<![CDATA[Abstract. Wildfire was a serious environmental problem in Indonesia, causing ecosystem damage, biodiversity loss, health problems from haze, reduced air quality, and economic losses. Wildfire monitoring widely utilizes satellite-based hotspot data from global monitoring systems. The objective of study is to find some region clusters in Indonesia based on wildfire vulnerability levels using hotspot data including three characteristics which are confidence, brightness temperature, and Fire Radiative Power (FRP). The Improved Clustering Large Applications based on Randomized Search (I-CLARANS) method was demonstrated to fo find the region clusters. The advantage of method was on initial medoid selection quality improvements, clustering efficiency on large datasets, and robustness againts outliers. The searching of cluster optimal number was tried by 2, 3, 4, 5 and 6 clusters by using silhouette coefficient criteria. The results show that 2 clusters was an optimal number of cluster with a maximum silhouette coefficient which is 0.5249, indicating good cluster structure. The first cluster consists of 2,411 hotspots with relatively low brightness, confidence, and FRP values. This cluster was labeled as low to moderate wildfire vulnerability. The second cluster consists of 742 hotspots with higher values, labeled as areas with high wildfire vulnerability. Abstrak. Kebakaran hutan dan lahan (karhutla) merupakan permasalahan lingkungan serius di Indonesia yang berdampak pada kerusakan ekosistem, hilangnya keanekaragaman hayati, gangguan kesehatan akibat kabut asap, penurunan kualitas udara, dan kerugian ekonomi. Pemantauan karhutla banyak memanfaatkan data titik panas (hotspot) berbasis satelit dari sistem pemantauan global. Penelitian ini bertujuan mengelompokkan wilayah di Indonesia berdasarkan tingkat kerentanan karhutla menggunakan data hotspot dengan variabel confidence, brightness temperature, dan Fire Radiative Power (FRP). Metode yang digunakan adalah Improved Clustering Large Applications based on Randomized Search (I-CLARANS), pengembangan dari CLARANS yang mampu meningkatkan kualitas pemilihan medoid awal, efisiensi pengelompokan pada dataset besar, serta ketahanan terhadap outlier. Jumlah klaster optimal ditentukan dengan menguji nilai k = 2 hingga 6 berdasarkan silhouette coefficient tertinggi. Hasil penelitian menunjukkan dua klaster optimal dengan silhouette coefficient 0,5249, mengindikasikan struktur klaster yang baik. Klaster pertama terdiri dari 2.411 titik panas dengan nilai brightness, confidence, dan FRP relatif rendah, dikategorikan sebagai wilayah berkerentanan karhutla rendah hingga sedang. Klaster kedua terdiri dari 742 titik panas dengan nilai yang lebih tinggi, dikategorikan sebagai wilayah berkerentanan karhutla tinggi.]]>
