Decision-Making Under Pressure during My PhD: Lessons from whale songs and ocean noise

Decision-Making Under Pressure during My PhD: Lessons from whale songs and ocean noise

Throughout most of my PhD, progress has been a slow, steady, and iterative process. But when fieldwork is involved, it doesn’t always follow that script. Experimental conditions change and plans fall apart, and you might need to make quick decisions to try to salvage the situation. Looking back, a few of my research breakthroughs came not from the plan, but from trying to adapt when things didn’t go as expected.

That’s what happened on the last day of a field expedition in Monterey Bay. My research in Distributed Acoustic Sensing (DAS) led to this field deployment, as part of an effort to understand the full potential of this relatively new technology. We’re still exploring its capabilities, limitations, and the kinds of real-world problems it’s well positioned to solve. On this trip, my colleagues and I were out on a research boat, hoping to record whale vocalizations using our sensing system. But the ocean environment proved much harder to work in than we predicted. The noise levels were overwhelming, and none of our test signals were coming through clearly. I was anxious that we might return with no “good” data. But on the very last day, a small, spur-of-the-moment adjustment turned things around.

Monitoring wind energy infrastructure and whales

Our project was motivated by the growing interest in floating offshore wind turbines (FOWTs) – turbines mounted on floating platforms and anchored to the seabed by mooring lines. As these technologies scale up, so do concerns about their environmental impact, especially on marine mammals like whales. Monitoring both the structural behavior of FOWTs (Maples et al., 2013; Kim et al., 2019; Hubbard et al., 2021; Xu et al., 2024) and their acoustic footprint in the ocean (Madsen et al, 2006; Bailey et al., 2010; Maxwell et al., 2022) is going to be critical for sustainable development.

We explored whether DAS could help. DAS turns fiber optic cables into dense arrays of vibration sensors, capable of detecting pressure waves and acoustic energy along the length of the fiber. In other words, it’s a way to turn a cable into thousands of virtual microphones. Research has shown that DAS can be used to detect whale vocalizations using seafloor telecommunication cables, but those systems operate in relatively quiet environments. Our question was: what happens when DAS is deployed in the much noisier, more dynamic upper ocean – the same environment where FOWTs will operate? In particular, we wanted to understand how DAS performs when deployed on floating platforms or vessels, where wave action and motion introduce complex, time-varying loads on the cable – similar to what FOWTs might experience.

A vertical cable in the ocean

To mimic a possible future DAS deployment on a FOWT mooring line, we lowered a fiber optic cable vertically into the water using an A-frame mounted on a boat. The DAS interrogator – the instrument that sends laser pulses down the fiber and listens for backscattered signals – was placed inside the boat cabin and connected to the cable reel on deck. We attached a weight to the end of the cable, aiming to keep it vertical and taut.

Our research boat with A-frame on the boat deck to lower the fiber optic cable into the water. (Source: Jeremy Snyder)

In practice, maintaining that vertical position proved difficult. Wave currents rocked the vessel, dragged and jolted the cable. At times, the cable wrapped around the mooring line, forming coils. Because the currents were so strong, the cable could be dragged to a nearly horizontal position. We later increased the weight to counteract this to keep it vertical, but it remained seemingly impossible to keep the cable from vibrating significantly and getting jolted around.

These physical disturbances had a clear effect on data quality. Most of the DAS data was dominated by high-amplitude, broadband noise. To make matters more difficult, the boat had to keep its engine running for a lot of the time due to weather protocols, introducing constant engine noise that seemingly overwhelmed the DAS system. Even during quiet periods, measurements were full of rapid fluctuations caused by wave-induced motion, turbulence, and boat activity. We also deployed an underwater speaker to play test signals, including tones, chirps, and a humpback whale song, and a co-located hydrophone to serve as a ground-truth reference. But even with known input signals, the DAS data was messy and unpredictable.

Raw DAS data plotted after frequency band extraction (FBE) with fiber optic cable in a linear configuration. The cable began at its connection to the DAS interrogator inside the boat. High noise levels were observed in the submerged section of the cable, starting at ~130m along the cable. (Source: Saw et al., 2025)

From a data science standpoint, the challenge was clear: how to detect weak biological signals in the presence of strong, nonstationary noise. We experimented with filtering, spectrogram analysis, and comparisons across depth channels, but the signal-to-noise ratio wasn’t high enough to support confident detection. Still, this data was giving us insight into how the cable responded to wave motion, what strain patterns were most common, and which frequency bands were especially noisy. By the final day, we had a clearer picture of what wasn’t working — and that pointed us toward what might.

A small adjustment that helped more than we expected

One idea came up late in the expedition: what if we introduced loops in the cable? In typical DAS applications on land, like roadside traffic monitoring, loops are usually something to avoid. They make it harder to determine the exact spatial location of signals along the fiber. My default approach in DAS research had been trying to keep the fiber as linear as possible.

But on the last day with only a few hours left to record data, we figured we could try something different. We looped three sections of fiber, using zip ties and tape. The idea was that loosely coiled fiber might be more decoupled from water movement and less sensitive to high-frequency vibrations and strain that had been overwhelming our recordings.

