A giant raptor (Aves: Accipitridae) from the Pleistocene of southern Australia

The giant accipitrid Dynatoaetus gaffae gen. et sp. nov. is described from existing and newly collected material. Initial fossil remains were collected from Mairs Cave (Flinders Ranges, South Australia) in 1956 and 1969, and comprised a sternum, distal humerus and two ungual phalanges. A further 28 bones from this individual—including the neurocranium, vertebrae, furculum, and additional wing and leg bones, most of which were incomplete—were discovered at the site in 2021. This allowed identification of additional fossils from the same species in collections from Cooper Creek (Lake Eyre Basin, SA), Victoria Fossil Cave (Naracoorte, SA) and Wellington Caves (Wellington, NSW). Dynatoaetus has variable similarity across elements to those of living species in the Perninae, Gypaetinae, Circaetinae and Aegypiinae. Parsimony and Bayesian phylogenetic analyses of combined morphological and DNA data resolved it as the immediate sister-group to the Aegypiinae within the Circaetinae + Aegypiinae clade. The robust and eagle-like morphology of the lower hindlimbs suggest that the species was a predator, rather than a scavenger, and thus functionally similar to large circaetines such as the Philippine Eagle Pithecophaga jefferyi. Furthermore, this new species is the largest known bird of prey from Australia, much larger than the modern Wedge-Tailed Eagle Aquila audax. It is outsized in Australasia only by female Hieraaetus moorei (the extinct Haast’s Eagle from New Zealand). It is inferred to have been Australia’s top terrestrial avian predator during the Pleistocene, ranging from arid inland Australia to the more temperate coast, and likely became extinct around the time of the megafaunal mass extinction which peaked around 50 Ka. Its extinction in the late Pleistocene, along with the recently described scavenging vulture Cryptogyps lacertosus, marked a distinct decline in the diversity and function of Australia’s raptor guild.


Fossil record of Australian accipitrids
The earliest known Australian accipitrids (eagles, hawks, and Old World vultures) are the late Oligocene species Archaehierax sylvestris Mather et al. (2021) from the Namba Formation at Lake Pinpa in South Australia and Pengana robertbolesi Boles (1993) from Riversleigh in north-western Queensland. These species were both adapted for forest habitats, with A. sylvestris having a shortened wingspan for its size  and P. robertbolesi possessing highly flexible legs that would likely have been capable of reaching into tree hollows to pull out prey (Boles 1993).
The post-Oligocene fossil record for Accipitridae remains quite limited. One species is known from the late Miocene 11-9 Ma, Aquila bullockensis Gaff and Boles (2010), which was based on an isolated distal humerus and is currently the oldest species of Aquila known from Australia. An undescribed eagle species was recorded from the ~ 8 Ma Alcoota site (Worthy and Yates 2018).
The Plio-Pleistocene Australian fossil record currently has two accepted extinct species of Accipitridae: Cryptogyps lacertosus (de Vis 1905; see Mather et al. 2022) and Necrastur alacer de Vis (1892). Several other accipitrid species were described by de Vis in the late nineteenth and early twentieth century but reviews of the original fossil material have either synonymized them with living raptors or discovered them to be non-avian van Tets and Rich 1990;Boles 2006). Additional accipitrid fossils have been recognised from multiple cave sites across the continent, including several undescribed taxa (Gaff 2002;Mather 2021).

Mairs cave
Mairs Cave is located in Buckalowie Gorge in the southern Flinders Ranges, South Australia (Fig. 1). It is approximately 400 m long, with the entrance chamber being 120 m long by 10 m wide (Kraehenbuehl et al. 1997;Treble et al. 2017). The floor of this entrance chamber slopes down towards the east and is covered in large piles of boulders in many places, among which fossils can be found. The sections of the chamber farthest from the entrance lack stratified sediment, and therein most of the larger vertebrate fossils have been found scattered amid the rocks. This presents a severe limitation in trying to determine the age of fossils, as there is little evidence to help estimate when they were originally deposited (Liddle et al. 2018).
Several speleothems from Mairs Cave have been dated in various studies that provide some insight on the age of surface deposits. Gould-Whaley et al. (2021) dated subaqueous speleothem growth in a former pool in the entrance chamber to around 67 ka (kiloannum, thousand years) and 48 ka, and Treble et al. (2017) found evidence for wet phases during the last glacial maximum during the periods 23-18.9 ka and 18.9-15.8 ka in small surface speleothems. However, much older speleothem growth is known in the cave, and is the focus of ongoing research (J. Woodhead pers. comm. to THW).
A wide range of mammals, birds and reptiles are known from this site and a faunal list expanded as a result of collections made during this study, listed in SI.1. The extinct megafaunal component, e.g. the apex predator marsupial Thylacoleo carnifex, suggests the fauna in part dates to older than 40 ka (see Saltre et al. 2016;Hocknull et al. 2020).

Wellington Caves
Located in central-western New South Wales less than 10 km south of the town of Wellington (Fig. 1), the Wellington Caves are a source of fossiliferous deposits that are poorly dated but are suggested to span from the early Pliocene to the late Pleistocene (Dawson 1985(Dawson , 1995Fischer 1997;Dawson et al. 1999;Osborne 2001;Nipperess 2002;Bell 2004;Megirian et al. 2010). Multiple caves have produced fossils, including Cathedral Cave, Bone Cave, Phosphate Mine and Mitchell's Cave (Dawson 1985;Osborne 1991Osborne , 1997. Many fossils from this site, particularly those from Mitchell's Cave and Phosphate mine, have poor provenance data. These fossils, known as the "Old Collection", were collected from the caves from the middle of the nineteenth century to the early twentieth century when the caves were under the management of the NSW Department of Mines (Dawson 1985). Among this "Old Collection" are several accipitrid fossils, some of which have been referred to Cryptogyps lacertosus (see Mather et al. 2022), while others have yet to be described.

