Mangrove classification using airborne hyperspectral AVIRIS-NG and comparing with other spaceborne hyperspectral and multispectral data

Abstract

Application of remote sensing makes the assessment and monitoring of mangroves both time and cost-effective. In this study, the capacity of AVIRIS-NG data in discriminating different mangrove species of Lothian Island of Indian Sundarbans has been evaluated and compared with hyperspectral (Hyperion) and multispectral dataset (Landsat 8 OLI and Sentinel-2). Spectral signatures of mangrove species were retrieved, and spectral libraries were created. With the corrected images and spectral libraries, mangroves were classified using appropriate classification techniques. For multispectral datasets (Landsat 8 OLI and Sentinel-2) and hyperspectral coarser-resolution Hyperion datasets, K-means classification followed by knowledge-based classification was adopted. For fine resolution hyperspectral AVIRIS-NG dataset, classification was accomplished using Support Vector Machine (SVM). The overall accuracy for the classification is significantly high in case of AVIRIS-NG data (87.61%) compared to the Landsat 8 OLI (76.42%), Sentinel-2 (79.81%), and Hyperion data (81.98%). The results showed that AVIRIS-NG hyperspectral dataset has the potential to classify not only the genus level but also species-level with satisfactory accuracy in a complex mangrove forest.