Abstract: Mangrove forests, as vital blue carbon ecosystems, play a crucial role in global carbon cycling and climate change mitigation. Chonburi Province, Thailand, a highly industrialized and urbanized coastal region, has undergone significant mangrove degradation and partial recovery over the past three decades. Yet, long-term assessments of carbon storage dynamics remain limited. This study developed a comprehensive ground-UAV-satellite monitoring framework by integrating Sentinel-1/2 imagery, UAV-LiDAR, UAV oblique photography, and field plot surveys. Stand parameters were derived through individual tree segmentation and allometric equations, while mangrove distribution and carbon stocks were estimated using a Random Forest model. The proposed approach achieved high accuracy, with an overall classification accuracy of 92.37% and a Kappa coefficient of 0.89. Biomass inversion performed robustly (R²=0.87, RMSE=12.46 Mg/ha), enabling reliable characterization of mangrove structure and biomass. Results showed that aboveground biomass and carbon stock declined from 163.97×10³ Mg and 155.77 kt in 1996 to 129.49×10³ Mg and 123.01 kt in 2015 (a 21% reduction), before recovering to 145.24×10³ Mg and 137.98 kt by 2023, reaching 85%–88% of 1996 levels. Carbon density per unit area remained stable (92.78–98.32 t/ha), indicating that changes in total carbon storage were driven mainly by mangrove area dynamics rather than density shifts. This study demonstrates the effectiveness of multi-source remote sensing and AI-based methods for quantifying spatiotemporal changes in mangrove carbon sinks. The findings provide a robust scientific basis for carbon accounting and offer valuable support for mangrove conservation and restoration in Thailand and similar coastal regions.