Acoustic
Environment of the Haro Strait:
Preliminary
propagation modeling and data analysis
Christopher D. Jones,
Michael A. Wolfson
Applied Physics Laboratory
University of Washington
Seattle, WA 98105
Date: April 2006
Abstract: This report presents a preliminary analysis of the acoustic environments of the southern resident killer whales in the Haro Strait of the Puget Sound, combining analysis of field measurements and acoustic propagation modeling for the frequency range 1-10 kHz. The Haro Strait is a highly variable acoustic environment with active commercial shipping, whale watching, and Naval activity. Southern resident killer whales are of unique public concern in this area because of increasing anthropogenic noise levels that may interfere with the animals foraging strategies and behavior. Predictive acoustic modeling in combination with field measurements can be used as a tool for understanding the mechanisms of impact and assessment of the risk, providing a quantitative evaluation of sound source levels in the context of complicated acoustic environments, changing background sound levels, and emerging management issues. Of principle concern in this report is background sound levels created by commercial shipping traffic or other persistent sound sources that propagate from the main shipping channel. The scope of the modeling effort encompasses numerical modeling of transmission loss and propagation at ranges of less than 10 km. Preliminary modeling results are analyzed and compared with recordings of ship noise collected in the spring/summer of 2004.
Contents
1. Introduction
2. Environmental Characterization
2.1. Bathymetry
2.3. Sound Speed Profiles
3. Ship Traffic
3.1. Vessel Tracking Operations Support System (VTOSS)
4. Measurements of Underwater Sound
4.1. Passive Aquatic Listeners (PAL)
4.2. PAL Deployment in the Haro Strait
4.3. Ship Signatures
5. Acoustic Propagation Modeling
5.1. Model Description
5.2. Model Inputs
5.2.1. Geo-acoustic Parameters of the Sea Floor
5.2.2. Sound Speed Profile
5.2.3. Rough Sea Surface and Sea Floor
5.2.4. Monte-Carlo Simulations
5.3. Model Outputs
5.4. Comparison of Model Results with Field Measurements
5.4.1. Single Ship Comparisons
5.4.2. Shipping Lanes
6.2. Estimation of Total Shipping Noise
1. Introduction
The Haro Strait is a complex, shallow water acoustic environment with steep bathymetric relief combined with an active shipping channel, frequent small boat activity, and Naval operations. The western side of San Juan Island is also a primary foraging area for the Southern Resident killer whales[1]. Consequently, these animals are of unique public concern in this area because of the potentially high impact of human activity on their environment. Questions regarding the acoustic environment of these animals have arisen as recreational whale watching, commercial shipping, and Naval activity[2] have grown in this area. A reasonable question to ask in this context is whether increasing underwater noise levels affect the killer whales ability to forage for prey by echo-location. For example, the analysis of the echo-location signals from killer whales[3] indicate that backscattered signal levels from salmon can be very low and comparable in level to natural background noise levels.
This report will address specific aspects of modeling the propagation of sound sources in the Haro Strait, focusing on the numerical estimation of transmission loss in the open channel. In particular, we will investigate the propagation of sound generated by large commercial ship traffic in the Strait and the estimation of sound source levels of individual ships. We will illustrate the role of modeling as a tool for model/data comparisons and the interpretation of field measurements of underwater sound. In this process we will employ a variety of compiled databases of the environment, information on ship traffic and vessel tracking, and field measurements of underwater noise collected recently in the Haro Strait in an area frequented by killer whales.
For purposes of this report, acoustic modeling is used to complement field measurements, as the shallow water environment of the Haro Strait is far too complex, and the geo-acoustic parameters of the area are not characterized well enough to rely on modeling alone. When modeling is constrained by measurements it can provide a useful tool to fill the gaps in measurements in both space and time. For example, measured data will be shown for a specific receiver location and time, and modeling results will be compared with this data to infer source levels of individual large commercial ships. If model results compare favorably and confidence is developed in the modeling strategy for this particular area, then the model may be used to estimate sound pressure levels at other locations in the region where measurements are not available.
Objectives and Scope:
The objective of the this report is to determine the feasibility of modeling the sound propagation environment of the southern resident killer whales in the Haro Strait and to compare initial model results with acoustic measurements taken in June and July 2004. Model/data analysis is limited to data provided by Jeff Nystuen recorded on the PAL system.[4] A significant portion of this effort involves collecting and compiling a database of environmental parameters required for acoustic modeling. The longer-term objectives are to extend these methodologies for model/data analysis by incorporating new acoustic data, more detailed environmental data, and new information on sound sources (e.g. shipping data) as they become available.
The scope of the modeling effort encompasses propagation modeling using readily available methods and codes[5] and the interpretation of existing acoustic data sets. Modeling and data analysis are focused at the frequency of 3.6 kHz, which is representative of the frequency range of 1-10 kHz (within killer whale auditory response). The modeling can be extended to lower frequencies. However, extending the models to higher frequencies (>10 kHz) is problematic due to the sensitivity of the model to uncertainties in the geo-acoustic environment at high spatial scales. Modeling high frequency propagation (>10 kHz) and reverberation is beyond the scope of this study.
The area of interest will be limited to the Haro Strait with propagation ranges less than 10 kilometers. However, the methods can be applied to larger scale studies such as in the coastal ocean or different regions (e.g. beaked whale habitat). We will investigate the effects of canyons and steep walls on forward propagation combined with randomness in the sea surface and the seafloor. We will include the effects of temporal and spatial variability in the environment to model and gain insight on how sound propagation may change as a function of time and location.
Technical Approach:
For the purpose of this report, the acoustic environment is characterized by the propagation loss only. Defined in terms of the standard sonar equation,[6] propagation loss is the amount of signal intensity lost as it propagates from a source to a receiver location. The numerical simulations will provide an estimate of the mean propagation loss between two positions and the variability of the estimate as a function of randomness and uncertainty in the environment. Both the mean and the associated variability (uncertainty bounds) of the estimate are necessary when comparing simulation results with field measurements.
