Quantifying uncertainties in model predictions is critical for evaluating the forecasting capabilities of ecosystem models and improving e confidence on ecological forecasting. The uncertainty in observed data is one of the main sources of forecasting uncertainty. Based on ten years data collected at Duke Forest FACE facilities, we accessed the uncertainties in model parameters and predictions by Markov Chain Monte Carlo approach.

Results/Conclusions

The results showed that high measurement errors (standard deviation, SD) led to high modeling uncertainties generally. For well constrained parameters, highly precise parameter estimation and model prediction were achieved by reducing SD. However, for the parameters that were not constrained, reduced SD still couldn’t result in a better constraint. The predicted carbon content in the pools was constrained well except the two litter pools and the microbial C pool. Reduced observation SD led to low uncertainties in simulated carbon content except for the microbial C pool, which had little changes with SD. The confidence intervals of forecasting increased with in model projections, especially for the pools with low fluctuation with climate, such as woody biomass, soil carbon. But for the pools that are sensitive to environmental fluctuations, they increased little. This study indicated that the observation errors can affect the confidence intervals of estimated model parameters and model forecasting. Generally, the higher precise the data are, the less uncertainties in model parameters and forecasting. The results indicate that quantifying uncertainties of modeling can improve our insights on the capabilities and limitations of ecosystem models in forecasting.