Hybrid Method for Monitoring Sundarbans Mangrove Forest Using Satellite Images

A hybrid method for monitoring Sundarbans mangrove forest using satellite images with machine learning techniques is presented in this paper. The Random Forest algorithm, data scaling, cloud, and shadow masking are used to enhance the precision of land cover classification over the Sundarbans. The study utilizes Landsat 8 imagery to classify land into deep forest, light forest, barren land, and water bodies. Significant fluctuations in land cover types over the analyzed period, with notable variations in forest density, barren land, and water bodies are identified. Normalized Difference Vegetation Index and Normalized Difference Water Index are two popular methods for observing vegetation health and water bodies in remote sensing. But it fails to classify different objects. This paper focuses on solving this issue using machine learning technique by combining the two methods along with the other band's reflectance as a feature to improve the classification and show a great f1 and recall. This study uses a hybrid model to monitor land cover changes in the Sundarbans mangrove forest from 2013 to 2022, achieving an overall accuracy of 0.98. Significant findings include a deep forest cover increase from 52.82% to 55.735%, fluctuations in barren land from 2.04% to a peak of 8.38% in 2021, and a decrease in water bodies from 32.18% to 28.82%, indicating possible changes in river dynamics and hydrological conditions.