Dynamic Data Driven Application for Forest Fire Spread Prediction

This work describes a two stages prediction method for wildland fire growth prediction. Proposed method takes advantege of genetic algorithms in order to develope a high performance and scalable application. Usually, a prediction is made using a forest fire simulator which receives several inputs (fire environment description) and it returns the state of the fire for a later instant of time. Having initial fire line and environmental characteristics, simulator uses some fire propagation model in order to simulate fire behavior (Figure 1 (a)). Taking into account this classical prediction method, we can see that it has the advantage of performing just one simulation (what means low processor time requirements). But this advantage is in a sense the main weak point of the method: final prediction quality depends on the suitability of the unique simulation (that means, using a unique input parameters set). The accuracy of the input parameters are really open to debate due to having its actual values is not easy, some times it is impossible. Consequently, we develope a method where a search of better parameter values is performed in order to reduce input data uncertainty. This