Listening to the deep water soundscape off East Antarctica
Thursday 10 May 2018
This week’s AAD seminar will be presented by Dr Brian Miller (Australian Antarctic Division). Brian is a marine mammal acoustician interested in passive acoustic population surveys, localisation, and tracking, as well as the effects of man-made noise on marine mammals. In his talk Brian will provide an overview of the acoustic data collection and outline how these data are used.
Abstract: Since 2004, the Australian Antarctic Division (AAD) has collected over 100,000 hours of underwater sound recordings in the Southern Ocean off East Antarctica. The bulk of these recordings were made by mooring autonomous acoustic recorders at deep water sites along the annual resupply routes to Australia's three research stations in East Antarctica. Many sites have now yielded several consecutive and/or non-consecutive years of data. Here we present a brief overview of this rich acoustic dataset including samples of the many weird and wonderful sounds that we typically detect.
While early data were originally collected to listen for critically endangered Antarctic blue and endangered fin whales, over the years the number of species recorded has increased in step with improvements in hardware and digital storage. Since 2013, acoustic recorders have provided information on the presence and behaviour of many Antarctic top predators including crabeater, leopard, Ross and Weddell seals, sperm whales, and all species of high-latitude, southern ocean baleen whales (Antarctic blue, fin, humpback, Antarctic minke, sei and southern right whales). We will explore the advantages and challenges of remote acoustic monitoring of marine mammals in the Southern Ocean. We begin with investigation of environmental noise, which in the Antarctic is driven largely by wind and ice. Then we present the highly seasonal contributions of blue, fin and minke whales and leopard seals to the soundscape. Lastly, we present systematic observations of sperm whales and preliminary observations of crabeater seals and humpback whales. We aim to continue acoustic data collection and analysis via the Southern Ocean Hydrophone Network (SOHN), a joint long-term project of the IWC's Southern Ocean Research Partnership (IWC-SORP) and the Southern Ocean Observing System (SOOS).
Please come along and learn more about this fascinating topic! As always, the seminar will be held in the theatrette of the AAD (ground floor). All welcome!
AAD Seminar Team
Listening to the deep water soundscape off East Antarctica
Video transcriptMike: Welcome to the AAD seminar series. Today we’ve got Brian Miller who will give our seminar. Brian did his undergrad at Boston University and graduated in 2003 – so you can to the maths. He did a Bachelor of Science in biomedical engineering and he also did a Bachelor of Arts in biology. I don’t know what to say about that… maybe it’s about how the US think about science, I don’t know. He then went on to complete a PhD at Otago University and it was titled “Acoustically derived growth rates and three-dimensional localization of sperm whales in Kaikuora”. Brian joined the AAD in 2011 as our cetacean acoustician replacing Jason Gedamke, who many of you will know. Brian almost immediate became an integral member of the whales team and has led acoustic teams on domestic and three Antarctic voyages and will lead another team in January 2019 on AAD’s RV Investigator’s voyage down to East Antarctica. He’s also a leader of the Southern Ocean Hydrophone Network group which is an initiative of the International Whaling Commission’s Southern Ocean Research Partnership. Brian’s talk today is title ‘Listening to the deep water soundscape of East Antarctica’.
Brian: Thanks, Mike. Okay, everyone can hear me okay? I’m really happy to be here today, to be able to share this talk with you.
Just want to talk a little about AAD’s passive acoustics program. It’s literally sound science so passive acoustics means listening to sounds and we focus on underwater sounds. Our stakeholders are government, industry and the science community, and as Mike mentioned, one of the primary stakeholders is the International Whaling Commission and we deliver our results through the Southern Ocean Research Partnership. Early this year, our group joined the Southern Ocean Observing System [SOOS] as well. So we’ve now created a technical working group – a capability working group – for the SOOS. Of course our work is delivered through the Australian Antarctic Science Program. There are a couple of common themes that keep arising with passive acoustics work. The most basic theme is simply passive acoustic surveys. This is listening for and counting animals and determining where they are. It doesn’t have to be animals; passive acoustics originally, I believe, started more as a military discipline monitoring for submarines. And geologists for a long time listened to the sounds of the Earth. All of those things could be considered passive acoustics. Animal communication and behavior is a really important aspect of passive acoustics. The more you understand about why the animals produce the sounds the more use you can get out of listening. A more recent topic that has become prevalent in this community is the effect of sound, of man-made sounds, on marine animals.
