In this paper, we address some of these issues by presenting a re-investigation of the chronology of the To'aga site using a suite of new and precise shell and bone AMS 14 C dates taken from key contexts. Based on these new 14

The To'aga excavation and radiocarbon dates
The To'aga site (AS-13-1), situated on the southern coast of Ofu Island, was first test excavated in 1986 with more extensive excavations being undertaken in 1987 and 1989 [19]. The site consists of stratified cultural deposits situated within a coastal beach terrace located on the southern side of Ofu Island. The terrace was archaeologically investigated primarily through the excavation of 1-m 2 test pits arrayed along a series of six transects running perpendicular to the coast (Fig 2).
These transects revealed buried cultural deposits, some of which contained PPW ceramics, shell fishhooks, ornaments, and other artefacts and associated faunal remains. During the 1987 field season, an expanded trench was opened up to further sample the deeply buried, ceramic-bearing deposits, and is referred to below as the "Main Excavation". The original suite of radiocarbon dates from all three excavation seasons are presented in Kirch [20]. Radiocarbon dates and associated information is given in Table 1.

Comment on Dates Charcoal
Short-lived nut charcoal samples with only 1 year of growth are considered to be one of the most reliable dating materials, assuming there has been minimal stratigraphic displacement. It is also well-established that most wood charcoal determinations will date earlier than the event by an unknown amount [26]. This could be by a few years, or several hundred years, but Pacific research has indicated that, except in the highest precision analyses, minor inbuilt age typically goes unnoticed [27] [28]. A large number of extant radiocarbon dates from Pacific contexts are on charcoal that has not been identified to short-lived materials, and removal of these unidentified charcoal dates from chronological evaluations would result in a dataset composed of relatively few 11 dates. It has been demonstrated in Bayesian models that a small dataset has a greater detrimental impact on chronological resolution than the potential of unidentified charcoals to skew to older ages [29]. Outlier analysis specifically designed to account for this inbuilt age in charcoal is promising [5] [29], but requires additional constraints (e.g., dates on short-lived materials or stratigraphy) against which to anchor the model (see also [30] where the model allows a small number of charcoal samples with inbuilt age to be younger than the context they represent, as would be the case with intrusive material).

Bone
Rattus exulans (Pacific rat) have a notoriety stemming from anomalously early dates from New Zealand archaeological contexts [31]. A range of theories have been put forward for these dates including the small size of these bones and potential laboratory contamination at the time of dating We are not aware of any rat specific 14 C studies into dietary corrections for island environments, but evaluation of 14 C dates of other omnivorous animals (i.e., humans, pigs and chickens) from Pacific contexts [9][34] suggests a similar correction methodology is required.

Marine Shell
Marine shells that were gathered for food precisely date the timing of this activity. However, shell radiocarbon dates remain problematic because of uncertainties over 14  (2016), support these observations (Fig 4).

Bayesian analysis
To provide the most probable chronology for To'aga, we conducted a Bayesian Sequence Analysis using OxCal 4.3.2 whereby radiocarbon ages are ordered on the basis of stratigraphic evidence [59]. In this model we have grouped the dates into two phases; Early (Transect 9, Unit 23 and Transect 5, Unit 28) and Main Excavation Layer (Units 6, 9 and 10) separated by a contiguous boundary. Within this "Early" phase, Unit 23 is further divided into three phases within a separate sequence that overlaps with the dates from Unit 28.
To assess the likelihood of any one sample being an outlier, a General t-type Outlier Model is inset into the sequence [60]. This enables outliers to be either too young or too old, and downweights their influence in the model [26]. These dates are assigned a prior outlier probability of

Results
There are differences in modelled age depending on whether a R value of -47±76 or -161±47 14 C years has been used (

Discussion
This multiphase sequence and outlier analysis provides the most secure age for the earliest deposits at To'aga available so far, and places first use of the site between 2782-2667 cal BP. This earliest phase of activity continues until 2613 cal BP (represented by boundary "Transition" in Table 2 and Fig 5).
Subsequent activities in the Main Excavation take place between 2727 and 2104 cal BP (boundaries "Transition" and "End"), possibly incorporating material from many phases of activity that have become mixed together as a result of continual activity in the same general location. How do these new ages compare to the chronology of Lapita and PPW in Samoa and Tonga and is it possible to shed any further light on the key chronological questions for the region: when was the earliest occupation; how fast did people spread; was settlement continuous; and from what direction did settlement spread? In an attempt to answer these questions, and to highlight discrepancies in our current knowledge of regional chronology and radiocarbon methodology, the following regional comparison has been undertaken. Even though we can refine the chronology movement through the islands using U/Th dating, as has been done for Tongan Lapita sites, this has not been possible for To'aga, and has not been as successful for defining PPW. In part, this is because of limited numbers of suitable culturally modified corals from key archaeological sites. Even the corpus of 14 C dates on shortlived materials from early sites is limited (see [21]). While it is difficult to give any 21 recommendations as to the number of dates required to improve our findings further, Schmid et al.
( [29] pg 67) has suggested in such large-scale phase models ~280 dates will produce results of the highest precision, while the most accurate results are achieved where sampling density is uniformly distributed. Moreover, while many researchers question the usefulness of using 14 C dating especially across the "Radiocarbon Plateau" (ca. 2650-2350 cal BP) there is structure present during this 300-year flat section of the calibration curve which could be utilised if larger numbers of precise (±20 years or better) dates on short-lived material were obtained from secure multiphase contexts. Clearly, more work is required across this region and time period.

Conclusions
Our new dates and re-analysis of the site chronology indicates that the best estimation of the initial use of the To'aga site is between 2782 and 2667 cal BP. This confirms the antiquity of the site relative to other PPW sites on Ofu Island in Samoa, but we cannot confirm or refute an overlap in settlement timing with the Lapita site of Mulifanua on 'Upolu. Our findings also suggest that the initial occupation at To'aga was contemporary with the terminal Lapita/PPW transition in Tonga, but we cannot definitively say which came first. What is apparent, is that settlement in Samoa occurred early and is likely to have continued in a near unbroken sequence since Lapita times.
Over the decades, radiocarbon dates and the interpretation of those dates have become more sophisticated, but as research themes develop and new dating technologies become involved in the debates it has more than ever become necessary to refine the issues -both in archaeological research and radiocarbon methodology -that still plague the chronological interpretation.
Chronometric hygiene methodologies initially provided the means to some clarity, but unfortunately removed a high proportion of dates from consideration. The limited number of early 22 sites throughout the Pacific means we cannot afford to ignore the evidence already collected and where data are no longer considered of highest precision it is essential that extant excavations with curated samples are revisited. Bayesian methodologies now offer new opportunities to test these assumptions, as does refinement in 14 C variation in both the marine and terrestrial reservoirs.