Rating Wildfire Danger by Big Weather Data and Machine Learning: A Case Study in Eastern China

Abstract

Wildfire, as a natural phenomenon, shapes global ecosystems and threatens human communities. The meteorological conditions of the forest environment is one of the critical factors, which exists obvious effect on the risk of wildfire for a given landscape and fuel type. It is high challenge to predict wildland fire risks owing to the huge amount of meteorological paramenters with volume,variety,value and velocity. Driven by the emerging Artificial Intelligence and Big Data analytics, this work proposes a machine learning model (Fuzzy C-Means algorithm) to assess the probability of wildland fire. By training the historical weather and fire data in Eastern China (Jiangsu Province), the complex relationships between the weather parameter and the rating of wildland fire danger are quantified. The Fire Weather Index system are applied to evaluate the accuracy of predictions and compare them with the standards in the literature. This system further reveals that the spring season is the peak period for wildland fires in Eastern China. Such a prediction is also in accordance with the practice of wildfire occurrence that, in practice, can provide early warning to local residences and forest services.