An exceptional partial skeleton of a new basal raptor (Aves: Accipitridae) from the late Oligocene Namba formation, South Australia

ABSTRACT The Australian pre-Pleistocene fossil record of Accipitridae (eagles, hawks, old-world vultures) comprises one latest Oligocene or early Miocene and one middle Miocene species, each represented by partial bones. Globally, most fossil accipitrids are based on single bones. The recent discovery of an older and considerably more complete accipitrid from late Oligocene sediments in Australia is therefore significant. It is derived from the Pinpa Local Fauna from the Namba Formation at Lake Pinpa, South Australia (~26–24 Ma). The fossil, described as Archaehierax sylvestris gen. et sp. nov., represents a raptor that was larger than the black-breasted buzzard Hamirostra melanosternon but smaller and more gracile than the wedge-tailed eagle Aquila audax. Comprehensive morphological and molecular phylogenetic analyses resolved Archaehierax as a basal accipitrid, not closely related to any living subfamily and perhaps the sister taxon to all other accipitrids exclusive of elanines. Relatively short wings similar to species of Spizaetus and Spilornis suggest it was adapted for flight within enclosed forests. Additional accipitrid fossils from the Namba Formation, a distal femur and a distal humerus, are incomparable with the holotype of A. sylvestris; they may represent distinct species or smaller individuals of the new taxon. lsid:zoobank.org:pub:6A25C569-3E9F-43B8-AAF8-F36CE405C06E


The Accipitridae and kin
The Accipitriformes comprises four extant families and 259 species: the New-World vultures (Cathartidae, seven species), secretary birds (Sagittariidae, one species), ospreys (Pandionidae, one species) and the eagles, hawks and Old-World vultures (Accipitridae, ~250 species) (Dickinson and Remsen 2013). The accipitrids are the most widely distributed family in the order, being widespread on every continent except for Antarctica, and play key roles as apex predators and scavengers in many environments.
From the 19 th to early 21 st century, the Accipitridae were usually placed in Falconiformes with the Falconidae (falcons), Cathartidae, Sagittariidae and Pandionidae (Sharpe 1874;Ridgway 1874;Gadow 1891;Sushkin 1905;Peters 1934;Jollie 1976;Stresemann and Amadon 1979;Lerner and Mindell 2005;Hackett et al. 2008). Molecular data justified recognising Falconidae and Pandionidae as separate families from Accipitridae, though Pandion is very closely related to the Accipitridae (Sibley and Ahlquist 1990;Wink and Sauer-Gürth 2004;Lerner and Mindell 2005;Griffiths et al. 2007;Hackett et al. 2008). Accipitriformes was resurrected to include Accipitridae, Pandionidae, Sagittariidae and Cathartidae by Christidis and Boles (2008) and this usage has been followed thereafter (Gill et al. 2010;Dickinson and Remsen 2013). Accipitridae is recognised to include multiple subfamilies, though similarities in plumage and morphology (see Peters 1934;Amadon 1964;Jollie 1976) have obscured relationships and composition which are only recently being revealed by analyses of molecular data (refer to Ferguson-Lees and Christie 2001;Lerner and Mindell 2005;Dickinson and Remsen 2013;Mindell et al. 2018). Here, we use Tirari Sub-basin of the Lake Eyre Basin (Alley 1998). The Namba Formation contains three primary Local Faunas (LFs): the Pinpa LF, the Ericmas LF, and the Tarkarooloo LF (Callen and Tedford 1976;Rich and Archer 1979;Rich et al. 1991; Thorn et al. 2021). The Pinpa LF derives from beds of olive and orange mottled clay and white dolomitic mudstone stained with manganese at the top of the lower member of the Namba Formation, which crops out at Lake Pinpa and Billeroo Creek (Tedford et al. 1977;Rich et al. 1991; Thorn et al. 2021). The Ericmas LF derives from fluvial sands deposited in channels cut into the lacustrine units of the Namba Formation, so is younger than the Pinpa LF, and derives from Ericmas Quarry and South Prospect Quarries at Lake Namba a few kilometres south of Lake Pinpa (Tedford et al. 1977;Rich et al. 1991). There is much overlap of species composition between Pinpa and Ericmas LFs suggesting broad similarity in age. While Rich et al. (1991) recognised the Ericmas LF to occur at Lake Pinpa, recent work by THW and ABC recognises only one local fauna from Lake Pinpa and restricts the Ericmas LF to that derived from Ericmas and South Prospect quarries at nearby Lake Namba (Thorn et al. 2021). The Tarkarooloo LF derives from fluvial sands in Tom O's Quarry by Lake Tarkarooloo about 10 kilometres west of Lake Pinpa, but the temporal relationship to the Ericmas LF is unknown, though co-occurrence of some species suggests a broadly similar age .
The accepted age of the Namba and Etadunna Formations has varied (see Pledge 2016). The Etadunna Formation, and therefore by proxy the Namba Formation, was initially considered to be of Oligocene age when it was first identified (Stirton et al. 1961). Subsequent analysis of the Pinpa and Ericmas LFs of the Namba Formation considered their likely age to be middle Miocene (12-16 Ma) (e.g., Tedford et al. 1977;Woodburne et al. 1985) based on the identification of grass pollen in the basal Namba Formation in Wooltana-1 bore that thereby indicated the presence of extensive grasslands (Callen and Tedford 1976). However, later studies of the same pollen slides determined that grass pollen was exceedingly rare and that of Restionaceae or sedges, which occur wherever wetlands are present, was common along with a host of rainforest taxa (Martin 1990). Therefore, Martin (1990) interpreted the palaeoenvironment as a rainforest surrounding swamps and via correlation with floras elsewhere, inferred a late Oligocene -early Miocene age for the basal Namba Formation. Woodburne et al. (1994) detailed the biostratigraphy and revised the age of the Etadunna Formation, using four lines of evidence: 1, the presence of Oligocene age (28-24 Ma) foraminiferal fauna in the Etadunna Formation (Lindsay 1987); 2, a reported Rb-Sr age of 25 Ma on an authigenic illite (Norrish and Pickering 1983); 3, the presence of land mammal fossils consistent with those of known Oligocene age; and 4, the magnetostratigraphic record, to conclude that a late Oligocene age for the Etadunna Formation was most likely. As Woodburne et al. (1994) determined that the Pinpa LF correlated with the most basal Zone A 'Wynyardiid' Fauna of the Etadunna Formation, a late Oligocene age was also inferred for it. Woodburne et al. (1994) did not cite Martin (1990) and were apparently unaware of it, as the pollen data provided compelling evidence to support their inferred late Oligocene age. This work was accepted by Megirian et al. (2010) who established a comprehensive land mammal biostratigraphy and advocated a late Oligocene age (28-24 Ma) for the Namba Formation. Some of the ages accepted by Megirian et al. (2010) have been robustly confirmed by direct dating of sites from Faunal Zones B and C at Riversleigh, northwest Queensland, to the early and middle Miocene, respectively (Woodhead et al. 2016), supporting the prior correlation of Faunal Zone B sites with the Kutjamarpu LF (Wipajiri Fm) from Lake Ngapakaldi, Lake Eyre Basin on biochronological grounds; the Kutjamarpu LF is slightly younger than the uppermost Etadunna local fauna -the Ngama LF -on biochronological grounds (Megirian et al. 2010). Therefore, we use the age range of 26-24 Ma advocated by Woodhead et al. (2016) for the Pinpa + Ericmas Local Faunas.
Aquatic or semiaquatic vertebrates are common in both the Pinpa and Ericmas LFs in the Namba Formation, including fish, turtles, crocodilians, and dolphins (Tedford et al. 1977;Fordyce 1983;Rich et al. 1991), revealing the presence of a permanent lake in the basin during this time period. The dolomitic layers at the site suggest periods of time when sections of the lake would seasonally dry out (Callen 1977). The afore-mentioned pollen record from the Namba Formation reveals that the depositional environment was dominated by a mix of rainforest and temperate forest surrounding sedge-lined lakes in the late Oligocene to early Miocene when the Pinpa LF was present (Martin 1990). The range in habitats resulted in a diverse vertebrate fauna inhabiting this area in the late Oligocene, resulting in a fossil fauna that is a key snapshot into our understanding of the evolution of Australian animals.
From the dolomitic beds of the Namba Formation exposed at Site 12 on Lake Pinpa (Figure 1), and hence forming part of the Pinpa LF, a partial skeleton of an accipitrid was recovered in 2016 by a Flinders University expedition, led by THW and ABC. Sixty-three recognisable bones are represented, spanning the tip of the rostrum to the pedal digits ( Figure 2). The pelvic limb elements are well preserved, except for the femur for which only the caput is preserved. This skeleton represents not only the oldest fossil representative of Accipitridae in Australia but also the most complete (cf. Boles 1993;Gaff and Boles 2010). This relatively complete state is also rare on a global scale, and only observed in a handful of cases, such as the Miocene vultures of China which are near complete (Hou et al. 2000;Zhang et al. 2010;Li et al. 2016), Palaeoplancus sternbergi (see Wetmore 1933;Mayr and Perner 2020) and Aviraptor longicrus (see Mayr and Hurum 2020). Besides the new skeleton, there are another two accipitrid fossils recognised from the Namba Formation; a distal right humerus, from the dolomitic layers at Site 11, Lake Pinpa (also of the Pinpa LF) and a distal femur from Ericmas Quarry, Lake Namba, of the Ericmas LF. It is the aim of this contribution to formally describe these fossils and determine their phylogenetic relationships in the Accipitridae.

Nomenclature
The anatomical nomenclature advocated by Baumel and Witmer (1993) is followed for all bones except for the os carpale radiale, which follows Mayr (2014), and the quadrate, which follows Elzanowski and Zelenkov (2015). Taxonomic nomenclature follows Dickinson and Remsen (2013) and Gill et al. (2020) for composition of Accipitriformes, and Nagy and Tökölyi (2014) for subfamilial composition.

Measurements
Bones were measured with an accuracy of 0.1 mm using digital callipers.

Photography
Photographs were taken using a focus stacking method using a Canon 5DS-r digital camera 50.0 MP with either a Canon EF 100 mm or a 65 mm f2.8 IS USM professional macro lens with multiple images then compiled into a single photo using the program Zerene Stacker. Some fossil specimens were whitened with ammonium chloride powder before imaging (see Feldmann 1989) to retain shape as the primary feature captured rather than variable staining and reflective surfaces.

