Towards a Wildfire Propagation Forecasting System Enhanced by UAV Swarms

Abstract

Wildfires have increased in frequency, duration and intensity during the past few decades primarily due to global changes in climate patterns. Wildfires are highly disastrous and impose significant negative economic, environmental, and societal impacts. Hence, even though their risk cannot be fully eliminated, developing reliable management and mitigation strategies is essential. Yet, accurate predictions of wildfire propagation dynamics are critical to enable reliable management, mitigate fire propagation and make timely decisions. Nevertheless, even at small scales, a thorough physics-based simulation of this dynamic behaviour can be costly and time-consuming, especially when fire-wind interactions are considered. A different group of models are developed to bypass this restriction by reducing or ignoring some physical rules, often at the expense of accuracy, to produce faster-than-real-time estimates of fire behaviour. However, predictions from these models are more questionable since they are decoupled from the atmosphere. This research proposes a shift towards exploring the applications of unmanned aerial vehicle (UAV) swarms in enhancing existing wildfire emergency response systems, focusing on their capabilities in improving the accuracy of faster-than-real-time wildfire predictions. While UAVs have been utilized and proven successful in performing wildfire detection and monitoring tasks, their integration into existing wildfire emergency response systems has not been clearly explored, nor have their applications in aiding operational wildfire propagation forecasting models by providing real-time wind measurements. This research primarily aims to develop solutions enabled by UAV swarms and artificial intelligence (AI) to improve the performance of existing decoupled wildfire propagation prediction models to assist emergency response management and decision making. As a secondary contribution, it also identifies opportunities for better integration of UAVs into wildfire emergency response systems.
Leveraging technologies such as UAV-based wind measurement, deep learning (DL), and computational fluid dynamics (CFD), this project involves designing and developing a framework to improve the accuracy of operational wildfire predictions. The framework is designed to estimate the high-resolution near-surface fire-driven wind field from the sparse UAV swarm-based wind measurements taken at the flight altitude. The estimated wind field would serve as input to decoupled operational models to improve their estimation accuracy and reduce uncertainty by mimicking the effect of fire-wind interaction on wildfire propagation. Consequently, the contributions of this research include: (i) developing a conceptual framework for the integration of UAVs within the existing wildfire emergency response systems, highlighting opportunities for maximizing the benefits of leveraging UAVs, (ii) designing and developing a framework to improve the accuracy of decoupled wildfire prediction models through continuously feeding model with wind fields constructed from UAV swarm measurements, (iii) design and develop an AI-driven framework for estimating the near-surface wind field from sparse UAV-based wind measurements, (iv) a machine learning framework based on combining data engineering, automated model fine-tuning, and utilizing advanced ML models for improving the accuracy of vegetation classification from cartographic data with the overall aim of improving wildfire prediction models reconciling the importance of vegetation type as input into the wildfire model, and finally (v) a detailed study of the effect of vegetation characteristics on wildfire propagation dynamics in grasslands, highlighting the importance of developing comprehensive understanding of fire dynamics to enable accurate predictions.
The developed frameworks are expected to impact society and wildfire response systems positively by providing more accurate estimations of wildfire propagation as well as extracting the full potential of UAVs in wildfire fighting operations.

Date of Award	Jul 2025
Original language	English
Awarding Institution	Coventry University
Supervisor	Mauro Innocente (Supervisor) & Jesper Christensen (Supervisor)