Daniel sent us this one — he's been digging through seven million air pollution data points from Israel's Ministry of Environmental Protection, which he pulled through some publicly accessible API that Claude Code sniffed out for him. He ran a time series comparing the workday-to-Shabbat ratio in Jerusalem against London and New York, and found something interesting: when you control for weather using a technique called deweathering, the Shabbat effect becomes more pronounced, not less. So he's asking two things. First, how does deweathering actually work when researchers are trying to isolate anthropogenic pollution from natural events? And second, to what extent are natural phenomena — Saharan dust, seasonal patterns, ozone formation — actually driving the metrics we use to benchmark air quality, independent of anything humans are doing?
This is such a good question, because it cuts against the instinct most people have. You'd think controlling for weather would wash out the Shabbat signal — make it harder to see. Instead it sharpens it. And that tells you something important about how weather and human activity interact in the data.
The weather was masking the thing we wanted to see, not creating it.
So let me start with deweathering, because the name makes it sound like you're subtracting weather, and that's not quite right. What you're actually doing is building a predictive model that learns the relationship between meteorological conditions and pollutant concentrations, and then you ask: what would the pollution have been on this day if the weather had been average?
You're not removing weather. You're normalizing it to some baseline.
The technical term is meteorological normalization. Deweathering is the name that stuck because a group at the University of Leeds published an R package called deweather about six years ago. The approach uses random forest regression or gradient boosting — you feed the model a training dataset with pollution levels and meteorological variables: wind speed, wind direction, temperature, humidity, atmospheric pressure, boundary layer height, precipitation.
Boundary layer height. That's how high the mixing layer of the atmosphere goes on a given day?
And it's hugely important. On a day with a low boundary layer, emissions get trapped near the surface and concentrations spike. With a deep mixed layer, the same emissions get diluted into a much larger volume. If you don't account for that, you might look at two days with identical traffic and think one was mysteriously dirtier. The model learns all these relationships. Then, for each day, you run the model with the actual meteorological conditions replaced by a long-term average — or you resample from the full distribution of weather many times and take the mean prediction. The difference between the raw measurement and the deweathered prediction tells you what the weather contributed.
On a day when a dust storm blows through Jerusalem and PM2.5 hits a hundred and fifty micrograms per cubic meter, the model says: given average weather, this day would have been maybe forty. The extra hundred and ten is weather-driven.
That's the idea. And the reason this matters for Daniel's Shabbat question is that weather in Jerusalem has its own weekly patterns. Not because weather cares about Shabbat, but because the urban heat island effect weakens when traffic drops, and there are subtle interactions between aerosol loading and local meteorology. If you just compare raw weekday to Shabbat concentrations, you're partly measuring the pollution reduction and partly measuring whatever meteorological conditions happened to prevail on those days. Deweathering strips that out and leaves you with the signal attributable to changes in emissions.
Which is the thing you care about if you're trying to isolate the human contribution.
And this method has been used in some really elegant studies. Researchers at the University of Birmingham applied meteorological normalization to air quality data from ten cities during the spring twenty-twenty lockdowns. They found that raw PM2.5 dropped by widely varying amounts depending on the city, but after deweathering, the emission-driven component was much more consistent. Weather had been obscuring the true lockdown signal.
The COVID natural experiment is basically a larger-scale version of what Daniel's doing with Shabbat. A known, scheduled reduction in human activity.
It's the same logic. In fact, there's a paper specifically on Shabbat air quality in Israel — researchers from the Technion and Hebrew University published it a few years ago. They looked at nitrogen dioxide, a more direct tracer of combustion than PM2.5, and found that NO2 drops by about twenty to thirty percent on Shabbat in Jerusalem and Tel Aviv. But the interesting part is that when they applied meteorological normalization, the drop became more consistent week to week. Raw data had some Shabbats where NO2 barely budged because of stagnant air, and some weekdays where it looked artificially clean because a strong wind swept through. Deweathering revealed the true emissions pattern.
Twenty to thirty percent is substantial but not enormous. You still have a lot of background NO2.