In general, propagation between two locations in the ocean includes both the direct propagation path between a source and a receiver and reverberation. Reverberation is the reflection and scattering of an acoustic signal as a result of its interaction with inhomogeneities and boundaries in the ocean. In this report acoustic propagation modeling is performed using two-dimensional parabolic equation (PE) numerical methods.[7] This type of propagation modeling includes only that component of reverberation in the forward direction, such as forward scattering from the sea surface and bottom. No backscattering is included, which includes the echoes back from a canyon wall, for example.
The application of PE simulations, as typically used in lower frequency, open-ocean modeling, requires special attention when used in shallow water environments. Improper application will likely produce results that will be difficult to compare with field measurements. Modeling issues that are given special attention include:
1) Effects of random roughness at the sea surface and sea bottom that will impact propagation at the frequencies of interest in this study.
2) Analysis of acoustic variability due to such randomness and the definition of uncertainty bounds for the model predictions.
3) High spatial resolution characterization of the geo-acoustic parameters (i.e. sediment properties, high resolution bathymetry, sound speed profiles).
2. Environmental Characterization
2.1 Bathymetry
The bathymetry of Haro Strait is characterized by a relatively deep canyon with a very steep wall at the western coast of San Juan Island. The channel rises to a relatively shallow region on the west side of the channel (Figure 2.1 and 2.2). Since bathymetry is a critical component of understanding acoustic propagation, a high-resolution bathymetry database of Haro Strait has been compiled for the Haro Strait region comprised of data from several sources. The highest resolution bathymetry (to our knowledge) is a recent multi-beam survey conducted by MBML[8] providing partial coverage of the area of interest with a 5 meter grid spacing resolution. The other primary source of lower resolution bathymetric data include NOAA and USGS.[9] These data where combined to provide a continuous bathymetric grid of the region at grid scales up to 5 meters. In practice, 20 meter grid spacing was adequate for modeling.
Figure 2.1 Perspective view of the bathymetry of the Haro Strait.
Figure 2.2 Bathymetry of the Haro Strait, 20 meter contours.
2.2 Geo-Acoustic parameters
The Haro Strait was glacially formed. The steep walls about the west side of San Juan Island are exposed rock. Silt and sand material lie on the bottom of the channel, but because of the strong and variable currents, the thickness of this silty sediment layer will vary temporally and spatially. More precise details of the bottom properties are known for very small sections of Haro Strait as a result of recent geo-acoustic inverse studies[10]. Here the bottom was surveyed using echo sounders, and sediment samples taken with grab samples and cores.
Since the bottom is likely variable and little data is available, we modeled the geo-acoustic environment with three different parameterizations, each within a bound we felt reasonable from the limited amount of geology known and measurements taken. Crudely speaking, the bottom acts as a sink of acoustic energy, with silt and sand absorbing much more sound energy than hard rock. The limiting cases are a thick layer of silt (thick compared to the wavelength of the sound waves) and exposed hard rock. In the case of hard exposed rock with large slopes, such as off the west coast of San Juan Island, backscattering and reverberation can be an important concern, but is beyond the scope of the present study. We also do not include erratic blocks in the geo-acoustic modeling as their population density is expected to be too low.
For the acoustic propagation model the relevant geo-acoustic parameters are 1) the sound speed profile, 2) the density profile of the bottom, and 3) bottom attenuation profile. Since we are mainly concerned with the deep channel of Haro Strait, we assumed a nominal sediment thickness of twenty-five meters and used a critical slope of fourteen degrees to set the bottom type to either a sand-mud-gravel composition or exposed rock. In other words, if the slope of the bottom (determined from the bathymetric data described in Section 2.1) is greater than this critical slope, we assumed the bottom would be scoured by the strong tidal currents so that it would remain as exposed rock. Otherwise the material is assumed to be a layer of sand, gravel, and mud. Tables in Section 5.2 summarize the geo-acoustic parameterization used for the modeling, as discussed further in Section 5.
2.3 Sound Speed profiles
Conductivity, temperature and depth (CTD) data were collected over many years in the Haro Strait region. We obtained this data for the last twenty years from DFO-Canada.[11] There were seventy CTD locations for the months of May and June, the time during which the acoustic data was collected; see Figure 2.3. Of these CTD locations, only locations that were within fifteen kilometers of the receiver location were used to produce sound speed profiles according to the equation of state given by Del Grosso.[12] The resulting sound speed profiles from May and June data taken over a twelve year span in the Haro Strait are shown in Figure 2.4. As seen in Figure 2.4, the sound speed profiles reveal a nearly invariant nature with both geographic location and detph. Seasonal variations exist (not shown), and may be of interest for future studies. Considering the short ranges of propagation (< 10 km) along with the very weak variation in the sound speed environment (< 0.5 %), we feel justified in using a single sound speed profile for our modeling. We chose 1480 m/s for our sound speed in our propagation model (see Section 5.).
Figure 2.3. Map of Haro Strait showing positions where temperature and conductivity data (CTD) were collected.
Figure 2.4. Sound speed profiles derived from CTD casts collected in May and June between the years spanning 1990 to 2002. Only profiles that spanned the full water column are show. The average of these profiles is shown in green.
3. Ship Traffic
3.1 Vessel Tracking System (VTOSS)
The Marine Communications and Traffic Services of the Canadian Coast Guard operate the proprietary Vessel Traffic Operations Support System (VTOSS). VTOSS collects radar signatures of vessels greater than twenty meters length, providing position, course, and speed. The system also collects electronic handoff data from vessels that are approaching within one hour of an exchange line or upon departure for vessels that are berthed one hour from an exchange line. The handoff data include the vessel name, call sign, type, number and type of barges (loaded or empty), port of origin and destination, speed, exchange line time of arrival estimate, and possibly additional information that might be useful to MCTS in regards to safe-guarding vessel traffic. Brian Bain, the Officer in Charge of MCTS Victoria, gave us permission to use data from radar signatures of ships that pass in the vicinity of Haro Strait. Ian Wade (DFO, A/OIC Victoria MCTS Centre) provided us a xbase database file of vessel tracking data that we parced for the data fields we desired. The database file contained a maximum of twenty-eight fields of data for ships operating in the Haro Strait region over the period of May 27 to June 30, 2004 (note May 26, June 9,19, and 20 were not supplied due to software errors).
