Data Assimilation for Wildfire Spread Simulation with Sparse and Partial Observation Data from Unmanned Areial Vehicles

Wildfire spread simulation approximates the dynamic spread of wildfires by modeling fire spread behavior under various fuel, terrain, and weather conditions. To achieve real-time fire spread prediction using wildfire spread simulation models, it is essential to assimilate real-time observation data from active fires into the simulation models. Unmanned Aerial Vehicles (UAVs) has been increasingly used to collect data from wildfires. Data assimilation methods that assimilate UAV-based observation data are widely used for supporting wildfire spread simulation. This dissertation formulates and establishes data assimilation that works with a discrete event wildfire spread model and UAV-based observation data. Particle filter (PF)-based data assimilation algorithms are developed to carry out the data assimilation task. The PF-based data assimilation algorithms are effective when the UAV-based observation data is sparse, partially observed, and noisy. Additionally, an advanced PF approach based on time-space confidence of the UAV observation data is established which can handle more challenging scenarios. Experiment results show that the developed data assimilation method for calibrating UAV-based fire data for wildfire spread simulation is effective and robust.