Right, because power plants and industrial sources don't shut down for Shabbat, and NO2 can travel. But for PM2.5, the Shabbat signal is trickier to isolate, and that's where Daniel's second question becomes really important — the natural baseline. Let me give you some numbers. The WHO's air quality guideline for annual mean PM2.5 is five micrograms per cubic meter. Most places on Earth exceed that, and some exceed it without any human help.
What's the natural background? If you removed every car and factory, what are we breathing?
It varies enormously by region. Over the open ocean, PM2.5 can be as low as one to two micrograms per cubic meter — mostly sea salt and sulfates from marine phytoplankton. In the Amazon basin during the wet season, you might see three to five. But in arid and semi-arid regions, the natural baseline is much higher. The Middle East sits in the global dust belt, which stretches from the Sahara across the Arabian Peninsula to Central Asia. In Jerusalem, even on a day with minimal human activity, the natural dust background probably contributes something like ten to twenty micrograms per cubic meter annually, and much higher during dust events.
The WHO guideline of five is physically impossible in this part of the world, regardless of policy.
And this is a real policy tension. The WHO guidelines are aspirational but not regionally adjusted. Israel's annual average PM2.5 is around twenty to twenty-two micrograms per cubic meter. Even if you eliminated every anthropogenic source, you'd probably still be above ten, maybe fifteen, just from dust. That doesn't mean we shouldn't reduce what we can, but it does mean the benchmark is misleading if you don't account for geography.
The dust belt is basically the geological equivalent of "it's not you, it's the location.
The Sahara is the big driver. Something like five hundred to seven hundred teragrams of dust leave the Sahara every year — a teragram is a million metric tons. About a quarter goes west across the Atlantic, fertilizing the Amazon, and a significant fraction goes east across the Mediterranean, hitting Israel, Lebanon, Cyprus, Greece. During a major dust event, PM10 in Jerusalem can exceed a thousand micrograms per cubic meter. The twenty-four-hour WHO guideline for PM10 is fifty. You get twenty times that.
5, the stuff that gets into your lungs and bloodstream, also spikes during these events.
During a Saharan dust outbreak, PM2.5 can jump from maybe twenty to well over a hundred in a matter of hours. And here's the thing about dust PM2.5 versus combustion PM2.5 — the toxicity is different. Dust particles tend to be more mineralogical, less coated in polycyclic aromatic hydrocarbons and transition metals than particles from diesel exhaust or coal burning. That doesn't mean dust is harmless — it's associated with respiratory and cardiovascular hospital admissions — but per microgram, the health impact appears to be lower than combustion-derived PM2.The epidemiological evidence is still developing, but it complicates the simple "PM2.5 is PM2.
Even if two cities have the same annual average PM2.5, the health burden could be completely different depending on whether the source is dust or traffic.
And this gets lost in the air quality discourse. We talk about AQI as if it's one number that tells you how bad the air is. But an AQI of a hundred and fifty from a dust storm is not the same health risk as an AQI of a hundred and fifty from a temperature inversion trapping diesel exhaust over a city.
The AQI is the glockenspiel of public health communication — it makes one clear sound, and that sound is "the air is bad," but it doesn't tell you why or what to do about it.
And ozone, which Daniel mentioned he's particularly interested in, has its own natural component completely separate from dust. Ozone isn't emitted directly — it forms photochemically when nitrogen oxides and volatile organic compounds react in sunlight. But there's also a natural background of tropospheric ozone from stratospheric intrusions, lightning-produced NOx, and biogenic VOCs — trees emit isoprene, which contributes to ozone formation.
Trees are polluting. The ultimate NIMBY twist.
They're not "polluting" in a moral sense, but biogenic emissions are real and significant. In the southeastern United States, ozone levels in rural areas can be surprisingly high not because of cars but because of oak and pine forests pumping out isoprene and terpenes. The natural background of ozone is typically around twenty to forty parts per billion. The WHO guideline for peak season ozone is sixty parts per billion. So even with zero human emissions, you're already halfway or more to the guideline just from natural processes.
Which makes the margin for anthropogenic ozone pretty thin.
