Department of Environmental Conservation

M.S. –  Wildlife and Fisheries Conservation

Project:

Marbled Salamander Photographic Identification

Contact:

mchesser’at’eco.umass.edumchesser(at)eco.umass.edu
704-619-7044
Ag Engineering Annex A

Support:

National Science Foundation

Faculty Advisor:

Dr. Kevin McGarigal

Project abstract:

My research takes place within the context of a continuing vernal pool amphibian ecology study spread across 14 vernal pools in the Holyoke mountain range. My proposed research will build on the pioneering work of Gamble and Ravela (2008), who developed a novel photo-sorting algorithm to analyze the results of an ongoing capture-mark-recapture study with marbled salamanders (Ambystoma opacum). As one of only a few studies to use computer assistance and the photographic identification marking technique to estimate population parameters of a terrestrial species, while simultaneously maximizing image quality, our study is primed to address issues related to error rates (outside photographic quality) associated with this increasingly popular technique. I hope to utilize a database of more than 9,000 high-quality images of breeding adults across 10 years to address the impact of incremental increases in database size, probability of a match, and time between recaptures on error rate.

Objectives :
(1) to quantify the advantages and disadvantages between two methods of photographic identification: human-visual vs. computer-assisted
(2) determine if either method of identification is biased in the type of errors made
(3) determine if there is a difference between methods in terms of the frequency of errors made (more specifically to quantify the supposed decrease in error frequency gained by using computer-assistance)

With respect to the first objective, three characteristics (dependent variables) will be examined: speed (how quickly images can be processed), accuracy (how close the top-ranked found match is to “true” matches in the dataset), and precision (how close the top 10 found matches are to true matches in the dataset). Toward the second objective we will measure the types of error as either a false positive (incorrectly identifying two different individuals as the same), or a false negative (incorrectly identifying two captures of the same individual as different).