Agent-based modelling for wildfire behaviour prediction

The Third United Nations World Conference on Disaster Risk Reduction, held in Sendai in 2015, has defined a global strategy directed at enhancing risk-exposed communities’ resilience. In line with those needs, the study intends to improve and optimize decision-making processes in wildfire risk management by implementing predictive spatially distributed models of wildfire behaviour.

The proposed methodology has been applied to simulate some large and fully documented wildfire events in Umbria and Sardinia regions, in Central and Southern Italy respectively. The predictive model for wildfire behaviour is based on the reviewed Rothermel’s quasi-empirical mathematical model, which investigates propagation-driving parameters, i.e. the local geomorphometrical and meteorological parameters along with the pyrological and phenological characteristics of the local plant communities, to estimate the rate of spread of the fire. Propagation-driving parameters and their spatiotemporal variability have been estimated in the pre-fire environment by applying and adapting empirical relationships well-established in literature. Remote sensing-derived data have been analysed over phenologically distinct periods, along with ancillary data, to elicit information necessary to distinguish the mosaic of fuel model types and to monitor spatiotemporal variations in either live or dead fuel moisture content. According to input data availability, the methodology has been adapted to different case studies, focusing major attention on MODIS instrument by NASA on board the Terra satellite as well as on Sentinel constellations of satellites of the ESA Copernicus programme due to their accessibility and to their medium-high spatial and temporal resolution. A two-dimensional Agent-Based Model with a hexagonal grid, which, given a map of the rate of spread and an ignition point as inputs, returns a map of the cumulative propagation time, has been developed in order to simulate the wildland surface fire behaviour.

Satellite estimated propagation-driving parameters have been compared with information collected in the field and recorded by the regional annual reports on wildfire events, revealing a good predictive ability. Likewise, the wildfire behaviour model has provided accurate predictions, up to 70% in terms of morphological matching between obtained simulations and respective documented historical events boundaries, also if compared with results from other well-known wildfire simulation toolset and software. Obtained results suggest the developed wildfire behaviour model could represent a promising tool in prioritizing firefighting interventions in near-real time.