Machine learning-driven analysis of greenhouse gas emissions from rice production in major Chinese provinces: Identifying key factors and developing reduction strategies

Abstract

Rice cultivation is a significant contributor to global greenhouse gas (GHG) emissions. However, the complex nonlinear relationship between driving factors and GHG emission intensity (GHGI) remains poorly understood, and effective reduction strategies are still needed. This study integrates machine learning models and SHapley Additive Explanations (SHAP) to assess the nonlinear relationship and design GHGI reduction strategies based on data from 14 provinces in China from 2012 to 2022. The key findings are as follows. (1) For GHGI reduction, the optimal conditions include an annual average sunshine duration of 47–75 days, an annual average temperature of 15.3–17.9℃, annual average precipitation levels of either 1000.0–1368.4 or 1680.0–2004.7 mm, soil pH below 5.6 or above 6.5, soil total nitrogen content of 17.0–20.3 g/kg, and soil organic carbon content of 15.0–22.5 g/kg. The recommended application rates for nitrogen, phosphate, and potassium fertilizers are 160.0–311.0 kg/ha, 124.9–129.9 kg/ha and 144.0–194.3 kg/ha, respectively. Agricultural practices such as transplanting, mixed farming, tillage and mid-season drainage demonstrate higher GHGI reduction potential compared to other measures. (2) For lowest-cost GHGI reduction strategies in major provinces, Heilongjiang, Jilin, and Liaoning provinces could reduce GHGI to 0.28, 0.15, and 0.05 tCO₂e/t, respectively, by adjusting sunshine conditions. Hainan, Guangdong, Fujian, Jiangsu, Jiangxi, Zhejiang and Guangxi provinces could achieve GHGI reductions to 0.62, 0.31, 0.21, 0.47, 0.57, 0.92 and 0.28 tCO₂e/t, respectively, by optimizing nitrogen fertilizer application and labor practices. Hunan and Anhui provinces could reduce GHGI to 0.57 and 0.85 tCO₂e/t by adjusting irrigation modes. Implementing these strategies would result in an average GHGI reduction of 28.75 %, although production costs per mu for early, mid-to-late indica and japonica rice in major provinces would increase by 28.87 %, 27.95 % and 27.38 %, respectively, compared to the original production costs. These findings provide valuable insights and a scientific basis for developing GHGI reduction strategies in rice production and enhancing the sustainability of this critical agricultural sector.