So what are the sounds, the underwater sounds, in the Antarctic? As I mentioned in the previous slide there are physical sounds, there are man-made sounds and there are biological sounds. Physical sounds can be sound from wind, precipitation, whether rain or snow, the sound of sea ice, as well as earth quakes. Man-made noise can come from ships whether that’s the propulsion, cavitation from the propellers, or sounds of machinery onboard the ship that are transmitted through the hull, or instruments that go ping or make other noises. But the sounds that I like to focus on are the biological sounds. A lot of different types of animals make noise; in the Antarctic the main sounds we are listening for are sounds from seals, but predominantly sounds from whales and dolphins were the sounds that really drove the conception and the bread and butter of our program. There are four seals in the Antarctic that make noise underwater regularly. These are crabeater, leopard, Weddell and Ross seals and there are about a dozen whale species that make sounds. So all baleen and toothed whales make underwater sounds and the instruments we are putting out at the moment can record sounds from blue, fin, humpback, Antarctic minke, southern right, sperm whales and killer whales.
This slide is just to illustrate the variety of different platforms that we can use to collect passive acoustic data. Passive acoustics, the main sensor that we use is called a hydrophone and it’s essentially an underwater microphone; it records the pressure wave form of sound underwater. In the top left corner, we have a simple directional hydrophone. It’s just a handheld device with some headphones. Here in the top right we have a more complex device which is AAD’s moored acoustic recorder. Some of you may be familiar with that or even helped to create or deploy that. At the bottom left we have sonar buoys; sonar buoys are directional listening devices that can be deployed from a ship and send their sounds back to the ship via radio link. And in the bottom right we have a towed radiophone array. These are devices that can be towed behind the ship for thousands of kilometers to listen for sounds while the ship is underway. So all of these include hydrophones, all of these are different forms of passive acoustic monitoring.
It’s a great time to be developing small sensors! In addition to the traditional forms of acoustic monitoring that I showed, there is a whole variety of platforms that we can put hydrophones on as well. These include autonomous underwater vehicles, profiling floats, autonomous surface vehicles and even on suction-cup tags that can be deployed on animals directly. So it’s an exciting time to be in this field.
What is it that we can learn from passive acoustics? Well, the main thing that we can learn is what species of animals are present and calling. We don’t learn anything about animals that are silent but we can about what species are present and calling. Most platforms allow us some idea about where those calling animals are located. For certain species or certain populations, we can also learn what those animals are doing. Perhaps there are sounds that are associated with feeding or with mating. Certain populations within a species might have distinct vocalization. So might learn about the population or the sex of the animal that is making these sounds. Males might make different sounds to females perhaps and if we know enough about the biology of the animals and what drives this sound production, we can learn about the abundance of those animals as well.
For the rest of the talk I’m going to focus on long-term acoustic recordings that have been made by the Australian Antarctic Division. The small map in the corner there is showing the location of sites where AAD has conducted long-term acoustic recording. This table is showing years for which we have long-term acoustic recordings. So in this table, a black X indicates a recording site that has approximately one year of data, the empty spaces indicate years when we didn’t attempt to collect data, and the little red sad faces there are unfortunately years we attempted to collect data but weren’t successful for some reason or another. Leaving instruments in the ocean for a year is always an activity that’s – leaving anything in the ocean for a year – is an activity that is fraught. Over the years we’ve used a variety of different instruments to collect these sorts of data. In the early years, we borrow instruments from SCRIPS Institute of Oceanography. This instrument is called and ARP or acoustic recording pack and these were deployed from 2004 to 2006. From 2006 to 2009, a device called a CMST Logger from Curtin University was deployed, and in 2013 the AAD deployed its first AAD moored acoustic recorder. Now these moored acoustic recorders were developed, they were basically conceived, designed and developed and operated entirely by the AAD and at this point I just like to thank the AAD’s STS group for their role in doing all of that. So we’re really, really lucky to have the STS group here. We’re able to take an idea, design precision-engineered instruments to do this job. The instruments that the AAD has designed allow us to record with better fidelity, longer duration than any of the instruments that came before.