Typically, only 20 fields held data. Table 3.1 shows the field labels and two examples from two different ships. The first field signifies the date and time the radar signature was recorded; it is stored in the format yyyy mm dd hhmm (year/month/hour/minute), and the signatures are recorded in six minute intervals. Other fields we used were the ship name (field 3. in Table 3.1), the ship latitude and longitude (fields 17. and 18. respectively in Table 3.1), and course and speed (fields 27. and 28. respectively in Table 3.1). The original data file was quite large of order 100 MB, so we separated it into smaller files according to the day. These were then filtered to obtain ship tracks, which were used for our data model comparisons described in Section 5.
The VTOSS database comprises a relatively complete record of large ship traffic in the region. Figure 3.1 illustrates categorization of traffic by ship type during the period of interest (of May 27 to June 30, 2004). Figures 3.2 illustrate the geographic sorting of vessel tracks by type during this same period. Figure 3.3 illustrates an estimate of the mean north and south shipping lanes for commercial cargo ships (not including Tugs).
|
field # |
Field label |
example #1 |
example #2 |
|
1. |
LAST_UDDTG |
"200405170005", |
"200405170005", |
|
2. |
VSL_ID |
"CSTL19931231000495", |
"CSTL19931231000494", |
|
3. |
NAME |
"JACQUES CARTIER", |
"CAPTAIN COOK", |
|
4. |
CALLSIGN |
"CY6103", |
"CY7903", |
|
5. |
LLOYDS_ID |
"0314837", |
"6613483", |
|
6. |
FLAG |
"CA", |
"CA", |
|
7. |
SATCOMNUM |
"A", |
"A", |
|
8. |
TYPE_ENC |
, |
, |
|
9. |
TYPE_DEC |
"TUG", |
TUG", |
|
10. |
LOA |
19.40, |
22.90, |
|
11. |
GRT |
72.00, |
124.00 |
|
12. |
TOW_ENC |
"1HE", |
"1 EMPTY BULK BARGE", |
|
13. |
TOW_DEC |
, |
|
|
14. |
IS_DC |
, |
, |
|
15. |
IS_DD |
, |
, |
|
16. |
IS_SPI |
, |
, |
|
17. |
POS_LAT |
49.42, |
49.36, |
|
18. |
POS_LON |
123.96, |
123.90, |
|
19. |
POS_RDRDTG |
"200405170005", |
"200405170005", |
|
20. |
POS_CIP |
, |
, |
|
21. |
POS_CIPDTG |
, |
, |
|
22. |
POS_SRC |
, |
, |
|
23. |
CVTOSS_ZONE |
"RDR", |
RDR", |
|
24. |
FROM_AT |
"VIC", |
VIC", |
|
25. |
NEXT_TO |
"LAF", |
LAF", |
|
26. |
SERVICE |
"JER", |
MID", |
|
27. |
COURSE |
251.00, |
298.00, |
|
28. |
SPEED |
15.6 |
8.3 |
Table 3.1.
Examples from two lines of the VTOSS data file.
|
SHIP TYPE |
COUNT |
PERCENTAGE |
|
BULK CARRIER |
5090 |
23.6 |
|
TUG |
3195 |
14.8 |
|
CONTAINER SHIP |
2901 |
13.4 |
|
FERRY |
2089 |
9.7 |
|
FISHING VESSEL |
1920 |
8.9 |
|
GENERAL CARGO |
1385 |
6.4 |
|
MISCELLANEOUS |
929 |
4.3 |
|
GOVERNMENT VESSEL |
730 |
3.4 |
|
WARSHIP |
726 |
3.4 |
|
PASSENGER |
494 |
2.3 |
|
CHEMICAL TANKER |
465 |
2.2 |
|
MOTOR YACHT |
433 |
2.0 |
|
SAILING VESSEL |
410 |
1.9 |
|
OCEAN OIL TANKER |
320 |
1.5 |
|
PLEASURE CRAFT |
110 |
0.5 |
|
LOG SHIP |
105 |
0.5 |
|
CHARTER VESSEL |
98 |
0.5 |
|
COASTAL FREIGHTER |
86 |
0.4 |
|
COMBINATION CARRIER (OBO) |
64 |
0.3 |
|
FISH FACTORY |
19 |
0.1 |
|
FISH PROCESSOR |
5 |
0.0 |
Figure 3.1 Vessel types in Haro Strait derived from VTOSS for May 27 to June 30, 2004.
Figure 3.2 Vessel tracks in Haro Strait
derived from VTOSS for May 27 to June 30, 2004.
Figure 3.3. North and south commercial cargo shipping lanes in the Haro Strait obtained from VTOSS data for of May 27 to June 30, 2004.
4. Measurements of Underwater Sound
4.1 Passive Aquatic Listerners (PAL)
Acoustic data used in this report was recorded with the Passive Aquatic Listeners (PAL). PALs are autonomous acoustic recorders designed to be attached to ocean moorings consisting of a broadband, low noise hydrophone, a signal processing board, a low-power microprocessor with a 100 kHz A/D digitizer, a 2 GByte memory card and a 48 Amp-hour battery pack. Physically a PAL is a cylindrical instrument 30 inches long by 6 inches in diameter. The hydrophone extends from one end. It is typically mounted in a cage to avoid damage by possible fishing lines. The weight in water is about 10 lbs, making it deployable on almost any type of mooring line. The new casings are more robust and will increase the weight to about 20 lbs in water.
A PAL is autonomous and depends on internal batteries for operation. The temporal sampling strategy is designed to allow the instrument to record data for up to one year.[13] In order to achieve this, the PAL is designed to enter a low power mode "sleep mode" between each data sample. The principal power usage is from the microprocessor when it is awake, drawing 43 ma. The microprocessor needs to be in this mode for roughly 30 s for each sample. The microprocessor only draws 0.3 ma when asleep. The hydrophone, pre-amp and signal processing board draw 12 ma when on and 1 ma when off. These only need to be on for about 2 s per sample, and so the power cost of each sample is 3.8 x 10-4 amp-hours. The total power cost of the expected 100,000 samples during a one-year deployment is roughly 42 amp-hours (AH). This power demand is met by using 3 stacks of 10 alkaline D-cell batteries, each with 1.6 AH of energy, a total of 48 AH. Data storage capacity is met using 2 GB flash memory cards.