Victoria Fossil Cave
Part of the World Heritage listed Naracoorte Caves in southeastern South Australia, Victoria Fossil Cave is the source of multiple fossil deposits, most notably Fossil Chamber, Grant Hall and the Ossuaries (Reed 2003(Reed , 2006Fraser and Wells 2006). The fossils within this cave date from the last 500 ka of the Pleistocene (Grün et al. 2001;Prideaux et al. 2007;Macken et al. 2011. This cave is one of the most significant sites for Pleistocene vertebrate fossils in Australia and has provided a valuable record of mostly mammalian species, with some reptilian and avian elements also preserved. Fossil Chamber, the source of some of the material described in this study, has fossil material dating between 500 and 213 ka (Ayliffe et al. 1998;Arnold et al. 2022). A recent study has indicated a potential upper (older) age limit of 600 ka for vertebrate fossils in the Naracoorte Cave Complex based on pollen and charcoal records . Accipitrid fossils are known from the very large and speciesrich deposit in Fossil Chamber but are extremely rare, with only eight bones currently identified (see Mather 2021).

Cooper Creek
The Cooper Creek, which flows across south-western Queensland into north-eastern South Australia before ending at Kati Thanda-Lake Eyre, is part of the Lake Eyre Basin. Most fossil sites along the 'lower' Cooper Creek (within South Australia) are outcrops of the Quaternaryaged Kutjitara and Katipiri Formations (Tedford and Wells 1990). Both formations were formed in a fluvial environment (Tedford and Wells 1990). The Kutjitara Formation is possibly middle Pleistocene in age based on thermoluminescence and electron spin resonance dating (see Alley 1998 and sources therein). Deposition of the Katipiri Formation mainly occurred in two periods around ~ 270-220 ka and ~ 120-90 ka (Callen and Nanson 1992;Nanson et al. 1992;Maroulis et al. 2007), with the youngest fossil-bearing deposits possibly being ~ 75-65 ka (see Magee et al. 1995).
Fossil accipitrids from the lower Cooper include a large distal tibiotarsus from Waralamanko Waterhole, specimens of Elanus scriptus (see Gaff 2002), and the type specimen (a lower left humerus) of Aviceda gracilis (de Vis 1905) (now considered an unidentified species of living Accipiter, see van Tets and Rich 1990;Gaff 2002).

Mairs Cave accipitrid
The presence of a very large, undescribed accipitrid in the fossil deposits of Mairs Cave has been acknowledged in multiple publications Baird 1991;Baird et al. 1991). It was known from a nearly complete sternum, a distal humerus and two ungual phalanges that were collected in 1956 and 1969. This material was recovered roughly 55 m [60 yards] from the cave entrance, as per the collection data of ungual phalanx SAMA P.17139. Most specimens were covered in a thin layer of calcite, indicating that they had lain on the surface of the cave floor rather than being buried. The sternum and ungual phalanges were determined to represent an Old World vulture by Gaff (2002); at the time Aegypiinae and Gypaetinae were not separated into two clades. This material was also assessed by Mather (2021), who included the distal humerus and agreed with Gaff's conclusion, with the additional clarification that the species was a gypaetine vulture.
On the 4th of December 2021, a team of Flinders University recreational speleologists and palaeontologists including EKM, ABC and THW revisited Mairs Cave to search for more fossil material of this accipitrid. Using the recorded location data associated with SAMA P.17139, the site of the original find was located and a further 28 bones were collected. These included the neurocranium, fragments of the mandible, several vertebrae, the extremitas sternalis and extremitas omalis of both sides of the furcula, proximal humerus, distal ulna, carpometacarpus, tibiotarsus shaft, tarsometatarsus, and several pedal phalanges from digits one, two and four. The bones were located scattered amid rocks against the south-eastern cave wall; a small concentration of bones represented the decomposition site, but several specimens had fallen about 3 m down through the rocks to a cave floor where periodic water flow had redistributed some up to several metres along the floor and enabled calcite deposition on them. Incredibly, a fragment of a distal humerus was found that was able to be reattached to the distal humerus that was collected over 50 years prior in 1969. Most of the material was very well-preserved, with many bones almost completely intact. With this discovery, the Mairs Cave accipitrid is now the best represented Australian fossil accipitrid from the Pleistocene. Additional fossils found in the vicinity of the accipitrid skeleton (and likely contemporaneous with it) have also resulted in expansion of the fossil fauna known from Mairs Cave (SI.1).
The aim of this study was to describe the new accipitrid species represented by the Mairs Cave skeleton and to assess its relationships. Additionally, isolated fossils from other sites are allocated to the same taxon, thereby extending its geographic and temporal range.

Terminology
The terminology for the osteological features is derived from Baumel and Witmer (1993). 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
Measurements were conducted using digital callipers, with results rounded to the nearest 0.1 mm.

