Spatial Clustering of Wildfire Hotspot Intensity in West Kalimantan, Indonesia: A K-Means Approach Using Google Earth Engine Data

Abstract

Recurrent vegetation and peatland fires in Kalimantan, Indonesia, generate transboundary haze that affects public health and regional air quality every dry season. This study presents a reproducible, remote-sensing-based workflow for characterising the spatial pattern of fire activity across the twelve regencies (kabupaten/kota) of West Kalimantan Province during September 2026, a month of intensified burning under El Niño-influenced dry conditions. Satellite-derived active-fire hotspot counts were aggregated to administrative boundaries using Google Earth Engine, and regency centroids were extracted from the associated polygon geometries. A K-means clustering model was fitted to a three-dimensional feature space combining geographic location (latitude, longitude) and log-transformed hotspot intensity, with the optimal number of clusters selected through the elbow method and silhouette analysis. The analysis identified two statistically and spatially coherent clusters. A high-intensity cluster comprising five regencies (Ketapang, Pontianak, Melawi, Sintang and Kapuas Hulu) accounted for 687 of 712 recorded hotspots (96.5%), with Ketapang Regency alone contributing 483 hotspots (67.8% of the provincial total). A low-intensity cluster of six regencies in the northern and central parts of the province accounted for the remaining 3.5%. These results delineate a concentrated fire-risk corridor running through the southern peatland- and plantation-dominated regencies of West Kalimantan and support the prioritisation of monitoring and mitigation resources toward a small number of administrative units rather than uniform province-wide deployment. The proposed workflow is lightweight, fully reproducible from open satellite data, and can be re-run monthly to track the evolution of provincial fire-risk geography.