Global Patterns in Island Colonization during the Holocene

Early Holocene agropastoral settlement of the insular Mediterranean was essentially contemporary with the full realization of Neolithic lifeways in the Levantine core of the Southwest Asian Neolithic and its expansion westward. Cyprus, closest to such Pre-Pottery Neolithic (PPN) communities, was settled first, at around 9000 BC, possibly capitalizing on prior Levantine HGF experience of the island (Knapp, 2010). This early spur out into the Mediterranean represented a cul-de-sac, however, as subsequent island colonization further west only began two or more millennia later, and was initially closely associated with the arrival of Neolithic lifestyles in the Aegean littoral. Radiocarbon dates from the basal stratum at Knossos (still the earliest known Neolithic community on Crete) suggest first settlement at around the same time as at western Anatolian sites such as Ulucak and the arrival of farming in mainland Greece (Brami, 2015; Douka et al., 2017), between 7000 and 6500 BC. This contemporaneity, and discontinuities in cultural sequences at sites with both Mesolithic and Neolithic deposits (for example, Munro & Stiner, 2015), support the likelihood that Aegean farming spread primarily through targeted demographic movements, often partly maritime in nature, from Anatolia (despite continued reservations in some quarters: Sampson, 2018). Yet, aside from Crete and Gökçeada (Atici et al., 2017)—the former the fifth largest Mediterranean island, and the latter only c.10 km from the mainland—the first Aegean farmers were apparently largely uninterested in island living, despite knowledge inherited from HGF exploration and their ongoing exploitation of insular resources such as obsidian (Çilingiroğlu & Çakırlar, 2013).

An initial preference for large islands, and/or those readily accessible from mainlands, continued as agropastoralism expanded westwards during the Middle Holocene. Sicily, Sardinia, and Corsica all have evidence for farming settlement by 6000–5700 BC (Freund et al., 2015; Lugliè, 2017), with minimal evidence for cultural, technological, or genetic overlap with antecedent HGF populations (Marcus et al., 2019). Accordingly, between c. 9000 and 5700 BC, the ‘big five’ Mediterranean islands had all been permanently settled by agropastoral communities. These communities manifestly had knowledge of, and the capacity to reach, smaller islands, even those further from the mainland, as extensive early Neolithic use of desirable obsidians from Melos and Lipari testifies. Yet at this stage smaller islands were rarely if ever desirable as targets for settlement, save for a few so close to the mainland (or to quasi-peninsular islands such as Sicily) that proximity dampened any disadvantages of smaller size (for example, certain Dalmatian and other Adriatic islands, plus Corfu). The salient exception is the Maltese archipelago, which appears to have been reached by the mid sixth millennium (Malone et al., 2019; Robb, 2001), although, as we discuss below, more substantive suites of ¹⁴C dates may indicate that this was not an ultimately successful venture. Tiny mid-Adriatic Palagruža, with a few sherds of Early Neolithic Impressed pottery (again, broadly sixth millennium BC), can be thought of as a waystation linking the Balkans to the southern Italian Neolithic farming hub of the Tavoliere, rather than permanently settled (Forenbaher & Kaiser, 2011).

The first widespread horizon of agropastoral life on the smaller, drier, and more remote islands of the Mediterranean dates most convincingly to the fifth millennium BC, both in the Aegean and the small archipelagoes that orbit Sicily. In the Aegean, this millennium witnesses the appearance of farming villages in the Cyclades, situated to take advantage of local sweet spots (Broodbank, 2000, pp. 117–49). Further west, the radiometric and ceramic-derived dates are less secure. In particular, the presence of ‘Stentinello’ wares (essentially local variations on an Impressed theme), which on Sicily can date as early as the mid sixth millennium, has traditionally been understood to indicate similarly dated settlement on the small Tyrrhenian islands (for example, Mannino, 1998). A recent analysis, however, provides Stentinello with a broader chronology of 5600–4000 BC (Freund et al., 2015, p. 208), overlapping with dates on Lipari attributed to the later ‘Diana’ horizon (Dawson, 2014, table 4.7). Evidently, a broader robust program of redating is in order. Notwithstanding a possible radiocarbon hiatus that hints at discontinuities in the colonization cycles of Malta and Gozo (Malone et al., 2019), the overall period 5000–4000 BC was one clearly of expanding insular horizons in the central Mediterranean.

The fourth and third millennia BC witness the final, most extensive expansion of settlement in the insular Mediterranean. This process involved an infilling of the last remaining gaps in the Aegean, related to a diversification in settlement types: the appearance of small-scale ‘farmsteads’ beyond the environs of larger village sites (Halstead, 2008), while the latter evolved to make new claims (in often new locations) over maritime knowledge and practices (Broodbank, 2000, pp. 211–319). Further west, the high tide of this expansion of island lifeways is witnessed by increasingly convincing evidence for permanent occupation (rather than sporadic exploitation) in the smallest Tyrrhenian islands and the relatively remote islands in the Strait of Sicily—Pantelleria, Lampedusa, and Linosa, as well as various other in-shore islets along the north- and southwestern Mediterranean littoral. The data from Pantelleria are equivocal, and permanent settlement may only, in the final analysis, prove to be a second-millennium phenomenon (Ardesia et al., 2006); but there was clearly a much earlier exploitation of the island’s obsidian, possibly mediated by maritime North African groups as well as people from the more usually assumed northern side of the Sicilian strait (Broodbank & Lucarini, 2019, pp. 221–23).

The only wrinkle in this otherwise fairly smooth pattern comes from the Balearics (Mallorca, Menorca, Ibiza, and Formentera), all relatively remote islands in the extreme west of the Mediterranean, but the first two of substantial size. The dates are contested, but the bulk of the evidence most plausibly suggests a colonization of the Gymnesics (the two larger islands) in the late third millennium BC, and of the Pityusics (smaller Ibiza and Formentera) over the course of the second or even first millennium BC (Alcover, 2008; Cherry & Leppard, 2018b; Ramis et al., 2002), within the regional Copper and Bronze Ages respectively. This interpretation may of course change, and palaeoenvironmental data do hint at earlier Middle Holocene biotic disturbance, which need not, however, be anthropogenic (Burjachs et al., 2017). In any case, the relatively late colonization of the Balearics presents an exception to Mediterranean patterning that still requires explanation.

Although the vast majority of Mediterranean islands were colonized by the beginning of the Late Holocene, and the terminal third millennium BC in that sense represents a natural stopping point for our discussion, it is important to note that the subsequent second millennium BC witnessed a prodigious shift in inter-island Mediterranean dynamics, and that further cycles of island abandonment and resurgent settlement extended longer still, with turnovers at the thresholds to and from the Classical and Roman world (earlier first millennium BC and later first millennium AD). Although the vast majority of initial insular colonizing activity had occurred well before the appearance of sailing vessels, first reliably evidenced in the East Mediterranean (probably deriving from a Nilotic tradition) around 2500–2000 BC (Broodbank, 2010), the expansion of this distance-shrinking technology integrated islands and transformed trans-basin relationships (Broodbank, 2013; Leppard et al., 2021), not least as the first Mediterranean urban societies conducted many of their interactions by sea. The establishment of the current, semi-arid Mediterranean climate regime during the Late Holocene may also have played a role in these transformations and in the periodic evidence for island abandonment, as small, dry islands were rendered more environmentally marginal, as well as easier to by-pass by increasingly long-range shipping technologies.

That the earliest dates for a human presence on the Caribbean islands come from Trinidad (c. 6000 BC) is unsurprising in biogeographic terms given its size and proximity to mainland South America (Boomert, 2000); contemporary sea-levels suggest a much shorter crossing and even a vestigial physical connection is conceivable. The other islands in the Southern Caribbean, running parallel to the South American coast, are smaller and more distant (c. 25–150 km) from the mainland, and were settled much later than Trinidad. Within this southern archipelago, there does seem to be a relationship between colonization date and island size. The larger island of Margarita has the earliest dates (c. 5000 BC) beyond Trinidad and the mid-sized islands of Aruba, Curaçao, Cubagua, and Tobago have considerably later dates (c. 3200–2400 BC), while several of the smallest and/or most remote islands were only ever seasonally occupied, or fully settled much later, or indeed never colonized at all (Hofman & Antczak, 2019).

Turning to the Greater Antilles, the earliest radiocarbon dates are also Middle Holocene and derive from Cuba, Hispaniola, and Puerto Rico (Cooper, 2010; Fernandes et al., 2020; Naegele et al., 2020; Napolitano et al., 2019; Rodríguez Ramos, 2010; Ulloa Hung & Valcárcel Rojas, 2013). The approximate date ranges for the initial colonizations of these three islands are broadly comparable (c. 4200–3200 BC), and consistent with an initial colonization of Cuba followed by Hispaniola and Puerto Rico, so plausibly from west to east. This sequence corresponds to biogeographic expectations, assuming a direct crossing from the Central American mainland first to Cuba (the largest island) and then following the islands’ configuration to Hispaniola (second largest) followed by Puerto Rico, although the scarcity of archaeological sites from this period means that the chronology is subject to debate (Ulloa Hung & Valcárcel Rojas, 2013). Current evidence indicates a strikingly later colonization (c. AD 550–950) for Jamaica (Allsworth-Jones, 2008; Callaghan, 2008), although several factors may have obscured earlier evidence (Keegan & Hofman, 2017); the likely origin points are neighboring Cuba and/or Hispaniola. In this regard, Jamaica displays similarities with the Bahamian archipelago, which also seems to have been initially peopled by colonists from both Cuba and Hispaniola in the region’s Late Ceramic Age (c. AD 700–1100) (Berman & Gnivecki, 1995; Keegan et al., 2008).

