Multitemporal Landsat Imagery with Optimum Band Ratio Techniques for Deciduous Forest Classification

Abstract

The objective of the study is to evaluate optimum band ratio combinations data set derived from monthly Landsat 8 imageries for forest type classification around the Sirikit dam reservoir using supervised classification with Maximum Likelihood Classifier (MLC). In this study, imageries data acquired from January 2014 to November 2017 were used to create the monthly band ratio data set of Normalized Difference Vegetation Indices (NDVI), Normalized Difference Moisture Indices (NDMI), and Normalized Burn Ratios (NBR) and used to create the monthly multispectral (MS) data set represented as a case of without applying band ratio techniques. In classifying deciduous forest type, four data sets were used to classify two classes of deciduous forests, namely mixed deciduous forest and dry deciduous dipterocarp forest. After the accuracy assessment, the result showed that the overall accuracy and kappa coefficient of all data sets were between 78.33% – 86.21% and between 44.32% – 62.83%, respectively. Herein, the monthly NDVI multitemporal data set provided the highest overall accuracy and kappa coefficient which were better than the monthly MS multitemporal data set about 4% and 8%, respectively. In conclusion, applying monthly multitemporal data of Landsat 8 with band ratio technique, especially NDVI, can increase the accuracy of deciduous forest type classification.