Assimilating soil organic carbon measurements to improve model projections

Projecting and forecasting the global terrestrial soil carbon (SOC) pools is an important task due to its size as well as interaction with vegetation and climate systems. However, it is a challenge because of inherent uncertainties in both model dynamics and initial state measurements. Furthermore the majority of existing observations of the total soil carbon cannot be used for detailed modeling as the exact observations needed to accurately initialize the model are not commonly done.

To address this issue, we use an established statistical method known as state data assimilation to combine information from measurements and model projections to create an improved state estimate. We tested this concept using the long continuous measurement time series from the Fallow multi-decadal data set. Here the Yasso SOC model was the chosen model and Ensemble Adjustment Kalman Filter (EAKF) the state data assimilation method. The results show drastic improvements in SOC forecasts already after only the first few observations. In addition they establish aspects such as data weights and covariances that need to be considered during implementation.