This is my attempt to illustrate how we can detect whale calls on an ocean bottom seismometer.  But this is not typically how people measure sound underwater!  Usually, they use an instrument called a hydrophone.  Here’s an example of a Reson TC4033:

Pretty, isn’t it? (A little aside: I chose this one because I used it to collect data for my master’s project and I also worked at Reson for two years.  Oh memories…).

Hydrophones are basically underwater microphones. They contain a special crystal called a piezoelectric transducer (my friend Alexis made a sweet blog post describing these: magic crystal blog post) Remember what a transducer is? It converts between different forms of energy, in this case acoustic to electric.

Good quality hydrophones are calibrated so that if you measure a certain voltage (at a certain frequency), you can say exactly how loud the sound wave was when it hit the transducer. There are all kinds of reasons why measuring the loudness (amplitude) of a sound wave is useful when you’re underwater. For example if you measure a really loud sound, it might be closer, or a really quiet sound, it might be further away.  It’s really much more complicated, because sound travels in all sorts of weird and wonderful ways underwater, but you can think of it like that for now.

A lot of underwater acoustic methods were developed to try to track submarines. (photo from Wikipedia, taken by an employee of the US Navy)

But guess what?  Measuring the loudness of a fin whale call (or any whale for that matter) can tell us lots of useful things.  Like how far can they communicate with each other?  How will ship noise disturb their communication?  How far away are they from that hydrophone you’ve got dangling off the side of your boat?

The problem with trying to do this on seismometers is that they are not calibrated to measure the loudness of sound – they just measure how much the ground moves.  So how do we deal with that?  Find out in another (future) post!  Ooh, it’s cliffhanger!

Here’s the basic idea:  an earthquake happens, at some unknown location.  Seismic waves are generated, and these waves travel through the earth.  The waves arrive at each ocean bottom seismometer (OBS) in the network at a slightly different time.  The OBSs record the signal, and then the differences in arrival time between each of them are used to back out both where and when the quake happened.

I’m not going to go into the math-y goodness here, because it’s complicated and I’d likely make loads of mistakes and it would all end in tears.  But one thing that is very, very important is timing.  We need to know those arrival times REALLY accurately.  One really handy-dandy way to deal with timing is by using GPS time.  Many GPS receivers, even pretty inexpensive ones, are equipped with a very accurate timer that puts out one pulse exactly once every second (aka, PPS!).  Problem is, you don’t necessarily know which second it is… until you get a GPS fix!  Then you are all set.

Oh wait.  Guess where GPS doesn’t work so well?  At the bottom of the ocean.

Scientists deal with that little speed bump by calibrating the time immediately before deployment and again immediately after recovery.  A correction is then applied to all the times in between based on characteristics of the instrument (or some other black magic).  Clock drift is just one of those things you’ve got to deal with when working with instruments that sit underwater for any length of time.

BONUS!

Rob posted an interesting comment, and I thought it might be good to just put it up into the body of the post:

Clock drift is a very annoying little problem that us seismo-dweebs like to brush under the rug as much as possible. The problem comes about, basically, because the crystal that is used inside the instrument to keep track of time is inherently flawed in some way. There are people developing new seismometers with “better” crystals that will mitigate this problem, but it is likely an issue that won’t be going away anytime soon.

I started with the question, “what is a seismometer?” because, well, that’s where it all begins.  Our dataset was not collected to measure whales, oh no.  It was collected to measure earthquakes.  Seismometers on the seafloor have essentially the same function as a seismometer on land, except for some obvious things, like being super-duper water-tight for one.  Electronics + water?  Not usually a good combo.

You’ll notice that my OBS drawing is a red box.  Not so in real life, my friends!  I’m just too lazy to draw one of THESE puppies (look for the part drilled into the rock):

Two reasons:  (a) Too hard to draw, and (b) too late to change it now.  Never mind though.  Henceforth, when I draw a seismometer, it will be a red box.  OBSs are often just deployed over the side of a ship, and wherever they land is where they’ll stay.  But the ones that were used in our study were installed using an ROV (remotely operated vehicle), and were either buried in sediments or drilled into basalt, like in the photo above.

The seismometers that we use were deployed for a year at a time, three years in a row.  So they had to have the ability to store a year’s worth of data.  That’s 128 samples recorded each second for each of the three axes of the accelerometer.  There are about 31.5 million seconds in a year, so that’s in the neighbourhood of 12 billion samples.  Holy smokes, that is a lot.

I’m not sure if the world has been formally introduced to the Quakes and Whales team:  World, Quakes and Whales.  Quakes and Whales, world.  We have signed ourselves up for a whole world of hurt, aka STP, or the Group Health Seattle to Portland Bicycle Classic.  See my one previous entry about it here.

The scene described in the comic above is pretty much true.  And it happened BEFORE we even started our actual ride.  So Dax tricked us into climbing an enormous hill (he claims not to have planned it, but he is ridiculously fond of biking up giant hills, and of convincing others to join his mad hijinks).  After we biked back down, we finally embarked on our planned route, which was a 43-mile loop between Redmond and Carnation.

