Validation of DNDC for 22 long-term N 2O field emission measurements

Abstract

Twenty-two long-term measurements of direct N 2O emissions from soils in an intensive agricultural area were used for the validation of the process-based DNDC model (version 8.3P). Model simulations were evaluated for temporal patterns of N 2O, NH 4 +, NO 3 − and water-filled pore space (WFPS) and total N 2O emissions. Several soil and crop input parameter adjustments to the model were evaluated but only the recalculation of the WFPS at wilting point and at field capacity, using pedotransfer functions, resulted in a clear improvement of the simulated variables (WFPS in all cases, N 2O in some cases). Therefore, only this adjustment was made to DNDC 8.3P. This change, however, resulted for some cases (both cropland and grassland) in retardation of nitrate leaching and to a lesser extent of NH 4 + to the deeper soil layers. The goodness of fit of the simulated temporal pattern of N 2O varied considerably between sites. The total simulated N 2O emissions from cropland showed a good agreement with the measurements, although there was a systematic overestimation of 7.4 kg N 2O-N ha −1. Grassland soils, in contrast, gave a low agreement between total simulated and measured N 2O losses. On the basis of all measured data a regional emission factor of 3.16 with a 95% confidence interval of −0.89 to 7.21 could be calculated. DNDC simulations resulted in an emission factor of 6.49 with a 95% confidence interval of 4.04–8.93. The overall outcome of the N 2O emission measurements and DNDC simulations were compared with several empirical regression models, which may be applicable for a temperate climate system. All of the tested regression models showed reliable results up to a N 2O emission of 10 kg N 2O-N ha −1. Higher emissions, however, were systematically underestimated. Though DNDC both under- and overestimated specific sites, the general agreement, over the whole range between measurements and simulations of total N 2O losses (simulations=0.82×meas.+6.2), was better than for the different regression models.