Discovery and excavation

In March 2019, farmhand Michael Johnston alerted Tūhura Otago Museum to a trackway made up of tridactyl footprints exposed in the Kyeburn River upstream of the State Highway 85 bridge (−45.148442 ⁰S, 170.262257 ⁰E; WGS84) on the Maniototo plains of Central Otago (). The imprints were preserved on a bedding surface of indurated silty clay on the riverbed and were likely exposed by a large flow event in November 2018. There were five visible prints in the trackway exposed in a line parallel to the riverbank (see and ) and another two that were partially obscured by the overhanging bank. A decision was made to extract the prints given the ephemeral nature of the site and the wish to preserve them for future generations. The footprints were under 1.3 m of water, so the entire river required diversion around the site to extract them. On 10 May 2019, after the appropriate resource consents and permissions were granted, a team consisting of members from: Tūhura Otago Museum, the University of Otago’s Department of Geology, Kāti Huirapa Rūnaka ki Puketeraki and Rūnaka o Ōtākou, excavated six of the footprints. Extraction was conducted using a masonry saw and chainsaw, with each footprint being removed individually.

Figure 1. Map showing the location of the moa footprint trackway site and excavation site in relation to the Kyeburn Bridge that crosses the Kyeburn River on State Highway 85 in Central Otago, New Zealand. Temporary river diversion was ∼20 m to the west of the trackway site through already disturbed river gravels. Localities and field numbers of sediment samples extracted for cosmogenic nuclide dating, and extent of Maniototo Conglomerate outcrops are indicated.

Figure 2. Orthographic image of the moa trackway found in the Kyeburn River immediately prior to extraction (OMNZ AV11896). Visible are five of the six footprints that were recovered as well as the very faint footprint discovered during the review of the orthographic imagery after the excavation and recovery. The larger shallow footprint can be seen between the first two footprints to the left. R3 and L4 are obscured from sight by the bank.

Registration and specimen information

The six footprints are lodged in the Tūhura Otago Museum, Dunedin (OMNZ AV11896). Locality and collection information is available on the New Zealand Fossil Record File Database (www.fred.org.nz) under fossil record number I42/f34. Subsequent inspection of the site survey imagery collected during the excavation captured the details of a larger moa footprint crossing the trackway (see between prints L1 and R1, footprints are labelled as per . Larger photographs of the individual footprints can be seen in and ).

Figure 3. Side by side comparisons of the OMNZ AV11896 dried footprint blocks after the extraction from the Kyeburn River, Central Otago, New Zealand. Not to scale.

Figure 4. Close up of the associated larger footprint found in the digital imagery after the Kyeburn footprint excavation from Kyeburn, Central Otago, New Zealand.

Geological setting

The moa trackway was found exposed on a bedding plane of buff-coloured indurated silty clay. The mudstone lithofacies can be traced upstream ∼200 m to where 30 m high cliffs on the true left side of the river valley meet the river’s bank. Here the riverbank outcrops are conformable with the Maniototo Conglomerate that comprises the cliffs (see Bishop, 1979). The Maniototo Conglomerate Formation (Youngson et al. 1998) is a series of indurated clays, sands and gravels dominated by greywacke clasts (Bishop 1974, pages 32–34; Bishop 1979, pages 13–16). These sediments represent a fluvial plain deposited by braided river systems that drained the newly uprising greywacke mountains of the Kakanui, Ida and Hawkdun ranges. These ranges were uplifting along the Waihemo-Hawkdun Fault Zone, which bounds the northern and north-eastern margins of the Maniototo Basin (Youngson et al. 1998; Forsyth 2001; Craw et al. 2016). Imbrication of Maniototo Conglomerate clasts at nearby localities indicates flow from the northeast indicating an emerging Kakanui range as the source (Bishop 1974). Youngson et al. (1998) inferred the fine-grained massive beds in the formation are loess deposits; wind-blown sediments that accumulated on the river braid plain. We infer that moa walked across the poorly compacted, fine sediments on the ground as they navigated the river braid bars and banks of the braid complex, leaving deep, clear pedal impressions. Further aeolian deposition rapidly buried the tracks, preserving them in high detail. Subtle differences in cohesion and/or grain size allowed the overlying horizon to be differentially eroded from the paleosurface by the recent flooding of the Kyeburn River. Liesegang rings surround the footprint depressions (, and ), indicating that during diagenesis groundwater pooled on this irregular contact within the rock sequence, and resulted in the iron oxidation concentrations around the footprints. This extra lithification preserved the impressions’ surface morphology, resulting in the trackway withstanding the fluvial erosion that recently exposed the horizon. Note that print L1 was badly eroded by abrasion from the gravel bedload in the river during high water flow ( and ) and L4 was heavily damaged from overlying vegetation that was growing through the overlying sediment layers (not included in imagery or excavated).

