An object-oriented classification method for mapping mangroves in Guinea, West Africa, using multipolarized ALOS PALSAR L-band data

摘要

The principal objective of this study was to determine the accuracy of an object-based image analysis (OBIA) approach in classifying mangroves from spaceborne synthetic aperture radar (SAR) data, specifically Advanced Land Observation Satellite (ALOS), phased array L-band synthetic aperture radar (PALSAR), and single-polarized (HH) and dual-polarized (HH + HV) L-bands. The accuracy of the object parameters was examined to determine the optimal colour and shape ratios for the hierarchical classification. At the first level of classification (mangroves from non-mangroves), the results indicate that it is possible to accurately separate mangrove areas from saltpan and water/shallow zones using both sets of SAR images for the Mabala and Yelitono islands of southern Guinea. The final accuracies, based on the most optimal object parameters, were 91.1% and 92.3% for the single- and dual-polarized data, respectively. At the second level of classification, separation among the three mangrove classes identified was most accurate when using the dual-polarized data, at an overall accuracy of only 63.4%. The three mangrove classes identified included tall red mangrove (Rhizophora racemosa), dwarf red mangrove (R. mangle and R. harisonii), and black mangrove (Avicennia ger-minans). Using the optimal combination of parameters, the extent to which a filter could be used to improve the accuracy was examined. At this level, it was determined that the dual-polarized data, filtered with a 3 × 3 Lee speckle filter and a segmentation scale of 5, resulted in an overall accuracy of 64.9%. Consequently, it is recommended that for persistently cloud-covered regions, such as Guinea, ALOS PALSAR data using an OBIA could be useful as a quick method for mapping and monitoring mangroves.