Several authors have produced detailed tectonic reconstructions of the broad SE Asia–Australasia Region that show the tectonic history of New Guinea (e.g. Hall 2002, 2012; Hill and Hall 2003; Zahirovic et al. 2016; van de Lagemaat and van Hinsbergen 2024). While the varying plate motion histories presented by these regional models are useful for understanding the global position of New Guinea through time, they do not capture the detail required to understand the emergence and linkage history of individual islands and land-masses. To address this, high-resolution tectonic reconstructions have been developed for both Indonesian New Guinea (Melia 2020) and Papua New Guinea (Holm et al. 2016). These reconstructions track the tectonic history of New Guinea back to 20 Ma in ~ 1–2 Million year (My) time slices and are based on high-resolution multi-beam and seismic datasets, new geological and isotopic dating studies and re-interpretations of sedimentary provenance and biostratigraphic data (e.g. Holm et al. 2015; Gold et al. 2017; Webb et al. 2020; Ryburn et al. 2024). While these reconstructions focus on different parts of New Guinea, their relative plate motions and records of collision are comparable, meaning that we have combined them here as the base map of our palaeogeographic and emergent land reconstruction of New Guinea.

To produce a unified palaeogeographic reconstruction of New Guinea showing the potential for emergent land and species dispersal over the last 35 My, we used a combination of detailed sedimentology, biostratigraphy, geochronology, thermochronology and lithological descriptions to assess whether emergent land was present in a given region before applying this to the base tectonic map. From here on, we distinguish between ‘New Guinea’ (Fig. 1B), the present-day geography of the island following final uplift of the Central Range in the Pliocene and ‘proto-New Guinea’ (Fig. 2), which refers to instances of continental and oceanic crust along the northern margin of the Australian Plate. This ‘proto-New Guinea’ went through cycles of emergence and submergence prior to present island formation. Criteria used to assess the emergence of land includes the timing of collision and resultant uplift rates from thermochronology and isotopic dating studies (e.g. Hill and Gleadow 1989; Martin et al. 2023; Ryburn et al. 2024), geological map data that records the presence of terrestrial sedimentation or sub-aerial volcanism at a given time period (e.g. Pieters et al. (1983)) and foraminifera biostratigraphic data that records the presence of differing marine environments surrounding areas of emergence (e.g. Gold et al. (2017, 2020)). To further quantify changes in the extent of emergent land-masses through time, we used ImageJ to produce estimates of emergent land area (km²) for each of the palaeogeographic reconstructions shown in Fig. 2. These estimates should be treated as minimum areas of emergence and include all emergent land-masses within the outline of proto-New Guinea at a given time slice along with outlying areas that will eventually form the present-day island of New Guinea (e.g. the Sepik Terrane and Papuan Peninsula, Fig. 2a), but excludes islands that are only briefly or never connected to New Guinea (e.g. Halmahera and New Britain). The estimates do not account for orbital-scale sea-level oscillations that would have either increased or decreased emergent land-mass over a given interval.

To further understand the effects of orbital scale forcing, we then integrated the palaeogeographic and tectonic data with Cenozoic global sea-level reconstructions (Miller et al. 2020, 2024; Rohling et al. 2022) as a guide to identify marine lowstands and highstands. We defined lowstand and highstand events as minima and maxima, respectively, below present mean sea-level (MSL) or above MSL. While different modelled outcomes for relative sea-levels in the Miocene are presented (Rohling et al. 2022; Miller et al. 2024) for our interpretation, the amplitude changes are critical to evaluate emergent states of shallow (ca. <-50 m MSL) marine carbonate platforms. We reason that corals grew vertically with sea-level rise and generated extensive atolls and shallow reef platforms. These platforms then became frequently exposed and connected during sea-level falls of > 20 m to form expanded land-masses that acted as crucial corridors and stepping stones for weakly dispersing birds, such as the birds-of-paradise. It is only during extreme highstand events > 20 m (e.g. the Mid Miocene Climatic Optimum or the Pliocene Climatic Optimum) that sea-levels could have risen too fast and too high for corals to keep up (Petrick et al. 2025), meaning that, during the lowstands in these periods, reef platforms may not have been exposed, thereby increasing isolation and promoting diversification. By not limiting the focus of this work to any single geological proxy and by integrating detailed tectonic reconstructions with new geological data from across the island and global sea-level fluctuations, we can produce a robust interpretation of the extent of emergent land and the linking points between different tectonic fragments.

