FDTD Model of Noise Pollution from an AI Data Center

arXiv:2610.14899 · physics.comp-ph, cs.LG · Submitted 2026-10-10 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: AI Radio. Generated commentary on the latest Artificial Intelligence papers.

Tom: Today's paper: "FDTD Model of Noise Pollution from an AI Data Center".

Jane: The gist The FDTD model simulates sound propagation over topography near an AI data center to predict noise pollution levels,

Tom: First, who's behind it and why it matters.

Paper summary: Jane: We've looked at the "FDTD Model of Noise Pollution from an AI Data Center" paper, and it’s been clear that this FDTD model uses finite-difference time-domain equations to simulate sound propagation over topography near an AI data center <ref:2610.14899#pg3>.

Tom: The authors are essentially demonstrating how sudden elevations and topographical features cause localized regions of higher noise pollution, which they calculate in decibels <ref:2610.14899#pg3>. It’s about showing that the environment dictates where the noise lands, not just a straight line away from the source <ref:2610.14899#pg3>.

Lu: The paper is focused on capturing the dynamic pressure velocity equations of sound waves propagating through realistic terrain <ref:2610.14899#pg3>, and they address acoustic refraction and vegetation as factors that affect this sound movement <ref:2610.14899#pg3>.

Meng: What's the actual implication for us right now? If we're looking at deploying these AI centers, does this paper give us any concrete guidance on mitigation strategies based on terrain?

Tom: It suggests that understanding the role that topography plays in noise pollution, and capturing it in specific topographies, can help identify areas where noise mitigation strategies might be best employed <ref:2610.14899#pg1>.

Jane: So, the implication is that we need to map the terrain before we worry about just building a data center; you need to consider how the land shapes the sound pollution profile <ref:2610.14899#pg3>.

Lalam: I'm confident this work points toward better spatial planning for where these types of energy-intensive operations are sited relative to human communities <ref:2610.14899#pg2>.

Tom: Exactly. The paper, titled "FDTD Model of Noise Pollution from an AI Data Center," shows that the physical shape of the land is a major variable we need to factor into our noise assessments <ref:2610.14899#pg3>.

Conclusion: Tom: So we’ve looked at how this FDTD model simulates sound waves over topography near an AI data center, and now we need to look at what this whole thing means for us <ref:2610.14899#pg3>.

Jane: Right. The title is "FDTD Model of Noise Pollution from an AI Data Center." It sounds very technical, but the core idea is pretty simple: they are modeling how sound bounces off the land near these massive data centers and how that changes the noise level you actually hear.

Lu: Yeah, what’s interesting is that they aren't just modeling a straight line path for sound. They’re using the FDTD method to capture all these dynamic pressure velocity equations of sound waves moving over real terrain <ref:2610.14899#pg3>.

Meng: So if you boil it down, this paper is showing that sudden hills and specific land features really matter where the noise pollution lands, not just how far away the data center is.

Tom: Exactly. They’re looking at how acoustic refraction and vegetation influence that sound movement <ref:2610.14899#pg3>. It’s about showing that the environment dictates where the noise hits <ref:2610.14899#pg3>.

Jane: And what they found is localized regions of much higher noise pollution when sound waves reflect off those sudden elevations <ref:2610.14899#pg3>. It’s not just a uniform noise floor spreading out everywhere.

Lalam: From where I'm sitting, this suggests that if we’re planning where to put these AI facilities, you can't just look at the distance from a nearby town <ref:2610.14899#pg2>. You have to map the local topography first.

Tom: So, what does this change for the people listening right now? It means that when you think about noise pollution from these centers, you need to consider the actual shape of the ground where they are built <ref:2610.14899#pg3>.

Jane: Right. This leads us to thinking about how we actually apply this kind of spatial awareness in real-world planning for these AI operations.

Gavin Buxton

Robert Morris University · Pennsylvania Department of Conservation and Natural Resources · Pennsylvania Spatial Data Access

physics.comp-ph, cs.LG

Submitted: 2026-10-10

Updated: 2026-10-10

The gist: The gist The FDTD model simulates sound propagation over topography near an AI data center to predict noise pollution levels, showing that sudden elevations and topographical features significantly

Key concepts

Finite-Difference Time-Domain (FDTD)
FDTD is a numerical method used to solve the dynamic pressure velocity equations for sound waves. It works by discretizing space and time into small steps, allowing researchers to track how sound pressure changes over time as it moves through a defined physical area.
Topographical Effects
This refers to how the physical shape of the land, such as sudden hills or slopes, affects noise propagation. The model demonstrated that when sound waves encounter these sudden elevations, they cause reflections and focusing effects that result in localized areas with much higher noise pollution than expected from simple spreading.
Perfectly Matched Layers (PML)
PML are special boundary conditions used in the simulation to mimic an infinitely large environment. They absorb outgoing sound waves at the edges of the computational domain, preventing reflections from bouncing back into the main simulation area and ensuring accurate results for wave propagation.
Tortuosity ($ au$)
Tortuosity is a physical parameter used to modify sound speed in porous media. It represents how winding or convoluted a path is through the material. In this model, increasing tortuosity reduces the effective speed of sound, which affects how noise travels through the surrounding terrain.

