Understanding the Challenge of Underwater Acoustics

Capturing clean underwater audio requires a fundamentally different approach than recording in air. Water is roughly 800 times denser than air, and sound travels nearly four times faster underwater. This dense medium transmits vibrations with far greater efficiency, but it also amplifies and disperses noise in ways that can overwhelm the target signal. For marine biologists, oceanographers, documentary filmmakers, and environmental monitoring teams, the ability to minimize noise in underwater audio recordings directly determines whether the data reveals meaningful information or remains unusable.

Noise in underwater recordings is never the result of a single cause. It accumulates from physical water movement, biological activity, mechanical artifacts in the recording chain, and distant human infrastructure. Without a systematic strategy to identify and mitigate each layer of interference, even the most sensitive hydrophone will produce muddy, artifact-laden audio. The objective is not to eliminate all sound—ambient noise is a natural component of any marine environment—but to suppress unwanted artifacts so that the target signal emerges with clarity and precision.

Sources of Underwater Noise: A Detailed Breakdown

Effective noise reduction begins with understanding what you are filtering. Underwater noise typically falls into four categories, each demanding a distinct mitigation approach.

Physical and Ambient Noise

Water movement generates continuous low-frequency rumble. Waves breaking at the surface, tidal currents rushing past submerged structures, and turbulence around the hydrophone itself all contribute to a baseline hum. Rain, wind, and surface chop add broadband noise that can saturate the recording channel if gain is set too high. In shallow or coastal environments, sediment transport and entrained bubbles introduce short-duration pops and crackles that mimic biological sounds. Even the motion of the hydrophone cable through water produces frictional noise, known as cable strum, that contaminates the low-frequency range.

Biological Noise

Marine life is both the target of many recordings and a primary source of interference. Snapping shrimp produce high-frequency clicks exceeding 200 decibels, completely drowning out quieter species within meters. Fish choruses, whale vocalizations, and invertebrate stridulations all occupy overlapping frequency bands. When multiple species call simultaneously, the resulting cacophony can obscure the specific sound you are trying to isolate. Understanding the acoustic ecology of your recording site is essential for scheduling and processing decisions. For example, recording during the quiet hours between dawn and dusk chorus can reduce biological noise contamination.

Mechanical and Equipment Noise

Every component in the recording chain introduces some level of self-noise. Hydrophone preamplifiers generate electronic hiss, especially in low-cost models where the noise figure exceeds 30 dB re 1 µPa. Cables pick up electromagnetic interference from nearby motors, inverters, or power lines. Buoy-mounted recorders sway in currents, producing cable strum and frictional noise against the mounting line. Even the hydrophone housing can resonate if not properly damped with elastomeric isolators. Connectors, especially those subject to saltwater corrosion, introduce intermittent crackling artifacts.

Anthropogenic Noise

Shipping traffic, dredging operations, seismic surveys, and recreational boating produce intense low-frequency noise that travels hundreds of kilometers underwater. In busy waterways, this continuous drone can raise the ambient noise floor by 20 dB or more. Pile driving and sonar pulses create short-duration, high-amplitude events that clip the recording if input levels are not managed. As coastal development increases, anthropogenic noise has become the dominant challenge for long-term underwater monitoring programs. Real-time vessel tracking tools such as those from Marine Vessel Traffic allow researchers to schedule recordings around heavy traffic periods.

Selecting and Deploying Hydrophones for Minimum Self-Noise

The most effective noise reduction begins before the first recording. Choosing the right hydrophone and deploying it correctly eliminates many problems that cannot be fixed in post-production. Consider the specific application: a omni-directional hydrophone is suitable for ambient sound monitoring, while a directional model can reject off-axis noise sources.