Geological age – previous work

There are few biochronostratigraphic constraints for the Maniototo Conglomerate, which is stratigraphically thick (thickness over 450 m in some places; Bishop 1974) and is likely diachronous across Central Otago. Fossil pollen was not recovered from samples analysed from the moa trackway horizon (E. Crouch, GNS Science, pers. comm., 2019) and adjacent outcrops (New Zealand Fossil Record file numbers I41/f8, f9); all samples appeared to be too weathered to preserve fossil pollen. Fossil pollen dating from blue-grey outcrops ∼15 km north in the upper Kyeburn River (New Zealand Fossil Record file number I41/f8596) yielded an age of the New Zealand Stage Opoitian or younger (Pliocene to present; Bishop 1974). These outcrops were interpreted by Youngson et al. (1998) to be the unweathered Maniototo Conglomerate. The formation was given a ∼11 Ma age range of Late Miocene to upper Pleistocene by Bishop (1974, 1979) and Youngson et al. (1998), though a Pliocene age was suggested by Forsyth (2001) and Craw et al. (2016, 2022).

Nomenclature

Where possible, local Kai Tahu Māori placenames are given first in their spelling as per Māori pronunciation followed by anglicised spelling or the English name. Geological nomenclature that uses Māori placenames follows spelling from previous scientific publications, where any anglicised spelling is retained here for continuity in the literature.

Geological dating

Minimum and best-fit ages were determined from the overlying sand and gravel sequences using cosmogenic nuclide burial dating (see Granger and Muzikar 2001 for a summary of the method). Cosmogenic nuclides are produced in near-surface rocks and sediments when high-energy cosmic radiation interacts with nuclei in rocks and minerals – in this case producing the radioactive isotopes ²⁶Al and ¹⁰Be from oxygen in quartz. If the quartz is within the upper few metres of the surface, these nuclides will build up in the minerals (Dunai 2010). However, subsequent burial blocks the incident cosmic radiation with the result that the concentration of each isotope will reduce based on its half-life, 1.39 Ma for ¹⁰Be (Chemeleff et al. 2010) and 700 ka for ²⁶Al (Norris et al. 1983). This technique has been used in the past to date alluvial sequences and moraines, including those containing early human artefacts and remains (see review by Granger 2006).

We collected two samples from a 30 m high cliff face (KB21-01 & KB21-02) ∼12 m stratigraphically above the clay layer containing the tracks. Sample KB21-01 (442069E, 5000353S, UTM Zone 59G) was 20.4 m below the top of the cliff and consisted of ∼12 well-rounded quartz pebbles, up to 2 cm in diameter, weighing a total of ∼200 g. Sample KB21-02 (442073E, 5000375S) was 22 m below the top of the cliff and consisted of ∼1 kg of fine to medium, well-rounded sand. A third sample, KB21-03 (441814E, 4999866S), was taken ∼100 m downstream of the tracks on the modern riverbank. The sample consisted of fine-grained sand. The samples were processed using standard procedures (Kohl and Nishiizumi 1992). They were crushed and sieved to 0.25–0.5 mm, magnetically separated and subsequently treated with HCl and HF to purify the quartz and remove any meteoric components. Quartz dissolution and extraction of the Be and Al fraction was carried out at the Australian Nuclear Science and Technology Organisation’s (ANSTO) cosmogenic geochemistry laboratory using ∼300 µg of ⁹Be from a low-level beryl carrier solution added prior to complete HF dissolution. Accelerator mass spectrometry (AMS) measurements of Be were performed at the 6MV SIRIUS accelerator facility (Wilcken et al. 2017, 2019). Results for ¹⁰Be were normalised to KN-5-3 with a reference value of 6.320e-12 (Nishiizumi et al. 2007) and ²⁶Al analysis were normalised to KN-4-2 with a reference value of 3.096e-11 (Nishiizumi et al. 2004).

