Processed ERA5 Regional Series, Model Outputs, and Event-Definition-Aware Reliability Audit for High-Runoff Hindcasting in Chongqing, China

This dataset supports the manuscript “Event-Definition-Aware Reliability Auditing of ERA5-Based High-Runoff Hindcasts in Chongqing, China.” Version 2.0 contains processed municipality-scale ERA5 daily and monthly series for 2014–2024, calendar exceedance and rainfall–runoff event catalogues, gradient-boosting and five-member LSTM ensemble predictions, seed-specific LSTM results, conformalized quantile regression intervals, complete model configurations, results for all 27 event definitions, 2,000 paired 14-day moving-block bootstrap replicates, six flow-regime assessments, upper-tail event results, and the analysis code used in the study. ERA5 runoff is treated as a reference proxy and should not be interpreted as observed river discharge. No test event exceeded the maximum runoff observed during model development; therefore, the upper-tail audit does not represent genuine beyond-range extrapolation. Original ERA5 NetCDF files and Tianditu administrative-boundary files are not redistributed; source information, acquisition scripts, and processing instructions are provided. Processed data, numerical tables, metadata, and documentation are licensed under CC BY 4.0. Original analysis code is licensed under the MIT License. Third-party source data remain subject to the terms of their original providers.