Jerusalem Air Quality — Weekday vs Friday vs Shabbat (preliminary)
12 months of 5-minute resolution data (2025-04-28 → 2026-04-28) for 12 active
Israeli MoEP / Envista ARM stations in Jerusalem, pulled from
https://air-papi.sviva.gov.il/v1/envista/... (see
research/api-air-sviva-gov-il.md).
Daytime window = 07:00–22:00 local (Asia/Jerusalem). Friday evening through Saturday morning is collapsed into "Sat" since Shabbat-observant traffic is already off the road by then; this slightly understates the true Shabbat drop because it includes Friday morning rush in "Fri".
Each station has ~50 Fridays and ~50 Saturdays in the window (station 623 joined Aug 2025, station 635 joined Feb 2026 — small samples flagged).
Signal: traffic-derived gases drop sharply on Shabbat
Median Shabbat % drop vs Sun-Thu daytime mean, across the 12 stations:
| Pollutant | Friday % drop | Shabbat % drop |
|---|---|---|
| NO | ~40 % | ~55 % |
| NOx | ~34 % | ~52 % |
| NO₂ | ~30 % | ~47 % |
| Toluene | ~44 % | ~38 % |
| Benzene | ~25 % | ~21 % |
| CO | ~7 % | ~8 % |
The signal is consistent, monotonic (Sun-Thu > Fri > Sat for every gaseous traffic pollutant at every station except a couple of edge cases), and strongest in the most observant neighbourhoods.
Where the drop is biggest
Top stations by NOx Shabbat drop:
| Station ID | Location | NOx weekday | NOx Shabbat | % drop |
|---|---|---|---|---|
| 635 | Bezalel St (city centre, near Geula) | 34.0 | 9.0 | 73 % |
| 310 | Central Bus Station Platform 10 | 130.0 | 40.5 | 69 % |
| 458 | Devorah HaNevia St (residential) | 39.6 | 15.5 | 61 % |
| 328 | Atarot industrial zone | 38.8 | 15.8 | 59 % |
| 547 | Highway 16 / Beit interchange | 18.1 | 7.8 | 57 % |
| 5 | Bar Ilan St (arterial) | 26.4 | 11.9 | 55 % |
| 193 | Romema (ultra-Orthodox) | 14.3 | 6.8 | 52 % |
| 607 | Beit Safafa | 18.6 | 9.3 | 50 % |
| 13 | Baka (residential mixed) | 13.9 | 8.5 | 39 % |
| 36 | Safra Square (city centre) | 11.1 | 7.5 | 33 % |
| 623 | Atarot Eden (8 mo. only) | 12.1 | 9.1 | 25 % |
| 568 | Rockefeller Museum | 33.1 | 28.7 | 13 % |
Rockefeller Museum (568) is the surprise outlier — much smaller drop than neighbours. Worth checking whether the inlet sits next to a road that remains busy on Shabbat (Sultan Suleiman / Salah a-Din area carries a lot of East-Jerusalem traffic, which doesn't observe Shabbat).
Conversely, the very tight grouping at 50–70 % drops in West-Jerusalem arterial / residential stations is a strong correlate of the population's observance of vehicular Shabbat.
Friday is genuinely intermediate
Friday daytime sits roughly half-way between weekday and Shabbat for
NOx at most stations — consistent with reduced commercial activity but
ongoing pre-Shabbat shopping/errands until early afternoon.
scripts/shabbat_correlation.py also writes station-{id}-friday-split.csv
breaking Friday into AM (07–13) and PM (14–22) — the AM block looks like a
near-normal weekday; the drop concentrates in PM.
PM2.5 / PM10 do not drop on Shabbat — they often rise
Particulate matter increases on Saturdays at almost every station (-30 % to -65 % "drops", i.e. Sat is higher than weekday). Several plausible contributors, none ruled out yet:
- April–May 2026 included multiple Saharan dust intrusions; 12 months is not enough to wash that out and Saturdays happened to coincide with some of them.
- Charcoal grilling (mangal) is a well-documented Israeli Friday-afternoon / Saturday source of PM2.5 and PM10.
- Reduced wind / different boundary-layer dynamics on weekends.
This is a real finding for the project but it cuts the other way from the gaseous traffic story and deserves its own analysis (probably needs to controlling for synoptic dust days).
Caveats
- Daytime defined as 07–22 ignores the actual Shabbat onset/conclusion times, which vary 17:00–19:00 by season. A Hebcal-driven exact window would tighten the signal.
- Calendar holidays (Pesach 2025-04, Shavuot, Rosh Hashana, Yom Kippur, Sukkot, Hanukkah) are not excluded — Yom Kippur in particular is the cleanest air day of the year nationwide and inflates "weekday" cleanness for any year that includes it (Wed 2025-10-01 was YK 2025).
- 12 months × 5-min data is ~700k rows per station, ~2 GB total raw JSON; the raw is gitignored and regenerable via the API; only summary CSVs are committed.
How to reproduce
# 1. Pull raw data (mints Guest tokens, ~2 GB)
.venv/bin/python -m pip install pandas numpy # one-off
bash scripts/fetch_historical.sh # (TODO: lift the inline pull from session)
# 2. Run analysis
.venv/bin/python scripts/shabbat_correlation.py
# → writes analysis/headline-3way.csv + per-station CSVs