Skip to main content
Studies · Case Study

How Far Does the Sound Travel?

The US East Coast has planned up to 30 GW of offshore wind. Each foundation requires pile driving — the loudest routine construction activity in the ocean. The North Atlantic Right Whale (~350 remaining) migrates through the construction zones. The regulatory framework sets a shutdown zone at 5 km. But pile driving prevents whales from hearing each other at 80 km. The regulations are looking at the wrong scale.

The regulatory question isn't whether pile driving hurts whales. It's whether it silences them.

3 Sites
Vineyard Wind · CVOW · Revolution
200
Monte Carlo Draws
~350
Right Whales Remaining
14×
Masking vs Shutdown Zone
Context

11 Wind Farms, One Noise Field

Before the investigations: the scale of what’s planned, and why the regulatory framework is looking at the wrong scale.

Each dot is a wind farm. Each circle is the 80 km masking zone — the area where pile driving noise prevents whale communication. The NARW migration corridor runs through all of them.

Kitty Hawk CVOW Skipjack Ocean Wind Atlantic Shores Empire Wind Sunrise Revolution South Fork Vineyard Wind Mayflower NC VA MD NJ NY CT/RI MA Wind farm location 80 km masking zone NARW corridor

Schematic (approximate positions, not to geographic scale). Masking zone = 80 km radius (RAM PE, 10 dB bubble curtain, 50–500 Hz). Migration corridor from Roberts et al. (2024) NARW density models. Farm locations approximate, based on BOEM lease area centroids.

The Model

One Model, Eight Questions

One parabolic equation propagation model (RAM PE, Collins 1993), validated against field measurements at Vineyard Wind before being applied to anything else. Three sites on the US East Coast. All inputs from public sources: CUDEM bathymetry (31 m), WOA2023 sound speed profiles, Roberts et al. (2024) whale density data. 200 Monte Carlo draws for uncertainty quantification. Every chart on this page uses real computed data — RAM PE output, MCMC posterior samples, or brute-force optimization results.

RAM PE Propagation Model

Model Type
Split-step Padé PE

Collins (1993) parabolic equation; handles range-dependent environments without the wide-angle restrictions of simpler PE formulations.

Frequency Range
50–500 Hz

Baleen whale communication band. Pile driving energy peaks at 100–300 Hz — directly overlapping NARW upcall frequencies.

Geometry
2D Range-Dependent

Single radial transect per bearing. Captures depth-varying bathymetry and sound speed along each propagation path.

Computation
~0.3 s / run

Fast enough for 200-draw Monte Carlo in minutes and a 408-run Bayesian inversion lookup table in under three hours.

Data Inputs

All inputs are public and reproducible. Bathymetry: CUDEM (NOAA NCEI, 31 m resolution). Sound speed profiles: WOA2023 monthly temperature and salinity at 0.25° resolution, converted to sound speed via the Mackenzie (1981) equation. Sediment parameters: Hamilton (1980) geoacoustic regression from sediment grain size. Whale density: Roberts et al. (2024) NARW habitat-based density models for the US Atlantic.

Monte Carlo Design

200 draws, perturbing bottom properties across their plausible ranges: sediment grain size φ ± 1.5 (phi units), bulk density ± 5%, and bottom sound speed ± 30 m/s. This design was chosen because Investigation 02 showed that bottom properties dominate shutdown zone uncertainty by more than 20× compared to any other input.

Bayesian Inversion

Investigation 08 ran 408 RAM PE simulations as a lookup table for MCMC posterior estimation. The inversion constrains the effective bottom at Vineyard Wind to φ = 6.5 ± 0.8 — softer and more absorptive than Hamilton’s published regression predicts for the site geology. The data-constrained RMSE is 1.7 dB, compared to 6.6 dB for the forward model using Hamilton priors alone.

Validation
Validated at Vineyard Wind (RMSE 6.6 dB, conservative bias) and cross-validated at South Fork Wind (7.1 dB RMSE) before being applied to any other site.
Investigations

Eight Questions, One Chain

Each answer enabled the next question. One propagation model (RAM PE), validated against real pile driving measurements, applied across three sites and extended to questions the regulatory framework doesn’t ask.

Investigation 01

Can We Reproduce Measured Sound Propagation?

RMSE = 6.6 dB against Rand Acoustics measurements at Vineyard Wind. Cross-validated against South Fork Wind. The model is conservative — it over-predicts sound levels, producing larger safety zones than measured.

Validation · RAM PE
Investigation 02

Which Input Drives the Regulatory Decision?

