Machine Learning-Based Mapping of Daily Methane Concentration in Rice Paddies Using Meteorological Data and Satellite Images: A Case of South Korea

Methane from rice fields has a strong greenhouse effect and its accurate estimation is essential to combat climate change.In this study, we conducted an analysis based on the Gradient Boosting Machine (GBM) model using Local Data Assimilation and Prediction System (LDAPS) data, Normalized Difference Vegetation Index (NDVI), and Normalized Difference Water Index (NDWI) from VIIRS and Moderate Resolution Imaging Spectroradiometer (MODIS), and FluxNet ground observations of the Cheorwon rice paddy region.This was used to estimate methane emissions from rice paddy fields in South Korea and to create a gridded spatial information map of methane concentrations.Using data with a spatial resolution of 1.5 kilometers, we identified detailed changes within the region and generated daily maps to analyze daily changes and seasonal characteristics.To predict methane concentration, we considered the correlation between meteorological factors such as latent heat flux, humidity, soil moisture, and soil temperature and methane emissions as key variables.Latent heat flux and humidity were selected as key variables considering that the migration of methane gas is affected by the evapotranspiration process.In addition, soil moisture, which creates the anaerobic conditions necessary for methane production, and soil temperature, which affects the activity of methanogenic microorganisms, were included in the analysis.Taking these various factors into consideration, we analyzed methane emission data from rice fields in Korea and visualized them on a map to understand the pattern of methane production in response to changing weather conditions.The developed model showed a correlation coefficient of 0.91 and Mean Absolute Error (MAE) of 28.97 in the 5-fold cross-validation and an average correlation coefficient of 0.87 and MAE of 35.46 in the Leave One Year Out (LOYO) cross-validation.These results are expected to contribute to the understanding of methane generation patterns under changing weather conditions and accurate methane emission estimation.In addition, the developed model and the constructed methane concentration map can be utilized as an important basis for establishing greenhouse gas reduction policies in the agricultural sector and effective climate change response strategies in the future.