Development of modelling approach to estimate mangrove carbon stock in Southern Thailand

Abstract

Blue carbon ecosystems, particularly mangrove forests, are vital for mitigating carbon emissions and strengthening institutional responses to climate change. Estimating mangrove carbon stocks is essential to understanding their contribution to climate change mitigation. However, limited advancement has been made in developing techniques to overcome the challenges of conventional field-based methods in regions, such as in Thailand. This study aimed at developing a carbon estimation approach using machine-learning-based and multi-sensor remote sensing in the Banlaem mangrove forest, Nakhon Si Thammarat province, Thailand. We employed active (Sentinel-1, ALOS2/PALSAR2) and passive (UAV) remote sensing data to generate aboveground biomass (AGB) models using machine learning models optimized using a genetic algorithm (GA), additionally incorporating mangrove type classification. The findings revealed that the SVR model exhibited superior performance in estimating mangrove AGB by producing an R2 of 0.92 and an RMSE of 18.24 ton∙ha−1. Notably, we found that incorporating mangrove type through the classification analysis into the modelling process substantially reduced the RMSE to approximately 10.50 to 15.87 ton∙ha−1, representing a significant advancement in Southeast Asia. SVR and XGBoost proved to be the most effective models for developing AGB estimates by mangrove type. This research introduces the integration of synthetic aperture radar (SAR) and UAV for mangrove carbon estimation in Thailand, along with mangrove type classification in the modelling process. It presents an effective approach for mangrove carbon stock assessment that supports national and global climate goals.