# Chinook salmon and southern resident killer whales occupy similar depths in the Salish Sea

New paper by UW colleagues entitled “Interpreting vertical movement behavior with holistic examination of depth distribution: a novel method reveals cryptic diel activity patterns of Chinook salmon in the Salish Sea” shows some results from Vemco receivers I deployed in the San Juan Islands. Young adult Chinook favor depths less than ~30 meters, with some seasonal variability in their diel activity patterns. Overall, they go deeper and vary more in the depths at night.

Dive profiles for two Salish Sea Chinook salmon during the summer and fall.

Interestingly, according to a report to NOAA/NWFSC by Baird et al, 2003 (STUDIES OF FORAGING IN “SOUTHERN RESIDENT” KILLER WHALES DURING JULY 2002: DIVE DEPTHS, BURSTS IN SPEED, AND THE USE OF A “CRITTERCAM” SYSTEM FOR EXAMINING SUB-SURFACE BEHAVIOR) SRKWs spend >97% of their time at depths of less than 30m.

This suggests any future deployment of horizontal echosounders should aim to ensonify a depth range centered on ~25m (e.g. 5-45m or 10-40 m).  Compared to the estimated orientation and surveyed depth range of our 2008-9 salmon-SRKW echosounder pilot studies, we may want to measure inclination more carefully to (a) center the survey on the mean summertime depth range of Chinook and (b) avoid ping reflections from surface waves, boats, and bubbles (which may have confused interpretations of targets >100 m from the transducer).  Here’s my diagram for the situation in 2008-9 in which we were centered on 15 m and ensonified a maximum depth range of ~0-30m (in other words, we may have been aiming a little high):

Screen grab from the 2009 ASA presentation showing echosounder geometry

With our ship source level paper (pre-print) finally revised in Overleaf and accepted by PeerJ, it’s time to get ready to help translate the manuscript (which ended up on the long side) into simpler formats to convey our key findings to the public and press.  My first impulse was to use this as an opportunity to do something I’ve long-desired for my teaching: create a table and/or info-graphic that communicates the (very confusing) difference between underwater and in-air decibel scales.

A more urgent goal for today — with press phone calls/interviews looming — is distilling some simple facts from the paper and articulating them without scientific jargon.  These same highlights and summary statistics are destined for a blog post at Beam Reach that will convey the key findings, potential implications — both for killer whale communication and echolocation masking, and for mitigation of underwater noise pollution from ships.

## How “loud” is ship noise?

Build tables that extract key results from the paper and contextualize them for the public, with attention to the confusion about underwater vs in-air decibels. First confirmed via DoSiTs that we subtract 62 when converting underwater decibels to in-air decibels (NOAA’s underwater acoustics page is not clear [and maybe wrong] on this topic!?). (Erbe, 2010) also could be improved by being explicit about how to do and derive this conversion…

 Ship class Underwater broadband source level In-air broadband source level Familiar noise source of equivalent intensity dB re 1 microPa @ 1 meter dB re 20 microPa @ 1 meter (from various Googled decibel scale images) “Max” (mean+4sigma) 201 138 Jet take-off Max of center 95% (mean+2sigma) 187 124 Jack hammer @ 1 m Mean of all ships in study 173 +/- 7 ~110 Rock concert (or earbuds at full volume? or auto horn) Max of center 95% of container ships (loudest class, mean+2sigma) 186 124 Jack hammer @ 1 m Mean of container ships (loudest class) 178 +/- 4 116 Ambulance siren Mean of military ships (one of quietest classes) 161 +/- 10 99 Chain saw

Of course, the killer whales are typically at least 100-1000 meters from these ships and we measured frequency-independent spreading rates at study site of about -18 log (R), so we should present a table of receive levels that depicts what ship noise experience is typical for SRKWs…

 Receiving situation Underwater broadband source level In-air r broadband source level Familiar noise source of equivalent intensity dB re 1 microPa dB re 20 microPa Typical background level at Lime Kiln Typical ship passing Lime Kiln “Loud” ship passing Lime Kiln 1 m from “loudish” ship 175 113 Rock concert 10 m from “loudish” ship (175-18) 157 95 Lawn mower 100 m from “loudish” ship (175-18*2) 139 77 Busy street 1,000 m from “loudish” ship (175-18*3) 121 59 Vacuum cleaner 10,000 m from “loudish” ship 103 41

