Detection and estimation of rainfall from broadband acoustic signals

Abstract

Jeff Nystuen contributed major advances in our understanding of the physical acoustics of rain falling on the ocean, culminating in his pioneering work quantifying rainfall from the shape of the acoustic spectrum. Ma and Nystuen [JAOT (2005)] exploited Vagle’s wind spectrum to self-calibrate hydrophones for long deployments as acoustic rain gauges. Their algorithm predominantly relied on 3 narrowband frequencies to detect rainfall, and correlated the rainfall amount with the power spectral density (PSD) at 5kHz. Recent research at UMass Dartmouth built upon the foundational work of Nystuen, Ma and others to examine how much additional information can be gleaned from broadband acoustic spectra. These algorithms exploit Principal Component Analysis and Linear Discriminant Analysis for rainfall detection, coupled with Error-Correcting Output Codes (ECOC) for quantizing rainfall estimation. Testing on 5 months of 3-minute PSDs from a noisy cove found a detection probability of 78 ± 5% with a 1.1 ± 0.3% false alarm rate. Moreover, ECOC-based hourly rainfall estimates achieved 0.97 ± 0.01 correlation with rainfall measurements at a co-located meteorological station. This talk plans to present results from a recent 6 week coastal deployment in deeper water. [Work supported by ONR/UMassD MUST program]