Does agroecosystem model improvement increase simulation accuracy for agricultural N2O emissions?

Abstract

In recent decades, agroecosystem models have been developed to simulate agricultural nitrous oxide (N2O) emissions. Coefficients of determination (R2) and root mean square error (RMSE) are widely used as metrics to assess the explanatory power and simulation accuracy of models and to provide perspectives on model improvement as models evolve. This study aimed to determine whether the fitting accuracy of three agroecosystem models to simulate agricultural N2O emissions improved with advancing versions of the models. We used several quality evaluation criteria to extract 94 and 97 reported R2 and RMSE values, respectively, from 32 published articles related to the use of three of the most-used agroecosystem models in the research field of N2O emissions [i.e., DNDC (DeNitrification-DeComposition), DayCent, and APSIM (Agricultural Production Systems sIMulator)]. Results showed that there was (1) no significant improvement in simulating N2O emissions between DNDC9.3 and DNDC9.5; and (2) no significant difference between the simulation abilities of the original models and the user-defined revised models for these widely-used models. These findings may be mainly a result of the offsetting consequences of changes in publication bias and increased focus on complex agricultural issues. The study also found that the simulation accuracy of DNDC was better under conditions of higher annual mean temperature and soil bulk density and lower soil total nitrogen, mainly caused by the formulas and data used to build and validate the model. The study results suggest that the suitability of a model for simulating N2O emissions depends on the climatic and soil conditions at the location of its application. Improving the simulation accuracy of agroecosystem models will require further targeted corrective and development actions in the future.