Short term forecasting of large scale wind-driven wildfires using thermal imaging and inverse modelling techniques
Document typePart of book or chapter of book
PublisherImprensa da Universidade de Coimbra
Rights accessRestricted access - publisher's policy
A key factor in decision-making process during a wildfire incident is counting on the forecast of how the fire is likely to behave in different fuels, weather conditions and terrain. Wildfire models and simulators attempt to assist fire responders in gaining understanding of the fire behaviour. The main hurdle to overcome when applying such technologies at operational level is the lack of a complete model that describes wildfire governing physics and the trade-off between accuracy and computing time. A forecasting prediction must be delivered within a positive lead time and current physical models are far beyond this requirement. Inverse modelling and data assimilation techniques offer a great potential of operational applicability in wildfires, coupling fire monitoring and fire behaviour forecast at real time. With this approach, a better description of the processes simulated by the fire behaviour models can be achieved when adding real-state information of the system, since discrepancies between simulated fire behaviour variables and observed variables are minimized. The use of this approach accelerates fire simulations without loss of forecast accuracy. In this paper we explore the adaptation to real fire scenarios of a synthetic-data-based inverse modelling structure for fire behaviour forecast. Improvements are investigated to extrapolate the already existing algorithm to real data assimilation from IR aerial monitoring. The technique explores elliptical Huygens expansion coupled with simple -yet effective- semi-empirical wildfire models. The algorithm assimilates fire fronts positions extracted from airborne thermal imaging and additional available data as wind speed and direction or fuel characteristics. The invariants -set of governing parameters that are mutually independent and constant for a significant amount of time- are resolved by means of forward model and linear tangent minimization. The technique has been adapted to be employed in large-scale mallee-heath shrubland fires experiments conducted in South Australia in 2008. Fires were filmed with a helicopter transported TIR camera. The IR images were processed to obtain the position of the fire perimeter at a maximum frequency of one isochrone every 10 seconds. The algorithm shows great capability to simulate fire fronts observations and opens the door to keep developing a fully automatic data assimilation algorithm with forecasting capacity.
CitationRios, O.; Pastor, E.; Planas, E. Short term forecasting of large scale wind-driven wildfires using thermal imaging and inverse modelling techniques. A: "Advances in Forest Fire Research". Coimbra: Imprensa da Universidade de Coimbra, 2014.