Supporting data and code for: Next-Day PM2.5 Forecasting in Nanjing: Random Forest Cuts Persistence Error by 19.4% under a Frozen Chronological Protocol

Reproducibility package for a manuscript submitted to Environmental Monitoring and Assessment: a leakage-controlled chronological benchmark comparing random forest, XGBoost and LSTM against persistence baselines for next-day PM2.5 in Nanjing, with SHAP analysis, Diebold–Mariano and bootstrap inference, a seven-city cross-city extension and an expanding-window test-year evaluation. It contains the analysis code, per-day model predictions, metrics, tuning and verification records, the manuscript sources, and SHA-256 checksums plus retrieval URLs for every raw input. The third-party raw archives and the derived daily tables are not redistributed here (they are available from the corresponding author); see SOURCES-AND-TERMS.md.