44 percent of US urban areas above the new fine particle standard would go undetected — about 20 million people

What is happening?

A study published on 2024-10-15 in Environmental Science and Technology Letters compared the federal fine particle monitoring network against modeled concentrations for every US census tract and found that 44 percent of urban areas exceeding the new annual PM2.5 standard of 9.0 micrograms per cubic meter — about 20 million people — would remain undetected because of gaps in current monitoring. The comparison used regulatory monitoring data for 2017 through 2019 against census-tract estimates from an empirical model that fuses satellite data with land-use variables.

Inside that finding sits a smaller and sharper number. 2.8 million people live in census tracts that are themselves above the standard while sitting inside areas the network classifies as attainment. The study calls these uncaptured hotspots and reports that they hold substantially higher shares of minority and low-income residents than the country as a whole. The study also identifies 10 priority locations whose monitoring would cut the uncaptured hotspot population by 67 percent.

The standard itself is in force and contested. EPA lowered the annual limit from 12.0 to 9.0 micrograms per cubic meter on 2024-02-07, 26 states and industry groups petitioned for review on 2024-03-06, EPA asked the court to vacate its own rule on 2025-11-24, and the DC Circuit upheld the standard on 2026-06-26, rejecting every challenge including the one filed by the agency. Designations are a separate track and that track has not moved: as of 2026-07-31 the EPA Green Book carried no page for this standard at all, and a federal district court ordered on 2026-07-17 that designations be promulgated by 2027-02-06 after the statutory deadline of 2026-02-07 passed with none issued. So the map of which areas violate the standard has not been drawn, and when it is drawn it will be drawn on the network as it already stands — the EPA monitoring fact sheet for the 2024 rule states that the initial designations likely will not include data from any newly sited monitors, because three years of data are required. That same fact sheet states that tightening the annual standard by a quarter produces about a 1 percent increase in the number of minimally required monitors, against a deployed network of almost 1,000 PM2.5 stations.

Whose problem is this?

The agency that decides how much measurement is legally required is the same agency that would have to act on what the measurement showed. Nothing outside that loop compels a monitor to be placed where no monitor has ever read.

Where does this problem end?

The boundary matters here because nobody disputes that the standard is in force. What is missing is the measurement that would show where it is being broken.

What is the state now, and what should it be?

Now

Needs a new measurementthe target state: no source opened here names a coverage target for the monitoring network. Appendix D states monitoring objectives such as maximum concentration, at-risk community and near-road placement, but no opened source states what fraction of the exceeding population the network is supposed to detect. There is therefore no official number against which 44 percent undetected registers as a failure — the benchmark it fails is the one the researchers supplied, not one the regulator set.

How big is it?

Between 2.8 million and 20 million people, on the 2017 through 2019 data window used by the study. The two bounds answer different questions. The upper bound of about 20 million counts everyone living in an urban area that exceeds the standard and that the network would not flag, including residents of tracts inside that area whose own air may be cleaner. What those people share is that no area-level protection is triggered for the place they live. The lower bound of 2.8 million counts only people individually misclassified at census-tract resolution: their own air is above the standard and the network reports their area as meeting it. Neither figure carries a published confidence interval in any source opened here.

What the range does not count: rural and small-county America, because the study frame is urban areas, while the 2026 county-scale work finds nearly 80 percent of US counties with no ground monitor and 63 unmonitored counties at failing-grade levels; people beside a source in a monitored city whose area design value is measured elsewhere and reads clean; short-term exposure, since the analysis is annual-average and the 24-hour standard of 35 micrograms per cubic meter is out of frame; and the separate population living in areas that are monitored, do exceed, and have gone without a designation since 2026-02-07, which no opened source sizes.

A count of people whose air has never been measured cannot itself be a measurement. Every figure in this block is a modeled estimate attributable to one study on one three-year window, and it should be read as one.

Under what conditions does it arise?

1. The rule of decision is a measurement. Nonattainment status, and with it the state implementation plan obligations, permitting offsets and control requirements that constitute the entire consequence of the standard, attach only where a design value exists. 42 U.S.C. 7407 defines a third category for any area that cannot be classified on the basis of available information as meeting or not meeting the standard. An area with no monitor generates no design value and so cannot be shown to violate.

2. The obligation to measure is keyed to what measurement already found. The minimum number of required sites in 40 CFR Part 58 Appendix D turns on two axes, the population of the metropolitan statistical area and the most recent three-year design value. The design value is itself the output of monitoring, so the requirement to look is a function of what looking has already found.

3. The equity clause added in 2024 cannot reach an unwatched place. Appendix D directs that where an area has additional required sites, one is to be placed in an at-risk community with poor air quality. That direction only operates where the table has already required a third site, which happens only in the largest metropolitan areas whose measured design value is at or above 85 percent of a standard. In an unmonitored area it never triggers.

