Gayana 68(2) supl. t.I. Proc. : 266-271, 2004 ISSN 0717-652X

USE OF AMBIENT SOUND MEASUREMENTS IN AN INTEGRATED SYSTEM FOR OCEAN MONITORING

Trevor Guymer, Graham Quartly & Keith Birch

Southampton Oceanography Centre, Empress Dock, Southampton, UK SO14 3ZH email: thg@soc.soton.ac.uk

ABSTRACT

Measurements of underwater sound were made at 16 frequencies in the range 500 Hz ­ 50 kHz using 3 moored hydrophones located off the west coast of Ireland during Oct-Nov 2003. These data have been analysed in relation to environmental conditions, particularly wind speed and rainfall. Although there are some unexplained offsets in the overall acoustic levels compared with those obtained by other researchers, it has proved possible to develop a classification scheme enabling wind, drizzle, rain, and contamination to be separately identified. The acoustic data can be used to infer wind speed to within 2 m s^-1. There is also useful skill in the detection of drizzle and rain events, and some encouraging results in the quantitative estimation of rain-rate have been achieved. Comparisons have been made with deployments of the hydrophones off SW England and Canada. It is found that when precipitation is absent, underwater sound can be predicted from wind speed alone to an accuracy of ±1dB provided the winds are not light. Suggestions are made as to how such a subsurface technique might be combined with spaceborne measurements to enable estimates of underwater sound levels globally. The feasibility of using hydrophones within the Argo float programme should be explored.

INTRODUCTION

Ambient sound measurements are important because, in the absence of wildlife and anthropogenic sources, they provide a record of area-averaged meteorological conditions, such as wind and rain. In particular, such measurements can help constrain estimates of the balance between evaporation and precipitation, which affects the salinity and hence the density of surface waters, and thus the overturning circulation of the oceans. The ability to determine the relative levels of environmental and man-made noise at any frequency is also important for underwater listening systems. Doubt concerning the validity of the methods used operationally for predicting rain and wind noise in sonar performance calculations, means that the subject needs to be put on a firmer scientific footing. A growing issue for consideration, both on the national and international agenda, is the effects of man-made underwater noise on living marine resources, particularly marine mammals. Some conservation organisations have suggested that damage has already been, or is likely to be caused by, active sonar and air-guns. In this context it is desirable to quantify the ambient noise so as (i) to be able to discriminate better between anthropogenic and natural noise and (ii) to be able to reduce the emitted power of sound sources as much as possible whilst still retrieving useful information.

The Acoustic Rain Gauge (ARG) project has been underway at Southampton Oceanography Centre (SOC) since 1995. Early trials were in sheltered Scottish lochs (Quartly et al., 2000), with the first deployment in exposed conditions being off Aberporth in Oct-Nov 2001 (Quartly et al., 2002). Funding under the NERC/MoD Joint Grant Scheme has enabled us to construct the Mark IV ARGs and trial them on the Scotian Shelf (off the east coast of Canada), off Plymouth, and, most recently, off Galway. It is this last deployment that is the subject of this paper. The specific aims were to obtain an extended time series of measurements from several ARGs in a location exposed to the Atlantic westerlies and in the presence of a full suite of ancillary information, including wave data.

METHODOLOGY

Instrumentation

The basic acoustic measurement system is based on a transducer and electronics produced by Metocean and modified by SOC. Acoustic signals are sampled every 90 seconds at 16 frequencies spanning the range 500 Hz to 50 kHz. One of the main aims of the development programme has been to translate the acoustic technology from short-term disposable buoys, to enable long-term deployment with internal recording and Orbcomm satellite data transmission. Details of the technology development programme and the three experiments conducted with the present Mark IV system are given in Quartly et al (2004).

The Galway deployment demonstrated not just the ARGs but a combination of technologies and infrastructure to provide near real-time display of the data via the web. Local ancillary measurements were provided by a meteorological buoy and a WaveRider, both with data transmission via satellite. All the data were archived at SOC, with automatic online display of the latest data, plus an automatic mooring monitoring system using SMS text messaging. Unfortunately, the met. buoy ceased working within a few hours of deployment so for validation we have used data from a routine met. buoy (M1) operated by the Irish weather service and located 110 km WNW of our experiment site. Rain radar data for the period were obtained from the UK Met Office.

Basis of analysis

The basic spectral shape of wind-induced underwater noise is an approximately linear decrease with frequency over the range of interest, with higher wind speeds leading to higher acoustic intensities across the whole frequency range. Two of the key parameters used in the rain detection (spectral slope and d₁₄) are illustrated in Fig. 1. The spectral slope is usually evaluated from 0.8 to 10 kHz but for Galway only 0.8 to 6 kHz was used. (The 8 and 10 kHz channels were ignored, because they suffered from occasional enhanced levels.) An extrapolation to higher frequencies is used to determine d₁₄, the excess observed at 14 kHz above that expected for a linear spectrum; this is used to indicate the presence of drizzle.