Comparative material
Skeletons of a broad range of accipitrids and outgroup taxa were loaned from museums and other institutions from Australia and internationally to compare to the fossil as follows.   (1867), showing the bones preserved in the fossil accipitrid specimen SAMA P.54998 shaded in grey. The illustrated taxon and fossil material are not identical in terms of the morphology of individual elements. MNH S.1972.1.59, MNH S.1896.2.16.120, MNH S.1952

Molecular data
Molecular data from Burleigh et al. (2015) was added to the morphological data to improve estimated relationships between living species (Lerner and Mindell 2005;Nagy and Tökölyi 2014;Burleigh et al. 2015). This allows the fossil taxa to be placed phylogenetically according to the signal in the morphological data, but in the context of a DNA-informed tree for living taxa which better accommodates homoplasy in skeletal morphology (see Holdaway 1994;Griffiths et al. 2007). The following genes, well-sampled in accipitrids, were

Phylogenetic analysis
Phylogenetic comparisons were aimed primarily at determining the relationships of the fossil specimen SAMA P.54998. A total of 47 species of Accipitridae, and one species each of Pandionidae, Sagittariidae, Cathartidae, Threskiornithidae, and Ciconiidae were sampled. The non-accipitrid species were selected for the following reasons; Pandionidae, Sagittariidae and Cathartidae are successive sister-taxa to the Accipitridae within the Accipitriformes; the species of Ciconiidae and Threskiornithidae (Ciconiiformes), are examples of bird families outside of Accipitriformes that share similar size and flight morphology, as well as a history of grouping with the Cathartidae in older phylogenies (see Sibley and Ahlquist 1990;Wink 1995).
Both parsimony and Bayesian analyses were used to explore the data. The parsimony analyses of the morphological, molecular, and combined morphological-molecular datasets used PAUP 4.0b10, and heuristic searches. Each search was comprised of 1000 random addition replicates, and enabled TBR branch swapping, with NCHUCK set to 1000. Characters that were inapplicable to a specimen were coded using '-', while missing data were coded as '?'. The taxa Threskiornis spinicollis, Ciconia ciconia, Coragyps atratus, Sagittarius serpentarius and Pandion haliaetus were all defined as outgroup taxa in all analyses, with Threskiornis spinicollis and Ciconia ciconia being the most basal outgroups. Once the heuristic searches had generated a set of most parsimonious trees (MPT), a strict consensus tree was created from them. The support for clades on these trees were then assessed using bootstrapping, with 1000 replicates, and majority-rule consensus trees set to conlevel 50 (support shown if >50%) (see SI.3).
For the Bayesian analyses, MrBayes 3.2.7 was used via the CIPRES platform (Miller et al. 2010). The morphological partition used the standard (Lewis) model for discrete data, with correction for non-sampling of invariant characters. The among-character rate variability was modelled using the gamma parameter, with distribution approximated using four categories. The molecular partitioning scheme and substitution models were identified using PartitionFinder (Lanfear et al. 2016), using BIC. The data was thus treated as three partitions: morphological data (morph); molecular partition 1 (pfinder Molec1), which contains Cyt-B codons 1 and 2, CO1 codons 1 and 2, ND2 codons 1 and 2, 12s, Rag-1 codons 1, 2 and 3, and FGBint67; and molecular partition 2 (pfinder Molec2), which contains Cyt-B codon 3, CO1 codon 3, and ND2 codon 3. The Molec1 and Molec2 partitions each had a GTR model using a Dirichlet prior for the state frequencies. The amongcharacter rates were set to InvGamma, with the gamma distribution approximated as above. All substitution parameters were unlinked across these molecular partitions (see SI.4).
Each analysis entailed four runs, each run comprising four MCMC chains (incrementally heated to 0.1), the number of generations set to 50,000,000, the sample and print frequency set to 5000. Burnin was set to 20% (and confirmed sufficient using PSRF and SDSF values in MrBayes) and the majorityrule consensus tree was obtained from all post-burnin samples. During the MrBayes runs, Ciconia ciconia was set as the sole outgroup taxon, due to limitations of MrBayes, but trees were later rerooted so that both Ciconia ciconia and Threskiornis spinicollis were the most distal outgroup clade. The Bayesian analyses were performed twice: with morphological and molecular branch lengths linked or unlinked.

Ecomorphological analyses
Measurements of selected elements for a sample of extant accipitrids were used to correlate morphology with ecology of living forms and so to retrodict the ecology and feeding strategy of the fossil taxon. Measurements used reflected their availability in the fossil: height of the quadrate; length and proximal width of the carpometacarpus; length, shaft width and distal width of the ulna; length, shaft width, distal width, height of the condylus lateralis, depth of the condylus lateralis, height of the condylus medialis and depth of the condylus medialis of the tibiotarsus; length, shaft width, distal width, and width and height of trochleae metatarsorum 2, 3 and 4 of the tarsometatarsus; length of the first phalanx of pedal digit 1; and the length of the first and second phalanges of pedal digit 2. The fossil species was compared to Elanus scriptus, Hamirostra melanosternon, Pernis apivorus, Lophoictinia isura, Neophron percnopterus, Aegypius monachus, Gyps coprotheres, Spilornis cheela, Haliaeetus leucogaster, Aquila audax, Hieraaetus morphnoides, Spizaetus tyrannus, and Circus assimilis in Principal Components Analysis (PCA). These species were chosen as they were considered to be either, exemplars of the different hunting strategies of Accipitridae, or were potential analogues for the fossil. The PCA was performed on measurements of the wings and legs: raw, log-transformed, and standardised for size by division by the height of the quadrate in each individual (as this was possible for the fossil). The extant taxa were classed based on their preferred habitat (open, woodland, or forest) as determined from the literature (Brown and Amadon 1968;Ferguson-Lees and Christie 2001).
Widths of the distal femur, distal humerus, proximal humerus and distal tibiotarsus were measured for both extant and fossil specimens so that the relationships between these values could be used to predict the width of the distal humerus and distal femur for SAMA P.58917, and thereby to assess whether size precluded the isolated distal humerus from Site 11 Lake Pinpa and the distal femur from Ericmas Quarry from being the same taxon as SAMA P.58917.

Remarks
The fossil is identified as an accipitrid due to the following combination of characters: Skull -Rostrum deep and narrow, with hooked tip and a large, broad nasal aperture; Tibiotarsus -Pons supratendineus ossified, aligned steeply transversely, medially placed, with unbranched canalis tendinosus, and distal condyles much wider than craniocaudally deep; Tarsometatarsus -Robust, with monosulcate hypotarsus, the lateral and medial hypotarsal crests widely separated and trochleae metatarsorum splayed both medially and laterally, and dorsally arched in distal view; Foot -Four digits with raptorial unguals, those of digits 1 and 2 relatively large; Digit IVphalanges 2 and 3 are very short compared to phalanx 4.
The fossil can be excluded from Falconiformes (Falconidae) and the other families of Accipitriformes (Cathartidae, Sagittariidae, Pandionidae) by the morphology of the tarsometatarsal hypotarsus cristae and sulcus. The cristae are fused or partially fused together to enclose the sulcus in Cathartidae, Sagittariidae, and Pandionidae, while in Falconidae the medial crista is connected to the shaft by a ridge that extends two-thirds of its length, features that are absent in the fossil.

Diagnosis
Accipitrids in which the following autapomorphic features are found: the pila medialis of the sternum dorsally separates two deep pneumatic fossae, the humerus has the caput humeri only slightly elevated proximally past the tuberculum ventralis, the tip of the processus procoracoideus of the coracoid sharply curves inwards ventrally towards the medial face of the bone, the tibiotarsus has the lateral/distal retinaculum scar in a deep fossa, the tarsometatarsus is relatively elongate with narrow trochleae metatarsorum that are separated by wide incisurae, and the incisura for the m. flexor hallucis brevis tendon is large, distinct, and extends distal to the fossa metatarsi I. In addition to this, the following features occur: the rostral tip of the rostrum is hooked below the tomial margin at a relatively shallow 30-40° angle, the quadrate has a deep, distinct foramen pneumaticum caudomediale, and the sternum has the apex carinae displaced caudally from the base of the spina externa., Genus Archaehierax Mather, Lee, Camens and Worthy gen. nov.

Diagnosis
An accipitrid distinguished by the combination of the following features; Rostrum. (1) The nares are large and fully open, (2) processus maxillopalatini not fused; Quadrate. (3) the condylus pterygoideus projects less medially than the condylus medialis, (4) a deep, distinct fossa caudomedialis with a small amount of pneumatism; Sternum. (5) The apex carinae is displaced caudally from the base of the spina externa, (6) the medial crista on the carina does not extend to the spina externa; (7) The pila medialis on the dorsal face separates two deep fossae (autapomorphy); Humerus. (8) The caput humeri is only slightly elevated proximally past the tuberculum ventralis (autapomorphy); Os carpale ulnare. (9) Deepened depression on ulnaris face; Tarsometatarsus. (10) The trochleae metatarsorum are splayed and separated by wide incisurae, especially laterally, with the individual trochleae themselves quite narrow in width (autapomorphy); (11) The incisura for the m. flexor hallucis brevis tendon is large, distinct, and extends distal to the fossa metatarsi I (autapomorphy); Phalanx IV.4. (12) The distal articular end that articulates with phalanx IV.5, is considerably wider than the shaft.

Etymology
The species name 'sylvestris' is derived from the Greek world 'sylvas', meaning forest, and the Latin suffix '-estris', meaning 'belonging to'. Type locality/Stratigraphy and age As per genus.

