Wildland and Forest Fire Prediction in Thailand using Satellite Data

Fire, either manmade or natural causes, poses significant threats to ecosystems, infrastructure, and all lives. Early prediction and monitoring are critical for fire management and mitigation. However, a lack of workforce, and it is insufficient to rely only on conventional monitoring techniques, such as employing human staff to operate lookout towers for fires or waiting for someone to call for an emergency. In previous work, satellite data can be used for fire prediction in many countries. In this work, we exploited 10 years-hot spots data from NASA satellite and Thai meteorological data from 2012-2022 and applied machine learning techniques for fire prediction in Thailand. In this work, we used the Extratrees BAG L2 model (this model consists of ExtraTrees, Bootstrap Aggregating, and Regularization) for fire prediction and evaluated prediction results using Mean squared error(MSE), Root mean squared error(RMSE), Mean absolute error (MAE) and R square. We obtained the values of MSE, RMSE, MAE and R square are 0.0057, 0.075, 0.04355 and 0.6869, respectively.