Air quality monitoring data: low-cost sensors, calibration, and publishing

Reference article · 6 min read

When a council asks for neighborhood-level air data, the request often arrives with an assumption: that the agency already measures air quality everywhere and simply has not published it. In practice, regulatory monitoring networks are small and are sited to answer regional questions about attainment and trends, not block-level ones. Low-cost sensors fill a different role. They are inexpensive enough to deploy in quantity and useful for showing spatial and temporal patterns, but they are not reference instruments and should never be presented as though they were.

Running a sensor network is less an equipment purchase than an ongoing data operation. Hardware is usually the cheapest part. The sustained cost sits in siting, power, connectivity, calibration, quality assurance, storage, and the public-facing explanation of what the numbers mean. Agencies that budget only for sensors tend to end up with a map that quietly goes stale.

Decide what the network is for before you buy anything

Write down the purpose, because it determines density, placement, and how much calibration work you sign up for. Common defensible purposes include screening for possible hot spots near roadways or industrial operations, giving residents timely awareness during smoke episodes, supporting complaint response, and building a record that justifies asking the agency with regulatory jurisdiction to place a reference monitor.

What a low-cost network cannot do is determine compliance. Under federal air monitoring rules, regulatory determinations rest on instruments designated as Federal Reference Methods or Federal Equivalent Methods and operated under prescribed quality assurance procedures. Sensor data does not substitute for that. Requirements and designations change, so confirm the current framework with your state or regional air agency and with counsel before making any claim about what your data supports.

Siting, power, and connectivity

• Placement. Aim for unobstructed airflow at roughly breathing height where practical, away from walls, parapets, and roof eddies. Document the coordinates, height, and surroundings for every site.
• Local confounders. Screen for drive-through lanes, loading docks, dryer vents, barbecues, irrigation spray, and idling areas. A sensor measuring one parked truck will be read as a neighborhood finding.
• Power. Mains power or Power over Ethernet is the most reliable. Solar needs battery sizing for short winter days and for smoke-darkened skies, which is exactly when you least want an outage.
• Connectivity. Cellular units with their own data plans avoid attaching unmanaged devices to host networks. If you use site Wi-Fi, involve IT early and place the devices on a segmented network with no route to internal systems.
• Host agreements. Schools, libraries, fire stations, and community organizations make good hosts. Get written permission covering access for maintenance, power and network use, removal, and a named contact.

Calibration, drift, and data gaps

Most low-cost particulate sensors estimate mass from light scattering, which makes them sensitive to humidity and to particle composition. Low-cost gas sensors drift with temperature, age, and cross-interference. Treat correction as a standing program, not a setup step.

• Co-locate. Run new units next to a reference monitor before deployment and return a rotating subset periodically. Ask the agency operating the nearest regulatory site whether co-location space is available.
• Correct, then keep both. Store raw readings immutably and apply corrections as a versioned, reproducible transformation, typically a regression including humidity and temperature terms. EPA's air sensor program has published correction approaches for widely used particulate sensors, including the one applied on its public smoke mapping tool. Being able to reprocess history when a correction improves matters more than getting it right the first time.
• Flag, do not fill. Outages from power, cellular, firmware updates, and heat are normal. Define minimum completeness for any average you publish, and show gaps as gaps rather than interpolating silently.
• Automated quality checks. Watch for out-of-range values, flatlined channels, disagreement between dual channels in the same unit, and a sensor diverging persistently from its neighbors. Retire units that fail repeatedly instead of quietly correcting them further.

From raw readings to something a resident can act on

The Air Quality Index is defined by EPA for specific pollutants, computed from concentrations over specified averaging periods, and reported in named categories from Good through Hazardous, with a near-real-time variant used for hourly reporting. Feeding uncorrected sensor output into that arithmetic produces a number that looks official and is not.

Publish corrected concentrations alongside a plainly labeled category, and pair each category with an action: what sensitive groups should do, whether outdoor activity should be moved or shortened, and where the official advisory lives. Label every view as sensor-derived and non-regulatory in the same visual space as the number, not in a footer.

Publishing, caveats, and smoke events

Draft the caveat language once, with counsel, and reuse it verbatim across the map, the API, and any exports. It should state the data source, that readings are uncorrected or corrected and how, that they are not used for regulatory purposes, and that official health guidance comes from the designated air agency and public health department.

Smoke episodes are the stress test. Traffic spikes, sensors can read outside the range where correction models were fit, and residents treat your page as an advisory. Plan for cached or static fallback pages, pre-written messaging reviewed by your public information officer, and a standing agreement with the regional air district and emergency management so your page points to their advisories rather than competing with them.

Retention, feeds, and integration

Raw sensor data is generally a public record. Set a retention schedule with your records officer under state archive rules, and keep the metadata that makes old data interpretable: site description, height, install and removal dates, firmware versions, and calibration history. Plan for bulk requests by publishing downloads rather than fielding them one at a time.

For integration, coordinate with the regional or state air agency before pushing data anywhere national. EPA provides pathways for sharing sensor data with its public tools, and those relationships usually run through the agency with monitoring jurisdiction.

What to ask vendors and require in the contract

1. Where does raw data live, and can we retrieve all of it, including history, by open API and bulk export at no extra charge?
2. What correction is applied, is it documented, and can we obtain both raw and corrected values for the same timestamp?
3. What are the calibration and warranty terms, expected service life, and per-unit replacement cost after the initial deployment?
4. What happens to our data and our devices if we do not renew? Require that hosted data returns to us in a documented format.
5. What is the security posture of the device and its cloud service, including firmware update handling and any required network access?

Require a quality assurance project plan as a deliverable, name the agency staff responsible for co-location and review, and fund maintenance for the full expected life of the network rather than the first year.

HQT works with state and local agencies on data platforms, integration, and public-facing reporting systems.

Frequently asked questions