Investigating extreme fire behavior in complex terrain using high-resolution large-eddy simulations on ML-enabled compute infrastructure

Wildland fires in regions of complex terrain are often associated with extreme fire behavior.
Understanding the interaction between complex terrain and atmospheric flows that results in these extreme conditions is thus an important endeavor. In particular, sloped terrain can lead to highly dynamic wind patterns, thereby enhancing turbulence which in turn directly affect all modes of heat transfer, resulting in more severe fire behavior. Thus, by leveraging a recently developed physics-based solver Swirl-LM based on TensorFlow and running on Tensor Processing Units, we investigate extreme fire behavior in complex terrain in a prescribed fire scenario. We discuss how thermal instabilities contribute to turbulence generation, and how coupled fire-atmosphere interactions generate a circulation of the convective smoke column.
Comparison with experimental data allows us to validate our numerical models and opens pathways for simulations of extreme fires in complex terrain at affordable computational cost.