Air quality monitoring looks like a data problem: put a sensor on a pole, read a number, plot it. In practice, it is an engineering problem. The signal you get from a gas or particulate sensor is shaped by temperature, humidity, ageing materials and other gases in the air. Turning that raw signal into data that an environmental manager can defend in front of a regulator takes calibration, quality control, and a well-designed data pipeline.

This article walks through how modern air quality monitoring networks are built, from the physics of the sensor to the API that feeds your dashboards.

Why Air Quality Monitoring Is Harder Than It Looks

A low-power electrochemical or optical sensor does not measure “NO₂” or “PM2.5” directly. It produces a voltage or a light-scattering response, and software converts that into a concentration. Three effects make that conversion unreliable if left unmanaged:

  • Drift: Sensing elements change over time. Electrolytes deplete, optical chambers collect dust, and the baseline slowly moves. A sensor that was accurate in spring may read consistently high or low by winter.
  • Cross-sensitivity: Many gas sensors respond to more than one compound. An NO₂ cell can react to ozone, and a CO cell can respond to hydrogen or ethanol. Without correction, one pollutant shows up inside another’s reading.
  • Meteorological effects: Temperature and relative humidity alter sensor response directly. For particulate sensors, high humidity makes particles absorb water and grow, which inflates optical readings unless it is modelled.

None of these problems means air quality sensors are unusable. They mean the raw output cannot be trusted by itself, and that the quality of the data depends on everything done around the sensing element. These limitations are also documented in a 2019 review of low-cost sensor performance.

Reference Stations vs Dense Sensor Networks

Regulatory reference stations are the benchmark. They use certified analysers, controlled shelters and trained operators, and they produce the data that legal limits are checked against. They are also expensive to install and maintain, so a city or an industrial site usually has very few of them.

That creates a coverage gap. Pollution is highly local: a road junction, a loading dock or a stockpile can have a very different profile from a station a kilometre away. Dense networks of compact monitors address this in three ways:

Spatial coverage: more measurement points per unit area show gradients that a single station cannot.

Response time: continuous, short-interval readings from the source area allow alerts within minutes rather than reports after the fact.

Cost of deployment: smaller units with lower power needs can be installed on existing infrastructure, which makes it practical to monitor a fence line or a port perimeter.

The two approaches are complementary. Reference stations anchor the network’s accuracy, and denser monitoring shows where, and when, to look more closely.

How Sensors Reach Near-Reference Performance

Near-reference sensors are not a different physical principle. They are the result of a disciplined process applied to the same sensing technologies. The main steps are:

  • Factory calibration: Each sensor is characterised under controlled conditions across a range of concentrations, temperatures, and humidity levels. This produces the initial compensation models for cross-sensitivity and meteorological effects.

  • Co-location with reference analysers: Units are then run next to certified reference instruments in real ambient air. Comparing the two datasets validates the models under realistic conditions and reveals residual bias that a laboratory cannot reproduce. Co-location is the step where the numbers become defensible.

  • Remote baseline and span adjustment: Because drift is inevitable, the calibration cannot be a one-off event. Networks that stay accurate over years apply baseline (zero) and span (sensitivity) corrections remotely, using reference data or statistical methods, instead of sending a technician to every site.

  • Continuous QA/QC routines: Automated checks flag stuck values, out-of-range readings, sudden steps, unexplained deviations between neighbouring units, and sensor health indicators. Data is validated, flagged, or corrected before anyone relies on it.

Professional-grade air quality monitors such as Kunak AIR Pro are designed around exactly this chain, which is why they are discussed as near-reference instruments rather than simple indicators. Ultimately, data quality depends less on any single component than on how consistently the whole calibration and validation process is applied across the fleet and over time.

The Data Pipeline: From the Pole to Your Systems

Accurate readings are only useful if they reach the right people and systems reliably. A typical IoT environmental monitoring architecture has four layers:

  • Connectivity: Monitors need to work where infrastructure is uneven: 4G for flexible, wireless deployment, Ethernet for fixed sites, and Modbus or other industrial protocols for integration with local equipment. Local buffering matters too, so that a network outage does not become a data gap.

  • Cloud processing: Raw signals are compensated, validated, and aggregated (for example, into hourly and daily averages that match regulatory reporting periods). Keeping this logic centralised means a correction can be applied to the whole network at once and history can be reprocessed when a model improves.

