(Color online) Concatenated spectra of detected events recorded in the presence of long-beaked common dolphins during a CalCOFI cruise. Spectra are sorted by peak frequency which are highlighted by black points.
(Color online) Original and reconstructed spectrum of a long-beaked common dolphin click. The inner figure shows a portion of the cepstrum associated with the echolocation click. The outer figure shows the 99 bin original spectrum (solid black line) derived from the time series and two zero-padded reconstructed spectra using the first 4 (dash) and 14 points (dashed-dotted line) of the cepstrum. The 0th cepstral coefficient was retained to preserve the frequency offset for illustrative purposes.
Histogram showing distribution of overall error rate in 100 trials of a randomized three-fold cross-validation experiment (total N = 300). The mean error is 0.22 ± 0.11σ with a median of 0.21.
(Color online) Distribution of error in 100 randomized three-fold cross-validation experiments reported by species. Mean, standard deviation, and median error rates are given in Table II.
Summary of data used for the species identification task showing the recording platform and year, number of sightings (sight) and echolocation clicks used, and the custom preamp board. Abbreviations: CalCOFI, California Cooperative Oceanic Fisheries Investigations oceanographic survey; SCI, San Clemente Island rigid hull inflatable boat survey, SC, Southern California Instrumentation cruises, FLIP, R/V Flip moored recordings, and HARP, High-frequency Acoustic Recording Package deployment.
Error rate statistics summarizing 100 three-fold cross-validation trials. The mean error (μ), standard deviation (σ), and median for each species represents single-species identification rate over all trials. In contrast, the mean overall error rate is computed by determining the percentage of misclassified click groups across all species. Cuvier’s beaked whale, Risso’s dolphin, and Pacific white-sided dolphins all had mean error rates below 7%.
Correlation between per species error rates across 100 three-fold cross-validation trials. More negative values indicate higher levels of misclassification.
Mean group size rounded to the nearest integer and the number of visual observations contributing to the mean as reported by Barlow and Forney (2007). Species with high mean group sizes aggregate in large groups at times.
Article metrics loading...
Full text loading...