Detecting blue whales from acoustic data

By Dawn Barlow, PhD student, OSU Department of Fisheries and Wildlife, Geospatial Ecology of Marine Megafauna Lab

In January of 2016, five underwater recording units were dropped to the seafloor in New Zealand to listen for blue whales (Fig. 1). These hydrophones sat listening for two years, brought to the surface only briefly every six months to swap out batteries and offload the data. Through all seasons and conditions when scientists couldn’t be on the water, they recorded the soundscape, generating a wealth of acoustic data with the potential to greatly expand our knowledge of blue whale ecology

Figure 1. Locations of the five Marine Autonomous Recording Units (MARUs) in the South Taranaki Bight region of New Zealand.

We have established that blue whales are present in New Zealand waters year-round 1. However, many questions remain regarding their distribution across daily, seasonal, and yearly scales. Our two-year acoustic dataset from five hydrophones throughout the STB region is a goldmine of information on blue whale occurrence patterns and the soundscape they inhabit. Having year-round occurrence data will allow us to examine what environmental and anthropogenic factors may influence blue whale distribution patterns. The hydrophones were listening for whales around the clock, every day, while we were on the other side of the world awaiting the recovery of the data to answer our questions.

Before any questions of seasonal distribution or anthropogenic impacts and noise can be addressed, however, we need to know something far more basic: when and where did we record blue whale vocalizations? This may seem like a simple, stepping-stone question, but it is actually quite involved, and the reason I spent the last month working with a team of acousticians at Cornell University’s Center for Conservation Bioacoustics. The expert research group here at Cornell, led by Dr. Holger Klinck, have been instrumental in our New Zealand blue whale research, including developing and building the recording units, hydrophone deployment and recovery, data processing, analysis, and advice. I am thrilled to work with all of them, and had an incredibly productive month of learning about acoustics from the best.

Blue whales produce multiple vocalizations that we are interested in documenting. The New Zealand song (Fig. 2A) is highly stereotyped and unique to the Southwest Pacific Ocean 2,3. Low-frequency downsweeps, or “D calls” (Fig. 2B), are far more variable and produced by blue whale populations around the world 4. Furthermore, Antarctic blue whales produce a highly-stereotyped “Z call” (Fig. 2C) and are known to be present in New Zealand waters occasionally 5.

Figure 2. Spectrograms of (A) the New Zealand blue whale song, (B), D calls, and (C) Antarctic Z calls.

One way to determine when blue whales were vocalizing is for an analyst to manually review the entirety of the two years of sound recordings for each of the five hydrophones by hand to scan for and select individual vocalizations. An alternative approach is to develop a detector algorithm to locate calls in the data based on their stereotypical characteristics. Over the past month I built, tested, and ran detectors for each blue whale call type using what is called a data template detector. This technique uses example signals from the data that the analyst selects as templates. The templates should be clear signals, and representative of the variation in calls contained in the dataset. Then, by comparing pixel characteristics between the template spectrograms and the spectrogram of the recording of interest using certain matching criteria (e.g. threshold for spectrogram correlation, detection frequency range), the algorithm searches for other signals like the templates in the full dataset. For example, in Fig. 3 you can see units of blue whale song I selected as templates for my detector.

Figure 3. Spectrogram of selected sound clips of New Zealand blue whale song, with units used as templates for a detector shown inside the teal boxes.

Testing the performance of a detector algorithm is critical. Therefore, a dataset is needed where calls were identified by an analyst and then used as the “ground truth”, to which the detector results are compared. For my ground truth dataset, I took a subset of 52 days and hand-browsed the spectrograms to identify and log New Zealand blue whale song, D calls, and Antarctic Z calls. In evaluating detector performance, there are three important metrics that need to be weighed: precision (the proportion of detections that are true), recall (the proportion of true calls identified by the detector), and false alarm rate (the number of false positive detections per hour). Ideally, the detector should be optimized to maximize precision and recall and minimize the false positives.

The STB region is highly industrial, and our two-year acoustic dataset contains periods of pervasive seismic airgun noise from oil and gas exploration. Ideally, a detector would be able to identify blue whale vocalizations even in the presence of airgun operations that dominate the soundscape for months. For blue whale song, the detector did quite well! With a precision of 0.91 and recall of 0.93, the detector could pick out song units over airgun noise (Fig. 4). A false alarm rate of 8 false positives per hour is a sacrifice worth making to identify song during seismic operations (and the false positives will be removed in a subsequent step). For D calls, seismic survey activity presented a different challenge. While the detector did well at identifying D calls during airgun operation, the first several detector attempts also logged every single airgun blast as a blue whale vocalization—clearly problematic. Through an iterative process of selecting template signals, and adjusting the number of templates used and the correlation threshold, I was able to come up with a detector which selected D calls and missed most airgun blasts. This success felt like a victory.

