Summary

  • Columbia University climate researcher Zoe Yiyuan Zhou and colleagues report in the Proceedings of the National Academy of Sciences that satellite-measured air pollution rose 24% to 33% across 46 cities in the months after the State Department ended its embassy air quality monitoring program in March 2025, with the researchers estimating the increase corresponds to a roughly 4% rise in PM2.5.
  • The pre/post design documents no significant pollution trends in the 300 days preceding the shutdown and reports the post-shutdown pollution rise as concentrating in cities with fewer alternative monitoring sources, evidence the researchers cite against gradual-drift and common-shock alternatives.
  • Three candidate causal channels — loss of third-party auditing, loss of a regulatory trigger, and the opportunity for polluters to skip compliance costs — appear in the researchers’ framing without direct measurement, and the article reports no confidence interval for the 4% PM2.5 calibration, leaving mechanism-level attribution and magnitude precision underspecified.
  • Andrea La Nauze of Deakin University, who has separately researched the program, called the study “high quality” and the 4% PM2.5 magnitude “certainly concerning”; the State Department, which the article reports cited monetary costs as the rationale for the shutdown, did not respond to a request for comment in time for publication.

A study published Monday in the Proceedings of the National Academy of Sciences by Columbia University climate researcher Zoe Yiyuan Zhou and colleagues reports that satellite-measured air pollution rose 24% to 33% across 46 cities in the months after the State Department ended its embassy air quality monitoring program in March 2025, with the researchers estimating the satellite readings translate to a roughly 4% increase in PM2.5. Increases were highest in Southeast Asian cities — Manila in the Philippines, Yangon in Myanmar, and Vientiane in Laos — and in cities with fewer independent monitoring alternatives. The State Department, which the article reports cited monetary costs as the rationale for the shutdown, did not respond to a request for comment in time for publication.

Empirical claim and proposed causal structure

The researchers’ interpretation treats the cessation of embassy monitoring as the cause (X) and the pollution increase as the effect (Y), with the causal arrow running from monitoring absence, through reduced regulatory enforcement and reduced public-accountability pressure, to higher emissions. Zhou described the research opportunity as unusual: “It’s such a rare case to study this drawback. While there is ample evidence of the benefits of increased monitoring, there are not many examples of a sudden falloff of surveillance.” Earlier research suggested the program contributed to measurable improvements in air quality, especially in countries where alternative monitoring was scarce, giving the monitoring-removal story prior empirical support. The article does not identify which mechanisms that prior literature itself credited, leaving open whether the present researchers’ three proposed mechanisms correspond to the channels the earlier work documented.

Three candidate mediators appear in the researchers’ framing: the loss of third-party auditing, the loss of a regulatory trigger, and the opportunity for polluters to skip compliance costs. These mediators are described in the discussion rather than directly measured; the study’s measurement layer sits one step upstream at the satellite pollution readings themselves. The researchers describe the design as an intervention-level natural experiment. The treatment is the shutdown of U.S. embassy monitoring. The primary outcome is satellite-measured pollution, with the article reporting that researchers estimate their satellite measurements translate to roughly a 4% increase in PM2.5.

What the design rules out

The article documents one direct check on temporal confounding: “no significant trends in air pollution over the 300 days preceding the end of U.S. reporting.” That flat pre-trend strengthens the case against gradual drift over the year before the shutdown. A moderator — the number of alternative monitoring sources available in each city — is explicitly examined, with the finding that pollution “spiked more sharply in cities with fewer monitoring alternatives.” That differential response is the strongest piece of evidence in the article against common-shock alternatives, because a single global or seasonal driver affecting all 46 cities simultaneously would not, in general, respect the monitoring-alternatives gradient the researchers report.

What the design does not rule out

The pre/post design does not, on its own, exclude a shared time-varying cause, a coincident global shock, or selection-driven regression toward a higher city-specific mean.

Seasonal confound. If the monitored cities share seasonal pollution cycles whose troughs align with the pre-period and whose peaks align with the post-period, time-varying weather patterns would produce the same observation without any causal role for the monitoring program. For the Southeast Asian cities the article identifies as worst-affected, the seasonal cycle most commonly implicated for regional haze is the burning of Indonesian and regional peatland and slash-and-burn fires, which typically intensifies during the dry season. That window partly overlaps the three-to-four-month post-shutdown peak the article reports. A 300-day pre-treatment window can mask seasonal structure if dominated by other noise.

Concurrent global events. A common cause operating in mid-2025 could drive both monitoring cessation decisions and emissions changes. Energy price movements, industrial recovery cycles, or geopolitical disruptions in regions that host many of the 46 cities could shift emissions independently of any single monitoring decision. The article does not name a specific concurrent event, and the study design does not include a control group of cities that never had embassy monitoring, so any such global shock would not be separately identified.

Latent regulatory environment. An unobserved regulatory-capacity factor could create the appearance of a causal link. Countries with weaker regulatory environments may have benefited more from embassy monitoring and may also be more susceptible to pollution increases once that monitoring ends. Under this structure, the monitoring program’s end is correlated with, but not the cause of, the pollution rise; both are downstream of the unobserved factor.

