Enhancing Wildfire Propagation Model Predictions Using UAV Swarm-Based Real-Time Wind Measurements: A Conceptual Framework

Abstract

The dynamic behaviour of wildfires is mainly influenced by weather, fuel, and topography. Based on fundamental conservation laws involving numerous physical processes and large scales, atmospheric models require substantial computational resources. Therefore, coupling wildfire and atmospheric models is impractical for high resolutions, even for small scales. Instead, a static atmospheric wind field is typically input into the wildfire model, either as boundary conditions on the control surface or distributed throughout the control volume. Wildfire propagation models may be (𝑖) data-driven; (𝑖𝑖) theoretical; or (𝑖𝑖𝑖) mechanistic surrogates. Data-driven models are beyond the scope of this paper. Theoretical models are based on conservation laws (species, energy, mass, momentum) and therefore are computationally expensive—e.g. the Fire Dynamics Simulator (FDS). Mechanistic surrogate models do not closely follow fire dynamics laws but a different set of somewhat related laws observed to make predictions more efficiently and with sufficient accuracy—e.g. FARSITE and the Level-Set FDS (LS-FDS). Whether theoretical or mechanistic surrogate, these wildfire models may be coupled with or decoupled from wind models (e.g. the Navier-Stokes equations). Only coupled models account for the effect of the fire on the wind field. In this paper, a series of simulations of wildfire propagation on grassland are performed using LS-FDS to study the impact of the fire-induced wind on the fire propagation dynamics. Simulations of coupled and decoupled models are compared for different terrain slopes and atmospheric wind velocities. Results indicate that different coupling strategies can lead to the Rate of Spread (RoS) differing by as much as 23% for 30% up-slope terrains. Aiming to capture the fire-wind interaction without the hefty cost of solving the Navier-Stokes equations, a conceptual framework is proposed: 1) a swarm of UAVs measure wind velocities at a certain altitude; 2) the wind field is constructed with the acquired data; and 3) the wind field is downscaled and fed into the wildfire model periodically. As proof of concept, a series of simulations are performed using an in-house decoupled physics-based reduced-order fire propagation model (FireProM-F) enhanced by wind field "measurements". In this paper, wind velocities are not actually measured but extracted from high-fidelity physics-based Large Eddy Simulations within FDS (taken as ground truth). As expected, higher measurement frequencies lead to more accurate predictions of the RoS, and to more realistic formations of the propagating fire front. Finally, an initial attempt is made at studying the effect of wind measurement uncertainty on the model predictions by adding Gaussian noise. Preliminary results show that higher noise leads to marginally slower propagation of the fire front.