4. The money that operates monitors has been shrinking and was proposed for elimination. Federal funding for state and local air quality management grants, the line that covers the operation, repair and maintenance of air monitors, fell by about 20 percent in real terms between 2004 and the 2020 assessment. The FY2026 budget request proposed eliminating that line outright, at an FY2025 enacted level of USD 235.6 million. Congress did not enact the elimination in the aggregate, raising categorical grants within the account by USD 3.5 million while cutting the EPA total by 3.5 percent.

5. Even where measurement exists, the act of designation has not happened. EPA promulgated no designations under this standard by the statutory deadline of 2026-02-07 and had to be ordered to act by 2027-02-06. That is an enforcement failure stacked on top of the measurement failure, and the court order fixes only the date.

What has been tried?

Two directions have been tried in parallel, and neither addresses the network. One is to move the number that defines a violation, which happened in 2024 and survived challenge in 2026. The other is to force the agency to publish the map, which happened in 2026 and is now on a clock. The count of required monitors moved by about 1 percent through both.

What was found?

Why is it still unsolved?

Measurement-absent — the obligation to measure is set by what measurement already found, so a place that has never been measured is treated exactly like a place that was measured and came back clean.

The minimum required sites under 40 CFR Part 58 Appendix D, transcribed from the current edition of the regulation:

The fourth footnote to that table is the whole mechanism in one line: These minimum monitoring requirements apply in the absence of a design value. The column that ends in zero is the column that governs when nobody has measured. Not measuring and measuring clean arrive at this table as the same input.

The first movement is that the rule of decision is a measurement rather than a prediction. Nonattainment, and everything that follows from it, attaches only where a design value exists, and a design value is what a monitor produces over three years. Where there is no monitor the statute supplies a third box, for areas that cannot be classified on available information, and no opened source states what obligations attach to that box under this standard.

The second movement is the circularity. A metropolitan area of between 50,000 and 500,000 people with no monitor is required to operate zero sites; an area whose urbanized core falls under 50,000 appears in no row at all. The requirement to look is a function of what looking has already found, and the regulation nowhere explains why absence of data was assigned to the low column rather than the high one. The same shape runs in reverse in the rule that governs taking a monitor out: a PM2.5 site may be discontinued where it has shown attainment for five years and carries a low probability of approaching the standard, which forecloses the possibility of that site ever reading dirty. That discretion is shared with EPA rather than held by a state alone, and no opened source documents a siting or removal decision made in order to avoid a designation, which is why this is recorded as a measurement failure rather than an inverted incentive.

The third movement is that the 2024 rule tightened the number without tightening the looking. A 25 percent reduction in the standard produced about a 1 percent increase in minimum required monitors, by the estimate of the agency that wrote both. The equity siting direction added in the same rule applies only to areas that already have additional required sites, so it reaches no community the network does not already watch. And the designations now due 2027-02-06 must rest on three years of data, which means they will be drawn from the network as it stood across the very window in which the gap was measured.

What observation would mean it is solved?

Candidates — (a) the minimum monitoring requirement stops keying to the measured design value, so that a modeled or satellite-estimated exceedance can itself trigger an obligation to install a monitor (b) a repeat of the 2024 analysis on recent data shows the undetected share of exceeding urban areas falling substantially (c) EPA promulgates designations under the 9.0 standard and the number of areas placed in the unclassifiable category is small.

(a) alone is weaker than it looks. A trigger written on modeled data still has to survive the question of which model, at what resolution, with what bias correction. A requirement can also be written and then not funded, and the grant line that pays to operate a site fell by about 20 percent in real terms between 2004 and 2020 without any requirement changing.

(b) alone is worse. The undetected share can fall because the network grew or because concentrations fell, and those two have opposite meanings. National concentrations have been declining for two decades on the account the agency gave when it announced its 2025 reconsideration, while wildfire smoke has pushed them the other way in some regions, and no opened source separates the two. A falling share would need that separation before it says anything about coverage.

(c) alone counts paper. A small unclassifiable count could mean the network is adequate or it could mean unmonitored areas were placed in attainment by default, which is the failure restated as a result rather than fixed. The Green Book has no page for this standard yet, so there is no baseline against which to read the first count when it arrives.

None of these three observations can be read alone, and the first two depend on data that does not yet exist.

What is it connected to?

