The role of measurement uncertainties for the simulation of grassland net ecosystem exchange (NEE) in Europe

Ecosystem models are used to explore scientific questions and as a means of decision support. It is therefore essential to assess the quality of the outputs from these. One important measure of quality of model outputs is their uncertainty that arise from uncertainties in measurements. In this study, uncertainty associated with simulation of NEE (net ecosystem exchange) by the PaSim (pasture simulation model) model was tested at four grassland sites. The effect of measurement uncertainties in the main input factors for climate (temperature, precipitation, global radiation), atmospheric CO2 concentration, soil characteristics (bulk density, clay fraction, pH, carbon content of different pools), and management (N input) on output uncertainty of NEE prediction was explored. Monte Carlo runs were performed for 2 years, 2002 and 2003, using Latin Hypercube sampling from probability density functions (PDF) for each input factor. Global uncertainty, represented by the standard deviation of the NEE estimates, was generally higher in 2003 than in 2002 per site with full data records for both years. There was considerable variation in global uncertainty from site to site and between years. These results indicate that output uncertainty does not depend solely on absolute input uncertainties. The mean NEE of the global uncertainty distribution was also different from the NEE result of the single model run using the default set of input factors. The results clearly indicate the dependency of factor importance and uncertainty in model simulations on the environmental conditions of the system under study. One indication is that more environmentally constrained systems produce higher uncertainties in model results. We conclude that any application of the model within decision support should be accompanied by a case specific uncertainty analysis presenting the mean of the output distribution as model results. The consequences of these findings for the application of PaSim at the continental scale are discussed.