The islands assessed for palaeo-connectivity in this study coincide with present-day occurrences of the study species Lycocorax pyrrhopterus on the island of Halmahera, L. obiensis on Obi and Semioptera wallacii on Halmahera (Fig. 1). Furthermore, we focus on the islands of Waigeo and Batanta that are home to two species (Paradisaea rubra and Diphyllodes respublica) that both have close relatives on mainland New Guinea. These islands comprise individual micro-tectonic land-masses (MTLs), that remained coherent land units composed of both continental and oceanic crust over geological time and that moved independently from each other (Fig. 2). Ultimately several MTLs merged together over millions of years to form the present-day islands. The palaeogeographic positions of MTLs were plotted for time steps of 1 My over the past 20 My as.proj files in Gplates (using the reconstruction provided in Melia [2020] and Suppl. materials 1, 2) and Geographic Information Software (GIS). We exported the GPlates 2.0 .proj files into .shp files and calculated the Euclidean distances between selected island pairs in ArcMap per 1 My time steps over 20 My. A time window of the last 20 My was chosen, based on the palaeo-positions of the MTLs and to capture important phylogenetic timings. MTLs were projected in WGS_1984_World_Equidistant Cylindrical projected coordinate system, for minimal distortion at the Equator. An uncertainty zone of 10 km was applied to represent the relatively low-resolution of the model reconstruction, based on a 1 mm accuracy of a maximum projected scale for the published tectonic maps at 1: 1,000,000 (Melia 2020), equating to a maximum accuracy of 10 km. In addition, a buffer zone of 10 km of the MTLs outer boundaries was used with the “Buffer” tool in ArcMap when creating MTL features, thereby accounting for the level of additional spatial uncertainty due to coastal changes that likely occurred over the study period. Based on this, we inferred the most probable biogeographic connection for the study taxa when MTL pairs approached each other within a distance of less than 20 km. We then analysed how the distance between MTL pairs changed over time and compared periods of high connectivity with the phylogenetic tree of the study taxa and checked for probable speciation events that were coeval with the timing of closest island pairs.

Our new palaeo-tectonic model with reconstructions of connectivity between mainland New Guinea and Obi/Halmahera identifies two time points when the opportunity for dispersal was particularly favourable; i) between 16.8–12.8 Ma, when the distance between New Guinea and Halmahera was the shortest ever (< 20 km at 13.8 Ma; based on Melia (2020)) and ii) around 8.2 Ma, when low sea-levels reduced the distance between these land-masses significantly. Further back in time (> 17 Ma), Halmahera was more than 100 km distant from New Guinea. More recently (12.8–3 Ma), Halmahera and New Guinea were separated by 80–100 km, but in between these islands, Obi and the Koor (Tosem) Block formed stepping stones effectively reducing the distance to < 50 km. Additionally, sea-level lowstands around 8.2 Ma reduced this distance even further. During the Plio-Pleistocene, Halmahera started drifting further away from New Guinea and despite sea-level drops during the Plio-Pleistocene, the distance between New Guinea and Halmahera remained more than 100 km (Fig. 3). Thus, opportunity for dispersal from New Guinea to Halmahera, leading to the evolution of the endemic bird-of-paradise genera Lycocorax and Semioptera, is most likely to have existed at either around 13.8 Ma or 8.2 Ma.