It does, especially in sunny climates. Israel has high ozone because of abundant sunshine, high temperatures, and the right precursor mix. Jerusalem's elevation — about seven hundred fifty meters — also matters because you're closer to the stratospheric ozone reservoir. During stratospheric intrusion events, you get pulses of ozone from the upper atmosphere mixing down that have nothing to do with local emissions.
If we zoom out, the natural contributors to what we measure as air pollution include Saharan dust, sea salt aerosols, biogenic VOCs, stratospheric ozone intrusions, lightning-produced NOx, volcanic emissions, and non-human-caused wildfires.
Pollen, which contributes to PM10 and PM2.5 during certain seasons. And secondary organic aerosol formation from biogenic VOCs — the famous blue haze over the Blue Ridge Mountains is natural photochemical smog from tree emissions. The point is that the baseline isn't zero. It's not even close to zero in many parts of the world.
Let's go back to deweathering for a moment, because I want to understand the limitations. If you're training a random forest model on historical data, you're learning the relationship between weather and pollution during a period when human emissions were also happening. How do you know you're not learning something about the human-weather interaction rather than pure weather effects?
That's a sharp question, and it's one of the known challenges. If traffic patterns change with weather — fewer people drive when it rains — the model might attribute some of the emission reduction to the rain rather than to the behavior change. The standard approach is to include temporal variables — day of week, hour of day, month — to capture emission patterns separately from meteorology. But you're right, it's not perfect. There's a residual coupling.
Deweathering gives you an estimate, not a clean separation.
It gives you a much better estimate than raw data, but yes, there's uncertainty. The other limitation is that these models are trained on the range of weather that occurred during your training period. If you get an unprecedented heat wave or unusual wind pattern, the model is extrapolating, and random forests are notoriously bad at extrapolation.
Which in a changing climate is not a theoretical concern.
Not at all. As extreme weather events become more frequent, the historical relationship between meteorology and air quality may not hold. There's active research on whether deweathering models trained on past data remain reliable under climate change. The early answer seems to be that they degrade but not catastrophically, at least for moderate deviations.
Let me pull us back to Daniel's specific experiment. He's comparing the workday-to-Shabbat ratio in Jerusalem against the weekend-to-weekday ratio in London and New York, and he's finding that when he dewears the data, the Jerusalem ratio becomes more pronounced relative to the other cities. What does that actually tell us?
It suggests that in Jerusalem, weather variability was partially masking the Shabbat signal. Maybe dust events were happening disproportionately on certain days of the week in his study period, or seasonal patterns in atmospheric stability coincided with the weekly cycle in ways that obscured the true emission reduction. London and New York have different meteorological regimes — less dust, more cloud, different boundary layer dynamics — and the weekend effect in those cities is already well-studied. In New York, PM2.5 drops by maybe five to ten percent on weekends. In London, it's similar. If Jerusalem's Shabbat drop is larger — say fifteen to twenty-five percent for PM2.5 — and deweathering makes that gap wider, that's evidence that the emissions reduction is real and substantial, but weather was making it look smaller than it is.
Because a random dust storm on a Saturday could swamp the signal from all those parked cars.
One Saharan dust event on a Shabbat can make it look like Shabbat had no effect at all, or even made things worse. Deweathering removes that one anomalous day by asking: what would this Shabbat have looked like with average weather?
The answer is: cleaner than the weekday, by a margin larger than what you see in London or New York.
That's the hypothesis. And it makes intuitive sense. In Jerusalem, the activity reduction on Shabbat is more comprehensive than a typical Western weekend. Public transit stops. Many businesses close. There's a cultural and legal dimension that goes beyond people just having a day off.
The city literally powers down in a way that London on a Sunday doesn't.
And you can see this in other data streams. Electricity demand in Israel drops by about ten to fifteen percent on Shabbat. Traffic counts on major Jerusalem roads drop by fifty to seventy percent. These are real, measurable reductions in activity, and they should show up in air quality if you can isolate the signal from the noise.
Which is what deweathering lets you do.