OAT screening says bottom properties dominate (11 km zone spread). Sobol decomposition (28,672 evaluations) reveals: source level and mitigation explain 74% and 61% of take variance. 56% comes from interactions OAT cannot see.

OAT + Sobol · 6 inputs · 28K evals
Investigation 03

Does Season Affect the Risk?

The shutdown zone barely changes with season (0.5 km). But whale density varies 21× between August and February. The IHA’s January–May pile driving restriction avoids 77% of annual risk. Season matters because of biology, not acoustics.

Seasonal Analysis
Investigation 04

What Is the Defensible Take Estimate?

Monte Carlo (200 draws): median 4.1 takes, P(exceed IHA’s 7) = 26%. Required bubble curtain performance for 10% exceedance risk: 10.5 dB. Measured performance at CVOW (14–17 dB) is sufficient.

Monte Carlo · 200 Draws
Investigation 05

Can Whales Still Hear Each Other?

The communication masking zone is 14× larger than the shutdown zone. A whale 50 km from the pile — safe by every regulatory standard — cannot hear another whale. With 5+ projects piling simultaneously, 100% of the 660 km migration corridor is masked.

Communication Masking · RAM PE
Investigation 06

Can Scheduling Create Quiet Windows?

Coordinated piling schedules create 9 guaranteed quiet hours per day for whale communication. Start date optimization reduces total takes by 3.3×. Both are scheduling decisions — zero additional cost.

Schedule Optimization
Investigation 07

Which Technology Solves the Problem?

Bubble curtains cut the shutdown zone by 12× but masking by only 2×. The masking threshold is 45 dB below the Level B threshold — a gap no attenuation system can bridge. Best available delivers ~20 dB, which solves harassment but barely dents masking. Only pile-free foundations (suction buckets, gravity bases) bring masking below 2 km.

Technology Comparison
Investigation 08

What Does the Data Say About the Bottom?

Bayesian inversion with 408 RAM PE runs constrains the effective bottom to phi = 6.5 ± 0.8 — softer than the published geology. The data-constrained RMSE is 1.7 dB. The real bottom is rougher and more absorptive than Hamilton’s regression predicts.

Bayesian Inference · MCMC
Fidelity Lesson

What Changes When You Zoom Out

The analysis started with a standard acoustic propagation model. The surprising findings came not from making the model more complex, but from asking what it means at larger scales. Each row in the table below uses the same RAM PE model — the only thing that changes is the question.

Level What’s Included Scale What Changes
Regulatory threshold Single-frequency TL, fixed bottom, Level B (160 dB) 5 km Standard shutdown zone — the number in every IHA
With uncertainty Monte Carlo bottom properties, 200 draws 3–8 km The zone is a range, not a number. Bottom properties dominate.
With communication masking Same propagation + whale call SL vs. ambient noise floor 80 km Impact extends 14× beyond regulatory scope. Whales can’t hear each other.
With corridor Multi-project cumulative masking, 5+ simultaneous piling operations 660 km 100% of the NARW migration corridor masked. 132× the regulatory zone.

Same model, same physics, same code. The answer changes by 132× because the question changes.

The right scale changes the answer

The shutdown zone (5 km) and masking zone (80 km) come from the same propagation model. The physics didn’t change — the question did. When you ask “where is it loud?” you get one answer. When you ask “where can whales still communicate?” you get a very different one.

Investigations 01 & 05

The bottom matters more than the ocean

Across three sites spanning Virginia to Massachusetts, bottom properties drove 11+ km of shutdown zone uncertainty. Seasonal sound speed profiles contributed 0.5 km. The decision about what surveys to fund follows directly from this ranking.

Investigation 02

Attenuation solves harassment but not masking

Bubble curtains reduce the shutdown zone by 12× but the masking zone by only 2×. The masking threshold sits 45 dB below the Level B threshold — no attenuation system can bridge that gap. The only solutions are pile-free foundations or coordinated scheduling.

Investigations 06 & 07

Per-project review misses the corridor

Every IHA evaluates one project in isolation. At the 5 km shutdown scale, adjacent farms don’t interact. At the 80 km masking scale, five farms create a continuous noise field across the entire southern New England shelf. The regulatory unit of analysis is too small.

Investigation 05
Sensitivity

Which Inputs Move the Answer?

OAT screening identified bottom properties as the dominant environmental driver. A full Sobol decomposition (28,672 evaluations, 6 inputs) reveals what OAT misses: source level and mitigation interactions account for 56% of take variance.

