• Amitai, E., , Nystuen J. A. , , and Liao L. , 2004a: Listening to the rain. Bull. Amer. Meteor. Soc., 85, 15.

  • Amitai, E., , Nystuen J. A. , , Liao L. , , Meneghini R. , , and Morin E. , 2004b: Uniting space, ground, and underwater measurements for improved estimates of rain rate. IEEE Geosci. Remote Sens. Lett., 1, 3538, doi:10.1109/LGRS.2004.824767.

    • Search Google Scholar
    • Export Citation
  • Amitai, E., , Nystuen J. A. , , Anagnostou E. N. , , and Anagnostou M. N. , 2007: Comparison of deep underwater measurements and radar observations of rainfall. IEEE Geosci. Remote Sens. Lett., 4, 406410, doi:10.1109/LGRS.2007.895681.

    • Search Google Scholar
    • Export Citation
  • Anagnostou, M. N., , Nystuen J. A. , , Anagnostou E. N. , , Nikolopoulos E. I. , , and Amitai E. , 2008: Evaluation of underwater rainfall measurements during the Ionian Sea Rainfall Experiment. IEEE Trans. Geosci. Remote Sens., 46, 29362946, doi:10.1109/TGRS.2008.2000756.

    • Search Google Scholar
    • Export Citation
  • Anagnostou, M. N., , Nystuen J. A. , , Anagnostou E. N. , , Papadopoulos A. , , and Lykousis V. , 2011: Passive aquatic listener (PAL): An adoptive underwater acoustic recording system for the marine environment. Nucl. Instrum. Methods Phys. Res., 626–627A, S94–S98, doi:10.1016/j.nima.2010.04.140.

    • Search Google Scholar
    • Export Citation
  • Andrew, R. A., , Howe B. M. , , Mercer J. A. , , and Dzieciuch M. A. , 2002: Ocean ambient sound: Comparing the 1960s with the 1990s for a receiver off the California coast. Acoust. Res. Lett. Online, 3, 6570, doi:10.1121/1.1461915.

    • Search Google Scholar
    • Export Citation
  • Dushaw, B., and et al. , 2010: A global ocean acoustic observing network. Proceedings of OceanObs’09 Conference: Sustained Ocean Observations and Information for Society, J. Hall, D. E. Harrison, and D. Stammer, Eds., Vol. 2, ESA Publ. WPP-306, doi:10.5270/OceanObs09.cwp.25.

  • Gavrilov, A., , McCauley R. , , Thomas F. , , Klaka K. , , Kent C. S. , , Perry M. , , and Duncan A. , 2009: Acoustic observatories of the Australian marine observing system. Proc. Third Int. Conf. and Exhibition on Underwater Acoustic Measurements: Technologies and Results, Nafplion, Greece, IACM/FORTH, 32-3. [Available online at http://promitheas.iacm.forth.gr/UAM_Proceedings/uam2009/32-3.pdf.]

  • Hildebrand, J. A., 2009: Anthropogenic and natural sources of underwater noise in the ocean. Mar. Ecol. Prog. Ser., 395, 520, doi:10.3354/meps08353.

    • Search Google Scholar
    • Export Citation
  • Ma, B. B., , and Nystuen J. A. , 2005: Passive acoustic detection and measurement of rainfall at sea. J. Atmos. Oceanic Technol., 22, 12251248, doi:10.1175/JTECH1773.1.

    • Search Google Scholar
    • Export Citation
  • Medwin, H., , Nystuen J. A. , , Jacobus P. W. , , Ostwald L. H. , , and Synder D. E. , 1992: The anatomy of underwater rain noise. J. Acoust. Soc. Amer., 92, 16131623, doi:10.1121/1.403902.

    • Search Google Scholar
    • Export Citation
  • Miksis-Olds, J. L., , Nystuen J. A. , , and Parks S. E. , 2010: Detecting marine mammals with an adaptive sub-sampling recorder in the Bering Sea. J. Appl. Acoust., 71, 10871092, doi:10.1016/j.apacoust.2010.05.010.

    • Search Google Scholar
    • Export Citation
  • Nittis, K., , Zervakis V. , , Papageorgiou E. , , and Perivoliotis L. , 2002: Atmospheric and oceanic observations from the POSEIDON Buoy Network: Initial results. J. Atmos. Ocean Sci., 8, 137149, doi:10.1080/10236730290004076.

    • Search Google Scholar
    • Export Citation
  • Nystuen, J. A., 2001: Listening to raindrops from underwater: An acoustic disdrometer. J. Atmos. Oceanic Technol., 18, 16401657, doi:10.1175/1520-0426(2001)018<1640:LTRFUA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Nystuen, J. A., , Amitai E. , , Anagnostou E. N. , , and Anagnostou M. N. , 2008: Spatial averaging of oceanic rainfall variability using underwater sound: Ionian Sea Rainfall Experiment 2004. J. Acoust. Soc. Amer., 123, 19521962, doi:10.1121/1.2871485.

    • Search Google Scholar
    • Export Citation
  • Tsabaris, C., and et al. , 2011: A marine groundwater spring in Stoupa, Greece: Shallow water instrumentation comparing radon and ambient sound with discharge rate. Procedia Earth Planet. Sci., 4, 39, doi:10.1016/j.proeps.2011.11.002.

    • Search Google Scholar
    • Export Citation
  • Vagle, S., , Large W. G. , , and Farmer D. M. , 1990: An evaluation of the WOTAN technique for inferring oceanic wind from underwater sound. J. Atmos. Oceanic Technol., 7, 576595, doi:10.1175/1520-0426(1990)007<0576:AEOTWT>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Locations of POSEIDON moorings with PALs (Pylos and Athos), weather radar location (Andravida), and rain gauge sites (Aktio and Methoni).

  • View in gallery

    Example taken from a single sample of the PAL sound bite data. (top) Time series of 4.5 s length, sampled at 100 kHz. Eight subsamples (each 10.24 ms) are evaluated to determine if the sample is stationary. If it is stationary, then the mean spectrum is saved and the original time series is discarded. (bottom) The eight selected subsamples spectra are shown with the mean spectrum (thick black line). This is the acoustic signature classified from the algorithm as drizzle in low (~3 m s−1) wind conditions.

  • View in gallery

    Example of a nonstationary sample taken from the PAL’s sound bites. (top) Time series of pressure (Pa) over 4.5 s with a sample rate of 100 kHz. (middle) FFT of (top) indicating high-frequency echolocation clicking and then a whistle from a cetacean in the northern Aegean Sea are recorded. (bottom) Spectra for eight subsamples at the times shown in (top).

  • View in gallery

    (top) Example of a 20-day time series plot of the PAL ambient total sound records (gray dots) taken from Athos in the northern Aegean Sea indicating a slowly varying background sound that is punctuated by shorter duration acoustic events, such as rainstorms (dots), ship passages (stars), and whale vocalizations (x). (bottom) Geophysical interpretation of the sound record, which is the desired product for many users. Classification includes the detection and removal of noises, such as ships, and then the subsequent quantification of the physical processes producing the sound. Note that after day 358, the surface anemometer on this mooring failed, but the acoustical wind speed measurement continued throughout the duration of the deployment.

  • View in gallery

    Scatterplots of acoustic parameters using (a) 2 vs 20 kHz, (b) 8 vs 20 kHz, (c) 8 kHz vs 2–8 kHz slope, and (d) 8 kHz vs 8–15 kHz slope. Data are shown from the Athos mooring PAL for days 350–370 of 2008, with the wind sound source data highlighted. The units are dB relative 1 μPa2 Hz−1 for sound level and dB decade−1 for spectral slope.

