Improvements in detection, localization, and tracking of sperm whales (Physeter macrocephalus) in Kauai

Since 2002, passive acoustic monitoring at the Pacific Missile Range Facility off Kauai, HI has yielded minimum density estimates and disturbance analyses for various whale species, but the ability to do so for sperm whales (Physeter macrocephalus) has been limited. Such efforts have traditionally suffered from false positives and decreased localization accuracy with increased click density. Recent software developments have aimed to address these issues and attempts were made to quantitatively and qualitatively assess improvement using archived 2014 data (n = 1063 hours of recording). Algorithm changes generated 2.4 times more detections (SD = 3.9x) and 1.4 times more localizations (SD = 2.4x) on average. Metrics indicating localization accuracy also improved. The number of detections per localization increased from a median of 5.5 (IQR = 5.2–5.8) to 7.0 (IQR = 6.0–8.3) and the percent of localizations theoretically capable of being tracked increased from 5.9% (IQR = 1.4%–11.6%) to 31.2% (IQR = 8.0%–46.9%). Visual comparison of localizations indicates much lower false positive rates, while other software developments have enabled tracking slow clickers and identifying foraging groups. Ultimately, track-level analyses should permit a more stable metric for minimum density estimates and disturbance tests and help establish baseline kinematics and behavior.Since 2002, passive acoustic monitoring at the Pacific Missile Range Facility off Kauai, HI has yielded minimum density estimates and disturbance analyses for various whale species, but the ability to do so for sperm whales (Physeter macrocephalus) has been limited. Such efforts have traditionally suffered from false positives and decreased localization accuracy with increased click density. Recent software developments have aimed to address these issues and attempts were made to quantitatively and qualitatively assess improvement using archived 2014 data (n = 1063 hours of recording). Algorithm changes generated 2.4 times more detections (SD = 3.9x) and 1.4 times more localizations (SD = 2.4x) on average. Metrics indicating localization accuracy also improved. The number of detections per localization increased from a median of 5.5 (IQR = 5.2–5.8) to 7.0 (IQR = 6.0–8.3) and the percent of localizations theoretically capable of being tracked increased from 5.9% (IQR = 1.4%–11.6%) to 31.2% (IQR = 8.0%–46.9...