## Broadband levels associated with the quantiles

Email with Val and Wikipedia remind us that Gaussian distributions hold 95% of their values in a range of the mean +/- twice the standard deviation (s.d., or “sigma”, or σ).  Here’s a comparison that Val ran down in Portland:

quantile(data_1_3_BB_noNA$ddB_SL_Emp,probs = c(0.025,0.05,0.25,0.5,0.75,0.95,0.975))  2.5% 5% 25% 50% 75% 95% 97.5% 155.6082 160.8810 169.6775 174.4650 178.3650 183.5545 184.8073 > mean(data_1_3_BB_noNA$ddB_SL_Emp)
[1] 173.5007
> sd(data_1_3_BB_noNA\$ddB_SL_Emp)
[1] 7.403251
Mean - 2 sigma
173.5 - 2*7.4
[1] 158.7      compare with 155.6   (2.5% quantile above)

Mean + 2 sigma
173.5 + 2*7.4
[1] 188.3	this one is about 3 dB above the 97.5% quantile above


So the 2*sigma estimates in the table are close enough, but may be slight (3 dB) over-estimates of the broadband levels associated with the  97.5% quantiles…

Erbe, C. (2010). Underwater acoustics: Noise and the effects on marine mammals. Pocketbook, printed by JASCO Applied Sciences, Brisbane, QLD, Australia. Retrieved from http://oalib.hlsresearch.com/PocketBook%%203rd%%20ed.pdf

# Ship noise paper progress: spectral figures

Back on 3/4/14, Val and I decided to bounce this strategy off Jason: rather than using only night-time data (to exclude the 50 kHz spike and hump we see in the 95% (and to a lesser extent 75%) quantiles of ship and background RL, present quantiles for the whole population and then underlay a “nighttime-only” 95% quantile curve to show how the 50kHz humps go away… If he balks (or reviewers do down the line), then we ~halve data set by only using nighttime data..

Working with Val in Seattle on final details of receive level figure:

• Hz with sci notation in x-axis, rather than kHz (for ship noise audience)
• Increase horizontal grid lines to make it easier to read off values and differences between background and ship levels.
• Work out how to get a legend in ggplot (involves “melting” data series into single column of data with variable name in adjacent column) and position it inside the plot

We sketched out Tufte-ian solution to the ship population source level plot that will show per Hz, 1/12-octave band, and 1/3-octave band levels.

Scott’s sketch of what a plot comparing spectrum, 1/12-, and 1/3-octave levels of ship noise.

About a month later, we’ve got that figure finalized and are working on the last piece of the paper: whether there’s anything interesting to say about acoustic outliers within a particular class of ships.  Here are notes I took as I visually analyzed plots Val made of noise spectrum levels for each of our ship classes.  Each plot has 25, 50, and 70% quantiles from the population within that class, as well as clouds of data points for each ship measurement made within that class.

Scanning through the DropBox folder called SL_by_Class, here are some highlights:

1) First a data processing sanity check: What are the diagonal features/artifacts (with slope of ~10 dB/decade) showing up in the densest of data point clouds in the 1/12 (but not 1/3!) octave level plots, e.g.–

ddB_abs_vs_ 1_12_octave band _ Bulk carrier __.png
and
ddB_abs_vs_ 1_12_octave band _ Container ship __.png

See this example at HF end in this zoomed screen grab  —

and more worrying example at LF because of the reflection of the slope in the 1/12 octave levels, along with what may be steps in the population distribution of levels near 25 Hz and 45 Hz –