To determine the most-likely burial age, we constructed a model to track nuclide concentrations through time for different burial depths and durations (see supplemental file 1 for model details). The model accounts for nuclide production at depth due to neutrons and fast and stopped muons following Schaller et al. (2002), and decay based on the nuclide half-lives. The initial ²⁶Al and ¹⁰Be concentrations were calculated assuming background erosion rates of 0.001–0.01 cm/yr. We allowed initial burial depths to vary from a maximum of 150 m to a minimum of the current depth of the sample. The time of initial burial was allowed to vary from 8 to 0 Ma. Finally, the samples are ‘exhumed’ at any time between the initial burial and today. The model sampled uniformly from these parameters 500,000 times. Goodness of fit was calculated as the sum of the absolute error between the modelled and measured ²⁶Al and ¹⁰Be concentrations (i.e. a 1-sigma error indicates that the modelled concentration was within the 1-sigma error of the measured concentration). We modelled each sample independently using their respective concentrations and current depths.

Trackway measurements

Using photogrammetry, we created an orthographic model of the site before the excavation of the footprints. The model was calibrated against actual measurements of the footprints as well as a scale in the images to ensure its accuracy. Trackway descriptor measurements include stride length, step progression, and the width of the angulation pattern (WAP) as well as the pace angulation as per Bishop et al. (2017). The WAP was calculated using the cosine rule and trigonometry from the stride length and the step progression length. shows trackway measurement parameters and labelling. The numbering of the footprints is in order of occurrence from the first left footprint, this being L1 and the first right footprint being R1. Because of the complexity of working at the bottom of a riverbed of a large flowing river with a trackway that extended into a bank, not all footprints were able to be captured in the digital image. For footprints numbered R3 and L4 the trackway measurements were taken manually prior to removal from the riverbed.

Figure 5. Methodology of footprint and trackway labelling and parameter measurements. The footprints represented in A and B are of a left foot. A, Total footprint length as measured from the tip of the claw of digit III to the caudal edge of the heel imprint (TL) and total width which is measured as the widest measurement from the end of digit II to III (TW). Total digit length is from the front of the claw or furthest end point of the digit impression to the centre of the caudal end of the heel imprint. In the schematic drawing, there is a faint external track outline around the heel pad imprint. The heel measurement was taken from within the footprint proper. Digit divarication between digits II-III (y) and III-IV (x) were taken from the mid-line of the digits and measured in degrees. B, Digit lengths from the tip of the claw to the rise between pad impressions (d) and widths at the thickest point of the digit (w). C, Trackway parameters. Labelling of the trackway starts with L1. LP and RP are the left and right paces respectively. S is the stride length. Width of the angulation pattern (WAP) which is measured perpendicular to the stride length. Progression for completely regular trackways would be half the stride length. The reference point used for the trackway parameter measurements is the tip of the third digit. Please note that this is a stylised schematic and is not drawn to scale.

Footprint measurements

The following measurements from the trackway footprints were taken in millimetres: length of digit II, III and IV from the tip of the claw to the rise of the last pad impression of the digit (digit I was not possible to measure as it is the phalange that points backwards and it is not preserved in many moa). This rise is obvious on inspection of the footprints and formed an obvious landmark and can be seen in , and . Digit width at the thickest point which was consistently located approximately two-thirds down the digit (see , and ). Total length of each digit from the claw tip to the most caudal part of the heel impression for digits II, III and IV. Total length (note, this measurement is the same as the total length for digit III but separate refence is useful for interpreting other footprints) and total width of the footprint measured as the widest possible measurement from within the footprint impression. Divarication angle between digits II-III, III-IV and II-IV was taken as the angle between the midlines of the digits. Measurements taken are shown in . Maximum depth of footprint impression at the deepest point. All measurements were taken from the physical footprints themselves before they were dried. For the last footprint in the trackway (L4), the only measurement that was practicable was the depth of the print because of its location within the riverbank. For this reason, it was excluded in comparisons. To measure the single footprint made by a secondary print maker we used a calibrated orthographic model rather than the physical footprint.

Hip height and speed of travel estimation

The trackmaker hip height was estimated by the formulae developed by Alexander (1976). The calculation for hip height (acetabulum height [h]) for bipedal track makers is; $h=4\times footprintlength$As there are multiple footprints, an average length of the footprints was used.

