Implementation dbscan algorithm to clustering satellite surface temperature data in indonesia

Abstract

Forest and land fires are national and international problems. The frequency of fires in one of Indonesia's provinces, Riau, is a significant problem. Knowing where to repair the burn is essential to prevent more massive fires. Fires occur because of a fire triangle, namely fuel, oxygen, and heat. The third factor can be seen through remote sensing. Using the Landsat-8 satellite, named the Enhanced Vegetation Index (EVI) variable, Normalized Burn Area (NBR), Normal Difference Humidity Index (NDMI), Normal Difference Difference Vegetation Index (NDVI), Soil Adapted Vegetation Index (SAVI), and Soil Surface Temperature (LST). DBSCAN, as a grouping algorithm that can group the data into several groups based on data density. This is used because of the density of existing fire data, according to the character of this algorithm. The selected cluster is the best cluster that uses Silhouette Coefficients, eps, and minutes value extracted from each variable, so there is no noise in the resulting cluster. The result is more than 0, and the highest is the best cluster result. There are 5 clusters formed by clustering, each of which has its members. This cluster is formed enough to represent the real conditions, cluster which has a high LST value or has an NBR value. A high LST value indicates an increase in the area's temperature; a high NBR value indicates a fire has occurred in the area. The combination of LST and NBR values indicates the area has experienced forest and land fires. This study shows that DBSCAN clustered fire and surface temperature data following data from the Central Statistics Agency of Riau Province. Proven DBSCAN can cluster satellite imagery data in Riau province into several clusters that have a high incidence of land fires.