Uncertainty propagation in soil greenhouse gas emission models: An experiment using the DNDC model and at the Oensingen cropland site

Abstract

The increase of green house gas (GHG) concentrations in the atmosphere is predominantly caused by the anthropogenic activities of fossil fuel burning and land use change. The flux of GHGs from soils and ecosystems to the atmosphere is large, and any errors in estimating these fluxes have a significant impact on our quantification of the relative importance of land use in contributing to global warming. Numerical models have been developed to estimate the net flux of the biogenic GHGs: CO 2, N 2O and CH 4, for various agricultural management practices. These models have been developed using data from many different experimental sites around the world, encompassing different crops, farm management systems, soil and climatic conditions. Crop experiments and GHG flux measurements are expensive and last several years if not decades so these models are often used to test hypothesis about the effect of future conditions, land use scenarios and also to predict the effect of novel land management scenarios to reduce emissions. However, uncertainties in the input soil parameters and meteorological data that drive these models propagates through them, resulting in uncertainties in the predictions of biogenic GHG emissions. This paper describes an experiment that investigates how well the commonly used de-nitrification de-composition (DNDC) soil model performs when used to predict the eddy-covariance CO 2 fluxes and crop yields measured in the first full year of the Oensingen cropland site in Switzerland. DNDC N 2O predictions are compared to the IPCC emissions factors for arable land. This study includes an estimation of the uncertainty of soil input parameters, a sensitivity study as to their effect on predicted GHG emissions and the propagation of their uncertainty through the model. This study considers uncertainty in meteorological measurements and the impact of using subsets of this data in the model. In particular the effect of using monthly meteorological parameters to generate daily time series for input into the model is investigated and the error propagation quantified. The overall impact of uncertainty in input parameters on predicted biogenic GHG emissions is relatively small with the PDF of the uncertainties indicating that the NEE is over estimated by 3.6% and has a SD of 3.6% of the actual NEE. Nitrous oxide emissions are not biased but have a larger SD of 23% of emissions, which when the global warming impact is considered is only 3% of net flux. DNDC can therefore be used with confidence to predict emissions, with the caveat that the biomass production needs to be match to local conditions.