Our project, our work here at the AAD is part of a bigger picture and you see here the filled circles showing the AAD recording sites that were the sites from that previous map. The open circles show all of the other recording sites that have been conducted in the Antarctic south of 60 [degrees latitude] by other nations. This map is showing all of the recording sites. The first Antarctic recorders were deployed in 2001. The AAD started in 2004, so we are just a few years behind there. This is part of what Mike mentioned already, the Southern Ocean Hydrophone Network or the SOHN and this is an initiative of the International Whaling Commission’s Southern Ocean Research Partnership and also the Southern Ocean Observing System. So the SOHN has a really long-term focus. We’re really aiming to characterize changes over long time periods. We’re looking at recording sound for more than a decade at each of these sites where we’re able to. It has a circumpolar focus; we’re building up to that circumpolar focus. You can see there are regions on this map that have no recording sites and we are keen to engage partners to fill in those gaps. International collaboration for the SOHN is absolutely important to deliver on these goals and the sorts of data we aim to collect with the SOHN are baseline ocean noise data to see how the ambient ocean noise changes over time. But as I already mentioned many of the group members focus on the distribution, occupancy and abundance of marine mammals or at least marine mammal sounds that we hear.
These lovely people are the IWC-SORP Acoustic Trends Working Group. These are the members who collect data and analyse data for the SOHN, and the goal of the working group is to use passive acoustics – this is the overarching goal – to use passive acoustics to deliver cost-effective information on Southern Ocean whales. The specific tasks that the group has been undertaking in recent years have involved coordinating, trying to match people who have ship-time to deploy instruments with people who have instruments, people who have analytical capacity to analyse the data. The group has also been working towards standardizing the data collection and analysis of these data. Each nation can potentially develop their own instruments and have slightly different standards and slightly different analytical approaches. That can frustrate a more holistic circumpolar approach to that analysis. The group also aims to develop novel systems, new instruments, new software, new analytical techniques to facilitate the collection and analysis of the data. And lastly the group has been working to provide advice and build international capacity to expand the network of recording sites.
Getting back towards the sound we hear in East Antarctica, marine mammals produce a wide variety of underwater sounds and I know in the abstract I promised we’ll play some sounds and we’re just about there. We can identify distinctive sounds from different species. There are a lot of sounds that are difficult to classify to species. So some species sound similar to each other but for the most part, every species that we have recorded has some sort of distinctive sound that we are able to identify and distinguish them. We’re not typically – this is a real common question so I just get out in front of it – we not typically able to identify individual singers. We can say that’s blue whale, that’s a fin whale, that’s a humpback whale but we can’t at this point say that’s Bobby, that’s Franky, that’s Harry. For some species passive acoustics can be a really, really cost-effective means of studying them and unsurprisingly the species that are most cost effective are the species that are rarely seen by other methods and that are highly vocal. So animals that make a lot of noise are perfectly amenable to passive acoustics, especially if they are rare and rarely seen from ships. In the Antarctic, these species pretty much are sperm whales, fin whales and blue whales. Alright, that’s enough background information, why don’t we have a listen to some of these animals. I’m going to play these sounds. There’s going to be a little animation called a spectrogram that shows a visual representation of the sounds and see if you can guess what animals are making these sounds.
[sound playing, spectrogram showing but sound difficult to hear] I started with an easy one. Any guesses? Just shout it out… Yeah, that’s right, that’s a humpback whale
[next sound playing…] Not a thunderstorm… it’s actually noise from ice. Ice can be really, really noisy, especially when it’s crushing into other ice or breaking up.