Figure 4. An example of spectrograms of simultaneous recordings from the five hydrophones illustrating seismic airgun noise (strong broadband signals that appear as repetitive black, vertical lines) overlapping New Zealand blue whale song. The red boxes are detection events selected by the detector, demonstrating its ability to capture song even during airgun operation.

After this detector development and validation process, I ran each detector on the full two-year acoustic dataset for all five recording units. This step was a good exercise in patience as I eagerly awaited the outputs for the many hours they took to run. The next step in the process will be for me to go through and validate each detector event to eliminate any false positives. However, running the detectors on the full dataset has allowed for exciting preliminary examinations of seasonal blue whale acoustic patterns, which need to be refined and expanded upon as the analysis continues. For example, sometimes the New Zealand song dominates the recordings on all hydrophones (Fig. 5), whereas other times of year song is less common. Similarly, there appear to be seasonal patterns in D calls and Antarctic Z calls, with peaks and dips in detections during different times of year.

Figure 5. An example spectrogram of simultaneous recordings from all five hydrophones during a time when New Zealand blue whale song dominated the recordings, with numerous, overlapping calls.

As with many things, the more questions you ask, the more questions you come up with. From preliminary explorations of the acoustic data my head is buzzing with ideas for further analysis and with new questions I hadn’t thought to ask of the data before. My curiosity has been fueled by scrolling through spectrograms, looking, and listening, and I am as excited as ever to continue researching blue whale ecology. I would like to thank the team at the Center for Conservation Bioacoustics for their support and guidance over the past month, and I look forward to digging deeper into the stories being told in the acoustic data!

Figure 6. A pair of blue whales observed in February 2017 in the South Taranaki Bight. Photo: L. Torres.

References

1.          Barlow, D. R. et al. Documentation of a New Zealand blue whale population based on multiple lines of evidence. Endanger. Species Res. 36, 27–40 (2018).

2.          McDonald, M. A., Mesnick, S. L. & Hildebrand, J. A. Biogeographic characterisation of blue whale song worldwide: using song to identify populations. J. Cetacean Res. Manag. 8, 55–65 (2006).

3.          Balcazar, N. E. et al. Calls reveal population structure of blue whales across the Southeast Indian Ocean and the Southwest Pacific Ocean. J. Mammal. 96, 1184–1193 (2015).

4.          Oleson, E. M. et al. Behavioral context of call production by eastern North Pacific blue whales. Mar. Ecol. Prog. Ser. 330, 269–284 (2007).

5.          McDonald, M. A. An acoustic survey of baleen whales off Great Barrier Island, New Zealand. New Zeal. J. Mar. Freshw. Res. 40, 519–529 (2006).


The significance of blubber hormone sampling in conservation and monitoring of marine mammals

By: Alexa Kownacki, Ph.D. Student, OSU Department of Fisheries and Wildlife, Geospatial Ecology of Marine Megafauna Lab

Marine mammals are challenging to study for many reasons, and specifically because they inhabit the areas of the Earth that are uninhabited by people: the oceans. Monitoring marine mammal populations to gather baselines on their health condition and reproductive status is not as simple as trap and release, which is a method often conducted for terrestrial animals. Marine mammals are constantly moving in vast areas below the surface. Moreover, cetaceans, which do not spend time on land, are arguably the most challenging to sample.

One component of my project, based in California, USA, is a health assessment analyzing hormones of the bottlenose dolphins that frequent both the coastal and the offshore waters. Therefore, I am all too familiar with the hurdles of collecting health data from living marine mammals, especially cetaceans. However, the past few decades have seen major advancements in technology both in the laboratory and with equipment, including one tool that continues to be critical in understanding cetacean health: blubber biopsies.

