علوم زیستی دریا
Mojdeh Miraki; Hormoz Sohrabi; Sima Sadeghi; Parviz Fatehi; Markus Immitzer
Abstract
Advances in remote sensing enable fast mangrove mapping the less need for intensive fieldwork, complex and heavy processing, and skill-based classification techniques. In this research, mangrove forest mapping was performed using Sentinel-2 satellite images in Google Earth Engine (GEE) in Hormozgan province ...
Read More
Advances in remote sensing enable fast mangrove mapping the less need for intensive fieldwork, complex and heavy processing, and skill-based classification techniques. In this research, mangrove forest mapping was performed using Sentinel-2 satellite images in Google Earth Engine (GEE) in Hormozgan province in three ecosystems of Qeshm, Khamir, and Sirik. For this purpose, all steps of mapping these forests, including pre-processing and classification were performed in the GEE. The Modular Mangrove Recognition Index (MMRI) and classic spectral indices were also used to highlight the spectral differentiation of mangrove cover from the surroundings. To classify the image of the study area, three land cover classes were used: mangrove, non-mangrove, and sea (water). The classification was performed based on the random forest algorithm and accuracy assessment was evaluated in R software based on the K-fold validation method. The Qeshm site was demonstrated the highest accuracy among the three ecosystems with an overall accuracy of 98% and a kappa of 0.73. Khamir and Sirik sites were shown an overall accuracy of 97% and a kappa value of 0.71 and 0.70, respectively. The MMRI index was the most important variable in the RF classification in Qeshm and Khamir, while in Sirik, the SAVI index was the most important spectral index in mangrove map providing. The overall accuracy of over 95% at all three sites indicates that combining Sentinel-2 data using appropriate indices in the GEE is an effective approach to mangrove forest mapping