Saturday, 27 September 2014

Forecasting the movement of a wildfire

I have a dream too. That one day we could predict the movement of a fire and be able to inform the Fire Service ahead of time by using just an iphone.

Hopefully our recent paper (published in open access) in Natural Hazards and Earth System Sciences moves us closer to that end, as it reports an algorithm that can forecast the movement of a wildfire when observations of the locations of the flames are available.

Five assimilated fire fronts with 1 min intervals (black solid lines). The first guess (red dashed line) is taken to be far from the true invariants vector to check the algorithm capability to converge. A 10 min forecast (blue solid lines) is also calculated using fuel depth as sensor data.
The title of the paper is "Forecasting Wind-Driven Wildfires Using An Inverse Modelling Approach". The fire model at the core of the forecast algorithm combines the classical theory of Rothermel's rate of spread with a perimeter expansion model (based on Huygens principle for the propagation of waves). We then pose the  problem as an optimisation and force our fire model to predict well any past observations of the real fire that might be avaible at that moment. Observations of the location of the fire can be produced from any system like from personnel in the field (deployed fire fighters), drones, airplanes or satellites. Once all the past observations are predicted, we consider that we have found the true characteristics of this particular fire and launch forecasts into the future. Where would be this fire be in 10 min or 1 h?

We have investigated the skills of the algorithm using synthetic data (not real fire data) and the results show it is very quick and decently accurate, and predicts the location of the front ahead of time. It needs further work to increase its accuracy, of course, but we already see the greatest strengths of our method are lightness, speed and flexibility. We specifically tailor the forecast to be efficient and computationally cheap so it can be used in mobile systems for deployment by the Fire Service. For example, in an iphone.

Hope Apple knock on our door one day.