Biopsy dart hitting a bottlenose dolphin below the dorsal fin. Image Source: NMFS

Blubber biopsies are typically obtained via low-powered crossbow with a bumper affixed to the arrow to de-power it once it hits the skin. The arrow tip has a small, pronged metal attachment to collect an eraser-tipped size amount of tissue with surface blubber and skin. I compare this to a skin punch biopsies in humans; it’s small, minimally-invasive, and requires no follow-up care. With a small team of scientists, we use small, rigid-inflatable vessels to survey the known locations of where the bottlenose dolphins tend to gather. Then, we assess the conditions of the seas and of the animals, first making sure we are collecting from animals without potentially lowered immune systems (no large, visible wounds) or calves (less than one years old). Once we have photographed the individual’s dorsal fin to identify the individual, one person assembles the biopsy dart and crossbow apparatus following sterile procedures when attaching the biopsy tips to avoid infection. Another person prepares to photograph the animal to match the biopsy information to the individual dolphin. One scientist aims the crossbow for the body of the dolphin, directly below the dorsal fin, while the another photographs the biopsy dart hitting the animal and watches where it bounces off. Then, the boat maneuvers to the floating biopsy dart to recover the dart and the sample. Finally, the tip with blubber and skin tissue is collected, again using sterile procedures, and the sample is archived for further processing. A similar process, using an air gun instead of a crossbow can be viewed below:

GEMM Lab members using an air gun loaded with a biopsy dart to procure marine mammal blubber from a blue whale in New Zealand. Video Source: GEMM Laboratory.

Part of the biopsy process is holding ourselves to the highest standards in our minimally-invasive technique, which requires constant practice, even on land.

Alexa practicing proper crossbow technique on land under supervision. Image Source: Alexa Kownacki

Blubber is the lipid-rich, vascularized tissue under the epidermis that is used in thermoregulation and fat storage for marine mammals. Blubber is an ideal matrix for storing lipophilic (fat-loving) steroid hormones because of its high fat content. Steroid hormones, such as cortisol, progesterone, and testosterone, are naturally circulating in the blood stream and are released in high concentrations during specific events. Unlike blood, blubber is less dynamic and therefore tells a much longer history of the animal’s nutritional state, environmental exposure, stress level, and life history status. Blubber is the cribs-notes version of a marine mammal’s biography over its previous few months of life. Blood, on the other hand, is the news story from the last 24 hours. Both matrices serve a specific purpose in telling the story, but blubber is much more feasible to obtain from a cetacean and provides a longer time frame in terms of information on the past.

A simplified depiction of marine mammal blubber starting from the top (most exterior surface) being the skin surface down to the muscle (most interior). Image Source: schoolnet.org.za

I use blubber biopsies for assessing cortisol, testosterone, and progesterone in the bottlenose dolphins. Cortisol is a glucocorticoid that is frequently associated with stress, including in humans. Marine mammals utilize the same hypothalamic-pituitary-adrenal (HPA) axis that is responsible for the fight-or-flight response, as well as other metabolic regulations. During prolonged stressful events, cortisol levels will remain elevated, which has long-term repercussions for an animal’s health, such as lowered immune systems and decreased ability to respond to predators. Testosterone and progesterone are sex hormones, which can be used to indicate sex of the individual and determine reproductive status. This reproductive information allows us to assess the population’s composition and structure of males and females, as well as potential growth or decline in population (West et al. 2014).

Alexa using a crossbow from a small boat off of San Diego, CA. Image Source: Alexa Kownacki

The coastal and offshore bottlenose dolphin ecotypes of interest in my research occupy different locations and are therefore exposed to different health threats. This is a primary reason for conducting health assessments, specifically analyzing blubber hormone levels. The offshore ecotype is found many kilometers offshore and is most often encountered around the southern Channel Islands. In contrast, the coastal ecotype is found within 2 kilometers of shore (Lowther-Thieleking et al. 2015) where they are subjected to more human exposure, both directly and indirectly, because of their close proximity to the mainland of the United States. Coastal dolphins have a higher likelihood of fishery-related mortality, the negative effects of urbanization including coastal runoff and habitat degradation, and recreational activities (Hwang et al. 2014). The blubber hormone data from my project will inform which demographics are most at-risk. From this information, I can provide data supporting why specific resources should be allocated differently and therefore help vulnerable populations. Further proving that the small amount of tissue from a blubber biopsy can help secure a better future for population by adjusting and informing conservation strategies.

Literature Cited:

Hwang, Alice, Richard H Defran, Maddalena Bearzi, Daniela. Maldini, Charles A Saylan, Aime ́e R Lang, Kimberly J Dudzik, Oscar R Guzo n-Zatarain, Dennis L Kelly, and David W Weller. 2014. “Coastal Range and Movements of Common Bottlenose Dolphins (Tursiops Truncatus) off California and Baja California, Mexico.” Bulletin of the Southern California Academy of Sciences 113 (1): 1–13. https://doi.org/10.3390/toxins6010211.