[next sound playing…] So you can see I had to play the sound that faster than it was originally recorded because it wouldn’t be audible otherwise. So it’s a very low frequency sound. Any guesses? These are 20 Hertz pulses produced by fin whales.
[next sound playing…] That went on a bit longer than I remembered. Any guesses? [audience: a seal of some sort?] It’s not a seal… it’s a big dolphin, it’s a killer whale. They’re just big dolphins.
[next sound playing…] It is mechanical, absolutely. These are the bow thrusters of the Akademik Treshnikov, a Russian ice breaker that I was on during the Antarctic circumnavigation expedition last year.
[next sound playing…] Anyone recognize that? Those are echo location clicks of sperm whales.
[next sound playing…] No guesses? This funny sound is called the bio-duck produced by Antarctic minke whales. The bio-duck was only recently linked to Antarctic minke whales due to an acoustic suction-cup tag being placed on a minke whale four years ago. The sound had been recorded by submariners for decades and they were never really sure what it was. They thought it might have even been some sort of Russian communication system but it turns out that it was Antarctic minke whales.
[next sound playing…] That was produced by a leopard seal.
[next sound playing…] That’s one I have played many times before in seminars. Anyone remember? No? It’s a blue whale, of course. It’s an Antarctic blue whale, that’s the song of Antarctic blue whales; they repeat the same units again and again and again for hours on end.
So I mentioned with passive acoustics we can learn what species is calling; they’re all sound quite different and distinctive from each other and we can learn about the occupancy and the temporal and spatial distribution. I said we wanted to focus on long-term recording sites and the figure at the bottom is what we call a long-term spectral average. It’s a way of visualizing an entire year of sound and all at once. This is a long-term spectral average from our site at the southern Kerguelen Plateau. We’ve got time on this x-axis here and we’ve got the frequency of sound on the y-axis. So it’s like looking at one of those spectrograms, one of those visualizations that I just showed but it’s quite compressed and it’s showing the average. Each vertical pixel corresponds to one hour of sound. So there is a lot of compression going on because the resolution of the display is far less than the number of hours in a year. But it gives us a good sense of what’s making noise. These horizontal bands of sound are tonal calls from animals, and the vertical bands are sounds, increased background noise, from elevated wind and waves. Just to illustrate the effect of wind, these moored acoustic recorders, they sit deep in the ocean - they are two to three kilometres deep – but we still pick up the effects of storms that pass overhead. The top graph is now showing the average wind speed at this site in six hour increments and the wind speed was measured from remotely sensed data. So it’s a product that gives us sort of two and half degrees spatial resolution and six hour temporal resolution. If you look closely, you can see some of these peaks in the wind speed correspond to some of the peaks in the long-term spectral average. If you don’t want to correlate it by eye, we can actually plot it up here now we have wind speed on the x-axis and we have the sound energy on the y-axis and you can see that there is a positive relationship there.
So where to the animals fit in? Antarctic blue whales, the low frequency sounds – we had to speed them up – so these come in at the bottom of the spectrogram here. Fin whales actually have a lot of overlap with blue whales; you can see these low frequency pulses also show up on the long-term spectral average towards the bottom. But if you see this little blip up here, this also shows up on the long-term spectral average as that narrow, yellow band from April to about July. So this is the season for blue and fin whale song.
Leopard seal sounds. We really mainly detect them throughout December and early January and if I recall correctly, this is during their breeding season. So these are highly seasonal calls.
Antarctic minke whale, the bio-duck sound, is kind of the odd one out. They are the only species really making noise throughout the winter.
