Land Cover Classification of Anzali Wetland Using Fusion of Sentinel 1 and ALOS/PALSAR 2 Images

Anzali Wetland in Iran as one of the most valuable wetlands registered in the Ramsar Convention is being destroyed by environmental factors and human activities. In the last two decades, among various satellite images, radar images have played a special role in wetland monitoring. Radar is an all-weather sensor and it is sensitive to surface roughness and moisture, they serve as a valuable source for quick and accurate monitoring of wetlands. However, similarities in backscattering coefficients of different wetland classes and relatively difficult processing – in comparison to optical images- are the most important factors that limit their application. In this study, the capabilities of SAR images in the classification of Anzali wetland and the three main land use classes around the wetland (i.e. agricultural lands, reeds, and built-up areas) were evaluated. Two radar images; Advanced Land Observing Satellite/Phased Array L-band Synthetic Aperture Radar (ALOS/PALSAR) and Sentinel 1 captured in 2018 were used. The texture parameters of the two images have been extracted. The images and their extracted texture layers have been fused by the feature-level method and further classified by the random forest method. The overall accuracy of feature-level fusion is equal to 75% and the kappa coefficient is equal to 0.62. The evaluation results related to producer and user accuracy are 100% and 83.33%, respectively, show the high capability of radar images in the classification and detection of wetlands. However, some errors have been observed in the separation of agricultural lands, reeds, and built-up areas.