Phylogenetic methods
A total of 300 morphological characters were compared in the fossil and selected living species, sampling from the following elements: neurocranium, sternum, scapula, humerus, ulna, carpometacarpus, digital phalanges, pelvis, femur, tibiotarsus, tarsometatarsus and pedal phalanges. Over half of these characters (154) were sourced from Migotto (2013), 2 from Elzanowski and Stidham (2010), 1 from Elzanowski and Zelenkov (2015), 8 from Gaff and Boles (2010), 1 from Worthy et al. (2016) and 3 from Mayr (2014), with the remaining 127 characters from Mather et al. (2021). The list of characters and character states can be viewed in SI.2. For the coding of the data, inapplicable characters were identified using '-', though these are treated identically to missing data coded as '?'. Both the holotype and all referred fossil specimens were scored, with data available for viewing in SI.3.
Molecular data from Burleigh et al. (2015) were sourced for the following genes: cytochrome b, cytochrome oxidase 1, fibrinogen B beta introns 6-7, NADH dehydrogenase 2, RAG 1 and 12 s RNA. This was used to ensure that the relationships between the living species in this study would be resolved as accurately as possible. Sequences for the following species were sampled from the dataset of Burleigh et al. (2015) Mather et al. 2021). Due to a lack of genomic data for Threskiornis spinicollis and Elanus caeruleus, data from Platalea leucorodia and Elanus scriptus were used, respectively, as stand-ins, as they are closely related taxa (see Campbell and Lapointe 2009 regarding usage).
Phylogenetic comparisons were aimed primarily at determining the relationships of the fossil species with respect to living taxa. A total of 47 species of modern Accipitridae, 1 species each of Pandionidae, Sagittariidae, Cathartidae, Threskiornithidae and Ciconiidae were sampled. The nonaccipitrid species were selected for the following reasons: Pandionidae, Sagittariidae and Cathartidae are all members of the order Accipitriformes along with the Accipitridae; the Ciconiidae and Threskiornithidae are examples of bird families that fall outside of Accipitriformes but are similar in size and flight morphology, as well as having a history of grouping with the Cathartidae morphologically .
Parsimony and Bayesian phylogenetic analyses were used to analyse the data, and the main discussion focuses on combined morphology and molecular datasets, with morphology-only results in SI. The parsimony analyses of the morphological and combined morphological-molecular matrices were analysed using PAUP 4.0b10, using heuristic searches. Each search comprised 1000 random addition replicates, and enabled TBR branch swapping, with NCHUCK set to 1000 (see Mather et al 2021). The taxa Ciconia ciconia, Threskiornis spinicollis, Coragyps atratus and Sagittarius serpentarius were set as the most distal outgroup. Once the heuristic searches had generated a set of most parsimonious trees (MPT), a strict consensus tree was created from them. The clades formed by these trees were then assessed using bootstrapping (200 replicates), and the majority-rule consensus was set to a conlevel of 50 (only clades > 50% shown). Parsimony analyses were performed with Dynatoaetus alone added to the matrix with living taxa, and with both Dynatoaetus and Cryptogyps added.
For the Bayesian analyses, the software MrBayes 3.2.7 was used via the platform CIPRES (www. phylo. org). The analyses employed three partitions: the morphological data, and two DNA partitions, which were obtained from Parti-tionFinder (Lanfear et al. 2017) using the settings in Mather et al. (2021). These partitions were termed pfinder_molec1, which included Cyt-B codons 1 and 2, CO1 codons 1 and 2, ND2 codons 1 and 2, 12 s, Rag-1 codons 1, 2 and 3, and FGBint67; and pfinder_molec2, which included Cyt-B codon 3, CO1 codon 3, and ND2 codon 3. The Morph partition for discrete numerical states had among-character rate variability set to gamma, with distribution approximated using four categories. The Molec1 partition had the GTR model, with a Nst of 6, with substitution rates and frequencies having (separate) Dirichlet priors. The rates were set to InvGamma, with the gamma distribution approximated using four categories. The Molec2 partition also had a (separate) GTR model. The rates were set to Gamma, with the distribution approximated based on four categories. The number of MCMC chains was set to 4 (incrementally heated at 0.1), the number of generations set to 50,000,000, the sample frequency set to 5000. Each analysis was run four times (in parallel), and after suitable burnin was determined, a consensus tree was then derived from all four runs. Ciconia ciconia was initially set as the sole outgroup taxon, due to limitations with MrBayes, but trees were later re-rooted on Ciconia + Threskiornis. The Bayesian analyses included both Dynatoaetus and Cryptogyps.
The files containing the data matrix in PAUP and MrBayes format can be found in SI.4 and SI.5, respectively.

Systematic palaeontology
Accipitriformes Vieillot 1816 Accipitridae Vigors 1824 Dynatoaetus gen. nov. Z o o b a n k I D : u r n : l s i d : z o o b a n k . org:pub:236A128E-4FFD-4B0D-BEC1-3F6DF133E390 Diagnosis: A large accipitrid distinguishable from other genera by the following combination of characters: a neurocranium that is (1) relatively short and wide compared to Aquila audax; (2) has the line linking the tip of the processus zygomaticus to the ventral tip of the processus paroccipitalis aligned about 45 degrees to the plane of basioccipital-parasphenoidal plane and ventrally encloses a tympanic recess that is longer than high; (3) the condylus occipitalis is relatively large; (4) the mamillar tuberosities (tubercula basilaria) are robust, and prominent caudally; (5) the foramen magnum is near-perpendicular to the basioccipital plane (~ 10 degrees off perpendicular); (6) has prominent tubercula for the insertion for m. pseudotemporalis superficialis on the facies orbitalis. A sternum with (7) distinct and prominently projecting processus labrum internum; (8) a spina externa that is narrower than the apex carinae; (9) a crista medialis carinae that does not extend to the base of the spina externa. A humerus with (10) a weakly ventrally projecting epicondylus ventralis; (11) the distance between the interior margin of the tuberculum supracondylare ventrale and the proximal tip of the condylus dorsalis is equal to that between the tip of the condylus dorsalis and the dorsal margin; and (12) a deep insertion pit for the distal head of the m. pronator profundus. An ulna with (13) a tuberculum carpale that has little prominence cranioventrally. A tarsometatarsus that (14) is robust; (15) has a trochlea metatarsi II that is broad and more distally elongate than trochlea metatarsi III; (16) has the trochleae metatarsorum II and IV with robust plantar flanges and with (17) the plantar openings of the foramen vasculare proximale medialis positioned on the lateral side of the crista medialis hypotarsi; and (18) has the medial side of the shaft deeply excavated caudally by the fossa parahypotarsalis medialis such that the medial margin is very thin.

