UVM Theses and Dissertations
Format:
Print
Author:
Brauer, Corinne L.
Title:
Dept./Program:
Natural Resources
Year:
2012
Degree:
MS
Abstract:
Anurans are often the focus of monitoring, so it is important that the best methods be chosen to meet the needs of a particular program. Some have begun to use autonomous recording units (ARUs) which can collect audio data for extended periods of time with little maintenance and at sites where traditional call surveys might be difficult. Another new method is use of computer software that automatically identifies calls ofdifferentspecies, allowing for data spanning large areas orlong periods oftime to be processed more expediently. Although these technologies offer many benefits, with increased automation comes increased error. Several different variables affect errors in detection, but we examined four in this study: ARU type, method of species identification, focal species, and background noise. We collected higher and lower quality audio data with two different types of ARUs at wetland sites in Vermont and New York. We identified calls on these recordings with the tradition method of human identification as well as with the software package Song Scope which automatically identifies calls.
Results were compared to the calls known to be on the recordings in order to find instances where species were incorrectly identified (false positives) and where species were missed (false negatives). We performed a logistic regression in order to determine the effects of ARl, J type, species identification method, and focal species, as well as variables associated with background noise on the rates of these two kinds of error. The results showed a significant three-way interaction between ARU type, species identification method, and focal species. The detection of some species saw a large increase in error when using automatic species identification (as high as a 40% increase in the likeliness of an error) versus human identification, while others, like A. americanus were less effected (less than it 10% increase). For most species there was little difference in error between the two recording units, but for L. clamitans detections made by the automatic species identification software, there was 10% more probability of error when using lower quality ARU rather than a higher quality commercial one. We also found main effects of the background noise produced by traffic and rain, as well as the effect of how many individuals were calling. This indicates that all variables must be considered when choosing a monitoring method.
Results were compared to the calls known to be on the recordings in order to find instances where species were incorrectly identified (false positives) and where species were missed (false negatives). We performed a logistic regression in order to determine the effects of ARl, J type, species identification method, and focal species, as well as variables associated with background noise on the rates of these two kinds of error. The results showed a significant three-way interaction between ARU type, species identification method, and focal species. The detection of some species saw a large increase in error when using automatic species identification (as high as a 40% increase in the likeliness of an error) versus human identification, while others, like A. americanus were less effected (less than it 10% increase). For most species there was little difference in error between the two recording units, but for L. clamitans detections made by the automatic species identification software, there was 10% more probability of error when using lower quality ARU rather than a higher quality commercial one. We also found main effects of the background noise produced by traffic and rain, as well as the effect of how many individuals were calling. This indicates that all variables must be considered when choosing a monitoring method.