And Daniel's approach of comparing against London and New York is clever because it gives you a control. If you just looked at Jerusalem alone, you might wonder whether the Shabbat effect is just the normal weekend effect that every city has. By benchmarking against cities where the weekend is secular and less comprehensive, you can quantify how much extra reduction you get from the religious observance dimension.
It's a natural experiment with a built-in control group.
There's a literature on this. Beyond the Technion paper I mentioned, there's been work on air quality during Ramadan in Middle Eastern cities, during the Hajj in Mecca, during Chinese New Year in Beijing — all these cultural and religious events that create predictable changes in human activity. Shabbat is a recurring natural experiment that's been running every week for millennia.
The longest-running A/B test in history.
Until recently, nobody had the data infrastructure to run it. What Daniel's doing — pulling seven million data points through an API, running deweathering algorithms, doing cross-city comparisons — that would have been a multi-year PhD project twenty years ago. Now it's something a motivated person with Claude Code can do in an afternoon.
The barrier to entry for this kind of analysis has collapsed.
And that's exciting, but it also means we need more people who understand what the tools are actually doing. You can run deweathering as a black box and get a number, but if you don't understand what meteorological normalization means or what its limitations are, you might overinterpret the results.
Which is why we're talking about it.
Let's go deeper on the natural phenomena side. You mentioned the Saharan dust contribution. What about seasonal patterns? I'd imagine the natural baseline isn't constant throughout the year.
It's highly seasonal, and the seasonality varies by pollutant and by region. 5 and PM10 in Israel, the dust season runs roughly from November to May, with peaks in spring and autumn. Summer is actually relatively dust-free because the atmospheric circulation patterns that transport Saharan dust eastward are less active. But summer brings its own natural pollution — higher ozone because of more intense photochemistry, and more frequent stratospheric intrusions.
If you're looking at a full year of data, you've got winter-spring dust, summer ozone, and then whatever the anthropogenic sources are doing on top of that.
And the anthropogenic sources have their own seasonality. In winter, more residential heating emissions. In summer, more air conditioning demand means more power plant emissions. Traffic patterns shift with school schedules and holidays. Untangling all of this is what makes air quality epidemiology so challenging.
This is where deweathering can actually help, because it lets you hold meteorology constant and see what's left.
And the seasonal natural patterns are big enough that they can completely dominate the annual average in some years. There was a study of PM10 in the eastern Mediterranean that found interannual variability was driven almost entirely by the frequency and intensity of dust events, not by changes in local emissions. A year with five major dust storms will have a much higher annual average than a year with two, even if nothing about human activity changed.
Which means if you're a policymaker trying to track whether your clean air policies are working, you need to control for dust events, or you might conclude your policies failed when really you just had a dusty year.
That's exactly the problem. And this is why meteorological normalization is becoming standard practice in air quality trend analysis. Regulatory agencies in Europe and North America are increasingly using these methods to separate meteorology from emissions when assessing compliance with air quality standards. The US EPA has a whole workflow for this.
Let me ask about something Daniel mentioned in passing — the idea that even without human activity, we wouldn't see a "perfect" AQI anywhere. What would the AQI be in a world without us?
It's a fun thought experiment. Over the open ocean, you'd see AQI in the green, single digits. In the global dust belt, you'd routinely see yellow and orange, maybe red during major dust storms. Near active volcanoes, you'd see purple and maroon. During wildfire season in places like the western US or Australia — and pre-human lightning-ignited wildfires were real — you'd see very high AQI episodically. The natural world is not clean in the sense of having perfectly pure air.
The pristine baseline is a myth.
It's a myth with a kernel of truth. Pre-industrial PM2.5 was certainly lower than today in most places, but not zero, and not uniformly low. Ice core records show that dust deposition has varied enormously over geological time, with glacial periods being much dustier than interglacials. The Holocene has been relatively dust-free by geological standards, which is one reason human civilization flourished in this period.
We got lucky with the climate window, and then proceeded to fill it with diesel particulates.
That's one way to put it. But the point is that natural variability is real and large, and understanding it is essential for making sense of air quality data. When Daniel looks at his seven million data points and sees a spike, the first question should be: was there a dust storm? The second: was there a temperature inversion? The third: was it a weekday or Shabbat? The order matters.
