How Shabbat Cleans Jerusalem's Air (For One Day A Week)

Air quality is a silent killer - as is noise pollution.

In Jerusalem, we're saturated in both.

Israel has average levels of air pollution that are concerningly considerably worse than those encounterd throughout much of Europe.

Even on a good day, average levels of PM2.5 and PM10 bear more resemblance to parts of industrialised China than to, say, London or New York.

Thankfully, there is a blissful weekly reprieve built into the Jewish calendar: Shabbat when traditionally observant Jews refrain from using electricity, motoring, and engaging in other Biblically-prohibited activities. The incessant construction (to build the light rail, to build up high rises) also grinds to a temporary halt. Public transport stops running.

The deluge of PM2.5 and PM10 that otherwise gets trapped in this topographic ridge on the edge of the desert suddenly stops being ommitted.

I've often paused for thought on Shabbat and wondered how much more peaceful our streets would be, and cleaner our air, if this artificial weekly calm happened ... well, every day. Cities around the world have experimented with measures designed to curb nuisance honking and keep polluting vehicles out of city centers and shown sizeable reductions in circulating small particles - linked, beyond doubt, to countless diseases.

I figured that Jerusalem might just make the perfect case study for analyzing, even at first glance, the correlation between cars, road activity, construction, and quality of air.

Method

To test my hypothesis, I put my monthly Claude subscription to even harder work - starting by downloading seven million data points from the Ministry of the Environment (is it scraping if your tax payers are paying for the data collection!?).

Then, the air quality readings from the Ministry of the Environment were marked with Booleans: is_shabbat and is_yom_tov and is_yom_kippur to test for the most striking day of the Israeli and Jewish calendar when cars are almost entirely absent from roads.

Then, I gathered data for the same time period from London and New York. I chose these cities as controls because the ratio of work-day to work-week activity could be made comparative by mapping such that the extra decrease in activity brought about by the Shabbat strictures, absent in those capitals, could be the delta between the two. For anyone motivated to take this analysis further, or test against other cities, the excellent World Air Quality Index aggregates governmental efforts around the world, among other sources.

To attempt to compare like with like I defined a few time periods using Heb Cal API to retrieve historical Shabbat times:

• The pre-Shabbat window: two hours before Shabbat.

• Shabbat and hagim (chol ha'moed was excluded because it does not have the same strictures)

• Havdala taken, for this purpose, to be the two hour window after Shabbat

Based upon that I defined as time series:

• Israeli workweek: Sunday to Thursday
• Israeli non-Shabbat weekend: Friday through to start of Shabbat

The ratio of these two was then compared with what I defined as the comparative measurive in non-Shabbat-keeping geographies: a classic Monday to Friday work week compared against the weekend period (Saturday and Sunday) as an aggregate.

Segmenting the Israeli weekend into its Shabbat and non-Shabbat elements allows for further testing of the Shabbat variable.

• Dataset: danielrosehill/Jerusalem-Air-Quality-Shabbat
• Findings (interactive): Space — Jerusalem-Air-Quality-Shabbat-Findings
• Analysis code: github.com/danielrosehill/JLM-Air-Quality-Analysis

Why Jerusalem is unusual

Jerusalem has an unusually strong weekly traffic cycle. From candle-lighting on Friday evening to tzeit hakochavim on Saturday night, large parts of the city — particularly Shabbat-observant neighbourhoods — empty of cars almost completely. The same pattern, more universally, applies to the biblical chagim on which driving is prohibited, and most starkly to Yom Kippur, when traffic effectively ceases nationwide, including among the non-observant.

This makes Jerusalem a near-textbook natural experiment for one specific question: how much of urban air pollution is anthropogenic, and how much of that is traffic?

If the hypothesis is right, the halachic calendar should produce a periodic, predictable dip in traffic-derived pollutants (NO, NO₂, NOx) that exceeds the generic "weekend dip" seen in any city.

Methodology

Data

• Jerusalem: 12 air-quality monitoring stations, hourly cadence, sourced from the Israeli Ministry of Environmental Protection (air.sviva.gov.il).
• London: 3 OpenAQ stations (NO₂, O₃, PM10, PM2.5, SO₂; CO at 2 of 3).
• New York: 4 OpenAQ stations (NO₂, NOx, NO, CO, O₃, PM10, PM2.5, SO₂).
• Window: 2025-04-28 → 2026-04-28 — 50 Shabbatot, 9 Yom Tov windows, 1 Yom Kippur.

