Interannual variations in marine phytoplankton impact the variability of available living marine resources, as well as the carbon cycle processes. Here we use a novel suite of retrospective, initialized ensemble forecasts with an Earth system model (ESM) to predict the evolution of surface chlorophyll anomalies. We demonstrate that the forecast system skillfully predicts marine surface chlorophyll variations 1-3 year in advance from low latitude to mid latitude, showing a substantial improvement over a simple persistence forecast and uninitialized forecast. We analyze how phytoplankton community composition, chlorophyll synthesis and growth rate contribute to its predictability. The potential predictability in chlorophyll at low-middle latitude derives mainly from the predictability in the nutrient limitation and at high latitude stems from temperature and light. We further analysis the mechanisms over five regions with low potential predictability of marine surface chlorophyll, indicating that the main driver of the potential predictability in chlorophyll is different in different regions.