Contributing Mechanisms to Skillful Seasonal Forecasts of Spring Carbon Uptake

Forecasting the uptake of carbon at seasonal lead times is challenging because of the uncertainties in seasonal meteorological forecasts and the complex feedbacks between the energy, water, and carbon cycles. Here, using a state-of-the-art seasonal forecast system, we demonstrate an ability to accurately forecast spring carbon uptake at multi-month lead times. Twenty 6-month forecasts of meteorology from NASAs subseasonal-to-seasonal (S2S) ensemble forecast system, each forecast beginning in a particular January during 2001-2020, are used to drive an offline terrestrial biosphere model. The resulting prediction of spring Gross Primary Production (GPP) is then evaluated against a fully independent, observational dataset derived from high quality MODerate-resolution Imaging Spectroradiometer (MODIS) reflectances. We find skillful forecasts of spring GPP in western North America, Europe, and parts of Asia. We can attribute the skill in western North America during boreal spring to meteorological variability, which is naturally tied to the amount of snow at the start of the forecast and its coupling to air temperature; the skill largely occurs where we have skillful forecasts of snow cover removal date and greening onset date. The forecast skill in Europe and central Asia is presumably attributable in part to a realistic forecast of soil moisture availability through a proper snow and soil moisture initialization. In late spring/early summer, the initialization in our offline forecasts of the carbon and vegetation variables appears to contribute to GPP prediction skill in Europe and parts of east Asia. Overall, this study highlights the significance of accurate land initialization (in particular, the initialization of the midwinter snow and the carbon and vegetation prognostic variables) for forecasts of GPP at seasonal lead times.