Halachic time-window labeling

The source dataset's coarse calendar flags (is_friday, is_shabbat derived from day-of-week) are superseded by exact halachic windows pulled from Hebcal, using Jerusalem coordinates and customs: 40-minute candle-lighting offset, tzeit hakochavim for end-of-Shabbat andm end-of-chag.

scripts/relabel.py walks the Hebcal candle-lighting and havdalah events, pairs them into closed windows, classifies each window by memo, and writes the labeled parquet with these flags:

A derived segment column resolves these into a single mutually-exclusive bucket (weekday, pre_shabbat, shabbat, post_havdalah, yom_tov, yom_kippur) for clean grouping. Chol HaMoed is intentionally not flagged — driving is permitted on those days, so they are indistinguishable from weekdays for this hypothesis.

Cross-city segmentation

Each city uses its own off-day convention — that's the apples-to-apples comparison ("what happens when residents are on their conventional rest days?"):

NO₂ values were converted from ppb (Jerusalem source) to µg/m³ (OpenAQ convention, 1 ppb ≈ 1.88 µg/m³ at 25 °C, 1 atm) for the overlay charts.

Aggregates

For a given pollutant $p$, station $s$, and segment $g \in {\text{workweek}, \text{off-day}}$, the per-station segment median is

$$m_{p,s,g}=\mathrm{median}(\{x_{p,s,t}:t\in g\})m\_\{p,s,g\}\; =\; \backslash operatorname\{median\}\backslash bigl(\backslash \{x\_\{p,s,t\}\; :\; t\; \backslash in\; g\backslash \}\backslash bigr)$$

where $x_{p,s,t}$ is the hourly reading at station $s$ and timestamp $t$. The city-level segment value pools across stations:

$$M_{p,g}={\mathrm{median}}_{s\in \mathcal{S}}\left(m_{p,s,g}\right)M\_\{p,g\}\; =\; \backslash operatorname\{median\}\_\{s\; \backslash in\; \backslash mathcal\{S\}\}\backslash bigl(m\_\{p,s,g\}\backslash bigr)$$

The headline off-day vs workweek change is then

$${\mathrm{\Delta }}_p=\frac{M_{p,\text{off}}-M_{p,\text{work}}}{M_{p,\text{work}}}\times 100\mathrm{\%}\backslash Delta\_p\; \backslash ;=\backslash ;\; \backslash frac\{M\_\{p,\backslash text\{off\}\}\; -\; M\_\{p,\backslash text\{work\}\}\}\{M\_\{p,\backslash text\{work\}\}\}\; \backslash times\; 100\backslash \%$$

with the segment definitions per city as in the table above — for Jerusalem, $\text{off} = \text{shabbat}$ over the exact halachic candle-lighting → tzeit window; for London and NYC, $\text{off} = {\text{Sat}, \text{Sun}}$. Yom Tov and Yom Kippur are reported as separate segments rather than folded into the Jerusalem off-day set, so they don't inflate the headline Shabbat number.

Headline result

NO₂ off-day vs workweek, median across stations:

Jerusalem's Shabbat NO₂ drop is roughly 2× New York's weekend drop and 3× London's. The same direction holds for NO and NOx; PM2.5 is much weaker and noisier; ozone moves the opposite way (it rises slightly on off-days everywhere, consistent with reduced NOx scavenging).

Stratified by pollutant

A few things stand out:

• NOx-family (NO, NO₂, NOx) moves as the hypothesis predicts. Jerusalem's drop is roughly twice NYC's and three times London's. This is the central evidence.
• Ozone rises on off-days in all three cities by 5–8%. NOx-saturated urban photochemistry: less NO means less O₃ scavenging at ground level. The fact that Jerusalem's O₃ bump is in the same range as the controls despite a far larger NOx drop is a clean signature of non-linear ozone production.
• PM2.5 is weak everywhere. Fine particulates here are dominated by long-range transport (Saharan dust, regional smoke, secondary aerosol) on day-to-week timescales — largely insensitive to a 25-hour traffic suspension. Jerusalem manages a small −7%, London and New York actually go up slightly on weekends.
• PM10 does respond in Jerusalem (−16%) — coarser primary particles (road dust, brake/tyre wear) track traffic more directly than secondary PM2.5.
• CO is at-or-near floor in Jerusalem and NYC (catalytic converters); not much signal left to lose.
• SO₂ is only reported by London stations and shows a −45% weekend drop — likely commercial/industrial rather than passenger traffic.

