WaterMAP: A Scalable Machine Learning Framework for Emission-Factor-Derived Spatiotemporal GHG Prediction and...

Greenhouse gas (GHG) emissions from wastewater treatment have gained increasing attention in global climate governance. However, conventional emission-factor (EF)-derived inventories lacked the ability to capture nonlinear variations, while existing machine learning (ML) models remained fragmented in scale. Here, we introduce WaterMAP (Wastewater AI Treatment Emission Regression for Multi-scale Accounting and Prediction), a novel ML framework designed for spatiotemporal EF-derived GHG prediction and mitigation evaluation. Using 40,722 inventory-based GHG records from 5155 WWTPs in China during 2009-2019 as a case study, the National_Total_Model achieved a test RMSE of 0.19 kg CO2-e/m3 when predicting the total EF-derived intensity. National_Type_Model showed that scope 1 totaled 7.6 Mt CO2-e, scope 2 accounted for 18.9 Mt CO2-e, and scope 3 contributed 0.5 Mt CO2-e in 2019. We then estimated cumulative GHGs for 2024 to be 36.8 Mt CO2-e, with an average intensity of 0.51 kg CO2-e/m3. Treated volume, TNinf, sludge yield, latitude, and CODinf were identified as the key predictors influencing EF-derived GHG emissions. We further proposed a GHG spatial heterogeneity index using Provincial_Type_Model, reflecting urban development. Under exploratory sensitivity analysis toward 2060, WaterMAP-guided projections suggested 9.6-34.0% potential reduction. Overall, WaterMAP offers a scalable framework for EF-derived GHG emissions screening and prediction, supporting GHG mitigation assessment.