With emergences of predictive skills in climate simulations, we currently stand at the horizon of being able to forecast marine ecosystems. However, the predicting high trophic levels from multiple biophysical bottom-up drivers remains understudied. Here, we evaluate the potential predictability of bottom-up biophysical drivers of fish in decadal Earth system model outcomes. Pelagic and bottom temperatures show approximately 6- and 10-year predictability compared to mesozooplankton and particulate organic carbon flux, about 1 and 2 years globally. We further examine fish biomass predictability using a spatially explicit mechanistic model for three fish functional types. Fish biomass demonstrates skillful predictability within 2 years globally, extended regionally for more years. In select marine ecosystems, forage fishes (anomaly correlation coefficient: r ~ 0.8), large pelagics (r ~ 0.9), and demersals (r ~ 0.8) are predictable in multiyear. These findings expand on prior research, shedding light on developing operational forecasts of living marine resources.