Our ice noise. You can hear ice throughout the year in the Antarctic whether it’s frozen solid or whether it’s being broken up. But what I really like to illustrate with this is that during the months when the recorder is not covered by ice, we actually have a real increase in low frequency background noise, and during the months when the recorder is covered by solid ice we have a real decrease in that low frequency noise level. The ice is a really important part of the soundscape in East Antarctica. So putting it all together, we can look at this long-term spectral average and this can basically be considered the soundscape of East Antarctica. We can kind of work out when and where species are based on simply comparing these long-term spectral averages. The top figure is showing the spectral average for the Casey site, a site along the Casey resupply route, and the bottom figure is showing the spectral average of the southern Kerguelen Plateau and one of the main things that seems to be different between these two figures is the colour of this band down here. This is representing find whale calls; these seem to be absent at the Casey site. Both of these spectral averages are showing the calendar year 2014. There was a little bit of an offset; the Casey recording was put in first, the one on the Kerguelen Plateau was set up a little bit later but otherwise there is a lot of overlap here. Another thing that seems to be more present on the south Kerguelen Plateau is the minke whales. They are a lot fainter at our Casey site.
We can also compare multiple years of recording within the same site. Now we’re looking at three years of long-term spectral averages, all from the southern Kerguelen Plateau, from 2014 to 2016, and you can see the intensity of the fin whale band seems to be quite variable throughout the year. Minke whales were very present in 2014 and 2015 but they seemed to be a little bit fainter in 2016. And the leopard seals, you can see again really producing sound during that breeding season and all of the sounds seem to be quite seasonal when they are present.
So I’ve been talking for a while now, and I’ve got a bit more to go but I try to get through this quickly. Some of the species that are really dominant in the soundscape we’ve had a listen to and looked at, and these long-term spectral averages are really a synoptic picture of a recording site for a year. But really, fully make use of the data we actually have to … we might be interested in sounds that are produced less frequently, that are intermittent or possibly rare. A lot of species don’t make noise as frequently as the ones that dominate the soundscape. Some species do make noise quite frequently but because of the characteristics of the sound they produce, they don’t show up on these long-term spectral averages. So other analytical techniques are required.
The sort of techniques we like to employ are often called Detection, Classification, Localisation and Density Estimation. That’s a lot to say so it does get abbreviated to DCLDE which is still a lot to say. Just conceptually, Detection can be thought of as separating out signals of interest from the background noise. Classification can be thought of as identifying different types of sound. Localisation is simply the determination of a sound source, and potentially the movement and tracking of a sound source, and Density Estimation is just a fancy way of counting, saying that we want to quantify the number of sounds or animals that we hear.
This suite of techniques is really a multi-disciplinary endeavor and it draws from engineering, computer and data science, physics, biology and statistics. You really can’t focus on any one of these fields and get anywhere in terms of actually understanding passive acoustics. You have to involve all of them to some degree. I won’t talk too much about this but happy to talk afterwards. Just wanted to show it’s truly a multi-disciplinary endeavor and it’s not purely a biological study, it’s not purely physics, it’s not purely engineering but it’s a combination of all of these things.
Just to give a quick illustration of an example, you know sperm whales are a very, very vocal animal. But they weren’t present in that long-term spectral average. There is no indication that there were sperm whales in that long-term spectral average. Here are sperm whale clicks, here’s a spectrogram of sperm whale clicks showing 10 seconds of data, ten seconds of audio and the clicks are these vertical bands. So the majority of time there is no signal and for very brief instances these echo location clicks are present. So we can create a detector by looking at the energy in these vertical bands. We can think of the energy in the clicks as a detection function. The bottom is now showing our detection function and we can apply a threshold to that detection function – represented by that red line – and where the detection function exceeds that threshold we can call that a sperm whale click. We can extract those clicks and we can basically detection and classification can be thought of as transforming the audio into a signal that is more friendly. So it’s signal processing and data science.