The phylogenomic tree of birds-of-paradise generated by Blom et al. (2024) recovers Semioptera wallacii as a relatively deep branch. This relationship differs from other studies that also included the majority of birds-of-paradise species (Irestedt et al. 2009; Jønsson et al. 2016; McCullough et al. 2022). Consequently, the divergence times of Lycocorax and Semioptera from their respective sister lineages are very similar, which suggests that the origin of these genera may trace back to dispersal and divergence during a single favourable geological/climatic period. Dispersal around 13.8 Ma is undoubtedly the most favourable time geologically and climatically, as colonisation of Halmahera required crossing a minimal water barrier. However, several recent publications, using fossil calibration points or molecular clock estimates, recover younger estimates for the bird-of-paradise radiation (Johansson et al. in press; McCullough et al. 2022; Stiller et al. 2024; Luo et al. 2025), which are in line with the dispersal estimates at 8.2 Ma.

We used a recent phylogenomic tree for the birds-of-paradise based on concatenation of densely sampled 10 kb. genomic window alignments (Blom et al. 2024). The summary-coalescent species tree, based on the same window alignments, was largely congruent. However, both approaches do not provide an estimate of divergence times and a Bayesian analysis of the same dataset is computationally intractable. We therefore randomly sampled 1000 of the 10 kb. genomic window alignments (autosomal only) and repeated this process of random subsampling five times. We created a concatenated alignment for each of the five datasets and used BEAUti (v.2.7.7) to produce an XML file for BEAST2 (Bouckaert et al. 2019; v.2.7.7). The five datasets were generated to assess consistency and approximate possible random sampling variation. To be able to assess whether Lycocorax and Semioptera colonised Halmahera during one or two geological/climatic events, we calibrated the tree using a single calibration point of the bird-of-paradise tree, namely the split between Lycocorax and Manucodia/Phonygammus. We produced two dated trees, reflecting ideal dispersal times from New Guinea to Obi/Halmahera; at 13.8 Ma (95% Highest Posterior Density (HPD): 12.5–14.9 Ma) reflecting one ideal dispersal time to Obi/Halmahera and a more recent ideal dispersal time at 8.2 Ma (95% HPD: 7.6–8.9 Ma). To do so, we used a normal distribution with a mean of 8.2 and 13.8 and a sigma of 0.35 and 0.6, respectively. We inferred each time-calibrated phylogeny using a strict molecular clock, the GTR+G substitution model (4 gamma rate categories) and the calibrated Yule model as tree prior. For each of the five random loci subsets, we ran three independent replicate runs (each with a different starting seed) for 5 million MCMC generations. Each replicate was visually inspected for convergence using Tracer (v.1.7.2). We excluded the first 10–20% of each replicate as burn-in and then, for each of the five datasets, combined replicates that converged on the same optimum. Finally, for each dataset, we generated a maximum-clade credibility tree with TreeAnnotator (v.2.7.7). For both dated trees, the resulting five time-calibrated phylogenies were highly concordant and only differed marginally in terms of topology or divergence time estimates.

To understand the conditions that controlled the emergent state of proto-New Guinea, a re-assessment of its geology and prevailing sea-level dynamics through time is essential. Prior to 34 Ma, there is little evidence for emergence in proto-New Guinea due to the dominance of marine carbonate deposition within the New Guinea Limestone Group (NGLG - Pieters et al. (1983); Gold et al. (2017)). The New Guinea Limestone Group comprises a series of limestone formations that variably record carbonate shoals, barrier reefs, lagoons and emergent atolls, like the present-day Great Barrier Reef, along with deeper marine pelagic carbonates. In the Palaeocene to Eocene, the NGLG was dominated by deeper water formations, such as the Imskin Limestone (Gold et al. 2017). From the Oligocene onwards, this developed into more shallow water facies and carbonate reefs, alongside terrestrial deposition and volcanism (Gold et al. 2017). Water depths would have been up to 50 m to allow for the growth of corals and shallower conditions are represented by sandy carbonates (calci-arenites), indicative of shallow wave regimes surrounding areas of tectonically driven emergence. Key to our understanding of proto-New Guinea emergence is that, during global sea-level falls and tectonic uplift, this shallow reef area would have emerged, forming a vast area of low-lying terrestrial land, expanding over the present-day region of New Guinea and northern Australia.