Etymology:
The genus name is a combination of the ancient Greek words δῠνᾰτός (dynatós), meaning strong, mighty, or powerful, and ᾱ̓ ετός (āetós) meaning eagle.
Type Type locality and stratigraphy: Mairs Cave, Buckalowie Gorge, Flinders Ranges, SA, Australia, 32° 10′ 30 S, 138° 52′ 23 E (see Fig. 1). The bones of the Mairs Cave individual were 55 m from the entrance against the east wall of the main passage and were scattered between rocks over 3 m vertically and along the passage floor below the rockpile. It, therefore, was essentially a surface deposit and bones on the floor were associated with extinct Pleistocene mammal species.

Etymology:
The name 'gaffae' honours Priscilla Gaff, who first discussed the fossil material of this species in her 2002 thesis revising Australian accipitrids.

Comparisons with Australasian Pleistocene taxa
Dynatoaetus gaffae can be separated from the aegypiine Cryptogyps lacertosus by the following features: it is significantly larger than C. lacertosus; the distance between the proximal tip of the condylus dorsalis and the margin of the tuberculum supracondylare ventrale on the humerus is equal to the distance between the condylus dorsalis and the lateral margin of the face between the tuberculum supracondylare dorsale and the epicondylus dorsalis (greater in C. lacertosus); the tarsometatarsus is moderately elongate (stout in C. lacertosus), the sulcus hypotarsi is broad (narrow), the base of the cristae hypotarsi are connected to a flat ventral surface (connected to a raised ridge), the foramen vasculare proximale medialis is located lateral to the crista medialis hypotarsi (located medial to the crista), the impressiones retinaculi extensorii form two prominent ridges with a deepened depression between them (ridges absent, no deepening), and the medial shaft is very thin dorsoplantarly (thick).
Dynatoaetus gaffae is only matched in size in the Australasian region by the extinct New Zealand eagle Hieraaetus moorei, but can be distinguished from the latter via the following features (Hieraaetus state in brackets): the sternum has prominent processus labrum internum present (absent/ very short in H. moorei); in the humerus, the incisura capitis is shallow (deep in H. moorei), the palmar insertion for the m. extensor metacarpi radialis forms a distinct line (circle-shaped in H. moorei), the ventral margin of the fossa forms a curved line parallel to the shaft margin (straight in H. moorei), the distance between the ventral margin of the fossa brachialis and the ventral shaft margin is between 1/3 and 1/4 shaft width (less than 1/4 shaft width in H. moorei), the distance between the tip of the condylus dorsalis and the margin of the tuberculum supracondylare ventrale is equal to the distance between the condylus dorsalis and the lateral margin of the face between the tuberculum supracondylare dorsale and the epicondylus dorsalis (greater width in H. moorei), the interior margin of the tuberculum supracondylare ventrale is oriented at least 45-50° across the shaft (oriented parallel to the shaft in H. moorei); in the ulna, the tuberculum carpale is weakly projecting cranially (moderate projection into a distinct peak); in the tarsometatarsus, the foramen vasculare proximale medialis is located lateral to the crista medialis hypotarsi (located medial to the crista) and is positioned adjacent to the base of the crista (positioned proximal to the base of the crista), the tuberositas m. tibialis cranialis is directly distal to the foramina vascularia proximalia on the dorsal facies (well separated distally), the impressiones retinaculi extensorii form two prominent ridges with a deepened depression between them (ridges absent, no deepening), the impressio lig. collateralis lateralis forms a prominent inflation on the lateral facies (flattened scar), and the second and third trochleae have equal distal extent (third trochlea longer distally). Thus, Dynatoaetus gaffae differs substantially from H. moorei.

Neurocranium (Fig. 3A-E)
The neurocranium is covered in a thin layer of calcite. It is preserved anteriorly from the craniofacial hinge as revealed by the preserved sulcus for the right side of processus frontalis of the premaxilla. Anteriorly, breakage results in the lacrimal articular facets not being preserved; similar both processus postorbitales (Fig. 3, PrPO) are broken. Dorsocaudally, a combination of corrosion and abrasion has removed the prominentia cerebellaris and areas lateral of it, although the foramen magnum is entire. The right squamosal area is damaged such that squamosal width is compromised. The following features are observable: There is an interorbital sulcus (Fig. 3, SI) extending caudally from the craniofacial hinge to the mid-length of the orbits. The lacrimal articular zone, while the facet is not preserved, must be relatively short caudal to the sulcus for the processus frontalis, and not more than 12 mm of the orbit length of ~ 35 mm, suggesting that it was relatively small. Relative to the plane of the basioccipital-parasphenoid, the cranial cavity is tilted about 40 degrees (Aquila is about 60 degrees) and related to this is the steep plane of the foramen 1 3 magnum (Fig. 3, FM) (~ 80 degrees to basioccipital plane) and the shallow angle of the line joining in the processus zygomaticus (Fig. 3, PrZ) and the ventral tip of the processus paroccipitalis (Fig. 3, PrPa). This results in the cavum tympanicum being rather longer than it is high. A raised facet within the tympanic cavity that buttresses the caudal side of the otic capitulum of the quadrate (when it is articulated) appears relatively low and small compared to that of A. audax, and similar to the condition observed in Aegypius and Gyps. However, this may be exaggerated by breakage.

Mandible (
The processus medialis (Fig. 3, PM) is short but prominently protruding medially. The articular area across the lateral and medial cotylae is robust, ~ 19.5 mm wide, forming an articular surface wider than long (most Accipitridae except Aegypiinae).

Vertebrae (Fig. 4)
Seven vertebrae are preserved, which include three thoracic vertebrae ( Fig. 4A-C), two caudal vertebrae (Fig. 4D, G), and two cervical vertebrae (Fig. 4F, G). Due to their limited number, their precise position along the spine is difficult to determine.

