FIREfly project: forest fire monitoring and prediction using machine learning

Forest fires and extreme wildfire events pose a major threat to ecosystems worldwide. This paper implements various machine learning algorithms for the prediction of forest fires in Northern Thailand, a region which is severely impacted by fire events and the resulting pollution. Using publicly available satellite data of fires and weather information, two prediction models, namely the Random Forest and Support Vector Machines, were developed and tested for their accuracy in forecasting forest fire occurrences. Initial results indicate that both models have an accuracy of approximately 60% in predicting forest fires. The real-time prediction data based on current weather conditions is further displayed in a dashboard. The online dashboard has been integrated with Project FIREfly which is a collaboration with Chiang Mai University and the University of Glasgow to visualize real-time data of forest fires. Through the integration of the predictive models, the online dashboard is able to show the probability of forest fires which improves situational awareness for emergency response services and enables them to take proactive measures in managing forest fires.