Modeling of Greenhouse Gas Emissions in Irrigated Lowland Rice by the Alternate Wetting-Drying (AWD) System

Abstract

Water management in rice farming significantly affects greenhouse gas (GHG) emissions, especially in irrigated lowlands. The alternate wetting and drying (AWD) method can reduce emissions without yield loss. To quantity GHG levels by AWD, the proper model is required under dynamic field conditions. Thus, the study aims to develop GHG emission model with AWD systems utilizing bio-physical environmental parameters with an artificial neural network (ANN) as well as compared with other irrigations system. The empirical data were observed according to three irrigation regimes, i.e., continuous flooding (FL), wetting (WT), and drying (DR) as representing AWD. Two main gases, methane (CH₄) and nitrous oxide (N₂O), were measured weekly by standard methods, alongside environmental parameters as the inputs model. The ANN model demonstrated excellent performance in predicting CH₄ emissions, with a coefficient of determination (R²) of 0.8544, indicating high predictive accuracy. The model also showed acceptable performance in predicting N₂O emissions (R² = 0.7258), although with greater variability and sensitivity to extreme values. These results suggest that the ANN is effective for modeling CH₄ emissions, while further refinement—such as data transformation and model separation—may enhance the accuracy of N₂O predictions.