Furcula (Fig. 5F)
The left and right scapus claviculae and the fused extremitas sternalis claviculae of the furcula are preserved in three pieces. The following features can be observed: The scapus claviculae are broad (29 mm wide) with robust facies articularis acrocoracoidea (Fig. 5, FAA). The extremitas omalis claviculae (Fig. 5, EOC) have large, circular impressions which are pneumatic (some Aegypiinae, most Gypaetinae, Haliaeetinae). The distance between the processus acromialis and the ventral facies of the articularis acrocoracoidea is relatively short (31 mm), which equals width at the base of the processus (Gypaetinae, Aegypiinae, Haliaeetinae). The apophysis furculae (Fig. 5, AF) is worn but its gracile nature suggests it had very little ventral projection (Perninae, Gypaetinae, Haliaeetinae). The furcula forms a broad U-shape in dorsal aspect (Perninae, Gypaetinae, Haliaeetinae). There is little to no deepening of the cranial facies directly dorsal to the apophysis furculae (Gypaetinae, Aegypiinae).
Distally, the fossa brachialis (Fig. 6, FB) has a deepened, circular ventral scar, with a shallow section extending proximodorsally from it (Perninae, Gypaetinae, Aegypiinae, Haliaeetinae). The fossa brachialis is not pneumatic (all Accipitridae). The dorsal margin of the fossa brachialis is broadly separated from the shaft margin (9 mm) by a distance less than a quarter of the fossa width (some Perninae, some Gypaetinae, Aegypiinae). The ventral margin of the fossa is curved and is not parallel to the shaft margin (Gypaetinae, Circaetinae). The palmar insertion scar for the m. extensor metacarpi radialis (Fig. 6, MEMR) is a broad (9 mm wide dorsoventrally), oval shape and separated from the condylus dorsalis by ~ 6 mm. The dorsal insertion scar for the m. extensor metacarpi radialis is small, shallow and circular/round (Perninae, Gypaetinae, Circaetinae, Buteoninae) and is only slightly prominent dorsally of the shaft proximal to it. Thus, the tuberculum supracondylare dorsale (Fig. 6, TSD) is low (few Perninae, some Gypaetinae, some Circaetinae, some Aegypiinae, Harpiinae, few Aquilinae). The dorsal profile expands dorsally more distally, such that the epicondylus dorsalis (Fig. 6, ED), even in its incompletely preserved state, was more prominent dorsally than the tuberculum supracondylare dorsale (some Gypaetinae, Aegypiinae). The interior margin of the tuberculum supracondylare ventrale (Fig. 6, TSV) is aligned at a 45-50° proximodorsally across the shaft, and the proximal margin of the tuberculum prominently projects cranially (most Perninae, Gypaetinae, Circaetinae, Aquilinae, Haliaeetinae, Buteoninae). The distance between the interior margin of the tuberculum supracondylare ventrale and the proximal tip of the condylus dorsalis (Fig. 6, CD) is roughly equal to the distance between the tip of the condylus dorsalis and the dorsal margin (some Perninae, Gypaetinae). The proximal insertion for the m. pronator superficialis (Fig. 6, MPS) is large, circular, deep, and close to the tuberculum supracondylare ventral (some Perninae, Circaetinae, Aquilinae, Haliaeetinae). The distal insertion for the m. pronator superficialis is deep, distinct, and located adjacent to the ventrodistal end of the tuberculum. The epicondylus ventralis (Fig. 6, EV) has very little ventral prominence relative to the shaft, such that CD condylus dorsalis, FADA facies articularis digiti alulae, OMMi os metacarpale minus, PE processus extensorius, PP processus pisiformis, SI sulcus intercondylaris, ST sulcus tendinosus, TC tuberculum carpale. Scale bars: A-C is 50 mm, E-G is 10 mm the shaft profile is straight level with the tuberculum supracondylare ventrale, rather than concave (Perninae, some Gypaetinae). On the epicondyle, the insertion for the m. pronator profundus (Fig. 6, MPP) is deeper than the insertion for the m. extensor metacarpi ulnaris (Fig. 6, MEMU) (most Perninae, Gypaetinae, Circaetinae, Aegypiinae, Aquilinae). Ulna (Fig. 7C, D) The distal end of the left ulna is well-preserved but is covered in a thin layer of calcite. The following features are notable: The tuberculum carpale (Fig. 7, TC) barely projects cranioventrally. The sulcus intercondylaris (Fig. 7, SI) is deep and V-shaped (most Accipitridae). The proximal margin of the condylus dorsalis (Fig. 7, CD) does not project ventrally and connects to the shaft in a continuous line in caudal view (some Gypaetinae, Aegypiinae). The length of the condylus dorsalis (24.7 mm, 1-2 mm missing) is slightly greater than its craniocaudal depth (22.5) in dorsal aspect (some Perninae, Aegypiinae, Circaetinae, Aquilinae). Fig. 7A, B) The entire left carpometacarpus is preserved except for the tip of the articular process for phalanx digiti minoris, with the ventral surface coated with a thin veneer of calcite. The following features can be observed:
Phalanx proximalis digiti majoris (Fig. 7F) The length between articular surfaces is 55.7 mm; the caudal margin is broken so its width is indeterminate. Femur (Fig. 8A-D) The femur SAM P.41514 is referred to this species based on its appropriate and distinctive large size and co-occurrence with SAMA P.28008, a partial tarsometatarsus identical to the holotype tarsometatarsus specimen.