Hydrophone Sensitivity and Frequency Response

Not all hydrophones are suited for all applications. A research-grade hydrophone with a flat frequency response from 10 Hz to 100 kHz is ideal for capturing the full bandwidth of marine sounds, but it may also expose excessive low-frequency flow noise. For targeted recordings, consider a hydrophone with a built-in high-pass filter that attenuates rumble below 100 Hz. Check the manufacturer’s specifications for self-noise, expressed as equivalent noise level relative to 1 µPa. Lower values indicate a quieter preamplifier. For example, the Brüel & Kjær 8106 hydrophone has a self-noise of around 40 dB re 1 µPa at 1 kHz, sufficient for many offshore applications, while the more sensitive HTI-96-MIN from High Tech Inc. offers a self-noise of 30 dB re 1 µPa, better for quiet freshwater environments.

Mounting and Damping

How you mount the hydrophone directly affects the signal-to-noise ratio. A rigid attachment to a boat hull or dock transmits vibrations directly into the transducer. Use a suspension system—a rubber bungee cord, a foam float, or a weighted line—to decouple the hydrophone from its support structure. For stationary bottom-mount deployments, bury the cable in sediment for the first few meters to prevent strumming. If recording from a drifting platform, use a slack tether and attach a drogue to reduce vertical motion. Additionally, use a mechanical isolator such as a neoprene sleeve between the hydrophone and any metal frame to dampen structural vibrations.

Placement Away from Turbulence

Turbulent flow across the hydrophone element produces low-frequency pressure fluctuations that masquerade as sound. Place the hydrophone in a region of laminar flow, typically at least two meters from any solid surface that could generate eddies. Orient the sensitive axis parallel to the expected current direction to minimize flow noise. In tidal zones, deploy the hydrophone below the depth of maximum current shear, where water movement is more uniform. Using a foam flow shield around the hydrophone element can further reduce flow-induced noise, but be aware that such shields also attenuate high-frequency signals above 10 kHz.

Optimizing the Recording Chain

The hydrophone is only the first link. Every component between the transducer and the storage medium must be chosen and configured to preserve signal integrity and minimize added noise.

Preamplifier Gain and Dynamic Range

Set the gain so that the loudest expected sound uses approximately 80% of the available dynamic range. Too low, and the signal is buried in the noise floor after digitization. Too high, and peaks clip, creating harmonic distortion that cannot be removed. Use a preamplifier with a low noise figure and a wide gain range. If your recorder has adjustable input impedance, match it to the hydrophone output impedance for maximum power transfer and minimum noise pickup. For battery-powered setups, consider a preamplifier with a low-power mode to extend deployment duration without sacrificing noise performance.

Cable Selection and Shielding

Underwater cables are subject to corrosion, flexing, and electromagnetic interference. Use a balanced, shielded cable with a braided copper shield and a waterproof connector. Keep the cable as short as practical—long cables act as antennas for electrical noise. If the recorder is in a boat or onshore, route the cable away from engines, inverters, and radio transmitters. Ferrite beads clamped at both ends can suppress radio frequency interference. For long-term deployments, use cables with Kevlar strength members to prevent stretch and microphonic noise from cable movement.

Power Supply Quality

Noisy power is a common but overlooked source of hum. Battery-powered recorders should use linear regulators rather than switching converters if possible. If the recorder must run on mains power, use a medical-grade isolation transformer to break ground loops. For long-term autonomous deployments, choose a recorder with a low-ripple voltage regulator and sufficient battery capacity to avoid voltage droop near the end of the recording cycle. Avoid using inverters or generators within 50 meters of the hydrophone cable.

Environmental Scheduling and Site Selection

No amount of equipment upgrades can compensate for recording in the wrong place at the wrong time. Environmental conditions are the dominant factor in underwater noise levels.

Timing with Tidal Cycles

In coastal waters, tidal currents are the primary driver of flow noise. Slack tide—the 30- to 60-minute window between flood and ebb—offers the lowest current velocities and thus the lowest ambient noise. Slack times shift by approximately 50 minutes each day, so plan recordings around local tide tables. During spring tides, currents are stronger and turbulence is higher; neap tides produce quieter conditions overall. Use a tidal prediction app or NOAA tide tables to identify optimal windows.