The earliest dates for the Lesser Antilles cluster on several Leeward Islands, notably Antigua, Saba and St Martin (c. 3500–2500 BC), and indicate a later Middle Holocene Archaic colonization (although Napolitano et al. [2019] would down-date this horizon). Ostensibly, this might represent an eastward extension of the aforementioned Greater Antillean processes ultimately originating in Central America (Hofman & Antczak, 2019). However, somewhat later dates for initial colonization of most of the other Leeward Islands (c. 2800–1000 BC) and several of the Virgin Islands (c. 2200–1100 BC) complicate this model. This may suggest that multiple islands were bypassed in such a first peopling of the northeastern Caribbean (Fitzpatrick, 2013; Hofman & Antczak, 2019; Napolitano et al., 2019), with various islands along the way simply visited but not initially settled, leaving little detectable trace (Bérard et al., 2016). Alternative or complementary explanations for the colonization of the Lesser Antilles advocate a separate process originating in South America, either involving a direct open-ocean crossing from the mainland to the Leeward Islands (Callaghan, 2010; Rodríguez Ramos, 2010), or a stepping stone colonization via the intervening Windwards. Regarding this last option, there is admittedly scant archaeological evidence for comparably early activity in the Windward Islands (excepting Tobago [Boomert, 2000] and Barbados [Fitzpatrick, 2011]), and for several islands the earliest dates remain Ceramic Age (Callaghan, 2010; Fitzpatrick, 2013; Giovas & Fitzpatrick, 2014; Napolitano et al., 2019), thus prompting the possibility of their colonization north-to-south from the Leewards (the ‘Southward Route Hypothesis’: Fitzpatrick, 2013). However, recent multi-island palaeoenvironmental projects provide new evidence to suggest that the initial colonization of the Windwards may have occurred earlier than previously recognized (Siegel et al., 2015, 2018). In some cases (for example, Grenada, Martinique, Marie-Galante) the first dates for landscape modification at c. 3500–3000 BC pre-date the earliest dates from archaeological sites by several thousand years, and may therefore provide support for a south-to-north stepping-stone expansion, with origins in South America (Siegel et al., 2018). In this model, taphonomic and other post-depositional processes are assumed to have destroyed or obscured the archaeological evidence of earliest settlement in the Windwards, contributing, alongside a lack of systematic surveys, coring, and palaeoenvironmental reconstruction, to the problems that bedevil much reconstruction of dynamics across the eastern Caribbean (Hofman & Hoogland, 2018). Clearly, the ‘Southward Route Hypothesis’ (Fitzpatrick, 2013; Napolitano et al., 2019) and the model proposed by Siegel et al. (2018) are to a large degree mutually exclusive, and this problem requires resolution (complicated further by the impact on radiocarbon assays of high levels of environmental variability: DiNapoli et al., 2021).

The Early Ceramic Age (c. 400 BC to AD 600/800) colonization of the Caribbean differed in several important respects from previous Archaic processes. It is generally agreed that these so-called ‘Saladoid’ migrations originated from northeastern South America and represent an expansion of Arawakan peoples (Bérard, 2013; Fernandes et al., 2020; Heckenberger, 2013; Rouse, 1992). The current earliest dates for Early Ceramic sites derive, however, from Puerto Rico and the northern Leeward Islands, a situation with strong parallels in the apparent pattern and subsequent distribution of Archaic colonizing populations (Napolitano et al., 2019). This may reflect: (1) direct voyaging from the mainland and/or Trinidad, bypassing the intervening Windwards (Callaghan, 2010; Rodríguez Ramos, 2010); and/or (2) interactions between extant Archaic populations and incoming groups (Hofman et al., 2018; Rodríguez Ramos et al., 2008). In any case, because most Early Ceramic Age sites cluster in the northeastern Caribbean, this colonization process does not correspond closely with biogeographic expectations. That apart, throughout the Early Ceramic Age there is clear evidence for population growth and settlement expansion (Curet, 2005; Rouse, 1992). This general pattern continues into the Late Ceramic Age (c. AD 600/800–1500), during which nearly all the previously unsettled islands were finally occupied, including many of the relatively small members of the Virgin Islands and the Grenadines (Giovas & Fitzpatrick, 2014; Hofman & Hoogland, 2018; Keegan et al., 2008), in addition to Jamaica and the Bahamas.

The presence of pre-established HGF populations rendered Holocene island occupation dynamics in this theater unusual in global terms. As seas rose during and after the terminal Pleistocene, extant HGF groups on newly formed islands on the Sunda Shelf (such as Borneo, Java, and Sumatra) increasingly specialized in hunting and gathering resources from rainforests and swamps, habitats that expanded with climatic amelioration (Amano et al., 2016; Rabett et al., 2013). Equivalent persistence of occupation is observed in New Guinea (Roberts et al., 2017). Early to Middle Holocene people in the Philippines, which had remained insular throughout the Pleistocene, responded to similar environmental changes by hunting tropical forest species (Lewis et al., 2008), and continued to practice open sea fishing (Pawlik, 2021). Local forms of agroforestry and cultivation developed across the region, in the case of New Guinea arguably eventually creating an epicenter of indigenous horticultural expansion westward into other islands (Denham, 2018a). Smaller islands, whether shrinking or newly forming as sea-level rose (Williams et al., 2018), were either abandoned in the Early to Middle Holocene (O’Connor, 1992; Shaw et al., 2020), or formed refugia for various biota, including people, who increasingly pursued littoral lifeways (Veth et al., 2017). Higher levels of maritime mobility and exchange over this timespan may have been prompted by the flooding of large tracts of continental shelf (Soares et al., 2008), the greater productivity of warming seas (Pawlik et al., 2014), and the technological innovation of dugout canoes made with shell adzes (Fredericksen et al., 1993; Shipton et al., 2020). Viewed synoptically, overlapping maritime interaction spheres appear to have emerged at this time around Wallacea (Shipton et al., 2020) and in the Bismarck Archipelago (Summerhayes, 2007).

As established, all the large continental islands in ISEA and Near Oceania were already occupied by the Early–Middle Holocene. However, from c. 3500 BC in Taiwan, and c. 2200 BC further south into ISEA proper, ways of life underwent large-scale changes that suggest population mobility and arguably often the arrival of new populations. The most obvious novel characteristics are pottery-making technologies and a degree of reliance on domesticated taxa. Given a parallel step-change in maritime crossing distances across ISEA and Near Oceania, it is widely assumed (though so far not directly substantiated) that this is also associated with new canoe technologies utilizing sails and potentially outriggers, whether developed locally or slightly earlier in the broader East Asian sphere. Populations with these characteristics rapidly colonized smaller islands. Examples include the Batanes Islands in the Philippines, Kayoa and the Banda Islands in Indonesia, along with the Duke of York Islands, and the Arawe, Siassi, Anir, and St Matthias groups in the Bismarcks, as well as New Georgia in the Solomons. These groups also settled the coastal fringes of larger islands; the coastal zones and river valleys of islands such as Luzon, Borneo, and Sumatra experienced colonization in the form of village settlements, often alongside the gradual replacement of HGF communities (Matsumura et al., 2018). These new groups brought with them domesticated animals (Piper et al., 2009) alongside pottery technology in the form of red-slipped earthenware, occasionally with incised, impressed, and dentate-stamped motifs (Carson et al., 2013); they also practiced fishing, arboriculture, and root-crop horticulture (although rice agriculture was probably introduced after these initial dispersals [Alam et al., 2021; Gutaker et al., 2020]). Eastwards, into Near Oceania, initial colonization by the first pottery-producing groups is associated with distinctive dentate-stamped Lapita pottery, first evident in the St Matthias Group (Kirch, 2021), and later moving into the rest of the Bismarck Archipelago between about 1600 and 1300 BC (Rieth & Athens, 2019). Lapita groups, like those in ISEA, targeted islands that were usually small relative to adjacent continental islands, and near key resources (Lepofsky, 1988). Near Oceanic Lapita settlements were densely packed, varied in size between 500 and 80,000 m² (Specht, 2007), and were situated on dunes or lagoon margins, in some cases apparently with stilt houses built over the water (Summerhayes et al., 2019).

As suggested above, this Late Holocene patterning in ISEA and Near Oceanic island occupation involved complex migration flows, also well attested via historical linguistics (Gray & Jordan, 2000) and genetics (McColl et al., 2018)—although their routes and nature, and the finer correlations with linguistic, material culture, and subsistence changes, continue to be debated. The ‘Out-of-Taiwan’ model, arguably the most parsimonious option (purporting to account for correlations in linguistic, genetic, and archaeological evidence) posits mobile, pottery-making communities associated with the Austronesian languages migrating through the region, from Taiwan and (from c. 2200 BC) first to the Philippines and then Indonesia, New Guinea, the Bismarcks, Solomons, and ultimately Remote Oceania (Bellwood, 2017). In contrast, others argue for a broader sphere of inception and interaction around ISEA, wherein correlations in language, technology, and genetics along a west-to-east cline reflect diffusions through extant social networks (Denham, 2018b). The most recent analysis (Cochrane et al., 2021), utilizing a Bayesian framework to assess geographic and chronological patterning in the initial appearance of pottery, suggests that the earliest pottery-bearing deposits in ISEA beyond Taiwan may occur in western Borneo and the northern Philippines. This discrete geography lends support to the suggestion that processes of biological, linguistic, and material-culture change within western ISEA during the Middle and initial Late Holocene may have been more complex, and less linear and neatly correlated, than is often supposed (a conclusion partly supported by recent genetic work; for example, Choin et al., 2021; Larena et al., 2021).