Diagnosis
As for genus.
The rostral section of the rostrum maxillare is preserved in reasonably good condition. Morphology of its rostral tip, tomial margin, rostral margin of the nares, and the palatines is visible.
The rostrum has a preserved length of 32.6 mm from the rostral tip to the posterior base of the nares, and a preserved depth of 17.3 mm from the tomial margin to the dorsal side of the rostrum taken at the rostral end of the nares. (Trait 1) The rostral tip of the rostrum is hooked, descending below the tomial margin at a 30-40° angle.
(2) The lateral tomial margin ( Figure 3A: CT), positioned distal to the nares, is ventrally convex. (3) The nasal aperture ( Figure 3A: N) is large (height 8.7 mm) and fully open as in most accipitrids, spanning just over half the rostrum depth. (4) The ossified section rostral to the incisura ventromedialis is of small to moderate size (10.1 mm preserved length) relative to the total length of the rostrum. (5) In ventral aspect, an incisura ventromedialis ( Figure 3B: IV) (sensu Livezey and Zusi 2007) is present (sediment filled) extending from where damage destroys it rostrally to the preserved caudal end of the palate; it is narrow and widens caudally, rather than being closed forming a fenestra. The pars maxillaris palatini ( Figure 3B: PM) are unfused and diverge slightly caudally, the left being least fragmented although it has broken from the adjacent lateral margin creating a false incision. A small fragment of bone preserved between the pars maxillaris palatini is interpreted as a displaced fragment of the processus maxillopalatinus. (6) Damage precludes ascertaining the presence/form of the fenestra ventrolaterale.
Other accipitrid subfamilies differ as follows: (Trait 1) Compared to the fossil, the tip of the rostrum is much more sharply hooked ventrally in most subfamilies. Only members of Elaninae, Perninae (except Chondrohierax uncinatus, which is sharper), Buteoninae, and Gypaetinae have similar or shallower angled tips.
The rostrum maxillare is overall most similar to that of species of Buteoninae (see SI.2 for more detailed differential comparisons).

Quadrate (Figure 3(C, D))
The left quadrate has considerable breakage affecting the lateral side ventrally, the medial side of the processus oticus, and loss of the processus orbitalis. Preservation of morphological detail is best medially. On the processus oticus, only about half of the capitulum squamosum is preserved and the dorsal half of the crista medialis is lost. Of the processus orbitalis, only the well-preserved base remains. Both the condylus pterygoideus and medialis are intact, but the entire condylus lateralis and caudal half of the condylus caudalis are lost.
(Trait 1) The processus oticus ( Figure 3D: POt) is short and broad leading up to the capitulum squamosum. (2) The capitulum squamosum ( Figure 3D: CS), as preserved is relatively small, and has a tuberculum subcapitulare forming a distinct hook projecting ventrally on its cranial margin.  FPC) is present just medial of the ventral-most point of the crista medialis. (10) The condylus pterygoideus ( Figure 3D: CP) is distinct, high-set and well separated from the condylus medialis. (11) The condylus medialis ( Figure 3C: CM) is well preserved, showing a large (4.0 mm wide by 2.8 mm deep), semi-ovular facet, with a pointed medial margin that extends further medially than the condylus pterygoideus. (12) The portion of the condylus caudalis ( Figure 3C: CC) preserved indicates a facet of a similar size to the condylus medialis, with a semicircular shape.
Other accipitrid subfamilies differ as follows (variable characters excluded): (Trait 8) The sulcus running along the ventral crista is broad and indistinct in all other taxa except in species of Milvus and Haliaeetus (Haliaeetinae) where it is narrow and indistinct. (9) The fossa caudomediale is practically absent in the subfamilies Circaetinae and Aegypiinae, as well as the species in Hamirostra, Lophoictinia (Perninae), Neophron (Gypaetinae), and Haliaeetus (Haliaeetinae); shallow in Aquilinae and species of Elanoides, Chondrohierax (Perninae), Polyboroides and Gypohierax (Gypaetinae); and deep in Elaninae, Accipitrinae and Buteoninae, as well as species in Pernis (Perninae). The depressio was indistinctly shaped (i.e. a gradually deepened area rather than a defined pit) in all species, and apneumatic in all species except in Pernis and Haliaeetus.
The quadrate is overall most similar to that of species of Aegypiinae (see SI.2 for more detailed differential comparisons).

Vertebrae (Figure 3E-J)
The atlas vertebra of SAMA P.54998 is 10.7 mm wide by 10.7 mm high (from the proximal margin of arcus atlantis to the distal margin of fossa condyloidea). (Trait 1) The arcus atlantis ( Figure 3G: Arc. At.) forms a low, flat arch, which dorsally has a maximum proximodistal width of 3.2 mm at the centre, overhanging the fossa condyloidea cranially. The rest of the distal corpus atlantis is worn away and cannot be assessed, but what is present suggests that few additional structures were present, and that there was some asymmetry in the shape of the lateral fossa condyloidea.
The axis vertebra is quite fragmented, with most of the neural spine, facies articularis atlantica, dens, facies articularis caudalis, processus ventralis and the area of incisurae caudales arcus broken away. Its width is 12.7 mm across the processus articulares caudales and mid-line length of the corpus vertebra is 8 mm. The facies articularis on zygopophyses caudales are 3.1 mm wide by 3.8-4.1 mm long. (10) A short but caudally prominent projection is present dorsal to each facet ( Figure 3H: FA), measuring 2-2.5 mm in width. (11) There is no evidence of a bridge enclosing the incisurae caudales arcus ( Figure 3F, H: PAC).
The three known caudal vertebrae are too broken to identify their position in the caudal series.
The vertebrae are overall most similar to that of species of Elaninae (see SI.2 for more detailed differential comparisons).

Sternum (Figure 4(A, B, C))
The cranial section of the sternum of SAMA P.54998 is preserved, retaining the structure of the spina externa, pila carinae, crista medialis carinae, apex carinae, the left sulcus articularis coracoideus, and the left labrum internum. It is characterised by: (Trait 1) A spina interna is absent; a small notch exists in its place.
(2) The spina externa ( Figure 4B: SE) is 6.4 mm wide at its base, 4.7 mm wide at its blunt tip and 4.2 mm long. In cranial view, the spina externa is triangular as a medial crista forming a lobe projecting 3.6 mm ventrally. The base of the spina externa is broader than the apex carinae (4.0 mm). Extant accipitrid subfamilies differ as follows: (3) The crista medialis extends to the base of the spina externa in members of all subfamilies except Aegypiinae. (8) The apex carinae lies directly ventral to the base of the spina externa, or projects more craniad, in species in all subfamilies except for Gypaetinae and Aegypiinae. (11) The sulci articularis coracoidei overlap in all species except for Gampsonyx swainsonii (Elaninae) and Sarcogyps calvus (Aegypiinae). (13) No species in any subfamily has a distinct pila medialis separating pneumatic fossae in the pars cardiaca, which is thus an autapomorphy suggesting subfamilial distinction for the new species.
The sternum is overall most similar to that of species of Aegypiinae (see SI.2 for more detailed differential comparisons).
Coracoid (Figure 4(F, G, H, I)) The well-preserved left and right omal ends and fragments of both sternal ends of the coracoids of SAMA P.54998 were recovered. They reveal the following: (Trait 1) A foramen nervi supracoracoidei ( Figure 4G: FoNS) is present and located adjacent to the shaft rather than near the medial margin of the processus procoracoideus; (2) The foramen lacks an opening into the corpus; and (3) it is small, about 1 mm in width, and positioned just sternal of the cotyla scapularis. (4) A large (6 mm wide) pneumatic foramen is present in the sulcus m. supracoracoidei ( Figure 4F: SMS). The width of the sulcus is approximately 14.5 mm from the ventrosternal corner of the facies articularis clavicularis to the facies articularis humeralis, and 12.8 mm from the medial margin to the laterodorsal margin immediately cranial to the cotyla scapularis. (5) The facies articularis clavicularis ( Figure 4F: FAC) is large, broad, and clearly delineated sternally by a crista that dorsally overhangs the aforementioned foramen, and ventrally is a low non-overhanging crista. The sternal margin of this facet is straight with no notch nor dorsal or ventral projections directed sternally. (6) The cotyla scapularis ( Figure 4G: CtS), preserved on the right omal fragment, is deep and large (6.7 mm wide by 5.7 mm long) in relation to the processus procoracoideus and triangular shaped. (7) The facies articularis humeralis ( Figure 4G: FAH) is 7.7 mm wide and 12.8 mm long. (8) The impressio lig. acrocoracohumeralis ( Figure 4H: ILA), best seen on the left specimen, forms a distinct sulcus ~7 mm wide by 18.1 mm long on the processus acrocoracoideus, although this may be exaggerated by damage to the fossil. (9) The processus procoracoideus ( Figure 4I: P. Procor.) forms a short projection medially, barely as long again as the cotyla scapularis width, with its tip sharply angling ventrally towards the medial face to partly enclose the triosseal canal. (10) The best preserved sternal-end fragment shows that the angulus medialis is acute, forming a 30-45° angle. (11) The medial side of the facies articularis sternalis is 6.5 mm wide at its broadest point, and shallow, with little deepening towards the dorsal margin.
The coracoid is overall most similar to that of species of Aegypiinae, Accipitrinae and Buteoninae (see SI.2 for more detailed differential comparisons).