Once we have perhaps detected the clicks, we can perform this feature extraction – so additional transformations – to separate out clicks, any sort of clicks, so our energy detector will not just detect sperm whale clicks but it might detect clicks that come from ice or other species. So we can take the extracted clicks, the detected clicks, and transform them yet again, in order to try to better represent the signal of interest. So here we’ve got our long-term spectral average again. The pink X shows a time period that has been manually inspected and has been determined by an analyst to have sperm whales present. The red cross indicates a time period when there are no sperm whales present. we can extract features from the audio for each of those times to separate out these signals better and then transform the audio into these features and – low and behold – after transformation we’ve got a dark spot on here that gives a good indication that sperm whales were present in this time period but not in that [points to empty space] time period. I don’t want to go into the details but happy to talk afterwards. If we apply these methods to the entire data set we can start to answer questions about the presence of sperm whales. So now top figure is showing, the blue line is showing the proportion of hours per day for which our detector and classifier have determined sperm whales are indeed present and the grey boxes are showing the proportion of days per month for which our classifier detected sperm whales. There is a faint red line in here that shows the proportion of ice cover and you can see that sperm whales are really not detected in the winter at all.
So this is just an example of detection and classification. If we actually want to count animals, we move beyond just detecting the presence of animals. We have to start localizing animals. A conceptually simple way to think about localization is just what is the detection range of these sounds. It’s conceptually simply to think about but actually to answer this question is really involved. It requires understanding the characteristic of the signal, the noise and you have to understand how sound travels through the environment, and you have to have good models for all of these things.
Just to illustrate how challenging this can be, this is a figure from a recent publication by
Benda-Beckmann and colleagues. This is illustrating two different models and two different thresholds for the detection of sperm whale clicks in an effort to determine the range. You’ve got the range of the clicks on the x-axis and the probability of detection on the y-axis and you can see that the worst performing model detects sperm whales out to about four kilometres, and the best performing model detects sperm whales out to about twelve kilometres. This is assuming a fixed signal, fixed noise and a fixed environment. In reality, the noise, signal and environment all change over time. So this can get quite challenging.
Finally, if we want to estimate the density of animals and not just the number of calls we detect we actually have to have some behavioural information about the animals. Just over here we have our density estimation equation. It says that the density is equal to the number of calls divided by the time, area and cue rate. So we can define the time span as one hour. Let’s assume that we can measure the detection range – not something that’s very easy but let’s assume that on average the detection range is ten kilometres. That gives us a detection area of pi r squared which is about 314 kilometres [squared]. From previous studies, people have determined that the click rate of an individual sperm whale is about 1.27 clicks per second. Each animals clicks at that rate, but they don’t click all of the time. In fact, each animal clicks only about 60 percent of the time. So with that individual click rate and duty cycle we can calculate what the cue rate is and we now have all of the information we need to estimate the density of animals. Assuming from our example that there were 2600 sperm whale clicks detected in an hour, a random given hour, we can put that into our density equation and estimate that this works out to three animals per 1000 square kilometres per hour. Now we have to be a little bit careful. We might think that we can estimate from that the total number of sperm whales that visit our site. However, if we want to extrapolate over that entire three year time period, we really need to be careful. And the situation we are trying to avoid is – I think I could explain it – but I think it’s really well expressed by this cartoon which has been hanging up in the science tea room for as long as I’ve been here. That’s the situation we have to avoid when we are trying to extrapolate density estimates over time. We can’t identify individuals, we don’t know whether we’re counting the same whale multiple times.
That’s probably a good place to end. Just to sum up, there is an increasing amount of acoustic data available. There is an exciting number of platforms and sensors and as sensors get smaller and computers get faster, more power efficient and data storage gets better we are only going to improve in that regard. There’s a suit of analytical methods available for analyzing these data but very few of these are general purpose methods or are standardized. So a lot of the work involves developing detectors that are specific to a species, classifiers that can distinguish between different species and we can do density estimation and abundance using acoustics but there is a really severe limitation because to do density estimation effectively we need to know a lot about the acoustic behavior of the animals that we’re listening to. We can conduct dedicated behavioral studies to better understand the acoustic behavior of animals. You can put acoustic tags on animals; that gives you a great source of information about their behavior but also sighting survey, simultaneous visual and acoustic surveys and focal follows of animals can be a real good source of information. This might still require some dedicated ship time but not necessarily entire dedicated voyages. This is the sum of the information we are hoping to collect on the upcoming voyage on the Investigator in January.
That’s all I’ve got. Thank you very much for listening and I am happy to try and answer any of your questions.