Global glaciations started with the unipolar icehouse state of Earth at the beginning of the Oligocene, 34 Ma, when Antarctica was glaciated for the first time – this led to a sea-level fall of at least 50 m (Miller et al. 2020). Coincidentally, this Eocene to Oligocene sea-level fall occurred during the onset of the Peninsular Orogeny in proto-New Guinea from ~ 35 to 30 Ma (van Ufford and Cloos 2005; Zahirovic et al. 2016; Ryburn et al. 2024), (Fig. 2a), which developed during the initial collision of the Sepik Terrane (a composite volcanic arc terrane) with the Australian Plate (Mahoney et al. 2019; Ryburn et al. 2024). This orogenic event led to metamorphism and subaerial exposure of sediments from both the Australian Plate and Sepik Terrane, resulting in a palaeogeography of isolated land-masses like the present-day configuration of New Caledonia and equated to a total emergent land area of ~ 330,000 km² (Davies and Jaques 1984; van Ufford and Cloos 2005; Ryburn et al. 2024). The continued impact of this orogenic event, coupled with Oligocene marine lowstands, led to the deposition of terrestrial to shallow marine sediments in western and eastern New Guinea, associated with the formation of low-relief islands with total emergent land area increasing from ~ 220,000 to 320,000 km² across the Oligo-Miocene boundary (van Ufford 1996; Gunawan 2013), (Fig. 2B, C). Orogenesis and sea-level fall also led to the emergence of a land bridge in the orogenic hinterland connecting northern Australia and present-day southern New Guinea at ~ 34 Ma (Norvick 2002). This low-lying land bridge extended across the Arafura Sea and along the Aru-Merauke Ridge and Cape York Peninsular from ~ 34 to 5 Ma. Evidence for the land bridge is recorded by fluvial gravels deposited in the Gulf of Carpentaria drainage basin, a widespread disconformity in the Papuan Fold Belt and deposition of the clastic Adi Sandstone from an emergent high (Norvick 2002; Harahap 2012), (Fig. 2C). Despite this, fluctuations in relative sea-level likely meant that palaeo-connectivity between these emergent regions was intermittent (e.g. Fig. 2C, D, again ranging from ~ 320,000 to 210,000 km²), with fluctuations of ± 50 m being driven by both Milankovitch periodicity (Miller et al. 2020) and tectonic uplift (Norvick 2002; Zahirovic et al. 2016). The ideal time periods for colonisation as identified by the geological and climatic history align with consistent matches between the periods of marine high and lowstands, tectonic uplift, geological evidence for emergence and MTL connectivity.

The palaeo-connectivity analysis revealed a high level of interaction between MTL pairs throughout the last 20 My (Figs 2, 3; Suppl. material 2), whereby the bulk of the tectonically-driven movement of MTLs occurred along the northern margin of the Australian continent (proto-New Guinea). The highest frequency of interactions occurred between proto-New Guinea and the Philippine-Caroline volcanic island-arc MTLs (including the islands of Halmahera, Obi, Bacan, Morotai and Waigeo), which were translated along the left-lateral strike-slip Sorong Fault Zone. This dominance of strike-slip fault movement occurred after the Peninsular Orogeny during rotation of the Philippine Sea Plate from the Miocene onwards (Hill and Hall 2003) and led to a distinct change in tectonic style across proto-New Guinea. This included periods of subduction beneath proto-New Guinea along the Maramuni Arc, recorded by continental volcanism throughout the Miocene (~ 18 to 12 Ma). This volcanism formed multiple emergent volcanic islands flanked by shallow seas and carbonate reefs along the northern margin of proto-New Guinea (Hill and Hall 2003; van Ufford and Cloos 2005; Holm et al. 2015; Gold et al. 2017; Webb et al. 2020; Tu et al. 2025), (Fig. 2D). This volcanic island configuration is important for understanding palaeo-connectivity because it occurred during the Mid Miocene Climate Optimum (MMCO) at a time when emergent land in proto-New Guinea was at its lowest (~ 210,000 km²). During the MMCO, global icesheet volumes perished and sea-levels reached more than 50 m above the present (Miller et al. 2020); however, tectonic uplift and sustained volcanism allowed pockets of emergent land to persist in proto-New Guinea, as recorded by subaerial eruption of volcanic rocks (Moon Volcanics) and their erosion into marginal marine depositional environments (the Koor Formation) in the north-eastern Bird’s Head (Webb et al. 2020), (Fig. 2D). These pockets of emergent land possibly acted as stepping-stones facilitating birds-of-paradise dispersal in spite of the global sea-level highstand. After the MMCO, global temperatures lowered and the East Antarctic ice sheet started to take shape, this transitional period being denoted as the Middle Miocene Climate Transition (MMCT) and occurred between ~ 14.8 Ma to ~ 12.8 Ma (Leutert et al. 2020; Miller et al. 2020). The MMCT was characterised by a global fall in mean sea-levels up to 30 m with a maximum low stand of about -50 m relative to the present at 13.8 Ma, based on astronomically tuned records (Miller et al. 2020).

