To our knowledge the present study represents the first application of a predictive equation derived solely from whole-body CT to the problem of body mass estimation for extinct animals. Previous volumetric mass estimate studies have relied upon articulated museum skeletons of extant species to derive a calibration equation (Brassey et al., 2013; Sellers et al., 2012). Yet articulated skeletons are often missing crucial specimen information, such as a recorded body mass. By working with frozen carcasses, body mass is directly measurable and uncertainties associated with mounting and posing of the skeletons can be avoided (Brassey & Sellers, 2014).

Our dataset consists of both ‘intact’ and ‘eviscerated’ pigeons as determined by examination of CT scans. Previous analyses of carcass composition have found eviscerated mass to represent 62–66% of live body mass in rock doves (Ibrahim & Bashrat, 2009; Omojola, Isa & Jibir, 2012), yet no data exist regarding the possible scaling of internal organ mass across a range of body sizes in the Columbiformes. As can be seen in Fig. 2, there is no consistent disparity between intact and eviscerated specimens, and the relationship between M_b and CH_vol in intact pigeons is relatively weak (r² = 70, p = 0.019). This correlation improves considerably when accounting for phylogeny (Supplemental Information 1), but remains weaker than the relationship between M_b and CH_vol characterizing eviscerated specimens. Live body mass has been shown to vary considerably in wild animals due to hydration, nutrition and gut content (Roth, 1990) and therefore some degree of scatter is to be expected in intact carcasses. Particularly striking is the variability in gizzard contents between similar-sized specimens visible in CT scans (see Fig. 5).

This suggests intact pigeon M_b cannot be corrected for the presence of internal organs using a single factor representing average percentage eviscerated mass as a function of live mass (i.e., multiplying by values of 0.62 or 0.66 previously found in the literature). Additionally, attempting to correct intact M_b by substituting intact CH_vol into the eviscerated regression model would be highly circular and result in artificially inflated correlation coefficients, if the equation were used in a predictive capacity. Therefore, we apply both the uncorrected OLS eviscerated model and combined (eviscerated and intact) model to bracket the range of likely dodo body masses. Interestingly, the very high correlation coefficient and low mean squared error of the eviscerated equation suggest that once the variability associated with fluid and gut content is removed, the relationship between the mass of the remaining musculoskeletal system and CH_vol is more tightly constrained.

No reliable records of the body mass of dodo exist prior to its extinction in the 17th Century and subsequent mass estimates have varied considerably. Early accounts of the flightless bird suggested an average mass of 50 lb (22 kg) (Herbert, 1634), although such accounts “have a tendency towards exaggeration” (Hume, 2006). More recently a ‘slim’ dodo (mean 10.2 kg) was proposed on the basis of femoral, tibiotarsal and tarsometatarsal length scaling in modern birds (Angst, Buffetaut & Abourachid, 2011a). However, hind limb bone length has been shown to correlate poorly with body mass relative to other cross-sectional geometric properties and frequently contains a strong functional signal (Field et al., 2013; Kirkwood et al., 1989; Campbell & Marcus, 1992; Brassey et al., 2013). Alternatively, a predictive equation based on femoral and tibiotarsal least circumference in ground-dwelling birds has suggested mass estimates between 9.5–12.3 kg (Louchart & Mourer-Chauviré, 2011).

The application of volumetric mass estimation techniques to the dodo has been rare. A sculpted scale model of a ‘slim’ dodo based upon mean skeletal measures was created to replicate sketches dating contemporaneously to its survival on Mauritius and resulted in mass estimated of 12.5–16.1 kg (Kitchener, 1993). In the same study, a ‘fat’ dodo model based on later ‘exaggerated’ artworks was predicted to weigh between 21.7–27.8 kg.

Here we estimate mean eviscerated body masses for articulated composite dodo skeletons of between 8.0–10.8 kg. Without further information regarding the effect of within-subject variability in gizzard, crop or gut content or interspecific scaling of viscera mass, any extrapolation to a live mass should be treated with caution. However, with this caveat in mind, a 33% increase in mass to account for missing organs (as quantified in extant C. livia) would take our results to 10.6–14.3 kg. This overlaps with the slim sculpted model based on contemporaneous accounts (Kitchener, 1993). Including our 95% prediction intervals takes both the NHMUK Tring and South Kensington skeletons to a maximum of 18.2 kg and 19.9 kg whole body masses, still considerably below the 22 kg suggested historically (Herbert, 1634). In contrast, the 95% prediction intervals of the Edinburgh dodo include 22 kg once multiplied by 1.33.

