Jump to Main Content
Assessing the performance of three frequently used biogeochemical models when simulating N2O emissions from a range of soil types and fertiliser treatments
- Zimmermann, J., Carolan, R., Forrestal, P., Harty, M., Lanigan, G., Richards, K.G., Roche, L., Whitfield, M.G., Jones, M.B.
- Geoderma 2018 v.331 pp. 53-69
- Monte Carlo method, application rate, cropland, fertilizer application, grasslands, grazing, greenhouse gas emissions, greenhouse gases, land use, model validation, models, nitrous oxide, regression analysis, soil types, spring barley, uncertainty
- Biogeochemical models have the potential to provide insights into the key drivers of greenhouse gas (GHG) dynamics, and may be used in Tier 2 and 3 GHG emission reporting. Modelling nitrous oxide (N2O) emissions from agriculture, however, is still subject to large uncertainties. In the present study we analysed the performance of the three semi-mechanistic models, DailyDayCent (DayCent), DeNitrification-DeComposition (DNDC 9.4 and 9.5), and ECOSSE when simulating N2O fluxes from two different land uses (simulated grazing and spring barley) under a range of fertiliser types and application rates. Model performance was assessed using linear regression analysis, root mean square error (RMSE), and relative error. Monte Carlo analysis was carried out to assess the sensitivity of DayCent and ECOSSE to changes in the timing of management events.The results show high variability in model performance. The performance of each model was dependent on both site and treatment, with no model showing consistently good performance. When averaged across all sites and treatments DayCent simulations produced the lowest number of significant total and relative errors. When looking at the relationship between modelled and measured N2O fluxes, ECOSSE performed best with a significant relationship in 61.8% of all simulations and an average r2 of 0.2. However, outputs from this model displayed the largest total and relative errors. Performance when simulating cumulative fluxes was generally poor. The Monte Carlo analysis showed that shifts in timing of management events by ±7 days lead to annual N2O fluxes varying by 5.6 ± 7.6% and 2.8 ± 4.2% for grassland and cropland, respectively. However the impact of the timing of single events can lead to much larger responses in N2O emissions. The large variation in model performance suggests that further development and calibration is required before using models in GHG reporting.