neuralsorcerer/air-quality

Air Quality & Meteorology Dataset

Dataset Description

This corpus contains 87 672 hourly records (10 variables + timestamp) that realistically emulate air‑quality and local‑weather conditions for Kolkata, West Bengal, India.
Patterns, trends and extreme events (Diwali fireworks, COVID‑19 lockdown, cyclones, heat‑waves) are calibrated to published CPCB, IMD and peer‑reviewed summaries, making the data suitable for benchmarking, forecasting, policy‑impact simulations and educational research.

The data are suitable for time‑series forecasting, machine learning, environmental research, and air‑quality policy simulation while containing no real personal or proprietary information.

---

File Information

File	Records	Approx. Size
air_quality.csv	87 672 (hourly)	~ 16 MB

(Rows = 10 years × 365 days (+ leap) × 24 h ≈ 87.7 k)

---

Columns & Descriptions

Column	Unit / Range	Description
datetime	ISO 8601 (IST)	Hour start timestamp (UTC + 05:30).
Pollutants
pm25	µg m⁻³ (15–600)	Particulate Matter < 2.5 µm. Seasonally highest in winter; Diwali + COVID effects embedded.
pm10	µg m⁻³ (30–900)	Particulate Matter < 10 µm (≈ 1.8 × PM₂.₅).
no2	µg m⁻³ (5–80)	Nitrogen dioxide, traffic proxy. Morning/evening rush‑hour peaks.
co	mg m⁻³ (0.05–4)	Carbon monoxide from incomplete combustion.
so2	µg m⁻³ (1–20)	Sulphur dioxide, low in Kolkata; slight fireworks/cyclone dips.
o3	µg m⁻³ (5–120)	Surface ozone; midday photochemical peaks and decade‑long upward trend.
Meteorology
temp	°C (12–45)	Dry‑bulb air temperature. Heat‑wave days reach > 41 °C.
rh	% (20–100)	Relative humidity. Near‑saturation in monsoon nights.
wind	m s⁻¹ (0.1–30)	10 m wind speed. Cyclones push gusts > 20 m s⁻¹.
rain	mm h⁻¹ (0–150)	Hourly precipitation. Heavy bursts during monsoon and cyclones.

---

Usage

This dataset is ideal for:

• Environmental Research – Analyse seasonal/diurnal pollution dynamics or meteorological drivers.
• Machine‑Learning Benchmarks – Train forecasting, anomaly‑detection or imputation models on multivariate, event‑rich time‑series.
• Policy / “What‑If” Simulation – evaluate NCAP / BS‑VI scenarios without privacy constraints.
• Extreme‑Event Studies – Stress‑test models on Diwali spikes, cyclone wash‑outs, or heat‑wave ozone episodes.
• Teaching & Exploration – Demonstrate correlation analysis, time‑series decomposition, ARIMA/LSTM modelling, etc.

---

Example Workflows

Task	Example
Forecasting	Predict next‑day PM₂.₅ using past 72 h pollutants + weather.
Classification	Flag hours exceeding national PM₂.₅ 24‑h standard (60 µg m⁻³).
Clustering	Segment days by pollution profile (clean‑monsoon, moderate, Diwali‑spike).
Causal Inference	Quantify lockdown‑attributable reduction in NO₂ via difference‑in‑differences.
Fairness	Compare model accuracy across seasons or meteorological regimes.

---

Data Pre‑processing Tips

• Timestamp Indexing – Set datetime as Pandas DateTimeIndex for resampling (df.resample('D').mean()), lag features, etc.
• Log/Box‑Cox – Stabilise heavy‑tailed pollutant distributions before regression.
• Feature Engineering – Add sine/cosine hour‑of‑day and month‑of‑year terms to capture periodicity.
• Outlier Handling – Retain extreme Diwali and cyclone values for robustness tests rather than trimming.
• Train/Test Split – Use rolling or time‑based splits (e.g., train 2015‑21, validate 2022, test 2023‑24) to respect temporal order.

---

License

Released under the Creative Commons CC0 1.0 Universal license – free for research, educational, or commercial use.
The data are algorithmically generated and do not contain any real measurements or personal information.

---

References

1. Guttikunda, S. K., & Jawahar, P. (2020). “Exceedances and trends of particulate matter (PM₂.₅) in five Indian megacities.” Atmospheric Environment, 222, 117125.
2. Centre for Science and Environment (CSE). (2016). Night-time air turns toxic in Kolkata winter: Rapid assessment briefing note. New Delhi: CSE.
3. IQAir. (2020). World Air Quality Report 2019 – City profile: Kolkata, India.
4. Biswas, T., Saha, D., et al. (2022). “Strict lockdown measures reduced PM₂.₅ concentrations during the COVID-19 pandemic in Kolkata, India.” Sustainable Water Resources Management, 8 (2), 45.
5. Times of India. (28 Oct 2019). “Diwali brings Kolkata AQI down from ‘moderate’ to ‘poor’.”
6. Chaudhari, S., & Sudhakar, S. (2013). “Ambient air quality during Diwali festival over an Indian metro.” Aerosol and Air Quality Research, 13 (4), 1133–1144.
7. Bhattacharya, S. (2020). “West Bengal faces the brunt of Cyclone Amphan.” Mongabay-India (News article, 26 May 2020).
8. Ghosh, A., & Pal, R. (2023). “Effect of the 2022 heatwave on ambient ozone levels in Kolkata, India.” Pre-print, ResearchGate.
9. Times of India. (2 Jan 2025). “Winter 2024-25 records Kolkata’s cleanest PM₂.₅ levels since 2019.”
10. Greenpeace India. (2023). NO₂ pollution and health risks in seven major Indian cities: Status update 2022. Bengaluru: Greenpeace South Asia.
11. Banerjee, A., & Srivastava, A. K. (2016). “Trace-element composition of PM₂.₅ and PM₁₀ from Kolkata during different seasons.” Atmospheric Pollution Research, 7 (5), 885-894.
12. Mukherjee, M., & Sarkar, S. (2019). “Evaluating variability, transport and periodicity of particulate matter over Kolkata, India.” Environmental Monitoring and Assessment, 191, 541.

---

Acknowledgements

The dataset was built from public literature: CPCB reports, IMD climate normals, peer‑reviewed studies on Kolkata air quality (2013‑2024), and official event timelines.
If you use this dataset, please cite “Synthetic Kolkata Air‑Quality Time‑Series (2015–2024)” and link back to this repository.