The femur is extremely large and robust compared to all living Australasian accipitrids (see measurements), with only specimens of the extinct Haast's eagle Hieraaetus moorei outsizing it. The fovea lig. capitis (Fig. 8, FLC) is deep (Aegypiinae, Circaetinae, Aquilinae, Haliaeetinae, Buteoninae) and large relative to the caput (Perninae, Gypaetinae, Aegypiinae, Circaetinae, Aquilinae, Haliaeetinae, Buteoninae). The fossa trochanteris is very shallow (Perninae, most Gypaetinae, Circaetinae, Aquilinae, Haliaeetinae, Buteoninae). The crista trochanteris (Fig. 8, CT) has a little breakage affecting the proximocranial part, but is low (most Perninae, Aquilinae, Haliaeetinae, Buteoninae), the cranial surface medial to it is flat rather than concave (Perninae, some Gypaetinae, Circaetinae, Aquilinae, Haliaeetinae), and it has one relatively small pneumatic foramen (Fig. 8, F) penetrating it medially in its distal third of length (some Gypaetinae, some Aegypiinae, Circaetinae, Buteoninae). The depression distad to the facies articularis antitrochanterica on the caudal face is very shallow (Perninae, most Gypaetinae, Aegypiinae, Circaetinae, Aquilinae, Haliaeetinae). The linea intermuscularis cranialis (Fig. 8, LIC) is positioned laterally (most Gypaetinae, Aegypiinae, Circaetinae). Laterally on the proximal end, the insertions of various muscles are well-marked. Using the terminology from Matsuoka and Seoka (2021), they are as follows: The scar for the insertion of m. obturatorius medialis (Fig. 8, MOM) occupies a 9 mm wide bulge level with the collum and in the caudal half of the lateral facies, with radial scarring present across a continuous area of 16.2 mm. The insertion of the m. obturatorius lateralis (Fig. 8, MOL) forms a deep oval scar distal to the m. obturatorius medialis (similar to, but much deeper and more distinct than, Aquilinae and Haliaeetinae). More distally a larger and deep scar centred on the shaft is for the insertion of the m. ischiofemoralis extensorius (Fig. 8, MIE) (most Accipitridae). The linear insertion scar of m. iliotrochantericus caudalis (Fig. 8, MICa) is about 18 mm long and runs parallel to and about 6 mm caudal to the crista trochanteris (similar to Perninae, Gypaetinae, Aegypiinae); it is well separated from the margin of the crista trochanteris (Perninae, Gypaetinae,
These differ from sympatric Aquila audax in details as follows (Aquila in brackets): that for m. obturatorius medialis is elevated (in depression) and forms one large,  continuous scar (split into several smaller attachment points); that for m. iliotrochantericus caudalis is longer ending distally level with m. iliotrochantericus caudalis (ends distinctly more proximally); m. iliotrochanterici medius is rounded (distinctly elongate and extends more proximally); m. ischiofemoralis is well separated from m. iliotrochanterici medius (close together).
Pedal phalanges (Fig. 9) The right metatarsal 1, LI.1, RII.1. RII.2, RVI.1, and 3 ungual phalanges are preserved. The following features can be observed: The metatarsal (Fig. 9, OM) is heavily eroded around its margins, making some features difficult to observe. The lateral margin leading to the trochlea metatarsi I (Fig. 9, TMI) has a largely straight profile, only curving outward laterally directly proximal to the trochlea. The main feature is that all phalanges are relatively short and very robust, such that lengths do not differ much from those of a female Aquila audax and they are very much stouter.
The phalanx I.1 (Fig. 9, PI.1) is missing the distal trochlea beyond its plantar base, nevertheless it was likely not longer than that of Aquila audax, having a very much shorter midsection. The medial plantar tuberculum (Fig. 8, MPT) is very large and robust.
The phalanx II.1 (Fig. 9, PII.1) is not fused with II.2 (all accipitrids excluding Haliaeetinae). Concomitant with its robust nature, the trochlea articularis of II.1 is broad and deep for reception of the tuberculum extensorium of II.2.
The phalanx II.2 (Fig. 9, PII.2) is robust and relatively short, being roughly twice the length of the II.1, and it tightly interlocks with it precluding rotation at the joint. The corpus phalangis is swollen both medially and laterally just proximal to the foveae ligamenti collaterales, much more so than in Aquila audax.
The phalanx IV.1 (Fig. 9, PIV.1) has a moderate amount of lateral expansion. The foveae ligamenti collaterales distally are shallow and poorly defined. In side view, the dorsal surface is deeply concave because of a much enlarged dorsal margin to the proximal articular surface. In plantar view, a robust tuberculum flexorium forms a ridge extending plantarly along the length of the lateral side of the digit. Medially, a robust and short tuberculum flexorium is proximally expanded.
Ungual phalanges (Fig. 9) The three ungual phalanges are each nearly perfectly preserved, missing only the very tip of the distal end. Due to a lack of associated pedal phalanges, it is uncertain to which digits these unguals belong, but ungual phalanges SAMA P.59525 and SAMA P.19157 seem to be a pair based on similar morphology. All specimens are notably larger than ungual phalanges one and two in the observed specimens of Aquila audax. Additionally, the following features are observed: The height of the articular facet is greater than its width. In P.17139, the width of the tuberculum flexorium is less than its height thanks to deepened depressions distal of the foramina, and the foramina on the sides of the tuberculum flexorium are distinct and deep. The width of the tuberculum flexorium in P.19157 is roughly equal to its height.

