Predictability of variations in the ocean and land carbon sinks in a multi-model framework

The strength of the ocean and land carbon sinks have a pronounced natural variability on inter-annual to decadal time scales. Together with anthropogenic carbon emissions, this natural variability co-determines the growth rate of atmospheric CO₂. Thus, accurate predictions of year-to-year variations in atmospheric CO₂require knowledge on predictability of carbon sinks. Several major modeling centers already include the land and ocean carbon cycle components into their decadal and seasonal prediction systems. These prediction systems have different assimilation and initialization designs.They do not assimilate any ocean or land biogeochemistry variables. The added value of predictions arises from the improvement of the initialized predictions compared to uninitialized projections. Here we examine the predictability of the ocean and land carbon sinks from several prediction systems based on Earth system models. We evaluate the predictive skill against the Global Carbon Budget estimate. A predictive skill of up to 3 years is determined for the air-sea CO₂ flux. Despite differences in model architectures, we find consistent spatial patterns and longer regional predictability for the ocean carbon uptake in all analyzed prediction systems. Predictive skill of the air-land CO₂ flux based on only two available models is up to 2 years. Land carbon uptake predictability mainly arises from the tropical and sub-tropical regions. Furthermore, we provide first estimates of anomalous atmospheric CO₂ growth rate predictability inferred from internal variability of the land and ocean carbon sinks. Our analysis indicates that ESM-based prediction systems have the potential to predict variations in atmospheric CO₂growth rate at lead time of 2 years.