2) Here is a list of per Hz (with absorption) png files that show *really* interesting outliers which I’ve grouped into a couple “categories of interest” —

ddB_abs_vs_ hz _ Bulk carrier __.png
1 super-high, 2 high, and 1 low outliers at HF
The super-high and unusually low outliers define a huge range from lowest to highest at uppermost HFs — about 60 dB for bulk carriers!
Other classes have less variability, e.g. Cargo ships and tankers ~30 dB, Container and military ships ~40 dB, Vehicle carriers ~45 dB, Tugs ~50 dB
For perspective, Ross (Ross, 1976) shows a plot of WWII sub cavitation inception raising levels in a 10kHz-30kHz band 50 dB as the speed of a sub:
(at 7m) increased from ~3kt to ~5kt…
(at 16m) increased from ~4kt to ~6kt…
(at 91m) increased from ~8kt to 12kt.

ddB_abs_vs_ hz _ Cargo __.png
1 high outlier at HF

ddB_abs_vs_ hz _ Container ship __.png
2 similarly high outliers at HF

ddB_abs_vs_ hz _ Tanker __.png
1 of upper outlier at HF shows some very interesting wiggles, akin to those that Hildebrand captured from the Hanjin
Here they are looking very wiggly (in the no-abosorption, per Hz version of the plot) –

ddB_abs_vs_ hz _ Tug __.png
2 high outliers at HF, one or both of which have some wiggles

ddB_abs_vs_ hz _ Vehicle carrier __.png
1 particularly high outlier

3) Other random nifty observations

Different fisheries boats show cool peaks near common transducer frequencies (e.g. 38 and 50 kHz):

There was more variability in HF levels for military boats than I expected.  I wonder if this is different ship/prop designs or different speeds — maybe a good class in which to look for correlations?

Ross, D. (1976). Mechanics of underwater noise. Pergamon Press.

# High frequency noise in the Salish Sea

After registering for ORCID and figshare, I worked with Val today in Seattle on trying to understand some of the subtleties in our ship and background quantile curves.  In particular, there are some peaks — both broad and narrow — in the 75th and 95th % quantiles from ~20-96kHz that may represent signals from sources other than commercial ships.  If they are being generated by “outlier” ships, then we need to put some effort into understanding which ships are the source(s) and what mechanisms may be generating the sound.

Here’s a grab of the HF part of Val’s preliminary figure (thanks to R and ggplot2) —

HF ship (blue) and background (black) 5,25,50,75,95% quantiles

There is a clear spike at 50 kHz that Jason thinks it is coming from depth sounders and/or fish-finders mounted on recreational fishing boats that happen to be nearby when a ship passes our study site and get recorded.  Though our ship recordings are only 2 seconds long, it’s possible that a nearby fishing boat could generate a ping during that interval because most sounders and finders have ping rates of 1-5 pulses per second.  The alternative is that depth sounders on the commercial ships are responsible for this peak.

To test those two hypotheses, Val wrote an R script that goes through our data set and looks for strong cross-correlations between a template (a power spectrum with a good strong peak at 50 kHz) and the rest of the ship recordings.  We looked through the resulting graphs for high correlation coefficients and visual similarity between the spectra.  Noting the month and time of day of these spectra, we found that all but one occurred during daylight hours and during the peak of the bottomfish & salmon fishing seasons (March-October).

Histogram of month of year in which ship spectra contain a prominent peak at 50 kHz.

Histogram of time of day at which ship spectra contain a prominent peak at 50 kHz.

## Recreational fishing depth sounders and fish-finders?

The bulk of these recordings were made during daylight hours.  Even the sample taken in the very early morning (5:50 a.m.) occurred during daylight; sunrise on that August morning was at 5:38.  One possible exception is the recording made at 18:50; it occurred on November 1, 2012, when sunset was at 17:51…

Because of the prevalence of daylight recordings, we presume that the more likely source of most of the 50 kHz pings is depth sounders or fish-finders mounted on recreational fishing vessels.  These boats are often observed drifting or trolling off of the west side of San Juan Island, particularly during the summer months.  They frequently operate very near shore, putting them directly over or within a 500 meter radius of our hydrophone.

## Commercial ship depth sounders

The one nighttime spectrum is also anomalous seasonally, occurring in January when odds are very good no one is out fishing in Haro Strait at 9:20 p.m.  Although we’re finding a great variety of pulse widths and rates among the spectra with 50 kHz peaks, this one stands out as having a clear first arrival and subsequent reverberant bottom-bounces.  It also has a slow pulse rate: about 1 pulse every 1.5 seconds, or 0.68 pulses/sec.