Fills with researchenvironmental justice screening tools and how they treat unmonitored areas, the low-cost sensor networks that operate outside the regulatory framework, ozone and air toxics monitoring coverage, wildfire smoke surveillance, and the same design-value circularity in other pollutant networks. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• Why absence of data was assigned to the lower requirement. Nothing in Appendix D explains the fourth footnote. No source opened here, including the EPA fact sheet for the 2024 rule, acknowledges that this was a choice, defends it, or records that the opposite default was ever considered.
• What coverage the network is supposed to achieve. No source states a completeness target. Appendix D names monitoring objectives but never a fraction of the exceeding population that must be detected, so the 44 percent figure has no official benchmark to fail.
• What one percent is in monitors. The fact sheet states the increase as a percentage and never converts it to a count, never says which areas gain a site, and never states the current minimum required number against which the percentage is computed. Almost 1,000 stations is the deployed count, and no opened source reconciles deployed with required.
• What the ten-monitor fix would cost. The finding that ten locations would cut the uncaptured hotspot population by 67 percent appears in the abstract and in every piece of coverage. Not one gives a capital or annual operating cost for a regulatory PM2.5 site, and the grant page states a total without a per-site figure, so no reader can compare the fix to anything.
• Whether the gap has changed since 2019. Every downstream citation through 2026 repeats the 2017 through 2019 window. No opened source re-runs the analysis on 2022 through 2024 data, which are the years that will generate the design values for the designations now due.
• What the five-year network assessments found. Every air agency must assess its network every five years and determine whether new sites are needed, with a round due in 2025. These go to regional offices. No national roll-up was found, and the EPA page for network assessments returned a not-found error. The legally mandated inventory of where the network is thin exists and is not aggregated anywhere the public can read.
• What happens to unmonitored areas in the coming designations. No source states how many areas EPA will designate unclassifiable for want of data rather than attainment, or what obligations attach to that category under this standard. The court order fixes the date and nothing else.
• Whether the monitoring grants survived. Neither the congressional research summary nor EPA publishes what share of the grant line goes to monitor operation or how many sites it sustains, and the FY2026 level for that specific line is not broken out. The page describing the Inflation Reduction Act monitoring awards, updated nine months into an administration that terminated hundreds of grants, mentions neither terminations nor an award list.

See the evidence

No primary journal article was read in full. The abstract of the core study was obtained complete and verbatim through an open bibliographic interface because the publisher returns a forbidden response, and its bibliographic fields were cross-checked against a second index for the same identifier. The regulatory text is primary and was read directly: the monitoring table and siting rules from the published Code of Federal Regulations, the statute and two rule sections from a law-school rendering that matched the government text cell for cell, and the EPA fact sheet and the filed complaint as published documents. Three sources disagree or drift, and the disagreement is left visible rather than resolved. First, the secondary coverage renders the central finding as 44 percent of nonattainment areas, while the study itself says urban areas; the study wording is used here, because as of 2026-08-08 there are no designated nonattainment areas under this standard and the other phrasing would be undefined. Second, two opened sources give the court-ordered deadline as 2027-02-06 and a third describes 2027-02-07 through the statutory extension; both dates are recorded. Third, a secondary source claimed the air grant line was eliminated in the enacted appropriation, which the congressional research summary does not support, so the claim is not carried and the absence of a line-level breakout is recorded as a silence instead. Two separate figures of 44 percent circulate in this subject area — the share of exceeding urban areas in the study, and a national share of Americans in counties with failing grades from a different organization on a different threshold — and only the first appears in this document. This is a Path A output, so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range 2,800,000–20,000,000 As of 2017-2019

Sensitivity The width of the range is not statistical. The two bounds are two different definitions from the same study on the same 2017 through 2019 data window, and neither carries a published confidence interval in any source opened here. The upper bound asks how many people live where area-level protection is not triggered; the lower bound asks how many people breathe air above the standard while their area is recorded as meeting it. What the range fails to count: rural and small-county America, because the study frame is urban areas, while 2026 county-scale work finds nearly 80 percent of US counties with no ground monitor, about 30 percent of the US population unmonitored, and 63 unmonitored counties at levels matching failing grades, none of which is converted to a headcount here because no population base was verified; people living beside a source inside a monitored city whose area design value is measured elsewhere and reads clean; short-term exposure, since the analysis is annual-average and the 24-hour standard of 35 micrograms per cubic meter is out of frame; and the separate population living in areas that are monitored, do exceed the standard, and have gone without any designation since the statutory deadline of 2026-02-07 passed, which no opened source sizes. The limit running in the opposite direction is the model. Both bounds rest on an empirical model that fuses satellite data with land-use variables being unbiased at census-tract scale in exactly the places no monitor observes. If that model overstates concentrations where monitoring is absent the counts fall, and if it understates them the counts rise. No opened source quantifies that sensitivity. Any figure taken from here should be attributed to the study by name and year and never presented as a measured count, because by construction it cannot be measured, and that is the problem itself.

Regional breakdown No source opened here publishes a state-level or region-level split of the undetected population. The study reports national totals for urban areas and census tracts, and the ten priority monitoring locations it names are places rather than population figures, with coverage listing states including Texas, Ohio, California, Wisconsin, Pennsylvania, Missouri and Idaho without attaching a count to any of them. Splitting the national figure by state population would be proportional allocation and would be wrong in both directions at once, because the gap is concentrated where monitors are absent rather than distributed with population.