Loops were introduced into the cable. Three sections of the cable were looped to minimize excessive cable vibrations and strain effects that could distort the recorded signals. Each looped section was ~51-55cm in diameter. (Source: Saw et al., 2025)

It worked better than we expected. During acquisition, we noticed that the looped sections had a visibly lower noise floor, and in the post-expedition analysis, we finally saw signals with structure and frequency content matching humpback whale vocalizations – something that hadn’t been clearly detectable in the earlier linear deployments. The loops didn’t eliminate all the noise, but they created “quieter” sections of the cable where weak biological signals could be detected. After converting the measurements to audio and checking with marine biologists, we confirmed that our DAS system had indeed captured (humpback) whale songs.

Raw DAS data plotted after frequency band extraction (FBE) with fiber optic cable in the revised, looped configuration. Lower recorded noise was observed in all three looped sections: top (136-151m), middle (313-333m), and bottom (482-498m). (Source: Saw et al., 2025)

Examples of humpback whale songs captured by DAS. This data was plotted after spectral baseline subtraction and averaging across channels in the bottom looped section. (Source: Saw et al., 2025)

What our study demonstrated

This project showed that DAS can work even in the harsh, noisy environments expected around floating offshore wind infrastructure. It demonstrated that a vertical DAS deployment – using a cable suspended from a mooring line – can detect marine mammal vocalizations even in the presence of engine noise, wave action, and cable movements.

The looped configuration was especially promising. It turned out that by intentionally coiling parts of the fiber, we could create localized sensing nodes with better signal clarity. These loops acted almost like acoustic isolators, reducing the influence of boat-induced motion and helping us “listen” more clearly at specific depths. And because fiber is relatively inexpensive, adding extra cable length to enable such designs is feasible and scalable.

This matters not just for whale monitoring, but for the broader vision of multi-functional sensing from FOWT infrastructure. A vertical DAS system co-deployed with a mooring line could potentially monitor both the structure’s health and the marine acoustic environment, offering a dual-purpose solution for renewable energy and environmental impact assessment.

Earlier this year, we published a paper describing this study (Saw et al., 2025), and we’re working on a follow-up that focuses on automating whale call detection in noisy DAS data using a CNN-based approach, integrating data from the co-deployed hydrophone to improve performance.

A final reflection

When we first saw the results from the looped sections, I remember thinking: “Ugh, why didn’t we try that earlier?” But that frustration gave way to a more generous realization. That idea only became obvious because of everything we had tried beforehand – because we had spent so much time struggling with the noise, analyzing it, and learning how different factors affected the signals. In hindsight, that decision we made under pressure wasn’t just a lucky development, but rather a result of what we had observed and struggled through up to that point.

Over the course of my PhD, I’ve come to see that progress can come from slow, methodological improvements as well as unexpected breakthroughs that arise under pressure. My PhD has had its unpredictable moments, and I’ve often felt unsure whether certain efforts would lead to progress or fall short of its original goals. But over time, I’ve gotten a little better at navigating those uncertainties – not just in research, but in life as well (I hope).

References

1. Bailey, Helen, Bridget Senior, Dave Simmons, Jan Rusin, Gordon Picken, and Paul M. Thompson. "Assessing underwater noise levels during pile-driving at an offshore windfarm and its potential effects on marine mammals." Marine Pollution Bulletin 60, no. 6 (2010): 888-897.

2. Hubbard, Peter G., James Xu, Shenghan Zhang, Matthew Dejong, Linqing Luo, Kenichi Soga, Carlo Papa et al. "Dynamic structural health monitoring of a model wind turbine tower using distributed acoustic sensing (DAS)." Journal of Civil Structural Health Monitoring 11, no. 3 (2021): 833-849.

3. Kim, Hyoung-Chul, Moo-Hyun Kim, and Do-Eun Choe. "Structural health monitoring of towers and blades for floating offshore wind turbines using operational modal analysis and modal properties with numerical-sensor signals." Ocean Engineering 188 (2019): 106226.

4. Madsen, Peter T., Magnus Wahlberg, Jakob Tougaard, Klaus Lucke, and Peter Tyack. "Wind turbine underwater noise and marine mammals: implications of current knowledge and data needs." Marine Ecology Progress Series 309 (2006): 279-295.

5. Maples, Ben, Genevieve Saur, Maureen Hand, R. Van De Pietermen, and T. Obdam. Installation, operation, and maintenance strategies to reduce the cost of offshore wind energy. No. NREL/TP-5000-57403. National Renewable Energy Laboratory (NREL), Golden, CO (United States), 2013.

6. Maxwell, Sara M., Francine Kershaw, Cameron C. Locke, Melinda G. Conners, Cyndi Dawson, Sandy Aylesworth, Rebecca Loomis, and Andrew F. Johnson. "Potential impacts of floating wind turbine technology for marine species and habitats." Journal of Environmental Management 307 (2022): 114577.

7. Saw, Jaewon, Linqing Luo, Kristy Chu, John Ryan, Kenichi Soga, and Yuxin Wu. "Distributed acoustic sensing for whale vocalization monitoring: A vertical deployment field test." Seismological Research Letters 96, no. 2A (2025): 801-815.

8. Xu, James T., Linqing Luo, Jaewon Saw, Chien-Chih Wang, Sumeet K. Sinha, Ryan Wolfe, Kenichi Soga, Yuxin Wu, and Matthew DeJong. "Structural health monitoring of offshore wind turbines using distributed acoustic sensing (DAS)." Journal of Civil Structural Health Monitoring 15, no. 2 (2025): 445-463.