|Gaillard, Richard - University Of Wisconsin|
|Jones, Curtis - University Of Maryland|
|Ingraham, Peter - Applied Geosolutions, Llc|
|Collier, Sarah - University Of Wisconsin|
|Izaurralde, Robert - University Of Maryland|
|Osterholz, William - University Of Wisconsin|
|Salas, William - Applied Geosolutions, Llc|
|Ruark, Matthew - University Of Wisconsin|
Submitted to: Ecological Applications
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/27/2017
Publication Date: 4/2/2018
Citation: Gaillard, R.K., Jones, C.D., Ingraham, P., Collier, S., Izaurralde, R.C., Jokela, W.E., Osterholz, W., Salas, W., Vadas, P.A., Ruark, M. 2018. Underestimation of simulated N2O flux in a model comparison of DayCent, DNDC, and EPIC. Ecological Applications. 28:694-708.
Interpretive Summary: Computer models are increasingly used to study complex agroecosystems, including the greenhouse gas, N2O, emissions from agricultural fields. These models need to be periodically evaluated and compared to make sure they are reliable and useful. We evaluated N2O flux from three models (DayCent, DNDC, and EPIC) using data from two Wisconsin research sites that represent cropping systems and nitrogen fertilizer management strategies common on dairy farms. Results showed that the models tend to underpredict N2O flux when measured flux is high. How the models simulated soil temperature and water content, which influence N2O emissions, did not explain model underprediction. We found that the models varied a lot in how they simulated other soil processes that influence N2O emissions, including microbial respiration, denitrification, and soil nitrogen availability. This indicates that field experiments should be measuring these other variables so they can be used to test the models. For now, results suggest there are limits to how well the models can be used for policy or management recommendations, and that scientists need to work together to improve the models.
Technical Abstract: Process-based models are increasingly used as tool for studying complex agroecosystem interactions N2O emissions from agricultural fields. The widespread use of these models to conduct research and inform policy benefits from periodic model comparisons that assess the state of agroecosystem modeling and indicate areas for model improvement. Multi-institutional collaborations provide an opportunity to access large datasets and perform broad model evaluations that inform both model development and policy applications. This work provides an evaluation of simulated N2O flux from three process-based models: DayCent, DNDC, and EPIC. The models were calibrated and validated using data collected from two research sites over five years that represent cropping systems and nitrogen fertilizer management strategies common to dairy cropping systems. With the exception of DNDC, calibration for daily N2O flux produced positive Nash-Sutcliffe evaluations of individual model efficiency (E = 0.30 and 0.42 for DayCent and EPIC, respectively), though efficiency over mean observation in validation was reduced (E = 0.03 for both DayCent and EPIC). We also evaluated the use of a multi-model ensemble strategy, which outperformed individual models in validation (E = 0.10) but not calibration (E = 0.15). Regression analysis indicated a cross-model bias in the estimation of high magnitude flux events, where for every observed g N2O-N ha-1 day-1, the models underestimated flux by an ensemble average of 0.73 g N2O-N. Model estimations of observed soil temperature and water content did not sufficiently explain model bias, and we found significant variation in model estimates of heterotrophic respiration, denitrification, soil NH4+, and soil NO3-, which may indicate that additional types of observed data are required to evaluate model performance and possible biases. Our results suggest a bias in the modeling of N2O flux from agroecosystems that limits the extension of models beyond calibration and as instruments of policy development. This highlights a growing need for the modeling and measurement communities to collaborate in the collection and analysis of the data necessary to improve models and coordinate future development.