Predicting Anuran Abundance Using an Automated Acoustics Approach
Estimating abundance is critical in many areas of ecology and conservation biology. Despite this, methodological limitations prevent population monitoring at large spatial scales. Automated acoustic methods, such as recording devices and sound recognition models, can determine occupancy and phenology but have not been utilized to estimate abundance in amphibians. Here we evaluate a method to estimate the number of breeding female Wood Frogs (Lithobates sylvaticus) on the basis of the relationship between egg mass abundance and the acoustic activity of males, using an automated recognizer to count and quantify their vocalizations. We collected egg mass abundance data and recorded acoustic activity at sites in New Brunswick, Canada and the northern United States in 2015 and 2016. Egg mass abundance at a site was best predicted by the maximum daily number of calls on the basis of a model selection approach and the predictive ability of models (root mean square error ± 16–19 egg masses or 59–70%). The number of males in the pond was best explained by mean calls per recording in 2016 but in 2015 no acoustic metric explained male abundance well. Although this method shows promise, variation in 1) the quality of the recordings, 2) the timing of vocalization activity, and 3) the sex ratios among sites and years combine to increase the error in abundance estimates. With additional research, this method could be useful for monitoring at large spatial scales and we encourage additional trials of acoustic estimates of abundance, especially where estimates of abundance already exist using other methods.Abstract
Map showing the location of the study sites in New Brunswick, Canada and in Michigan, New York, and Connecticut, USA.
Relationship between the maximum number of recognizer hits made on the night with the most recognizer hits and egg mass abundance for Wood Frogs (Lithobates sylvaticus) for all sites and years. The different regions (New Brunswick, Canada and New York, Michigan, and Connecticut, USA) are identified by the legend in the plot. Data from 2015 are red and data from 2016 are blue. The solid black line represents the fitted values of the model based on all sites and years. The dotted black line is the fitted values from all sites in 2015 (plotting just New Brunswick results made no appreciable difference between fitted lines). The dashed black line is the fitted values from all sites in 2016.
Relationship between the number of recognizer matches and the number of true calls of the Wood Frog (Lithobates sylvaticus) in a recording from two sites in New Brunswick (NB) and the Michigan (MI) site from 2015 and 2016. Color codes are as follows: NB pond 2015, purple (coefficient = 0.252, r2 = 0.674); NB pond 2016, red (coefficient = 0.353, r2 = 0.508). MI pond 2015, green (coefficient = 0.884, r2 = 0.806); MI pond 2016, blue (coefficient = 0.818, r2 = 0.882). Shaded areas show 95% confidence intervals.
Timing of entry for male (solid lines) and female (dashed lines) Wood Frogs (Lithobates sylvaticus) into the Michigan breeding pond (A), mean air temperature (B), mean number of estimated vocalization per 5-min recording (C), and maximum number of vocalizations in a single 5-min recording by days into the breeding season in 2015 (red) and 2016 (blue; D). The recorder failed on day 11 in 2016; the data from this day have been interpolated for graphical representation purposes only (not used in the analysis) by taking the mean of day 10 and day 12 for 2016.
Contributor Notes