Summary

The gist The FDTD model simulates sound propagation over topography near an AI data center to predict noise pollution levels, showing that sudden elevations and topographical features significantly influence noise levels.

Introduction and Problem Statement

Low-frequency noise pollution from AI data centers is a concern for local residents because it can carry significant health impacts. The rapid growth of AI has caused an unprecedented increase in electricity demand which stresses the capacity of local grid infrastructure. Traditional data centers have been found to generate low frequency noise pollution exceeding 60 dB in the local community, that detrimentally impacts the health of local residents. Environmental noise can lead to non-auditory health effects ranging from annoyance and sleep disruption to an elevated risk of cardiovascular disease.

Modeling Approach

The finite-difference time-domain (FDTD) method is used to capture the dynamic pressure velocity equations of sound waves propagating from a central region over realistic terrain. The site of an AI data center currently under construction in Beaver County, PA and the surrounding terrain constitutes the computational domain and the effects of acoustic refraction and vegetation are considered. The system size in two-dimensions is 29154×3000, with a spatial resolution of ∆x = 0.1715 m and a time step of ∆t = 1 × 10−4 s. The FDTD equation for updating the local pressure is of the form p n+1 i,j = p n i,j − ρ´i,j c´2 i,j∆t [v n+ 1/2 x,i+1,j − v n+ 1/2 x,i,j ∆x + v n+ 1/2 y,i+j+1 − v n+ 1/2 y,i,j ∆y].

Physical Parameters and Modifications

The density of air, ρ, depends on the temperature, T, and is given by ρ = P / RairT 3. The speed of sound also depends on the air temperature, and is given by c = 331.32√ [1 + TC 273.15] where TC = T −273.15. The modified density is given by ρ´ = ρτ / ϕ where ρ is the density of air, τ is the tortuosity, and ϕ is the porosity of the media that the sound wave is propagating through. Similarly the speed of sound is reduced, as τ is increased: c´ = c / √τ. The flow resistivity, R, has units of kPa s/m2.

Boundary Conditions and Source

To terminate the FDTD model we require absorbing layers at the boundary that restrict reflection. The perfectly matched layers (PML) boundary conditions are the standard manner of achieving this. Near the boundary, 300 lattice steps deep, is a region where the waves propagating in from the simulation domain are gradually absorbed with little reflection to mimic an infinitely large domain. The source of the sound wave is taken to be a region 30 m high on top of a 5 m by 5 m structure, to mimic the emission from the rooftop of an AI data center. The source is a 100 Hz sound wave in a Gaussian envelope given by v n+1/2 i,j = A0 sin (2πf · t) exp (–(t − t0)2 / 2σ2t).

Results and Topographical Effects

The predicted sound level at an angle of 140◦ from the AI data center is plotted in Fig. 1b. The black line is the sound level in decibels, the light gray line represents the spherical spreading of sound waves, and the colored line is the elevation, as a function of distance from the AI data center. The sound waves traveling from the AI data center, and reflecting from other surfaces, impact these regions of sudden elevation and result in localized regions of higher noise pollution.

Improvements for AI systems

  1. Topography-Aware Noise Mitigation Design: AI systems can incorporate topographical data, such as that used in this FDTD model, into their site selection algorithms to proactively identify areas more susceptible to high noise levels and avoid them when deploying critical infrastructure.

  2. Localized Acoustic Risk Assessment: The system can utilize the predicted sound level contours (Fig. 2) to provide specific risk assessments for local residents, identifying areas where noise pollution may be significant versus those where it can experience very low noise pollution.

  3. Vegetation and Atmospheric Correction Modules: To improve prediction accuracy, the AI system can integrate modules that account for environmental factors, such as the damping effects of vegetation (Figure 3) and localized acoustic refraction caused by temperature gradients (Fig. 4).

  4. Targeted Mitigation Strategy Recommendation: Based on the simulation results, the system can recommend specific mitigation strategies tailored to topographical features, such as identifying areas where noise mitigation strategies might be best employed.

Sources

Related papers