Maramuni Arc volcanism in proto-New Guinea ceased at ~ 10 Ma, approximately the same time as carbonate deposition within the New Guinea Limestone Group ended (Pieters et al. 1983; Holm et al. 2015; Webb et al. 2020). This cessation coincides with collision of the Philippine-Caroline volcanic island-arc MTLs with proto-New Guinea from ~ 15–5 Ma (Dow and Sukamto 1984; Crowhurst et al. 1996; Monnier et al. 1999; Hall 2002; Hill and Hall 2003; Webb et al. 2019). This collision propagated westwards bringing islands such as Halmahera into contact with emergent land in northern proto-New Guinea throughout the Miocene and Pliocene (Hall 2002; Hill and Hall 2003; Webb et al. 2019). Collision between Obi, Halmahera and northern proto-New Guinea first occurred at ~ 14 Ma (Baker and Malaihollo 1996; Hall 2002; Melia 2020), (Fig. 2D), before further collision between the Koor (Tosem) Block, Arfak Block, Obi and the Bird’s Head Region from ~ 9 to 6 Ma (Sutriyono et al. 1997; Hall 2002; Webb et al. 2019; Melia 2020), (Fig. 2C, D). When assessing the palaeo-connectivity of these collisional events, we see that the distances between Morotai, Halmahera and Waigeo remain static relative to each other (Figs 2, 3). Morotai and Halmahera remain at a distance of less than 20 km, are considered connected in our analysis and denoted together as the Halmahera Complex, which has ultimately been translated over 1200 km along the northern margin of proto-New Guinea to its present position. Waigeo, on the other hand, always remained isolated from the Halmahera Complex over ca. 200 km. At ~ 17 Ma Obi moved northwards 50 km along a fault strand to connect with the Halmahera Complex and remained close until its collision with proto-New Guinea at 14 Ma (Fig. 2D). While the Halmahera Complex and proto-New Guinea did not make direct contact with each other, connectivity was first established in a stepping-stone fashion via both Obi and Bacan North (Fig. 2D). This connectivity is supported by isotopic data that records periods of uplift in the Obi MTL at 14 and 10 Ma (Baker and Malaihollo 1996). These collisional events coincide with the MMCT sea-level low stands that would have further influenced species exchange between Halmahera, Obi and New Guinea.