Small reef islands in Near Oceania were colonized substantially after the Lapita horizon, from around two millennia ago. These islands formed more recently through the Late Holocene and represent precarious environments at the margins of horticultural sustainability (McNiven, 2015), although often with reliable marine resources (Bayliss-Smith, 1990). Several tiny volcanic and coral islands, including Takuu, Luangiua (Ontong Java), Nukumanu, Sikaiana, Bellona, and Rennell, were subject to ‘back-colonization’ from Polynesia (see following section), resulting in the so-called ‘Polynesian Outliers’ (Kirch, 1984). Irwin (1992, pp. 183–9) suggests that these islands were at the extreme limits of accessibility and utility for permanent settlement and, to mitigate such stresses, their islanders were highly mobile, maintaining widespread contacts within Near Oceania, as well as north into Micronesia and east into Polynesia.

Increasingly, small islands seem to have been selected in the Late Holocene for several reasons, a principal one being that they contained resources suitable for the specialized landscape management and subsistence practices of the colonizing groups. The continental biogeographic zones of the Sunda and Sahul shelves were far more biodiverse than, for example, the smaller circum-New Guinea islands (Roos et al., 2004), but many smaller islands proved ideal for colonists in terms of their access to marine and littoral resources, as well as space for cultivation and foraging. The gradual decline in terrestrial biodiversity as groups moved through Wallacea, the Solomons and Bismarcks did, however, prompt changes to subsistence strategies, including increased gathering from the reef edge (Walter & Sheppard, 2017) and translocations of marsupials and cassowaries from larger islands to ensure protein resources (for example, Heinsohn, 2010; Summerhayes et al., 2009). Access to geological resources was also essential to successful colonization. Potting communities often settled close to clay and temper deposits or had access to more distant sources during initial colonization (Heath et al., 2017). Moreover, after the Mt Witori (W-K2) eruption c. 1300 BC in New Guinea (Machida et al., 1996), modes of obsidian acquisition and exchange altered, following the appearance of Lapita communities. The Bismarck obsidian sources, particularly Talasea, were widely distributed to Lapita communities in Remote Oceania as far east as Fiji (Best, 1987), and as far west as contemporary communities in Borneo (Chia, 2003). Maintenance of long-range movement of such resources was essential to Late Holocene colonization strategies, and mitigated the stresses of occupying new environments.

The second reason why smaller islands were selected preferentially in the Late Holocene may relate to conflict avoidance or mitigation. As noted above, Late Holocene groups moving into the region often encountered already established long-term populations (Torrence et al., 2009). Thomas’ (2008) concept of ‘friction landscapes’ is helpful here, with much of the friction in this context potentially deriving from interactions with these existing populations. Spriggs (1997, p. 88) identifies a ‘defensive posture’ in Lapita colonization, whereby settlements were positioned at arm’s length from established mainland groups. Subsequent centuries, however, witnessed increasing interaction and admixture (Green, 1991b); on the north and south coasts of New Guinea, Middle to Late Lapita period pottery suggests that Austronesian-speaking groups were active on the mainland (McNiven et al., 2011; Summerhayes, 2019), and pottery finds in the New Guinea Highlands also suggest that potting technology had been adapted to inland clay sources (Gaffney et al., 2015).

Following the initial Late Holocene migrations and subsequent mixing and interaction, long-distance exchange networks linking ISEA with Near and Remote Oceania fragmented into smaller but increasingly specialized trading systems (Allen, 1985; Tanudirio, 2006), certainly by c. AD 1000 and possibly earlier. This saw strategic relocations to monopolize access to resources and the colonization of precariously small islands in critical intermediary locations to facilitate redistribution (Irwin, 1985; Lilley, 1988; Shaw et al., 2016). Along the eastern coasts of New Guinea, a number of tiny uplifted coral islands with limited horticultural capacity were settled within the last millennium and formed the base for mobile canoe voyagers who produced specialty products to exchange for subsistence crops (Gaffney et al., 2020; Gaffney, 2022). Meanwhile, around Halmahera and western New Guinea, maritime groups positioned themselves strategically to exploit bird-of-paradise feathers, raw metals, betelnut, glass beads, pottery, slaves, textiles, bronze axes, and spices (Swadling, 1996), in this case clearly engaging with wider Southeast Asian, East Asian and Indian Ocean trading systems and consumption centers for exotic goods—an extra-insular driving theme to which we will return.

After a pause following their initial appearance in Near Oceania, Lapita settlements appear further east in the Reef and Santa Cruz Islands, Vanuatu, New Caledonia and Fiji, contemporaneously around 1050 BC (Sheppard, 2011; Sheppard et al., 2015), and indicating the first human colonists to cross the Near–Remote Oceania biogeographic boundary (Green, 1991a). Within 150 years of the first Lapita excursions into Remote Oceania, colonists reached Tonga in West Polynesia c. 900 BC and quickly populated the north–south extent of the archipelago (Burley et al., 2015). The only strictly Lapita site in Sāmoa postdates the Tongan landfall, at c. 800 BC (Petchey, 2001), and there are very few additional occupation deposits in Sāmoa for the next few centuries (Clark et al., 2016). The timespan over which Lapita pottery-making populations moved from the Bismarck Archipelago to their farthest eastern extent in Remote Oceania is thus between four and seven centuries. The first Remote Oceanic Lapita sites, with the possible exception of Sāmoa, all appear to be slightly earlier than those of the Northern and Western Solomons or the New Guinea south coast, even though the former are much farther from the geographic origin of Lapita (Sheppard, 2011). The numerous Lapita sites in the westernmost archipelagoes of Remote Oceania suggest successful colonization over a large area, and population continuity to European arrival is indicated by an archaeological record largely without hiatus (cf. Addison & Matisoo-Smith, 2010; Cochrane et al., 2016), although further demographic inputs from the west are likely (Harris et al., 2020). Additionally, the distribution of Lapita decorative motifs indicates continuing movement between Near and Remote Oceania, and between different Remote Oceanic archipelagoes after colonization (Cochrane & Lipo, 2010; Green, 1979), although this inter-archipelagic sailing practice seems to have ended by approximately 700 BC.

The colonization of the Mariana Islands in western Micronesia c. 1200 BC occurred after the appearance of Lapita pottery in the Bismarcks, but most probably slightly preceded the Lapita expansion into Remote Oceania. One of the earliest and best-dated sites in the Marianas is Unai Bapot, but there is debate about the timing of its occupation. Petchey and colleagues (Clark et al., 2010; Petchey et al., 2016) have shown that dates of 1500 BC or older at the site (Carson, 2008) are derived from a single shellfish species, the dating of which is affected by limestone-derived carbon (Petchey et al., 2018; Carson [2020] does not fully address Petchey et al.’s analysis, and accordingly his conclusions may be erroneous; see now Petchey & Clark, 2021). Although an earlier date cannot be categorically excluded (Petchey & Clark, 2021), all other species of shellfish dated, and all charcoal dates, point to about 1200 BC as the colonization horizon (see also Rieth & Athens, 2019). Multiple sites across the Marianas whose date ranges encompass 1200 BC (Carson, 2014, table 4.1) suggest a successful colonization of some geographic breadth, although many of these sites also suffer from dating issues (Rieth & Athens, 2019, p. 8).

The Palauan archipelago, southwest of the Marianas, was definitely reached by approximately 1000 BC, if not slightly earlier, with evidence for occupation in a variety of environments across the archipelago at this time (Clark et al., 2006; Fitzpatrick, 2018). Admittedly, paleoenvironmental research has revealed a signature interpreted as human landscape modification a substantial 1500 years earlier (Athens & Ward, 2001), but there is a complete absence of similarly aged artifact-bearing deposits. Dickinson and Athens (2007) argue that the discrepancy is caused by subsidence and inundation of early coastal sites, but the existence of unambiguous artifactual signatures around 1200–1000 BC for both the large west Micronesian archipelagoes suggests that these dates are more robust indicators of earliest colonization. In both cases, the colonists almost certainly arrived from ISEA. Linguistically, both modern Palauan and Marianas Chamorro are Malayo-Polynesian languages related to others in ISEA, although not themselves closely related (Pawley, 2018). The earliest pottery in the Marianas, though not in Palau, has clear stylistic affinities with red-slipped and impressed-decorated pottery from ISEA (Carson et al., 2013; Cochrane et al., 2021; Swete-Kelly & Winter, 2022). In short, it is clear that western Micronesia was colonized from the west by groups that were in general terms culturally related.

