Unveiling non-linear thresholds of life-cycle carbon emissions in small reservoirs: an integrated stochastic EI-LCA and machine learning framework.

Small reservoirs are critical for rural water security, yet their life-cycle carbon emissions remain poorly quantified due to extreme data scarcity and non-linear emission dynamics. Conventional Life Cycle Assessments (LCA) often rely on static assumptions, failing to capture complex multi-variable interactions. To overcome this, we propose an integrated stochastic framework coupling Environmentally Extended Input-Output LCA (EI-LCA) with Monte Carlo simulations, Extreme Gradient Boosting (XGBoost), and Shapley Additive Explanations (SHAP). Applied to the Huxingdi Reservoir in China, the framework effectively mitigated uncertainty from sparse upstream data. Results unveil a distinct emission profile for small reservoirs, fundamentally diverging from mega-dams, with heavy machinery operation (construction phase) and localized biogenic GHG fluxes (operational phase) acting as co-dominant drivers. More importantly, unlike methods that only produce abstract feature rankings, our SHAP-based piecewise regression can identify statistically robust operational thresholds. We determined an "economic carbon radius" of 41.3 km for material transportation. Carbon emissions increase nonlinearly once transport distances exceed this threshold. In addition, we detected critical tipping points of machinery efficiency. These points establish quantitative reference benchmarks to guide equipment retrofitting. We further discuss the methodological limitations of empirical biogenic models (e.g., G-res) and the boundaries of SHAP attribution. Ultimately, this study transitions data-driven LCA from macro-scale prediction to granular, actionable guidelines, providing a highly scalable blueprint for low-carbon engineering procurement and watershed management.