Mangrove mapping using machine learning techniques on satellite imageries

Abstract

Conservation and management are ponderous to monitor; the mangrove ecosystems are evident to be ecologically and economically significant to anthropomorphic populations. In spite of being viable and crucial biome services, mangroves frequently face threats from environmental disaster. The extraction for details to sustain and inspect the mangroves by adopting Gis and Hyperspectral data via Machine Learning. Previous research works have various dominant revelations in this particular domain and exhibit precision of 95% by this modus. The spectral profiles for 8 mangrove species were collected from various literatures. In the present study, Hyperion data of Pichavaram mangrove forest area is evaluated. The pre- processing of hyperspectral data is the prime emphasis, 242 bands were identified with bad pixels and fixed by replacing them with mean value of neighbouring pixels, where 179 bands are selected by MAD (Mean Median Absolute deviation) method. Endmember extraction was performed by deploying an hourglass workflow. MNF and PPI images were plotted in nD Visualiser and 48 endmembers were extracted. As they match the endmember with the ground truth data, distinct classifications are performed to map mangroves. By using SA classification, the abstraction of the mean spectra for definite mangrove pixels is obtained. Other unsupervised algorithms such as ISO clustering, K means Clustering are performed to map the study area. The findings of the study indicate the efficiency in pre-processing of hyperspectral data and absolute mappings of mangrove vegetations.