Challenges and Insights for Simulating Nitrous Oxide Emissions in Eastern Canada: Evaluating an Agroecosystem Model Ensemble

Biogeochemical models continue to be improved in their ability to account for the impacts of agricultural management, soil characteristics, and climate on crop productivity and greenhouse gas emissions. Depending on the model, limitations still exist including the ability to characterize a limited range of management practices, the oversimplifications of crop physiology, and an inadequate simulation of soil microbial environments. This study uses a long-term 22-year field experiment in eastern Canada to calibrate and evaluate several agroecosystem models, including DayCent, DNDC, DSSAT, and STICS, for their ability to simulate crop productivity and nitrous oxide (N₂O) emissions. Model performance was assessed against near-continuous N₂O measurements using flux towers. Corn, wheat, soybean, and canola were grown over the 22 years for several treatments including manure versus inorganic fertilizer, fertilizer rate, timing of fertilizer applications, early and late planting, and use urease and nitrification inhibitors. Findings suggest that the ensemble of models could accurately predict corn, wheat and soybean yields in contrast to the general overprediction of canola yields. Growing season N₂O emissions are generally well-simulated at the Ottawa site with weekly performance statistics showing Wilmot d values of 0.7 for conventional management and 0.75 for BMP management. However, challenges persist in accurately capturing daily emission patterns and estimating emissions during the spring-thaw period. The DSSAT and STICS models, which do not have explicit soil mechanisms related to spring thaw, simulated low N₂O emissions and thus it is recommended that these mechanisms be incorporated in the future. Difficulties in modeling the timing of denitrification events highlighted limitations in the representation of microsite-level pedoclimatic conditions, diffusion processes, and the simulation of microbial activity. The model ensemble simulated an acceptable level of annual N₂O emissions for most treatments with 5.8% overprediction across 22 years, with the overestimation mainly from the manure and dual inhibitor treatments. Comparing model strengths and weaknesses across different locations provides valuable insights for future model improvements.