Quantification of methane emissions in rice paddies: A machine learning approach with rice cultivar and transplanting date

Estimates of methane (CH 4 ) emissions from rice paddies are crucial to assess the contribution of rice production in mitigating or adapting climate change.This study aimed to develop a predictive model for estimating the CH 4 emissions in rice paddies, and suggest rice production strategies for sustainable agriculture.The parameters for constructing the CH 4 emissions predictive model were obtained from field-based experiments from 2022 to 2023 in Daejeon, South Korea and separated as follows: climatic factors (i.e., mean temperatures of atmosphere, soil, and water, annual total precipitation), cultivation factors (i.e., cultivar and transplanting date), and soil properties (i.e., pH, organic matter, and bulk density).Additionally, before the developing the predictive model, each variable underwent significance test (p < 0.05).The statistical relationship and regression analysis of the CH 4 emissions were followed the Pearson correlation method and multiple linear regression (MLR) algorithm, respectively.Consequently, the CH 4 emissions had the negative correlation with the mean water temperature, total precipitation, transplanting date, and bulk density, where the mean water temperature had the highest negative coefficient at -0.824.In contrast, the CH 4 emissions were raised by the increase in the mean atmospheric temperature (+0.231) and organic matter content (+0.391).The predictive model for forecasting the CH 4 emissions had the coefficient of determination (R 2 ) value of 0.836, which exhibits the reliability of 84% for estimating the CH 4 emissions.In predictive model, soil bulk density (+18.2933) had the highest influence, while total precipitation exhibited the lowest coefficient at -0.0426.The statistical metrics, which showed the range and occurrence of residuals in the predictive model, were evaluated at 8.802 kg ha -1 and 7.235%, respectively.These findings serve as a valuable tool for establishing strategies to estimate the CH 4 emissions in rice production, whereas expanding the dataset to include various experimental regions was needed to enhance the model's performance.