Figure 4.1. PAL in deployment cage.
Electronically, PALs consist of a low-noise wideband hydrophone (either an ITC-8263 or a Hi-Tech-92WB), signal pre-amplifiers and a recording computer (Tattletale-8). The nominal sensitivity of these instruments is -160 dB relative to 1 V/mPa and the equivalent oceanic background noise level of the pre-amplifier system is about 28 dB relative to 1 mPa2Hz-1. Band-pass filters are present to reduce saturation from low frequency sound (high pass at 300 Hz) and aliasing from above 50 kHz (low pass at 40 kHz). The hydrophone sensitivity also rolls off above its resonance frequency, about 40 kHz. A data collection sequence consists of a four-second time series collected at 100 kHz. This time series is then sub-sampled at four times generating four 1024 pt or 10.24 ms short time series. Each of these sub-samples is fast Fourier transformed (FFT) to obtain a 512-point (0-50 kHz) power spectrum. These four spectra are averaged together and spectrally compressed to 64 frequency bins, with frequency resolution of 200 Hz from 100-3000 Hz and 1 kHz from 3-50 kHz. These spectra are evaluated individually to determine the acoustic source and then are recorded internally. The time interval between data collection sequences is variable depending on the acoustic source detected and the mission requirements.
The PAL is not a continuous acoustic sampler. The basic PAL data are a time series of spectral levels between 200 Hz and 50 kHz. The interval between samples is chosen by the user depending on the intent of the deployment. Typically this interval is several minutes, but is variable depending on the sound source detected. Between data samples, the PAL processor enters a deep sleep mode to conserve batteries.
4.2 PAL Deployment in the Haro Strait
The PAL was deployed in the Haro Strait for a 6 week period in early summer of 2004 (May 27 to July 2004). The mooring was located along the western coast of San Juan Island in approximately 300 meters of water [48 30.186 N, 123 08.896 W], as shown in Figure 4.2. The PAL was attached to a bottom mooring with the recording hydrophone at an approximate depth of 100 meters below the surface. The mooring configuration is illustrated in Figure 4.3.
Figure 4.2.
Position of PAL deployment in Haro Strait. 50 meter depth contours
shown.
Figure 4.3. Haro Strait mooring configuration.
4.3 Ship Signatures
Looking in detail at the time-frequency data provided by the PAL recordings reveals a variety of acoustic signatures that are likely representative of the underwater acoustic environment of the Haro Strait. Since the data available for this study is recorded in the form of time averaged spectra, the data does not typically show transient acoustic signals of duration on the order of ~5 seconds or less, such as individual whale vocalizations. However, it does show (as it was designed) longer time scale features such as the acoustic signatures of passing ships and natural sound sources such as wind and rail. In this sense, the PAL data is representative of persistent background sound levels that an animal would experience in a particular area and at a particular depth.
Consider the spectrogram for a single day (day 151) recorded by the PAL instrument as show in Figure 4.4, representing a typical day during the period of deployment. During the day light hours increased acoustic activity is observed as the density of spectral lines increases relative to the density during night time. Ships and boats are likely the primary source of background sound levels. Ship traffic also continues during night time hours. Larger ship signatures are characteristically loader at lower frequencies with broad spectral bandwidths. The higher frequency content of the ship signatures is typically lower for ships at longer ranges from the receiver, as higher frequencies will attenuate more with range. In general, however, the acoustic signatures of ships are complex and can vary greatly with ship type, speed and orientation.[14] Other sound sources are also observed in Figure 4.4. For example, a signature of rain noise in the range of 10 to 40 kHz is observed at around 5:00 am[15], as circled.
Figure 4.4. PAL data for a single day.
Of particular interest in the PAL data is the correlation of specific acoustic signatures with ship tracks provides by the VTOSS database. This allows the correlation of specific acoustic recording with the ship type, location, speed and orientation. The recorded signatures of identified ships can then be used (in combination with propagation modeling) to infer absolute ship source levels and their contribution to background sound levels in particular areas. This analysis will be discussed in the following sections. However, several examples are discussed here to illustrate the methodology of ship source analysis.
Example 1: Cargo ship passing within 1 km of mooring:
The signature of a ship passing close to the mooring is characterized by higher received levels with an increase in the higher frequency content of the signal. It should be noted that the characteristics of a specific signature (spectral level and content) include both propagation effects and the unique spectral signature of the source. Different ships have different spectral signatures that are a function of speed and orientation. However, in general higher frequencies tend to attenuate more with range.
In this example a single ship track and acoustic signature is isolated. The spectral signature is shown (circled) in the PAL spectrogram in Figure 4.5. The associated VTOSS ship track illustrated in Figure 4.6a shows that the ship passed within 1 kilometer of the mooring at approximately 15:00 on day 151, 2004. The spectral signature as a function of time (as the ship passes) for two frequencies of interest are shown (3.6 kHz and 10.4 kHz) in Figure 4.6b.
Figure 4.5. PAL data for a single day, with cargo ship identified (ellipse) from VTOSS data at hour 15 (3:00 pm).
Figure 4.6 a) Track of a cargo ship (blue) derived from VTOSS data in the vicinity of the PAL mooring (green star). Concentric circles about the mooring of radii 1,2,3,4, and 5 km are shown in green. b) PAL intensity data for two particular frequencies over the time interval when the cargo ship was in the vicinity of the mooring.
Example 2: Cargo ship passing at a range of ~5 km from mooring:
The signature of a ship passing at a larger distance from the mooring is characterized by lower received levels with less higher frequencies in the signal. In this example the ship track and acoustic signature is also isolated, and the spectral signature is delineated in Figure 4.7.
Figure 4.7. PAL data for a single day, with cargo ship identified (ellipse) from VTOSS data near hour 6 (6:00 am).
