Overview

ROCCA

Overview

The ROCCA (Real-time Odontocete Call Classification Algorithm) module is intended to provide a set of measured variables for detected acoustic events. The variables differ depending on whether the acoustic detection is a delphinid whistle or click. ROCCA uses these parameters in conjunction with a user-selected random forest classification model to classify the whistle or click to species. Three random forest models are available; one for odontocete species in the northwest Atlantic, one for species in the temperate Pacific and one for species in the waters surrounding Hawaii.

Individual whistle/click detections and classifications are grouped into user-defined Encounter Numbers. An encounter is defined as a set of whistles and/or clicks that can all be attributed to one school of odontocetes. Variables describing the acoustic behavior of the encounter are measured from the whistle and click detections for that encounter. Each encounter is classified based on information from detected whistles, clicks and ancillary variables using the encounter classifier, or as the species with the highest number of random forest tree votes if no event classifier has been selected.

Delphinid Whistles

ROCCA classifies delphinid whistles based on spectrographic measurements extracted from whistle contours. The contour extraction algorithm measures 50 different variables for each contour. Whistles can be classified based on measurements taken manually or based on measurements taken automatically. Manual classifiers are available for species in the tropical Pacific Ocean (TropPacWhist.model) and for species in the northwest Atlantic (NWAtlWhist.model). Automated whistle classifiers are available for species in the northwest Atlantic (NWAtlWhist_auto.model), the temperate Pacific (TempPacWhist_auto.model) and the waters surrounding Hawaii (HIWhist_auto.model).

The manual classifier requires seven main steps:

  1. User selects a whistle from the spectrogram display by drawing a box around it using the cursor

  2. The whistle is captured by ROCCA and displayed in a new spectrogram window window

  3. The whistle contour is traced either automatically by ROCCA or manually using the cursor. The user can manipulate the trace, if desired

  4. The time-frequency points are automatically extracted from the contour

  5. The contour parameters are automatically measured

  6. The contour is classified using the currently-loaded classifier model

  7. The classified whistle statistics are added to the specified Encounter in the GUI display and the saved data files

ROCCA can also be used in conjunction with the Whistle & Moan Detector module to measure and classify whistles detected and extracted automatically.

Odontocete Clicks

ROCCA measures 17 variables from clicks passed to it by the Click Detector module. Each click is classified to species based on these variables and the currently loaded classifier model, and the click statistics are added to the specified Encounter in the GUI display and the saved data files.

Classifiers are available for species in the tropical Pacific Ocean (TropPacClick.model), the northwest Atlantic (NWAtlClick.model), the temperate Pacific (TempPacClick.model) and the waters surrounding Hawaii (HIClick.model).

Encounters

ROCCA measures 17 variables from the whistles and clicks classified within each encounter. The encounter is classified to species based on these variables and the currently loaded classifier model, and the results are passed to the GUI display and the saved data files. Note that if no encounter classifier has been specified, the encounter will be classified based on the species in the encounter with the highest cumulative percentage of random forest tree votes.

Classifiers are available for encounters in the tropical Pacific Ocean (TropPacEnc.model), the northwest Atlantic (NWAtlEnc.model), the temperate Pacific (TempPacEnc.model) and the waters surrounding Hawaii (HIEnc.model).

References

Oswald, J. N., Rankin, S., Barlow, J., and Lammers, M. O. (2007). �A tool for real-time acoustic species identification of delphinid whistles,� The Journal of the Acoustical Society of America, 122, 587.

Next: Configuring ROCCA