An important part of the acoustic processing is the determination of the dominant source. If the sound spectrum is consistent with wind-only or wind and drizzle conditions, then there is a well-established simple inversion of the acoustic intensity at 8 kHz to give the wind speed. We use the equation for U_10n (wind speed at 10 m above the sea surface, assuming neutral stability conditions) as specified by Vagle et al. (1990). If the dominant sound source is believed to be heavy rain, the wind speed retrieval is invalid, however Nystuen et al. (1993) showed the rain rate is simply related to the sound level at 5 kHz. In general, rain is much louder than wind, but a wind speed of 20 m s^-1 will mask the sound of rain below a few mm hr^-1.

Figure 1: Determination of spectral slope using channels 2 to 10, and calculation of the 14 kHz acoustic anomaly (excess above fitted line).

RESULTS

Wind spectra

The simple algorithm of Vagle et al. (1990) relates the observed acoustic intensities at 8 kHz to U_10n. Significant biases between these estimates and direct wind measurements occurred. Figure 2a shows a representative time series for a 3-day period, with the ARG estimates approximately a factor of two greater than the validation records. This is true of all the other Mark IV ARG datasets and is not explained by the separation of the ARG and the met. buoy. An alternative perspective (Fig. 2b) is that the acoustic intensity at 8 kHz is on average 7 dB higher than would be expected for the given wind speed, assuming the relationship detailed in Vagle et al. (1990). The Galway results are compared with previous deployments of the ARGs in Fig. 3. After the 7 dB adjustment of the algorithm of Vagle et al. (1990), it agrees most closely with the data collected over the Scotian Shelf. The Aberporth data (from the Mark III ARGs) follow the same curve, but are a few dB higher. The Plymouth site was found unsuitable for trials as its general sound levels were much higher, exhibited more short-term variability (not shown here), and revealed only a weak correspondence with wind speed. The Galway data, which are our most extensive dataset (in duration and range of conditions experienced), suggest a shallower curve, with higher noise levels at low wind speed and lower values at high wind speed. However, as we were making use of a validation sensor on M1 some 110 km away, there will be a broad range of wind speeds at the ARG location for any given wind speed at M1 such unavoidable smearing of wind speed classes will tend to flatten the curve, so this does not imply a different acoustic response to wind speed at this location compared to the others.

Figure 2: Evaluation of ARG wind records off Galway (Nov. 2003). a) Illustrative 3-day segment. b) Scatter plot of entire deployment acoustic intensities at 8 kHz against wind speed from met buoy (rain-free conditions only).

Figure 3: Relationship of 8 kHz signal to wind speed for various open ocean trials. The curve representing the algorithm of Vagle et al. (1990) has been shifted up by 7 dB.

The data can also be analysed from a different perspective: what are the acoustic intensities to be expected from knowledge of the wind speed? Figure 4 shows the accuracy of acoustic intensity predictions, assuming rain is absent. The observations for Galway have a scatter of 2-3 dB about the mean for a given wind speed, with much larger variability for the highest frequencies. Analysis for Aberporth and Scotian Shelf gave values of ±1 dB for wind speeds ab<ove ~5 m s^-1. It is likely that the Galway signals are harder to predict simply because the wind measurements being used are from 110 km away, although it is possible that other factors, such as whether the wind direction is with or against the prevailing swell, have some effect

Figure 4: Accuracy of acoustic predictions at Galway for each frequency as a function of wind speed. Colour indicates standard deviation for each wind speed (assumed rain-free).

Rain detection and quantification

As mentioned in section 2.2, a simple well-established algorithm exists for inferring rain rate using the acoustic intensity at 5 kHz but the key question is: when can it be applied? If the spectrum shows a significant value for d₁₄, the acoustic excess at 14 kHz, then drizzle is adjudged to be present and the rain rate set to 1 mm hr^-1, as no further increase in accuracy can be achieved. Heavier rain rates are associated with changes in the spectral slope. The spectral slope varies only slightly with wind speed; a threshold of -14.5 dB/decade was chosen to be indicative of heavy rain.