Bottom Properties Critical

Sediment grain size, bulk density, and bottom sound speed. Perturbing these across their plausible ranges produces ±11 km of shutdown zone spread at every site. A $200K geotechnical survey resolves more uncertainty than any other measurement campaign.

±11 km zone spread
Ambient Noise Floor High

The masking threshold depends on ambient noise (~105 dB, Wenz 1962). Actual ambient varies 10–15 dB with shipping, wind state, and biological noise. A 10 dB swing reshapes the masking zone substantially — but has no effect on the shutdown zone.

±10–15 dB on masking range
Seasonal Sound Speed Minor

Summer and winter sound speed profiles produce different propagation paths. The shutdown zone changes by only 0.5 km. Season matters because of whale density (21× variation), not acoustics.

±0.5 km zone spread
Bathymetry Resolution Negligible

Varying CUDEM resolution from 31 m to 90 m changes the shutdown zone by 0.2 km. The propagation model is insensitive to fine-scale bathymetric features at these frequencies.

±0.2 km zone spread

The answer depends on the question. For environmental inputs alone (OAT), bottom properties dominate. But for the regulatory decision (Sobol on take), source level (ST=74%) and mitigation (ST=61%) dwarf everything else — and interact strongly. The optimal strategy: monitor actual source levels during initial piles, then invest in geotechnical surveys.

Validation

Can You Trust These Numbers?

The model was validated against real pile driving measurements before being applied to any other question. Conservative bias means safety zones are over-estimated, not under-estimated.

Model Prediction Transmission loss at Vineyard Wind using Hamilton (1980) bottom priors
Measured Data Rand Acoustics (2023) field measurements during pile driving: RMSE 6.6 dB
Conservative
Model over-predicts sound levels — producing larger safety zones than measured. Bias direction is safe.
Model Prediction Same model applied to South Fork Wind (different site, same methodology)
Measured Data NOAA SFV report: RMSE 7.1 dB. Cross-validation confirms transferability.
Validated
Model trained at one site generalizes to a second site 80 km away with comparable accuracy
Model Prediction Bayesian inversion constrains effective bottom to φ = 6.5 ± 0.8
Measured Data Data-constrained RMSE drops to 1.7 dB (from 6.6 dB with Hamilton priors)
Validated
408 RAM PE runs + MCMC. The real bottom is softer than published geology predicts.
Validation Summary
Validated at one site, cross-validated at a second, then applied to a third. Conservative bias throughout — every safety zone in this study is an over-estimate, not an under-estimate.
Limitations

What This Cannot Tell You

Every model has boundaries. These are ours.

2D Propagation Only

RAM PE computes transmission loss along a single radial transect. Azimuthal variability (3D bathymetric focusing, lateral refraction) is not captured. Actual zone shapes are not circles.

Single-Transect Validation

The model was validated at Vineyard Wind and cross-validated at South Fork Wind. Performance at CVOW and Revolution Wind is inferred from the same methodology, not measured directly.

Ambient Noise Uncertainty

The masking threshold depends on an ambient noise floor of ~105 dB (Wenz 1962). Actual ambient on the US East Coast varies 10–15 dB with shipping density, wind state, and biological noise. Masking zones could be substantially larger or smaller.

Whale Call Variability

The ~160 dB upcall source level (Parks et al. 2011) is a population median. Individual calls range from ~145 to ~175 dB. Quieter whales have larger effective masking zones.

200 Monte Carlo Draws

Adequate for central estimates (median, P75) but marginal for tail statistics. Exceedance probabilities carry ±3% sampling uncertainty.

Global Context

Where Else Does This Apply?

The masking framework is specific to baleen whales — species that communicate at the same low frequencies (50–500 Hz) as pile driving noise. For dolphins and porpoises, the threat is habitat displacement, not communication masking. The mechanism differs, but the recommendation is the same: manage construction acoustics at the corridor scale, not the project scale.

This is not an argument against offshore wind. It is an argument for matching the scale of environmental management to the scale of acoustic impact. The transition to clean energy and marine mammal conservation are both essential. They can coexist — but only if the regulatory process looks beyond individual project boundaries.

Recommendations

What the Analysis Supports

Five recommendations — each traceable to a specific investigation and grounded in a model validated against field measurements at two sites.

  • 01

    Regulate at the masking scale, not the harassment scale

    The shutdown zone (5 km) and the communication masking zone (80 km) come from the same model. Current IHAs are structured around Level B harassment thresholds. Communication masking at 80 km affects ~350 remaining right whales even when no harassment occurs. Regulations that use only the 5 km threshold are missing the impact that matters most for population viability.

