Air quality monitoring has a structural blind spot. Reference monitoring stations measure the atmosphere precisely and expensively, which means cities have very few of them. A metropolitan area of two million people might have a handful, spaced kilometres apart, and the readings they produce are accurate for the spot where the instrument sits.
What they cannot tell you is how anyone felt.
That gap is where the interest in community health data comes from. If pollution has human costs, and the argument for spending money on cleaner air rests entirely on those costs, then measuring only the chemistry leaves the more persuasive half of the case unmeasured.
This has been tried, successfully, once at real scale. It is worth looking closely at what made it work, because the details are less convenient than the pitch usually suggests.
The precedent: AIR Louisville
Louisville, Kentucky has one of the highest asthma burdens in the United States. Between 2015 and 2017, the city government, a nonprofit called the Institute for Healthy Air Water and Soil, and the digital health company Propeller Health ran a project to find out whether crowdsourced health data could actually change anything.
They enrolled 1,147 residents and fitted their rescue inhalers with electronic sensors that recorded when and where each dose was taken. The data was matched against local environmental conditions.
The results, published in Health Affairs in 2018, were substantial on both fronts.
For participants: a 78% reduction in rescue inhaler use and a 48% improvement in symptom-free days over twelve months. Simply seeing your own pattern changes how you manage the condition.
For the city: the aggregated data produced concrete municipal policy recommendations, including expanding tree canopy, mitigating tree removal, zoning emission buffers around pollution sources, revising recommended truck routes, and building a community asthma notification system.
That is the strongest evidence available that this approach works. It is also the reason to be careful, because the conditions that produced it are specific.
What made Louisville work
Four things, none of which is “people used an app.”
The measurement was objective. An inhaler sensor records a physical event. Nobody had to remember, estimate, or rate anything. That single design choice removes most of the bias that plagues self-reported health data.
Participants were enrolled, not self-selected. Recruitment ran through employers, health plans and nonprofits, targeting people with diagnosed asthma. That gives you a defined population you can describe, rather than whoever happened to download something.
The city was a partner from the start. Louisville Metro Government was in the collaboration, not the audience for a report handed over afterwards. Data does not persuade institutions from the outside nearly as well as it does from the inside.
It was published and peer reviewed. Findings that survive review get treated as evidence. Findings on a company blog get treated as marketing.
What self-reported symptom data can and cannot prove
Here is the part most articles on this subject skip, and skipping it is how citizen science projects get dismissed by the people they are trying to persuade.
Self-selection. The researchers behind Smell Pittsburgh, a citizen science app for reporting pollution odours, state plainly in their own paper that people who install such an app already have high awareness of air pollution, so their sample does not represent the general population. The same is true of anyone who downloads an air quality app. This is not fixable by having more users. More users means a larger biased sample.
Awareness bias. If you know it is a smog day, you are more likely to notice and log a headache. This is the hardest problem in the field, because the app that shows you the air quality is also the app collecting your symptom report. The information flows both ways.
Subjectivity. A 3 out of 5 headache means different things to different people, and different things to the same person in different moods. Sensor data does not have this problem. Symptom ratings do.
Geographic patchiness. Community data is dense where enthusiastic users live and empty everywhere else. The neighbourhoods with the worst air are frequently the ones with the fewest app users, which inverts exactly the pattern you were hoping to reveal.
Regulatory standing. Even calibrated low-cost sensors currently fall short of what is required to support official measures under the revised EU Ambient Air Quality Directive. Self-reported symptoms are far further from that bar. Nobody is going to enforce a limit value based on how a neighbourhood said it felt.
Take those seriously and a useful distinction emerges, one that the low-cost sensor community already uses: some applications need relative signals and some need absolute ones.
Relative works. Trends over time. Before-and-after comparisons around an intervention. Daily cycles. Which of two periods was worse. Raising public awareness.
Absolute does not. Regulatory compliance. Formal health risk assessment. Emissions modelling. Legal claims about specific concentrations.