Hip height estimates for the track maker was used in conjunction with Alexander's (1976) equation where SL = stride, h = hip height, g = acceleration of free fall;

$speed\approx 0.25speed\cong 0.25\surd g\times S{L}^{1.67}\times h^{-1.67}\cong 0.7826\times S{L}^{1.67}\times h^{-1.17}$

Mass of the Kyeburn moa

We have predicted the mass of both moa individuals represented as tracks based upon the allometric linear regression models developed by Alexander (1983) which is in line with other estimates of moa by Duncan and Holdaway (1989). We used the linear regression equation and constants from Alexander (1983) to predict the weight based on the longest digit with the equations and values as follows; $y/y_o=(m/m_o{)}^b$where y is the length of the longest digit in mm, m is the body mass in kg, y_o (178), m_o (92), are geometric mean values calculated by Alexander (1983) for Anomalopterygidae (synonym of Emeidae, see Checklist Committee 2022) for toe length. b (0.26) is used from the Alexander (1983) Model I. For the trackway maker, the results presented represent the mean length of digit III with the maximum and minimum values based on the length of the same digit from the footprints. Digit length is used because the equations produced by Alexander (1983) were derived for skeletal material rather than footprints. For the other larger footprint, we present the weight as a single value rather than a range as we only had one footprint to measure digit III from.

Comparative ichnology

The measurements of the footprints were compared to complete articulated moa foot skeletons from the collections of Tūhura – Otago Museum, Southland Museum and Art Gallery and Waitaki Museum. Only moa foot skeletons that were complete enough to get most of the measurements were used. We measured individuals from the species of moa that were from lowland South Island taxa according to Bunce et al. (2009). These are little bush moa (Anomalopteryx didiformis), stout legged moa (Euryapteryx curtus gravis), eastern moa (Emeus crassus), heavy-footed moa (Pachyornis elephantopus) and South Island Giant Moa (Dinornis robustus). Due to sexual dimorphism, we have treated male and female D. robusus separately. We have also included in this analysis upland moa (Megalapteryx didinis) as, although unlikely, there is a possibility that the Pliocene-Early Pleistocene habitat of the Maniototo could have overlapped with the Holocene habitats known for this species. From these moa skeletons, we measured the length of digits (sum of phalanges including the claw), width at the thickest point of the digit, as per B (all specimen registration numbers and raw data are presented in supplementary file 2).

Please note that due to the comparison that we are making here between the footprints and skeletal material we have not performed any statistical comparisons and have drawn our conclusions from the variation and observations alone, we have not done other measurements due to the historical inaccuracy of mounted moa skeletons, for example the angle of the phalanges and the digits lacking cartilage. When interpreting results, it should be noted that mummified remains show that the thickness of flesh on each side of the digits and areas of the foot lacking fat pads can be up to 5 mm thick in the smallest moa, A. didiformis (K. Fleury pers. obs. 2020 – mummified remains from Otago Museum (OMNZ AV7474) and Southland Museum (E80.4)).

Geological age

Samples KB21-01 and KB21-02 resulted in indistinguishable outputs and are both different from KB21-03 which is part of a different sedimentary complex (). Of the 500,000 model runs, over 3000 matched the measured concentrations within 1-sigma (). The model results demonstrate the covariance between the palaeo-denudation rate (the erosion rate of the catchment that supplied the quartz) and the modelled burial age. Faster erosion delivers lower ²⁶Al and ¹⁰Be concentrations, requiring shorter burial durations to decay down to the measured concentrations. Note, however, that there is a limit to this relationship () with palaeo-denudation rates faster than 0.005 cm/yr yielding fewer plausible burial scenarios. While low for tectonically active regions, this corresponds well with the global mean denudation rate from modern catchments (Portenga and Bierman 2011). The model results also suggest that the samples were initially buried to a depth of 74 (+44/−34) m and that exhumation to 2040 cm below the surface occurred at 410 (+330/−280) ka, however, the wide distribution of the burial depth and exhumation time limits their usefulness. The burial age, and hence the minimum age for the moa tracks is better constrained with a mean of 3.57 Ma (+1.62/−1.18) with a mode of 2.9 Ma. No acceptable models for KB21-01 and KB21-02 resulted in ages younger than 1.4 Ma ().

Figure 6. Results of the Monte Carlo burial model for KB21-01. The top panel (a) includes all 500,000 model results, demonstrating the overall poor fits for younger ages. The colour scale is the goodness of fit as measured by standard deviation from the measured concentrations. Panel (b) shows only those results that were within 1-sigma of the measured values. Note the covariance between paleo-denudation rates and modelled ages.

Figure 7. Frequency and cumulative probability of best fit modelled ages from sediment samples KB21-01 & KB21-02. The mean of the distribution is 3.57 Ma (+1.62/−1.18 at 1-sigma). The distribution is skewed with a mode of 2.9 Ma and a long tail towards older ages.

Trackway description

The trackway featured seven large tridactyl footprints that were on average 46.3 mm deep (sd = 5.8) and were between 272 and 300 mm wide and between 194 and 260 mm in length that were made by a single individual (detailed measurements are presented in and ). The average divarication angle in degrees between digits II-III was 48.50 (sd = 2.4) and III-IV 54.17 (sd = 3.4).

