Seagrass Percent Cover Mapping Around Teluk Terima Using Machine Learning for Blue Carbon Stock Estimation

Seagrass ecosystems play a crucial role in blue carbon sequestration and in supporting various coastal ecosystem services. However, spatial information on seagrass distribution and aboveground carbon (AGC) stocks remains limited in Indonesia, particularly in the Teluk Terima area. This study aims to map seagrass cover and estimate AGC using PlanetScope SuperDove imagery combined with a Random Forest (RF) machine learning approach. Field data were collected using a photo-transect method within 1 × 1 m2 grids to generate training and validation datasets. Pre-processing included sunglint correction, followed by pixel-based classification using the RF algorithm. The classification achieved an overall accuracy of 59.1 and a user’s accuracy of 59.4 for the seagrass class. Seagrass cover was found to be dominant in the Tanjung Kotal and Labuan Lalang areas, with density values ranging from 0 to 82.6. AGC was estimated using a linear regression model (AGC = 0.4386 × PC + 1.0413), where PC represents the seagrass cover percentage. The estimated AGC values ranged from 1.04 to 37.28 g C m-2 with an R2 of 0.4657 and an RMSE of 10.9 g C m-2 indicating a moderate relationship between modelled and reference AGC. These findings demonstrate that the RF algorithm, supported by representative field data, can effectively map seagrass spatial distribution and predict carbon stocks with acceptable accuracy. The results provide valuable insights for spatially explicit seagrass conservation and climate change mitigation planning in coastal regions. © ACRS 2025.All rights reserved.