Community symptom data belongs firmly in the first column. That is a real place to be, not a consolation prize. Most advocacy is an argument about direction and priority rather than about a number to three decimal places.
What this data is genuinely good for
Telling you where to put a real monitor. Reference instruments are expensive, so siting them well matters enormously. A cluster of reports along a freight corridor is a hypothesis worth spending a monitor on. This is the highest-value use, and it treats community data as a pointer rather than as proof.
Before-and-after around a change. A street closure, a new low-emission zone, a plant modification. Relative comparison within the same population over time cancels out a lot of the bias that makes cross-sectional comparison unreliable.
Corroboration. When forty people in the same valley independently report the same thing on the same still winter night, that is real information even though each individual report is imprecise. Convergence across independent observers is the mechanism doing the work.
Political weight. This one is uncomfortable to say out loud but it is true: a chart of concentrations moves fewer councillors than the testimony of residents. The most consequential air quality decision in recent British history came from a coroner’s inquest that named air pollution as a cause of a nine-year-old’s death, not from a monitoring report. Human evidence changes what institutions treat as urgent.
Where SharedSky stands today
Being direct, because this is where similar articles tend to overstate.
SharedSky does not currently publish community maps or aggregated symptom heat maps. What you log is yours. You can see your own history, and you can delete your data at any time. The app is anonymous by default and requires no account.
Community-level insight is a goal, and it is the reason the project exists rather than a feature that quietly shipped. Building it responsibly is harder than building it, and we would rather say that than imply otherwise.
If and when it exists, these are the commitments it will be held to:
- Opt-in, separately. Contributing to any aggregate will be a distinct choice, not a consequence of using the app. Consent buried in a privacy policy is not consent.
- Published method. If we cannot describe precisely how aggregation and suppression work, in language a sceptical reader can check, we should not be publishing the output. Terms like differential privacy and k-anonymity mean specific things, and using them loosely is worse than not using them.
- Suppression thresholds. Areas with too few contributors get no output at all, because a sparse map is a re-identification risk wearing a disguise.
- Stated limitations attached to every output. Any aggregate we publish will carry its sample size and its known biases. Data presented without its limitations is advocacy pretending to be evidence, and it does not survive contact with anyone who knows the field.
What you can do now
None of the above is a prerequisite for the parts that already work.
Keep your own record. Two weeks of logs, including the good days, will tell you more about your own triggers than any city-level map. That is worth doing for its own sake.
Take it to your doctor. A dated symptom record with environmental context attached is genuinely useful clinical material, and it is the single most valuable thing your logs will do this year.
Find the people already doing this locally. Most cities have a clean air group, a cycling campaign or a residents’ association working on traffic and pollution. Organised local pressure is what moves councils, and one committed resident with a coherent argument outperforms a thousand passive data points.
Support reference monitoring. Arguing for more official monitors in under-covered neighbourhoods is unglamorous and it is the intervention with the clearest path to regulatory consequence.
SharedSky is not a medical device and does not provide medical advice. See our Medical Disclaimer.
Sources
- Barrett et al., AIR Louisville: Addressing Asthma With Technology, Crowdsourcing, Cross-Sector Collaboration, And Policy, Health Affairs (2018). https://www.healthaffairs.org/doi/10.1377/hlthaff.2017.1315
- OECD Observatory of Public Sector Innovation, AIR Louisville. https://oecd-opsi.org/innovations/air-louisville/
- Hsu et al., Smell Pittsburgh: Engaging Community Citizen Science for Air Quality. https://arxiv.org/pdf/1912.11936
- European Commission, Air pollution data from citizen-operated sensors improved by novel quality control system (2025). https://environment.ec.europa.eu/news/air-pollution-data-citizen-operated-sensors-improved-novel-quality-control-system-2025-11-06_en
- Evolving trends in application of low-cost air quality sensor networks: challenges and future directions, npj Climate and Atmospheric Science (2025). https://www.nature.com/articles/s41612-025-01216-4
Want to help build clean air awareness in your city? Log your daily baseline in SharedSky and join the community movement today.