Mangrove ecosystems are vital for coastal resilience, biodiversity, and climate regulation. This study assessed the spatial and temporal dynamics of mangrove cover in Jardines de la Reina National Park (JRNP), Cuba, between 2014 and 2024, using Landsat-8 imagery and five machine learning classifiers. Random Forest (RF) achieved the highest accuracy (97.60 %), with rigorous uncertainty propagation via Monte Carlo simulation, setting a new benchmark for mangrove mapping in the data-poor insular Caribbean. This method was selected for generating annual maps. Results revealed a loss of over 1500 ha of mangrove forest-an 18.65 % reduction-primarily in the western sector, especially Breton and Alcatraz keys. NDVI trend analysis confirmed significant degradation in these areas, while central keys remained more stable. Environmental factor analysis identified mean sea level (MSL) as the dominant driver of mangrove loss, followed by annual precipitation. Limited freshwater and sediment input, exacerbated by damming and droughts, likely impaired mangrove resilience. Patchy dieback patterns were observed, with localized mortality within otherwise healthy stands. Herbivory by hutia (Capromys pilorides) may contribute to stress, but recent data are lacking. Although JRNP is a protected area, the dominance of external environmental drivers-particularly sea-level rise and reduced precipitation-poses challenges that may exceed current local conservation management capabilities. The study highlights the need for integrated field and remote sensing approaches to monitor ecosystem health. Future research should focus on sediment accretion, primary productivity, herbivory impacts, and hydrological connectivity. This framework offers a model for holistic mangrove in marine protected areas across the Caribbean and supports adaptive management strategies to address the challenges posed by climate change.