Modeling N2O emissions of complex cropland management in Western Europe using DayCent: Performance and scope for improvement

Under the United Nations Framework Convention on Climate Change (UNFCCC), industrialized countries and countries with economies in transition (so called Annex 1 countries) are encouraged to move towards more sophisticated approaches for national greenhouse gas reporting. To develop a model-based approach for estimating nitrous oxide (N2O) emissions from agricultural soils, model calibration is one of the first important steps. Extensive multisite field observations are necessary for this purpose, as agricultural management in Western Europe is complex (e.g., diverse crop rotations, different types of fertilizer and soil tillage). In the present study, we used ca. 24,000 daily N2O flux observations from six cropland sites, two in France and four in Switzerland, to conduct an automatic data-driven calibration of the biogeochemical model DayCent. This model is planned to be used for greenhouse gas reporting in the entire European Union as well as in Switzerland. After a site-specific calibration, a leave-one-out (LOO) cross-evaluation was conducted to assess the model’s ability to predict N2O emissions for sites it was not calibrated for. Mean observed N2O fluxes for 54 interactions of crop cycles, field studies and treatments were used to evaluate the model. The LOO cross-evaluation resulted in a R2 of 0.63 for the prediction of mean N2O fluxes per crop cycle, compared to an R2 of 0.51 obtained with default parameterization. Our results showed that the improvement in N2O predictions was associated with the adjustment of only seven parameters controlling the N cycle in soil (e.g., the maximum daily nitrification amount and the inflection point for the effect of water-filled pore space on denitrification) out of several hundred parameters. These parameters showed a wide range of values between sites, revealing an important challenge for calibration-based improvement of N2O simulations. Despite the remaining uncertainty, our model-based estimates of N2O emission per crop cycle (2.64 kg N ha-1) were clearly closer to measurements (2.67 kg N ha-1) than commonly used emission factor approaches (1.60–1.71 kg N ha-1). Based on extensive field observations, our results suggest that, after data-driven calibration of only few N cycle parameters, DayCent simulations are useful for reporting N2O emissions of complex cropland management. These model based-estimates were more accurate, because they consider key drivers that are disregarded by simpler approaches. Moving towards more complex methods of N2O reporting, is therefore expected to improve the accuracy and additionally allows to assess mitigation options.