This threshold was then implemented, with the Nystuen et al. (1993) algorithm being used to infer rain rate from the 5 kHz signal. However, once again our recorded sound levels were much larger than envisaged by the rain algorithm; to produce reasonable rain-rates it has been necessary to reduce the measured 5 kHz intensities by 14 dB prior to its application. A section of the resultant time series is shown in Fig. 5 together with rain radar estimates. The wind-only spectra have a slight convex curvature, so the «normal» value for d₁₄ is around -2.5 dB, with values above -1 dB being deemed «drizzle». The existence of a 14 kHz drizzle peak (Fig. 5c) occurs more frequently than a shallow spectral slope (Fig. 5b), which is in keeping with the low rain rates recorded by the rain radar (Fig. 5d). The quality of the comparison was improved by the discarding of spectra that were not quasi-linear between 0.8 and 6 kHz. Such spectra often have major contamination at the lowest frequencies, causing a steep spectral slope, and consequently making the observed 14 kHz signal appear as a positive anomaly. However, acoustic contamination did sometimes cause shallow slopes too. The choice of a detection threshold for drizzle is not easy, as the value of the anomaly in drizzle-free conditions does appear to vary with other meteorological parameters. For example, compare the values for day 322 with those for day 324, which is after a major wind change passed through (Fig. 5a), bringing lengthy periods of light rain (Fig. 5d). This probably also corresponded to a change in wind direction relative to the swell; such details require further investigation.

A subsequent comparison of rain rates from a land-based capacitance rain gauge with the rain radar showed similar levels of disagreement which suggests that an appreciable part of the discrepancy is likely to be due to the use of sensors based on different techniques and having different space-time sampling.

Figure 5 : Example of wind speed, rain and acoustic records during Galway deployment. a) Wind speed (acoustic data corrected by 7 dB beforehand). b) Spectral slope, with red dots when shallower than -14.5 dB/decade. c) Drizzle peak, d₁₄, with magenta dots when greater than -1 dB. d) Rain rate, with acoustic records (dots) when deemed significant. [ÊLight blue indicates when spectrum between 0.8 and 6 kHz is not close to linear, and thus derived slopes and anomalies are invalid for geophysical inversion. ].

DISCUSSION: COMBINING ACOUSTIC AND SATELLITE DATA

The usefulness of our localized field trials can be greatly extended by examining the correspondence between acoustic records and pertinent satellite data, both to use ARGs as validation sensors for the satellite algorithms, and also to relate spaceborne measurements directly to the associated underwater sound levels. There are a number of different satellite sensors that can be used to measure wind and waves. Wind, through its effect on the roughness of the sea surface, changes the reflectivity as perceived by all active microwave systems. It also affects the background thermal emission of water, through such effects as sea surface roughness and the generation of whitecapping. (Wilheit and Chang, 1980).

Similarly, a large number of spaceborne sensors (including infra-red, passive microwave and altimeters) show sensitivity to rain systems, and many have been used to develop algorithms for inferring rain rate (Richards and Arkin, 1981; Wilheit et al., 1991; Ferraro and Marks, 1995; Kummerow et al., 1998; and Quartly et al., 1999).

The data from ARGs and satellites are complementary. Buoy measurements are extremely limited in spatial sampling but can provide time series of acoustic data with a resolution of a few minutes, representative of a surface area of O(10⁴ m²). Satellites provide uniform near-global coverage of quantities related to acoustic noise but, depending on swath width, in a rather sporadic way. There is obviously scope for the in situ measurements to be used in calibrating and validating wind and rainfall estimates derived from satellite sensors. However, there is also the possibility of using the two together to relate underwater noise levels at particular frequencies directly to the signatures measured by the satellite sensors, e.g. radar backscatter or radiometric brightness temperature. (Such approaches are used in meteorology where satellite radiances are assimilated directly into numerical weather prediction models without the intermediate step of conversion to geophysical parameters.) Since this means that coincident meteorological data are no longer required it gives us two strategies for deployment of hydrophones.

ARGs on fixed moorings

It is useful to consider how many moorings would be required to calibrate the satellite wind and rain estimates and to validate the algorithms. They should span the different precipitation regimes from the convectively-dominated tropics, through the large-scale ascent typical of mid-latitude fronts, to the polar latitudes where snowfall may occur. Also, account should be taken of variations in sea-state. For example, in mid-latitudes the eastern sides of ocean basins experience higher mean waveheights than the western sides because of the much longer fetch in the prevailing westerlies. As a minimum, this suggests having 6 or 7 moorings to cover an ocean basin (3 on the eastern side, 3 on the western and perhaps one in the centre. This implies 30-35 for global coverage. Based on our experiences, maintaining such coverage of both acoustic and ancillary instruments on an operational basis would be difficult and expensive and could probably only be envisaged as part of an observing network set up for other purposes. However, it is not obvious that each basin needs to be sampled at the same density. One basin, the North Atlantic say, could have the full suite supplemented by a few moorings at specific sites where it is considered that relationships based on the basin array might not be sufficiently representative.

