Satellite Active Fire Clustering in Northern Thailand

Abstract

Open burning is a primary cause of the haze problem in Thailand, particularly in the northern region, posing significant health risks. These fires are largely anthropogenic, originating from both agricultural and forested areas. Despite policies prohibiting burning, illegal burning continues to occur, necessitating ongoing monitoring and mitigation efforts. Satellites provide a valuable data source, enabling the tracking of fires even in remote areas. However, the point-based nature of satellite hotspot data, rather than a continuous surface, limits our ability to estimate both the total burned area and the complete boundary of fire spread. This research clustered satellite-detected active hot spots and created polygons around the clusters to represent burning areas. Characteristics of these clusters between 2020 and 2024 were then analysed. Using an unsupervised machine learning technique known as Density-Based Spatial Clustering of Applications with Noise (DBSCAN), adjacent hot spots were grouped. Concave hulls and buffers were then employed to create polygons around these clusters representing fire perimeters for further analysis. Analysis of the northern region revealed that most fire perimeters in one day has a yearly average area between 1.664 and 3.223 square kilometers. The fire clusters expanded continuously over periods of less than 10 days, and spread covered up to less than 10 square kilometers. The majority occur in protected forest areas and national parks, while burning in fields and rice paddies is rare. The characteristics of these hot spot clusters can be used to develop a forecasting system for predicting the spread of wildfires.