What the Jerusalem-only panels show

The "Saturday hole"

A day-of-week × hour-of-day heatmap of NOx, pooled across all 12 stations.

The Shabbat block is visible without any statistical machinery.

Yom Kippur — the cleanest cut in the calendar

±36 hours bracketing Yom Kippur 2025. NOx and NO₂ collapse to a baseline floor for the duration of the fast — a near-complete cessation of combustion-source activity that would be hard to engineer experimentally.

Friday is intermediate, and the signal is in the afternoon

Friday morning behaves like a weekday; the pre-Shabbat afternoon is where the curve bends. The 2-hour pre_shabbat segment captures a real phenomenon — the erev-Shabbat shopping rush followed by a sharp drop into candle-lighting.

The PM paradox

NOx falls sharply on Shabbat; PM2.5 doesn't, and on a calendar-day resolution often rises. Useful pedagogically: it shows what kind of pollutant a Shabbat-style intervention can and can't move.

Follow-up: PM2.5 + PM10 combined

PM2.5 alone is the well-known paradox case — fine particulate often moves the "wrong" way on Shabbat because long-range dust and secondary aerosol overwhelm any local traffic signal. Adding PM10 brings the coarser, more locally-sourced fraction (road dust, brake / tyre wear, construction) into the picture.

Combined PM total (PM2.5 + PM10), daytime 07–22, station-hours where both species were observed:

Two readings worth noting:

• The drop on Shabbat is real once PM10 is included — −13%, vs the much noisier PM2.5-only signal. Coarse particulate from local sources tracks the traffic-and-construction lull that PM2.5 partly hides.
• Yom Kippur is in a different league — combined PM falls by roughly three quarters, a full step below any other segment in the calendar. Same city, same stations, same season; just nobody driving.

Follow-up: every pollutant correlated with is_shabbat

The headline NO₂ result is the strongest of a family. Running every non-meteorological pollutant against is_shabbat, with Cohen's d computed on values residualised by hour-of-day to strip out the diurnal cycle:

A clean split:

• Strongest negative effects are all combustion / vehicle-exhaust markers — NO₂, NOx, NO, CO, and the BTEX aromatics (toluene, benzene). These are the species you would expect a city-wide traffic suspension to move, and they all move in the right direction with effect sizes between 0.18 and 0.52.
• The two pollutants that go up on Shabbat are ozone and PM2.5, and both have well-known reasons. Ozone in NOx-saturated urban air is titrated by NO; remove the NO and ground-level O₃ rises. PM2.5 is dominated by long-range transport (regional dust, secondary aerosol) on weekly timescales and is decoupled from local emissions — the small positive coefficient is the dust-event lottery, not a Shabbat effect.

The point-biserial correlations have astronomical sample sizes (>10⁶ for NO₂), so significance is uninformative; the d column is the one to read.

scripts/extra_analyses.py produces the full table at data/shabbat_correlations.csv (also includes Mann-Whitney p values for robustness).

Follow-up: Jerusalem vs averaged Western controls, per pollutant

The cross-city headline number averages New York and London into a single "Western" comparator (each city contributes equal weight to the % drop). Per-pollutant:

NO₂ is the cleanest contrast — Jerusalem's drop is roughly two and a half times the averaged Western weekend drop. CO moves modestly more in Jerusalem than the controls. Ozone rises roughly equally in all three cities (the photochemistry is the same; the magnitude of the underlying NOx drop is different but the O₃ response is similar in % terms).

PM is the messy case. PM2.5 actually rises more on Shabbat than on the Western weekend — partly the Saharan-dust lottery already noted, partly that PM2.5's dominant sources are not the ones Shabbat suppresses. PM10 in Jerusalem also rises slightly on Shabbat at the daily-mean level, while London's PM10 falls on weekends; the discrepancy is consistent with Jerusalem's coarse particulate being driven heavily by region-scale dust events whose timing is uncorrelated with the Jewish calendar.

(Unit caveat: London control values for non-particulate species are µg/m³, NYC's are ppm, Jerusalem's are ppb. The % change within each city is unit-invariant, so the side-by-side bars compare like with like even though the absolute values aren't directly comparable.)