Selection on the outcome of interest. If embassies were sited in cities whose pollution was already elevated or trending upward, the post-shutdown rise could partly reflect regression toward a higher city-specific mean rather than a behavioral response to lost monitoring. This alternative appears, on general diplomatic grounds, weaker than the others: embassy siting is typically governed by strategic criteria — capital cities, consular districts, host-country permissions — rather than by pollution levels, which makes pollution-driven selection-on-outcome less plausible than the seasonal, concurrent-events, or latent-regulatory-environment alternatives.

What would discriminate among the alternatives

Three auxiliary observations would distinguish the alternatives on data the article does not report.

A control group of comparable cities that never had embassy monitoring would sharpen the inference: if their pollution followed the same seasonal or global pattern during the same window, the seasonal and concurrent-event alternatives gain weight; if their pollution did not, those alternatives weaken. City-level seasonal-decomposition analysis matched to local climate calendars would also discriminate: uniform spikes across cities in different climate zones weakens seasonal confounding, while concentration in cities with shared weather patterns strengthens it. A natural experiment in which monitoring is restored in a subset of cities would discriminate against the latent-regulatory-environment alternative: if pollution in restored cities reverses while unrestored cities remain elevated, the monitoring-removal story is supported against the unobserved-factor account.

The article’s coverage frames this as a limit the researchers acknowledge in their own framing: the study cannot determine why pollution increased after monitoring stopped. The intervention rung of the causal-inference ladder is reached for the average effect; the counterfactual rung is not.

Post-peak decline as a second empirical signature

The article reports a second empirical signature the proposed causal structure should also explain: pollution levels “appeared to decline somewhat after peaking about three to four months after monitoring stopped.” Zhou said she suspects it may reflect other forms of monitoring — by individuals or local governments — coming online in the wake of the U.S. withdrawal. That substitute-monitoring hypothesis treats the post-peak decline as the entry of a new monitoring node partially compensating for the removed U.S. monitoring, with its own effect on the auditing and regulatory channel. Zhou said it is “difficult to determine why this occurred or whether the trend will continue.”

The post-peak decline is also consistent with regression to a higher mean or with a time-varying confound that has since reversed. The timing of the decline is itself informative: if the decline coincides with monsoon onset or with the end of the regional burning season in the affected Southeast Asian cities, the seasonal-confound alternative gains weight; if the decline occurs uniformly across cities with different climate calendars, that alternative weakens. The decline does not, by itself, adjudicate among the candidate causal structures. The pre-shutdown-to-post-shutdown comparison is reported with quantified magnitudes; the post-peak decline is reported with hedged speculation. That asymmetry is the structure of what the substrate supports.

Mechanism-level attribution coverage

Eric Zou — an environmental economist at the University of Michigan, distinct from the study’s lead author Zoe Yiyuan Zhou — is quoted endorsing the third-party-auditing channel directly: “Maybe the U.S. Embassy was a really strong force, in terms of a third-party auditing mechanism, and once that’s removed, people are like ‘Now I can just pollute, right?’” The other two mechanisms — the regulatory-trigger channel and the skip-compliance-costs channel — appear in the researchers’ framing without on-record expert endorsement in the article.

Outside assessments

Andrea La Nauze, an environmental economist at Deakin University in Australia who has researched the benefits of the program but was not involved in this study, called the findings significant. “It’s a great study, high quality. And it’s a real pity,” she said, adding that the program “led to sustained improvements in air quality that were continuing to increase after 10 years.” La Nauze said the 4% PM2.5 magnitude is “certainly concerning” because “particulate matter is a serious and dangerous thing, and being exposed to it is not good.” Zhou said, “Data is just so important. Having access to information is necessary to make sure that these previous gains can continue. Public information creates accountability.”

Source limits and unresolved questions

The article notes that “satellite data is not a perfect substitute for ground-level monitors” and that “on-the-ground measurements were not consistently available for the post-shutdown period.” The measurement model is therefore the satellite-to-PM2.5 calibration reported by the researchers; identifiability of the treatment effect is conditional on that calibration. The article does not report a confidence interval for the 4% PM2.5 figure or characterize known satellite-to-ground biases, so the magnitude estimate carries a calibration uncertainty the substrate does not pin down.

The article also does not specify the full post-shutdown observation window beyond “roughly three to four months after reporting ended,” nor does it reconcile the program’s expansion to more than 80 embassies and consulates with the 46 cities included in the study’s analysis. The State Department’s stated rationale — “citing monetary costs,” per the article — is the only documented characterization of the shutdown decision on the public record, and the agency’s “no response” is recorded in the article rather than supplied.

The available structure supports a correlation between the monitoring shutdown and the observed pollution rise, and the differential response by monitoring-alternative density is the strongest piece of internal evidence against common-shock alternatives. The structure does not, on the article’s own account, support a fully identified causal claim about the mechanism or its magnitude with the precision of the 4% PM2.5 figure.

Analytical techniques used in this piece

This analysis applies the methods below. Each links to a short, plain-English explainer you can read and reuse.

Causal DAG
Maps cause and effect as an explicit directed graph, exposing confounders and mediators (Pearl).
BATNA
Your best alternative to a negotiated deal — the walk-away that sets your leverage (Fisher & Ury).