  • View in gallery

    Data points from the Athos PAL during days 350–370 of 2008 (15 Dec 2008–4 Jan 2009) showing the classification of the clicks (black x) from the wind speed (gray dots). Shown is cetacean (probably striped dolphin) detection using a comparison of the sound level at 20 kHz and at 30 kHz. Echolocation clicks have a broadband peak at 30 kHz (see Fig. 3).

  • View in gallery

    Time series plot showing the detection of high-frequency (~40 kHz) clicks in the PAL data. Clicking is subjectively confirmed by examining the saved original time series [sound bites (SB)] samples that contained transient sounds. The time of all SB recorded are shown on the bottom of the figure (circles). SBs (diamond) containing high-frequency clicks (see Fig. 3) confirm spectral click detection (cetaceans present). SBs (square) containing ship noise are also marked with loud sonar detected on day 339.

  • View in gallery

    Temporal records of ship passages of (a) close ship at 0750 UTC on day 377 (11 Jan 2009) and (b) distant ship at 2240 UTC on day 359 (24 Dec 2008).

  • View in gallery

    Comparison of buoy anemometer wind speed and acoustic wind speeds using Vagle et al.’s (1990) algorithm [Eq. (1)] and the new third-order fit [Eq. (2)].

  • View in gallery

    Comparison of 3-h averages of wind speed as measured by the Pylos PAL and the surface-mounted anemometer on the mooring.

  • View in gallery

    Comparison of rainfall accumulation from the Pylos PAL, the Andravida radar, and the Methoni daily rainfall reports during February 2009. Black line indicates time periods when the radar was on, which was during all significant rainfall events at Pylos during February. Also shown is PAL wind speed compared to the mooring anemometer.

  • View in gallery

    PAL ambient noise budget plot over the Athos’ site showing the relative dominance of the sound sources: wind, rain and shipping (mostly distant shipping). During the winter, high wind speeds make wind the dominant sound source.

  • View in gallery

    Mean sound levels for wind, rain, and shipping (distant shipping) at (a) 2 and (b) 20 kHz when that sound source is detected.

  • View in gallery

    Summary of cetacean detections at Athos site. Vocalizations, mostly high-frequency (30–40 kHz) clicking but also 10-kHz whistling (Fig. 3), are detected.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 98 98 19
PDF Downloads 80 80 15

Monitoring Greek Seas Using Passive Underwater Acoustics

View More View Less
  • 1 Hellenic Center for Marine Research, Anavyssos, Attica, Greece, and Applied Physics Laboratory, University of Washington, Seattle, Washington
  • | 2 National Observatory of Athens, Athens, Greece
  • | 3 Department of Civil and Environmental Engineering, University of Connecticut, Storrs, Connecticut
  • | 4 Hellenic Center for Marine Research, Anavyssos, Attica, Greece
© Get Permissions
Full access

Abstract

The Hellenic Center for Marine Research POSEIDON ocean monitoring and forecasting system has included passive underwater acoustic measurements as part of its real-time operations. Specifically, low-duty-cycle long-term passive acoustic listeners (PALs) are deployed on two operational buoys, one off Pylos in the Ionian Sea and the second off Athos in the northern Aegean Sea. The first step toward the quantitative use of passive ambient sound is the classification of the geophysical sources—for example, wind speed and rain rate—from the noise of shipping, from other anthropogenic activities, and from the natural sounds of marine animals. After classification, quantitative measurements of wind speed and precipitation are applied to the ambient sound data. Comparisons of acoustic quantitative measurements of wind speed with in situ buoy anemometer measurements were shown to be within 0.5 m s−1. The rainfall detection and quantification was also confirmed with collocated measurements of precipitation from a nearby coastal rain gauge and operational weather radar rainfall observations. The complicated condition of high sea states, including the influence of ambient bubble clouds, rain, and sea spray under high winds, was sorted acoustically, and shows promise for identifying and quantifying such conditions from underwater sound measurements. Long-term data were used in this study to derive sound budgets showing the percent occurrence of dominant sound sources (ships, marine mammals, wind, and rain), their relative intensity as a function of frequency, and statistical summaries of the retrieved rainfall amounts and wind speeds at the two buoy locations in the Aegean and Ionian Seas.

Corresponding author address: Dr. Jeffrey Nystuen, Applied Physics Laboratory, University of Washington, 1013 NE 40th Street, Seattle, WA 98105. E-mail: nystuen@apl.washington.edu

Abstract

The Hellenic Center for Marine Research POSEIDON ocean monitoring and forecasting system has included passive underwater acoustic measurements as part of its real-time operations. Specifically, low-duty-cycle long-term passive acoustic listeners (PALs) are deployed on two operational buoys, one off Pylos in the Ionian Sea and the second off Athos in the northern Aegean Sea. The first step toward the quantitative use of passive ambient sound is the classification of the geophysical sources—for example, wind speed and rain rate—from the noise of shipping, from other anthropogenic activities, and from the natural sounds of marine animals. After classification, quantitative measurements of wind speed and precipitation are applied to the ambient sound data. Comparisons of acoustic quantitative measurements of wind speed with in situ buoy anemometer measurements were shown to be within 0.5 m s−1. The rainfall detection and quantification was also confirmed with collocated measurements of precipitation from a nearby coastal rain gauge and operational weather radar rainfall observations. The complicated condition of high sea states, including the influence of ambient bubble clouds, rain, and sea spray under high winds, was sorted acoustically, and shows promise for identifying and quantifying such conditions from underwater sound measurements. Long-term data were used in this study to derive sound budgets showing the percent occurrence of dominant sound sources (ships, marine mammals, wind, and rain), their relative intensity as a function of frequency, and statistical summaries of the retrieved rainfall amounts and wind speeds at the two buoy locations in the Aegean and Ionian Seas.

Corresponding author address: Dr. Jeffrey Nystuen, Applied Physics Laboratory, University of Washington, 1013 NE 40th Street, Seattle, WA 98105. E-mail: nystuen@apl.washington.edu

1. Introduction

Underwater ambient sound contains quantifiable information about the sound source. In the frequency band from 500 to 50 kHz, dominant geophysical sources of underwater sound include the sound of breaking waves and rainfall. Usually the source of breaking waves is the local wind, allowing the amplitude of the ambient sound due to breaking waves to be closely correlated with wind speed (Vagle et al. 1990). The other dominant geophysical sound source is rainfall from raindrops splashing the ocean surface. This signal can be quantified to monitor rainfall rate (Nystuen 2001; Ma and Nystuen 2005; Nystuen et al. 2008). Other sound sources in the marine environment include geophysical sources (Anagnostou et al. 2011; Tsabaris et al. 2011) such as earthquake and volcanoes; biological sources (Miksis-Olds et al. 2010) such as cetaceans, pinnipeds, fish, and crustaceans; and anthropogenic noises (Hildebrand 2009; Andrew et al. 2002) such as shipping, industrial noise, and sonars. These sounds must be correctly identified and classified in the sound spectra prior to the application of quantitative algorithms for the measurement of geophysical quantities such as rainfall rate and wind speed.

As with other passive remote sensing techniques, the underwater acoustic measurement does not interfere with the process being monitored. Contrary to the surface-based sensors, the underwater sensor is away from the harsh environment of the air–sea interface. This has several advantages relative to surface-based observations. Because of the underwater deployment, the likelihood of vandalism, a surprisingly big problem for surface instrumentation even in remote ocean locations, is reduced. Because the measurement is passive, no potentially harmful sound is introduced into the marine environment. Furthermore, underwater observations do not suffer from any natural hazard (e.g., strong winds, lightning) errors typically introduced to observations from meteorological sensors (i.e., anemometers, rain gauges, etc.) installed on surface buoys. On the other hand, as in any remote sensing technique, sound source and quantification estimates from underwater acoustic data require validation prior to their use in marine applications.