Phylogenetic analyses
Parsimony analysis of morphological and molecular data produced three most parsimonious trees (MPTs) of length of 1835, the strict consensus of which is shown in SI.8.
The clade comprising D. gaffae and Aegypiinae was supported by nine character states that optimised unambiguously on that branch (character number in brackets): the alignment of the lateral edges of the processus paroccipitalis of the neurocranium being roughly parallel (29: 1 → 0); the foramen magnum of the neurocranium being positioned on the caudoventral plane (30: 2 → 1); the crista medialis carinae does not abut with the spina externa on the sternum (71: 0 → 1); the apex carinae is positioned caudal to the sulcus sellaris medialis in the dorsoventral axis on the sternum (75: 1 → 0); the depth of the carina is less than the depth of the sternum basin (89: 1 → 2); the fossa m. brachialis of the numerus is narrowly separated from the dorsal facies of the shaft (151: 1 → 2); the carpometacarpus lacks a deepened sulcus separating the processus pisiformis and processus extensorius (180: 1 → 0); the ridge bordering the distal end of the sulcus tendinous on the carpometacarpus has a low profile (185: 1 → 0); the medial shaft margin adjacent to the mid-length fossa parahypotarsalis medialis is dorsoplantarly compressed but slightly thickened (263: 2 → 1). Only one of these optimisation-unambiguous characters, 71, was also unique and unreversed (CI = 1.00). Of these character states, only those for 29, 71, 75, 89, 151, 185 and 263 are unique or mostly unique to the Aegypiinae in accipitrids (29, 75, 89, and 151 are seen in some Gypaetinae, 263 seen in a small number of Gypaetinae and Perninae).
The clade comprising D. gaffae, Aegypiinae and Circaetinae was diagnosed by three unambiguous characters: the fossa sternocoracoidei on the sternum extending to the fourth processus costalis (80: 3 → 2); the os metacarpale majus on the carpometacarpus has weak ventral projection (186: 1 → 0); the first phalanx of pedal digit I is equal to slightly more robust than most other phalanges (285: 2 → 1).
When Cryptogyps lacertosus was added to the phylogeny (Fig. 10), neither the position of D. gaffae nor support for the fossil-Aegypiinae clade changed notably (51.2%). The characters supporting this clade were unchanged from the previous analysis. The clade comprising C. lacertosus and Aegypiinae had good support (78.5%), but support for Aegypiinae-Circaetinae (50%) and the Circaetinae clade (44.4%) was low. The divergence of the fossils-Aegypiinae-Circaetinae clade dropped (65.2%), as did the divergence of the Harpiinae-Aquilinae from all other Accipitridae (58.8%) and the Harpiinae clade to a very small degree (64.2%). A morphology-only parsimony analysis tree using the same taxa is presented in SI.8. In this tree, Dynatoaetus resolves as the sister taxon of a group comprising both Aegypiinae (including Cryptogyps) and Gypaetinae, though DNA studies strongly suggest that the morphological similarities between Aegypiinae and Gypaetinae are due to convergence. Similarly, incorrect associations were seen in some other morphologically similar taxa; the circaetine Pithecophaga jefferyi and the harpiine Harpia harpyja were sister to each other and nested within the Aquilinae, while Chondrohierax uncinatus and Hamirostra melanosternon were separated from other Perninae. As the aim of the phylogenetic analyses was to place the fossils within the most widely accepted phylogeny of living taxa, this justifies our use of molecular data to effectively constrain the relationships of living taxa.
The Bayesian analysis (Fig. 11) of the data matrix including both Dynatoaetus and Cryptogyps produced trees mostly concordant with the parsimony analysis. Dynatoaetus gaffae was positioned basal to all the Aegypiinae, but with low support (pp [posterior probability] = 0.49). Cryptogyps lacertosus was separated from other Aegypiinae into its own branch, but with very weak support (0.35). Support for the Circaetinae-Aegypiinae clade (including D. gaffae), and the grouping of this clade with all other Accipitridae, was 0.52. Both fossil species were strongly excluded from the Aquilinae-Harpiinae-Accipitrinae-Haliaeetinae-Buteoninae clade, which had a pp of 1.0.

