Using Deep Learning Algorithms in Forecasting the Severe Haze Events in Southeast Asia

Adverse environmental and weather events cause substantial economic loss. For instance, particulate pollution has become a serious environmental and societal issue of many Southeast Asian countries in recent decades. Forecasting the occurrence of these high-impact events could allow mitigation measures to be implemented ahead of time and thus effectively minimize economic loss. Here I will present an exploratory effort of using deep convolutional neural networks (CNN) to forecast the occurrence of haze or severe particulate pollution in Southeast Asia. Deep learning directly connects raw data as input with the output through deep artificial neural networks, providing an “end-to-end” solution. This could actually benefit the efforts such as forecasting poorly-known low-probability extreme events by avoiding human abstraction that would likely mislead the design as well as training of the model. With nearly one million of neurons and 38 million parameters, the CNN developed for the effort, as called HazeNet, has achieved impressive performance, scored almost perfect accuracy in training while a 95.1% accuracy in validation. The details of CNN design, datasets, hyperparameter tuning, alongside attempt to advance science understanding using the validation results will be discussed.