Because if you ask the Shabbat question first, you might attribute a dust spike to human activity simply because it happened on a weekday.
And the deweathering approach handles this elegantly by asking the weather question first, statistically, and then leaving you with the residual that's more likely to be emission-driven.
I want to circle back to something you said about the toxicity difference between dust PM2.5 and combustion PM2.If the health impact per microgram is different, does that mean we should have different air quality standards for different regions?
That's a live debate in the environmental health community. The WHO guidelines are universal, but there's growing recognition that PM2.5 from different sources has different toxicity. Diesel particulate matter is probably the most harmful per microgram. Dust is on the lower end, though not zero. Sulfate particles from coal burning are somewhere in the middle. If you set a single standard, you're implicitly assuming all PM2.5 is equally toxic, which we know isn't true.
The alternative — source-specific standards — would be an administrative nightmare.
How do you enforce a standard that depends on what the particles are made of? You'd need continuous chemical speciation monitoring everywhere, which is expensive and technically demanding. The practical compromise is to keep the mass-based standard but recognize that the health benefits of reducing PM2.5 will depend on what you're reducing. A policy that cuts diesel emissions by ten micrograms per cubic meter is worth more, in health terms, than a policy that cuts dust by the same amount.
Which has implications for Israel specifically. If a large fraction of Israel's PM2.5 is natural dust, the cost-benefit calculus for reducing the anthropogenic fraction looks different than in a place where most PM2.5 is from combustion.
The marginal benefit of reducing anthropogenic PM2.5 in Israel is actually higher because you're targeting the more toxic fraction. But the total achievable reduction is capped by the natural background. You could eliminate every car and factory in Israel and still have PM2.5 above the WHO guideline during dust events. That's not an argument against reducing emissions — it's an argument for realistic expectations.
For communicating those expectations to the public. The AQI app on your phone doesn't tell you whether the red number is from Saharan dust or rush hour traffic.
No, and that's a real communication failure. During a dust storm, people see the AQI spike and think the air is toxic in the same way it would be during a smog event. The health advice should be different. During a dust storm, the main concern is respiratory irritation from mineral particles. During a smog event, you're dealing with a cocktail of carcinogens and cardiovascular stressors. Masking helps in both cases, but the long-term risk profile is different.
This connects to something we've talked about before — the AQI is a useful simplification, but it collapses multiple dimensions of risk into a single number. It's the BMI of environmental health.
That's a good comparison. BMI tells you something, but it doesn't tell you about body composition, metabolic health, or fitness. AQI tells you something, but it doesn't tell you about particle composition, gas-phase pollutants, or exposure duration. And just like BMI, it can be misleading at the individual level while being useful at the population level.
For Daniel's analysis, the deweathering step is essentially removing one of the dimensions that confounds the weekly pattern — and in doing so, it reveals that the Shabbat signal is real and substantial.
And the fact that the deweathered ratio in Jerusalem is more pronounced than the weekend ratio in London and New York suggests that the religious observance dimension — the comprehensive shutdown — produces an additional air quality benefit beyond what you get from a secular day of rest.
Which is a fascinating finding, if it holds up. It's not just that people drive less on weekends everywhere. It's that when an entire city substantially reduces activity on a fixed schedule, the air notices.
The air notices. And now, thanks to open data and modern analytical tools, we can notice the air noticing.
Should we talk about what this means for policy? If the Shabbat effect is real and measurable, does that have implications beyond being an interesting natural experiment?
I think it does, though you have to be careful not to overinterpret. The Shabbat effect demonstrates that large, coordinated reductions in activity produce measurable air quality improvements. That's not a new insight — we saw it during COVID, we saw it during the Beijing Olympics. But the weekly recurrence of Shabbat makes it a particularly clean signal. It's not a one-off intervention with confounding factors. It's the same intervention, every week, for thousands of years.
The reproducibility is built in.
And reproducibility is the gold standard in science. If you want to study what happens when a city reduces traffic by fifty to seventy percent, you can study Shabbat in Jerusalem every single week. You don't need a special event or a lockdown.
