Classification tree analysis (CTA) of smoke detection using Himawari_8 satellite data over Sumatera–Borneo Island, Indonesia

Abstract

This study proposed Classification Tree Analysis (CTA) for automatic smoke detection using Himawari_8 Satellite data over the Maritime Continent of Sumatera and Borneo Islands in Indonesia. Two timestamps of the Region of Interest (ROI) sampling, including Cumulonimbus (Cb) top, low-middle cloud, smoke, bare soil, cirrus cloud, vegetation, and water classes, were used as the input to determine the best CTA models. The CTA model classification was supervised using a collection of 21 single and transformation bands. The study also employed and compared two impurity measures, the Gini Index, and Entropy. The responses of the output of 4 CTA models (Entropy-09, Gini-09, Entropy-10, and Gini-10) were spatially, temporally, and statistically analysed. Furthermore, the CTA models were validated using METAR data (weather airport observation), with results showing that Entropy-10 have the highest Overall Accuracy value of 0.79, and lowest False Alarm Rate Value of 0.11. The computing time shows that Entropy-9 is the fastest with a mean of 19.8 s, followed by entropy-10 with 20.7 s. The accuracy assessment, spatial and temporal analyses, and computing process revealed that the Entropy-10 was the best model. The results of the CTA Entropy-10 are implemented over a small area, such as an airport to justify the work of weather observers and forecasters. This is often based on the objective satellite-based smoke detection product. Furthermore, they serve as information for aviation users in improving their situational awareness of adverse weather conditions related to safety.