Very few studies exist that provide validation data for acoustic measurements of geophysical parameters. In a past field experiment [Ionian Sea Rainfall Experiment (ISREX)] passive aquatic listeners (PALs) were deployed at the edge of the deep Ionian Sea west of the coastal city of Pylos to evaluate the retrieval of rainfall from underwater sound measurements (Nystuen et al. 2008). PALs are adaptive low-duty-cycle recorders designed to monitor the background sound field in the marine environment, and to detect acoustic events, such as rainstorms (Ma and Nystuen 2005), ship passages, and calling bouts from marine animals (Miksis-Olds et al. 2010). Four PALs were deployed in that experiment at depths of 60, 200, 1000, and 2000 m on a single mooring line (Nystuen et al. 2008). PAL rainfall measurements from the various depths were compared against high-resolution rainfall estimates from a mobile high-resolution dual-polarization X-band radar deployed at close range (17 km east of the mooring) based on six precipitation events that were captured during the field experiment. The studies based on these data (Anagnostou et al. 2008; Nystuen et al. 2008) demonstrated that the signal from rainfall is easily detected under a variety of wind and noise conditions, and that the signal is robust with respect to depth. A spatial averaging of the signal with depth was verified by comparing the acoustic rainfall-rate measurements to radar-derived rainfall rate averaged over different length scales (averaging areas). These results demonstrated that the length scale of the acoustic rainfall measurement is dependent on the depth of the listening instrument. Data from that experiment also produced a relationship between the sound pressure level (S; dB rel. 1 μPa2 Hz−1; hereinafter, this unit applies to all sound pressure levels) and the radar reflectivity (Ze; mm6 m−3) (Amitai et al. 2007). This relationship is more fundamental than the traditional relationship between rainfall rate (R; mm h−1) and Ze because the uncertainties associated with the nonlinear conversion of the sound to rain rate and reflectivity to rain rate is excluded. In general the authors confirmed prior results that were obtained from an experiment in shallow waters (Amitai et al. 2004a,b) indicating a very high S–Ze correlation even from the deepest (2000 m) PAL deployment.

The results from ISREX, however, were limited in terms of storm cases examined, which motivated the current study aimed to collect continuous and longer-term acoustic data by deploying in November 2008 two PALs (Anagnostou et al. 2011) at two buoy locations of the Hellenic Center for Marine Research real time operational ocean observation system, named POSEIDON, (Nittis et al. 2002, http://www.poseidon.hcmr.gr): one near Pylos at the edge of the Ionian Sea, and the second near Athos island, in a deep basin of the northern Aegean Sea (Fig. 1).

Fig. 1.
Fig. 1.

Locations of POSEIDON moorings with PALs (Pylos and Athos), weather radar location (Andravida), and rain gauge sites (Aktio and Methoni).

Citation: Journal of Atmospheric and Oceanic Technology 32, 2; 10.1175/JTECH-D-13-00264.1

The novelty of this field observation program is the integration for the first time of a real-time processing embedded software to the PAL system for the continuous acquisition, classification, and quantification of geophysical (wind and rain), biological, and synthetic sound sources, including the generation of noise budgets using the POSEIDON oceanographic observation system (Anagnostou et al. 2011). Similar efforts are led by the Laboratory of Applied Bioacoustics (LAB) of the Technical University of Catalonia [Universitat Politecnica de Cataluya, Barcelona Tech (UPC)] of an international program entitled “Listen to the Deep Ocean Environment (LIDO)” to monitor geophysical, biological, and human-generated sounds using passive acoustic systems mounted on integrated cabled deep sea platforms and moored stations (http://listentothedeep.com/acoustics/index.html). Other underwater passive acoustic systems focusing mainly on the near-real-time monitoring of biological and natural hazards (i.e., earthquakes, volcano eruptions, etc.) include the ALOHA observatory system, operated by the University of Hawai‘i at Mānoa (http://aco-ssds.soest.hawaii.edu/); the Perennial Acoustic Observatory in the Antarctic Ocean (PALAOA), operated by the Alfred Wegener Institute of Germany (http://dosits.org/technology/observermarineanimals/realtimepassiveacousticsensors/); and the passive acoustic system of the Integrated Marine Observing System (IMOS), led by the University of Tasmania on behalf of the Australian marine and climate science community (http://imos.org.au/home.html).

In this study we use the continuous and long-term data collected by the two POSEIDON PALs in the Aegean and Ionian Seas, aimed to evaluate the acoustic measurements in terms of their accuracy in the quantification of wind speed and rainfall rates and the detection of shipping and cetacean presence as a potential online product for operational ocean observation systems, such as POSEIDON. The error analysis in this study is based on surface wind measurements from the POSEIDON buoy anemometers and rainfall estimates from an operational weather radar and nearby coastal rain gauge measurements. In sections 2 and sections 3, we describe the observational setup, followed by a description of PAL instrumentation and sampling strategy. In sections 4 we discuss the acoustic sound source classification analysis and processing needed for geophysical quantification of wind speed and rainfall rate. In sections 5 we generate sound budgets and statistical summaries of retrieved rainfall amounts and wind speeds. The summary and conclusions are given in sections 6.

2. Experimental setup

Two PALs were deployed as part of the POSEIDON system. One PAL is at Pylos at 500-m depth at 36.8°N, 21.6°E (Fig. 1). The data reported here were collected from November 2008 to March 2009. A second PAL is on the POSEIDON mooring near Athos (40.0°N, 24.7°E), in the northern Aegean Sea (Fig. 1). The Athos PAL is attached at 200-m depth, and the data are from November 2008 to March 2009. Both moorings report wind speed and direction from surface-mounted anemometers. The Pylos mooring is within the Hellenic operational radar network coverage from the Andravida weather radar (37°55′N, 21°17′E) operated by the Hellenic National Meteorological Service (HNMS). It is a single-polarization C-band weather radar with operational range at 150 km in the high-resolution (150 m) mode and 300 km at the surveillance low-resolution mode (500m). The scanning strategy of the Hellenic Weather Radar Network includes the high-resolution mode that requires 10 min to complete a volume scan of 15 elevation angles and every hour one surveillance low-resolution scan that requires 5 min to complete four different elevations. The radar observations consist of horizontal-polarization reflectivity (ZH) and Doppler velocity measurements. At the range of the Pylos mooring (125 km), the lowest radar beam is 2 km above the sea surface. The radar rainfall applied in this study is the standard power-law reflectivity (ZH) to rainfall rain (R) relationship calibrated using long-term rain gauge rainfall observations from the Aktio (38°58′N, 20°46′E) meteorological station, located at roughly the same range as the Pylos mooring but in the northerly direction, and the Methoni (36°50′N, 21°42′E) station, which is near (15 km range) the Pylos mooring. Prior to rainfall estimations, the radar data were corrected for the atmospheric vertical profile of reflectivity (VPR) effect and for mean-field bias through comparison with daily rain gauge rainfall accumulations at the Methoni meteorological station.

In the following section, we provide information on the PAL system deployed in the POSEIDON buoys, including a description of the instrument and its communication with the buoy, sampling strategy, and examples of detection and data quality control.