Though the policy translation is tricky. You can't exactly mandate Shabbat observance for air quality reasons.
No, and you shouldn't. But you can look at the Shabbat data and ask: what specific activities are driving the improvement? Is it the reduction in private car travel? The closure of commercial districts? The reduction in industrial activity? If you can identify the components, you can design policies that target those components without the religious framework.
Congestion pricing, commercial vehicle restrictions, low-emission zones — these are all policies that try to replicate, in a targeted way, some of what Shabbat does organically.
And the Shabbat data gives you a real-world benchmark for what's achievable. 5 drops by X percent when traffic drops by Y percent on Shabbat, you can model what a congestion pricing scheme that reduces traffic by some fraction of Y would achieve.
Assuming the relationship is linear, which it probably isn't.
It almost certainly isn't. Air pollution chemistry is nonlinear. Ozone, especially, has a complicated relationship with its precursors. Reducing NOx can actually increase ozone in some urban cores because you're titrating less of it away. This is the weekend ozone effect — some cities see higher ozone on weekends despite lower NOx emissions because the chemistry shifts.
Reducing traffic might make some pollutants better and others worse.
And that's another reason why Daniel's analysis is interesting. If he's looking at multiple pollutants — PM2.5, PM10, NO2, ozone — across the Shabbat transition, he might see different patterns for each. NO2 should drop sharply because it's directly emitted by vehicles and has a short atmospheric lifetime. 5 should drop more modestly because of the dust background. Ozone might actually increase, or it might decrease depending on the local chemistry regime.
The Shabbat natural experiment could tell you which pollutants are most traffic-sensitive in Jerusalem specifically.
And that's locally actionable information. If you find that NO2 drops forty percent on Shabbat but PM2.5 only drops ten percent, that tells you traffic is the dominant source of NO2 but not the dominant source of PM2.Your policy priorities should follow that evidence.
You'd focus on traffic for NO2 reduction and look elsewhere — industry, transboundary pollution, dust management — for PM2.
And dust management is its own fascinating topic. You can't stop the Sahara from producing dust, but you can reduce the resuspension of dust by vehicles on roads, which is a significant contributor to urban PM10. Street sweeping, paving unpaved roads, reducing construction dust — these are all interventions that target the natural-anthropogenic hybrid fraction.
The dust that's natural in origin but human in mobilization.
A dust particle that blows in from the Sahara and settles on a road is natural. That same particle, pulverized and re-entrained into the air by a passing truck, is now anthropogenic. The boundary between natural and human-caused is fuzzier than it first appears.
Which makes deweathering even more valuable, because it doesn't care about the philosophical distinction. It just asks: given the weather, what would we expect? Everything else is something we might be able to influence.
The residual after meteorological normalization is the part of air pollution that's potentially within human control, whether it's direct emissions or resuspended dust or secondary aerosol formation driven by human-emitted precursors.
Let's talk about ozone a bit more, since Daniel mentioned he's particularly interested in it. You said the natural background is twenty to forty parts per billion. What drives the spikes above that in a place like Jerusalem?
Ozone formation requires three things: nitrogen oxides, volatile organic compounds, and sunlight. Jerusalem has all three in abundance. The NOx comes primarily from traffic. The VOCs come from traffic, industrial solvents, and also from vegetation — Jerusalem is surrounded by pine forests that emit terpenes. And the sunlight is intense, especially in summer. The photochemistry is fast. You can get ozone formation within hours of the precursor emissions.
Ozone is a regional pollutant, not just a local one.
Ozone has an atmospheric lifetime of days to weeks, so it can be transported hundreds of kilometers. Jerusalem's ozone is influenced by emissions in Tel Aviv, in the coastal plain, even in neighboring countries. During summer, the eastern Mediterranean often has elevated ozone across the entire region because of the combination of high pressure systems, abundant sunshine, and precursor emissions from multiple sources.
Even if Jerusalem eliminated all its own NOx and VOC emissions, it would still have ozone from upwind sources plus the natural background.