Scapula (Figure 4(D,E))
Both the left and right scapulae of Archaehierax sylvestris gen. et. sp. nov. are almost complete, lacking only the distal third or less of the corpus scapulae. In total, the preserved craniocaudal length of the scapulae is 56.8 mm (left) and 53.3 mm (right).
The proximal dorsoventral width of the scapula is 14.4 mm from the acromion to the ventral side of the facies articularis humeralis.
(1) The tuberculum coracoideum ( Figure 4E: TC) is low and barely cranially prominent dorsal of the facies articularis humeralis. (2) The acromion ( Figure 4D: Ac) has a distinct cranio-laterally oriented crista lig. acrocoracoacromiali dorsally, and a robust rounded medial prominence. The ligamental attachment on the margo dorsalis has a very small prominence and is not elevated dorsally above the rest of the margo dorsalis.
The scapula is overall most similar to that of species of Elaninae (see SI.2 for more detailed differential comparisons).
The humeri are poorly preserved in SAMA P.54998. Only the caput humeri, crus dorsale fossae, fossa pneumotricipitalis, and the incisura capitis of the proximal end of the right humerus is preserved. The left humerus is more complete, preserving about 60mm of proximodistal length of the proximal end including the caput humeri, crus dorsale fossae, fossa pneumotricipitalis, incisura capitis, sulcus lig. transversus, facies bicipitalis, crista deltopectoralis, and some of the proximal shaft. However, there is also significant breakage and fracturing of the bone surface in this specimen, which has resulted in the loss of the tuberculum dorsale, the ventral margin of the crista bicipitalis, tuberculum ventrale, and sulcus n. coracobrachialis. These specimens reveal the following: (1) The incisura capitis ( Figure Figure 5C: SLT), best seen in the left humerus, is shallow but well defined, and seems continuous between the ventral and dorsal sections. Ventrally, the sulcus is deep and round, measuring 6.3 mm wide by 4.5 mm long cranial to the incisura capitis. The crista deltopectoralis ( Figure 5B: CrD), while quite fractured, is preserved in its entirety in the left specimen. Preserved length is 42.7 mm from the assumed position of the tuberculum dorsale to its distal end. (7) The profile of the proximal section of the dorsal margin of the crista between its origin near the tuberculum dorsale and the angulus cristae of the crista deltopectoralis is flat in a ventro-cranial view. (8) The angulus cristae of the crista deltopectoralis is very prominent and distinctly triangular in dorsal view. (9) Distally, the crista deltopectoralis, while fractured, projected mainly cranially (shaft margin visible proximal to the distal point of crista).
Extant accipitrid subfamilies differ as follows: (Trait 1) In all subfamilies except Aegypiinae and Spilornis cheela in Circaetinae, species have a deep incisura capitis; (3) The caput humeri is more elevated proximal to the incisura capitis and tuberculum ventrale, ranging from a moderate (Elaninae) to a large proximal projection (all other subfamilies) so a low flattened caput is identified as an autapomorphy of the species.
The proximal humerus is overall most similar to that of species of Elaninae, Aegypiinae, Aquilinae, Haliaeetinae and Buteoninae (see SI.2 for more detailed differential comparisons).
Ulna (Figure 6(C, D, E)) SAMA P.54998 preserves a near-complete left ulna, reassembled from fragments, that is only missing the olecranon, parts of the ventral margin of the cotyla ventralis contiguous with the olecranon, and the caudodorsal margin of the cotyla dorsalis. The distal right ulna is also preserved with the condyles mostly intact, with only the ventrocaudal margin of the condylus dorsalis and condylus ventralis worn away. They reveal the following features: (1) The ulna is largely straight in dorsal and ventral view, with only very slight caudal curvature towards the proximal and distal ends. The processus cotylaris dorsalis projects distally of the cotyla ventralis ( Figure 6E: PCD), is 5.8 mm wide, and (2) is quadrangular in shape with a flattened dorsal tip between parallel equal-length  Abbreviations: CD, condylus dorsalis; CrI, crista intercotylaris; CtV, cotyla ventralis; CV, condylus ventralis; DL, depressio ligamentosa; DR, depressio radialis; FAR, facies articularis radiocarpalis; FAU, facies articularis ulnaris; IB, impressio brachialis; IR, incisura radialis; IST, impressio scapulotricipitis; IT, incisura tendinosa; ITC, incisura tuberculum carpalis; PCD, processus cotylaris dorsalis; SI, sulcus intercondylaris; ST, sulcus tendineus; T, tuberculum; TAV, tuberculum aponeurosis ventralis; TBR, tuberculum bicipitale radiale; TCr, tuberculum carpale; TLCV, tuberculum ligamentum collateralis ventralis. Scale bar 10 mm. depressed relative to the shaft and is 12 mm long proximodistally. The midshaft of the left specimen is 7.7 mm craniocaudally wide in dorsal aspect. (8) The papillae remigales caudales form low, barely prominent scars, which is typical of most accipitrids. The distal end of the ulna measures 12.9 mm wide (left) and 12.2 mm wide (right) between the cranial point of the tuberculum carpale ( Figure 6C: TCr) and the caudal margin of the condylus dorsalis ( Figure 6C: CD) in ventral aspect. (9) The tuberculum carpale is short and blunt, or rounded, in dorsal and ventral view, (10) with a flattened facet directed ventrodistally. (11) The incisura tuberculum carpale ( Figure 6C: ITC) forms a distinct notch separating the tuberculum carpale and condylus ventralis ( Figure 6C: CV) when viewed in dorsal aspect. The condylus dorsalis (left specimen) is 13.5 mm long proximodistally along its caudal margin in ventral aspect, and 9.3 mm deep from the caudal margin to the incisura tendinosa, in caudodorsal view. (12) The caudal margin of the condylus dorsalis forms a continuous curve in the proximal half, best visible in either craniodorsal or caudodorsal view, interrupted only by a small notch for the incisura tendinosa ( Figure 6E: IT). (13) The incisura tendinosa lies between the condylus dorsalis and the condylus ventralis (dorsal aspect), though it does not quite separate the two proximodistally. (14) The depressio radialis ( Figure 6C: DR) is shallow and not pneumatized. (15) The sulcus intercondylaris ( Figure 6C: SI) forms a relatively deep v-shape in ventral aspect. (16) The condylus ventralis distinctly projects distocranially, and measures 5.7 mm wide (in ventral aspect) by 11.2 mm deep (in cranial aspect).
The ulna is overall most similar to that of species of Circaetinae (see SI.2 for more detailed differential comparisons).

Radius (Figure 6(A, B))
In SAMA P.54998, the left radius is complete, preserving most features of the proximal (cotyla humeralis slightly worn ventrocaudally) and distal ends.
The cotyla humeralis is large, measuring 5.5 mm deep dorsoventrally, and 4.1 mm wide. It shows the following: A tuberculum bicipitale radiale ( Figure 6A: TBR) is located 5.4 mm distal of the facies articularis ulnaris on the dorsal face. The tuberculum has (1) a large, deep, non-pneumatic fossa (2.6 mm wide by 3.8 mm proximodistal length) on it, and (2) has a distinct profile in cranial view as a low, quadrangular ridge.  (Figure 6(A, B): FAU) forms a prominent bulb that projects out ventrally, and which has a deep notch on the proximal margin that gives it a doublepeaked appearance.
The radius is overall most similar to that of species of Haliaeetinae and Gypaetinae (see SI.2 for more detailed differential comparisons).
It has the following features, terminology from Mayr (2014) The os carpale radiale is overall most similar to that of species of Elaninae and Buteoninae (see SI.2 for more detailed differential comparisons).
Os carpale ulnare (Figure 7(I, J)) Both the right and left os carpale ulnare are complete in SAMA P.54998. Measurements (mm) right/left: dorsoventral width 12.2/11.8, craniocaudal depth (excluding cranial projection) 4.4/4.6, and proximodistal length 9.9/9.2. They show: (1) A distinct projection, roughly in the centre of the cranial face ( Figure 7J: CrPro), extends well cranially above the margin of the proximal ligament attachment, and lacks a prominent ridge that makes it contiguous with the proximodorsal corner.  Extant accipitrid subfamilies differ as follows: (Trait 2) The ventral cranial face is higher set cranially than the dorsal cranial face in all subfamilies except Accipitrinae. (3) There is no notch on the proximal end of the crus longus in all subfamilies except in the genera Aquila (Aquilinae), Haliaeetus (Haliaeetinae) and Buteo (Buteoninae), which have a shallow notch. (8) There is no deepened depression on the caudal face in all subfamilies except Aquilinae, Aegypius monachus (Aegypiinae), Accipitrinae, and the genus Buteo (Buteoninae). (9) There is no raised crista on the caudal face in all subfamilies except in the genus Buteo (Buteoninae).
The os carpale ulnare is overall most similar to that of species of Elaninae, Perninae, Gypaetinae and Aquilinae (see SI.2 for more detailed differential comparisons).
Carpometacarpus (Figure 7(A, B)) For SAMA P.54998, the left carpometacarpus is almost complete, missing only the cranial section of the facies articularis digitalis majus.
(2) A rounded ridge extending from the processus pisiformis to the trochlear rim separates the fossa infratrochlearis from (3) an extremely deep sulcus ( Figure 7A: SV) between the processus pisiformis and the processus extensorius. This sulcus is elongate and extends from the trochlea carpalis ( Figure 7B: TrC) to adjacent to the processus alularis ( Figure  Measurements -see Appendix 1 Table S1; overall, the carpometacarpus has a proximal craniocaudal width that is equivalent to 25% of the total length, which is moderately gracile. Extant accipitrid subfamilies differ as follows (Trait 1) The fossa infratrochlearis is shallow, except in Perninae (deep in Pernis apivorus and Chondrohierax uncinatus, the latter also pneumatic) and Gypaetinae (deep). (19) The proximal section of the os metacarpale minus has a shallower groove, except in P. typus (Gypaetinae, deep) and Haliaeetus leucocephalus (Haliaeetinae, deep), or in Elaninae (flat, ungrooved).
The carpometacarpus is overall most similar to that of species of Buteoninae (see SI.2 for more detailed differential comparisons).
(1) The proximal L MI.1 has a dorsoventral width of 6.6 mm and a craniocaudal depth of 6.5 mm, and from proximal view is triangular. (2) The cranial margin is tapered into a thin crista that continues along the length of the preserved bone. (3) A small tuberculum is present on the caudal margin of the ventral face, close to the proximal end. (4) The dorsal margin and face of the proximal end is much more protruding than the ventral margin and face and has two ligament attachment points on its dorsal surface. (5) The width of the bone narrows distally. (6) The L MIII.1 is 14.2 mm long and has a prominent caudal projection slightly less than halfway distally along its length, the tip of which is oriented caudoproximally. (7) Both the dorsal and ventral face lack any sort of depression or sulcus. Only the proximal end of the MII.1 is preserved, which is 8.3 mm wide by 6.9 mm deep.  The manus bones are overall most similar to those of species of Perninae (see SI.2 for more detailed differential comparisons).
Femur (Figure 8(A, B)) In SAMA P.54998, only the caput of both femora has been preserved, preserving the articularis acetabularis face and the fovea ligamenti capitis. The width of the caput is 7.6 mm. The fovea ligament capitis ( Figure 8A: FLC) is shallow and set in the proximal margin of the caput. The articularis acetabularis face ( Figure 8B: FAAce) is not well defined from the medial face.
Tibiotarsus (Figure 8(C, D, E, F, G)) Both the right and left tibiotarsi are preserved in SAMA P.54998. The left tibiotarsus is almost complete with only the proximal articular surfaces missing. It preserves the base of the cnemial crests and the entire crista fibularis, but damage to the distal end obscures the details of the pons supratendineus, the tuberculum retinaculum m. fibularis, and the distal insertion scar for the retinaculum extensorium tibiotarsi. The right tibiotarsus is missing the entire proximal end, and the distal end could not be reconnected to the shaft but is very wellpreserved revealing most features of interest.
As observed on the left element, (1) the impressio lig. collateralis medialis ( Figure 8C: ILCM) is a slightly elevated tuberculum on the medial face, measuring 4.5 mm wide and 8.9 mm long.  Figure 8F: DEL) is shallow, and bordered by a flattened, broad crista on the distal margin of the lateral condyle. The trochlea cartilaginis tibialis is difficult to assess due to breakage, (27) but appears largely flat.
The tibiotarsus is overall most similar to that of species of Buteoninae (see SI.2 for more detailed differential comparisons).

