Modeling Multiple Scattering in Volumetric Light Transport

16 concepts16 papers2 proposals

Imagine shining a flashlight through a clear room versus a room filled with thick fog. In the clear room, light travels in straight lines. In thin fog, a photon might hit exactly one water droplet before bouncing directly into your eye. This scenario is known as the single scattering approximation. It is mathematically elegant because the relationship between the light source, the scattering particle, and the observer is linear and easily traceable.

However, as the fog thickens into a dense cloud or as light travels deeper underwater, this approximation completely breaks down. A photon will strike a particle, bounce to another, then another, and another before finally escaping the medium. This phenomenon is multiple scattering.

Single Scattering:   [Light Source] ------> (Particle) ------> [Sensor]

Multiple Scattering: [Light Source] ------> (Particle) ----> (Particle) ----> (Particle) ------> [Sensor]

When light bounces repeatedly within a participating medium, directional information is scrambled. What was once a crisp, directional beam of light softens into a diffuse, isotropic glow. Understanding and modeling this complex behavior is the core challenge of volumetric light transport.

The Breakdown of the Single Scattering Assumption

For decades, fields ranging from computer graphics to remote sensing relied on the single scattering assumption to keep computations tractable. But when environments become highly turbid or dense, ignoring multi-bounce interactions leads to severe inaccuracies.

In real-time rendering, analytic models that rely exclusively on single scattering fail to capture the rich diffusion and volume density characteristic of environments like thick fog or heavy cloud layers, as noted by Sun et al.. The problem extends far beyond graphics:

  • Optical Communication: Traditional models for non-line-of-sight ultraviolet communication assume a single-bounce path. Ding et al. demonstrated that this simplification fails to predict signal path loss over long distances or in dense atmospheric conditions, where multiple scattering events are highly probable.
  • 3D Reconstruction: Computer vision techniques like photometric stereo assume light travels unhindered to an object and back. In turbid media, multiple scattering introduces non-linear paths that invalidate these geometric assumptions, requiring significantly more complex mathematical frameworks to resolve the true 3D shape, as shown by Murez et al..
  • Atmospheric Sensing: Forward models for lidar inversion often assume multiple scattering is negligible. Marais et al. observed that this assumption breaks down in high-density environments like thick clouds, necessitating complex radiative transfer modeling to accurately preserve cloud and aerosol boundaries.

Simulating the Chaos: Forward Models

To accurately render or simulate dense media, researchers must move from simple ray-tracing to solving the Radiative Transfer Equation (RTE). The RTE models how light energy is stochastically absorbed, scattered, and emitted as it travels through a uniform volume (Hu et al.).

One foundational way to solve this is through Monte Carlo simulations. As derived by Pattanaik et al., particle-based light transport can be extended into volumes by tracking how individual light particles stochastically lose energy or change direction upon every single collision within the medium.

Explicitly accounting for these multi-bounce interactions is what unlocks high-fidelity realism. For instance, Haber et al. demonstrated that the nuanced color distributions and atmospheric gradients of twilight skies cannot be captured by single scattering alone; the long path lengths of twilight amplify multiple scattering, which fundamentally dictates the sky's appearance.

Inverting the Blur: Reconstruction Through Dense Media

If multiple scattering scrambles light paths and obscures the source, how can we possibly "see" through dense media or reconstruct 3D volumes from 2D images? The key insight is that multiple scattering is not random noise—it is a deterministic physical process governed by the density and structure of the medium.

Differentiable Rendering and Generative Priors

A powerful modern paradigm pairs physically-based differentiable renderers with generative models to solve highly ill-posed inverse problems, such as reconstructing a 3D cloud from a single photograph.

Leonard et al. address this by using a neural diffusion model as a structural prior alongside a renderer that explicitly simulates multiple scattering. Because the forward renderer models how light actually bounces through the volume, the system can backpropagate the error between the rendered image and the photograph. This allows it to iteratively optimize and disentangle the object's physical density from the complex, scattered lighting environment.

Turning Scattering into Information

Instead of trying to eliminate scattered light, some frameworks treat it as a rich source of data. Nayar et al. rethought structured light scanning for murky environments. By modeling the physics of how a projected light pattern scatters off particles, the authors demonstrated that the resulting multiple scattering can be mathematically resolved to restore clear-air appearances and achieve accurate 3D reconstructions.

In extreme cases where high opacity completely destroys direct directional information, researchers shift from ray models to diffusion-based transport models. For example, Borcea showed that high-energy laser beams can be imaged and tracked in dense atmospheres by analyzing the statistical coherence and speckle patterns of off-axis scattered intensity, bypassing the need to be in the beam's direct footprint.

Computational Bottlenecks and Frontiers

While the physics of multiple scattering are well-understood, implementing them computationally remains a massive bottleneck. Simulating thousands of stochastically bouncing photons is incredibly expensive. Consequently, many volumetric rendering techniques still intentionally focus on single scattering and transmission to maintain interactive frame rates, sacrificing high-fidelity realism in dense volumes like clouds (Zhang et al.).

The challenges multiply when attempting to optimize or edit volumes programmatically:

  • Derivative Complexity: In volume stylization and painting pipelines, extending single scattering models to handle multiple scattering introduces severe computational and memory demands, particularly when calculating the complex derivatives required for optimization (Klehm et al.).
  • Representational Limits: Modern point-based representations, like 3D Gaussian Splatting, struggle with dense media. As noted by Chen et al., standard directional decomposition functions can capture basic anisotropic surface reflections, but they lack the mathematical flexibility required to approximate the high-dimensional light transport inherent to multi-bounce volumetric scattering.
  • Environmental Factors: Real-world media are rarely uniform. Standard radiance fields often ignore critical factors like the localized absorption and scattering properties of deep or turbid water, which are essential for accurately representing visibility and color shifts (Zhan et al.). Furthermore, micro-physical variations—such as whether atmospheric particles are water-coated or ice-coated—significantly alter single and multiple scattering dynamics, directly impacting how structured light beams attenuate in real-world weather (Cao et al.).

The trajectory of this research direction points toward a unified goal: replacing cheap linear approximations with deep physical models and differentiable frameworks, turning what was once considered destructive visual noise into a structured map of the physical world.

Key Foundational Work

Computation of global illumination in a participating medium by monte carlo simulation (1993) — contributed 4 concept elements to this direction.
Structured Light in Scattering Media (2005) — contributed 1 concept elements to this direction.
Physically-based simulation of twilight phenomena (2005) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_multiple_scattering_effects_cba69db0,
  title = {multiple scattering effects},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/multiple-scattering-effects/rdr_cba69db0241249a5944e6d42324f604c},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_multiple_scattering_effects_cba69db0_v1,
  title = {multiple scattering effects},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/multiple-scattering-effects/rdr_cba69db0241249a5944e6d42324f604c/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake