Beijing Air Tiantan PM2.5 Time Series Dataset

Derived hourly air-quality and meteorological time series dataset for the Tiantan station from the Beijing Multi-Site Air Quality data collection at UCI. Source: https://archive.ics.uci.edu/dataset/501/beijing+multi+site+air+quality+data. The record contains a canonical Parquet benchmark file and CSV mirror for forecasting and sensor-fault robustness evaluation. It retains the longest Tiantan segment selected by documented gap rules, contains no missing values or duplicate timestamps, and uses PM2.5 as the downstream forecast target. The Parquet and CSV files are byte-identical to those of version 2 of the earlier Kaggle record at https://www.kaggle.com/datasets/sensorfaultbench/beijing-air-tiantan-pm2-5-time-series-dataset/versions/2. The record includes a checksum manifest and Croissant 1.1 metadata with Responsible AI fields. Benchmark code and data guide: https://github.com/awindmann/SensorFault-Bench. Paper: https://neurips.cc/virtual/2026/loc/paris/poster/138912.