Abstract

Simon Njeudeng Tenku, Xiangming Xiao, Jinwei Dong

Up-to-date information on forest cover is important for every country. Radar remote sensing technology has considerable potential to acquire cloud-free images in moist tropical forests due to frequent cloud cover. The objective of this study was to investigate the potential of Phased Array Type L-band SAR (PALSAR) imagery for forest cover mapping. 50-m orthorectified mosaic image of 2009 from L-band PALSAR was used for the study. Geo-referenced field photos were taken as ground truth data points and were organized to obtain regions of interest. K-means and decision tree classifications were carried out to discriminate forests from other land covers. Decision tree algorithm was more reliable than K-means and showed an overall accuracy assessment of 91.7 % and Kappa coefficient of 0.86. Decision tree estimated 24.4 million hectares of forests, that is, 12.5 % higher than 21 million hectares reported by ITTO in 2011. Lband PALSAR modelled backscatter output for the difference phase showed a grey tone pattern for forest cover similar to that of decision tree than to K-means; implying a spatial reliability for the modelled phase. Improved PALSAR-based forest cover mapping necessitates data coverage in both rainy and dry seasons.