Fire risk assessment in WUI – WII implementing bayesian networks to infer fire spread probabilities

Abstract

The current state of development achieved in our society parallels our impact on the environment. Even though natural disasters have shaped human history, anthropogenic processes also catalyze large-scale disasters, being climate change a paradigmatic example: droughts and high temperatures trigger wildfires beyond what is acceptable to be environmentally sustainable. Unprecedented levels of industrialization and urbanization in history foster wildfire menace. Wildfire-Urban Interfaces (WUI) and Wildfire-Industrial Interfaces (WII) are relevants domains of wildfire impact, demanding efforts in all the aspects of the disaster management cycle to build resilience. In this scenario, fire modelling tools help assessing risks in those interfaces. Fire spread is a complex physical phenomenon. To that aim, WUI-WII fire spread models need to be physically sound and scalable to geographically extended areas, which demands alternative computational approaches to comprise every possible fire exposure risk. Dynamic Bayesian Networks (DBN) are a promising tool to infer fire spread probabilities, which depend on the landscape, wind, fuel constituents and weather variables. Our work explores novel efficient ways of interpreting fire spread as marginalizing node probabilities from network topologies representing geographical scenarios. DBN General-purpose commercial libraries are available, but they are not designed explicitly for modelling spatio-temporal physical problems. This limits their applicability to dynamical parameter updating or their impossibility to infer the main underlying physical parameters. To that aim, we propose a framework that will ultimately help to develop new open-source codes that may contribute to the field in future research.