AERSI — Air Exposure Severity Index
AERSI is a station-level, daily-updating air quality severity index for India. It complements standard AQI — which reports a single day's conditions — by combining WHO-normalized pollution load, 30-day exposure persistence, and day-to-day volatility into one composite severity score across CPCB monitoring stations nationwide. AERSI is live at aersi.live.
Formula
AERSI = PL^0.50 × EPF^0.25 × VSF^0.25
- PL — Pollution Load: WHO-normalized, soft-saturated, weight-renormalized pollutant concentration, computed from the 30-day rolling window mean per pollutant.
- EPF — Exposure Persistence Factor: how often AQI exceeded 100 over the observed rolling window, dampened by data coverage.
- VSF — Variability Severity Factor: median absolute day-to-day AQI change over consecutive calendar-day pairs, tanh-bounded.
Pollutant weights (PM2.5 0.40, PM10 0.20, NO2 0.15, Ozone 0.15, SO2 0.10) are derived from India-specific attributable DALYs in the Global Burden of Disease Study 2019 for PM2.5 and ozone, with the remaining pollutants estimated from global comparative-risk literature and renormalized. Full derivation in methodology.html.
Live System
- 543 stations scored daily across India
- Automated pipeline via GitHub Actions — updated 10:30 AM IST
- Interactive map, station explorer, and full methodology reference at aersi.live
Data Source
CPCB via data.gov.in (Resource ID: 3b01bcb8-0b14-4abf-b6f2-c1bfd384ba69)
Rolling 30-day window · ~530 active monitoring stations
Validation
- Formula reconstruction error < 0.001 across all stations
- Spearman(AERSI, latest-day AQI) = 0.719 — genuinely complementary, not redundant
- Exponent sensitivity: r = 0.993–0.999 across alpha 0.4–0.8
- EPF threshold sensitivity: r = 0.987–0.990 across WHO and CPCB thresholds
- Geographic validation: NCR belt (1.79) consistently above southern states (1.08)
Repository Structure
aersi-station-timeseries/
├── data/
│ ├── snapshots/ raw daily CPCB CSVs (dated, immutable)
│ ├── rolling/ merged 30-day rolling dataset
│ └── processed/ final scored output (aersi_station_scores.csv)
├── src/
│ ├── fetch/ Step 1 — CPCB API ingestion
│ ├── process/ Steps 2–4 — rolling window, AQI, AERSI computation
│ ├── map/ Step 5 — interactive map generation
│ ├── analysis/ sensitivity analysis and regression test suite
│ └── pipeline.py orchestrator — runs all steps in sequence
├── outputs/ generated interactive map (aersi_map.html)
├── css/, js/ frontend styles and shared layout/logic
├── *.html website pages (index, map, explore, why,
│ methodology, about, privacy)
└── .github/workflows/ daily pipeline CI/CD
License
- Code: MIT License
- Dataset (
data/processed/): CC BY 4.0
- Raw CPCB data: GODL-India
See LICENSE for full terms.
Citation
If you use this code or dataset in your research, please cite:
Dataset:
Chakrabarty D. (2026). AERSI — Air Exposure Severity Index Dataset v1.0 (May 24 – August 23, 2026). Zenodo. https://doi.org/10.5281/zenodo.22065550
Survey paper:
Marne P.M., Bhosale S.N., Chakrabarty D. (2026). Air Quality Index in the Era of Data Science: A Survey of Methods, Technologies and Exposure Trends. IEEE INDIACom 2026. DOI: 10.23919/INDIACom70271.2026.11526638
