91ÅÝܽ

Title: An Automated Click Classifier for Detecting Southern Resident Killer Whales in the Salish Sea: The Lasso or Trees?
Date: July 25th, 2025
Time: 10:00am
Location: LIB 7200
Supervised by: Ruth Joy and Robert Tibshirani

Abstract:

Several commercial shipping routes in the Salish Sea pass through critical habitat of the endangered Southern Resident killer whale (SRKW), a population increasingly threatened by elevated underwater noise levels. To mitigate the impacts of vessel noise on SRKW in a cost- effective manner, automated algorithms must differentiate SRKW from a sympatric population of West Coast Transient (WCT) killer whales. This study utilizes a multi-year dataset from the Boundary Pass Underwater Listening Station. Twenty-one machine learning classsifiers  were trained and evaluated using logistic regression with varying lasso-penalization or ensemble tree methods. A population-specific killer whale click detector-classifier is presented that integrates a Teager-Kaiser click detector with an eXtreme gradient boosting classifier to predict SRKW or WCT presence using 12 acoustic features  (Event-Level Precision: 0.83, Recall: 1.00, F2 Score: 0.96, MCC: 0.85, ROC-AUC: 0.98).  The findings of this study support the development of automated vessel alert systems for SRKW in the Salish Sea.