Unlike all previous volumetric studies, our convex hulling technique does not require a value for body density to be assigned from the literature. Instead, we directly derive the relationship between M_b and CH_vol in order to avoid uncertainty regarding assigning literature values, which have been shown to differ considerably across avian groups and with various methodologies for estimating body density (Seebacher, 2001). However, this does implicitly rely upon the predictive equation being applied to a fossil of an extinct species that is closely related to (and can therefore be assumed to share a similar body density to) the modern dataset from which the predictive equation was derived. This would include soft tissue density, integument density and skeleton density. In this case of estimating dodo mass based on extant pigeons, we believe this assumption can be upheld. In a micro-CT study of femoral and tibiotarsal mid-shaft cross-sectional geometry, the dodo and solitaire have been found to possess limb bone pneumasticity within the range of extant ground-dwelling and flighted species (Brassey et al., 2013) for example.

Alternatively, CH_vol may be multiplied by a given value of carcass density to give a hard lower limit to body mass (as carcass volume cannot be less than convex hull volume). The sole literature value for intact feathered pigeon density is 648 kg/m³ from Hamershock, Seamans & Bernhard (1993), producing hard lower bounds to estimated body mass 5.8 kg, 6.3 kg and 7.9 kg for the Tring, Kensington and Edinburgh composite skeletons respectively.

We consider the convex hulling technique to be superior to other sculpting-based volumetric methods (such as manual sculpting with clay (Kitchener, 1993) or digital sculpting with non-uniform rational B-spline (NURBs) curves (Bates et al., 2009)) for the purpose of mass estimation as soft tissues and hypothesized respiratory systems need not be reconstructed for fossils of extinct species, and the technique is entirely repeatable. When values for centre of mass (COM) and segment inertial properties are required for further biomechanical analyses, NURBs may be required in order to achieve a representative mass distribution across the skeleton. In such situations, it is essential that soft tissue reconstructions are based on quantitative comparative dissection data from relevant modern species in order to minimize subjectivity in model creation. However, for the sole purpose of mass estimation, convex hulling should be the preferred technique.

Previous authors have cautioned over the extrapolation of regression models beyond the limits of the extant dataset when applied in a predictive capacity (Henderson, 2006). To avoid this scenario, here we also apply a convex hull predictive model previously derived from ratites and ground-dwelling galloanserae birds (Brassey & Sellers, 2014) to the mounted dodo specimens. This results in mass estimates for the intact dodo ranging between 10.8–14.0 kg, remarkably similar to those values tentatively reconstructed by correcting the eviscerated pigeon model for missing viscera content. This further strengthens the argument for the reconstruction of a relatively slim dodo, and suggests extrapolation of the predictive equation beyond the range of modern pigeons does not, in this instance, result in implausible mass estimates.

Yet a predictive equation based upon cursorial ground-dwelling birds might also be considered inappropriate in light of the commonly-held perception of the dodo as being poor at locomotion, i.e., non-cursorial. The issue faced when assembling a modern calibration dataset on the basis of functional/behavioral similarities (as opposed to phylogenetic relatedness) is the requirement to assume a particular function/behaviour in a fossil species. In the case of the dodo, several ‘first-hand’ descriptions attest to the ‘tameness’ and ‘edibility’ of the bird (Parish, 2013, and references therein), yet very limited (and contradictory) accounts exist regarding its locomotor performance. Whilst some confirm the perception of dodo as fat and waddling:

Others suggest the dodo was capable of fast and ‘jaunty’ locomotion:

In light of this confusion, a more appropriate modern calibration dataset might therefore be selected on the basis of perceived evolutionary pressures (or lack thereof) to which the dodo was subjected, rather than assumed locomotor ability. Yet this also proves problematic, as the fates of many other recent flightless bird species that have evolved in the absence of native terrestrial predators have followed that of the dodo i.e., recent extinction. Possible extant candidates are limited to the kakapo (Strigops habroptilus), Galapagos cormorant (Phalacrocorax harrisi), Auckland Islands and Campbell Islands teal (Anas aucklandica and Anas nesiotis) and over a dozen species of Gruiformes. Given that the majority of the above species are categorized as threatened or extinct in the wild (IUCN, 2013), obtaining specimens and associated mass data is extremely challenging. Therefore whilst the dodo may have differed from our modern calibration datasets in being both flightless and non-cursorial, it must be recognized that a panacea for dodo mass estimation is unlikely to exist, and perhaps the most appropriate recent analogues are already extinct or nearing extinction.

