The present work describes a methodology based on Artificial Neural Networks (ANN) and multitemporal images from the MODIS/Terra-Aqua sensors in order to detect areas with high risk of forest fire in the Brazilian Amazon. The hypothesis of this work is that, due to the characteristics of land use and land cover change dynamics in the Amazon forest, the temporal spectral profile of forest areas preparing to be burned can be separated from other areas. A study case was carried out in three municipalities in the north region of Mato Grosso State, Brazilian Amazon. Feedforward ANNs, with different architectures, were trained with a backpropagation algorithm, taking as inputs the NDVI values calculated from MODIS images acquired during five different periods preceding the forest fire season. Samples were extracted from areas where forest fires were detected in 2005, and also from forest and agricultural areas. These samples were divided to train, to validate and to test the ANN. The tests results achieved a mean squared error of around 0.07. When simulated in an entire municipality, the ANN model was efficient in showing the spatial distribution of the forest fire probability, which was coherent with the fire events observed in the following months.
Maeda, E.E. et al. (2009): "Forest Fire Risk Mapping in the Brazilian Amazon Using MODIS Images and Artificial Neural Networks", Anais XIV Simpósio Brasileiro de Sensoriamento Remoto, Natal, Brasil, 25-30 abril 2009, INPE, p. 1425-1432.
