Technical Efficiency of Wildfire Detection and Machine Learning Predictions in Alberta, Canada

Abstract

Wildfire management agencies must continue to evolve and adjust to the dynamic nature of their industry. They face pressures that include a changing climate with the prospect of intense future fire seasons, tighter government budgets for wildfire detection and suppression, and the fast-pace of development of wildfire detection technologies such as remote-sensing devices, satellites, and drones. Alberta Wildfire, to fulfil its mandate of managing wildfire in the Forest Protection Area of Alberta, must adapt to these conditions by making cost-effective strategic decisions. As such, there is an increasing need for studies that examine the performance of Alberta’s wildfire detection system. This study offers two main contributions. First, we provide insights on the contribution of lookout towers in detecting wildfires and their role in the entire detection system. A production economics approach is employed to develop robust non-parametric Data Envelopment Analysis (DEA) models that estimate production frontiers that serve as a technical benchmark for lookout towers in Alberta’s detection system. Results from this analysis reveal a high-performing detection system and most lookout towers have high technical efficiency. Lookout towers operate close to the “state-of-the-art” technology frontier such that further productivity gains may require a new technology. Second, we explore the relationship between technical efficiency of lookout towers and local weather. We develop machine learning models that use local weather to classify lookout towers as technically (or productively) efficient or not. We find that weather can successfully predict a tower’s technical efficiency class, which suggests that there is a strong association between local weather and technical efficiency of wildfire detection.