The dodo specimens included in this study are composite skeletons, comprising skeletal material from more than one individual and including sculpted or cast elements. Therefore, our study is limited to estimating the body mass of the hypothetical animal represented by each articulated skeleton, rather than a known individual. Currently there exists only one near-complete dodo skeleton comprising a single individual (the Thirioux dodo), upon which research is currently still continuing (Randall et al., 2014).

Whilst the use of composite skeletons should clearly be treated with caution when used in biomechanical analyses, their composite nature does not entirely rule out their use, particularly in the case of mass estimation. A recent large-scale macroevolutionary (Benson et al., 2014) study of body size in a fossil lineage relied upon mass data derived solely from humeral and femoral circumferences of one individual specimen per species. While this approach is often the only one feasible given the highly fragmentary nature of the fossil record, taking one individual as being representative of an entire species leaves us vulnerable to the possibility of high levels of intraspecific variation.

In contrast, a volumetric reconstruction based on a composite skeleton may be more likely to reflect a species average by virtue of being a combination of several individuals and could be less skewed by isolated robust or gracile elements. If subsequent biomechanical analyses are to be carried out (such as finite element analysis on a particular musculoskeletal unit), then it is important that the body mass entered into the analyses is representative of that specific individual. However, for the case of volumetric body mass estimation alone, it ought to be possible to derive a representative species mean from a composite skeleton.

Of more concern is the frequency of missing, deformed or reconstructed material within a fossil mount. Known issues with the dodo mounts included in this study include missing ribs (Edinburgh skeleton), missing carpals (NHMUK South Kensington skeleton), deformation of the fragile pubis (NHMUK South Kensington skeleton) or the loss of the most of the ischium, pubis and caudal vertebrae (NHMUK Tring skeleton). For a given object, the extent of the convex hull fitted to that object is dictated solely by its geometric extremes. In many ways this is advantageous for volumetric fossil reconstructions as damage occurring within the bounds of the convex hull does not affect our volume estimate. However, when extremities are missing (such as the caudal tip of the pubis), the shape and volume of the convex hull are strongly affected. This is evident in the low percentage trunk volume of the NHMUK Tring skeleton (Fig. 5) compared to those of extant pigeons and other dodos.

Whilst some evidence of underdevelopment of pectoral elements and overdevelopment of pelvic elements in the dodo is discernable relative to extant volant pigeons, Fig. 5 predominantly illustrates the important contribution of trunk volume to total mass estimates. The Edinburgh skeleton has a proportionally more voluminous trunk than that of extant pigeons and other dodo skeletons, and therefore all other skeletal elements contribute proportionally less to total CH_vol. The more voluminous trunk relative to other specimens may be attributed to the anterior positioning of the sternum due to constraints associated with the armature supporting the mount. The opposite is true of the NHMUK Tring skeleton, in which damage to the extremities of the pelvic girdle result in a reduced trunk volume. This highlights the sensitivity of volumetric reconstructions of fossils of extinct species to trunk morphology, and should be a concern when working with both composite and complete fossil specimens. Whilst the inclusion of cast and/or sculpted material outside of the trunk may also introduce additional uncertainty into the reconstruction, it is unlikely to impact heavily upon mass estimates given their relatively minor contribution to overall volume (Fig. 4).

Here we present the first volumetric reconstruction of fossil body mass based entirely on modern whole-animal CT data. The eviscerated body mass of three articulated composite dodo skeletons is estimated to fall between 8.0–10.8 kg. When accounting for missing organ mass, our mean values still fall towards the lower range of previously published mass estimates. As the availability and cost of CT improves, we believe this non-subjective convex hull approach will become increasingly commonplace. Mass estimation of extinct species from fossils relies upon two key components; a reliable calibration equation derived from extant species, and an accurate reconstruction of the extinct individual from its fossil. We discuss the issues surrounding the use of articulated composite skeletons, and highlight the particular importance of trunk morphology to volume reconstructions. We suggest future efforts should focus on quantifying ribcage and sternal geometry in extant groups in order to bracket the possible trunk shape in fossils of extinct species.