Mangrove gross primary productivity (GPP) plays a critical role in coastal carbon cycling, yet its seasonal dynamics and spatial variability remain insufficiently characterized. Here, we integrated eight months of half-hourly eddy covariance measurements with 10-m Sentinel-2 imagery to quantify seasonal GPP, calibrate light use efficiency, and examine the spatial drivers of productivity in the Qinglangang mangrove forest, Hainan Island, China. Tower-derived GPP averaged 6.67 ± 1.81 g C m⁻² d⁻¹, peaking at 8.57 g C m⁻² d⁻¹ in July. Bootstrap-calibrated maximum light use efficiency (εmax) exhibited seasonal variation, with values of 1.491, 1.965, and 1.897 g C MJ⁻¹ APAR in spring, summer, and autumn, respectively. Incorporating sea surface temperature and salinity constraints did not improve half-hourly GPP predictions, indicating limited predictive value of monthly oceanic climatologies at sub-seasonal timescales. The seasonally calibrated model yielded landscape-scale GPP estimates of 3.27 ± 2.08, 7.14 ± 4.06, and 3.16 ± 1.78 g C m⁻² d⁻¹ across the three seasons, respectively. Spatial attribution using XGBoost and SHAP identified leaf area index as the dominant predictor, accounting for 69.9 % of total feature importance, whereas meteorological variables alone explained only 19–23 % of spatial GPP variability. Random forest analysis further indicated a greater contribution of distance from the coastline than elevation. These findings demonstrate the seasonal dependence of mangrove light use efficiency and the importance of canopy structure in shaping landscape-scale productivity, providing a basis for integrating flux observations with high-resolution satellite estimates of mangrove GPP.