  • 02

    Fund bottom surveys, not oceanographic campaigns

    Bottom properties drive 11 km of shutdown zone uncertainty across three sites. Seasonal sound speed contributes 0.5 km. If a project has a limited environmental survey budget, the sensitivity ranking says spend it on geotechnical characterization, not oceanographic campaigns. A $200K bottom survey resolves more uncertainty than a $100K oceanographic deployment.

  • 03

    Apply the IHA seasonal restriction universally

    The January–May pile driving restriction avoids 77% of annual acoustic risk. The shutdown zone itself barely changes with season (0.5 km variation). The restriction works because of whale biology (21× density variation), not acoustics. Every project in the corridor should face this restriction, not just projects with NARW habitat overlap in their immediate vicinity.

  • 04

    Require corridor-scale cumulative analysis in each IHA

    At the 5 km shutdown scale, adjacent farms don’t interact. At the 80 km masking scale, five simultaneous projects create a continuous noise field across the entire southern New England shelf. The regulatory unit of analysis is the individual project; the biological unit of concern is the 660 km migration corridor. NMFS should require cumulative masking analysis as part of each individual IHA application.

  • 05

    Prioritize pile-free foundation technology for shallow sandy sites

    Bubble curtains reduce the shutdown zone by 12× but the masking zone by only 2×. No attenuation system can bridge the 45 dB gap between the Level B threshold and the masking threshold. Suction bucket and gravity-base foundations eliminate pile driving entirely. For sites where these technologies are technically feasible (shallow water, sandy substrate), they should be the preferred option — not a mitigation measure of last resort.

References & Data Sources

Reproducible with Public Data

Every input comes from publicly available, peer-reviewed sources. Every analysis is reproducible with open-source tools.

Model

  • Collins, M.D. (1993). “A split-step Padé solution for the parabolic equation method.” J. Acoust. Soc. Am., 93(4), 1736–1742. doi:10.1121/1.405831 [RAM PE]

Environmental Data

  • NOAA NCEI. Continuously Updated Digital Elevation Model (CUDEM), 1/9 arc-second (~31 m). [Bathymetry]
  • NOAA NCEI. World Ocean Atlas 2023, monthly T/S, 0.25° resolution. Mackenzie (1981) sound speed equation. [Sound Speed]
  • Hamilton, E.L. (1980). “Geoacoustic modeling of the sea floor.” J. Acoust. Soc. Am., 68(5), 1313–1340. doi:10.1121/1.384967 [Sediment]
  • Roberts, J.J. et al. (2024). Habitat-based cetacean density models, US Atlantic. Duke University MGEL. See also Roberts et al. (2016), Scientific Reports, 6, 22615. doi:10.1038/srep22615 [NARW Density]

Validation

  • Rand Acoustics (2023). Underwater sound field verification during pile driving: Vineyard Wind 1. Prepared for Vineyard Wind LLC.
  • NOAA Fisheries (2024). Sound Field Verification Report: South Fork Wind. [Cross-validation]

Regulatory

  • NOAA Fisheries (2024). Technical Guidance for Assessing the Effects of Anthropogenic Sound on Marine Mammal Hearing, v3.0. [Thresholds]
  • Southall, B.L. et al. (2019). “Marine mammal noise exposure criteria: Updated scientific recommendations.” Aquat. Mamm., 45(2), 125–232. doi:10.1578/AM.45.2.2019.125
  • Vineyard Wind 1 (2024). Incidental Harassment Authorization, NMFS Permit No. 27224. [Take limits]

Communication Masking

  • Parks, S.E. et al. (2011). “Sound production behavior of individual North Atlantic right whales.” Endangered Species Research, 15, 63–76. doi:10.3354/esr00411 [Upcall SL]
  • Clark, C.W. et al. (2009). “Acoustic masking in marine ecosystems.” Mar. Ecol. Prog. Ser., 395, 201–222. doi:10.3354/meps08402
  • Amaral, J.L. et al. (2020). “Underwater sound from impact pile driving at the Block Island Wind Farm.” J. Acoust. Soc. Am., 147(4), 2323–2333. doi:10.1121/10.0001035

8 investigations · 3 sites (Vineyard Wind, CVOW, Revolution Wind) · 200 Monte Carlo draws · 408 RAM PE runs (Bayesian lookup) · Validated against Rand Acoustics VW1 (RMSE 6.6 dB) and South Fork Wind SFV · Real CUDEM bathymetry (31m res), WOA2023 sound speed, Roberts et al. 2024 NARW density · All code and data reproducible with public sources and open-source tools.