We present PhotonTracer, a newly developed light-scattering simulation in the geometric-optics limit accelerated on the Graphics Processing Unit (GPU). It is written in CUDA/C++, leveraging the efficient and hardware-accelerated ray-geometry intersection provided by the NVIDIA OptiX™ SDK. It is distributed as a Python package, which makes it accessible to the community. Thanks to the achieved performance, the simulation of light transport and reflectance in highly multiple-scattering semi-infinite media becomes possible on affordable consumer GPUs. A wide range of simulation parameters is accessible through customizable per-ray output buffers. This enables the calculation of many observables relevant to remote sensing, such as albedo, bidirectional reflectance distribution function, scattering phase matrices, or absorption depths. PhotonTracer is validated against two existing simulations. The scattering phase matrix elements for single particles are validated against the ray tracing simulation SIRIS4. The reflectance, transmission and absorption for a layered medium are validated against a Transfer Matrix Method simulation. To illustrate one possible application, the reflectance spectra of semi-infinite particulate ice samples with different mean particle diameters and shapes are simulated and compared to laboratory spectra of well-characterised analogues. We find a good agreement between the simulated and measured spectra, especially for the shape of the water absorption bands. PhotonTracer is well-suited to study the effects of a wide range of medium properties on light scattering. A speedup of 3 to 4 magnitudes over traditional CPU models enables the simulation of highly multiple-scattering systems without the need for simplifying assumptions. This makes it a valuable tool, which enables the inversion of remote sensing and laboratory data and can serve as a reference for analytical models. • A GPU-accelerated light-scattering model in the geometric optics limit is presented. • Highly multiple-scattering systems with negligible absorption can be simulated. • It is distributed as a Python package, making it easily accessible to the community. • A speedup of 5000x was achieved on the GPU over comparable CPU-based models.
Ottersberg et al. (Sun,) studied this question.