3. Passive aquatic listeners

a. Instrument description

PALs are cylindrical instruments 80 cm long × 15 cm in diameter. The weight in water is about 5 kg. PALs are autonomous underwater acoustic recorders with an internal battery designed to attach to any ocean moorings. The hydrophone extends from one end and it consists of a broadband, low-noise hydrophone; a low-power microprocessor with a 100-kHz analog-to-digital (A/D) digitizer; a memory card for internal storage; and a battery pack. An important feature of the microprocessor is a very low-power “sleep” mode to save energy between acoustic samples, allowing autonomous deployment for up to one year. On the Pylos mooring, an inductive modem link to the surface has been implemented, allowing real-time transmission (every 3 h) of the PAL data to shore from the surface buoy via a satellite link (Anagnostou et al. 2011).

b. Sampling strategy

A PAL is an acoustic sampler designed to record data for up to one year at sea. A single sample of PAL data consists of a 4.5-s time series collected at 100 kHz. This time series is spectrally processed to obtain a 64-frequency bin spectrum with a frequency resolution of 200 Hz from 100 to 3000 Hz and 1 kHz from 3 to 50 kHz. These spectra are evaluated individually to determine the acoustic source and then are recorded internally. Thus, the standard dataset is a time series of spectra, or spectral components, such as the sound levels at chosen frequencies, or the slopes of the sound spectra between chosen frequencies. To achieve yearlong deployments, the PAL is designed to enter a low power mode (sleep mode) between each data collection sequences. The time interval between data collection sequences is variable, depending on the acoustic source detected. Typically, the sleep interval is set to 5–10 min, with the sampling interval changing to 1 min during rain events or if an unusual noise is detected. In this experiment the sleep interval between nonevent samples was set to 5 min. This sample interval was changed to 1 min when loud noises such as rain, ships, or whale calls were detected. When the system detects drizzle or wind speed (which is the dominant sound source in our spectrum time series), the “sleep time” interval changes to 2 and 5 min, respectively. These sampling intervals were chosen to be frequent enough so that if a pod of whales or dolphins stays in the vicinity of the PAL for tens of minutes, detection is likely to occur (Miksis-Olds et al. 2010). The PAL acoustic sensor is able to distinguish each source by means of a detection algorithm (discussed in the following section) based on its acoustic signatures.

The original 4.5-s time series for each data collection sequence is discarded. However, these data can be used to provide an audio confirmation of the sound source identification. The PAL has an automated option to save selected samples if they contain interesting signals. Practically, these “sound bite” files are relatively large, about 1 Mb each, and cannot be transmitted in real time to shore. In fact, only a limited number of sound bites can be stored on a PAL during a typical deployment. These data are available after recovery of the PAL and are especially useful for verifying the detection of the calls from cetaceans, and the presence of ship noise or other interesting signals in the underwater soundscape.

c. Examples of detection

An example of a single data sample is shown in Fig. 2. This example shows a uniform (stationary) time series. Each of the eight subsamples reports the same spectrum, to within a user-defined tolerance, typically 12 dB, at all frequency bands. When this criterion is satisfied, an average spectrum is computed and stored. The original temporal sample is discarded. The average spectrum is evaluated to obtain a background sound source (wind, drizzle, rain, shipping, whale, noise). The adaptive sample interval (time to next sample) is then set based on this identification, and allows for a PAL to change the sampling strategy based on the sound source present.

Fig. 2.
Fig. 2.

Example taken from a single sample of the PAL sound bite data. (top) Time series of 4.5 s length, sampled at 100 kHz. Eight subsamples (each 10.24 ms) are evaluated to determine if the sample is stationary. If it is stationary, then the mean spectrum is saved and the original time series is discarded. (bottom) The eight selected subsamples spectra are shown with the mean spectrum (thick black line). This is the acoustic signature classified from the algorithm as drizzle in low (~3 m s−1) wind conditions.

Citation: Journal of Atmospheric and Oceanic Technology 32, 2; 10.1175/JTECH-D-13-00264.1

A second sample is shown in Fig. 3 containing transient sounds, that is, sounds lasting less than 4.5 s. In this case, it is echolocation clicking from a cetacean, probably a striped dolphin, and a whistle. If the transient sound occurs during one of the eight subsamples and it produces a spectrum that deviates from the mean spectrum (of all eight subsamples) by more than the user-specified tolerance, then a transient sound is detected. This detection can be tuned to particular frequency bands in order to detect specific sources, such as the 30–40-kHz echo location click of beaked whales or striped dolphins (Fig. 3), or the frequency band of a sonar, etc.

Fig. 3.
Fig. 3.

Example of a nonstationary sample taken from the PAL’s sound bites. (top) Time series of pressure (Pa) over 4.5 s with a sample rate of 100 kHz. (middle) FFT of (top) indicating high-frequency echolocation clicking and then a whistle from a cetacean in the northern Aegean Sea are recorded. (bottom) Spectra for eight subsamples at the times shown in (top).

Citation: Journal of Atmospheric and Oceanic Technology 32, 2; 10.1175/JTECH-D-13-00264.1

Once a transient is detected, a decision is made whether to save the original temporal sample. This is mainly a matter of storage space on the memory card, currently limited to 2 Gb. This allows only 2200 original samples to be stored during a deployment. These time series are referred to as sound bites. A rationing scheme has been established to assure that a daily quota of sound bites can be allowed. If many transients are detected early in a given day, then a 24-h coverage is not maintained. However, regardless of the decision to save the original temporal sample, the spectral components of the eight subsamples are saved to the spectral time series data file, assuring that the time series of spectral components does provide full daily coverage.

d. Acoustic data quality control

To prevent aliasing from high frequencies (over 50 kHz) and saturation from lower frequencies (less than 20 Hz), electronic bandpass filters are present in the PAL electronics. These are simple resistor–capacitor (RC) amplifier filters with half-power low- and high-pass frequencies at 300 and 40 000 Hz, respectively. At very low sound conditions, electronic spikes in the spectra due to electrical interference from the microprocessor are apparent. These are true noise, and they prevent low sound levels at these frequencies from being reported. Fortunately, these data spikes are confined to narrow frequency bands and the data can be presented with these frequency bands removed.

There is also sensitivity offset and a residual frequency-dependent sensitivity correction associated with each PAL sensor (the hydrophone sensitivity). A nominal offset is user-specified during the deployment, but the exact offset is calculated in postdeployment analysis using the surface-mounted anemometer on the mooring itself. It is known that the sound level is very highly correlated with the wind speed. The surface-mounted buoy anemometer data are used to identify periods of uniform wind (several hours of constant wind speed at different wind speeds). Spectral data from each of these time periods are obtained, and the sound level at 8 kHz is adjusted to minimize the difference between the observed anemometer wind speed and the acoustic wind speed. This value is the sensitivity offset for the particular PAL sensor.

The frequency-dependent component of the offset is obtained from the assumption that wind-generated spectra have known uniform spectral slopes for moderate wind speed conditions from roughly 1 to 40 kHz (Vagle et al. 1990; Ma and Nystuen 2005). Thus, uniform slope spectral fits are obtained for moderate wind speed conditions (4–10 m s−1), and deviations from this spectral fit are assumed to be the frequency-dependent component of the spectral sensitivity of the PAL. This is “validated” by noting that the application of this sensitivity correction to other nonwind sound sources, such as rainfall or shipping, removes the same frequency-dependent features from the recorded spectra. In other words, it is assumed that different sound sources do not share the same frequency-dependent spectral features. Note that this correction is small (usually less than 1 dB) relative to the variability of the sound levels (tens of dB).

