Estimation of the tree aboveground biomass and carbon stock of the Bai Tu Long National Park forest ecosystem from Sentinel-2 images via regression models

Forest ecosystems are vital for ensuring ecological balance and provide crucial services such as climate regulation and biodiversity conservation. Accurate biomass and carbon storage estimation is key for resource management and climate change mitigation. This study focuses on Bai Tu Long National Park, Vietnam, a biodiverse region, to estimate forest biomass via Sentinel-2 imagery and regression models. Sentinel-2 data from February 2024, alongside field data from 20 sample plots collected in March 2024, were used to develop and test linear regression models (log-log, log-lin, lin-log, lin-lin) for aboveground biomass (AGB) estimation. Model performance was rigorously assessed through Leave-One-Out Cross-Validation (LOOCV), utilizing the coefficient of determination (R2) and root mean square error (RMSE). The log-lin model yielded the best results (R² = 0.76, RMSE = 20.52 Mg/ha). The resulting biomass distribution map revealed that the biomass was mainly concentrated in Ba Mun Island and the Sau Island cluster, with values ranging from 25.32 to 110.35 Mg/ha. The average carbon stock within the forest ecosystem was estimated to be between 11.90 and 51.87 Mg C/ha. This research presents an effective method for estimating forest biomass via remote sensing and regression models, enhancing our understanding of forest ecosystems in Bai Tu Long National Park. These findings can inform sustainable strategies for managing forest resources and assessing carbon emission reduction potential. Future research should focus on improving biomass estimation accuracy by increasing field data collection, utilizing multitemporal remote sensing data, and formulating specific regression equations for each forest type.