  • Alerts: Threshold and rule-based notifications (by pollutant, location, wind sector or duration) move the system from reporting to operational response.

  • Integration: An open API lets the data flow into SCADA systems, business intelligence tools, digital twins or internal compliance reports without manual exports.

A dedicated air quality data platform such as Kunak AIR Cloud covers this chain in one place. The design principle holds for any stack, however: treat the measurement and the software as one system. If the cloud layer cannot trace a value back to its calibration state and QA flags, the numbers lose much of their value as evidence.

What “Good Data” Means: Standards and Regulation

“Accurate” is not a useful specification on its own. Several frameworks define what performance is expected, and in which context:

CEN/TS 17660: This European technical specification sets out performance evaluation for air quality sensor systems, in separate parts for gases and particulate matter. Class 1 is the most demanding tier, intended for sensor systems that can be used for indicative measurements.

MCERTS for indicative monitors: The UK Environment Agency’s MCERTS scheme includes certification for indicative ambient monitors, giving buyers independent assurance beyond the manufacturer’s own claims.

US EPA performance targets: The EPA publishes performance targets and testing guidance for air sensors used in supplemental and informational monitoring, with metrics such as precision, bias, linearity and error. Its Air Sensor Toolbox is a good starting point for evaluation protocols.

Independent evaluations: Public test programmes, such as the South Coast AQMD’s AQ-SPEC, compare sensor performance in field and laboratory conditions and are useful for cross-checking any claim.

The regulatory context is also moving. The EU’s Ambient Air Quality Directive 2024/2881 tightens limit values for 2030, bringing them closer to the WHO’s 2021 Global Air Quality Guidelines. For PM2.5, for example, the annual EU limit value drops to 10 µg/m³, while the WHO guideline is 5 µg/m³. Tighter limits mean smaller margins, and smaller margins make measurement uncertainty and good documentation more important. Member states are also expected to make better use of supplementary methods to improve spatial coverage, which opens a clearer role for well-characterised sensor networks alongside reference stations.

For teams comparing options, certification is a practical filter. For example, Kunak AIR Pro is certified under MCERTS as an indicative ambient particulate monitor for PM2.5 and PM10 (certificate CSA MC230418/00), and the manufacturer reports alignment with the Class 1 criteria of CEN/TS 17660. The point is not the product, it is that this kind of independent evidence is what separates a monitoring network from a collection of sensors.

Real-World Use Cases

Different sectors use the same technology for different questions:

Industrial fenceline monitoring: Monitors placed around a plant boundary show whether emissions reach neighbouring areas, and combined with wind data, they help identify the source of an episode.

Ports: Ship traffic, cargo handling and trucks create overlapping sources. Distributed monitoring helps separate them and supports mitigation planning.

Mining and quarries: Dust is the main concern. Real-time particulate data lets operators adjust watering, blasting schedules or haulage before limits are exceeded.

Cities: Dense networks complement reference stations to map hotspots near schools, traffic corridors or low-emission zones, and to evaluate the impact of mobility policies.

Data centers: Gaseous contaminants and particulates can affect equipment reliability, and indoor and perimeter monitoring supports facility management and audits.

A Practical Checklist for Planning a Deployment

Before choosing hardware, it helps to answer the following questions:

What is the decision? Compliance reporting, operational control and public information need different levels of data quality.

Which pollutants and averaging periods matter? Match them to the applicable regulation or permit.

What evidence of performance exists? Look for third-party certification, published co-location results, and independent evaluations, not just datasheets.

How is calibration maintained? Ask whether baseline and span corrections can be applied remotely, and how often they are verified.

How are data quality and traceability handled? You should be able to see which readings were validated, flagged, or corrected, and why.

How does the data leave the platform? Check the API, protocols (4G, Ethernet, Modbus), and compatibility with your SCADA, BI, or reporting tools.

What is the siting plan? Height, distance to sources, wind patterns, and power availability all affect representativeness.

Who owns the process? Assign responsibility for maintenance, data review, and escalation before the first unit is installed.

Air quality data only becomes useful when its limits are understood and managed. Teams that treat monitoring as an engineering system, covering the sensor, the calibration, the pipeline and the standards, end up with data they can act on and defend.

This article was published under HackerNoon's Business Blogging program.