Analyzing seasonally and wind-dependent performance of rain detection from underwater acoustic power spectral density.

Abstract

Rainfall can be empirically monitored by analyzing characteristic spectral features in the ocean's ambient sound. Previous work to detect and estimate rainfall from passive underwater acoustics used linear transformations of these features; Ma and Nystuen [J. Atmos. Oceanic Technol. 22, 1225-1248 (2005)] measured acoustic power at three narrowband frequencies, extended by Mallary, Berg, Buck, and Tandon [J. Acoust. Soc. Am. 154(1), 556-570 (2023)] and Berg [Master's thesis (2023)] to principal component analysis (PCA). PCA represents broadband spectra with small numbers of linear coefficients [Hotelling, J. Educ. Psychol. 24(6), 417-441 (1933)]. This research extends the PCA broadband approach to design detectors by rain, wind, and season to finely tune subspaces for each class. Five-minute power spectral density (PSDs) are computed using Welch's method and are separated into dry (<1 mm/hr) and rainy (≥1 mm/hr) recordings for each season and wind category. A linear dimension reduction matrix is defined for the dry PSDs of each season and wind category while preserving >99% of variance. Rainfall can then be detected using a likelihood ratio test of PSDs for each class. At a shallow-water site, separation of detectors by season and wind improves detection performance from 44.7% of rainfall volume to 55.1% (64.7% with false alarm filtering) with 1% false alarms. Accounting for wind and season may improve monitoring of natural processes such as rainfall from underwater acoustic recordings.