Modelling forest SOC change – calibration and validation challenges

Soil organic carbon (SOC) is the largest terrestrial carbon (C) pool with a vital role in the global C cycle. Considering it is one of five mandatory pools in national greenhouse gas (GHG) inventory reports, it is important to accurately assess SOC stocks and changes. Measuring SOC stock changes is challenging due to costly and destructive soil sampling, the high spatial variability of soil carbon and the slow process of soil C accumulation or loss. In order to reduce the uncertainty of SOC stock changes estimates, repeated national soil inventory is required. In the absence of repeated national inventories, SOC stock changes could be estimated using a modelling approach. The aim of our research is to calibrate and validate the terrestrial ecosystem model Biome-BGCMuSo for the simulation of SOC stock changes in lowland forests as an additional tool for use in national GHG inventory reporting.

In our work, we combine different data sources (chronosequence experiment and eddy-covariance (EC) site) and different data types and frequencies (long-term C stocks and high-frequency C fluxes) of various ecosystem variables (aboveground live wood C (AGC), forest floor C, SOC, Net Ecosystem Exchange (NEE), Gross Primary Productivity (GPP) and Ecosystem Respiration (RECO)). The model calibration was performed using the daily values of main ecosystem C fluxes from the EC tower in the Jastrebarsko pedunculate oak forest and annual data on AGC, forest floor C and SOC from permanent measurement plots in the footprint of the EC tower. For model validation, we used annual data on C stocks in the aboveground live wood biomass, forest floor and mineral soil in the top 30 cm from seven stands of pedunculate oak chronosequence in Jastrebarsko forest. All analyses were performed in R software.  

Measured SOC showed no age trend and high between-stand spatial variability of 19-30%, while for modelled SOC between-stand spatial variability was only 6% and a negative age trend was observed. The calibration using solely daily NEE fluxes resulted in a better overall agreement of model output with observations for this variable, but at the cost of the reduction in intra-seasonal variability. The calibration using aboveground and soil C stocks improved the agreement for these variables but caused greater discrepancies between measured and modelled daily NEE fluxes. The model validation showed a good agreement for C stock change in aboveground live wood biomass and mineral soil for most of the chronosequence stands, but with high disagreement between measured and modelled C stocks in the forest floor in general. Obtained results emphasize the importance of multi-variable calibration and validation to improve model accuracy and robustness across all simulated pools, fluxes and processes.