The associated VTOSS ship track of this ship is shown in Figure 4.8 where the ship passed within minimum range of 4-5 kilometer of the mooring at approximately 05:45 on day 151, 2004. The spectral signature as a function of time for two frequencies of interest are shown (3.6 kHz and 10.4 kHz). Note that the spectral levels at the same frequencies are approximately 20 dB/Hz lower as compared to a ship passing at a distance of 1 kilometer. Note also that the spectral signal shows much less variability.
Figure 4.8 a) Track of a cargo ship (blue) derived from VTOSS data in the vicinity of the PAL mooring (green star). Concentric circles about the mooring of radii 1,2,3,4, and 5 km are shown in green. b) PAL intensity data for two particular frequencies over the time interval when the cargo ship was in the vicinity of the mooring.
5. Acoustic Propagation Modeling
This section will discuss simulations of acoustic propagation and how it can be used to estimate acoustic transmission loss and variability associated with a particular source and receiver location. The previous sections discussed the shipping traffic database (VTOSS) and acoustic recordings (PAL), which allow one to associate individual ship tracks as acoustic sources of sound with data recorded on the PAL. Recordings provide the underwater sounds levels received at a specific location and time due to unknown sound sources. The VTOSS database provides a record of locations and times of the sound sources (the larger ship traffic, at least) but no direct information about their acoustic signatures and source levels. Simulations provide a mechanism to combine these data (recordings and ship tracks) to estimate the source levels of the individual ship tracks.
5.1 Model Description
The acoustic propagation model used in this study is a slightly modified version of the parabolic equation (PE) model that is part of the Navys Oceanographic and Atmospheric Master Library (OAML). The modified version we use allows for a rough surface, which can be important for higher acoustic frequency (> 1 kHz) signals. This model assumes an isotropic spreading of acoustic energy in azimuth about a locally cylindrical coordinate system with axis of symmetry running vertically through the horizontal position of the acoustic source (a ship). The model also assumes acoustic energy only propagates away from the source, and backscattering is neglected. These two approximations are reasonable for most situations and for the study we have done here.
There are expected to be regions near the west coast of San Juan Island with significant backscatter from the sloping walls. These regions are very difficult to model because: 1) efficient full-wave numerical models that accurately describe the backscattering are not readily available for the range scales of interest, and 2) the bottom properties are extremely difficult to obtain at the necessary resolution for performing accurate model predictions. There are ways around this, but these issues lie outside the scope of this report. In Section 6 a methodology to address this issue for a future study is proposed. However, we feel that the PE model is appropriate for this study because most of the propagation modeling is constrained to the channel, and the sound that has interacted with the steep canyon walls will be significantly lower in intensity compared with what has propagated from a ship source in the channel.
Like all PE models, the basic equation the model solves is an approximation to the Helmholtz wave equation, whereby one assumes that the envelope of the evolving acoustic pressure field varies slowly on the order of the acoustic wavelength. This allows a one-way wave equation for the envelope of the acoustic pressure field, which can be numerically solved as an initial value problem. The standard boundary conditions are pressure equal to zero at the surface and the normal derivative of the pressure equal to zero at the computational bottom. Note the computational bottom is often hundreds of meters greater than the ocean bottom to allow for sediment and basement interactions, and the bottom is treated as a fluid. A profile of sound speed is required in the ocean volume domain, and a profile for sound speed, density, and attenuation is required at and below the ocean-sediment interface. The frequency dependent attenuation due to boric acid and magnesium sulfate ionic relaxation processes, which are significant at acoustic frequencies at and above one kilohertz are automatically handled within the model.
Some particular features that the model does address is rough surface and rough bottom scattering. Rough bottom scattering is handled by modifying the bathymetric data base described in Section 2.1 to include random displacements at horizontal scales of the order of 1 meter. The rough surface is addressed similarly, but the random surface displacements are treated more physically, by creating realizations based on their spectral characterization from wind forcing. The surface is considered frozen in the model, which is a valid approximation since the phase speed of the acoustic waves is much greater than that of the surface waves. Also, since the model treats the bottom and surface as deterministic, one cannot expect point-wise convergence between the model and data; i.e. we can only obtain realizations of the bottom and surface that are statistically similar to what actually existed when the data was collected, so we can only expected statistical agreement between the model output and data. At lower frequencies of propagation, the effects of the rough surface and rough bottom become less important, and one might expected point-wise convergence to be obtained. However, at these lower frequencies (hundreds of Hertz), the effects of the sub-bottom on the propagation become important, and little is known about the geo-acoustics of the Haro Strait.
Finally, regarding the sub-bottom, the model allows for a jump in both density and sound speed at the water sediment interface, and treats the sub-bottom as a fluid (shear waves are ignored). The sound speed in the sediment layer is allowed to increase linearly with depth, and a frequency dependent bottom loss is included. The thickness of the sediment layer is allowed to vary with range, independent of the bathymetry, and a highly absorbing basement layer is modeled below the sediment.
For a more thorough explanation of the PE model used in this study, the reader should review the papers by Collins[16] and Rosenberg.[17]
5.2 Model Inputs
Since the environment is not well characterized in the Haro Strait region, simulations are performed for a realistic range of input values. In effect this provides bounds on the model outputs, given the uncertainty of the inputs. Model inputs consist of
1) Geo-acoustic parameters of the seafloor,
2) Sound speed profiles of the water column,
3) Bathymetry and bottom roughness,
4) Sea surface roughness (due to wind).
5.2.1 Geo-acoustic parameters of the seafloor:
For the region of interest, the geo-acoustic parameters of the seafloor are the least characterized of the model inputs. Two bottom types are assumed: 1) mixed sand/mud sediment, and 2) mixed rock/sand sediment. The model inputs for these two bottom types are shown in Table 5.1.
|
Geo-acoustic Parameter |
Sand/Mud |
Rock/Sand |
|
Sound speed (m/s) |
1550 |
1800 |
|
Sound speed gradient |
2 |
0 |
|
Density (g/cc) |
1.7 |
2.0 |
|
Attenuation (dB/lambda) |
0.129 |
0.7 |
Table 5.1. Geo-acoustic parameters used for PE modeling in Haro Strait.
5.2.2
Sound speed profile:
The sound speed profile for the region is assumed to be linear with sound speed of 1483 m/s at the surface and 1480 at the bottom. These values are taken from the mean sound speed profile derived in Section 2.2.
