DigitalInterpretabilityofAnnualTile-basedMosaicof Landsat-8OLI for Time-series Land Cover Analysis in the Central Part of Sumatra

Abstract

This paper presented a digital interpretability of the annual tile-based mosaic (TBM) images for the operational purposes of time-series land cover analysis. The primary data used were the TBM images of Landsat-8 OLI of the central part of Sumatra, acquired from January 2015 to June 2017. The method used was comparing the overall accuracies of the results of the TBM images of land cover classification that using the master training samples of 2016 data with that using the training sample from each year of the three-years of data. The classifications were performed using four groups of spectral bands, namely Band 6-5-4-3-2, Band 6-5-4, Band 6-5-2, and Band 5-4. In order to improve the overall accuracies (OA), the classification results were afterward reclassified using fewer class number, based on Jefferies Matusita (JM) distance approach. The digital interpretability of the images could be deliberated through the average of overall accuracy (AOA) scores, which is Good with a score of > 80%, Fair between 70.0% - 79.9% and Poor if < 70%. The results showed that the use of the group of the Bands 6-5-4-3-2 performed at Good overall accuracy, consistency level with an AOA score of 86% for six object classes. Whereas the classifications using the groups of the Bands 6-5-4-3-2, Bands 6-5-4, and Bands 6-5 indicated Good accuracy, the consistency level for four object classes, with AOA scores of 89%, 82%, and 81%, respectively. It means that the annual mosaic image could be accepted through the digital interpretability of the land cover classification with AOA > 80% for six and four object classes. To support operational requirements, the use of group Bands 6-5 could also be recommended as the most efficient group of bands selected for land cover analysis with four object classes.