4. Results

The results presented here are from postprocessing of the raw recordings (i.e., spectrum and sound bites) retrieved from the memory disk of the PAL after the recovery of the instrument from the two POSEIDON buoy moorings. However, the same processing procedure applies in the real-time processing system. Figure 4 shows a typical time series record (of 20 days) of the total sound level of ambient sound in the marine environment of the Athos site. Different processes producing sound have different time scales, and so the time record can be thought of as a series of acoustic events imbedded in a background sound field. In the frequency range from a few hundred hertz to tens of kilohertz, most of the biological and anthropogenic sounds have very short time scales, or on the order of seconds, whereas geophysical processes producing sound, such as wind, rain, and sea ice (Dushaw et al. 2010; Gavrilov et al. 2009), have time scales of minutes to hours. Below, we describe the algorithmic steps and discuss the verification results.

Fig. 4.
Fig. 4.

(top) Example of a 20-day time series plot of the PAL ambient total sound records (gray dots) taken from Athos in the northern Aegean Sea indicating a slowly varying background sound that is punctuated by shorter duration acoustic events, such as rainstorms (dots), ship passages (stars), and whale vocalizations (x). (bottom) Geophysical interpretation of the sound record, which is the desired product for many users. Classification includes the detection and removal of noises, such as ships, and then the subsequent quantification of the physical processes producing the sound. Note that after day 358, the surface anemometer on this mooring failed, but the acoustical wind speed measurement continued throughout the duration of the deployment.

Citation: Journal of Atmospheric and Oceanic Technology 32, 2; 10.1175/JTECH-D-13-00264.1

a. Classification algorithm

A critical component of the quantitative use of the ambient sound field to describe the marine environment is the identification of the sound source. This analysis depends on the assumption that different sound sources have unique spectral characteristics that allow identification through multivariate analysis of spectral parameters, such as sound levels at various frequencies and spectral slopes in various frequency bands. This process is illustrated in Fig. 5. Different combinations of spectral parameters have been plotted with respect to one another for a 20-day period at the Athos PAL mooring. Several categories of sound sources are identified in their generic position on each subplot. These include geophysical (wind, drizzle, heavier rain, and very heavy convective rain), biological (sound sources generated from any marine life), and two or three categories of anthropogenic (close ships, distance ships, and distance ships noise during very calm sea conditions) sound sources. These classification sound sources are analyzed next.

Fig. 5.
Fig. 5.

Scatterplots of acoustic parameters using (a) 2 vs 20 kHz, (b) 8 vs 20 kHz, (c) 8 kHz vs 2–8 kHz slope, and (d) 8 kHz vs 8–15 kHz slope. Data are shown from the Athos mooring PAL for days 350–370 of 2008, with the wind sound source data highlighted. The units are dB relative 1 μPa2 Hz−1 for sound level and dB decade−1 for spectral slope.

Citation: Journal of Atmospheric and Oceanic Technology 32, 2; 10.1175/JTECH-D-13-00264.1

1) Geophysically generated sound sources

In nature the wind is always blowing apart from short time exceptional calm conditions (wind speed less than 3 m s−1). The sound source for wind is wind-caused breaking waves. However, in cases of calm conditions that do not produce waves, the wind is not acoustically detectable. This implies a physical lower limit for the acoustic wind speed algorithm of about 3 m s−1 (Vagle et al. 1990). In addition, PAL can detect sound produced by the raindrops splashing on the surface of the ocean and each kind of rainfall presents a unique spectrum. Wind-generated locus highlighted in Fig. 5 can be thought of as a continuous locus of data spanning a range of wind speeds from 3 to 15 m s−1. The combination of 8- and 20-kHz sound levels (Fig. 5b) is most useful for separating a wind source from precipitation, but a shipping signal is often difficult to be separated and identified from the wind/rain source. Therefore, the shipping signal is better detected using a combination of the 2- and 20-kHz sound levels (Fig. 5a), although there is still ambiguity between heavy rain and loud close ships. Comparing the spectral slope from 2 to 8 kHz with the sound level at 8 kHz (Fig. 5c) allows heavy rain and close ships sources to be identified, while using the slope between 8 and 15 kHz (Fig. 5d) easily identifies drizzle sound sources. However, the drizzle signal can be suppressed in cases of high wind, increasing uncertainties in this classification.

Under especially calm conditions (i.e., wind speeds < 3 m s−1), the sound level at 20 kHz can be very low in intensity (i.e., <30 dB), and the sound intensity at 2–5 kHz can be relatively high. This becomes a generic background spectrum, with individual ship “events” not necessarily evident. This situation (dead calm with distant shipping) is detected as a locus of points during a comparison between the 2-kHz and 20-kHz sound levels (Fig. 5a) with SPL20 < 30 dB and SPL2 > 45 dB, generating an interesting “foot” on the scatter diagrams. As the wind speed increases above 3 m s−1, this feature quickly disappears. The distant sound that was being detected can no longer be heard, and the locally generated 2–5-kHz sound from the wind waves is less than the distant background levels. The sound levels at 2–10 kHz can actually decrease, although the wind is increasing. Apparently the distant 2–5-kHz sound is no longer propagating, probably due to the small-scale surface roughness attenuating (scattering) sound in this frequency band. The dominant feature of wind is clearly visible in the frequency band 2–10 kHz, confirming the possibility of using the sound level at 8 kHz for quantitatively estimating the wind speed. From these diagrams (Fig. 5 and 6), multivariate classification tests are established (see the appendix).

Fig. 6.
Fig. 6.

Data points from the Athos PAL during days 350–370 of 2008 (15 Dec 2008–4 Jan 2009) showing the classification of the clicks (black x) from the wind speed (gray dots). Shown is cetacean (probably striped dolphin) detection using a comparison of the sound level at 20 kHz and at 30 kHz. Echolocation clicks have a broadband peak at 30 kHz (see Fig. 3).

Citation: Journal of Atmospheric and Oceanic Technology 32, 2; 10.1175/JTECH-D-13-00264.1

2) Biologically generated sound sources

Another signal present in underwater ambient sound data is the vocalizations of marine animals, especially cetaceans, as they communicate and forage for food. In particular, striped dolphins, common dolphins, and deep-diving beaked whales generate directional high-frequency clicks, typically at 35–40 kHz. Spectra consistent with this activity are present in the PAL data at both Athos and Pylos, both areas where these animals are known to be present. High-frequency clicks and marine mammal calls can be identified by examining acoustic levels at 20 and 30 kHz (Fig. 6). These samples can be found in cases where there are more intensive sound levels at 30 kHz relative to 20 kHz (consistent with a spectral peak at 35 kHz). Data from other sound sources (i.e., wind, rain, shipping, etc.) can be found on a different locus of points of the scatterplot.

To confirm the sound source, the recorded original time series samples (sound bites) were reviewed subjectively (Fig. 7). Sound bites containing recordings of clicking and calls associated with dolphins were heard and indicated. Other sound bites contained shipping noise or background geophysical noise, such as drizzle. Comparisons indicate that whenever sound bites contain clicks, the spectral classification algorithm detects those sounds.

Fig. 7.
Fig. 7.

Time series plot showing the detection of high-frequency (~40 kHz) clicks in the PAL data. Clicking is subjectively confirmed by examining the saved original time series [sound bites (SB)] samples that contained transient sounds. The time of all SB recorded are shown on the bottom of the figure (circles). SBs (diamond) containing high-frequency clicks (see Fig. 3) confirm spectral click detection (cetaceans present). SBs (square) containing ship noise are also marked with loud sonar detected on day 339.

