A machine learning approach for the discrimination of theropod and ornithischian dinosaur tracks.

Fossil tracks are important palaeobiological data sources. The quantitative analysis of their shape, however, has been hampered by their high variability and lack of discrete margins and landmarks. We here present the first approach using deep convolutional neural networks (DCNNs) to study fossil tracks, overcoming the limitations of previous statistical approaches. We employ a DCNN to discriminate between theropod and ornithischian dinosaur tracks based on a total of 1372 outline silhouettes. The DCNN consistently outperformed human experts on an independent test set. We also used the DCNN to classify tracks of a large tridactyl trackmaker from Lark Quarry, Australia, the identity of which has been subject to intense debate. The presented approach can only be considered a first step towards the wider application of machine learning in fossil track research, which is not limited to classification problems. Current limitations, such as the subjectivity and information loss inherent in interpretive outlines, may be overcome in the future by training neural networks on three-dimensional models directly, though this will require an increased uptake in digitization among workers in the field.

A new method to calculate limb phase from trackways reveals gaits of sauropod dinosaurs.

Limb phase, the timing of the footfalls in quadrupedal locomotion that describes common gaits such as the trot and the pace gait,1,2 is widely believed to be difficult or even impossible to estimate for extinct tetrapods.3-5 We here present a fundamentally new approach that allows for estimating limb phase based on variation patterns in long trackways. The approach is tested on trackways of modern mammals, where the estimates generally correspond well with the actually employed limb phase. We then estimate limb phases of giant wide-gauged sauropod dinosaurs based on three long trackways from the Lower Cretaceous of Arkansas, US.6,7 Gait selection at the largest body sizes is of considerable interest given the lack of modern analogs. Contrary to previous assumptions,8,9 our estimates suggest lateral sequence diagonal couplet walks, in which the footfalls of the diagonal limb pairs (e.g., right hind and left fore) are more closely related in time than those of the same side of the body (e.g., right hind and right fore). Such a gait selection allows for efficient walking while maintaining diagonal limb support throughout the step cycle, which is important for a giant, wide-gauged trackmaker.10 Estimations of limb phase may help to constrain other gait parameters, body size and shape, and, finally, potential trackmaker taxa.

Automatic generation of objective footprint outlines.

The objective definition of footprint margins poses a central problem in ichnology. The transition from the footprint to the surrounding sediment is often continuous, and the footprint wall complex, requiring interpolation, approximation, and a priori assumptions about trackmaker anatomy to arrive at feasible interpretations of footprint shapes. The degree of subjectivity of such interpretations is substantial, and outlines produced by separate researchers can differ greatly. As a consequence, statistical shape analysis, regardless if based on linear and angular measurements or on the shape as a whole, are neither fully repeatable nor objective. Here I present an algorithm implemented in the programming environment R that is able to generate continuous footprint outlines based on three-dimensional models-fully automatically, objectively, and repeatable. The approach, which is based on contour lines extracted from the model, traces the outline at the point where the slope of the track wall is steepest. An option for automatic landmark placement is implemented for tridactyl footprints. A case study was carried out on 13 footprints of a single trackway of a theropod trackmaker from the Lower Cretaceous of Münchehagen, Germany. Analysis of the landmark coordinates returned by the script did reproduce statistical results published in an earlier study that was based on human-made interpretative drawings, demonstrating the applicability of the present method for the objective and quantitative shape analysis of tracks. Although faint anatomical details are not always recorded and features not related to the foot anatomy may be included, the generated outlines tend to correspond with human-made interpretative drawings regarding the overall shape. While not suited as a full replacement of interpretative drawings, these generated outlines may be used as an objective basis for such interpretations.

Dinosaurs: Four Legs Good, Two Legs Bad.

The quadrupedal Sauropods - the biggest dinosaurs to walk the Earth - evolved from bipedal ancestors. Two new early sauropodomorphs from South Africa and Argentina indicate that very large, flexed-limbed sauropodomorphs coexisted with early columnar-limbed sauropods for 20 million years.

Geometric morphometric analysis of intratrackway variability: a case study on theropod and ornithopod dinosaur trackways from Münchehagen (Lower Cretaceous, Germany).

A profound understanding of the influence of trackmaker anatomy, foot movements and substrate properties is crucial for any interpretation of fossil tracks. In this case study we analyze variability of footprint shape within one large theropod (T3), one medium-sized theropod (T2) and one ornithopod (I1) trackway from the Lower Cretaceous of Münchehagen (Lower Saxony, Germany) in order to determine the informativeness of individual features and measurements for ichnotaxonomy, trackmaker identification, and the discrimination between left and right footprints. Landmark analysis is employed based on interpretative outline drawings derived from photogrammetric data, allowing for the location of variability within the footprint and the assessment of covariation of separate footprint parts. Objective methods to define the margins of a footprint are tested and shown to be sufficiently accurate to reproduce the most important results. The lateral hypex and the heel are the most variable regions in the two theropod trackways. As indicated by principal component analysis, a posterior shift of the lateral hypex is correlated with an anterior shift of the margin of the heel. This pattern is less pronounced in the ornithopod trackway, indicating that variation patterns can differ in separate trackways. In all trackways, hypices vary independently from each other, suggesting that their relative position a questionable feature for ichnotaxonomic purposes. Most criteria commonly employed to differentiate between left and right footprints assigned to theropods are found to be reasonably reliable. The described ornithopod footprints are asymmetrical, again allowing for a left-right differentiation. Strikingly, 12 out of 19 measured footprints of the T2 trackway are stepped over the trackway midline, rendering the trackway pattern a misleading left-right criterion for this trackway. Traditional measurements were unable to differentiate between the theropod and the ornithopod trackways. Geometric morphometric analysis reveals potential for improvement of existing discriminant methods.