Discussion
Dynatoaetus gaffae is a new genus and species of large accipitrid raptor from the Australian Pleistocene, based on relatively complete material comprising 32 bones of a partial skeleton and 4 other referred fossil bones from three other sites. It is up to twice the mass of the largest living Australian eagle, the Wedge-Tailed Eagle Aquila audax, and of all other fossil eagles known from the continent.

Phylogenetic affinities
Dynatoaetus gaffae exhibits a mixture of features found across species of Perninae, Gypaetinae, Circaetinae and Aegypiinae. These are likely a combination of phylogenetic and functional features, making assessment of the precise taxonomic position of the fossil difficult using only comparative morphology. All phylogenetic analyses (morphology only, parsimony and bayesian analyses of mixed molecular and morphology data) found D. gaffae to resolve within the clade of Circaetinae + Aegypiinae as sister to Aegypiinae. The Circaetinae include a large raptorial species (Pithecophaga jefferyi) in the Philippines.
Dynatoaetus gaffae can be distinguished from species of Circaetinae, the sister clade to Aegypiinae (see Mindell et al. 2018), found geographically close to Australia (notably Spilornis cheela and Pithecophaga jefferyi) by the following features (circaetine state in brackets): the processus postorbitalis ends dorsal to the processus zygomaticus (extends level), the foramen magnum is positioned caudoventrally on the caudal end of the skull (completely ventral), prominent tubercula are present on the caudal section of the area muscularis aspera (absent); on the sternum, the processus Fig. 11 Result of Bayesian analysis using combined molecular and morphological data (ordered), with the fossil species Cryptogyps lacertosus included. Molecular and morphological partition branch lengths were unlinked. Node values show posterior probability expressed as percentages labra internum prominently project cranially (absent in Pithecophaga, small projection in all others), the crista medialis terminates before the base of the spina externa (extends to spina externa), the apex carinae tip is positioned caudal to the sulci articulares coracoidei (adjacent to sulci), there are thought to have been six processus costales (seven in Pithecophaga + Spilornis); in the scapula, the tuberculum coracoideum barely projects cranially (prominent projection), a small foramen is present in the ventral facies of the cranial end (absent), the facies articularis humeralis is continuous with the ventral shaft margin (facies projects out from shaft); in the carpometacarpus, the facies alularis digiti alulae is continuous with the shaft (separated from shaft by a notch), a shallow sulcus separating the processus pisiformis and processus extensorius (deep sulcus); in the ulna, the tuberculum carpale is flattened cranially (moderately projecting); in the femur, the crista supracondylaris medialis is prominent (flat), and the impressio lig. collateralis lateralis spans roughly two-thirds the caudo-cranial depth of the condylus lateralis (a third of craniocaudal depth or less); in the tibiotarsus, the condylus medialis does not expand medially past the shaft margin (extends further medially); in the tarsometatarsus, the foramen vasculare proximalis medialis is positioned lateral to the crista medioplantaris (positioned medial to crista), there is a deep depression between the impressiones retinaculi extensorii (no depression).
Dynatoaetus gaffae differs from living species of Aegypiinae by the following features (aegypiine state in brackets): a lack of pneumatism in the distal wing bones and sternum. In the neurocranium, the lamina parasphenoidalis is unfused (fused), and there are prominent tubercles present on the area muscularis aspera (absent). In the sternum, the sulci articularis coracoidei overlap (do not overlap); the processus labrum internum are present and quite prominent (absent or greatly reduced). In the scapula, there is no foramen/pneumatisation present in the cranial region of the acromion (present); a foramen is present in the medial facies of the cranial end (absent); the facies articularis humeralis is continuous with the shaft margin (ends with a ventrally offset notch); the depth of the corpus scapulae increases caudally (remains largely the same). In the tarsometatarsus, the merged sulcus flexorius and fossa parahypotarsalis medialis forms a deepened sulcus (shallow); prominent attachment scars for the impressions retinaculi extensorii (greatly reduced/absent); the fossa infracotylaris dorsalis is deep (shallow); the flange on the trochlea metatarsi II prominently projects medioplantarly (little projection); the flange on trochlea metatarsi II prominently projects plantarly (little projection); the fovea ligamentum collateralis is deep (shallow). Furthermore, Dynatoaetus gaffae has traits quite unlike the living crown aegypiines, such as large raptorial ungual phalanges and a tarsometatarsus with typically eagle-like proportions and features.
The fossil differs from any living species of Gypaetinae (which associated with aegypiines in the morphology-only parsimony analysis) by the following features (gypaetine state in brackets): in the sternum, the spina externa is narrower than the apex carinae in cranial view (thicker or equal thickness), the crista medialis carinae terminating proximal to the sulci coracoidei (terminating adjacent); in the humerus, the proximal and distal heads of the m. pronator profundus are large and deep (small and shallow). In the tarsometatarsus, the foramina vascularia proximalia is positioned lateral to the crista medialis hypotarsi (positioned medial). The fossil is much larger than the small gypaetine vultures Neophron percnopterus and Gypohierax angolensis, and is also notably larger than Gypaetus barbatus, a species which is slightly larger than Aquila audax.
These differences reveal the highly distinct nature of Dynatoaetus gaffae and warrants the establishment of a new genus for the taxon.

Ecology
The known distribution of Dynatoaetus gaffae is currently limited to the southern parts of the continent (see Fig. 1) from the Cooper Creek of the Lake Eyre Basin, Mairs Cave, Wellington Caves and Naracoorte Caves region, in central Australia, mid-north South Australia, central eastern NSW and southeast South Australia, respectively. However, this may reflect abundance of fossil sites in these regions and the expected rarity of a top predator. A broader distribution over the continent would not be unusual, given the range of many modern Australian raptors (see Simpson and Day 2010).
Despite its much larger size, the pedal phalanges of Dynatoaetus gaffae do not seem to have been much longer than that of a large Aquila audax, suggesting similar foot span in both species. However, the phalanges are comparatively much more robust in D. gaffae, which could indicate a more powerful grip. This kind of morphology can be observed in accipitrid species such as Stephanoaetus coronatus, which is known to prey upon monkeys and small antelope (Sanders et al. 2003); Harpia harpyja, which primarily hunts sloths and monkeys but will also rarely take terrestrial animals like the collared peccary Tayassu tajacu (see Miranda 2018); and Hieraaetus moorei, which is known to have preyed upon large species of Dinornithiformes (Brathwaite 1992;Worthy and Holdaway 2002). This indicates that D. gaffae could also have been adapted for hunting relatively large prey, using its strength to maintain its grip on the struggling animal. The late Pleistocene Australian fauna included species like the sthenurine kangaroos, larger macropodine kangaroos, large snakes, giant megapodes, the flightless Genyornis, and the diprotodontids (Wroe et al. 2013), the young and sickly of which D. gaffae may have preyed upon.
While D. gaffae has a predatory morphology, it is likely that it also scavenged frequently upon the carcasses of megafaunal marsupials, as do many modern predatory eagles (see Kane et al. 2014;Margalida et al. 2017). It would have faced competition from specialist avian scavengers such as Cryptogyps lacertosus but the size of D. gaffae would have given it an advantage against this smaller species. This is observed in modern avian scavenging guilds, where the largest, most belligerent birds tend to maintain dominance around carcasses (Moreno-Opo et al. 2020). No taphonomic traces on fossil megafaunal material have been referred to a large raptor to date but this likely reflects bias in consideration of possible predators and scavengers.
ecosystems. The prey of A. audax typically ranges in size from rabbits to small kangaroos, and it frequently scavenges from carcasses (Brooker and Ridpath 1980;Olsen et al. 2010). It is unlikely that this generalist habit would have been possible when accipitrid species such as the giant predatory D. gaffae and the vulturine Cryptogyps lacertosus were still alive to provide competition. In modern ecosystems where multiple species of raptors are present, the partitioning of resources such as food and nesting sites allows such coexistence to occur (Krüger 2002;Giovanni et al. 2007;Olsen et al. 2010;Treinys et al. 2011;Whitfield et al. 2013). It is, therefore, possible that until the late Pleistocene, Aquila audax was limited to hunting a smaller range of prey, scavenged less frequently, and may have been more restricted in its range due to interspecific competition. The extinction of the Australian megafauna would have resulted in the subsequent extinction of scavengers like C. lacertosus and large predators like D. gaffae (see Galetti et al. 2018), leaving the surviving generalist A. audax free to exploit their former niches.