A Rule-Based Classification Method for Mapping Saltmarsh Land-Cover in South-Eastern Bangladesh from Landsat-8 OLI

Abstract

Wetland vegetation classification often treated the saltmarsh as a single type of land-cover (LCT). Mapping the dynamic and spatially complex coastal zones using optical remote sensing is still challenging. This study firstly analyzed the spectral properties of target objects generated by Landsat 8 (OLI), formulated new spectral indices and then proposes a rule-based approach to mapping five vegetated (saltmarsh, seagrass, mangrove, non-mangrove forest, and agricultural land) and three non-vegetated (wet sand, saltpan, and built-up areas) LCT in the study area, that is, large coasts located in the south-eastern coasts of Bangladesh. The thresholds of spectral indices were selected from the newly introduced spectral indices over the method development site (Bakkhali estuary). The rule-based LCT classification process followed a set of cascade rules of image thresholding and masking, based on a hierarchical tree in order to generate detailed thematic maps of saltmarsh land-cover. Overall accuracy (OA) and Kappa coefficient (K) of rule-based approach were 84.6% and 0.821, respectively. The reliability and robustness of the approach was tested over two independent external validation test sites: Karnaphuli river estuary and Teknaf peninsula and consistent accuracy results achieved: OA = 81.7% (K = 0.787) and OA = 84.6% (K = 0.821) respectively.