A multisensor satellite image classification for the detection of mangrove forests in Qeshm Island (Southern Iran)

Abstract

Mangrove forests in Iran are among the complex and productive ecosystems because these types of forests directly and indirectly play a significant role for humans and the environment. This study developed a parallel land cover classification method to identify mangrove forests (mangrove forests) in southern Iran using high-resolution (similar to 10 m) optical image of Sentinel-2 satellite and high-resolution (similar to 10 m) synthetic aperture radar image that presents ALOS-2 satellite (dipolar). Therefore, in this paper, ALOS-2 bipolar (VV, VH) was used for land cover classification and Sentinel-2 multispectral data as reference data. Generally, GLCM textures in different window sizes were applied to the SAR data, and then all of them were subjected to PCA transformation; finally, the first three components were used as input to the maximum likelihood classification (MLC) algorithm to classify the two mangrove classes. Other lands in addition, the backscatter image was also included separately in the MLC algorithm for land cover classification. The obtained statistical results showed that when the texture with different windows is placed in the input of the ML algorithm, it has a kappa coefficient value of 0.52, and when the input is a single backscattered image, it has a higher kappa coefficient value of about 0.83. In general, the results show that the map prepared by the only backscattered image performs better and similar to the optical image used. In addition, the accumulation of GLCM texture in the dimensions of the windows reduces the accuracy of the mangrove cover map, which as a result causes an exaggerated prominence in the land cover, especially the mangrove.