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 related laws observed to make predictions more efficiently and with sufficient accuracy—e.g. FARSITE, and FDS with the Level Set model (FDS-LS). Whether theoretical or mechanistic surrogate, these wildfire models may be coupled with or decoupled from wind models (e.g. 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 FDS-LS to study the impact of the fire-induced wind on the fire propagation dynamics. Results show that coupling leads to higher Rates of Spread (RoS), closer to the values reported from field experiments, with increasing wind speeds and higher terrain slopes strengthening this trend. Aiming to capture the fire–wind interaction without the hefty cost of solving Navier-Stokes equations, a conceptual framework is proposed: 1) a swarm of unmanned aerial vehicles measure wind velocities at flight height; 2) the wind field is constructed with the acquired data; 3) the high-altitude wind field is mapped to near-surface, and 4) the near-surface wind field is fed into the wildfire model periodically. 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 proof of concept, wind velocities are not actually measured but extracted from high-fidelity physics-based Large Eddy Simulations within FDS, which are taken as ground truth. Unsurprisingly, higher measurement frequencies lead to more accurate predictions of the RoS and to more realistic formations of the propagating fire front. An initial attempt is made to study the effect of wind measurement uncertainty on the model predictions by adding Gaussian noise. Preliminary results show that higher noise leads to the fire front displaying a less smooth shape and slower propagation.

A High-resolution Large-eddy Simulation Framework for Wildland Fire Predictions using TensorFlow

FireBench: A High-fidelity Ensemble Simulation Framework for Exploring Wildfire Behavior and Data-driven Modeling

Computational modeling of extreme wildland fire events: A synthesis of scientific understanding with applications to forecasting, land management, and firefighter safety

Evaluation of a data-driven wildland fire spread forecast model with spatially-distributed parameter estimation in simulations of the FireFlux I field-scale experiment

A wildland fire modeling and visualization environment

WRF simulation over complex terrain during a southern California wildfire event

Real time simulation of 2007 Santa Ana fires

Developing a Geospatial Data-Driven Solution for Rapid Natural Wildfire Risk Assessment

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

Toward an integrated system for fire, smoke, and air quality simulations

Emulating Wildfire Plume Injection Using Machine Learning Trained by Large Eddy Simulation (LES)

Coupled atmosphere-wildland fire modeling with WRF-Fire

Predicting Fire Propagation across Heterogeneous Landscapes Using WyoFire: A Monte Carlo-Driven Wildfire Model

An interpretable wildfire spreading model for real-time predictions

Fire and Smoke Digital Twin -- A computational framework for modeling fire incident outcomes

Uncertainty Aware Wildfire Management

Fire and smoke digital twin – A computational framework for modeling fire incident outcomes

Discrete Wildfire Simulation Case Study

Coupling Physical Model and Deep Learning for Near Real-Time Wildfire Detection.

Machine Learning and Artificial Intelligence for Wildfire Prediction

Physics-Informed Machine Learning Simulator for Wildfire Propagation