Evaluating and improving land surface model predictions using site-level observations

We evaluated the performance of the Community Land Model, version 4, using site-level observations compiled through the North American Carbon Program Site Synthesis activity. Simulations were performed at 14 forest sites, and predictions of carbon, water, and energy fluxes and ecosystem state variables were compared with site observations. We found strong evidence that the initial model implementation of nutrient limitation was inadequate to capture diurnal variation in net carbon flux. By introducing a plant nutrient storage capacity, we were able to significantly improve the model performance at both diurnal and seasonal time scales, without adversely impacting water and energy flux predictions. We were also able to identify two other factors leading to poor performance at a subset of the sites, and suggest here some additional modifications to the cold-season sensitivity of photosynthesis and the implementation of the prorgnostic fire module.