Citation: Journal of Atmospheric and Oceanic Technology 32, 2; 10.1175/JTECH-D-13-00264.1

3) Anthropogenic-generated sound sources

For geophysical and biological applications, the sound from shipping is noise. The detection of shipping noise is important, so that quantitative interpretation of the data is not falsely contaminated with such anthropogenic noise. In fact, anthropogenic noise comes in all forms, and there are no unique spectral features that define it. However, there are generalizations that can be made. Often shipping noise has more low-frequency components, resulting in relatively higher sound levels below 2 kHz than, for example, wind or rain. This allows shipping detection based on the slope of the spectrum from 2 to 8 kHz (Fig. 5c). Two types of shipping noise are recognized here: close ship passages and distance shipping detection. The close ship passage is generally a very loud event with a short time scale (Fig. 8a). The sound levels at all frequencies are high, with the lower frequencies persisting longer, as low-frequency sound is less attenuated in water. Thus, the first and last detection of a close ship passage is at the lower frequencies. This is also true for a “distant” ship passage. Again, the lower-frequency sound propagates over long ranges (up to tens of km), while the higher frequencies are absorbed. There is often no signal at all at the higher (>20 kHz) frequencies (Fig. 8b). Of course, the distant shipping signal is detected most often during low sea-state conditions, when the background geophysical sound levels are lowest.

Fig. 8.
Fig. 8.

Temporal records of ship passages of (a) close ship at 0750 UTC on day 377 (11 Jan 2009) and (b) distant ship at 2240 UTC on day 359 (24 Dec 2008).

Citation: Journal of Atmospheric and Oceanic Technology 32, 2; 10.1175/JTECH-D-13-00264.1

b. Quantification

Once the sound source categories are identified, quantitative algorithms for wind speed and rainfall-rate estimation can be applied. The quantitative wind speed algorithm from Vagle et al. (1990) uses the sound level at 8 kHz:
e1
where U is wind speed (m s−1) and SPL8 is the sound level at 8 kHz. This relationship is valid for 3.3 < U < 15 m s−1. Above 15 m s−1, the effect of bubble clouds absorbing sound causes the signal at 8 kHz to begin to underestimate wind speed. For wind conditions less than 2 m s−1, there is no wave breaking and thus no acoustic signal to measure the wind speed. Indeed, the algorithm has a minimum value of 2.2 m s−1. Identifying “calm” conditions acoustically indicates a wind speed of less than 3 m s−1, rather than a wind speed with an error bar of ±0.5 m s−1. To evaluate extending a quantitative wind speed algorithm to a wider wind speed range, the sound levels at 8 kHz were compared to the surface anemometer (Fig. 9) taken from high-temporal-resolution observations of the Pylos site. It is observed that there is a bias and offset at the low wind speeds (below 4 m s−1). A third-order fit to the data reduces the bias and offset. This new quantitative wind speed relationship is given by
e2
where U and SPL8 are as defined above, and the coefficients (a0, a1, a2, and a3) are (0.0005; −0.0310; 0.4904; 2.0871). The new algorithm is applied to the Pylos data and compared to the surface buoy anemometer data in Fig. 10. In this figure, the buoy anemometer data are averaged to a 3-h period and compared to the acoustic data averaged over the same time intervals. The correlation between acoustic wind estimates and the surface buoy anemometer measurements is high (r = 0.94 for wind speeds greater than 2 m s−1). Discrepancies can be attributed to the sampling differences of the two instruments. In fact, the buoy anemometer represents a point measurement, while PAL sound measurements at 500-m depth represent wind speed conditions over a ~8 km2 area around the mooring location. Other sources include environmental factors (e.g., wave height) affecting the relationship between sound and wind speed, and also random errors in the buoy anemometer measurements.
Fig. 9.
Fig. 9.

Comparison of buoy anemometer wind speed and acoustic wind speeds using Vagle et al.’s (1990) algorithm [Eq. (1)] and the new third-order fit [Eq. (2)].

Citation: Journal of Atmospheric and Oceanic Technology 32, 2; 10.1175/JTECH-D-13-00264.1

Fig. 10.
Fig. 10.

Comparison of 3-h averages of wind speed as measured by the Pylos PAL and the surface-mounted anemometer on the mooring.

Citation: Journal of Atmospheric and Oceanic Technology 32, 2; 10.1175/JTECH-D-13-00264.1

Three categories of precipitation can be recognized acoustically: “drizzle,” or rainfall containing only small droplets (less than 1.5-mm diameter); “rain,” heavier rain containing larger drops (over 2-mm diameter); and “heavy convective rain,” containing many large and very large drops (over 3-mm diameter) (Nystuen 2001; Medwin et al. 1992). The different drop sizes produce bubbles during their splashes at the sea surface, resulting in sound production at the resonant frequency of the bubbles. Because the bubble size distribution from raindrop splashes is different than the bubble size distribution from wind waves breaking, the spectral signal from wind and rain are unique from one another and the sound source can be identified. The signal from drizzle contains a distinctive spectral peak from 13 to 25 kHz. Heavier rainfall containing larger droplets generates high acoustic levels from 1 to 10 kHz. Very heavy, convective rainfall often contains very large raindrops (>4 mm) that produce bubbles that resonate from 1 to 2 kHz and produces an ambient bubble layer at the surface that absorbs high-frequency sound (>20 kHz) and shows a steep negative spectral slope in that frequency band (20–50 kHz).

c. Statistical evaluation of the rainfall algorithm

To validate acoustic detection and quantification of rainfall, the PAL data at the Pylos mooring were compared to rainfall data from the Andravida operational weather radar, located at a distance of 125 km. At this range, the lowest beam of the radar reflects off of raindrops at a height of 2 km above the sea surface. Detection of rainfall at this relatively low elevation is an indication of rainfall at the surface. The signal at this height is adjusted to an expected reflectivity at 500-m height based on observed atmospheric structure (VPR), and then corrected for monthly bias by comparison with daily rainfall data from the Methoni weather station, roughly 17 km from the Pylos mooring location. Table 1 summarizes monthly bulk radar rainfall error statistics (i.e., radar-gauge daily rainfall correlation and relative root-mean-square error) evaluated using the Aktio and Methoni rain gauges. Comparison of the radar estimates to the Aktio rain gauge give an overall correlation of 0.75 and a relative RMSE of about 100%, while the correlation and relative RMSE against the Methoni rain gauge observations are about 0.74% and 150%, respectively. Because of the uncertainty in high-temporal-resolution radar rainfall estimates, a quantitative comparison of rainfall is constrained to accumulated rainfall values. Specifically, Fig. 11 shows a comparison of accumulated rainfall values from PAL for February 2009 against coincident estimates by the Andravida weather radar (radar cell over the mooring) and the Methoni weather station (~15 km east of the mooring). The wind speeds from the buoy measurements and PAL estimates are also reported, allowing a qualitative evaluation of the effect of wind speed on the acoustic rainfall estimate. For example, during light wind conditions on 14 February, the PAL overestimated the rainfall accumulation relative to both radar and rain gauge, whereas during high wind conditions on 12 and 19 February, the PAL rain accumulation is low relative to the other rainfall measurements.

Table 1.

Monthly bulk radar rainfall error statistics (correlation and relative root-mean-square error) evaluated using the Aktio and Methoni gauge data.

Table 1.
Fig. 11.
Fig. 11.

Comparison of rainfall accumulation from the Pylos PAL, the Andravida radar, and the Methoni daily rainfall reports during February 2009. Black line indicates time periods when the radar was on, which was during all significant rainfall events at Pylos during February. Also shown is PAL wind speed compared to the mooring anemometer.