Fibula
The proximal ends of the left and right fibulae are preserved in Archaehierax sylvestris gen. et. sp. nov. Craniocaudal depth is about 10.2 mm, while width is 3.7 (right) and 4.0 mm (left).
(1) A shallow fossa is present in the cranial half of the proximal lateral face. (2) The caudal face has a very shallow depression just distal to the caudal projection. (3) The medial face has a broad but shallow sulcus that extends from near the proximal margin down the shaft.
Tarsometatarsus (Figure 9) The right and left tarsometatarsi are both imperfectly preserved in SAMA P.54998. The left tarsometatarsus preserves the original length of the bone, though the medial half and the proximal end, from mid-length on the medial side to and including the area proximal to the foramen vasculare proximale laterale, has been dorsolaterally twisted approximately 90° relative to the rest of the bone. The lateral half is thus undistorted from just proximal to the foramen vasculare proximale laterale distally. The cotyla lateralis is missing with about half of the eminentia intercotylaris. The foramen vasculare proximale mediale is obscured by the distortion on both the dorsal and plantar side.
The right specimen has the distal half preserved well with all features, but the proximal half is so badly fragmented that nearly all identifying features are lost. Only the crista medialis hypotarsi is recognisable. The specimens reveal the following features: (1) The length of the tarsometatarsus is about 66-75% of the length of the tibiotarsus (uncertainty allows for the missing proximal end of the tibiotarsus). (2) The length to distal width (maximal across trochleae) ratio is approximately 1:6 and shows the tarsometatarsus is moderately elongate among accipitrids. (3) The cotyla medialis ( Figure 9D: CtM) is deep and with a notably convex dorsal margin.
(4) The eminentia intercotylaris ( Figure 9D: EI) projects a few millimetres proximally to the rim of the cotyla medialis. (5) The crista lateralis flexoris hallucis longus (sensu Mayr 2016) (lateral hypotarsal crista) and the crista medialis flexoris digitorum longus (medial hypotarsal crista) ( Figure 9E: CMFDL) are not fused together plantarly, and so form a wide monosulcate hypotarsus. (6) The medial hypotarsal crista is distinctly proximodistally longer (8.9 mm from proximal margin to distal hook, 11.4 mm from proximal margin to distal termination point) than the lateral hypotarsal crista (6.0 mm). (7) The plantar depth of the lateral hypotarsal crista is 13.6 mm (76%) of The impressio retinaculi extensorii, preserved on the dislocated cotyla medialis, are a pair of distinctly projecting cristae, with the retinaculum itself unossified, which is the typical state among the accipitrids. (14) The fossa infracotylaris dorsalis is shallow in the undamaged section distal to the eminentia intercotylaris and towards the retinaculi. (15) There is a distinct sulcus extensorius ( Figure 9A: SExt) at mid-length, which opens to the medial face around two-thirds of the distance distally along the shaft. (16) The medial half of the proximal 40% of the shaft is highly compressed as it is in many accipitrids, forming a crista 1.3 mm thick. (17) The crista plantaris lateralis ( Figure 9G: CPL) is well-developed, extending from the hypotarsus to level with the fossa metatarsi I. (18) In lateral aspect the crista plantaris lateralis is markedly projecting plantarly, deepest just proximal to mid-length.  Figure 9H: TII) and IV are almost identical (9.1 mm and 9.5 mm respectively, measured from right specimen). (26) Trochlea metatarsi III ( Figure 9H: TIII) is located comparatively much higher dorsally, and while it has a depth of 7.7 mm, the plantar-most point of it is separated from that of trochlea metatarsi IV by about 6.6 mm. (27) Trochlea metatarsi II has a robust profile in distal view, with a short, robust plantar projection on its outer margin and a deep fovea ligamentosa collateralis. (28) Trochlea metatarsi III has a robust profile in distal view and has a shallow medial groove dorsodistally. (29) Trochlea metatarsi III is laterally directed relative to the shaft axis. (30) Trochlea metatarsi IV is the narrowest of the trochleae in distal view, with a short, thin plantar projection on the lateral margin. (31) The flange on trochlea metatarsi II is moderately projecting plantarly, (32) while the flange on trochlea metatarsi IV is quite prominent and plantar oriented. (33) The distal extent of the trochleae metatarsorum II and III is roughly equal and surpass distally the distal margin of IV.
Extant accipitrids differ across all subfamilies as follows: (9) The sulcus flexorius is shallower, with the cristae plantares lateralis and medialis relatively low in elanines, gypaetines, circaetines, haliaeetines, and buteonines. (21) The incisura m. flexor hallucis brevis is shallower and shorter ending at or proximal to the fossa metatarsi I in all observed species. (23) In all subfamilies, the incisura intertrochleae are relatively narrower and the autapomorphically wide incisura in the fossil is one of its most characteristic features.
The tarsometatarsus is overall most similar to that of species of Elaninae (although more elongate), Aquilinae and Circaetinae (see SI.2 for more detailed differential comparisons).
Digit I ( Figure 10A) The os metatarsale I in SAMA P.54998 is fairly robust, with the proximal end attenuated to a thin point. In dorsal and plantar view, the region immediately proximal to the medial side of the articular surface for I.1 is inflated, creating a > 90° angle just distal of the midlength point along the 'shaft'. The attachment facet for the tarsometatarsus is quite long, extending to be adjacent with the previously mentioned inflation, but is not prominent in lateral view. The sulcus for a tendon on the distal dorsal face is bordered by a reduced crest that is positioned slightly medial of centre in the metatarsal. The phalanx I.1 is long and moderately robust, with the enlarged proximal end much wider than the corpus. Plantarly, the tubercula flexoria are enlarged, enclosing a broad sulcus that extends as a shallow attachment point to roughly midway on the corpus distally. The lateral side of the face dorsalis of the proximal end is slightly inflated into a ligamental attachment point, forming a modest, rounded mound. The foveae lig. collaterales on the distal end are very deep, and there is only a very shallow indentation set between the two foveae on the dorsal face. The ungual phalanx I.2 (as seen in right side) is slightly larger than the ungual phalanx II.3, with mild curvature along the ungual phalanx. The phalanx I.1 and ungual I.2 exhibit notable hypertrophy in contrast to digits III and IV, a trait that is present in almost all Accipitridae (Fowler et al. 2009).
Digit II ( Figure 10B) Like most accipitrids, there is no fusion of the phalanges II.1 and II.3. The species in Haliaeetinae and Ictinia are notable for such fusion (see Jollie 1976). The phalanx II.1 is quite short compared to phalanx II.2, being just under half its length and considerably shortened lengthwise (not compressed lateromedially), which is a common trait in Accipitridae. All phalanges are notably hypertrophied compared to those in digits III and IV.
Digit III ( Figure 10C) Four phalanges are present. The medial face of the ungual phalanx (III.4) lacks the central ridge present in most accipitrids and falconids, though it is possibly that this feature has been poorly preserved.
Digit IV ( Figure 10D) The midshaft width of phalanx IV.4 is 3.2 mm, compared to the 3.8 mm of phalanx I, and the distal end of phalanx IV.4 is distinctly widened, measuring 4.6 mm. This dramatic shift in width along the digit does not appear in the sampled elanine genera (Elanus, Gampsonyx), species of Hieraaetus (though the section before the articular end is swollen) or Spizaetus, but appears to a lesser degree in the pernines, gypaetines, circaetines, aegypiines, species of Aquila, and species of Haliaeetus.

Summary
Archaehierax sylvestris shares a mosaic of characters across a broad range of taxa and thus the above comparisons do not reveal clear affinity with any one taxon. Different elements in the fossil skeleton differ markedly as to which subfamilies they most closely resemble: the rostrum maxillare -buteonines; the quadrate -aegypiines; the vertebrae -elanines; the sternum -aegypiines; the coracoidaegypiines, accipitrines, and buteonines; the scapula -elanines; the humerus -elanines, aegypiines, aquilines, haliaeetines and buteonines; the ulna -circaetines; the os carpi radiale -elanines and buteonines; the os carpi ulnare -elanines, pernines, gypaetines and aquilines; the carpometacarpus -buteonines; the tibiotarsus - buteonines; and the tarsometatarsus -elanines (fossil is more elongate), aquilines and circaetines. There are several autapomorphies which further differentiate it from all extant subfamilies. Notably these include the sternal basin having a medial bar separating deep pneumatic fossae, humerus with very low proximal projection of the caput, and tarsometatarsus with broad incisurae intertrochleae and the incisura for the m. flexor hallucis brevis tendon extending distal to the fossa metatarsi I. Together, these support differentiation of this taxon with separate subfamilial status, consistent with the phylogenetic results discussed below.

Comparison with fossil accipitrids
Australia has only two described pre-Pleistocene accipitrids. Pengana robertbolesi is from Sticky Beak Site in the Riversleigh World Heritage Area, of ? Late Oligocene -Early Miocene age, which is now considered one of the Faunal Zone A sites (Travouillon et al. 2006) that on biochronological grounds are slightly younger than the Pinpa LF (Woodhead et al. 2016). It is represented by a distal tibiotarsus (Boles 1993), and while of similar size, is easily distinguished from Archaehierax sylvestris by the following characters: the distal margin of the pons supratendineus is angled less steeply, ~30° relative to the long axis; the condyles have flattened sides in cranial aspect and are not medially and laterally expanded relative to the distal end of the shaft. Aquila bullockensis from the mid-Miocene Camfield Beds (12 Ma) of Bullock Creek (Gaff and Boles 2010;Megirian et al. 2010) was described from a distal humerus -which was not preserved in Archaehierax sylvestris. However, A. bullockensis is very much (>10 Ma) younger than Archaehierax sylvestris, much larger, and the morphology of the distal humerus was interpreted to be typical of species of Aquila. Archaehierax sylvestris has many features on other bone elements that exclude close affinity with both Aquila and the Aquilinae, so conspecificity with A. bullockensis can be ruled out.
In relation to Oligocene-age fossil accipitrids from elsewhere in the world, the geographic isolation of Australia makes it unlikely that any described species are closely related to Archaehierax sylvestris. As reviewed in the Introduction, most late Oligocene and early Miocene accipitrid species are found in North America and Europe. Nearly all of them are described from a single skeletal element, making assessment of relationships with Archaehierax sylvestris difficult. The late Oligocene -early Miocene accipitrids from North America, including the relatively complete Palaeoplancus sternbergi, are all easily distinguished from Archaehierax sylvestris by size, and by morphology of the distal tarsometatarsus; specifically, trochlea metatarsi II is relatively broader and/or the intertrochlear incisions are much narrower.
Four Oligocene fossil accipitrids are described from Europe, all but one of which is based on a single bone: Aquilavus hypogaea (Milne-Edwards, 1892), from the Quercy fissure fillings, is incomparable as it is based on a femur.
Aviraptor longicrus Mayr and Hurum, 2020, of early Oligocene age from Poland, is described from a complete skeleton. It is a very small accipitrid with highly elongate legs like those seen in species of Accipitrinae, which clearly distinguishes it from Archaehierax sylvestris.
Three fossil accipitrids are known from early Miocene deposits of Europe; Promilio incertus (Gaillard, 1939) was described from a right tarsometatarsus from Chavroches, France, which lacks the wide incisura of Archaehierax sylvestris, the flange on the trochlea metatarsi II is oriented more medially, and the hypotarsal crests are of roughly equal craniocaudal depth (Gaillard 1939, Figure 9).
The middle Miocene accipitrids from Asia are all aegypiine vultures (Hou et al. 2000;Zhang et al. 2010Zhang et al. , 2012Li et al. 2016) and so are not closely related (see phylogenetic analysis below).

