Machine learning-driven method for in-situ high-frequency CH measurement in paddy fields based on water-soil-air factors: A case study of the Yangtze River Basin.

Accurate and high-frequency monitoring of methane (CH) from rice paddies is crucial for effective carbon emission control but remains challenging due to fluctuant emissions and complex field environments. This study proposed a new in-situ high-frequency CH4 measurement method based on machine learning and sensor-measurable water-soil-air environment factors. The results show that: (1) soil and paddy water serve as critical media influencing CH production and transportation, with paddy water depth (H), soil electrical conductivity (EC), and soil temperature (T) being significantly positively correlated with CH emission flux, while soil redox potential (Eh) had a negative effect (p < 0.05). (2) The decision tree (DTR) showed the best accuracy for CH inversion, with soil factors being the optimal input group (R = 0.84), which was superior to water-soil (0.83), water-soil-air (0.55), and air-soil (0.45) groups; Eh, EC, soil pH, and T are the essential input variables (R>0.80). (3) Combining the immediacy of multi-sensor detection and the accuracy of machine learning, the new method demonstrates notable advantages in high frequency, high accuracy, synchronous multiparameter monitoring, and low cost. This method enables the real-time monitoring and control of CH emission from paddy fields, thereby offering new perspectives for CH monitoring in small water bodies (such as ditches, ponds, lakes, etc.).