Follow-up: why standard AQIs miss the Shabbat effect

The natural way to summarise air quality in one number is a composite index. I computed two of the major standards on the full 12-month window:

• US EPA AQI — the most widely cited. Max-of-sub-indices across PM2.5, PM10, O₃, NO₂, CO, with each pollutant having its own breakpoint table (PM2.5 and PM10 24-hour averaged, O₃ and CO 8-hour, NO₂ 1-hour). Designed for public-health communication: "what's the worst pollutant right now?"
• Canadian AQHI — the alternative philosophy. A linear additive index of NO₂, O₃, and PM2.5 (3-hour averaged), weighted by the epidemiological mortality coefficients from the Canadian cohort studies. Designed for health-risk communication: "how much excess mortality risk are you exposed to right now?" Because it's additive and includes NO₂, it should — in principle — respond to a traffic intervention even when PM doesn't move.

Then I added two custom activity-isolating indices to peel apart what Shabbat does and doesn't target:

• TCI (Traffic-Combustion Index): mean z-score of NO₂, NO, NOx, CO, benzene, toluene against the per-station workweek baseline.
• PD-Index (Photochemical/Dust): mean z-score of O₃ + PM2.5 — the "wrong-way" pollutants that go up on Shabbat.

By halachic segment, mean across all available station-hours:

The story the indices tell, in order of usefulness for this question:

• EPA AQI barely moves on Shabbat (64 → 67). It's structurally dominated by particulate matter, and PM in Jerusalem is dust-driven on weekly timescales. A citizen looking at their AQI app would never see the Shabbat traffic intervention.
• Canadian AQHI moves a little, but not much (4.7 → 4.4, a ~7% drop). Even though it's additive and includes NO₂, the PM2.5 term dominates the sum because Jerusalem's baseline PM2.5 is high enough that e^(0.000487·PM2.5) is much larger than the NO₂ term. AQHI is a better-designed index than EPA AQI for this question, but it's still smothered by the Jerusalem dust signal.
• TCI plunges by half a standard deviation on Shabbat (−0.10 → −0.61), and almost a full standard deviation on Yom Kippur (−0.90). That's the intervention the data is actually showing.
• PD-Index drifts the other way on Shabbat (−0.05 → +0.15) — the photochemistry and dust signal that isn't what Shabbat targets.

Yom Kippur is the only segment where every index agrees: EPA AQI 64→28, AQHI 4.7→3.0, TCI to floor, PD-Index also negative. That's the day when the dust lottery happened to coincide with the most complete emissions stop in the calendar — the only day all four lenses converge on the same story.

Looking at which pollutant drives EPA AQI on each station-hour sharpens the point further:

PM2.5 already dominates the EPA AQI on three quarters of weekday station-hours, and on Shabbat its share rises, because the only sub-index falling sharply is NO₂. The composite is structurally insensitive to a Shabbat-style intervention; you have to look beneath it to see the effect.

Why this matters beyond Shabbat. Standard AQIs are calibrated for typical-Western-city pollutant profiles, where traffic NOx and PM2.5 move together and PM2.5 sources are mostly local. In a city where PM2.5 is dominated by long-range dust transport, every standard composite becomes a "dust meter" that's blind to local emissions interventions — traffic days, low-emission zones, congestion charges, fleet electrification. The right way to evaluate such interventions is with an activity-attributable index like TCI, not a public-health composite.

Interpretation: what's probably driving each pollutant

For every pollutant that appears in both Jerusalem and at least one control city, here is a plain-language read of the likely / probable / possible causal stories — both for the source mix and for the Shabbat delta. This section is interpretive: the data is consistent with these mechanisms but doesn't prove any of them on its own.

Nitrogen dioxide (NO₂)

• Likely: Tailpipe exhaust from petrol and (especially) diesel road traffic is the dominant urban source of NO₂. Jerusalem's −61% Shabbat drop, vs −24% averaged for the Western controls, is consistent with a much steeper traffic suspension than a normal Western weekend. Buses, taxis, and private cars all stop in Shabbat-observant areas; the public transport network largely shuts down.
• Probable: Construction equipment and the active light-rail extension works also pause on Shabbat, removing a second diesel source that doesn't pause on a London or NYC Saturday.
• Possible: Some of the gap is meteorological — Shabbat sample is 50 Saturdays in one weather year and inversion / wind patterns aren't controlled for. A meteorology-residualised version is the obvious next step before claiming the −61% number is purely anthropogenic.