5.2.3 Rough sea surface and sea bottom:
Randomness in the medium is modeled as surface roughness at both the sea surface and the sea bottom. Bottom roughness is generated using a power-law spectral model[18] with inputs defining the RMS height of the roughness and correlation length. Estimates of these parameters for the Haro Strait are not available from measurements. Therefore, we estimated values based on similar types of bottoms. In all the cases shown we used for a bottom root-mean-square roughness height of 0.4 m and the bottom roughness correlation length of 10 meter.
Sea Surface roughness is generated using a Pierson-Moskowitz spectrum[19] spectral model for a 1D surface with wind speed U (m/s) and parameters alpha and beta. Three cases of sea surface conditions were considered: a flat surface (no wind); a typically sea surface with wind speed of 5 knots; and a rough surface with winds of 10 knots. The values used for the PE simulations are given in Table 5.2.
|
Sea Surface Roughness Parameters |
Flat |
Typical |
Rough |
|
Wind speed (m/s) |
0 |
5 |
10 |
|
Alpha |
0 |
8.10e-3; |
8.10e-3; |
|
Beta |
0.74 |
0.74 |
0.74 |
Table 5.2. PE model parameters used for simulating realizations of a rough sea surface due to wind forcing in Haro Strait.
5.2.4 Monte-Carlo simulations:
Inputs to the model that represent randomness in the environment require the generation of a statistical ensemble. The ensemble is used to describe the variability (e.g. mean and variance) in the model outputs due to the model inputs. For example, bottom roughness when added to the propagation modeling will create scattering at the sediment-water interface that will, in general, increases sound penetration into the seafloor, thus potentially increasing propagation loss between the source and the receiver. The addition of this type of randomness is an important aspect of the modeling effort. Statistical estimates are found using a Monte-Carlo method, in which multiple realizations of the surface roughness are generated for each source/receiver position and the model executed for each realization. Surface roughness realizations are generated by Fourier synthesis assuming Gaussian statistics for the roughness height.
Figure 5.1 illustrates the simulation output for one two-dimensional slice of the medium between the source (at a depth of 10 m in the upper left corner) and a grid of receiver positions along the propagation path. The upper surface in the plot is the sea surface (depth=0). The color scale represents acoustic intensity relative to the unit intensity source. The lower surface of the simulation illustrates the topography of the sea bottom, as the acoustic penetration into the bottom is highly attenuated, changing from higher intensity (red) to lower intensity (blue). In this case the position of the source corresponds to the position of a ship, and the far right boundary of the grid corresponds to the position of the PAL mooring. The lower panel of the Figure 5.1 illustrates the same simulation with bottom roughness added to the sediment-water interface. Increased randomness in the water and penetration into the sediment due to roughness at the sediment-water interface can be observed.
Monte-Carlo simulations provide an ensemble of transmission loss estimates for each source/receiver position and for a two-dimensional grid of values between the two positions. To compare model results with data (as will be discussed in the next section), descriptive statistics of the field at the receiver position must be evaluated. In this report two measures are used: 1) the central tendency will be defined in terms of the arithmetic mean and median of the field; and 2) dispersion of the model results will be defined by the interquartile range (IQR). The IQR of the model is used to avoid imposing Gaussian assumptions on the model data variability. In general, one would expect to observe a Rayleigh (exponential) distribution for the values of the acoustic intensity due to randomness (surface roughness) in the propagation environment, at least in the limit of long range. However, for the limited number a realization used in the Monte-Carlo simulations (10 to 20), as well as the limited range, we found the IQR to be more robust measure to highlight the variability of the acoustic intensity.
Figure 5.1. Comparison of transmission loss model results with and without sea surface roughness. Colors indicate intensity in dB, with red corresponding to high intensity and blue low. The dynamic range shown is 80 dB.
In addition to measuring dispersion of the model results over the Monte-Carlo realizations, the ensemble of output values is increased by considering a depth window for data analysis. In this case, all the grid values within a specified range of depths at the receiver location are added to the ensemble. As will be discussed further, there is considerable variability of the modeled intensity as a function of depth. Considering a window of depth (as opposed to a single depth value) also facilitates comparison of the model results with field data, as the exact depth of the PAL receiver is not known precisely.
5.3 Model Outputs
The propagation model described in this section provides an estimate of the propagation loss between a source and a receiver. The source is modeled as an omni-directional point source emitting a continuous (CW) signal at a single frequency. All the simulations performed in this report are performed at 3.6 kHz. This frequency was chosen to match the frequency bin provide by the PAL data. The model output consists of a two-dimensional grid of sound pressure levels (relative to the unit source level) at discrete depths and ranges from the source position. The complex sound pressure level (p) output from the PE model is expressed as transmission loss using the sonar equation:
,
where R is the range from the source to the receiver. Given an estimate of TL, the actual sound level received at the receiver location is found by adding the source level (SL in dB) to the transmission loss.
5.4 Comparison of Model Results with Field
Measurements
Simulations
provide an estimate of the propagation loss between a source and a receiver
(along with a measure of data dispersion). The measured field data provides absolute (calibrated) sound
pressure levels received at a specific location in space and time. A comparison of model simulations and
field data for known positions of acoustic sources (large ship tracks, in this
case) can provide an estimate of the level of sound emitted at the source. Once confidence is gained in the
modeling and measurement methodologies, the process of model/data comparison
can be used to estimate the propagation of other sources of anthropogenic sound
in the same area, to evaluate the potential variability of sound levels as a
function of location and time (seasonal), and to eventually provide estimates
of total shipping sound source levels in a limited region (such as in the main
shipping channel).
This section
provides a preliminary analysis of the model results compared with PAL data for
the limited period of time during the deployment of instruments in May-July
2004. The objective is to estimate
the source levels of large ships in the Haro Strait by comparing a variety of
model results of propagation loss with recorded sound levels. The comparison is useful to evaluate
applicability of model inputs and modeling strategies. Since the true ship source levels are
not known, and in general are quite complicated, the estimates formed here are
preliminary. Recommendations for
further studies will be discussed in Section 6.