Linguistic and ceramic affinities collectively suggest that central and eastern Micronesia were colonized from the south, by post-Lapita populations deriving from the former Lapita region, including islands near the Solomons, and not from western Micronesia (Kirch, 1987; Pawley, 2018). Pohnpei and Kosrae, two environmentally rich volcanic high islands, as well as the more environmentally limited eastern atolls such as Kiribati (which had likely only stabilized as islands several centuries before their colonization: Weisler et al., 2012), were colonized between approximately 50 BC and AD 150 (Athens, 2018). Kapingamarangi and Nukuoro atolls were colonized later than the rest of Micronesia and probably in the context of increased two-way voyaging contacts with West Polynesia, beginning about AD 1000 (this is reflected in the atolls’ Polynesian languages). Radiocarbon dates from Nukuoro are imperfectly reported; colonization of Kapingamarangi is perhaps better dated at AD 1200–1400, but these dates rely on unidentified wood charcoal and bulk sediment samples (Rieth & Cochrane, 2018).

The increase in voyaging around AD 1000 that led to late colonization of Kapingamarangi, Nukuoro and other ‘Polynesian Outliers’ (Kirch, 1984) closer to New Guinea (discussed in the previous section) also resulted at a more significant scale in the final population expansion eastward, northward and southward into previously unoccupied Pacific islands. The colonization of the Polynesian Triangle, from Hawaiʻi in the north, to Rapa Nui in the east, and Aotearoa New Zealand in the south, began about AD 1000, with all the major island groups settled by c. AD 1250 (Allen & Kahn, 2010; Kirch, 2010; Mulrooney et al., 2011; Walter et al., 2017; Wilmshurst et al., 2011). Although there is still debate about the timing of colonization on particular islands, the differences often amount to no more than a hundred years or so, and overall a clear pattern of rapid colonization across a vast region has emerged (although the markedly early colonization of eastern Micronesia also involved prodigious inter-island distances).

In the center of East Polynesia, the Society Islands have previously been beset with dating problems, but dating classification (Wilmshurst et al., 2011), paleoenvironmental (Stevenson et al., 2017), and excavation analyses (Kahn, 2012) have now generally converged on a colonization date of approximately AD 1000, making these islands the earliest colonized in East Polynesia. The Cook Islands, a dispersed set of atolls and small high islands, are closest to the West Polynesian homeland, yet their colonization may postdate that of the Societies: Allen and Morrison (2013), using a Bayesian approach, estimate Cook Island colonization at AD 1050–1270 (95% probability). To the southeast, islands in the Australs have varied colonization dates. The earliest is probably Rapa, estimated to be between AD 800 and AD 1300 (95% probability) (Anderson & Kennett, 2012), and more specifically about AD 1100 as indicated by paleoenvironmental evidence (Prebble et al., 2013). To the northeast, Marquesan colonization occurred shortly after the Societies, likely sometime between AD 1000 and AD 1200 (Allen & McAlister, 2013), while to the far southeast, Mangareva, the Pitcairn group, and Rapa Nui were settled around AD 1200 (Hunt & Lipo, 2006; Kirch et al., 2010; Weisler et al., 2012; Wilmshurst et al., 2011). Recent finds of chicken bones from late pre-Columbian South America indicate occasional onward voyaging even further to the east over the next few centuries (Storey et al., 2007).

The northern and southern extremes of Polynesia have clear colonization chronologies, due to the long history of research in these islands and concerted efforts at accurate and precise dating. In Aotearoa New Zealand, archaeological and paleoenvironmental evidence, including dates from the commensal Pacific rat (Rattus exulans) and rat-gnawed seeds (Wilmshurst & Higham, 2004), converge on a colonization estimate in the mid–late thirteenth century, which is also within a recent Bayesian estimate of AD 1270–1309 (95% probability) (Dye, 2015; although see Petchey & Schmid, 2020). In Hawaiʻi, a similar Bayesian approach combining archaeological and paleoenvironmental data estimates colonization at AD 940–1130 (95% probability), but most likely between AD 1000 and AD 1100 (Athens et al., 2014).

The Faroe Islands lie roughly midway between Norway and Iceland (respectively c. 600 and 440 km distant), and constitute the first stepping stone for a diaspora across the North Atlantic. The eighth-century Irish monk Dicuil suggested that anchorites had lived on certain North Atlantic islands before the Norse arrived (Tierney, 1967), and although the identification of the Faroes in his writing has been questioned (Arge, 1991), the earliest palaeoenvironmental dates do suggest small-scale human landscape modification between AD 400 and 600 (Jóhansen, 1979; Hannon et al., 2005; Edwards et al., 2005). Moreover, a pre-Norse presence has been directly confirmed by the earliest dates of non-native barley grains from Á Sondum on Sandoy (Church et al., 2013). Initial Norse settlement of Sandoy and Eysturoy is evident in the ninth century (Church et al., 2005; Stumman Hansen, 2013). Permanent early (AD 800–1000) sites are located on the coast, while seasonal later (AD 1000–1100) sites are also found inland (Malmros, 1990). The large islands were rapidly settled; the scarcity of data from smaller islands means that it is uncertain whether they were seasonally occupied, or never settled at all.

Iceland, too, presents the vexed issue of anchorite activity prior to the Norse arrival (Tierney, 1967). A few putative sixth to eighth century AD dates come from the small island of Vestman, off the south Icelandic coast (Hermanns-Auðardóttir, 1989), and from the southwest around Reykjavík (Nordahl, 1988). However, these derive from wood charcoal with inbuilt age and none lie stratigraphically below the Landnám Tephra Layer (LTL) of AD 877 ± 1 (Schmid et al., 2017). The earliest short-lived materials, by contrast, are consistent with tephrochronological evidence and support an initial colonization horizon in the mid–late ninth century (Ascough et al., 2007; Schmid et al., 2021). This mid–late ninth century colonization is also attested in literary sources. Íslendingabók, composed in the twelfth century (Grønlie, 2006), pinpoints the timing and speed of earliest Norse settlement, and indicates that all the medieval administrative divisions of Iceland were settled within 60 years (AD 870–930). This timeframe is further confirmed by ice-core data and historically dated tephra layers (Schmid et al., 2017). A recent comprehensive reassessment of the Icelandic evidence indicates that 550 archaeological sites can be assigned to three consecutive periods of settlement (Schmid, et al. 2021); only two seasonal sites in the southwest date to the pre-Landnám period and can be interpreted as residues of exploration. The earliest palaeoecological data from the same area suggest that non-native barley was cultivated before the deposition of the LTL, approximately between AD 830 and AD 880 (Schmid et al., 2018a). However, occupation evidence becomes far more extensive (Schmid et al., 2021) during the Landnám and post-Landnám periods (AD 877–938/939; AD 938/939 to 1104). Pre-Landnám sites are situated coastally, while early Landnám sites are located both in coastal and inland areas, including less optimal ones, and spread from the southwest to the north and then the east.

Overall, Iceland’s settlement was extremely rapid, homogeneous, and continuous, with all inhabitable parts occupied before AD 890 (Schmid et al., 2018b, 2021; Vésteinsson & McGovern, 2012). This speed relates in part to the scale of the process, perhaps involving as many as 24,000 initial settlers in less than 20 years (Vésteinsson & McGovern, 2012), although there was also later immigration (Vésteinsson & Gestsdóttir, 2014). Moreover, this explosive colonization was successful in the long-term; most settlements were not abandoned and indeed persist to this day. Smaller satellite islands were also settled; Vestman (regardless of the uncertain earlier dates) in the late ninth to tenth century AD (Hermanns-Auðardóttir, 1989), and Papey in the late tenth to eleventh century (Eldjárn & Sveinbjarnardóttir, 1989). In northwestern Iceland a strong correlation between a farmstead’s size and its establishment date can be discerned, suggesting that the largest and wealthiest farms were founded by the first settlers (Steinberg et al., 2016).

Grænlendingasaga and Erik’s Saga Rauða (written in c. AD 1190 and 1260 respectively) describe the voyages of the Norse to Greenland (whose nearest landfall lies c. 300 km distant from northwest Iceland) around AD 985 (Gilbert et al., 2008; Rasch & Jensen, 1997), during the Medieval Climate Optimum (AD 950–1250 [Jackson et al., 2019]). Unlike Iceland, Greenland was already occupied, thanks to prior expansion by circum-polar HGF groups across land, ice and sea. The Norse settlement unfolded east to west; the current evidence indicates that the larger Eastern and Middle Settlements (between 60 and 61 degrees north, and with 560 recorded farms), began before the smaller Western Settlement (around 64 degrees north, with some 75 farms) (Arneborg et al., 2012).

The earliest sheep, goat, and cattle (unequivocal evidence for Norse translocations) in secure contexts are from large Landnám settlements in the Eastern Settlement and, in radiocarbon terms, fall within a late eighth to tenth century AD range compatible with the textual tradition (Edwards et al., 2008). Other early permanent settlements are dated to the late tenth century (for example, Arneborg et al., 1999, 2012). Clarifying the initial colonization of the Western Settlement is more challenging. The earliest reindeer, harp seal, and walrus bone samples from Niaquusat in the Western Settlement also fall between the late eighth and tenth centuries. However, their contexts are most likely connected to the indigenous Dorset occupants, and might suggest Norse–Dorset interactions as readily as a small number of early Norse arrivals (Arneborg et al., 2012). Aside from these culturally ambiguous dates, permanent Norse settlements in the west date to the early eleventh century AD (Arneborg et al., 1999, 2012).

