Integrating optical and SAR data for improved mangrove mapping: Performance analysis of OSCMI versus random forest classifier

Mangrove Forest (MF) extents and distributions are fundamental for conservation and restoration efforts. According to previous studies, Landsat and Sentinel-2 imageries are widely used for mangrove area mapping. There are also several vegetation indices from optical data for mangroves used to discriminate mangrove from other vegetation. Still, their performance is not satisfactory, especially due to their presence in the intertidal region. To address this limitation, Sentinel-1A SAR images are integrated with optical datasets for mangrove extent mapping. In this study, we have compared mangrove area obtained from two methods, first one is supervised classification using random forest classifier integrating vegetation indices with Sentinel-2 optical datasets. Another approach is threshold segmentation using combined optical and SAR images, as well as the Combined Mangrove Index (CMCI) with VV polarization. The result shows OSCMI based classification yield best results and outperforms random forest classification by optical data. Mangrove area estimated for the study area by OSCMI classification was 2027 ha with the overall accuracy of 84.7 % and kappa coefficient of 0.66. This shows integrating SAR data for mangrove area extraction improves accuracy.