Citation: Journal of Atmospheric and Oceanic Technology 32, 2; 10.1175/JTECH-D-13-00264.1

5. Sound budgets

An important aspect of PAL is a long-term recorder designed to be deployed for many months at a time. This allows it to sample the marine environment over different seasons and for long periods. Geophysical analysis of wind and rainfall conditions over time can be monitored, and marine animal or shipping detection can be quantified. Direct measurements of the sound field are made, and can be attributed to different acoustic sources. Thus, the relative loudness or persistence of different sources can be quantified, generating a sound budget. For example, Fig. 12 shows the relative dominance of wind, rain, and distant shipping from November 2008 through March 2009 at the Athos mooring. During the winter, the sound of wind and rain dominate, but during the spring (and the very calm conditions), distant shipping is often detected; however, it is not particularly loud. [Note: close ship passages are relatively loud (Fig. 8a).] Figure 13 shows that this relative loudness (of wind, rain, and distant shipping) is a function of frequency. At 20 kHz, rainfall is clearly the loudest signal, and distant shipping is barely above the background instrument noise. Even at 2 kHz, the mean signal from rainfall is usually dominant (when rainfall is present). During the winter months, the mean signal from wind is louder than from distant shipping, indicating that distant shipping will only be detected during periods of relative quiet (between storms). This statement does not apply to close ship passages, which are easily detected even in high wind conditions (Fig. 8a). In the spring (after April), the sound levels from wind drop significantly and distant shipping detection is prevalent.

Fig. 12.
Fig. 12.

PAL ambient noise budget plot over the Athos’ site showing the relative dominance of the sound sources: wind, rain and shipping (mostly distant shipping). During the winter, high wind speeds make wind the dominant sound source.

Citation: Journal of Atmospheric and Oceanic Technology 32, 2; 10.1175/JTECH-D-13-00264.1

Fig. 13.
Fig. 13.

Mean sound levels for wind, rain, and shipping (distant shipping) at (a) 2 and (b) 20 kHz when that sound source is detected.

Citation: Journal of Atmospheric and Oceanic Technology 32, 2; 10.1175/JTECH-D-13-00264.1

Figure 14 summarizes cetacean detection at Athos, showing that animals are present (vocalizing) 1%–3% of the time, with a higher level of detection during December–February. Of course, they need to be vocalizing to be detected, and so this is the minimum level of presence. On the other hand, visual observations (the traditional means of detecting animals) are difficult or impossible during most winter periods because of high seas and darkness.

Fig. 14.
Fig. 14.

Summary of cetacean detections at Athos site. Vocalizations, mostly high-frequency (30–40 kHz) clicking but also 10-kHz whistling (Fig. 3), are detected.

Citation: Journal of Atmospheric and Oceanic Technology 32, 2; 10.1175/JTECH-D-13-00264.1

6. Conclusions

Underwater passive acoustic monitoring of the marine environment represents a new tool for oceanographers. This is the first time that passive acoustic monitoring of the marine environment is being incorporated into an operational ocean observation system. In this study we showed that quantitative interpretation of the sound field provides measurements of physical conditions at the sea surface, and allows for the monitoring of marine mammal presence and shipping noise, including persistence and loudness. Two low-duty-cycle adaptive sampling passive aquatic listeners (PALs) have been deployed on two POSEIDON ocean observation system moorings operated by the Hellenic Center for Marine Research. Quantitative acoustic measurements of wind speed, detection, and measurement of rainfall and detection of other acoustic events, including bouts of marine mammal vocalization, and of shipping detection are demonstrated using these PALs.

The study used long-term (one year) observations from the POSEIDON PAL deployments in the north Aegean and south Ionian Seas. Analysis indicates a strong correlation of PAL wind measurements with anemometer measurements from both buoy locations. A notable example of the value of underwater sound measurement is that the surface anemometer on one of the moorings failed due to high wind conditions (30 m s−1 reported by PAL), while the acoustical wind speed measurement continued throughout the duration of the deployment. We reported good correlations in rainfall accumulations between PAL, weather radar, and a nearby coastal rain gauge. Sound bites recorded by the PALs during the observation period were used to verify the acoustic classification of marine mammals. Sound bites also contained significant ship noise, which was used to verify the near or distant ship classification.

Sound budgets were generated for the whole observational period to determine the relative dominance of the various sound sources: wind, rain, and shipping. It was shown that for the north Aegean Sea, high wind speeds during winter make wind the dominant sound source, while ship trafficking dominates the sound budget in the spring months when winds are calmer. Rainfall represents less than 10% of the sound budget, with a peak in winter months, which is consistent with the rainfall climatology of the region. We also presented statistics of the sound levels at 2 and 20 kHz for the wind, rain, and shipping, showing louder sounds for rain and wind at 20 kHz relative to shipping, which is associated with low frequencies (<2 kHz). Finally, a summary of cetacean detections from the Athos mooring location was presented, showing a higher presence in the late fall to winter months (1.5%–3% of time).

We are working on several extensions of this study. First, data collection from the PAL deployments in the two POSEIDON buoys is continuing to derive long-term sound budget analysis and rainfall/wind climatologies for the region. A third PAL has been deployed from June 2011 to the present at a spar buoy moored in the Ligurian Sea (43°47.364′N, 9°9.798′E) and is operated by the Institute for Studying Intelligent Automation Systems of the Italian National Research Council. The spar buoy (about 50 m long and 11t) is equipped with a set of meteorological sensors providing rainfall and wind measurements that will be used to quantify the accuracy of the wind and rain retrievals from PAL under various sea and weather conditions. One of our future research objectives is to quantify the influence of wind and rainfall structure on the acoustic quantification of rainfall and to use this information to develop an improved rainfall estimation algorithm. Furthermore, use of raindrop size distribution by polarimetric radar can be used to evaluate the acoustic retrieval of rainfall drop size distribution in open ocean conditions. These studies will be facilitated by ongoing field observations and will be reported in subsequent publications.

Acknowledgments

The authors thank the POSEIDON group of the Hellenic Center for Marine Research for facilitating the PAL deployment at the Athos and Pylos buoys. We also thank Dr. George Poteriadis, Mr. George Alexakis, and Mrs. Artemis Papapetrou of the Hellenic National Meteorological Service for providing the weather radar and rain gauge data. This work was supported by an EU Marie Curie International Incoming Fellowship to Dr. Jeffrey Nystuen (PIIF-GA-2009-237297) and the National Science Foundation, Physical Oceanography Program. The fabrication of PALs used in this study was funded by the Hellenic Center for Marine Research.

APPENDIX

PAL Automated Classification and Rainfall/Wind Quantification Algorithm for Sound Sources in Open Water

This appendix presents an automated classification algorithm for underwater sound sources at midfrequencies (500 Hz to 50 kHz). It is based on Figs 5 and 6. A hierarchy of classifications is applied. First, the shipping and biological categories are determined. Then, these samples are removed and the rainfall categories are determined. Finally, the remaining samples are tested for wind.