Greenland Norse sites were located in fjords or near rivers and lakes. As in Iceland, there is a strong correlation between site size and establishment date; initial settlers founded larger and ultimately wealthier farmsteads than later settlers (Arneborg et al., 2012). Estimated total peak populations for the Greenland Norse settlements range from 2250 to 6800 people (Lynerup, 1996), yet all were abandoned before the mid fifteenth century AD, potentially due to climatic deterioration caused by the Little Ice Age, which may have triggered resource stress and competition between the Norse and indigenous groups (Dugmore et al., 2010). It has also been suggested that storm frequency intensified around AD 1425–1450, disrupting vital long-range interaction networks.

Newfoundland, a large island in the Canadian maritime archipelago, lies c. 1100 km south of Greenland. Grænlendingasaga and Erik’s Saga Rauða describe several Norse expeditions around AD 1000 to Vínland, widely understood to be Newfoundland, where they encountered indigenous people. So far only one Norse site (L’Anse aux Meadows) has been extensively excavated. The settlement persisted for only a short time before it collapsed, potentially due to competition between Norse and local American-Dorset groups (McGhee, 1984). The radiocarbon dates from peat, twigs and charcoal with inbuilt age do not provide narrow ranges, but indicate a short occupation (Wallace, 2003), according to recent Bayesian analysis beginning c. AD 910–1030 and ending AD 1030–1145 (Ledger et al., 2019).

Tectonic activity to the north of ISEA and Oceania has generated numerous islands, notably but not exclusively in the form of island arcs and chains (Fig. 4). The Japanese archipelago and surrounding islands, including the Ryukyus (between Kyushu and Taiwan) and Kurils (between Hokkaido and the Kamchatka Peninsula), have a particularly long and complex settlement sequence. Several saw Pleistocene occupation, in some instances continuing into later times. The four main Japanese islands, today totaling some 380,000 km² in area, all boast a robust Upper Palaeolithic. During the Last Glacial Maximum, the deep strait between Korea and Kyushu endured, but those between mainland Asia and Sakhalin, and between Sakhalin and Hokkaido, were closed by eustatic drawdown, while that between Hokkaido and Honshu was either drastically narrowed or turned into a tenuous land bridge (Oba & Irino, 2012). There is, accordingly, little or no need to invoke maritime colonization for the initial Japanese Palaeolithic, nor for the succeeding Jōmon Palaeolithic and its Holocene HGF successors, assuming that local cultural processes rather than new influxes are responsible for the latter (much the same applies to Sakhalin’s enduring HGF population). Palaeolithic and later Jōmon maritime activity is, however, implicated in the offshore activity on fringing islands (Ikeya, 2015; Kaifu et al., 2020), as well as the colonization of smaller true islands to the north and south. In the Ryukyus, following a maritime Upper Palaeolithic occupation but subsequent abandonment, a renewed Holocene HGF colonization of much of the chain (excepting the southernmost islands) began around 7000 BC (Takamiya, 2006). The most plentiful evidence comes from large (but distant) Okinawa; and interestingly, despite early evidence from intervening islands, by around 2000 BC settlement appears to have contracted southwards, with smaller islands such as Yaku abandoned (Takamiya et al., 2015). North of Hokkaido, the southern Kurils likewise appear to have been settled around 6000–5500 BC (Fitzhugh et al., 2016). The earliest evidence derives from Iturup, with later settlement from neighboring Kunashir and Shikotan (Kuzmin et al., 2012), all notably among the largest of the Kurils, as well as the most southerly of the islands and closest to Hokkaido.

The appearance of agropastoral lifeways in the Japanese archipelago during the first millennium BC (the Yayoi phenomenon) is usually attributed at least in part to a population influx from mainland East Asia, bringing rice agriculture (for example, Crawford, 2011; Hudson, 2013; Hudson et al., 2020). The initial Yayoi is now redated to c. 1000–900 BC (de Boer et al., 2020; Shoda, 2010), and initially farming was limited to Kyushu, Shikoku, and southern Honshu. This distribution probably reflects the axis of expansion and latitudinal constraints on rice cultivation rather than a preference for larger islands per se; for example, northerly Hokkaido, an enduring HGF bastion, is twice as large as Kyushu. In contrast to much of ISEA and Near Oceania, where horticulturalists entering regions with already established HGF occupants at first tended to avoid or skirt the latter populations (focusing on small islands or moving rapidly through), in much of the core of the Japanese archipelago HGF lifeways rapidly disappeared, save in northern Honshu, Hokkaido and off-shore islands, plus the more distant Kurils and Ryukyus (de Boer et al., 2020; Hudson et al., 2020).

Island colonization of the circum-Pacific east of the Bering Sea is ultimately a function of the arrival of people in the Americas, yet substantially postdates the primary phase of this process (as it does to a still more pronounced degree on the Caribbean flank). The eastern Aleutians (the Fox Islands) appear to have been reached by HGFs, presumably from Alaska, around c. 7000 BC (Knecht & Davis, 2001). Moving westward and out to sea, island size decreases from the Fox to the Andreanof, Rat, and Near groups, and both this and increasing distance correspond tightly with later colonization. The first settlement on Unalaska dates to c. 7000 BC, on the Andreanofs to 4000–3000 BC, on the Rat Islands to 1500 BC, and on the Near Islands to 500 BC (Veltre & Smith, 2010)—a globally rather rare example of human colonization patterns mapping onto simple biogeographic expectations without appreciable ‘noise’. From the Aleutians southwards, there are no substantial oceanic islands off the coast of the Americas, excepting the small eastern Pacific remotes such as the Revillagigedo Islands, the Alijos Rocks, the Galapagos, and the Juan Fernandez group, none of which were colonized prior to European arrival. The majority of the inshore islands situated on the continental shelf, from the Pacific Northwest to the Chilean archipelago, are so close to the American mainland that crossings are mostly trivial, though arguable exceptions include the HGF intakes of the Kodiak, Haida Gwaii, California Channel and Panamanian Pacific islands. In the Kodiaks, earliest settlement is now dated to c. 5500 BC at the Rice Ridge Site on Kodiak itself, the largest of the group (Kopperl, 2012). Further south, the Haida Gwaii archipelago appears to have been occupied (or seasonably exploited) from the early Holocene, c. 9000–8000 BC (Mackie et al., 2018). Initial activity on the California Channel Islands is even earlier, lying at the Pleistocene/Holocene boundary (Erlandson et al., 2011), when the archipelago was probably united into the palaeoisland Santarosae. The islands off Pacific Panama were first settled by HGFs slightly before 4000 BC, followed after a long hiatus by pottery-using groups from the last centuries BC (Martín et al. 2017). The similarity between the earliest dates here and in the Greater Antilles raises the intriguing possibility of a common stimulus in the changing lifeways and demographics of the early Archaic in continental Central America.

Among the Atlantic’s Macaronesian islands (Fig. 5), the remote Cape Verdes, Azores and Madeira were probably unknown prior to Portuguese expansion, although there are hints on Madeira (in the form of Mus musculus as a proxy for human arrival) that Norse voyages may have reached this far into the ocean at southerly latitudes (Mitchell, 2022, p. 63), perhaps as a consequence of voyages around the Iberian peninsula and into the Mediterranean. The Canaries, by contrast, lie only c. 90 km off the West African coast and were certainly colonized by agropastoralists prior to this, though precisely when has long been disputed. A recent synthetic analysis convincingly narrows the likely timeframe to between the later first millennium BC and the early first millennium AD (Nascimento et al., 2020), and Mitchell demonstrates that reliable radiocarbon dates become more plentiful in the latter half of this window (2022, pp. 76–82). The admittedly equivocal genetic evidence may suggest that this colonization derived from source populations in the Maghreb or perhaps further east (Fregel et al., 2019). That this timespan also witnessed a step change in economic integration and urbanization in the western Mediterranean and ‘Mediterranean Atlantic’ beyond the Gibraltar Strait, associated with Punic and later Roman maritime activity (Broodbank, 2013, pp. 561–575), is unlikely to be coincidental. There is an intriguing parallel for this co-development of maritime integrative processes and island colonization in the potentially contemporary settlement of Socotra, on the far side of the continent.

Further south, in the Gulf of Guinea and Bight of Biafra, stratified sequences and robust dates are largely absent. Indeed, São Tomé and Príncipe has recently been identified as the only modern nation that remains essentially devoid, to date, of archaeological investigation (Mitchell & Lunn-Rockcliffe, 2022). That said, Bioko (at 2000 km² the largest island in the Gulf) and smaller Corisco (only c. 20 km from the mainland) provide evidence for settlement by farming groups well within the timeframe of this study. On Corisco, a review of radiometric dates indicates likely settlement around 100 BC (Sánchez-Elipe Lorente et al., 2016). Bioko presents a more complicated situation: the Bantu-speaking Bubi (genetically similar to other Bantu-speaking groups: Gelabert et al., 2019) appear to have colonized the island at around the same time as Corisco (Clist, 1998), but Bubi mythohistoric accounts recall an extant indigenous people, the Balettérimo, who may conceivably be implicated in an otherwise unusual and archaeologically identified chipped stone tradition. Príncipe, São Tomé, and Annobón, lying increasingly far out into the Gulf (Annobón lies a considerable 350 km from the mainland), all appear to have been free of human settlement at the point of European arrival (Mitchell, 2022; Mitchell & Lunn-Rockcliffe, 2022).

