Initial Results on Landuse/Landcover Classification Using Pixel-Based Random Forest Algorithm on Sentinel-2 Imagery over Enrekang Region

Abstract

Abstract Land use classification is the basis for making further policy in many fields including agriculture. Effective methods in landuse/landcover (LULC) classification are essential for later application in policy making. The development of remote sensing technology has been increasing rapidly. The use of Earth Observing (EO) Sentinel-2 imagery can greatly help LULC mapping over large area. As the basic input on the assessment of land availability and suitability, it is important to perform LULC in such way that it is objective, replicable, and accurate. This study aim to performed state-of-the-art Random Forest algorithm on multitemporal Sentinle-2 imagery on LULC extraction over Enrekang Region. With its 10 m spatial resolution as well as multitemporal information, acquired on December as a representation of the rainy season and in July as a representation of the dry season, it is expected to produce a more optimal LULC maps. Confusion matrix were then performed using visually interpreted Pan-sharpened and orthorectified SPOT-6/7 imagery to calculate the accuracy. The output of LULC classification based were expected to reach 95% overall accuracy.

Access to Document

Other files and links

• Link to publication in Scopus
• Open Access Version Available

Fingerprint