From 13 Ma to 11 Ma, Obi moved westwards along proto-New Guinea with variable proximity (~ 60 to 20 km) and separated from Halmahera at 12 Ma, remaining at a distance of at least 60–70 km until present. Between 10.5 and 8 Ma, Obi sheared again against the Bird’s Head Peninsula during collision of the Koor (Tosem) Block, creating widespread uplift and emergence (~ 270,000 to 300,000 km² of emergent land), including local carbonate reefs (Sutriyono et al. 1997; Webb et al. 2019), (Fig. 2E). After 8 Ma, Obi became separated once more from New Guinea (~ 30 km distance; Fig. 2F) until 3 Ma when again it sheared against the western edge of the Bird’s Head Peninsula (Fig. 2G). Finally, in the last 2 My, Obi was translated more than 150 km east of the Bird’s Head Peninsula along the Sorong Fault Zone and remained isolated (Fig. 2H). During the Late Miocene, the Eastern Antarctic ice sheet stabilised more and this stabilisation coincided with a marked global cooling event (Herbert et al. 2016). Around this time, sea-levels fluctuated with relatively low amplitudes perhaps reaching -20 m lower than the present indicating cooler conditions and sea-levels rarely exceeded +25 m (Miller et al. 2020; Rohling et al. 2022). The harmonised and astronomically tuned sea-level curve reconstructed by Miller et al. (2020), (Fig. 3B) reveals a gradual lowering of mean sea-levels between 10 Ma and 7.8 Ma, culminating with a sea level fall of ca. > 50 m at 8.2 Ma, attributed to a 1 My scale Milankovitch event (Miller et al. 2005). Although the lowering of mean sea-levels below zero during the Late Miocene are contested (Rohling et al. 2022), the magnitude of sea-level amplitude change ranged between 25 and 40 m below the present (Stap et al. 2017; Rohling et al. 2022; Miller et al. 2024) and we use the ~ 8.2 Ma lowstand as a proxy anchor point for our reconstructions. These marine lowstands would expose all shallow marine reefs and atolls and lead to extensive land-masses that were used as stepping stones. Indeed, a lowstand succession in eroded carbonates in the Sarawak Basin is part of a major sequence-stratigraphic horizon SEA 72 and attributed to the 8.2 Ma lowstand (Morley et al. 2021). Additionally, Mathew et al. (2020) related seismic karstification stages of emerged carbonate platforms to marine lowstands occurring around 9.5 Ma and ~ 8.8 Ma. These lowered sea-levels, in combination with the collision of Obi and the Koor Block with the Bird’s Head, led to the emergence of shallow marine carbonate platforms. These formed extensive terrestrial land-masses marking a prolonged period of increased connectivity, including the second instance of stepping-stone connectivity at ~ 8 Ma between proto-New Guinea (here represented by the Bird’s Head and Koor Block), Obi and Halmahera to the north (Figs 2F, 3). At the Late Miocene to Early Pliocene transition, global temperatures rose again, leading to an increase in mean sea-levels reaching peak levels higher than 20 m relative to the present (Miller et al. 2020).

Uplift and emergence associated with these palaeo-connectivity events continued into the Plio-Pleistocene with a final collisional phase at ~ 3.5 Ma as fragments of the Melanesian Arc collided with present-day eastern New Guinea (the Adelbert and Finisterre terranes – Abbott (1995); Holm and Richards (2013)). This led to extensive uplift along the island and saw the transition from a proto-New Guinea consisting of isolated volcanic islands and variably connected land-bridges to the present-day topography of New Guinea as a continuous landmass of ~ 500,000 km² (e.g. Tu et al. 2025, Fig. 2G). Uplift rates varied across the island with the Central Range reportedly reaching its present altitude at ~ 3 Ma (Weiland and Cloos 1996), the Kemum Block of western New Guinea reaching present altitude within the last 2 My (Jost 2018) and uplift of eastern New Guinea occurring at ~ 4 Ma (Hill and Gleadow 1989). This variable uplift led to extensive erosion and re-deposition of sediment into the expansive lowlands seen at the present-day, which are directly impacted by global sea-level fluctuations.

After 5 Ma, temperature increases led to the partial melting of the Antarctic icesheets and culminated in the Pliocene Climate Optimum (PCO) period with mean sea-levels ranging between +20 to +23 m (Grant et al. 2019) and reaching maximum highstands up to 40 m between 3.9 and 3.3 Ma (Miller et al. 2020; Dumitru et al. 2021). The highstands during the PCO led to submergence of exposed carbonate reefs, which closed the window of opportunity for dispersal of birds-of-paradise during this period. Moreover, the extreme sea-level rise had led to drowning of carbonate platforms beyond the photic zone, leading to a cessation in coral growth denoted as the Pliocene Reef Gap (Petrick et al. 2025). This would also imply that lowstands during the Pliocene would not lead to exposure of carbonate platforms. With the onset of the ice ages in the Quaternary, mean sea-levels started to fall and the amplitudes of sea-level fluctuations started to increase. From 2.3 to 2.0 Ma, sea-levels reached -50 to -60 m, which would have led to frequent exposure of carbonate platforms (Miller et al. 2020). At around 0.9 Ma sea-level amplitudes increased leading to lowstands reaching -80 m and ultimately more than 130 m lower than at present during the last lowstand 22 ka ago (Lambeck et al. 2014).