Material
Distal right humeral fragment, preserving a relatively unworn distal end and 16.2 mm of shaft, and some associated fragments of the shaft, SAMA P.58917.

Remarks
The fossil can be excluded from other raptor families on the following features: Falconidae (falconid state in brackets): the condylus dorsalis is thickened and rounded distally (consistently narrow and rectangular); the processus flexorius ends proximal to the condylus ventralis (equidistant).
Sagittariidae (sagittariid state in brackets): the two fossae marking the attachment points for the lig. collaterale dorsale are positioned roughly adjacent to each other (cranial-most fossa slightly proximal to and abutting caudal fossa in sagittariids).
The fossil is broadly similar to accipitrids and displays the following features: (1) The tuberculum supracondylare dorsale ( Figure 11A: PSD) is located well-proximal to the condylus dorsalis ( Figure 11A: CD) and is small, barely projecting dorsally of the shaft, but projects slightly cranially as a proximodistally elongate rugosity; (2) the dorsal face/shaft margin between the tuberculum supracondylare dorsale and the epicondylus dorsalis is mildly inflated; Two shallow scars for the m. extensor carpi radialis are present on the tuberculum supracondylare dorsale ( Figure 11C: MECR), (3) the larger palmar attachment scar on the cranial face adjacent to the dorsal margin is oval (4) and the smaller dorsal scar is located on the dorsal face of the processus. (5) In caudal view, the processus flexorius (Figure 11: PF) terminates proximal to the condylus ventralis ( Figure 11A: CV) but is prominent ventrally. (6) The sulcus scapulotricipitalis ( Figure 11B: SST) forms a shallow but broad notch roughly 2 mm wide on the caudal face. (7) The fossa olecrani ( Figure 11B: FO) is moderately deep, defining well the dorsal margin to the processus flexorius but does not create a discontinuity with the sulcus humerotricipitalis. (8) The sulcus humerotricipitalis ( Figure 11B: SHT) is very shallow, and at 5.3 mm wide extends over half of shaft width of 9.7 mm at the same point. (9) The fossa m. brachialis ( Figure 11A: FB) is shallow but distinct, with a proximodistal length of 13.8 mm extending well proximal to the tuberculum supracondylare dorsale, and a maximum dorsoventral width of 7.3 mm level with the proximal margin of the The fossil has the most similarities to species from the subfamily Elaninae (see SI.2 for more detailed differential comparisons), but differs markedly in regards to the inflation of the dorsal face between the tuberculum supracondylare dorsale and the epicondylus dorsalis, the size and shape of the palmar and dorsal attachment scars for the m. extensor carpi radialis, the distinct depression in the section of dorsal face caudal to the tuberculum supracondylaris and the epicondylus dorsalis, the sulcus humerotricipitalis width, the fossa m. brachialis length, the configuration of the insertion scars on the distal epicondylus ventralis, the position of the distal margin of the condylus dorsalis relative to that of the condylus ventralis in cranial view, the ventral projection of the processus flexorius, and the connectivity of the condylus ventralis and entepicondyle in cranial view.
As the Archaehierax sylvestris specimen SAMA P.54998 lacks a preserved distal humerus, it cannot be compared to SAMA P.58917. However, it is not believed to belong to the same species due to the significantly smaller size of SAMA P.58917 from the humerus size predicted for SAMA P.54998 (see comparative measurements below).

Remarks
The specimen can be excluded from the Pandionidae and Cathartidae by the presence of a single muscular attachment on the planum popliteum, and from Falconidae and Sagittariidae by the linea intermuscularis caudalis remaining level and visible on the medial margin of the caudal face.
The femur is consistent with accipitrids and has the following morphology.
(Trait 1) The linea intermuscularis caudalis ( Figure 12C: LIC) is highly distinct, running along the medial border of the caudal shaft face in a raised line, (2) but is not continuous with the tuberculum m. gastrocnemialis medialis, so there is no crista supracondylaris medialis. The distal femur NMV P.222435 is from an accipitrid which exhibits the most similarity to those of species in Buteoninae, Aegypiinae, and most of Elaninae (see SI.2 for more detailed differential comparisons). It mainly differs from species in these subfamilies in lacking a prominent crista supracondylaris medialis, the position and shape of the attachment point on the planum popliteum, and the weak projection of the epicondylaris lateralis.
As the distal femur is not a highly diagnostic section of the accipitrid skeleton, and the distal femur is not preserved in Archaehierax sylvestris specimen SAMA P.54998, NMV P.222435 is regarded as gen. et. sp. indet. The size difference between NMV P.222435 and the predicted size of the distal femur of SAMA P.54998 is greater than would be predicted from typical sexual dimorphism, which makes it unlikely the two are representatives of the same species (see comparative measurements below).

Size comparisons of the three fossils
The width measurements of the proximal humerus, distal humerus, distal tibiotarsus and distal femur of extant taxa were compared (see Appendix 1, Table S2) and showed that the distal width of the humerus was between 80% and 90% of the proximal width of the humerus, while the distal width of the tibiotarsus was between 75% and 110% the distal width of the femur in extant accipitrids. If the bones of Archaehierax sylvestris had similar ratios, then it can be predicted that the width of the missing distal humerus should fall in the range 23.4-26.4 mm, while that of the missing distal femur should be between 15.8 and 22.0 mm broad. Based on this, both the isolated distal femur NMV P222435 and the isolated distal humerus SAMA P.58917 are too small to belong to an individual the size of the A. sylvestris holotype. However, sexual dimorphism is known to be considerable and common in accipitrids (Brown and Amadon 1968;Marchant and Higgins 1993) and raises the possibility that these isolated fossils may belong to a smaller sex of the one species if they fall within a certain size range. Field et al. (2013) devised multiple algorithms for predicting body mass from skeletal measurements, while Campbell and Marcus (1992) predicted body mass based on the femur and tibiotarsus circumference. Using these, the mass of the bird for the Archaehierax sylvestris holotype is estimated as 3.7 kg based on the length of the coracoid facies articularis humeralis, 4.6 kg by the least shaft diameter/width of the tarsometatarsus, and 3.2 kg based on tibiotarsus least shaft circumference. The mass of the bird represented by the distal femur is calculated at 2 kg based on femur shaft width/diameter, or 1.6 kg based on shaft circumference. The mass of the bird represented by the distal humerus is calculated at 1.5 kg based on shaft width/diameter, or 1.6 kg based on circumference. Assuming these predictions are accurate, the femur represents a bird 46-67% smaller than the skeleton specimen, and the humerus one 60-67% smaller. This would be pushing accipitrid sexual dimorphism to its extreme limits, making it unlikely that the fossils represent a single species. However, these mass predictions use different elements, limiting their comparability. Nevertheless, while considering it likely that at least two accipitrids are represented, we consider it unwise to describe the smaller as a second species when size would be the only distinguishing factor and their congeneric status cannot be assessed. Figure 13. PCA plots using length measurements of the carpometacarpus, ulna, humerus, tibiotarsus, tarsometatarsus, pedal digit 1 and pedal digit 2 treated in three ways. (A) Absolute data, (B) log-transformed data, (C) size standardised data with variables proportional to quadrate height. Directional arrows at top right indicate directionality of limb length (S, short and L, long) along the PC axes. Note: axes in A and B have been scaled for better visualisation, so 2D distances do not represent true 2D distances in PCA space. Abbreviations: Ae. mon., Aegypius monachus; Ar. syl., Archaehierax sylvestris; Aq. au., Aquila audax; Ci. as., Circus assimilis; El. scr., Elanus scriptus; Gy. cop., Gyps coprotheres; Ham. mel., Hamirostra melanosternon; Hal. leug., Haliaeetus leucogaster; Hi. mor., Hieraaetus morphnoides; Lo. isu., Lophoictinia isura; Ne. per., Neophron percnopterus; Pe. api., Pernis apivorus; Spl. ch., Spilornis cheela; Spz. tyr., Spizaetus tyrannus. Dark green, forested habitat; light green, woodland/open forest; orange, open habitat (grassland, savannah etc.). Fossil (Archaehierax) indicated by black square.