ARGs on drifting profiling floats

If we relax the constraint of having coincident wind- and rain-measuring systems at the surface, as discussed above, then the possibility emerges of deploying hydrophones on drifting subsurface platforms. This has the advantage of avoiding instrumentation permanently on the ocean surface where it would be vulnerable to storm and shipping damage. The Argo programme (Gould et al., 2004) is building a global array of 3000 floats, providing temperature and salinity profiles of the top 2000 m every 10 days, and which will operate on a sustained basis with near real-time data delivery. The high level of international commitment to Argo and its global coverage are stimulating thought about what other sensors could usefully be added to the basic float platform. Acoustic measurements are one such candidate. Issues requiring consideration are whether acoustic signatures can be acquired while profiling and/or drifting at parking depth, and incorporation of the acoustic measurements into the telemetry stream (this has been a limitation in present float configurations but advances in satellite communications are opening up possibilities for significantly increased volumes of transmitted data). If only 2% of the total Argo array were to be equipped with acoustic sensors an extra 60 platforms would be available for calibration and validation of underwater noise derived from satellite sensors.

CONCLUSIONS AND RECOMMENDATIONS

Our acoustic results show a systematic offset from measurements from previous researchers; its cause remains a matter of investigation but may, in part, be due to errors in calibration, extraneous sound sources and differences in sea-floor characteristics. Nevertheless, it can be possible, by making empirical adjustments, to develop a classification scheme enabling wind, drizzle, rain, and contamination to be separately identified.

Our results from the various deployments show that, when rain is absent, local wind speed can be used to predict underwater sound levels to an accuracy of ±1dB, provided that the wind speed is above ~5 m s^-1. Further, our results show that there is useful skill in the detection of drizzle and rain events and that it may be possible to estimate rain-rate quantitatively to within a factor of two if care is taken over spatio-temporal sampling.

Acoustic measurements on Argo floats could be used for calibrating satellite-derived estimates of underwater sound; this is an area that should be actively pursued. If deployment on floats is successful, this would avoid the necessity of using vulnerable surface moorings to suspend hydrophones and make ancillary wind, sea-state and precipitation measurements.

REFERENCES

Ferraro, R.R. & G.F. Marks, 1995. The development of SSM/I rain-rate retrieval algorithms using ground-based radar measurements. Journal of Atmospheric and Oceanic Technology, 12, 755-770. [ 1]

Gould, J. & 16 authors, 2004. Argo profiling floats bring new era of in situ ocean observations, Eos, 85, 170 & 190-191. [ 2]

Kummerow, C., W. Barnes, T. Kozu, J. Shiue & J. Simpson, 1998. The Tropical Rainfall Measuring Mission (TRMM) Sensor Package. Journal of Atmospheric and Oceanic Technology: Vol. 15, No. 3, pp. 809­817. [ 3]

Nystuen J.A, C.C. McGlothin & M.S. Cook, 1993. The underwater sound generated by heavy rainfall. Journal of Acoustical Society of America, 93, 3169-3177. [ 4]

Quartly, G. D., M.A. Srokosz & T.H. Guymer, 1999. Global precipitation statistics from dual-frequency TOPEX altimetry. Journal of Geophysical Research, 104, 31489-31516. [ 5]

Quartly, G.D., T.H. Guymer, K.G. Birch, J. Smithers, K. Goy & I. Waddington, 2000. Listening for rain: theory and practice. Proceedings of 5th European Conference on Underwater Acoustics, Lyon, France, 11-13 July 2000, Vol. I, 723-728. [ 6]

Quartly, G.D., K.M. Shannon, T.H. Guymer, K.G. Birch & J.M. Campbell, 2002. Testing acoustic rain gauges in inland and coastal waters. Proceedings of the Sixth European Conference on Underwater Acoustics, 24th-27th June 2002, Gdansk, 659-664. [ 7]

Quartly, G.D., T.H. Guymer, K.G. Birch, J.M. Campbell, I. Waddington & M.R. Edwards, 2004. Understanding the acoustic environment: Final report. Southampton Oceanography Centre Research and Consultancy Report, No. 91, 87pp. [ 8]

Richards, F. & P.A. Arkin, 1981. On the relationship between satellite-observed cloud cover and precipitation. Monthly Weather Review, 109, 1081-1093. [ 9]

Vagle S, W.G. Large, & D.M. Farmer, 1990. An evaluation of the WOTAN technique of inferring oceanic winds from underwater ambient sound. Journal of Atmospheric and Oceanic Technology, 7, 576-595. [ 10]

Wilheit, T.T. and A.T.C. Chang, 1980. An algorithm for retrieval of ocean surface and atmospheric parameters from the observations of the scanning multichannel microwave radiometer. Radio Science, 15, 525-544. [ 11]

Wilheit, T.T., A.T.C. Chang & L.S. Chiu, 1991. Retrieval of monthly rainfall indices from microwave radiometric measurements using probability distribution functions. Journal of Atmospheric and Oceanic Technology, 8, 118-136. [ 12]