Against the background of the normalized occurrence of green tides in the Yellow Sea, developing seasonal-scale forecasts to extend the forecast lead time and provide a sufficient early warning window for disaster response has become an urgent need for disaster prevention and mitigation. Identifying the sources of uncertainty and their propagation chains can guide error suppression and model optimization. Based on the Yellow Sea Green Tide Ecological Model (YSGTEM), this study introduces a stochastic perturbed parameterization scheme to construct the Yellow Sea Green Tide Ensemble Prediction System (YSGTEPS). Taking the 2016 green tide event as an example, seasonal-scale ensemble forecast experiments and uncertainty analyses were conducted. Evaluation using MODIS remote sensing data shows that both YSGTEM and YSGTEPS can well simulate the drift, growth, and decay characteristics of the 2016 green tide. YSGTEPS effectively reduces the simulation errors of YSGTEM and more accurately depicts the drift path and biomass dynamic distribution characteristics of the green tide over simulation periods longer than 15 days, with the path error reduced by 32 % on average and the IOU increased by 11.09 percentage points on average. The uncertainty of dynamic parameters in YSGTEM is the dominant factor causing differences in biomass and distribution simulations, accounting for approximately 80 % of the total. The uncertainty in dynamic processes propagates to simulation results through mismatches in the spatiotemporal matching of environmental factors. Improving the simulation accuracy of dynamic processes should be prioritized to enhance seasonal-scale simulation performance.