Model Selection to Estimate N2o Emissions Measured with Closed Chamber Method in Rice.

Model Selection to Estimate N₂O Emissions Measured with the Closed Chamber Method in Rice

Nitrous oxide (N₂O), a major greenhouse gas that contributes to global warming and ozone depletion, is produced during nitrification and denitrification in the soil. However, N₂O fluxes from soil measured with closed chamber vary depending on the model used for calculations. The methodology for N2O measurement is complex and often few samples are taken and the emission rates extrapolated over season and locations. We measured N₂O fluxes from rice fields in Bangladesh under different fertilizer and water management treatments using closed chamber technique. From each chamber, gas samples were collected at every 8 min interval during 40 min chamber closure time and repeated eight times a day. Emissions were calculated with two regression models, linear and exponential, and a diffusion deterministic model, the Hutchinson-Mosier (HM), using data from the Boro(dry season) 2014 to identify significant emission events meeting three conditions, i.e., appropriate goodness of fit, response treatments, and sensitivity to treatment differences. 

The linear regression model is good predicting the general trend of N₂O emissions, but fails to follow the actual concave pattern of N₂O concentrations in 29% of the cases resulting in overestimations.  In contrast, the exponential model with its varying slope made an accurate estimation of the emission rate, and added significant emission events.  The HM which operates with only three or four points identified less significant emission events than the linear or exponential models.  Depending of which subset of the 6 observed points is used, it can result in severe overestimations or underestimations.

Significant events from the three models were in agreement following the time pattern resulting from management practices and were sensitive to treatment differences.  The seasonal total N₂O emission from the prilled urea treatment was 520, 535, and 247 g ha^-1 by the linear, exponential, and HM models, respectively.  Considering that about 29% of the linear model predictions are overestimations, the exponential model can be consider the best for estimating N₂O emissions from rice fields.