The foregoing has outlined at an empirical level the spatial and temporal patterning of Holocene island colonization, theater by theater. Figure 7 presents the resultant consolidated global pattern, distinguishing between broadly HGF and farming populations, and providing the baseline for much of the interpretative discussion that follows. It is apparent that certain aspects of the pattern observed by Keegan and Diamond (1987) remain robust, while others need revision. On the former front their focus on three great nurseries of maritime mobility as engines of global island colonization remains justified. The Mediterranean, Caribbean, and ISEA with Near Oceania—the first and third with particularly densely interwoven continental and insular geographies—each witness exceptionally early island colonization, led off in all instances by HGF occupation that in the Eurasian examples stretches back unevenly into the Pleistocene, followed by variably later farming colonizations. Truly oceanic spaces were breached last, often only with distance-busting sail-based technologies, although the vectors differ; trans-(North) Atlantic, circum-Indian Ocean, and directly out into Remote Oceania, the last unmediated by continental contact until end-of-the-line landfalls in South America. Remote Oceania remains the theater in which island colonists demonstrably traveled furthest, although some Indian Ocean distances, if they could ever be shown to be unbroken, might rival these. Conversely Africa, with a long yet largely unindented coastline furnished with few offshore islands, still appears to come relatively late to the insular game, as Keegan and Diamond observed.

On the other hand, subsequent data have undoubtedly altered and complicated the picture since 1987, by bringing entirely new sequences to light, and challenging the previous pattern for certain long-known theaters. Major examples of the former include several Indian Ocean and other circum-African island groups, as well as the surprisingly early dates for a Holocene HGF presence on islands around the northern Pacific rim from Okinawa to the California Channel Islands. The latter is especially pertinent in the Caribbean, where the chronological data are slowly resolving into a seemingly irreducibly complex set of dynamics. It also obtains in Near and Remote Oceania, where Lapita has been temporally clarified and spatially expanded, including south towards the Torres Strait, while broadly contemporary starts in parts of Micronesia are now well documented and a clear, late signal has been affirmed for much of eastern Remote Oceania. Likewise, the North Atlantic theater has benefited from a major radiocarbon-driven tightening of chronological focus. Even in the Mediterranean, overall the most stable among the previously well-investigated theaters (cf. Broodbank, 2006), ultra-early Holocene dates on Cyprus and the case for an anomalously late arrival at the opposite end of the basin, in the Balearics, prompt reevaluation.

One ongoing challenge to establishing colonization dates that applies across several theaters also deserves note. This involves circumstances where the earliest dated cultural deposits and evidence for environmental perturbations or other ambiguously anthropogenic signatures do not chronologically align. When the temporal gap is small, this can satisfactorily be explained by taphonomic or recovery effects obscuring the earliest cultural evidence; this, for example, is probably the case in Malta, with a divergence of perhaps three or four centuries between large-scale environmental change and dated cultural deposits (Malone et al., 2019; Marriner et al., 2019). Where there is a more pronounced divergence, but more solid underpinning of the lower-dated, cultural horizon, we might well conclude that the antecedent environmental signature was non-anthropogenic, as for example on Palau (Athens & Ward, 2001; Clark et al., 2006). In at least two cases, however, dissonance between the archaeological and palaeoenvironmental data creates an interpretative impasse. First, in the Lesser Antilles dates from archaeological deposits support a ‘Southward Route Hypothesis’ (Fitzpatrick, 2013, 2015; Napolitano et al., 2019), but dated environmental disruption supports a northward ‘Stepping Stone Model’ (Siegel et al., 2018). Secondly, on Madagascar, where the earliest unambiguous archaeological deposits date to the first millennium AD (Anderson et al., 2018; Mitchell, 2020a, 2020b), purported butchery marks on avifauna are dated several millennia earlier (Hansford et al., 2018). These questions cannot currently be clarified or resolved without improved data, although broader informed contextualization of the kinds attempted here may be helpful in provisionally selecting the most plausible scenarios.

How much of the structure and variability now established do models grounded in biogeography capture? And how much instead requires recourse to other—cultural and social—modes of explanation? We begin with an exploratory quantitative analysis to assess the continued importance of biogeographic factors, employing a statistical evaluation of the impact of several quantifiable physiographic, geometrical, or configurational (sensu Keegan & Diamond, 1987) variables—plus one readily identifiable cultural variable—on colonization dates for a substantial sample of islands for which robust chronological data are available, and which span our main theaters. We follow this by considering the role of certain other broadly environmental factors that are harder to evaluate in strictly quantitative terms, yet which potentially served to shape global as well as distinctive regional patterns.

The quantitative analysis uses a sample of 49 islands selected for robust and chronologically refined data for first Holocene settlement (i.e., in cases like Okinawa, we ignored preceding Pleistocene occupation) and that span the theaters under discussion. Six variables were used to describe each island: size; latitude; elevation; coastal complexity index; isolation index; and the presence or absence of sailing technology as a readily measurable cultural variable (see Supplementary Information for data and R code). Coastal complexity index is generated by length of shoreline divided by area (essentially a measure of smoothness versus roughness). This variable is relevant in that degree of coastal indentedness has been convincingly associated with autocatalytic sequences (sensu Keegan & Diamond, 1987) in some contexts (for example, Broodbank, 2006), with a ‘flat’ configuration discouraging maritime experimentation—Atlantic Africa is a good example. Obviously, this attribute applies primarily to the source area, rather than target, of colonization; but as indented littorals tend to co-occur within associated coastal and insular regions, this variable provides a useful and justifiable cipher. The isolation index is defined as the sum of the square roots of the distances to the nearest continent, the nearest island group or archipelago, and the nearest equivalent or larger island (Dahl, 1991). This measure is preferable to raw distance from the nearest continent (for example, Itescu et al., 2019). For the present purpose, dates BC/AD were converted into time before present (BP), the dependent variable in our analyses.

A graphic exploration (Fig. 8) of independent variables and colonization time BP suggests that, of our six variables, the isolation index and known or presumed presence of the sail are the best predictors of colonization time. That is, the presence of sails (in Remote Oceania and the North Atlantic) is associated with very late colonization episodes. This is perhaps not surprising, but nevertheless useful as a formal affirmation that, beyond certain spatial thresholds, particular technological conditions were fundamentally necessary for oceanic colonization. Given the characteristics of the independent variables (which are not normally distributed), non-parametric tests of correlation were generated with the Mann–Whitney U test for the presence of sail and Kendall’s tau for all other variables. These tests indicate there are significant correlations between the sail’s presence/absence and BP (p < 0.00), isolation index and BP (p < 0.00), and a possibly significant correlation between size and BP (p = 0.10). All other p-values are between 0.15 and 0.39.

Generalized linear models (GLMs) allow the examination of the effects of more than one continuous dependent variable on BP. This can determine, for example, if a model that includes both isolation index and other variables better predicts BP than isolation index alone. GLMs are like multiple linear regression (but do not have the strictures of a normally distributed response variable—here, BP—among others: McElreath, 2020). They compare the regression of the response variable on various combinations of independent variables, with each combination considered a separate model. The efficacy of different models can then be determined using Akaike’s information criterion (AIC) and a Bayesian information criterion (BIC), which are used to rank models in terms of information loss (Symonds & Moussalli, 2011). The AIC for each model is calculated from the number of fitted parameters (independents), the residual sum of squares of the model or maximum likelihood estimate, the intercept as an additional parameter, and the variance estimate as a parameter. In general, lower AIC values are better considering a parsimony criterion where less complex models (e.g., with fewer parameters) are preferred over more complex ones. Twenty-six models were constructed of all possible combinations of independents (i.e., an all sub-set approach). The top three models by AIC are presented in Table 1.

Here, delta is the difference between a model’s AIC and the top model’s AIC. AIC is not just a function of the number of parameters (although it might look like it in this case). There are other models within the 26 evaluated that have lower AICs than either the second or third ranked models, but with greater degrees of freedom. The calculated weight (Akaike weight) allows us to compare how effectively the best-fitting model predicts the data, relative to other models (Symonds & Mousalli, 2011). Weight is a value between zero and one with the weights of all models evaluated summing to one, and is analogous to the probability that a model is the best approximating model. Thus—for example—the Isolation Index model only has about a 13% probability of being the best approximating model.

We can also use BIC to rank our models. BIC is very similar to AIC but is calculated in a slightly different way (Johnson & Omland, 2004), from several terms: (1) a formula using the value of the maximum likelihood function computed from the parameter values that create the highest likelihood of realizing the response variable value; (2) sample size of the response variable; and (3) the number of parameters. The top three models by BIC are shown in Table 2.

Again, the Isolation Index model seems to be the best. In a Bayesian framework weight is not akin to a probability such as for interpretations of AIC, but is an estimation of Bayes Factors, part of a method for determining support amongst competing Bayesian models (McElreath, 2020; Raftery, 1995). Figure 9 shows the Isolation Index model fitted to the data with a 95% confidence envelope, and the actual values. The data are very noisy, hence the low AIC and BIC weights.

Examining the residuals (Fig. 10) reveals two outliers with z-scores greater than 1 standard deviation, Okinawa (case 20), and Cyprus (case 41), both—interestingly—colonized earlier than predicted by the model. Intriguingly, both these large islands witness preceding HGF settlement and therefore present opportunities for knowledge feedback to later colonists.

