Optimizing digitalization effort in morphometrics

Abstract

Quantifying phenotypes is a common practice for addressing questions regarding morphological variation. The time dedicated to data acquisition can vary greatly depending on methods and on the required quantity of information. Optimizing digitization effort can be done either by pooling datasets among users, by automatizing data collection, or by reducing the number of measurements. Pooling datasets among users is not without risk since potential errors arising from multiple operators in data acquisition prevents combining morphometric datasets. We present an analytical workflow to estimate within and among operator biases and to assess whether morphometric datasets can be pooled. We show that pooling and sharing data requires careful examination of the errors occurring during data acquisition, that the choice of morphometric approach influences amount of error, and that in some cases pooling data should be avoided. The demonstration is based on a worked example (Sus scrofa teeth) using a combinations of 18 morphometric approaches and datasets for which we identified and quantified several potential sources of errors in the workflow. We show that it is possible to estimate the analytical power of a study using a small subset of data to select the best morphometric protocol and to optimize the number of variables necessary for analysis. In particular, we focus on semi-landmarks, which often produce an inflation of variables in contrast to the number of available observations use in statistical testing. We show how the workflow can be used for optimizing digitization efforts and provide recommendations for best practices in error management.

Sorting the Flock: Quantitative Identification of Sheep and Goat from Isolated Third Lower Molars and Mandibles Through Geometric Morphometrics

“…In our study we found that the cumulative error in picturing and landmarking is, at its highest, around 11%. This is relatively similar to other protocols such as the one used by Evin et al (2020) to measure pig third lower molars were inter-operator error averaged around 13% when both landmarks and sliding semi-landmarks were used. In addition, our study showed that the amount of error increases only marginally when pictures are acquired by multiple operators.…”

“…In addition, since two different people acquired the pictures of the remaining specimens in this study (MJ and AE), the differences between operators were also assessed. The measurement repeatability was quantified using Procrustes ANOVAs (Claude, 2008;Evin et al, 2020).…”

Incongruences between morphology and molecular phylogeny provide an insight into the diversification of the Crocidura poensis species complex

“…Due to the small size, the fragile nature, and the high likeliness of Crocidura specimens, using the outlines allowed the inclusion of more individuals while capturing the variability in overall skull shape. In addition, the use of an automated procedure avoided the biases caused by manual landmark placement 49 . 433 genotyped individuals of the C. poensis species complex were photographed using a NIKON D5600 and a 60 mm AF-S Micro NIKKOR lens.…”

Testing inter-observer error under a collaborative research framework for studying lithic shape variability

“…Comparatively, fewer studies have examined the levels of inter-observer error in outline based GMM methods. Evin et al (2020), in an investigation of error between morphometric approaches, found that although methods that employ landmarks were the most sensitive to error, outline data saw relatively lower levels of intra-observer error compared to inter-observer error, with photography being an in uential source of variance between observers. Digital photography is widely used in 2D GMM as it is inexpensive, easy to perform, and does not require extremely specialist knowledge or equipment, with the digitization of landmarks and/or outlines on the resulting images providing a 2D representation of the 3D object.…”

“…The focal length and speci cations of the lens used can, however, cause parallax error, optical distortion that occurs when the specimen is too close or not directly centered beneath the lens ( sheye). Nonetheless, several studies employing both landmark and outline methods suggest that 2D GMM data are minimally affected by parallax error, especially when the camera set-up is standardized and calibrated, with deviations small and constant enough for accurate analyses (Caple et suitable for collaborative research designs in studies of lithic shape, due to the objectivity of data capture, the fact that landmark methods have high rates of inter-observer error, though this is more pertinent during landmark digitization than object photography (Evin et al, 2020), and the potential to reduce inter-observer error through the standardization of the photography procedure.…”