5.4.1 Single Ship Comparisons
In cases where the acoustic signature of a single ship can be identified in the PAL data and the ship track identified by the VTOSS database, simulations of the propagation can provide a means of estimating the source level of the ship. An example of this process is illustrated with a single ship as identified in Section 4.3.
Simulations are performed for 10 ship positions corresponding to 1 km range increments as the ship passed by the stationary mooring, as illustrated in Figure 5.2. Figures 5.3 illustrate the model results (intensity in dB) and the two-dimensional simulation geometry for each bathymetric slice between the source positions and the receiver. The source is located in the upper left corner of each slice, and the receiver is at the 100 depth grid point at the right edge of the slice. The range between the source and receiver decreases and then increases as the ship passes the mooring.
.
Figure 5.2. Source positions for VTOSS ship track named EverUnison corresponding to the source position in the TL plots show below. The red line shows the path associated with the first source position. Depth contours at 50 m intervals are shown.
Figure 5.3. (Following) Plots of transmission loss between the ship source positions (1 through 10) shown in Figure 5.2. Color scale represents the field in dB units with a 80 dB dynamic range.
Propagation modeling is performed using Monte-Carlo simulations and bounds on the model inputs (as previously discussed). To determine model sensitivity to different types of geo-acoustics parameters and wind speeds, multiple simulations are performed. The central tendency and dispersion of transmission loss are estimated for each case and results compared with measured data. For each case the transmission loss estimates are fit to the data to provide an estimate of the ship source level.
Figures 5.4, 5.5, and 5.6 illustrate the model/data fits and source level (SL) estimates for three cases: 1) smoother sand/mud sediment, 2) rough sand/mud sediment, and 3) rough rock/sand sediment. Each figure shows the model results found for each source position as the ship passes by the mooring (10 positions total). At each source position three cases of wind speed (sea surface roughness) are simulated: 1) flat sea surface with wind speed 0 m/s, 2) rough sea surface with wind speed 5 m/s, and 3) rough sea surface with wind speed 10 m/s.
The model results are presented with the mean and median to express the central tendency of the simulated data and the IQR of the simulated data to express dispersion of the results. The IQR bounds represent the difference between the 75th and the 25th percentiles of the model data. Because of outliers in the data, the IQR was found to be more representative than the standard deviation as an estimate of the spread of the body of the data.
In all three cases of sediment type and wind speed the model results fit the measured results when an estimated ship source level of SL=175 dB was applied. In the last case (rough rock/sand sediment) the simulation results are high, implying that the transmission loss in this case is less than the other cases. This result is consistent with expectations as a rock/sand sediment type would be expected to attenuate sound propagation less than a sand/mud bottom type, in general. The results, however, are similar to within 3 dB for the various cases, implying that bottom type, sea surface roughness, and bottom roughness do not contribute greatly to the model results in this case. It may be case that the propagation distances between the ship source and receiver are so short that secondary effects of the bottom and surface conditions do not dominate the propagation loss estimates, with bathymetric effect dominating.
The estimated ship source level of SL=175 dB @ 3.6 kHz is consistent with expected source levels for a large ship traveling at slow speed. Although these estimate can not be verified since no independent measurements of source levels are available.
Variability in the model results is expressed in terms of the IRQ. In all three cases a range of values is on the order of 5 dB. Increased wind speed does not appear to increase variability. In the case of zero wind speed (flat sea surface), the observed variability is due to windowing in depth (as opposed to multiple realizations), indicating significant fluctuation as a function of depth (relative to the variability caused by surface roughness).
Figure 5.4. Comparison of model results and PAL data for a single ship track (EverUnison, day 151) with three wind speeds (0, 5, 10 m/s) and a smooth sand/mud bottom.
Figure 5.5. Comparison of model results and PAL data for a single ship track (EverUnison, day 151) with three wind speeds (0, 5, 10 m/s) and a rough sand/mud bottom.
Figure 5.6. Comparison of model results and PAL data for a single ship track (EverUnison, day 151) with three wind speeds and a rough rock/sand bottom.
5.4.2 Shipping Lanes
The modeling strategy applied to an individual ship track can also be applied to the average positions of ships traveling in the major shipping lanes. Large commercial ships typically travel within a narrow lane north and south with a predicable track (see Figure 3.3). The simulation of transmission loss for an average track along the shipping lane can provide a means of estimating ship source levels without performing a simulation of each individual ship.
In the case presented here, transmission loss estimates are formed for the northern ship lane derived from the VTOSS database for commercial cargo ships. Average ship positions along the center of the shipping lane are found at constant range increments of 1 kilometer from the mooring location, as illustrated in Figure 5.7. At each position along the shipping lane Monte-Carlo simulations are performed for a 3.6 kHz source at 10 meters depth. The geo-acoustic parameters for these simulations are limited to rough sand/mud sediment type. As shown previously, the bottom type does not appear to be a major source of variability in the model results. Two wind speed values (5 and 10 m/s) are used to estimate a model central tendency and data dispersion.
Figure 5.7.
Source positions for the northern shipping lane derived from VTOSS ship
tracks.
Figure 5.8 illustrates the model results for transmission loss between ship positions in the shipping lane as the ship travels north pass the mooring location. Source position numbers correspond to the position numbers in Figure 5.7. The transmission loss estimates show at least a 10 dB variation in received source levels as the ship passes the mooring. Variability due to sea surface state is estimated to be at a least 5 dB with the IQR of the model data.
Figure 5.8.
Model results for transmission loss associated with shipping lane
traffic (as shown in Figures 5.7).
The mean and standard deviation of the model results are obtained by
Monte-Carlo simulations with 20 realizations of a random sea surface and bottom
roughness.
6. Recommendations
The previous sections described the modeling effort. Here some recommendations for future directions of modeling and measurement strategies will be discussed.
6.1 Measurement Strategies
It is our recommendation that acoustic modeling in the Haro Strait is best used as a complement to field measurements, as the environment is far too complex, and the geo-acoustic parameters of the area are not characterized well enough to rely on modeling alone. However, to develop further confidence in the models and to enable them as a tool for prediction, several controlled measurements should be done, which to the best of our knowledge have not been performed. These experiments should be designed to measure quantities that can be compared directly with modeling results and specifically designed to investigate propagation from the shipping channel.
