In this work we have developed a procedure to predict fuel types for wildfires simulations. The procedure consists on the pre-processing of satellite imagery, its labeling and the prediction for each pixel with a neural network.