And this is the challenge with ozone. It's not like PM2.5 where you can point to a specific local source and regulate it. Ozone is a collective action problem. Every city in the region contributes to every other city's ozone.
Which makes the Shabbat effect on ozone particularly interesting. If traffic drops in Jerusalem on Shabbat, but traffic continues in surrounding areas, what happens to Jerusalem's ozone?
It could go either way. In a NOx-saturated urban core, reducing NOx can increase ozone because you're reducing the titration reaction where NO converts ozone back to NO2. This is the weekend ozone effect observed in Los Angeles and other cities. In a more rural or NOx-limited area, reducing NOx reduces ozone. Jerusalem might be in a transitional regime where the effect varies by season and by location within the city.
Daniel might find that ozone is actually higher on Shabbat in some parts of Jerusalem.
It's possible. And that wouldn't mean Shabbat is bad for air quality — it would mean the ozone chemistry is nonlinear and you need to understand the regime before you can interpret the data.
Which is a good general principle for this entire conversation. You need to understand the system before you can interpret the measurement.
That's really the thesis of everything we've been saying. The number on the AQI app is the end of a long chain of physical and chemical processes, some natural, some human, some a hybrid of both. Without understanding that chain, the number is just a number.
Deweathering is one tool for pulling the chain apart.
One very useful tool. Not the only one. There's also source apportionment, which uses chemical fingerprints to attribute PM2.5 to specific sources — diesel, gasoline, dust, sea salt, biomass burning. There's back-trajectory analysis, which traces air parcels backward in time to see where they came from. There's satellite remote sensing, which gives you spatial context that ground monitors miss. A complete analysis would combine all of these.
For Daniel's specific question — is the Shabbat effect real, and is it bigger than a generic weekend effect — deweathering plus cross-city comparison is a solid approach.
It's a great approach. And I'd love to see the results when he's done. The seven million data points, the deweathering, the Jerusalem-London-New York comparison — this is exactly the kind of citizen science that open data enables. Someone with curiosity and technical skills can produce genuinely novel insights.
The barrier isn't the data anymore. It's knowing what questions to ask and how to interpret the answers.
And that's where understanding the natural baseline, the meteorology, the chemistry — all of that contextual knowledge — becomes essential. You can run the algorithm without it, but you can't understand what the algorithm is telling you.
Which brings us back to where we started. The weather was obscuring the Shabbat signal. Controlling for it revealed the signal. That's counterintuitive to most people, but it makes perfect sense once you understand what deweathering actually does.
It's not magic. It's just careful statistical accounting. And it's becoming standard practice in air quality science for exactly this reason.
To summarize for the question — deweathering uses machine learning to predict what pollution would have been under average weather conditions, and the residual tells you what's likely driven by emissions. Natural phenomena — Saharan dust, sea salt, biogenic VOCs, stratospheric ozone, wildfires — contribute a significant baseline that varies by region and season, and in the Middle East that baseline can be half or more of the total PM2.5 in some years. The Shabbat effect in Jerusalem is real, it's larger than a generic weekend effect, and deweathering makes it clearer by removing the meteorological noise.
That's the episode.
Now: Hilbert's daily fun fact.
Hilbert: In nineteen twenty-four, the Soviet Union operated exactly one public timekeeping station serving the entire Caspian basin — a single mechanical clock in Baku that synchronized shipping schedules, train departures, and oil field shifts across an area larger than France.
One clock for an area the size of France.
The logistical confidence that implies is staggering.
So here's a forward-looking thought. As climate change shifts weather patterns, the natural baseline for air pollution in many regions is going to shift too. More dust storms, more wildfires, different ozone chemistry. The distinction between natural and anthropogenic is going to get even blurrier, and tools like deweathering are going to become even more important for understanding what we can control and what we can't.
The Shabbat natural experiment is a reminder that large-scale behavior change does show up in the data — when you know how to look for it.
This has been My Weird Prompts. Thanks to our producer Hilbert Flumingtop. You can find every episode at myweirdprompts.If you enjoyed this, leave us a review wherever you listen — it helps.
Until next time.
