Replication data and code for: Volume, not intensity: built environment and taxi peak-commuting carbon in central Shanghai

Derived data and analysis code supporting the manuscript "Volume, not intensity: built environment and taxi peak-commuting carbon in central Shanghai" (submitted to Transportation Research Part D: Transport and Environment). CONTENTS data/ — 500 m grid-level derived layers for central Shanghai (n = 1,494 cells): 29 built-environment indicators; commuting carbon outcomes (all-day occupied, morning peak, evening peak); empty-cruising (deadheading) carbon; trip counts, per-trip carbon and short-trip shares; opportunity-accessibility layers; multi-scale (250 m / 1 km) MAUP aggregates; robustness outputs (alternative centre sets, disjoint commuting definitions); external-validation summaries; the 500 m grid geometry (GeoPackage); and full-precision model hyperparameters. code/ — Tier-A analysis scripts that run entirely from the released data with relative paths: OLS / Moran's I / LISA, spatial Durbin model, GWR/MGWR, XGBoost with spatial-block cross-validation, SHAP, ALE, single-variable counterfactual scan, and NSGA-II bi-objective optimisation (F1 = minimise predicted commuting carbon; F2 = maximise opportunity accessibility). code_restricted/ — Tier-B provenance scripts requiring the restricted raw inputs (GPS trajectories, mobile-signalling OD, raw POI); these cannot be re-executed from this package. Their aggregated outputs are included in data/. output/ — Trained models, hyperparameters, spatial fold assignment and cross-validation metrics, enabling byte-level model reproduction. MANIFEST.csv — Full 32-character MD5 checksums and file sizes for every file.README_ZENODO.md — Data dictionary, units, reproduction instructions, and an explicit statement of what can and cannot be reproduced from this package. IMPORTANT — TWO CARBON BASELINESThe released carbon fields are computed with the reference emission factor EF = 2.98 kg CO2/L, on which all models were trained. The manuscript reports absolute carbon with a calibrated factor EF = 2.31 kg CO2/L (see Appendix H of the manuscript); grid_500m_carbon_ef231.csv provides these values directly (= released carbon x 0.775). All scale-invariant quantities — percentages, SHAP values, ALE profiles, correlations, Moran's I, accessibility, exchange rates — are identical under either factor. DATA AVAILABILITY AND RESTRICTIONSThe underlying taxi GPS trajectories and the mobile-signalling job-housing OD data are provided by third parties under authorisation and cannot be redistributed. Only de-identified, aggregate grid-level derivatives are released here. Consequently the derivation of carbon from raw trajectories cannot be re-executed from this package; the released grid data can, however, be verified for internal consistency and used to reproduce every statistical and modelling result reported in the manuscript. ACCESSFiles are restricted during peer review. Editors and reviewers may request access, or use the access link provided in the cover letter.