The sequence of the footprints shows the track-making moa was travelling in a north-north-east direction. The trackway sequence was greater than 2 m long and 0.4 m wide. The posterior section of the first footprint in the series (L1) shows evidence of substantial erosion due to it being covered in shingle in the river. The average sizes of the footprints in the sequence are presented in .

The stride length between the first left footprint and the second left footprint, L1 and L2, was 989 mm long, between R1 and R2 was 1020 mm and between L2 and L3 1030 mm. The physically measured stride length between R2 and R3 was 980 mm and between L3 and L4 was 1070 mm (please note that there were two more footprints that were obscured in from the orthographic image). Overall, there was a small direction change travelling to the right. The walking angular progression and the stride length are presented in . The WAP or gauge between L1-R1-L2 is 1450.05 mm, R1-L2-R2 is 164.91 mm and L2-R2-L3 is 127.29 mm showing how far each footprint extends out from the midline of progression.

Footprint descriptions

There was no impression for Digit I, the hallux, visible in or near any of the footprints. Digit II is shorter and thicker than digit IV, this and the following descriptive features can be seen in . Digit III is longer than II and IV and is directed forwards in the line of travel. Some of the footprints have what look to be claw impressions at the distal end of the phalanges where the impression terminates. The best examples of this are seen in footprint R1 and L3 where the claw marks are obvious for digits II and IV and appear rotated outwards. The impression for the ungual of digit III is rounded, lacking a narrowing pointed tip except for what is perhaps a faint mark at the distal end of footprint R2. The impression left by the fleshy metatarsal pad is asymmetrical with a shorter steeper inside margin than the outside one and well pronounced in all footprints except for the damaged footprint L1. shows enlarged views of all of the six footprints that were collected.

The plantar surface of the moa foot consists of fleshy integument that covers and protects the joints between the phalanges. There is a single undivided impression for the plantar pad of digit II in all the footprints. On digit III, there are two parts to the impression suggesting a divided fleshy pad evidenced by a faint widening of the wall and a discernible interpad space. We assume that there was a third pad that connected the distal phalanges to the claw, but it is not obvious. The impression at the distal end of digit III has collapsed around the claw which obscures the ability to clearly discern the interpad space and the claw impression of this digit. Of all the digits this is the hardest to discern specific details. Digit IV shows impressions for two fleshy digital pads and this is the clearest in footprints R1, L2 and L3.

The variations in length of the footprints are caused either by a collapse in the sediment, erosion within the river, or the growth of an iron-rich layer within the footprints which is harder than the surrounding sediment. Footprints R1 and L2 show the best preservation. No striations or other marks or skin textures were able to be discerned by the authors. There is in several of the footprints, a hollow under the surface where the claw was pushed into the mud. This indicates that during the stride as the moa lifted its foot, the digits were rotated plantar-posteriorly as the pressure of the foot was released, otherwise striations or drag marks should be visible. Each footprint has a very clear rounded metatarsal pad that has a defined mark that separates it from the digits. At the rear of each footprint is what looks to be a point where the metatarsal pad has rounded off the edge of the footprint as the foot was being placed onto the mud.

Secondary footprint description

The larger singular footprint was travelling in a south-east direction. It was only observed in the orthographic images taken when it was covered in water as a highly eroded impression that was barely visible (see for an enlarged view). It was a tridactyl print that was at its greatest dimensions 285 mm long from the end of the metatarsal pad to the end of the impression made by digit II and 448 mm wide. This footprint was made by an individual that had a wide interdigital angle and longer digit length than the other trackway maker. There were no obvious digital joint pad impressions but the rounded area at the back of the foot where the metatarsal pad would have been seen. This footprint can be seen in and enlarged in .

Estimated Hip height and speed of travel

The estimated hip height of the moa that created the track way is 1095.33 mm, with an average stride length of 1018 mm. The average estimated speed of travel when the footprints were deposited is 2.61 kmh⁻¹. The speed of the moa along the trackway did vary according to the stride length which at its minimum was 2.45 kmh⁻¹ and at its maximum 2.81 kmh⁻¹.

Mass of the moa

We calculated that the moa that left the trackway had a mass of 84.61 kg with an estimated minimum of 62.53 kg and a maximum of 118.24 kg based on the mean length of digit III and its minimum and maximum measurements from the trackway. For the moa that left the second, larger footprint, the estimated mass is 158.375 kg.

Species identification and comparison to articulated moa feet bones