RESUMEN
A 9-yr dataset of continuous sea noise recording made at the Cape Leeuwin station of the Comprehensive Nuclear-Test-Ban Treaty hydroacoustic network in 2002-2010 was processed to detect calls from pygmy blue whales and to analyze diurnal, seasonal, and interannual variations in their vocal activity. Because the conventional spectrogram correlation method for recognizing whale calls in sea noise resulted in a too high false detection rate, alternative algorithms were tested and the most robust one applied to the multi-year dataset. The detection method was based on multivariate classification using two spectrogram features of transients in sea noise and Fisher's linear discriminant, which provided a misclassification rate of approximately 1% for missed and false detections at moderate sensitivity settings. An analysis of the detection results revealed a consistent seasonal pattern in the whale presence and considerable interannual changes with a steady increase in the number of calls detected in 2002-2006. An apparent diurnal pattern of whales' vocal activity was also observed. The acoustic detection range for pygmy blue whales was estimated to vary from about 50 km to nearly 200 km from the Cape Leeuwin station, depending on the ambient noise level, source level, and azimuth to a vocalizing whale.