Lowther-Thieleking, Janet L., Frederick I. Archer, Aimee R. Lang, and David W. Weller. 2015. “Genetic Differentiation among Coastal and Offshore Common Bottlenose Dolphins, Tursiops Truncatus, in the Eastern North Pacific Ocean.” Marine Mammal Science 31 (1): 1–20. https://doi.org/10.1111/mms.12135.

West, Kristi L., Jan Ramer, Janine L. Brown, Jay Sweeney, Erin M. Hanahoe, Tom Reidarson, Jeffry Proudfoot, and Don R. Bergfelt. 2014. “Thyroid Hormone Concentrations in Relation to Age, Sex, Pregnancy, and Perinatal Loss in Bottlenose Dolphins (Tursiops Truncatus).” General and Comparative Endocrinology 197: 73–81. https://doi.org/10.1016/j.ygcen.2013.11.021.

Science (or the lack thereof) in the Midst of a Government Shutdown

By Alexa Kownacki, Ph.D. Student, OSU Department of Fisheries and Wildlife, Geospatial Ecology of Marine Megafauna Lab

In what is the longest government shutdown in the history of the United States, many people are impacted. Speaking from a scientist’s point of view, I acknowledge the scientific community is one of many groups that is being majorly obstructed. Here at the GEMM Laboratory, all of us are feeling the frustrations of the federal government grinding to a halt in different ways. Although our research spans great distances—from Dawn’s work on New Zealand blue whales that utilizes environmental data managed by our federal government, to new projects that cannot get federal permit approvals to state data collection, to many of Leigh’s projects on the Oregon coast of the USA that are funded and collaborate with federal agencies—we all recognize that our science is affected by the shutdown. My research on common bottlenose dolphins is no exception; my academic funding is through the US Department of Defense, my collaborators are NOAA employees who contribute NOAA data; I use publicly-available data for additional variables that are government-maintained; and I am part of a federally-funded public university. Ironically, my previous blog post about the intersection of science and politics seems to have become even more relevant in the past few weeks.

Many graduate students like me are feeling the crunch as federal agencies close their doors and operations. Most people have seen the headlines that allude to such funding-related issues. However, it’s important to understand what the funding in question is actually doing. Whether we see it or not, the daily operations of the United States Federal government helps science progress on a multitude of levels.

Federal research in the United States is critical. Most governmental branches support research with the most well-known agencies for doing so being the National Science Foundation (NSF), the US Department of Agriculture (USDA), the National Oceanic and Atmospheric Administration (NOAA), and the National Aeronautics and Space Administration. There are 137 executive agencies in the USA (cei.org). On a finer scale, NSF alone receives approximately 40,000 scientific proposals each year (nsf.gov).

If I play a word association game and I am given the word “science”, my response would be “data”. Data—even absence data—informs science. The largest aggregate of metadata with open resources lives in the centralized website, data.gov, which is maintained by the federal government and is no longer accessible and directs you to this message:Here are a few more examples of science that has stopped in its track from lesser-known research entities operated by the federal government:

Currently, the National Weather Service (NWS) is unable to maintain or improve its advanced weather models. Therefore, in addition to those of us who include weather or climate aspects into our research, forecasters are having less and less information on which to base their weather predictions. Prior to the shutdown, scientists were changing the data format of the Global Forecast System (GFS)—the most advanced mathematical, computer-based weather modeling prediction system in the USA. Unfortunately, the GFS currently does not recognize much of the input data it is receiving. A model is only as good as its input data (as I am sure Dawn can tell you), and currently that means the GFS is very limited. Many NWS models are upgraded January-June to prepare for storm season later in the year. Therefore, there are long-term ramifications for the lack of weather research advancement in terms of global health and safety. (https://www.washingtonpost.com/weather/2019/01/07/national-weather-service-is-open-your-forecast-is-worse-because-shutdown/?noredirect=on&utm_term=.5d4c4c3c1f59)

An example of one output from the GFS model. (Source: weather.gov)

