Webster, Trudi; Van Parijs, Sofie; Rayment, Will; Dawson, Steve (2019), Data from: Temporal variation in the vocal behaviour of southern right whales in the Auckland Islands, New Zealand, Dryad, Dataset, https://doi.org/10.5061/dryad.ps58g58
Autonomous recorders are frequently utilised for examining vocal behaviour of animals, and are particularly effective in remote habitats. Southern right whales are known to have an extensive acoustic repertoire. A recorder was moored at the isolated sub-Antarctic Auckland Islands for a year to examine whether the acoustic behaviour of southern right whales differed seasonally and throughout the day at their main calving ground in New Zealand. Recordings were made in each month except June, and vocalisations were audible in all months with recordings except January. A total of 35,487 calls were detected, of which upcalls were the most common (11,623). Call rate peaked in August (288±5.9 [SE] calls/hour) and July (194±8.3). Vocal behaviour varied diurnally with highest call rates detected at dusk and night, consistent with the concept that upcalls function primarily as contact calls. Zero-inflated model results confirmed that seasonal variation was the most important factor for explaining differences in vocal behaviour. An automated detector designed to expedite the analysis process for North Atlantic right whales correctly identified 80% of upcalls, although false detections were frequent, particularly when call rates were low. This study is the first to attempt year-round monitoring of southern right whale presence in New Zealand.
Summary of the analysis of the southern right whale vocalisation data
This spreadsheet contains data on the presence of right whale vocalisations for all of the recordings from the DSG recorder.
Subsequent tabs in the spreadsheet contain the analysis of calls by season, month, per day, diel period and hour and call type
Summary of the detector performance data
This spreadsheet contains data on the performance of the automated detector (the eXtensible BioAcoustic Tool (XBAT) written in MATLAB) which was developed for upcall vocalisations produced by North Atlantic right whales.
Manual and automated detections were compared to test the effectiveness and sensitivity of the algorithm for detecting SRW upcalls. We randomly selected 10% of the total number of recordings for verification analysis and annotated the detection events as either true or false based on the manual classification. Two measures of detector performance were applied: Positive Predictive Value (PPV) which represents the percentage of the upcalls logged by the detector that were judged by the human analyst to have been correctly assigned, and True Positive Rate (TPR) which represents the percentage of upcalls detected by the analyst also logged by the detector.
R script for Zero Inflated Model and Likelihood Ratio
A Zero Inflated Model with a negative binomial distribution was used to evaluate whether particular seasons and/or diel periods were important for explaining differences in right whale call presence and call rate.
This script runs the ZIM and conducts a log-likelihood ratio test to determine whether a poisson or negative binomial distribution best fitted the count data.
ZIM and likelihood ratio R script.txt
R script for jittering the upcall versus detector figure
The number of automatically detected upcalls per recording vs the number of upcalls detected by the analyst has been displayed in a figure. Points required jittering by 0.6 to offset and reveal overlapping points.
Jitter upcall v detector R script.txt