a. Shipping categories

  • Close ship passage if SPL8 > 60 dB and SLOPE2–8kHz < −13 dB decade−1
  • Distant shipping if (SPL20 < 0.93 × SPL2 − 16.5 and SPL2 < 63) or (SPL20 < −0.2448 × SPL22 + 31.9 × SPL2 − 996 and 63 dB < SPL2 < 68 dB) and SLOPE2–8kHz < −20 dB decade−1
  • Distant shipping if (SPL2 > 65 dB and SPL2 < 80 dB and SPL20 > −0.09 × SPL22 + 12.18 × SPL2 −363 and SPL20 < SPL2 − 17) and SLOPE2–8kHz < −18 dB decade−1

b. Biological categories

  • Whale echo location clicking if SPL30 > SPL20 and SPL30 > 38 dB

c. Rainfall categories

  • Medium rain (stratiform) if SPL20 > SPL5 × 0.75 + 5 and SPL5 ≤ 70 dB
  • Heavy rain (convective) if SPL8 > 60 dB and SLOPE2–8kHz > −13 dB decade−1 and SPL20 > 45 dB
  • Drizzle if SPL8 < 50 dB and SLOPE8–15kHz > −5 dB decade−1 and SPL20 > 35 dB and SPL20 > SPL8 × 0.9 dB
  • Rain with high winds if (SPL20 > −0.1144 × SPL82 + 12.728 × SPL8 − 307) and (SPL20 < −0.1 × SPL82 + 11.5 × SPL8 −281) and 51 dB < SPL8 < 64 dB and SLOPE2–8kHz > −13 dB decade−1

d. Wind categories

  • Wind if SPL2 < 65 dB and SPL20 < 0.73 × SPL2 + 1.3 and SPL20 > 0.93 × SPL2 − 16.5
  • High winds (with unresolved shipping noise) if SPL2 > 63 dB and (SPL20 > −0.2448 × SPL22 + 31.9 × SPL2 − 996 and SPL2 < 68 dB) and (SPL20 < −0.09 × SPL22 + 12.18 × SPL2 − 363)

e. Quantitative algorithms

Quantitative algorithms for wind speed and rainfall rate are given by the following:

  1. The wind speed algorithm is , where U is wind speed (m s−1), SPL8 is the sound level (dB rel. 1 μPa2 Hz−1) at 8 kHz, and the coefficients (a0, a1, a2, and a3) are (.0005; −0.0310; 0.4904; 2.0871).
  2. The rainfall-rate algorithm is , where R is rainfall rate (mm h−1), SPL5 is the sound level (dB rel. 1 μPa2 Hz−1) at 5 kHz, and the coefficients (b0 and b1) are (.0325; −1.4416) (Nystuen et al. 2008).

REFERENCES

  • Amitai, E., , Nystuen J. A. , , and Liao L. , 2004a: Listening to the rain. Bull. Amer. Meteor. Soc., 85, 15.

  • Amitai, E., , Nystuen J. A. , , Liao L. , , Meneghini R. , , and Morin E. , 2004b: Uniting space, ground, and underwater measurements for improved estimates of rain rate. IEEE Geosci. Remote Sens. Lett., 1, 3538, doi:10.1109/LGRS.2004.824767.

    • Search Google Scholar
    • Export Citation
  • Amitai, E., , Nystuen J. A. , , Anagnostou E. N. , , and Anagnostou M. N. , 2007: Comparison of deep underwater measurements and radar observations of rainfall. IEEE Geosci. Remote Sens. Lett., 4, 406410, doi:10.1109/LGRS.2007.895681.

    • Search Google Scholar
    • Export Citation
  • Anagnostou, M. N., , Nystuen J. A. , , Anagnostou E. N. , , Nikolopoulos E. I. , , and Amitai E. , 2008: Evaluation of underwater rainfall measurements during the Ionian Sea Rainfall Experiment. IEEE Trans. Geosci. Remote Sens., 46, 29362946, doi:10.1109/TGRS.2008.2000756.

    • Search Google Scholar
    • Export Citation
  • Anagnostou, M. N., , Nystuen J. A. , , Anagnostou E. N. , , Papadopoulos A. , , and Lykousis V. , 2011: Passive aquatic listener (PAL): An adoptive underwater acoustic recording system for the marine environment. Nucl. Instrum. Methods Phys. Res., 626–627A, S94–S98, doi:10.1016/j.nima.2010.04.140.

    • Search Google Scholar
    • Export Citation
  • Andrew, R. A., , Howe B. M. , , Mercer J. A. , , and Dzieciuch M. A. , 2002: Ocean ambient sound: Comparing the 1960s with the 1990s for a receiver off the California coast. Acoust. Res. Lett. Online, 3, 6570, doi:10.1121/1.1461915.

    • Search Google Scholar
    • Export Citation
  • Dushaw, B., and et al. , 2010: A global ocean acoustic observing network. Proceedings of OceanObs’09 Conference: Sustained Ocean Observations and Information for Society, J. Hall, D. E. Harrison, and D. Stammer, Eds., Vol. 2, ESA Publ. WPP-306, doi:10.5270/OceanObs09.cwp.25.

  • Gavrilov, A., , McCauley R. , , Thomas F. , , Klaka K. , , Kent C. S. , , Perry M. , , and Duncan A. , 2009: Acoustic observatories of the Australian marine observing system. Proc. Third Int. Conf. and Exhibition on Underwater Acoustic Measurements: Technologies and Results, Nafplion, Greece, IACM/FORTH, 32-3. [Available online at http://promitheas.iacm.forth.gr/UAM_Proceedings/uam2009/32-3.pdf.]

  • Hildebrand, J. A., 2009: Anthropogenic and natural sources of underwater noise in the ocean. Mar. Ecol. Prog. Ser., 395, 520, doi:10.3354/meps08353.

    • Search Google Scholar
    • Export Citation
  • Ma, B. B., , and Nystuen J. A. , 2005: Passive acoustic detection and measurement of rainfall at sea. J. Atmos. Oceanic Technol., 22, 12251248, doi:10.1175/JTECH1773.1.

    • Search Google Scholar
    • Export Citation
  • Medwin, H., , Nystuen J. A. , , Jacobus P. W. , , Ostwald L. H. , , and Synder D. E. , 1992: The anatomy of underwater rain noise. J. Acoust. Soc. Amer., 92, 16131623, doi:10.1121/1.403902.

    • Search Google Scholar
    • Export Citation
  • Miksis-Olds, J. L., , Nystuen J. A. , , and Parks S. E. , 2010: Detecting marine mammals with an adaptive sub-sampling recorder in the Bering Sea. J. Appl. Acoust., 71, 10871092, doi:10.1016/j.apacoust.2010.05.010.

    • Search Google Scholar
    • Export Citation
  • Nittis, K., , Zervakis V. , , Papageorgiou E. , , and Perivoliotis L. , 2002: Atmospheric and oceanic observations from the POSEIDON Buoy Network: Initial results. J. Atmos. Ocean Sci., 8, 137149, doi:10.1080/10236730290004076.

    • Search Google Scholar
    • Export Citation
  • Nystuen, J. A., 2001: Listening to raindrops from underwater: An acoustic disdrometer. J. Atmos. Oceanic Technol., 18, 16401657, doi:10.1175/1520-0426(2001)018<1640:LTRFUA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Nystuen, J. A., , Amitai E. , , Anagnostou E. N. , , and Anagnostou M. N. , 2008: Spatial averaging of oceanic rainfall variability using underwater sound: Ionian Sea Rainfall Experiment 2004. J. Acoust. Soc. Amer., 123, 19521962, doi:10.1121/1.2871485.

    • Search Google Scholar
    • Export Citation
  • Tsabaris, C., and et al. , 2011: A marine groundwater spring in Stoupa, Greece: Shallow water instrumentation comparing radon and ambient sound with discharge rate. Procedia Earth Planet. Sci., 4, 39, doi:10.1016/j.proeps.2011.11.002.

    • Search Google Scholar
    • Export Citation
  • Vagle, S., , Large W. G. , , and Farmer D. M. , 1990: An evaluation of the WOTAN technique for inferring oceanic wind from underwater sound. J. Atmos. Oceanic Technol., 7, 576595, doi:10.1175/1520-0426(1990)007<0576:AEOTWT>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
Save