Classification Tree Analysis (Gini-Index) Smoke Detection using Himawari_8 Satellite Data Over Sumatera-Borneo Maritime Continent Sout East Asia

Abstract

Classification tree analysis (CTA) automatic smoke detection was proposed using Himawari_8 Satellite data over Sumatera and Borneo Island Maritime Continent. Day Natural color and aerosol RGB composite were used to make Region of Interest (ROI) sampling of Cumulonimbus (Cb) top, low-mid cloud, smoke, bare soil, cirrus cloud, vegetation, and water. CTA - Gini index supervised classification being constructed with two different band collection as input. The result shows that CTA model 2, using 21 bands collection as input, has better overall accuracy value (about 0.75). Then the CTA model 1 only has an overall accuracy of about 0.63. The future research is still open in comparing the different impurity methods of CTA model, ensemble model of smoke detection using several CTA output models, and also a diurnal or seasonal variation of smoke using CTA model detection.