Quantifying Mangrove Dynamics using Sentinel-1 Multitemporal Data and Random Forest Algorithm

Mangroves are significant coastal ecosystems that are susceptible to changes in the environment. In this study, the distribution and size of mangrove expansions were predicted using multitemporal Sentinel-1 synthetic aperture radar (SAR) data and the random forest (RF) method. The data used are multitemporal Sentinel-1 SAR images of 2020, 2021, and 2022 on the coast of South Lampung Regency. We performed pre-processing on these data, including filtration and radiometric correction, to ensure good quality before use in the analysis. SAR features extracted from the multitemporal imagery provide valuable information for identifying and mapping mangroves. The results of Sentinel image extraction explain that vertical-vertical (VV) polarization in 2020 has the lowest value in July with a value of -19.62dB and the highest value in December with a value of -16.48dB; in 2021, the highest value is in November with a value of -16.37dB and the lowest in August with a value of -19.01dB. In 2022, the lowest value is in February with a value of -19.87dB and the highest is in May with a value of -16.44dB. Backscatter values in vertical-horizontal (VH) polarization have a relatively similar pattern. These backscatter values are input parameters for RF classification. The classification results are divided into Built-up Area, Road, Water body, sea, mangrove, vegetation, and agriculture classes. The highest accuracy evaluation was obtained in 2021 with a kappa value of 0.93 and an overall accuracy value of 93.3%. The usage of multitemporal Sentinel-1 SAR data and algorithms is significantly aided by this work.