Random Forest Classification for Mangrove Land Cover Mapping Using Landsat 5 TM and Alos Palsar Imageries

The objective of this research was to evaluate the accuracy of random forest classification rule using object based image analysis (OBIA) application (eCognition Developer) and the results were compared with common pixel-based classification algorithm (maximum likelihood/ML) for mangrove land cover mapping in Kembung River, Bengkalis Island, Indonesia. Seven data input model derived from Landsat 5TM bands, ALOS PALSAR FBD, and spectral transformations (NDVI, NDWI, NDBI) were examined by both classifiers. Feature objects statistical parameters were selected and implemented on random forest classifier. Overall accuracy (OA) as well as user and producer accuracies and Kappa statistic were used to compare classification results. Our results showed that the more data model used produced higher overall accuracy and kappa statistics for RF classifier. For each data input model, random forest classifier has higher overall accuracy than maximum likelihood. The best mangrove discrimination in RF classifier was achieved when the combination of Landsat 5 TM, SAR, and spectral transformation were used, while in ML classifier, the best mangrove discrimination was achieved when the combination of Landsat 5 TM and ALOS PALSAR was used. The overall accuracy achieved by RF classifier was 81.1% and 0.76 for Kappa statistic. Meanwhile, for ML classifier, the overall accuracy achieved was 77.7% and 0.71 for Kappa statistic.