Evaluation of denitrification from three biogeochemical models using laboratory measurements of N2, N2O and CO2

Abstract

Abstract. Biogeochemical models are useful for the prediction of nitrogen (N) cycling processes, but accurate description of the denitrification and decomposition sub-modules is critical. Current models were developed before suitable soil N2 flux data were available; new measurement techniques have enabled the collection of improved N2 data. We use measured data from two laboratory incubations to test the denitrification sub-modules of existing biogeochemical models. Two arable soils – a silt-loam and a sand – were incubated for 34 and 58 days, respectively. Fluxes of N2, N2O and CO2 were quantified using gas chromatography and isotope-ratio mass spectrometry (IRMS). For the loamy soil, seven moisture and three NO3− contents were included, with temperature changing during the incubation. The sandy soil was incubated with and without incorporation of litter (ryegrass), with temperature, water content and NO3− content changing during the incubation. Three common biogeochemical models (Coup, DNDC and DeNi) were tested using the data. No systematic calibration of the model parameters was conducted since our intention was to evaluate the general model structure or “default” model runs. As compared with measured fluxes, the average N2+N2O fluxes of the default runs for loamy soil were approximately 3 times higher for Deni, 105 times smaller for DNDC and 22 times smaller for Coup. For the sandy soils, default runs were 3 times higher for DeNi, 7 times smaller for DNDC and 12 times smaller for Coup. While measured fluxes were overestimated by DeNi and underestimated by DNDC and Coup, the temporal patterns of the measured and the modeled emissions were similar for the different treatments. None of the models was able to determine litter-induced decomposition correctly. The reason for the differences between the measured and modeled values can be traced back to model structure uncertainty and/or parameter uncertainty. Given the aim of our work – to assess existing model processes for further development and/or to identify missing processes within the models – these results provide valuable insights into avenues for future research. We conclude that the predicting power of the models could be improved through future experiments that collect data on denitrification activity with a concurrent focus on control parameter determination.