While dolphins can be detected in a variety of ways, dolphin detectors often refer to T/C-PODs and other signal processing algorithms used during Passive Acoustic Monitoring (PAM) studies.
While dolphins can be detected in a variety of ways, dolphin detectors often refer to T/C-PODs and other signal processing algorithms used during Passive Acoustic Monitoring (PAM) studies (www.passiveacousticmonitoring.com). Dolphin detectors monitor dolphin activity when visual monitoring is not conducive (e.g. poor weather conditions, fog, hours of darkness, and inconspicuous species). The limitation to all PAM is that the dolphins must be vocalising in order to be detected.
C-PODs (www.c-podclickdetector.com) are digital successors of T-PODs (www.t-pod.co.uk). A description of T-POD hardware, process of data collection and classification can be found in the materials and methods section of Todd et al. 2009 (www.osc.co.uk).
C-PODs contain omnidirectional hydrophones and work by detecting echolocation clicks produced by toothed whales, dolphins, and porpoises (collectively known as odontocete cetaceans). These animals produce sonar clicks to navigate (www.dosits.org), forage (www.dosits.org), as well as for communication. C-PODs can be used to detect all toothed cetaceans, with the exception of sperm whales (Physeter macrocephalus), because the frequency of their clicks (0.1 – 30 kHz) can fall below the C-POD frequency detection range (20 – 160 kHz).
C-PODs detect dolphin clicks using digital waveform characterisation. For each click, time, centre frequency, Sound Pressure Level (SPL), duration and bandwidth is logged. The advantage of C-PODs over other PAM technologies is that they only store data from clicks, which reduces data storage requirements, and allows for quicker post processing.
Specialised C-POD software is used to analyse click data (www.chelonia.co.uk) Given that tonal clicks made by dolphins are not always distinctive, C-PODs detect click trains, not just individual clicks. A click train is a series of similar clicks produced by a single animal in succession. A KERNO classifier is used to detect and classify click trains automatically. Click trains are classified into four classes:
The KERNO classifier is an improvement on the T-POD software train classifier, as it now uses additional non-parametric statistical methods. This reduces the effect of outliers (i.e. clicks that come from a different source). Other advantages to the KERNO classifier include:
These dolphin detectors are fully automated and extremely robust, meaning they can be left in the field for extended periods of time. See www.dolphindetectors.co.uk for some of the research that has been conducted with these dolphin detectors. There is also a Deep C-POD that can be deployed to a depth of 2 km.
Dolphin detectors are often incorporated into software packages used with towed hydrophone arrays (www.towedhydrophonearrays.com). There is a vast selection of software programmes available for marine mammal signal processing, e.g. IFAW suite (www.marineconservationresearch.co.uk), Ishmael (www.bioacoustics.us), Raven (www.birds.cornell.edu), PAMGuard (www.pamguard.org), custom written programmes in MATLAB® (www.mathworks.co.uk) etc. PAMGuard is a combination of the IFAW suite and Ishmael, and has become the industry standard for marine mammal mitigation. It is also available freely and is open source. An advantage of towed hydrophone arrays over C-PODs is that they can be used in real time, and can be towed over large areas.
When classifying dolphin whistles, many programmes use a collection of variables from whistle contours (e.g. see figure below). Statistical techniques are then used to classify these variables as belonging to a certain dolphin species.
Two detectors are used commonly in PAMGuard, a whistle and moan detector, and a click detector. Tonal vocalisations, produced by odontocetes (http://en.wikipedia.org) and mysticetes (http://en.wikipedia.org) are detected by the whistle and moan detector, whilst the click detector records high frequency click sounds from odontocetes such as harbour porpoises (Phocoena phocoena), beaked and sperm whales.
The PAMGuard whistle classifier collects information from multiple whistles and analyses them statistically as a group, instead of classifying single whistles. In order for the classifier to know what characteristics belong to each species, it must be trained with a sample of whistles beforehand. In reality, components of whistles can be missed by the detector, resulting in short fragments of whistles. The PAMGuard whistle classifier consequently accumulates statistics of three parameters (start frequency, slope, and curvature) over many fragments to identify the species.
PAMGuard also has a Real-time Odontocete Call Classification Algorithm (ROCCA) module that uses spectrographic measurements extracted from whistle contours to classify delphinid whistles. A whistle is selected from the spectrogram display by the user, and a contour extraction algorithm measures 54 different variables of the whistle contour. The variables are then classified using the user-selected classifier model.