Nitric oxide (NO) and nitrogen oxides (NOx)

• Likely: Same combustion sources as NO₂. NO is the more direct primary emission and converts to NO₂ in the atmosphere within minutes.
• Probable: Cold starts are a disproportionate share of NO emissions. Friday's morning rush still happens (NO peaks then) but Shabbat morning has effectively no cold starts — explaining why NO drops even more steeply than NO₂ (−78% vs −51%).
• Possible: A small fraction is from gas-fired water heaters and cooking; these largely keep running on Shabbat (timers, plata) so the residual non-zero Shabbat NOx is partly stationary combustion.

Ozone (O₃) — the "wrong-way" pollutant

• Likely: Ground-level O₃ in NOx-rich urban air is titrated by NO: NO + O₃ → NO₂ + O₂. Drop the NO and ground-level O₃ rises. Every city in the panel goes up on its off-day by 4–8%, regardless of the magnitude of the underlying NOx drop, which is the textbook signature of NOx-saturated photochemistry.
• Probable: Jerusalem's O₃ rise (+5%) being similar to NYC and London despite a far larger NOx drop indicates the urban centre sits well into the NOx-saturated regime — a regime where reducing NOx actually increases local ozone before eventually flipping to NOx-limited at much lower NOx levels.
• Possible: Some of the residual O₃ is regional / transported from upwind areas; the diurnal photochemistry in spring and summer is particularly active over the Judean hills.

Carbon monoxide (CO)

• Likely: CO is overwhelmingly a vehicle-exhaust pollutant in modern cities, but post-catalytic-converter fleets emit very little of it. CO at Jerusalem and NYC stations is already near the noise floor — there isn't much signal left to lose. Hence Jerusalem's modest −13% drop on Shabbat and the equally modest −9% Western-average weekend drop.
• Probable: Cold-start enrichment dominates the remaining CO signal; Shabbat removes the morning-cold-start cohort, hence the small but consistent negative.
• Possible: London's off-day CO drop being slightly larger than NYC's may reflect London's older diesel-heavy bus fleet; without per-station vehicle-mix data this is suggestive only.

Fine particulates (PM2.5)

• Likely: PM2.5 in Jerusalem is dominated on weekly timescales by long-range transport — Saharan and Arabian dust intrusions, regional smoke, and secondary aerosol formed from precursors emitted hundreds of kilometres away. The dominant local sources (residential heating in winter, some traffic) are a smaller fraction of the daily mean than regional inputs.
• Probable: This is why PM2.5 paradoxically rises on Shabbat in Jerusalem (+20%) — the dust lottery is uncorrelated with the Jewish calendar, and 50 Saturdays is too small a sample for the regional contribution to wash out. London and NYC weekend PM2.5 is essentially flat, consistent with their PM2.5 being more locally sourced and lightly traffic-coupled.
• Possible: Some of the Shabbat rise is real — gas-cooktop use during Friday-night and Saturday-lunch meals, and plata / blech operation, contribute primary particles indoors that bleed outdoors. Hard to separate from dust without isotopic or chemical-speciation data.

Coarse particulates (PM10)

• Likely: PM10 mixes a regional-dust component (same Saharan intrusions) with a much larger local primary component than PM2.5 — road dust, brake and tyre wear, construction-site dust, and unpaved edges. The traffic-and-construction lull on Shabbat should reduce the local fraction.
• Probable: At hourly resolution the Shabbat traffic suspension does show up — it's why combined PM2.5+PM10 falls 13% on Shabbat and 25% in the post-Havdalah lull, and why PM10 falls 13% on London weekends. The daily-mean PM10 in Jerusalem still goes slightly up on Shabbat (+12%) because the dust component dominates the daily average even when the hourly-resolution local component clearly drops.
• Possible: Construction is the under-appreciated contributor here. Several active large works (light-rail extensions, high-rise) sit close to several stations; their full pause on Shabbat is plausibly half the PM10 effect, with road dust the other half.

Why the cleanest natural experiment is Yom Kippur

Yom Kippur is the only segment where everyone, including the non-observant, stops driving — no internal control population, full nationwide compliance for ~25 hours. The PM-total drop of −74% on YK is larger than the entire dynamic range of the Shabbat experiment, and the NOx collapse to a near-floor value is, in effect, what a complete urban traffic ban looks like. It's a one-day-a-year glimpse of the maximal effect, and consistent with all the other pollutant-specific stories above.