The geological history of New Guinea has been highly contentious, resulting in multiple interpretations of the early divergences and build-up of avian diversity. Particularly, hypotheses regarding when and how passerine birds dispersed from their Australo-Papuan centre of origin to other continents have varied, often hinging on different interpretations of available palaeogeographic data (e.g. Jønsson et al. (2011); Moyle et al. (2016); McCullough et al. (2022)). Based on palaeogeographic reconstructions and sea-level models, some studies have proposed that New Guinea acted as a cradle of passerine bird diversity as early as 25 Ma (Jønsson et al. 2011; Aggerbeck et al. 2014). In contrast, others have argued that New Guinea was completely submerged until 15 Ma, implying a more recent colonisation (Moyle et al. 2016).

Our revised palaeogeographical model of the geological evolution of New Guinea, coupled with modelling of sea-level changes, indicates that extensive land-masses were present on the northern Australian Plate as early as 35 Ma (~ 330,000 km² of emergent land). Even during periods of high sea-levels, New Guinea persisted as an archipelago composed of large islands (e.g. proto-New Guinea). These islands were periodically connected during low sea-level stands and episodes of tectonic collision and uplift, creating conditions conducive to extensive diversification amongst early colonisers, including ancestral passerine lineages.

It has been proposed that forest-adapted species were restricted to isolated pockets of wet forest along the periphery of Australia as the continent became increasingly arid during the Miocene (Byrne et al. 2008, 2011). In this context, proto-New Guinea provided extensive wet forest refugia, potentially acting as an important engine for the generation of avian diversity. Our palaeogeographical modelling, which demonstrates the presence of extensive proto-New Guinean landmasses (~ 330,000 to 210,000 km²; Fig. 2) separated from Australia since the mid-Cenozoic, is further supported by biogeographical evidence. Recent studies on lizards (Oliver et al. 2018), insects (Bank et al. 2021), monotremes (Flannery et al. 2022) and pigeons (Oliver et al. 2023) are consistent with scenarios of early colonisation and subsequent diversification within proto-New Guinea.

Data from the new palaeo-tectonic model, coupled with information about sea-level changes through time, identified two optimal times for dispersal from proto-New Guinea to Halmahera (Fig. 3) at 13.8 Ma and at 8.2 Ma. At 13.8 Ma, the distance between these two land masses was the shortest ever (< 20 km) and low sea-levels (approx. 30 m lower than at present) further decreased the water gap. Thus, we first used the time interval of 13.8 Ma (95% HPD = 12.5–14.9) as a calibration point (for the divergence of Lycocorax and its sister group) to date the phylogenetic tree of birds-of-paradise. Using this calibration, we obtain a divergence time for the split between Semioptera wallacii and its sister group at 12.6 Ma (95% HPD = 11.6–13.7). Thus, our dated phylogenomic bird-of-paradise tree shows that the two independent dispersal events from New Guinea to Halmahera (Lycocorax and Semioptera wallacii) happened within a relatively narrow time window (12.6–13.8 Ma [95% HPD = 11.6–14.9]) (not shown). Factoring in the uncertainties surrounding such time estimates, it appears that both colonisation events took place during the same optimal geological/climatic time period.