MMSI 636091891 waveform and spectrogram from 1/1/12 at 21:21 showing 50 kHz pings.

Waveform and spectrogram showing 50 kHz ping sequence.

The TDOA is 0.1 seconds which is a path length difference of about 150m.  That’s the approximate depth of Haro Strait midway between the northbound shipping lane and Lime Kiln.  So in this case, the pings may indeed have come from the ship.  Here’s the sound file in case you want to load it into Audacity and listen to the pings; try playback at 0.01x – 0.13x speed.  This ship is a 334 m container ship named the Vancouver Express.

Since Val included in the cross-correlation assessment only ships that had transited our study site at least 10 times, the other strong evidence that pings are made by a ship rather than a nearby boat would be repetitions of similar ping sequences each time the ship passes.  This assumes that it is unlikely that the same type of recreational fishing sounder would be present during each transit of a particular cargo ship.  On the contrary, if the ship’s depth sounder is heard on one transit, we ought to detect that same sounder on other transits of the same ship.

Vancouver Express 636091891 transit on 2-11-12 at 11:50

Vancouver Express 636091891 transit on 9-7-11 at 6:47

Below is another fascinating example of ship-generated noise, including 50 and 79 kHz pulses that may have come from the ship, or some other vessel(s).

372649000 7-8-12 showing 79 & 50 kHz pings and cavitation noise to 96 kHz.

Next will come the tough part: excluding from our ship noise data set the recordings that have spectra contaminated by fishing boat pingers, system noise, and other non-ship sources…  Thankfully, Val is a programming wizard so this won’t take long!

# Late night notes on noise from ship propeller cavitation

In Val’s most recent plots of noise levels received from passing ships at our Lime Kiln study site, we’re seeing interesting peaks at high frequency (>20 kHz) that vary between ship classes. Most prominently, there is a peak near or just above ~40 kHz for many ship types that is sometimes quite narrow-band, other times really broad (+/- ~10 kHz).

What spectral patterns should we expect for modern commercial ships, particularly at frequencies between 10 kHz and 100 kHz?

Thus begins another dive into the ship noise literature… revealing:

1. Merchant ship propeller diameter (in meters) is about equal to their length in meters divided by 25 (Gray, Greeley, 1980).

From Gray and Greeley, 1980

2. Mean blade-rate frequency varies a bit with ship length, but basically for big ships a typical frequency is about 8 Hz.

Blade rate frequency for ~200m ships (from Gray and Greeley, 1980)

3. The blade-rate frequency shows up because cavitation noise is maximized when a propeller blade tip passes through the low velocity and pressure region of the wake field (the top of their rotation on single screw ships).  There is also a lot of energy at many harmonics of the blade-rate frequency.  Similar patterns of peaks are created by engines (firing rate harmonics) and generators (generator harmonics).
4. The following plots from (Arveson, Vendittis, 2000) show that underwater noise data collected <~600m from a 173 m long coal carrier contain these numerous harmonics.

1. All those peaks and additional broad-band noise from the collapse of the cavities add up to create spectra (e.g. 1/3-octave levels, below) which have led to the common characterization of shipping noise as having most of the energy at low frequencies (1-500 Hz).  As Areveson and Vendittis put it:“In addition to the blade rate harmonic series, cavitation generates a wideband spectrum due to the chaotic collapse of cavities. This spectrum has a broad, high-level ‘‘hump’’ centered at about 55 Hz, followed by a continuum that decreases by 6 dB per octave on a constant-bandwidth plot, or 3 dB per octave as seen on a 1/3-octave plot.”

