Capturing the dynamics of aboveground carbon stock in intertidal seagrass meadows using Sentinel-2 time-series imagery

One of the challenges associated with the monitoring of seagrass meadows is the seasonal variability in percent cover, which is closely linked to the aboveground biomass carbon stock (AGC). To gain a comprehensive understanding of seagrass dynamics, it is essential to obtain spatial and temporal information on seagrass AGC. The most effective approach for mapping the dynamics of seagrass AGC is remote sensing; however, limited robustness of the mapping model limits their applicability across different locations. To address this issue, we developed a robust model for mapping seagrass AGC, with the objective of capturing the dynamics of seagrass AGC in intertidal seagrass meadows. Using seagrass field data and assuming that pure seagrass and sand pixels have 100 % and 0 % seagrass cover, respectively, we trained stepwise, machine learning (random forest, support vector machine, and multivariate adaptive regression spline), and deep learning (dense neural network) regression models to convert Sentinel-2 reflectance into seagrass AGC. The accuracy of the models was evaluated at multiple sites with available field data, and the results demonstrated an RMSE ranging from 6.28 to 13.97 g C m⁻² and a correlation coefficient between 0.69 and 0.83. Overall, the SVM regression model exhibited the highest accuracy. The SVM model was subsequently applied to 13 seagrass sites across Indonesia over a 36-month period, revealing consistent and recurring monthly and bimonthly AGC patterns. The majority of seagrass meadows exhibited their highest AGC during the May–June period and their lowest during the September–October period. This study also represents the first time-series mapping of seagrass AGC in Indonesia on a monthly and bimonthly basis, marking a significant advancement in understanding seagrass's potential as a blue carbon sink. Additionally, to achieve more accurate assessments of seagrass changes, it is crucial to account for the monthly and seasonal dynamics in seagrass growth patterns.