Tomskiy universitet sistem upravleniya i radioelektroniki
Russian Federation
This paper presents a comparison of computational methods for calculating electromagnetic radiation scattering by spherical particles within the framework of Mie theory, applied in systems for retrieving microphysical parameters of atmospheric aerosols. Benchmark testing of three approaches was conducted: the PyMieScatt library, a vectorized NumPy implementation, and a fully vectorized GPU-based calculation using TensorFlow. It is shown that for parametric analysis tasks, the vectorized NumPy implementation provides a 49–50x speedup relative to sequential PyMieScatt calls. Limitations of GPU acceleration related to video memory consumption during full vectorization across all parameters are discussed.
Mie theory, light scattering, atmospheric aerosols, computational vectorization, GPU, optical modeling
1. Mishchenko M.I., Travis L.D., Lacis A.A. Scattering, Absorption, and Emission of Light by Small Particles. Cambridge University Press, 2002.
2. Bohren C.F., Huffman D.R. Absorption and Scattering of Light by Small Particles. Wiley, 1983.
3. Marichev V. N., Matvienko G. G., Yushkov V. A., Balugin N. V., Bochkovskiy D. A. Lidarno-ballonnyy eksperiment po issledovaniyu stratosfernogo aerozolya dlya klimaticheskih nablyudeniy i diagnosticheskih zadach // Meteorologiya i gidrologiya. 2022. № 11. S. 41-47.
4. Sumlin B.J., Heinson Y.W., Chakrabarty R.K. PyMieScatt: A Python package for Mie scattering calculations // Journal of Open Source Software. 2018. DOI:https://doi.org/10.21105/joss.00678
5. TensorFlow Documentation. Best practices for memory management. URL: tensorflow.org/guide/memory
6. Micikevicius P. et al. Mixed Precision Training // ICLR. 2018. arXiv: 1710.03740
