A COMPARATIVE STUDY OF MUSEUM DIGITIZATION STANDARDS FOR USE IN SCIENTIFIC RESEARCH

Specimen digitization efforts lack standardization across institutions but produce images with widespread availability. The effects of different protocols may have unintended consequences if used in research studies. I evaluated the ability of these photographs to be used in morphometric research. Specifically, I tested: 1) Do differences between standards/protocols distort or fail to accurately capture the specimen's shape information? 2) Do particular image differences affect statistical interpretations of population level variation?

I first conducted a specimen re-photographing project within four museums, representing zoology, entomology, herbaria, and paleontology using standards developed within the museums. I then re-photographed the 10 specimens as a researcher would for geometric morphometrics, resulting with five photographs each. Photographs were processed in Fiji, tpsUtil, tpsDig, and all analyses were conducted in R. Geometric morphometrics with Procrustes superimposition was used to quantify morphospace variance/shape variation for all specimens. Principal components analysis (PCA) and Canonical variates analysis (CVA) were conducted to evaluate if there were any differences between the images produced with different standards. Resolution of the uploaded image, different angling of specimens, and the plane of focus were the most significant factors that affected accurate landmarking. Protocol based differences were multiple vs. single specimen imaging and specimen handling. The least variation was seen either in flat, high contrast colored specimens or more 3D specimens with simple landmarking.

Exploring the resolution issue further, I took a separate sample of 26 ammonites and landmarked both web-served and higher resolution archived images. Both PCA and CVA tests showed large overlap and variation in procrustes distances within the morphospace between the different resolution samples. Analysis of variance relating to shape was significant (p<0.0001) when looking at number of specimens per photo but not with resolution. Ignoring single specimen photos, procrustes variance increased with the number of specimens in the photo. These results highlight that when conducting image based research, resulting shape differences can be influenced by imaging standards.