Ground sensing: particulate matter from low-cost sensors

A satellite sees the column above a location or the heat of a flame. It does not see how much smoke stands in a street, and in the afternoon over South-East Europe it sees nothing at all.

The ground layer measures where people, power and assets are, on street lamps in settlements and it measures continuously. This division of labour is stated openly: deep forest remains a matter for satellite and patrol; the boxes measure the smoke that reaches settlements.

The optical particulate sensor draws air through a chamber and counts particles from the scattering of a laser beam. Every box carries two of them as channel A and B, the same principle as the market leader among low-cost networks: the displayed value is the mean of both channels, both are always stored separately, and if one deviates beyond a threshold the box sets a divergence flag. Wear thus becomes visible without anyone driving out or placing a reference beside it.

Added to this are temperature, relative humidity, air pressure and a channel for volatile organic compounds. Pressure per station makes inversion weather situations detectable; the VOC channel is a second, independent smoke trace, particularly for smouldering fires, which release few particles and much gas.

Optical particle counters measure scattered light, not mass. At high humidity particles take up water and swell; fog droplets are counted along. Without correction such a sensor regularly reads too high in a valley-basin winter, precisely when the number matters.

That is why every box measures for at least two weeks beside an official reference station before field use. From this come slope, offset and a humidity-dependent correction model following the published approach of the US environmental agency for low-cost particulate sensors.

The model is versioned. A change of model produces a version step in the dataset, not a silent change of old values. Because the raw data is preserved unchanged, an improved correction model can also be applied retrospectively.

Three rules decide whether such data is scientifically usable.

First: raw data is inviolable. Every measurement is stored unchanged and append-only. Corrections, quality flags, humidity correction, calibration factors, are separate, versioned layers above it, each with documented methodology. Nobody ever “improves” a raw time series.

Second: the methodology is public. Sensor type, measurement cycle, correction models, calibration history per station and the known limits accompany every dataset.

Third: a measurement with low-cost sensors is an indicative measurement and a complement to official monitoring stations, never a replacement. That is how it is presented in the platform, and that is how it must be presented to authorities. Simulated demonstration stations are marked as simulated in the interface and in every export and appear in no published dataset.

Planning and prototype. The platform already carries stations and measurement series in its data model and in the research interface. Own boxes appear publicly only once the first one sends real values from the field; the simulated stations used for demonstrations are visible to signed-in agencies only and marked as a simulation.

The operating model follows sensor.community, which grew out of luftdaten.info: a citizen-run particulate matter network that has operated thousands of stations for years on open build instructions, fixed data formats and open data publication. samoEKO adds what is deliberately not provided there: a calibration regime and quality assurance at academic level.