The headline number: Jerusalem amplifies the weekend NO₂ drop by ~4×

The most intellectually honest way to read this whole study is not "the NO₂ drop on Shabbat is X%." It's a ratio: how much larger is the Jerusalem off-day drop than the averaged London + NYC weekend drop, after weather is controlled? Every city has a weekend cycle. The interesting quantity is the delta — what Shabbat does that a Western weekend doesn't.

After meteorological deweathering (which I'll describe in the next section), the answer is:

NO₂ at ~4× is the clean, defensible headline because both magnitudes are large. The PM ratios are inflated by tiny Western denominators (their weekends barely move PM at all) — so they're true in direction but not stable as a quantitative claim. CO ratio is ~1× because catalytic converters have pushed CO toward the floor in all three cities, leaving little signal to amplify. O₃ goes the other way — about which more below.

Interpretation: why the ratio is ~4×

The 4× number isn't really "Jerusalem has a stronger weekend cycle." It's better described as Shabbat being closer to a full traffic-plus- construction ban than a Western weekend is.

The Shabbat intervention removes essentially the entire urban NO₂ inventory at once:

• Private cars — sharp reduction city-wide, near-total in observant neighbourhoods.
• Diesel buses — Egged, Superbus, Dan, intercity coaches all stop. Buses are typically 15–25% of urban NO₂; in Jerusalem this whole category goes to zero.
• Light-rail and high-rise construction — Jerusalem is in an unusually active build-out phase. The Green and Blue Line works run diesel excavators, drills, and generators six days a week, plus dozens of active high-rise sites; all silent on Shabbat. Non-road mobile machinery (NRMM) is a known top-3 NO₂ source in growing cities.
• Trucks and freight delivery — near-zero on Shabbat.
• Sherut / shared taxis — stop in observant areas.

A typical Western weekend removes only part of that. London weekend traffic is ~25–30% lower (not zero), buses run reduced schedules, central- zone construction is restricted by planning rules but suburban and infrastructure works continue, and domestic gas heating + Heathrow aviation are unchanged. New York is similar but with one significant difference: building heating with #4/#6 fuel oil is uniquely large in NYC's NO₂ inventory and runs continuously regardless of the day of the week. JFK/LGA aviation, peaker power generation, and the city's bus service are all unchanged on weekends.

So the comparison is:

• Jerusalem Shabbat: commuter traffic + diesel buses + freight + construction — all suspended simultaneously.
• Western weekend: commuter traffic + some freight — and that's it.

A factor of ~4 between those two interventions is what a back-of-envelope source apportionment would predict, and it's what the deweathered data shows. The number isn't a coincidence; it's a direct consequence of how much broader the Shabbat shutdown is than a normal weekend.

Why O₃ amplification is less than 1

Ozone moves the opposite way to traffic NO₂: ground-level O₃ rises slightly on every off-day in every city in the panel, by 1–8%. The mechanism is NOx titration: in NOx-rich urban air, NO scavenges O₃ via NO + O₃ → NO₂ + O₂. Reduce NO and ground-level O₃ rises.

The interesting thing is that Jerusalem's O₃ bump is smaller than the Western weekend O₃ bump despite a much larger NOx drop (+1% vs +2–3%). This is the signature of NOx-saturated photochemistry: in this regime, a deeper NOx cut doesn't produce a proportionally larger O₃ rise because the chemistry flips into a different regime once NOx falls below a threshold. The Shabbat NOx cut is large enough to start doing that.

What the ratio predicts about Yom Kippur

Yom Kippur removes the residual that Shabbat doesn't fully reach: through-traffic on Highway 1, secular West Jerusalem traffic, and East Jerusalem traffic that doesn't observe the Jewish calendar. The amplification ratio for Yom Kippur should therefore be larger than for Shabbat — and that's what the data shows. Combined PM falls 74% on YK, NOx collapses to a near-floor, and the composite EPA AQI itself drops to 28 (Good band) which never happens on a regular Shabbat. Yom Kippur is the natural extension of the same source-removal logic with the final few categories peeled away.

Follow-up: how much of this is just weather?

The single biggest caveat on every previous number in this post is meteorology. 50 Saturdays in one weather year is a small sample, and if those 50 Saturdays happened to coincide with windy weather or low boundary-layer heights, what looks like a Shabbat effect could be a weather effect. The honest answer requires controlling for it.