In summary, the quantitative analyses suggest that the presence of sailing technology predicts late colonization time, for reasons associated with theater-specific adoption of technological innovations that facilitated open-ocean travel; a useful observation at regional but not global scales, and a reminder of the importance of building local specifics into any explanation of patterning. In a generalized linear model framework, isolation alone is the best predictor of colonization time. However, the associated scores are not high, and the data are very noisy. As a result, we suspect that the variables utilized in these models (although clearly important) are not the only ones that account for structure in global island colonization. So the explanatory net will, indeed, need to be widened beyond the purely physiographic, geometrical, or configurational properties we examine here.

The preceding analysis clearly demonstrates an explanatory role for degree of isolation and, to a lesser extent, island area—two variables, along with configuration, that are central to island biogeography. Our broader exploration further bears out the importance of these factors cross-culturally. This is especially the case in the Caribbean and Mediterranean (see now Plekhov et al., 2021), where a handful of very large islands (the Greater Antilles and the Mediterranean ‘big five’), or smaller islands with very low isolation indices (Trinidad; the Dalmatian islands), witness much earlier colonization than smaller, remoter islands. Distance, area, and configurational effects also seem relevant in the Gulf of Guinea, and the Ryukyus, Kurils and Aleutians, but less so in the open-ocean contexts of Remote Oceania, the Indian Ocean and the North Atlantic, as discussed further below.

These factors combine to make islands attractive in multiple ways. Accordingly, harder questions need to be asked as to exactly why, for example, big, reasonably easily reached islands were attractive. Was it because they were simply easier to discover; because they could support robust populations; or because they tended to have more varied environments and greater biodiversity? Again, in closely packed theaters (where most outsized islands are found, unsurprisingly given their continental geological origins), sheer distance does not seem to be an insurmountable barrier, with crossings of seldom much over 50 km (a day or two’s paddle) in the Mediterranean, to take one case. In this kind of context it does not seem to be their intrinsic discoverability that renders such larger islands attractive, as argued by the first applications of island biogeography to human colonization (Cherry, 1981, 1984; Keegan & Diamond, 1987). Rather, the key attractor was area and (in less abstract terms) area’s typical implications for demographic capacity, environmental diversity and sustainability (including of edible or otherwise exploitable animal and plant resources), geological and pedological variety, and maximum elevation (ensuring more varied ecozones, as well as, especially for farmers in semi-arid theaters, orographic rainfall). A big island, once colonized, also brought ‘commuter effect’ benefits to its neighbors (Keegan & Diamond, 1987, p. 59), in terms of larger overall populations and social networks on which to draw in adversity, or from which to recolonize after an abandonment.

The significance of area may offer insights into divergences between HGF and farming dynamics on islands. Food-acquiring strategies should be more sensitive to area than food-producing strategies, in that they are dependent on non-anthropogenic trophic systems (though this is not a binary division, as these strategies can co-mingle, especially in the tropics). As a result, we should expect insular area to impinge more severely on the colonization dynamics of HGFs. Food-producing communities should be released from these constraints, and yet, critically, still demonstrate preference for larger environments (Cherry & Leppard, 2018a) (we ignore here, to limit an otherwise lengthy discussion, the complicating factor of latitudinal effects on carrying capacity: Freeman et al., 2020). This general observation may be central to building a global account of the relative paucity of HGF versus farming settlement of islands.

Beyond biogeography, but still relating to the environmental and spatial contexts in which islands existed, two further points are worth emphasis. First, although intriguing regionally specific cases have been made that certain climatic conditions might have played a significant role in easing maritime travel to islands (notably a putatively less violent North Atlantic during the Medieval Climate Optimum, and the effect of El Niño events on facilitating eastward voyaging in Remote Oceania: Anderson et al., 2006), the sheer variety seen in the chronology of colonization sequences denies any global-level patterning beyond the overall identification of the Holocene as an island-colonizing window—certainly, Fig. 7 shows no inter-regional correlation of positive or negative activity with, for example, the well-known 8.2 kya climate event.

Second, and more positively, although our exploratory quantitative analysis suggests that absolute latitude has little impact on the timing of colonization at a global scale, several unrelated colonization episodes do display a preference for extension within a given latitudinal envelope. In the Mediterranean, earliest Caribbean, Lapita, initial Remote Oceanic, and Norse cases colonizers moved readily with the latitudinal grain, but less frequently across it, at least beyond broad bands of similarity. Why should this be the case? Coincidence is far from impossible: the Mediterranean has a pronounced east–west orientation, as do the first-colonized Greater Antilles. North Atlantic expansion operated between open ocean to the south and polar environments to the north; and the orientation of equatorial winds in the Pacific, prior to innovations in sail technology, may have militated in favor of west–east movement. Yet latitudinal gradients may equally have been culturally favored, whether for cosmological reasons (the comfort zone within the rising and setting of a key celestial navigation aid?), or, at least for island farmers, due to the limited tolerance of their domesticates for different growing conditions, and the time required to breed adapted strains. As Vander Linden and colleagues (2022) have pointed out, genetic lag in mutation rates (and consequent variation in hardiness or tolerance) of various European Neolithic staples may be relevant to explaining discontinuous range expansion, with the development of landraces increasingly tolerant of latitudinal difference enabling subsequent north–south expansion of farming. Considering the generally late dates for expansion across latitudinal bands in Near and Remote Oceania, such a model could perhaps be most effectively explored in this theater, as the genetic analysis of early Oceanic crops develops. Further possible explanations for the pulse-like nature of certain island colonization sequences are offered below.

Clearly, island biogeographical and broader environmental patterning can explain part but far from all of the behavioral patterning observed. One key point to note is that although successful in addressing the relative timing within regions (i.e., in which approximate order islands were colonized), purely environmental models are less able to shed much light on absolute timing at larger scales—especially given the limited degree to which inter-regional climate fluctuations and island colonization seem to be correlated. As we shall see, absolute timings, regionally and globally, owe more to wider social and cultural factors.

But first, what of autocatalysis, Keegan and Diamond’s explanatory model drawing upon both island configuration and human cognitive responses to shed light on why certain island colonizing sequences take off so spectacularly? This model seems exceptionally insightful, with clear relevance not only to the extraordinary take-off of island expansion from ISEA through to Remote Oceania, but also to the Norse Landnám across the North Atlantic, where earlier discovery of islands in relatively benign sailing nurseries (the Baltic and North Seas), allied with enormous dispersal range, led to the type of long-distance open-ocean exploration described in Erik’s Saga Rauða. As Cochrane (2018) argues, expansion through Near Oceania led to iterative colonizing behaviors; aided by innovations in canoe technology, this behavior brought into range Remote Oceania, with its sparser, more far-flung insular geography. In the sense that this exploratory behavior had been ‘selected for’, it was maintained, honed and enhanced, driving expansion far into the Pacific. Once the actual nature of the island and coastwise expansion that brought an Austronesian language to Madagascar is better understood, autocatalyzing processes may also find purchase in Indian Ocean explanations. At a much more modest level, local autocatalyzing sequences can be identified in many other island groups, including those of the Mediterranean and Caribbean.

It will be self-evident that the most spatially ambitious dispersals just discussed also involved the use of advanced seafaring technology, and specifically the greater speed, cargo-carrying capacity and (in terms of propulsive energy) calorie-light locomotion of sail-driven seacraft. Indeed the initial adoption and progressive refinement of such technology to meet ever-greater challenges in the world of islands can itself be understood as one element of autocatalysis, rendering distance more plastic in terms of human experience and capabilities (Broodbank, 2000, pp. 105–106; 2010). The impact of technology-dependent perceptions of maritime distance was such that, for example, in the absence of the sail little Annobón may have been as, or more, remote from the African mainland as Sāmoa from Tonga. It is unfortunate that our knowledge of the seacraft in which people reached islands is extremely uneven, and for many theaters, especially during the earlier Holocene, based on little more than inference. The minimal assumption is an environmentally suitable subset from a potential range of paddle-driven canoes, reed craft, skin boats or rafts, adequate for short-range and short-term crossings to and between islands. But as we have seen, later colonization dates are strongly associated with the existence of sailing technology, a function of the relatively recent global emergence and development of the sail and associated innovations in select, probably independent regions in western and eastern Eurasia from the third and second millennium BC onward, including modes of navigation suitable for oceanic space and the capacity to sail to windward (Anderson, 2010). This interrelated group of innovations was central to the Late Holocene island colonizing phases in ISEA and Near Oceania, plus the entirety of the Remote Oceanic, North Atlantic and (assuredly, if less directly evidenced) Indian Ocean expansions, though it proliferated too late to have an impact on initial island colonization in the Mediterranean and appears not to have existed in the Caribbean before AD 1492—an observation perhaps in implicit tension with claims for direct long-range colonization events into the heart of the islands from the South American mainland.

It is interesting to note that dramatically increased dispersal capacity did not release island colonists from the impacts of other physiographic variables. In consequence, easing access to remote and/or small islands may paradoxically have increasingly exposed such colonists to the greater risks associated with decreased size and increased remoteness. If so, we might expect exceptional long-range colonizing abilities to co-exist with higher frequencies of localized failures—as is indeed borne out by the examples of Henderson, Pitcairn, Kirimati, Norfolk, and other ‘mystery islands’ in the Pacific (Anderson, 2001), as well as larger islands (Greenland)—or higher rates of subsistence and social crises (for example, Rapa Iti and Rapa Nui: DiNapoli et al., 2018). This ‘bridge-too-far’ effect may account for other less spectacular instances of failed or unsustainable colonization on the outer edges of island worlds.