PCA analysis of limb measurements
Length data for a range of post cranial measurements were visualised in PCA plots to determine if there was any correlation between them and preferred habitat. All PCAs used a variance-covariance matrix, iterative imputation for missing data (in the case of I.2 length of Gyps coprotheres), and 1000 bootstrap replicates. See Appendix 2 for datasets, scree plots, biplots, PCA values.
The first PCA used absolute length measurements of the carpometacarpus, ulna, tibiotarsus, tarsometatarsus, pedal digit 1 and pedal digit 2 ( Figure 13A). In the resulting scatterplot PC1 (92.1% variance) was most strongly driven by the ulna, with some influence from the carpometacarpus (wings), the tarsometatarsus and tibiotarsus and PC2 (7% variance) by the tarsometatarsus and tibiotarsus (legs). Archaehierax sylvestris was positioned as a long-legged, short-winged taxon, well separated from other species. Both Spizaetus tyrannus and Spilornis cheela grouped closely together, creating a cluster for forest-habitat accipitrids. Circus assimilis, which inhabits grassland and open woodland, was positioned intermediate between Archaehierax sylvestris and the forest taxa.
A second PCA was run after log-transforming the measurements. In the resulting scatterplot ( Figure 13B) PC1 (91.4% variance), was driven by almost all measurements, with those of the tibiotarsus and tarsometatarsus having slightly more influence than those of the wings and digits, and PC2 (3.3% variance) revealed that species were separated most strongly based on the tarsometatarsus length, with lesser influence from the digit lengths and tibiotarsus length. Archaehierax sylvestris grouped with the long-legged and short-winged taxa, but the distribution of the extant taxa changed. Spizaetus tyrannus and Spilornis cheela were more widely separated, with the open-habitat taxon Circus assimilis positioned more closely to Spizaetus tyrannus.
As size dominated the first two PCAs, a third PCA was performed with measurements standardised for size, by division of postcranial data by the height of the quadrate, an element which correlates strongly with skull size and therefore body size (Elzanowski et al. 2001). In the resulting scatterplot ( Figure 13C), PC1 (67.2%) was most strongly driven by ulna length and to a lesser degree by carpometacarpus length, while PC2 (28.9%) was most strongly driven by tibiotarsus length and tarsometatarsus length. Archaehierax sylvestris occupied a more negative position on PC2 relative to Circus assimilis as the peak of the long-legged, shortwinged taxa, and Spizaetus and Spilornis clustered together closely once more. Archaehierax sylvestris fell intermediate between Circus assimilis and the forest accipitrid cluster.

Phylogenetic analyses
We performed phylogenetic analyses of morphological data only, and combined morphological and molecular data, using parsimony and Bayesian methods. We discuss all analyses below, but have most confidence in the analyses combining morphology and molecules, in particular the unlinked Bayesian analyses, for reasons discussed at the end.

Analysis 1: Parsimony, morphology only, unordered characters
The first analysis used only morphological data, with no ordering, constraints or weighting applied to the characters. The resulting 30 most parsimonious trees (hereafter MPTs) had a tree length of 1686 steps (SI.5Figure 1). Coragyps atratus, Ciconia ciconia, Threskiornis spinicollis, and Sagittarius serpentarius were rooted as the outgroup (PP = 97%), while Pandion resolved as sister to Accipitridae with a support value of 97%. This is broadly concordant with independent molecular phylogenetic studies.
Within Accipitridae, the tree is less congruent with DNA trees. The Accipitridae as a family had strong support (87%) with the non-Australian Perninae resolved as the most basal clade, which was strongly supported (87%) but had species left in a polytomy.
The fossil Archaehierax sylvestris n. gen. et sp. resolved as a branch between the Circaetinae-Harpiinae-Aquilinae clade and all other subfamilies higher up the tree. However, support for this position was very weak (<50%).

Analysis 2: Parsimony, morphology only, ordered characters
Analysis 2 differed from Analysis 1 by ordering certain multistate characters which formed morphoclines (see SI.1). This generated four MPTs with a tree length of 1720. The resulting strict consensus tree (SI.5Figure 2) is largely the same as for analysis 1, but with the following differences. The Accipitridae resolved with strong support slightly higher than the previous analysis (PP = 88%).
The fossil Archaehierax sylvestris was resolved as being between the Elaninae and the Australian endemic Perninae on the phylogenetic tree, though support for this position was very weak (<50%).

Analysis 3: Parsimony, morphology and DNA, ordered characters
As the analyses based on morphology failed to resolve the taxa in a way that reflects strongly supported clades based on comprehensive molecular data, and the primary aim of the analysis was to assess how the fossil related to the well-corroborated clades of modern taxa, molecular data from six genes was added for 47 taxa (see Methods) forming a combined morphology and molecular data matrix used in Analysis 3. Parsimony analysis of this matrix produced three MPTs with a tree length of 1831 (See SI.5Figure 3).
Given the molecular data largely constrains the tree to the relationships dictated by molecular data alone, relationships were mostly the same as those in recent molecular studies (Nagy and Tökölyi 2014;Mindell et al. 2018).
The position of the fossil Archaehierax sylvestris varied between the strict consensus tree and the bootstrap majority consensus tree of the same analysis. In the strict consensus tree (SI.5) the fossil resolved as nested within the Circaetinae, sister to Pithecophaga jefferyi. However, the bootstrap consensus tree resolved the fossil as its own branch between the Perninae-Gypaetinae and the Circaetinae-Aegypiinae clades with moderate (68%) support.

Analysis 4: Bayesian inference, morphology + DNA, ordered
The Bayesian analysis with molecular and morphological branch lengths unlinked produced a broadly similar tree for living taxa to the bootstrap consensus of the corresponding parsimony analysis, but with overall much stronger supports for higher-level clades ( Figure 14). All subfamilies resolved as monophyletic, and the divergence nodes for all subfamilies and major clades were greater than 70% except for one.
The fossil Archaehierax sylvestris resolved as a lineage between the Elaninae and the Perninae-Gypaetinae clades (i.e. non-elanine accipitrids). Support for Archaehierax plus a clade of all non-Elanine accipitrids was weak (44%), but there was moderate support (73%) for monophyly of all other non-elanine accipitrids excluding Archaehierax.
When the branch lengths for the molecular and morphological data were linked (SI.5Figure 4), the position of the fossil changed. Archaehierax sylvestris moved up the phylogeny and resolved as an independent branch above the Circaetinae-Aegypiinae clade but below the Harpiinae and relatives. Support for this node was stronger than that of the position resolved by the unlinked analysis, but still weak (56%).

Summary
All phylogenetic analyses resolved Archaehierax sylvestris with the Accipitridae, consistent with the conclusions drawn from the morphological descriptions, though its precise position within that family varied. Some analyses found it deeply nested within Accipitridae, closely related to, but outside buteonines, haliaeetines and accipitrines. These analyses include the morphology-only parsimony analyses, morphol-ogy+molecular parsimony and morphology+molecular Bayesian analysis with linked branch lengths. However, as discussed below, these deeply nested affinities for Archaehierax are problematic, and appear less plausible than the topology retrieved in the Bayesian analysis with branch lengths unlinked -where it was one of the most basal accipitrid lineages, with only Elaninae diverging before it ( Figure 14).
A more precise and robust position for Archaehierax sylvestris is perhaps prohibited by missing data. Even with the 63 preserved elements, there is still a significant amount of missing data. The mandible and cranium, most of the sternum, the distal ends of the humeri, the pelvis, and most of the femora were not preserved. Thus, only 45% (135/300) of phylogenetic characters could be assessed in SAMA P.54998.

Discussion
Pengana robertbolesi was previously the only accipitrid raptor known from the late Oligocene in Australia (Boles 1993;Worthy and Nguyen 2020), being slightly younger than Archaehierax sylvestris at 24-20 Ma (Travouillon et al. 2006;Woodhead et al. 2016), and represented only by a distal tibiotarsus making relationships within Accipitridae difficult to establish. The specimens from the late Oligocene Namba Formation are the oldest accipitrids in Australia and extend the fossil record of the Australian Accipitridae to 26-24 Ma, when Australia was much warmer and heavily forested.