1/3 octave levels vs ship speeds

5. What is commonly overlooked is the rise in high-frequency noise that occurs with increases in ship engine RPM (and correspondingly ship speed).  Note that the increase in noise between the highest two RPM levels at 30,000 Hz is 2-4x the increase at 30 or 300 Hz.
6. The authors note: “Above the cavitation inception speed the shape and peak frequency of the wideband cavitation spectrum do not change appreciably with ship speed. Only the overall level changes; it increases smoothly with speed according to 104 log (rpm), or about 31 dB per double speed.”
7. While these spectra are interesting, they are not showing much structure above 20 kHz…
8. At least for the cruise ships going 10 knots and measured at 500 yards by (Kipple, 2002) there are no dramatic peaks in the 10-40 kHz range, nor any indication that there is a peak near 40 kHz.

Cruise ship 1/3-octave spectra

Gray, L. M., & Greeley, D. S. (1980). Source level model for propeller blade rate radiation for the world’s merchant fleet. The Journal of the Acoustical Society of America, 67(2), 516–522. doi:10.1121/1.383916
Arveson, P. T., & Vendittis, D. J. (2000). Radiated noise characteristics of a modern cargo ship. The Journal of the Acoustical Society of America, 107(1), 118–129. doi:10.1121/1.428344
Kipple, B. (2002). Southeast Alaska Cruise Ship Underwater Acoustic Noise (Technical Report No. NSWCCD-71-TR-2002/574) (p. 92). Naval Surface Warfare Center – Detachment Bremerton. Retrieved from http://www.nps.gov/glba/naturescience/upload/CruiseShipSoundSignaturesSEAFAC.pdf

# Refining ship noise paper & figures

Working through figures that Val generated in PDX while we are both back in Seattle (12-6pm).  Trying to get figures updated with latest data set (northbound only, as our spreading experiment only extended into northbound lanes).

For comparison with our passenger vessel class, we digitized a few plots from (Kipple, 2002) and (Kipple, Kollars, 2004). Then we put the ship source spectral data in a Google spreadsheet.

Learned you can insert an OLE object into LibreOffice document, including a spreadsheet (linked if you like), or a chart.  Managing tables is a pain in LibreOffice, especially adjusting the column width.

Also learned that there is a serious LibO bug related to caching images in documents.  It probably explains why I was experiencing horrendous lags and jumps when trying to scroll through the figure section of the ship source level paper.  It seems to have helped to bump up all memory allocations by about an order of magnitude in the LibO Preferences.

Kipple, B. (2002). Southeast Alaska Cruise Ship Underwater Acoustic Noise (Technical Report No. NSWCCD-71-TR-2002/574) (p. 92). Naval Surface Warfare Center – Detachment Bremerton. Retrieved from http://www.nps.gov/glba/naturescience/upload/CruiseShipSoundSignaturesSEAFAC.pdf
Kipple, B., & Kollars, R. (2004). Volendam Underwater Acoustic Levels (Technical Report) (p. 7). Naval Surface Warfare Center – Detachment Bremerton.

# Hydra: a step towards marine telemetry data sharing

Today I finally was able to log on to Hydra a web site that facilitates data sharing among researchers who track the movements of aquatic animals in the Pacific Northwest, U.S.

The whole proprietary tag/receiver technology maintained by Vemco leaves a *lot* to be desired from an open science perspective.  By all rights they should be the ones making it easy for their users to share data.  But Vemco has dominated the market and played defensively for so long that the community here in the progressive Northwest was forced to create Hydra to enable some degree of organized collaboration and data sharing.

As we finish up the final field work for our study of how adult (lackmouth) salmon move in the San Juan Islands, today I’m cleaning up receivers, offloading data, uploading detection files to Hydra, and preparing the receivers to be returned to their owners.  Also, I organized and uploaded to Flickr a bunch of photos from the project.

Data management progress:

• Reset Panasonic toughbook clock to NIST
• Verified data offloaded from Iceberg and Patos receivers (101004, 110848) and prepared for storage
• Offloaded data from previous batch of recoveries (9 receivers) and backed up all data to cloud
• Still need to offload data from latest batch of recoveries (5 receivers)
• When interpreting results, be sure which time Vue exports: local or GMT!  Checking in the Vue software options, I see time is set to UTC.