I pulled hourly weather (temperature, relative humidity, wind speed, wind direction, surface pressure, boundary-layer height, precipitation) from Open-Meteo's Historical Weather API (ERA5-backed reanalysis) for Jerusalem, London, and New York across the full 12-month window. Then I fit a Random Forest per pollutant per city — this is a Python port of the openair R package's deweather::buildMod workflow.

The model includes weather covariates, station ID, and a seasonal day-of-year cycle. Hour-of-day and day-of-week are deliberately not in the feature set: keeping them would let the model absorb the diurnal traffic cycle and the Shabbat / weekend effect along with weather, which is exactly what we want to measure in the residuals. The deweathered series is value − RF_predicted + mean(value) — same overall mean as the raw signal, but with the variance the model could explain from meteorology stripped out.

R² on the meteorology+season model ranges from 0.69 (NO) to 0.93 (CO, O₃) for Jerusalem, and 0.79–0.98 for the controls.

What actually survives weather control

Three things to take away:

• Roughly half of the raw NO₂ drop is real, half is weather. The Shabbat effect on combustion-traffic emissions survives weather control with a clean step-change of about a third — still very large, still the dominant signal, but smaller than the raw 59% number suggested.
• Both particulate paradoxes dissolve once weather is controlled. PM2.5 going up on Shabbat (+28% raw) was almost entirely the Saharan-dust lottery: those 50 Saturdays happened to coincide with more dust. After deweathering it falls slightly (−4%), in line with the traffic story. PM10 swings even further — from +11% raw to −12% deweathered. The "PM doesn't care about Shabbat" story is partially wrong; what's really true is "PM is dominated by dust on weekly timescales, and once you remove the dust the Shabbat traffic signal is recoverable."
• Ozone barely moves either way. The raw +3% becomes a deweathered +1%. The NOx-titration mechanism predicts a small rise, and that's what survives.

Cross-city, after weather control

Jerusalem's deweathered NO₂ drop is about four times the deweathered weekend drop in either control city. This is the cleanest version of the headline finding: the gap isn't a weather artefact, and it isn't just a generic weekend cycle that happens harder in Jerusalem. Once the weather is stripped out of all three cities, Jerusalem's traffic suspension still stands apart by a factor of four.

PM2.5 and O₃ are essentially flat across all three cities after deweathering — confirming that the apparent cross-city differences in these species were noise.

The deweathering layer is now part of the dataset: see data/weather_hourly.parquet for the meteorology and data/jlm_deweathered.parquet / data/control_deweathered.parquet for the residualised pollutant series.

Caveats

• 12 months is one weather year. The 50 Shabbatot in the sample are bound to whatever meteorology and dust events happened on those Saturdays. Spring 2026 Saharan dust intrusions inflate PM2.5 on a handful of them.
• Meteorological residualisation is now done (figures 15–16) and reduces the headline NO₂ Shabbat drop from −59% to −34%. The deweathered numbers should be treated as the load-bearing version of every quantitative claim in this post; the raw numbers are kept above for transparency about how much weather contributes.
• Heuristic windows. The pre-Shabbat 2h and post-havdalah 2h windows probably vary by season and neighbourhood; a fixed offset is a simplification.
• Internal control. East Jerusalem stations (notably Rockefeller Museum / station 568) carry traffic that does not observe Shabbat — they show the smallest drops, which is itself a useful sanity check on the mechanism.

How to reproduce

# 1. Halachic relabel (reads source parquet + Hebcal JSON)
python scripts/relabel.py

# 2. Build London + NYC tidy parquet from OpenAQ JSONs
python scripts/build_controls.py

# 3. Generate all figures + the cross-city CSV
python scripts/generate_figures.py

Dependencies: pandas, pyarrow, numpy, matplotlib. No statsmodels or seaborn — kept deliberately light.

Sources & licensing

• Air quality (Jerusalem): Israeli Ministry of Environmental Protection — air.sviva.gov.il
• Air quality (London / New York): OpenAQ
• Halachic times: Hebcal — Jerusalem coords, 40-min candle-lighting offset, tzeit hakochavim for havdalah

Analysis code: MIT. Figures and labeled parquet: CC-BY-4.0, matching the upstream dataset. Underlying air-quality measurements remain attributable to the original sources above.

---

All Graphics

---

---

---

---

---

---

---

---

---

---

---

---

If you build on this — meteorology-residualised version, expansion to more control cities, or per-neighbourhood breakdown by station — I'd love to see it. PRs and discussion welcome on the Space and GitHub repo.