Relationships with fossil and extant Australian Accipitridae
Archaehierax sylvestris is unambiguously an accipitrid based on many skeletal features, but notably the morphology of the tarsometatarsal hypotarsus, the lack of a spina interna on the sternum, and the shortened second and third phalanges of the fourth digit. Unsurprisingly, Archaehierax sylvestris has multiple unique features of its skeletal morphology that distinguish it from other accipitrids, such as the low caput humeri (humerus), the two fossae in the pars caudale separated by a pila medialis (sternum), and the wide incisurae intertrochleares (tarsometatarsus). However, as summarised above, the different elements of Archaehierax sylvestris do not reveal a consistent closer relationship to the species in any one subfamily. Some elements, such as the rostrum, carpometacarpus and tibiotarsus, show much similarity to species in more derived subfamilies like the Buteoninae, while others, such as the quadrate, vertebrae, and sternum resemble those of more basal subfamilies like Elaninae and Aegypiinae. Other elements, like the humerus, ossa carpalia and the tarsometatarsus, share features with multiple subfamilies. The scapula, ulna, radius, carpometacarpus, carpal phalanges, fibula and pedal phalanges do not align well with the species of any one subfamily. The morphology of the os carpi radiale also excludes the fossil from an accipitrid clade comprising Harpiinae, Aquilinae, Haliaeetinae, Buteoninae and Accipitrinae (see Mayr 2014). This mix of affinities among characters contributes to understanding why the fossil does not group robustly in any subfamily in the phylogenetic analyses. This typifies many Palaeogene fossil bird species across multiple families (see Mayr 2009) and, along with the phylogenetic results, supports the idea that Archaehierax does not belong to an extant subfamily. Missing data probably exacerbates the problem as about 55% of characters could not be coded.
In our parsimony analysis using combined morphological and molecular data SI.5, Archaehierax sylvestris resolved either deeply nested within Circaetinae as sister to Pithecophaga jefferyi (strict consensus tree) or as a stem lineage situated between the clade Gypaetinae-Perninae and Aegypiinae-Circaetinae (bootstrap consensus tree). The Bayesian analysis of combined morphological and molecular data with morphology and molecular branch lengths linked had A. sylvestris resolved as above the Aegypiinae-Circaetinae clade but lower than the Harpiinae and Aquilinae. In contrast, the Bayesian analysis of combined morphological and molecular data, with unlinked molecular and morphological branch lengths, resolved Archaehierax sylvestris near the base of the Accipitridae, immediately above the Elaninae. The topology of the unlinked molecular and morphology branch lengths tree ( Figure 14, analysis 4a) is preferred for several reasons; firstly, given the age of the fossil, a more basal position on the accipitrid phylogenetic tree is more plausible. Dated molecular phylogenies imply that most of the extant accipitrid subfamilies had not diverged by the late Oligocene, with only the Elaninae, which diverged at 33.7 Ma (Mindell et al. 2018), likely present, as the Perninae +Gypaetinae clade diverged at 23.8 Ma from remaining accipitrids (Mindell et al. 2018). Other lineages emerged during or after the middle Miocene (Nagy and Tökölyi 2014;Oatley et al. 2015;Prum et al. 2015;Mindell et al. 2018). Secondly, while many analyses of combined morphological and molecular datasets link branch lengths between these data types (e.g. Ronquist et al. 2012), this might be justifiable only under certain circumstances. Duchéne et al. (2020) compared the effects of linking branch lengths of gene loci trees and demonstrated that partitioning and linking loci to create proportionate branch lengths gave the strongest support, while analyses that had unlinked loci, or loci that were linked to produce identical branch lengths, received weaker support. Goloboff et al. (2019) explored the question of whether assuming a common mechanism of evolution to both all genetrees and morphological data was warranted and concluded that morphological data was generally not compatible with the common clock assumption used when linking branch lengths, producing low levels of branch length correlation. Similarly, Barba-Montoya et al. (2021) found a poor linear relationship between branch lengths for morphological and molecular data, consistent with the idea that the morphological traits were evolving at much more variable rates compared to the molecular ones. Based on this, the results estimated by linking morphological and molecular branch lengths (SI. 5) should be regarded with caution.
The molecular-based divergence dates for Aquilinae (Nagy and Tökölyi 2014;Mindell et al. 2018) suggest that Aquila bullockensis, at 14-12 Ma (Woodhead et al. 2016), pre-dates the inferred age of the Aquila genus by at least 5 Ma. Morphologically the holotype distal humerus has several distinct differences from Aquila audax, including a reduced distal projection of the processus flexorius, a tuberculum supracondylare ventrale with less cranial projection and no proximal narrowing, little to no convexity between the processus supracondylare dorsale and epicondylus dorsalis, and the dorsal insertion for the m. extensor radii is positioned offset from the processus supracondylare dorsale. This does not necessarily mean that A. bullockensis is not an aquiline, but rather that it is unlikely to be a member of the crown Aquila or any other extant aquiline genus, and that the initial comparative descriptions were too limited to support referral of the species to Aquila. Since the description of the holotype, more fossil material that likely belongs to A. bullockensis has been discovered and is awaiting description, which may change interpretations of the relationship of A. bullockensis to the extant Aquilinae.
In relation to size, it is clear that Archaehierax sylvestris was a large accipitrid, smaller than the wedge-tailed eagle Aquila audax and the white-bellied sea eagle Haliaeetus leucogaster but larger than the black-breasted buzzard Hamirostra melanosternon among the extant Australian fauna. It is tempting to assume from this that it must belong to a lineage of large accipitrids. However, while size is sometimes useful in diagnosing clade membership, there are notable exceptions. Haast's eagle Hieraaetus moorei, one of the largest eagles ever known at an estimated 15 kg (Worthy and Holdaway 2002), is most closely related to the little eagle Hieraaetus morphnoides (Bunce et al. 2005;Knapp et al. 2019), which weighs under 1 kg, the two diverging from a common ancestor approximately 1 million years ago (Knapp et al. 2019). Another example is seen among extant species; the Philippine eagle Pithecophaga jeffreyi is morphologically convergent on the Harpiinae in terms of large body size, prey preference, and preferred habitat, but groups molecularly with the Circaetinae, most of which are medium-sized reptile specialists.
Regardless of its closest extant relative, Archaehierax sylvestris demonstrates the presence of Accipitridae in Australia since the late Oligocene and that there were at least two divergent clades (A. sylvestris and Pengana robertbolesi) in Australia around the Oligo-Miocene boundary. All well-sampled Australian faunas from the late Oligocene onwards are now known to have contained accipitrids Boles 1993;Gaff and Boles 2010;Rich and van Tets 1982;Louys and Price 2015;Worthy and Yates 2018). However, the geographic origin of these austral accipitrids is difficult to infer, given the presence of accipitrids across multiple continents at this time and a lack of phylogenetic analyses of them with fossils elsewhere.
Conspecificity of the isolated distal humerus and femur specimens from other sites in the Namba Formation with Archaehierax sylvestris could not be excluded based on morphology given the holotype lacks these elements, although their much smaller size makes this unlikely, exceeding differences attributable to sexual dimorphism. However, in the absence of overlapping skeletal elements, establishing whether they are congeneric or not is impossible, so we refrain from describing them as a new species.

Palaeobiology
Archaehierax sylvestris is inferred to have inhabited forested areas, based on the pollen records from the Namba Formation (Martin 1990) and the associated fauna in the Pinpa LF, which contains many arboreal taxa such as koalas (phascolarctids), and members of four families of possums and kin (phalangeriforms) (see above; Rich et al. 1991). Our principal component analyses show the fossil taxon grouped most closely to species with relatively shorter wings and longer legs. This body form is observed in forest dwelling eagles and hawks, such as species of Spizaetus, Spilornis, Harpia harpyja and Pithecophaga jeffreyi, which are adapted to flying through more constricted spaces among the trees and vegetation (Brown and Amadon 1968;Holdaway 1991 unpublished thesis). However, it is also present in the spotted harrier Circus assimilis, which Archaehierax sylvestris was also closely associated with in PCA plots, which favours open grassland and lightly wooded areas for its habitat (Brown and Amadon 1968;Marchant and Higgins 1993;Debus 1998). In the case of Circus assimilis, however, wingspan to leg length proportion is less the result of the wings being shortened, but more the product of the legs being hyper-elongate compared to other accipitrids, especially in the tarsometatarsus, to facilitate a specialised hunting strategy. Circus assimilis is known to forage by slowly flying less than five metres above vegetation (Aumann 2001) and has been documented pursuing small prey such as lizards on foot (Buij 2014). It is likely that the high ratio between the wing and leg length in C. assimilis is therefore being driven by a need to reach into grass cover to quickly grab small vertebrates before they can escape. As Archaehierax sylvestris does not exhibit the extreme elongate tarsometatarsus morphology observed in C. assimilis, and more closely resembles that of the crested serpent-eagle Spilornis cheela and the black hawk-eagle Spizaetus tyrannus, it can be inferred that the ecology of Archaehierax sylvestris was more akin to the latter species.
With its shorter wings allowing manoeuvrability, Archaehierax would not have been a particularly fast flier, but would have been capable of more agile twists and turns in flight than an accipitrid of its size with a typical wingspan. If we use extant forest eagle species such as those in Spizaetus as a morphological analogue, it can be assumed that Archaehierax sylvestris was likely an ambush hunter, waiting on a perch within forest cover until prey came into range, and then attacking with a quick burst of speed (Whitacre and Jenny 2013).
The potential diet of Archaehierax can be inferred based on that of living analogues, such as species in Spizaetus. A female ornate hawk-eagle Spizaetus ornatus was recorded feeding on the remains of an estimated 3.2 kg Central American agouti (Dasyprocta punctata) and later, on a great curassow (Crax rubra), which can weigh between 3.1 and 4.8 kg (though it was not directly observed killing these animals). Whitacre and Jenny (2013) recorded a male with an adult great tinamou (Tinamus major) around 1 kg in weight, which is also the average weight for male Spizaetus ornatus. Archaehierax sylvestris is notably larger than the species of Spizaetus observed in this project (Spizaetus ornatus, Spizaetus tyrannus), and assuming similar prey hunting abilities, would have been quite capable of preying on many of the mammals and birds known from the Pinpa Local Fauna.
Based on its larger physical size, phylogenetic position, and the proportions of the tibiotarsus to the tarsometatarsus, it is unlikely that Archaehierax sylvestris was restricted to preying on large invertebrates and small vertebrates as seen in the elanines and some of the pernines. The extant Australian elanines, the letterwinged kite Elanus scriptus and the black-shouldered kite Elanus axillaris, primarily feed upon small mammals (typically mousesized), lizards, and large insects such as beetles, grasshoppers and locusts (Brown and Amadon 1968;Marchant and Higgins 1993). The pernine kite Hamirostra melanosternon feeds upon small mammals (rabbit-sized at largest), reptiles, and birds, and has been observed to break open eggs of large ground-dwelling birds using either stones or its beak (Brown and Amadon 1968;Marchant and Higgins 1993). The square-tailed kite Lophoictinia isura preys on a wide array of small birds, reptiles, large insects, and even bird eggs from nests (Brown and Amadon 1968;Marchant and Higgins 1993).
However, Archaehierax sylvestris also lacks the robustness of legs seen in the species of aquilines and harpiines that feed on larger birds and small to medium mammals. The morphology and tibiotarsus-tarsometatarsus ratio of the fossil are also slenderer compared to the fish eagles/haliaeetines, which require the sturdiness to strike through water and the grip to maintain a hold on struggling prey, so despite living near a lake this bird likely did not fish like these species. Aquila audax feeds on mammals ranging in size from rabbits to small wallabies, and is also a frequent scavenger of roadkill, while Haliaeetus leucogaster near exclusively preys on large fish and sea-snakes (Brown and Amadon 1968;Marchant and Higgins 1993). Hieraaetus morphnoides, a smaller bird that is closely related to Aquila, preys upon small rabbits and other mammals of a similar size, as well as small ground birds (Brown and Amadon 1968;Marchant and Higgins 1993). Archaehierax is larger than H. morphnoides, but its more gracile morphology may have restricted it to prey of a similar size to that preferred by this species.
The reduced size of the flange on trochlea metatarsi II in Archaehierax sylvestris, as well as its strongly plantar orientation, differs greatly from most Accipitridae. As this marks the point where the musculature for digit II connects to the tarsometatarsus, this could indicate a reduced ability to manoeuvre this digit in the plantar-medial direction, which is the orientation present in most accipitrids. However, the wider spacing of the trochleae could indicate a greater foot span when the toes are extended for prey capture, which might compensate for the loss of potential manoeuvrability.
The Pinpa Local Fauna contains a diverse array of animals (see above), some of which would have been potential prey for Archaehierax sylvestris. If a diet of small to medium birds and mammals ranging in habitat from arboreal, terrestrial and littoral is inferred, prey species may have included Wilaru tedfordi (a presbyornithid), Ngawupodius minya (a dwarf megapode), smaller individuals and juveniles of Madakoala devisi (an early koala), a huge diversity of possums and many of the waterbirds that appear in abundance in the Local Fauna.
We also thank Gerald Mayr, Nikita Zelenkov, and an anonymous reviewer for looking over the manuscript before publication and providing valuable feedback and corrections.

Authors contributions
EKM and THW designed the study. EKM collected all data, compiled the morphological matrix and performed all analyses. MSYL assisted with analysis of the molecular data and the combined analyses of morphological and molecular data. THW and MSYL contributed to data interpretation. EKM wrote the manuscript, and all authors edited the manuscript. THW and ABC collected the fossil material of Archaehierax sylvestris while on field work.