Weather Windows

Wind speed correlates directly with underwater noise levels. At wind speeds above 10 knots, surface agitation generates spray and bubbles that produce audible hiss in the underwater sound field. Rain raises the noise floor across a broad frequency band. Check marine weather forecasts for wind speed, wave height, and precipitation probability. The quietest recordings occur during extended periods of high pressure with light winds and calm seas, often in the early morning hours before sea breezes develop.

Distance from Anthropogenic Sources

Ship noise propagates efficiently in shallow water, bouncing between the surface and the seabed. Even if you are kilometers from a shipping lane, the low-frequency drone of engines can contaminate your recording. Use real-time vessel tracking tools to identify nearby traffic and schedule breaks during heavy transit periods. In harbors or estuaries, consider recording at night when recreational and commercial traffic drops significantly. For stationary monitoring, place hydrophones at least 500 meters from known shipping lanes and pile-driving zones.

Post-Recording Noise Reduction Techniques

Even with perfect field technique, some noise will remain. Digital signal processing offers powerful tools to clean up the audio without distorting the signal of interest. Always keep an unprocessed backup of the original recording before applying any processing.

Spectral Subtraction and Adaptive Filtering

Spectral subtraction analyzes a segment of audio that contains only noise, creates a statistical model of that noise, and then subtracts it from the full recording. This technique works well for stationary noise sources like hums and hisses but can introduce artifacts called musical noise if the subtraction is too aggressive. To reduce musical noise, apply a small noise floor offset or use a minimum statistics approach. Adaptive filters continuously update the noise model as the recording progresses, making them effective against slowly varying noise such as changing current speed. Both methods require clean noise-only samples, so capture a few seconds of ambient sound before your target source begins.

Band-Pass and Notch Filters

If you know the frequency range of your target sound, apply a band-pass filter to remove everything outside that range. For example, many fish calls occupy frequencies between 200 Hz and 2 kHz, so filtering below 150 Hz and above 3 kHz eliminates flow noise and high-frequency snapping shrimp clicks without affecting the desired signal. Notch filters remove specific interfering tones, such as the 50 Hz or 60 Hz hum from mains power or the pulse of a sonar source. Use a spectrogram to visualize the noise spectrum and precisely set filter cutoff frequencies.

Wavelet Denoising

Wavelet denoising decomposes the audio into time-frequency components, thresholds the coefficients, and reconstructs the signal. This method excels at preserving transient sounds—clicks, pops, and short calls—that overlap spectrally with noise. Wavelet-based techniques require careful selection of the threshold level; too high, and you lose the transient; too low, and noise remains. MATLAB and Python libraries provide built-in wavelet denoising tools that are well-documented for bioacoustics applications. For real-time denoising, consider using a discrete wavelet transform with a symlet mother wavelet, which offers good time-frequency localization.

Adaptive Notch Filtering for Tonal Noise

For persistent tonal noise such as a boat engine or sonar ping, an adaptive notch filter can continuously track and cancel the offending frequency. This technique uses a reference input (e.g., from a second hydrophone in a quiet location) to generate an anti-phase signal that cancels the noise at the primary hydrophone. While complex to implement, it can remove tones without affecting broadband signals.

Practical Field Protocols for Clean Recordings

Beyond the technical measures, disciplined fieldwork habits prevent contamination that is impossible to remove later.