Several cost effective acoustic experiments seem appropriate at this time. In particular, focusing on shipping noise estimates, several measurements would be useful:
1. Direct measurements of transmission loss using a known source in the shipping channel: These measurements would address commercial shipping traffic noise and be directly related to the modeling approach illustrated in this report. Here measurements could be done using a calibrated source located in the shipping channel and calibrated receivers at several different locations of interest out of the shipping channel, such as along the coast of San Juan Island. Such measurements should include:
a. Narrow frequency band measurements (long tones) to better compare data with existing propagation models. This can be done cost effectively using a recording instrument like the PALs and a hand held source deployed over the side of a small boat. The source levels would be low (comparable to ship source levels) to avoid animal disturbance. Measures data would be directly comparable to simulations of transmission loss, providing insight into the validity and accuracy of a PE modeling approach.
b. Broadband sources (short pulses) to address the issue of reverberation and backscatter from the channel walls. In direct comparison with the above measures this would address the relative importance of reverberation in the area. The difficulty with these measurements, however, is that the source levels for the short pulses must be relatively high in order to make reliable measurements, compared to the longer tones. So care must be taken to avoid animal disturbance.
2. Longer term recordings of shipping noise: These types of measurements could be used to investigate:
a. Seasonal variation in the background noise levels that could be correlated with the changing oceanographic and weather conditions, and
b. Regional variation with receivers at several locations in the Strait to sample the wide range of geographic conditions, and
c. Correlation of seasonal and regional measures of background sound levels with records of killer whale locations over time.
Each of these measurements would require a cost effective acoustic receiver. To measure background sounds levels, averaged spectral data is sufficient and reduces the data storage requirements. Time series data is not necessary unless transient signals (i.e. whale vocalizations) are also desired. In the case of time series data, the receivers would require either a cable to shore or radio (wifi for short distance) connection to land. For time series data recording over long periods of time, data processing and storage requirements are significant and should be address up front. Given these limitation, it is recommended that spectral type recording (requiring much less data storage and processing) be used.
3. Direct measurements of broadband source levels of large ships and tugs in the main shipping traffic lanes: This is necessary because no data really exists on ship source levels. Therefore, the model-based inversions made from a distance are unvalidated. A simple way to do this is to put a mooring in the shipping channel (using a PAL or a similar instrument) to record ship signals as they pass overhead. With this geometry the ships sound source levels are measures a very short ranges and propagation effects are reduced significantly, allowing for good estimates the true source levels of the ships in situ. The VTOSS database could then be used to compile statistics on ship noise with ship type, speed and orientation. With good information on the true source levels of ships, model results and data from remote sites (not in the ship lanes) can then be used to address the sound exposure levels over larger areas and at different times of the year. Combined with models and other data, this would also provide a baseline measurement of sound levels in the Strait that could be used to address long term trends and impact.
Implicit in these recommends is the appreciation for the practical value of developing a modeling capability. This comes in part from experience, as measurements are generally much easier to make than they are to understand. Acoustic measurements always require thoughtful interpretation, which usually requires a propagation model, unless the recording are made at very short ranges or in very simple environments. With confidence in a modeling strategy, models can then be used to extrapolate data to other regions of the Strait where measurements are not available and to predict sound levels for potential scenarios such as Naval operations or increased commercial shipping traffic.
6.2 Estimation of Total Shipping Noise
Total background sound levels can always be measured directly with in-situ hydrophones. But to cover the relatively large area of the Haro Strait, many sensors would be required. In the final analysis, it is also desirable to know where the sound is coming from, as opposed to combined measurements of all the sound sources in an area. For example, a relevant question in the Straits may be: what is the total background noise levels due to commercial shipping vs. pleasure boats or commercial whale watching boats?
The combination of propagation modeling of the shipping lanes, selective in-situ measurements, and the VTOSS database can be used to address this issue of noise partitioning and provide an estimate of total shipping noise within the region at different times of the year. In principle, the basic method is straight forward: use the VTOSS database to identify the times and ranges of ships that pass within a specified area and then use the propagation model to estimate the sound levels at a specific location or set of locations.
The estimates ship source levels can be handled in two ways. First, one could relying completely on assumptions about ship source levels, or as mentioned above, use direct measurements of ship sound source levels to validate model/data inversions for a set of representative ship. Alternatively, one could use long-term recordings from a single location (such as done in this report using the PAL data) to invert for ship source levels. This can done using the VTOSS ship tracks and times to identify individual signatures in the PAL data, as illustrated in Section 4.3. Then the measured data can be fit to the transmission loss curve for the average shipping lane, providing source levels of the ships, as illustrated in Section 5.4.1.
Since the majority of large ships travel within the narrow shipping lane, numerical simulations for the average ship lane track should be sufficient, as opposed to performing numerical simulations for each individual ship track. Errors associated with the track deviation within the lane should be minimal, as variability due to surface roughness and bottom type uncertainty are on the order of 5 dB and likely as large as the deviation do to source position errors on the order of a 100 meters. Problem will arise when there is more than one ship contributing to the sound recorded at a specific time. However, by looking at enough data over time there will likely be enough times when individual ship signatures can be identified and a representative picture of ship traffic source levels should be able to be compiled. Alternatively, one could assume that the sound source add incoherently.
6.3 Further VTOSS Analysis
The VTOSS database of ship traffic itself is a valuable database for analysis. The analysis of this database directly could involve:
1. Developing an archiving of VTOSS data to compile statistics of shipping traffic in the region.
2. The correlation shipping data such as track density, ship type, speed, etc. with data on animal locations and movements.
3. Integration of other databases on the acoustic characteristics of ships with the VTOSS data to provide a merged database of ship acoustic source level and position. For example, the combined information on engine type, prop length, draft, hull length, etc. and information on ship speed and orientation (as provided by VTOSS) can be used to better predict source level. Classifying or estimating ship source levels from ship characteristics is a longer-term goal, where as direct measurements (as recommended above in 6.1.3) can be used in the shorter term and as a way to test predictions.