Trooper and I walked down to Greenlake yesterday evening.  The sun was out, the sky was blue, and it was not terribly cold out.  I tied Trooper to the leg of one of the patio tables outside Ben and Jerry’s and went in to get some ice cream.  As you know, Trooper is a dog.  He doesn’t have that much excitement in his life.  In fact, going for a walk is the highlight of his day.  So when it occurred to me that he might enjoy a cup of vanilla ice cream, I didn’t hesitate.  After paying for my order, I carried our ice cream out to Trooper, who was, by this time, drooling in anticipation.  I set the ice cream down and let him have at it.  Once he was all set up, I leaned back in the patio chair and started working on my own cone of strawberry cheesecake.  That’s when I noticed a couple walking past us with their dog.  Their dog looked very jealous of Trooper, naturally.  The owners, though – they were shooting me the stink eye.  They looked at the ice cream cup. They looked back at me, they looked at Trooper, and back to me.  I was officially being judged (negatively) for my dog-parenting skills.

Oh well.  As Lora put it later on, “If feeding my dog ice cream is wrong, I don’t want to be right”.

I’ve done one training ride (ie. a ride that’s longer than my commute).  It was last Tuesday with Dax.  Google maps tells me that it was about 30 miles in total.  It took me 2.5 hours.  Awful.  I thought I was dying, especially at the end.  My house is on a hill, so the last five miles or so were just a kick in the teeth.  As soon as I got home I wanted to eat like five pounds of poutine.  Which, luckily, was not available to me at that point.

Tomorrow, Sunday morning, we have another training ride planned.  This time we are making the (to me) very ambitious journey to the Red Hook brewery in Woodinville.  That’s a total of about 14 miles longer than what we did on Tuesday.  Only this time it’ll be with Rob and Lora too.    The very best part of this plan is that in the middle of those hellish 44 miles, we are stopping for food and beer.  Sweet nectar of the gods.

Here’s a crappy little drawing I made.  I hope it illustrates how I feel about the ride we’re supposed to do tomorrow morning.

Measuring source levels of marine mammal vocalizations is complicated, and even more so when using ocean bottom seismometers. I’m trying to look at some of these complications and sources of error and uncertainty.

One of these is the interference between the direct path arrival and the surface bounce. Because they take different paths to reach the receiver, they arrive at slightly different times. These offsets result in interference patterns – sometimes constructive and sometimes destructive.

R_1 and R_2 are the direct path and the surface bounce. D_s and D_r are the source depth and receiver depth. H is the horizontal distance between the source and the receiver.

I wrote some code to model these effects, and I’ll just start out with a little video clip that shows what I mean by constructive and destructive interference. I’m plotting the RMS (root-mean-square) amplitude of the received signal. You can see that in addition to the interference pattern, the amplitude of the input signals is decreasing – this is to account for transmission losses along the travel paths, modeled simply using a spherical spreading assumption (scaled by range).

Next, I wanted to see how this effect might manifest itself in our particular setup.  I ran the code using an approximate receiver depth of 2200 meters, a source depth varying between 0 and 100 meters, and a horizontal range of 0 to 2200 meters.  I chose the source depths based on what I think are likely depths from which a fin whale might call.  The horizontal ranges are restricted such that the incidence angles will be small(ish) – a constraint that is imposed in part to reduce ambiguity with later multipath arrivals, and in part because of the physics of converting ground motion back to an acoustic pressure level (details I won’t go into here).  The surface bounce is given a 180 phase flip (and no loss of amplitude) since the surface is treated as a perfect pressure release boundary.

Here are the results of that model:

Interference patterns for a series of horizontal ranges and source depths.

Wow, that is a lot of possible variability!  This has been just a quick little experiment, and there’s a significant possibility that I’m still doing something wonky in my code, but based on looking at examples from the literature (for example, Charif et al., 2002), it seems to be in the right ballpark.  Very interesting – this will definitely affect how I interpret my source level estimates.

There are a few options for doing this. From what I can gather, it’s not something that is easily done within Python (someone correct me if that’s wrong!). So instead, I’m creating a series of image files (.png’s), and converting to video outside Python. I looked at a couple of options and settled on ffmpeg, which seemed really simple.

I grabbed it from fink:

fink install ffmpeg

All of my image files were saved in a single folder, with names like 001.png, 002.png, … 198.png. Converting them to video format:

ffmpeg -i %03d.png -vb 1024K video.mpg

The -i flag is for input file(s) and -vb is video bitrate. Here’s the output:

The whales probably aren’t really trying to talk to us.  But there is a good chance that they use those low frequency sounds to talk to each other.  And being able to get their messages to each other is probably very important for their livelihood – they might be telling each other about a delicious krill buffet.  Or maybe they need to attract mates with a whale-themed pick-up line.  Either way, it’s kind of bad if the noise in the ocean is so loud that they can’t hear each other.

found at http://abbotlab.wordpress.com/2011/04/12/the-new-depths-of-noise-pollution/

HAPPY BIRTHDAY, DAX!