The Food and Drug Administration (FDA)—a federal agency of the Department of Health and Human Services—that is responsible for food safety, has reduced inspections. Because domestic meat and poultry are at the highest risk of contamination, their inspections continue, but by staff who are going without pay, according to the agency’s commissioner, Dr. Scott Gottlieb. Produce, dry foods, and other lower-risk consumables are being minimally-inspected, if at all.  Active research projects investigating food-borne illness that receive federal funding are at a standstill.  Is your stomach doing flips yet? (https://www.nytimes.com/2019/01/09/health/shutdown-fda-food-inspections.html?rref=collection%2Ftimestopic%2FFood%20and%20Drug%20Administration&action=click&contentCollection=timestopics&region=stream&module=stream_unit&version=latest&contentPlacement=2&pgtype=collection)

An FDA field inspector examines imported gingko nuts–a process that is likely not happening during the shutdown. (Source: FDA.gov)

The National Parks Service (NPS) recently made headlines with the post-shutdown acts of vandalism in the iconic Joshua Tree National Park. What you might not know is that the shutdown has also stopped a 40-year study that monitors how streams are recovering from acid rain. Scientists are barred from entering the park and conducting sampling efforts in remote streams of Shenandoah National Park, Virginia. (http://www.sciencemag.org/news/2019/01/us-government-shutdown-starts-take-bite-out-science)

A map of the sampling sites that have been monitored since the 1980s for the Shenandoah Watershed Study and Virginia Trout Stream Sensitivity Study that cannot be accessed because of the shutdown. (Source: swas.evsc.virginia.edu)

NASA’s Stratospheric Observatory for Infrared Astronomy (SOFIA), better known as the “flying telescope” has halted operations, which will require over a week to bring back online upon funding restoration. SOFIA usually soars into the stratosphere as a tool to study the solar system and collect data that ground-based telescopes cannot. (http://theconversation.com/science-gets-shut-down-right-along-with-the-federal-government-109690)

NASA’s Stratospheric Observatory for Infrared Astronomy (SOFIA) flies over the snowy Sierra Nevada mountains while the telescope gathers information. (Source: NASA/ Jim Ross).

It is important to remember that science happens outside of laboratories and field sites; it happens at meetings and conferences where collaborations with other great minds brainstorm and discover the best solutions to challenging questions. The shutdown has stopped most federal travel. The annual American Meteorological Society Meeting and American Astronomical Society meeting were two of the scientific conferences in the USA that attract federal employees and took place during the shutdown. Conferences like these are crucial opportunities with lasting impacts on science. Think of all the impressive science that could have sparked at those meetings. Instead, many sessions were cancelled, and most major agencies had zero representation (https://spacenews.com/ams-2019-overview/). Topics like lidar data applications—which are used in geospatial research, such as what the GEMM Laboratory uses in some its projects, could not be discussed. The cascade effects of the shutdown prove that science is interconnected and without advancement, everyone’s research suffers.

It should be noted, that early-career scientists are thought to be the most negatively impacted by this shutdown because of financial instability and job security—as well as casting a dark cloud on their futures in science: largely unknown if they can support themselves, their families, and their research. (https://eos.org/articles/federal-government-shutdown-stings-scientists-and-science). Graduate students, young professors, and new professionals are all in feeling the pressure. Our lives are based on our research. When the funds that cover our basic research requirements and human needs do not come through as promised, we naturally become stressed.

An adult and a juvenile common bottlenose dolphin, forage along the San Diego coastline in November 2018. (Source: Alexa Kownacki)

So, yes, funding—or the lack thereof—is hurting many of us. Federally-funded individuals are selling possessions to pay for rent, research projects are at a standstill, and people are at greater health and safety risks. But, also, science, with the hope for bettering the world and answering questions and using higher thinking, is going backwards. Every day without progress puts us two days behind. At first glance, you may not think that my research on bottlenose dolphins is imperative to you or that the implications of the shutdown on this project are important. But, consider this: my study aims to quantify contaminants in common bottlenose dolphins that either live in nearshore or offshore waters. Furthermore, I study the short-term and long-term impacts of contaminants and other health markers on dolphin hormone levels. The nearshore common bottlenose dolphin stocks inhabit the highly-populated coastlines that many of us utilize for fishing and recreation. Dolphins are mammals, that respond to stress and environmental hazards, in similar ways to humans. So, those blubber hormone levels and contamination results, might be more connected to your health and livelihood than at first glance. The fact that I cannot download data from ERDDAP, reach my collaborators, or even access my data (that starts in the early 1980s), does impact you. Nearly everyone’s research is connected to each other’s at some level, and that, in turn has lasting impacts on all people—scientists or not. As the shutdown persists, I continue to question how to work through these research hurdles. If anything, it has been a learning experience that I hope will end soon for many reasons—one being: for science.