Variation in low-frequency estimates of sound levels based on different units of analysis

The measurement and analysis of underwater sound is a complicated process because of the variable durations of contributing sources and constantly changing water column dynamics. Because the ambient sound distribution does not always follow a Gaussian structure and may be nonstationary in time, analysis over an extended period is required to accurately characterize the data. Utilizing recordings from the Indian Ocean, the temporal variation in ambient sound including transient signals was examined using multiple processing window lengths and subsampling intervals. Results illustrate the degree of uncertainty in sound levels based on different units of analysis. The average difference between sound level estimates in the 10-30 Hz band due to subsampling was 2 dB and as high as 4 dB. The difference in the full band (5-110 Hz) was as high as 6 dB. Longer averaging windows (200 vs 60 s) resulted in larger variations over different subsampling intervals. This work demonstrates how sampling protocols within a single dataset can influence results and acknowledges that comparative studies at the same location but with different sampling protocols can be substantial if signal processing parameters are not statistically accounted for to confirm interpretation of results and observed trends.