Machine learning-based approach for estimating greenhouse gas emissions in varying silicate-fertilized rice paddies

Tropical Plant Biology · Springer Nature · 2026-10-23 · DOI: 10.1007/s10333-026-01105-3

Аннотация

Accurate greenhouse gas (GHG) quantification from rice paddies is essential for climate change adaptation but is constrained by labor-intensive measurement methods. Silicate fertilizers have the potential to mitigate GHG emissions, yet predictive tools for evaluating their effectiveness remain underdeveloped. The objective of this study was to develop and validate machine learning (ML) models for predicting methane (CH 4 ), nitrous oxide (N 2 O), and carbon dioxide (CO 2 ) emissions from silicate-fertilized rice paddies using environmental covariates as input features. Field experiments were conducted over 2-year using three silicate fertilizers with different iron slag contents (i.e., 0.0%, 2.5%, and 5.0%) and a control treatment. Environmental data, including meteorological variables, soil chemical properties, and fertilizer compositions, were collected. Three ML algorithms (i.e., random-forest, gradient boosting, and k-nearest neighbor) were evaluated using the first-year data for model development and the second-year data for cross-year validation. CH 4 emissions showed strong negative correlations with soil redox potential (Eh, -0.60), whereas N 2 O emissions were positively correlated with soil Eh (+ 0.92). Although gradient boosting showed strong learning performance on the training data, random-forest demonstrated superior generalizability, with coefficient of determination (R 2 ) values of 0.86 (CH 4 ), 0.87 (N 2 O), and 0.81 (CO 2 ) during cross-year validation. Measured and predicted cumulative GHG emissions were generally in close agreement, although a noticeable discrepancy was identified for CH 4 emissions in the control treatment. This approach reduces reliance on labor-intensive measurements while improving the scalability of GHG monitoring. The integration of silicate fertilization with ML-based prediction provides a practical tool for evidence-based environmental management in rice production.

Читать у издателя →

Похожие статьи