From 17–13 Ma, Halmahera and Obi were connected, only to become separated again around 12 Ma by approximately 60 km of water (Fig. 3). Colonisation of Halmahera during this time would have involved stepping stone colonisation between proto-New Guinea and Halmahera via Obi and, thus, Obi must have been colonised as well. Our dated phylogenetic tree suggests, however, that separation of Lycocorax obiensis on Obi and Lycocorax pyrrhopterus on Halmahera happened much later, approximately 3.8 Ma during the Pliocene Climatic Optimum (PCO) period when sea-levels were generally high. This makes the timing of colonisation by L. obiensis of Obi from Halmahera implausible and, in turn, speaks against the early colonisation scenario of Halmahera/Obi at 13.8 Ma (Table 1). It is of course possible that the population of Lycocorax that colonised Obi, which remained a small, isolated island (except perhaps at 8.2 Ma when a time of low sea-levels re-connected Obi with Halmahera) became extinct during the Late Miocene. If this were the case, then we would expect a colonisation event from Halmahera to Obi to have led to the formation of two separate species on the two islands. However, this seems very unlikely at 3.8 Ma, as this was a time of particularly high sea-levels. Furthermore, this dated tree suggests that two taxa endemic to Waigeo and Batanta (Paradisaea rubra and Diphyllodes respublica) diverged from their closest relatives on mainland New Guinea around 3.3–3.8 Ma. This is very difficult to reconcile with the geological and climatic history, as Waigeo, at this time, was far away from New Guinea (> 100 km) and as it coincides with the Pliocene Climate Optimum, when sea-levels were particularly high and reef platforms likely never exposed, thus limiting dispersal further (Table 1).

At the second most favourable dispersal time at approximately 8.2 Ma, Halmahera and Obi had drifted past proto-New Guinea, creating a water gap of 60–70 km between Halmahera/Obi and the Bird’s Head part of proto-New Guinea (Fig. 3). However, during the Late Miocene cooling episode, the lowest sea-levels at 8.2 Ma were reached and must have led to extensive emergence of carbonate platforms (e.g. Mathew et al. (2020)) and a significantly reduced water gap. This is locally further supported by geological evidence with the Mio-Pliocene deposition of sandy limestones of the Anggai Formation along the northern edge of Obi and the Weda Formation in Halmahera, indicative of variably emergent carbonate platforms and lagoons (Sudana et al. 1994). We acknowledge ongoing debates regarding the relative sea-level positions between 10 and 7.9 Ma. In our reconstruction, we use 8.2 Ma as an anchor point, though we recognise that future improvements in data and modelling may allow for alternative anchor points in this period.

Dating the bird-of-paradise tree using a divergence time between Lycocorax and its sister-group at 8.2 Ma again shows that the two independent dispersal events from New Guinea to Halmahera (Lycocorax and Semioptera wallacii) happened within a relatively narrow time window (7.4–8.2 Ma [95% HPD = 6.7–8.9]) (Fig. 4), which strongly suggests that colonisation of Halmahera took place at a single favourable geological/climatic time period. At that time, colonisation of Halmahera at 8.2 Ma likely also included colonisation of Obi. This dated phylogenetic tree suggests that separation of Lycocorax obiensis on Obi and Lycocorax pyrrhopterus on Halmahera happened in the Early Pleistocene, approximately 1.9 Ma, when sea-level lowstands reached 60 m below the present. This seems likely as this time marks the onset of global cooling, which led to periodically lowered sea-levels that may have facilitated dispersal of Lycocorax from Halmahera to Obi even if dispersal across a small water barrier were required. Furthermore, this dated tree suggests that Paradisaea rubra and Diphyllodes respublica, both endemic to Waigeo and Batanta, diverged from their closest relatives on mainland New Guinea around 2.0–2.3 Ma. This coincides with a time when Waigeo and northern Batanta were as close as ever to the north-westernmost part of New Guinea (the Bird’s Head) during their transport along the Sorong Fault Zone (Fig. 2H) and fragments of transitional crust from the Bird’s Head had collided with and fused to the southern margin of Batanta Island (e.g. the Saranami Formation). Moreover, the island pairs Waigeo and Batanta were already connected intermittently during the Late Miocene via the variably emergent and karstic reefal Waigeo Formation (Charlton et al. 1991). During the Early Pleistocene, dramatic global periodic cooling and warming oscillations had initiated, which at times led to markedly reduced sea-levels that further facilitated incidental dispersal to Waigeo and Batanta via emergent land-bridges during the lowest lowstands that reached 60 m below the present (Table 1). These lowest lowstands occurred with a periodicity of several > 100 k years (Miller et al. 2020) and, thus, could also have led to subsequent isolations of these island forms on Batanta and Waigeo that ultimately led to the formation of distinct species on the two islands.