Trouble at the extremities of expansion apart, one key element of seafaring technology was not simply to enable people to reach islands but to sustain networks of mobility and connection that would allow such ventures to be sustained. Many episodes of island colonization by farming populations are, during their earlier phases, associated with apparently high degrees of interconnectivity. This is evident in the material record, where Lapita (and pre-Lapita in ISEA: Cochrane et al., 2021), Saladoid, and Impressed wares in the Pacific, Caribbean, and Mediterranean respectively index broadly shared potting traditions, as well as long-distance circulation of lithic resources (especially obsidians and cherts: Best, 1987; Çilingiroğlu & Çakırlar, 2013). That this cultural homogeneity reflects deeper inter-community mobility is now supported by various bioarchaeological datasets from the Caribbean and Remote Oceania (see references in Laffoon & Leppard, 2018), as well as by a similar diachronic meta-study from the Mediterranean (Leppard et al., 2020). Paleogenomic studies from the Caribbean, Mediterranean and Pacific suggest this mobility operated at very large scales (e.g., Fernandes et al., 2020; Mathieson et al., 2018; Nägele et al., 2020; Olalde et al., 2015; Skoglund et al., 2016). However, soon after the earliest phases of colonization, inter-island mobility appears to contract in these theaters (if sometimes only temporarily). Interaction networks in the Caribbean become more localized over time (Hofman et al., 2011), as also in the Mediterranean (Broodbank, 2000, pp. 163–164). Inter-island connectivity in the Pacific seems to become less frequent in the aftermath of local colonization horizons, and trading networks in the small fringing islands of New Guinea and Wallacea also operated at reduced spatial scales, albeit highly intensively, starting in the first millennium AD.

The initial pattern of extensive material inter-connections is most readily explicable in terms of the need for small, potentially fragile groups dispersing over extensive areas to maintain long-distance ties to avoid demographic crunches, for example through long-distance exogamy (Jordan et al., 2009; Leppard, 2015). Such marital residence mobility might express itself materially as broad areas of cultural homogeneity if craft production was at least in part a gendered activity. As populations grew after initial colonization, the need to offset the fragility of small populations via exogamy relaxed, long-distance mobility waned, and material culture reflects more parochial patterns.

The broader issue of the distribution of populations (both pre-existent and incoming) is otherwise relevant to some of the regional sequences. Specifically, extant populations may have rendered some options unattractive to newcomers, a factor recognized by Keegan and Diamond (1987) under the banner of ‘competition’ (distribution models in human behavioral ecology predict comparable effects, for example, Jazwa et al., 2019). Norse interaction with indigenous populations, for example, can be understood in these terms. In ISEA, as we have seen, previously established populations on the largest islands seem to have influenced the directions and spatial patterning of incoming maritime groups, who maintained spatially discrete settlement (even if pre-existing exchange networks became potentially attractive to newcomers).

Elsewhere, however, the situation appears to have been rather different. Allee (1931) demonstrated that in a low-density demographic environment an extant population may in fact be an attractor, and some established populations indeed appear to have been an attractive factor. In the Caribbean, the otherwise perplexing spatial overlap between long-established Late Archaic and incoming Early Ceramic Period settlement in the Leewards might be explicable in these terms. Recently established colonizing populations may have been equally attractive. In the Mediterranean, we have suggested that big islands were partly favored due to their capacity to support large, stable populations. Iceland also witnessed rapid in-filling, as did parts of Aotearoa New Zealand and optimal niches in the Hawaiian group.

Allee effects wane rapidly, however, as populations grow beyond the point that optimal habitat is fully occupied, and accordingly lower-ranked habitats start to fill. The North Atlantic is again instructive, with prime settlement sites in Iceland occupied first, and the rapid settlement of the whole island—including marginal regions—prompting further onward movement westwards (Schmid et al., 2021). This process of self-reinforcing growth, followed by increasing demographic pressure that renders the risk of further long-distance colonization increasingly tolerable, should have a discontinuous spatial and temporal signature. It is therefore interesting that temporally discontinuous aspects of island colonization have long been noted: the Remote Oceanian ‘long pause’ (cf. Keegan & Diamond, 1987, p. 67); highly discontinuous movement through the Caribbean (i.e., the Saladoid/post-Saladoid interface: Fitzpatrick, 2015); and arrhythmic (Guilaine, 2001) farming expansion through the Mediterranean. There are other viable explanations for this discontinuous pulsing, as discussed above, including climatic changes and latitudinal constraint—but its broad incidence may indicate some cross-cultural effects of growing island demography in the aftermath of colonization (cf. Leppard, 2014).

Our focus in this synthesis has been on islands, globally. However, as colonizing populations ultimately derived from neighboring continents, it is also instructive to relate processes in these maritime worlds back to those unfolding within continental cores. Our analysis suggests that, in addition to biogeography, and insular and wider maritime social and cultural factors, a range of large-scale processes within the Americas and Afro-Eurasia may be more closely implicated than is broadly recognized in the processes along and beyond their coasts that we witness archaeologically as island colonization.

Firstly, and most obviously, several regional episodes of island colonization should be understood as extensions of bigger, continental Neolithization processes. For reasons we do not address in detail here—although the insular data may hint at the nature of the process—the emergence of food-producing economies was often followed by spatial expansion (Bellwood, 2011; Diamond & Bellwood, 2003). In the Mediterranean, earliest island settlement by agropastoralists should be understood within this broader outflux around the basin (Broodbank, 2013; Leppard, 2021), and in the Gulf of Guinea early activity on Corisco and Bioko aligns well with the Bantu expansion. In the Caribbean, the arrival of more fully horticultural (food-producing) and Arawakan-speaking communities is arguably the outstretched limb of gradual processes of domestication and growth originating deep within the South American continental interior. The expansion of food-producing lifestyles on the coast and islands of East and Southeast Asia is admittedly more complicated and less fully understood, but again there is a suggestive similarity in timing with the terminal Mid and early Late Holocene island colonization horizon in ISEA.

Continental dynamics may be relevant in other ways too, most notably on several of the maritime margins of Eurasia. For example, 3000–1000 BC witnessed the florescence of large and sometimes urban polities in central East Asia (Liu, 2009). This period saw increased maritime mobility, including pelagic fishing and voyaging, in the East and South China Seas, and the initial Yayoi farming colonization of the Japanese main islands is now dated to at least 900 BC. Increased coastal activity in and after state formation on the continent finds parallels in the terms of the growth of maritime and specifically island networks in the Mediterranean world over a similar timespan, driven in part by reorientation of island resources towards new continental consumption centers (Broodbank, 2013). In both cases, targeted archaeological exploration of the critical deltaic junction points might shed light on the translation of continental practices and new riverine transport technologies to maritime and ultimately insular conditions. In ISEA further evidence for maritime mobility and reconfiguration of maritime networks in the early first millennium AD—perhaps linked to Indian Ocean activity of the kind glimpsed in settlement of Madagascar and the Nicobars—occurred in the context of the integrative processes immediately prior to the emergence of Iron Age states by mid millennium (Stark, 2006). In a similar way, the dry, northern circum-African islands of the Canaries and Socotra witness first colonization in the wake of the rise of Iron Age imperial states to their northeast and north, respectively, and the emergence of economic links between these and adjacent areas of Africa. Last but not least, the Norse expansion occurred in the wake of political centralization and integration within Scandinavia, a process that, as Barrett (2008) emphasizes, had ramifications for the negotiation of social status and that created not only winners but also losers, potentially liable to seek their fortunes elsewhere.

These temporal connections are probably too nebulous to demonstrate an unambiguous causality, but are nonetheless suggestive. Several island colonization episodes in the low–mid latitudes certainly correspond approximately with the emergence of large-scale continental societies (although there are clear exceptions, for example the lack of discernable impact of the emergence of the first Mesoamerican states on the Caribbean); and these episodes may be related directly or indirectly. In direct terms, these expansions away from burgeoning centers of political power may reflect Scott’s observation (2009, 2017) that state-making tends to be convulsive. Institutionalization and territorialization demand co-option, and rejection or avoidance of these processes in turn drives mobility. Increased mobility on the fringes of state-making may, then, represent flight away from incipient centers of power. In indirect terms, marginal colonization (i.e., movement into lower-ranked environments) is one in a suite of responses to the types of demographic packing widely attested in complex societies—that is, a common type of ecodemographic pressure is resolved along two distinct pathways with nonetheless comparable archaeological outcomes. Conversely, the new types of wealth and demand often attendant on urban/state societies, alongside expanding economic networks and the comparative ease of bulk water transport, may provide an impetus for the long-range exploitation of marine and insular environments and resources that simply did not exist in the absence of formalized social hierarchies.

These points are far from intended to reduce patterns in global island colonization to functions of mainland process, but adopting a global and deep-historical perspective undoubtedly encourages an evaluation of certain island colonizations in the light of transformative continental dynamics during the Middle to Late Holocene. Equally, it enables a productive move beyond both localized (‘insular’) and generalized (for example, purely biogeographic or environmental) models.