  • Calibrate before every deployment. Use a pistonphone or a calibrated reference source to record a known sound pressure level. This allows you to quantify the noise floor and verify system functionality. Store calibration tones in a separate file for documentation.
  • Log environmental metadata. Record wind speed, current velocity, water depth, temperature, and nearby vessel activity at the time of each recording. This metadata helps you interpret noise patterns and apply appropriate filters during post-processing.
  • Monitor the recording in real time if possible. Use headphones or a spectrogram display to catch problems as they occur. Cable chafe, loose connectors, and battery drain produce characteristic artifacts that are obvious during live monitoring but difficult to diagnose later.
  • Use multiple hydrophones in a synchronized array. With two or more spatially separated channels, you can apply beamforming to focus on sounds from a specific direction while rejecting noise from other angles. This technique is especially valuable in areas with strong directional noise sources like boat traffic.
  • Perform routine maintenance. Clean hydrophone elements with distilled water and a soft brush to remove biofilm and mineral deposits. Inspect connectors for corrosion and replace O-rings annually. A poorly maintained hydrophone introduces mechanical noise that degrades every recording.
  • Use a hydrophobic coating on the hydrophone element to reduce the adhesion of air bubbles, which can create pop artifacts when they release from the surface.
  • Deploy a secondary reference hydrophone in a location with minimal signal to capture the ambient noise field separately, enabling more accurate subtraction later.

Advanced Processing with Machine Learning

In recent years, machine learning models have become accessible tools for underwater audio denoising. Convolutional neural networks (CNNs) trained on large datasets of clean and noisy underwater audio can learn to separate signals from interference with remarkable accuracy. These models operate on spectrograms rather than raw waveforms, allowing them to exploit temporal and spectral patterns that traditional filters miss. Open-source frameworks such as TensorFlow and PyTorch now include pre-trained audio denoising models that can be fine-tuned on your specific recording environment.

The primary limitation is the need for a representative training dataset. If the model has not been exposed to your particular combination of noise sources, it may produce unpredictable results. For research groups running long-term monitoring programs, investing in a custom machine learning pipeline can reduce post-processing labor by orders of magnitude. A common approach is to use a U-Net architecture for time-frequency masking, where the model learns to suppress noise while preserving transients. Transfer learning from models trained on airborne speech denoising can be adapted to underwater sounds with additional training data from the target environment. Tools like Discovery of Sound in the Sea provide curated datasets that can serve as a starting point for training.

Balancing Noise Reduction with Signal Fidelity

Every noise reduction technique introduces a trade-off. Aggressive filtering removes noise but also attenuates or distorts the target signal. The key is to apply only as much processing as needed to make the recording usable for its intended purpose.

For quantitative acoustic analysis, such as measuring sound pressure levels or counting call rates, any processing that alters amplitude or temporal structure must be carefully documented and justified. For qualitative work, such as documentary audio or public outreach, listeners generally prefer a slightly processed recording that is pleasant to hear over a pristine but noisy one. Always keep an unprocessed backup of the original recording so you can revisit the data if analysis methods evolve. Use a consistent processing chain for all recordings in a study to ensure comparability.

One practical approach is to apply noise reduction in stages: start with minimal filtering (e.g., a gentle high-pass at 20 Hz to remove infrasonic noise), then listen critically, and only increase processing if the target signal remains obscured. Document each step with processing parameters saved in the metadata. For research publications, include a figure showing the spectrogram before and after processing to demonstrate the extent of noise reduction.

Final Checklist for Low-Noise Underwater Recordings

  • Use a hydrophone with self-noise at least 10 dB below the expected ambient noise floor at the recording site.
  • Suspend the hydrophone in a decoupled mount at least 2 meters from any rigid structure.
  • Record during slack tide, calm weather, and periods of low vessel traffic.
  • Set gain conservatively, leaving headroom for unexpected loud sounds.
  • Capture a noise-only sample at the beginning of each recording session.
  • Apply band-pass filters matched to the frequency range of your target sound.
  • Use spectral subtraction or wavelet denoising for persistent stationary noise.
  • Document all processing steps in the recording metadata.
  • Maintain equipment regularly and log environmental conditions.

Underwater audio is a window into a world that remains largely hidden from human senses. With careful equipment selection, disciplined field protocols, and thoughtful post-processing, you can cut through the noise and reveal the acoustic richness beneath the surface. The techniques outlined here form a practical framework that adapts to any environment, from a quiet freshwater lake to a bustling coastal ocean. Start with the fundamentals, listen critically at every stage, and refine your process with each deployment.