<![CDATA[Land fires are an environmental issue that requires data-driven monitoring to better understand their occurrence patterns at an operational level. Although satellite hotspot data provide rich spatial and temporal information, their analysis still faces challenges such as uneven data distribution, variations in density, and the presence of noise that can affect the identification of fire episodes. This study aims to develop a spatiotemporal clustering framework to extract wildfire hotspot episodes in Sulawesi using NASA’s Fire Information for Resource Management System (FIRMS) data from 2023. Each hotspot is represented as (lat, lon, t), with spatial proximity calculated using the Haversine distance and combined with temporal proximity in a spatiotemporal distance matrix based on precomputed distances. A grid search was conducted across 912 configurations and compared against three density-based clustering methods: HDBSCAN, DBSCAN, and OPTICS. Evaluation was performed using internal metrics—namely, the Silhouette coefficient and Density-Based Clustering Validation (DBCV)—while also considering noise proportion and coverage to ensure the resulting clusters remain representative. The results show that HDBSCAN provides the best balance by producing 26 clusters, 21.78% noise, 78.22% coverage, and competitive metric values, thus making it the primary method. Visualization of the results shows that episode activity increases from mid- to late-year and is concentrated in specific zones, while Fire Radiative Power (FRP) characterization reveals variations in intensity between episodes. This study contributes to providing a more transparent and representative hotspot analysis framework to support pattern analysis, the determination of priority monitoring areas, and data-driven decision-making in the fields of environmental informatics and geospatial analytics.]]>
