Wildfire Forecasting with Point Processes

Abstract: We develop a spatio-temporal Hawkes model for wildfire spread in Sicily over the period 2001--2025, with particular focus on a covariate-dependent branching ratio that quantifies the strength of self-excitation. The motivating idea is that, conditional on an initial fire, subsequent ignitions and detections are generated through a spread mechanism whose intensity varies with environmental conditions. Since satellite detections are observed only burstwise at irregular overpass times, standard continuous-time inference is not directly applicable. We therefore make use of the branching representation of the Hawkes process and assume that each ignition can be linked to its fire of origin, yielding a practical basis for inference from clustered overpass data. The resulting framework allows the triggering intensity to depend on meteorological covariates, fire-weather indices, local fuel composition, and topography.

Mathematics

Audience: general audience

Gothenburg PhD seminar

Series comments: Rooms and times may vary, please check the latest update. In-person only.