More progress on 11/14/13:

• Uploaded yesterday’s csv files to Hydra after verifying that drift was no more than 5 minutes for any deployment during our study.
• Updated metadata, including: made new sheet to match Receiver Deployment upload format, copied metadata to it, then converted recover dates to End_Date column; then converted all lon/lat pairs to decimal degree format (and checked/corrected all for accuracy via Google map of deployments)
• At first attempt at deployment metadata upload, I got “Manufacturer” error.  I tried re-formatting Google spreadsheet date as YYYY-MM-DD, removed a () from a site name, removed all blank lines before downloading the .csv file, and then simplified the file name (no spaces) before re-uploading.  That got me this error:
Unable to parse start date: Date string was 2011-11-15, and did not match YYYY-MM-DD HH:MM:SS or MM/DD/YYYY (time data '2011-11-15' does not match format '%Y-%m-%d %H:%M:%S')
• Then I tried switching back to the Google date format of MM/DD/YY even though the output CSV file does not display double digits for the first 9 months or days (e.g. 3/4/2013 not 03/04/2013).  Hydra didn’t seem to mind about that though –
• receivers-recovered-mmddyyyy.csv processed successfully, Deployment File, 32 rows, 0 rejected.
• Finished offloading data from last batch of 5 receivers (but did not open cases as they still need final cleaning): 101589, 100914, 101594, 101621, 108402.

More progress on 11/15/13:

• Backup and upload 5 most recent data files.
• Experiment with Hydra mapping and sharing functionality (seems to leave a lot to be desired wrt figuring out anything about the actual fish whose tags we’ve detected), finish feedback email to Hydra administrators, and send it.
• Emailed Anna+ with updates, links, and .zip of latest data files
• Upload to Google workbook and compare 2012-2013 tag list with rough notes on detections (column P).  I find zero overlap.  Bummer!

# Pondering noise levels from cavitation

Our spreading experiment results and ship receive levels are suggesting that different ships within a single ship class put out different amounts of high-frequency noise (at the same range from our hydrophone and the same speed over ground).  Could this be a manifestation of cavitation within many or all classes of ships?

A clue could come from experimental (e.g. laboratory) measurement of cavitation noise.  How loud is cavitation from propellers at the standard range of 1 meter?

Thus my literature search began.  (Blake, Wolpert, Geib, 1977) show some interesting figures of source spectra for cavitation that have minima near 6 kHz and increasing levels not only at higher frequencies (up to 80kHz) but also at lower frequencies (down to 1 kHz where levels are actually above those at 80kHz).   The figure notations in (Kipple, 2002) suggest that cavitation noise primarily occurs at frequencies above 10-20kHz.

They also show a plot of spectral density of cavitation noise at 31.5 kHz that could be compared to single ship spectrum presented by (Hildebrand, 2006), as well ours…  Hildebrand computed the 31.5 kHz source spectrum level of the Hanjin to be ~110 dB.  The corresponding Blake levels range from 75-100 dB, suggesting that the Hanjin is *really* cavitating!

# Towards open science at Beam Reach

Herein begins my first open science notebook.  Inspired by my friend, Eli Holmes, who shared her open notebook during the recent Federal furlough, I started reading about open notebooks and decided that maintaining one would fit into my on-going efforts to refine and open my scientific workflow.

As 2013 ends I’ll be working on three projects.  Collaborating with Val Veirs and Jason Wood, I will finish a paper on the underwater noise made by ships within the summer habitat of southern resident killer whales.  I will daft a synopsis of sound (noise and signals) in the Salish Sea for the Encyclopedia of Puget Sound.  And I will wrap up a salmon tracking project in the San Juan Islands that was organized by Tom Quinn (UW) and Kurt Fresh (NOAA/NWFSC) with Anna Kagley and others.  These projects will supplement the teaching and research (wiki) and blogging I conduct through my non-profit, the Beam Reach Marine Science and Sustainability School.

Along the way, I expect to continue exploring tools that boost my scientific productivity and support my long-term goal of opening science.  I am writing in Open Office and maintaining my references, bookmarks, and notes with Zotero, while keeping an eye on tools like ShareLaTeX.  I use GMT to make maps and we are using R to generate figures for the ship source level paper.  While surveying our open access publication options, we are watching the release of Libre (this November!?) in the hopes of engaging its open peer-review process.