How to Measure CO2: Indoor Air to Process Control
Carbon dioxide is the one gas where the measuring principle is effectively settled: it does not burn, it is not electrochemically convenient, and it absorbs infrared light strongly at 4.2 µm. That leaves non-dispersive infrared absorption as the practical method across every application from a classroom to a fermentation tank. What changes between applications is not the principle but the range and the accuracy you have to pay for.
Why CO2 is measured optically
A catalytic sensor needs something to oxidise, and carbon dioxide is already fully oxidised - there is no reaction left to detect. Electrochemical cells for CO2 exist but consume a reagent and drift as it depletes. Infrared absorption has neither problem: the molecule has a strong, well-isolated absorption band, nothing is consumed, and the same sensor works at 400 ppm and at 100% vol with only a change of optical path and calibration.
The historical objection was power and cost, both of which came from the infrared source. LED-based emitters removed that objection: average currents in the tens of microamps make battery-powered and wireless CO2 monitoring practical, which is why ventilation sensors have appeared in places where running a cable was never an option.
The mechanism worth understanding is the dual-wavelength scheme. A reference wavelength that CO2 does not absorb travels the same optical path as the measuring wavelength. Ageing of the emitter, dust on the windows and general contamination affect both signals almost equally and cancel in the ratio. Without that arrangement an optical CO2 sensor drifts; with it, calibration holds for years.
The historical objection was power and cost, both of which came from the infrared source. LED-based emitters removed that objection: average currents in the tens of microamps make battery-powered and wireless CO2 monitoring practical, which is why ventilation sensors have appeared in places where running a cable was never an option.
The mechanism worth understanding is the dual-wavelength scheme. A reference wavelength that CO2 does not absorb travels the same optical path as the measuring wavelength. Ageing of the emitter, dust on the windows and general contamination affect both signals almost equally and cancel in the ratio. Without that arrangement an optical CO2 sensor drifts; with it, calibration holds for years.
Ranges and accuracy by application
The single most common specification error is buying accuracy that the application cannot use, or a range that the application will exceed. This table is the shortcut.
Two rules of thumb follow from it. Below about 2 000 ppm the useful accuracy figure is expressed in ppm plus a percentage of reading, because the absolute error dominates at low concentrations. Above a few percent by volume the percentage term dominates and the ppm term stops mattering.
| Application | Typical range | Useful accuracy | What decides the choice | |
| Ventilation and indoor air quality | 0–5 000 ppm | ±50 ppm or ±3% of reading | Long-term stability; drift matters more than initial accuracy | |
| Classrooms, offices, meeting rooms | 0–2 000 ppm | ±50 ppm or ±3% of reading | Resolution around the 800–1 400 ppm decision band | |
| Cold storage, cellars, beverage plants (safety) | 0–5% vol | ±0.1% vol | Fast response and low placement; alarms at 0.5 and 1.5% vol | |
| Greenhouse CO2 enrichment | 0–3 000 ppm | ±50 ppm or ±3% of reading | Stability - drift wastes gas continuously | |
| Incubators and controlled atmosphere | 0–20% vol | ±2% of reading | Humidity tolerance and repeatability | |
| Fermentation, biogas, purge verification | 0–100% vol | ±2–3% of reading | Linearity across the full range; wet gas handling | |
| Emission and stack monitoring | Application-specific | Per regulation | Traceability and regulatory method compliance- |
Indoor air quality: what the numbers actually mean
Outdoor air sits near 420 ppm today, and that is the floor any indoor measurement is compared against. Indoors, CO2 is not primarily a toxicity question - it is a proxy for ventilation. People exhale it, so its concentration tracks how much fresh air per person a room is getting.
The commonly used interpretation: up to about 800 ppm is well ventilated; 800–1 200 ppm is acceptable but shows ventilation working at its limit; above roughly 1 400–1 500 ppm complaints about stuffiness and reduced concentration become common, and this is the band where cognitive-performance studies report measurable effects. Occupational limits are much higher - ACGIH publishes a 5 000 ppm time-weighted average and a 30 000 ppm short-term limit - so a room that feels unbearable is still far from an occupational exposure problem.
For demand-controlled ventilation the specification is therefore modest: a 0–5 000 ppm range, accuracy in the region of ±50 ppm plus 3% of reading, and - more important than either - stability, because a drifting sensor makes a ventilation system oscillate or quietly stop working.
The commonly used interpretation: up to about 800 ppm is well ventilated; 800–1 200 ppm is acceptable but shows ventilation working at its limit; above roughly 1 400–1 500 ppm complaints about stuffiness and reduced concentration become common, and this is the band where cognitive-performance studies report measurable effects. Occupational limits are much higher - ACGIH publishes a 5 000 ppm time-weighted average and a 30 000 ppm short-term limit - so a room that feels unbearable is still far from an occupational exposure problem.
For demand-controlled ventilation the specification is therefore modest: a 0–5 000 ppm range, accuracy in the region of ±50 ppm plus 3% of reading, and - more important than either - stability, because a drifting sensor makes a ventilation system oscillate or quietly stop working.
Safety applications: where CO2 becomes the hazard
CO2 is heavier than air and accumulates in low, unventilated spaces. Three environments account for most incidents: beverage and brewing plants with bulk CO2 supply, cold storage and dry-ice handling, and cellars, pits and tanks where fermentation or natural degassing occurs.
The numbers that matter here are different from ventilation numbers. Around 4% vol (40 000 ppm) is immediately dangerous to life and health; well before that, at a few percent, breathing rate and heart rate rise sharply and judgement degrades. A safety instrument therefore needs a range to at least 5% vol, alarms in the region of 0.5% and 1.5% vol, and a response time short enough to warn someone walking into a filling cellar - sub-20-second T90 is achievable and worth specifying.
One practical point that is often missed: sensor placement follows the gas, not the room. CO2 sinks, so detectors belong low, near the floor and near the accumulation point, which is the opposite of where methane detectors go.
The numbers that matter here are different from ventilation numbers. Around 4% vol (40 000 ppm) is immediately dangerous to life and health; well before that, at a few percent, breathing rate and heart rate rise sharply and judgement degrades. A safety instrument therefore needs a range to at least 5% vol, alarms in the region of 0.5% and 1.5% vol, and a response time short enough to warn someone walking into a filling cellar - sub-20-second T90 is achievable and worth specifying.
One practical point that is often missed: sensor placement follows the gas, not the room. CO2 sinks, so detectors belong low, near the floor and near the accumulation point, which is the opposite of where methane detectors go.
Process control and quality measurement
Above a few percent by volume the application changes character. Fermentation control in breweries and biogas plants, controlled-atmosphere storage of fruit, greenhouse CO2 enrichment, incubators, and purge or blanketing verification all need percent-level measurement, and some need the full range to 100% vol.
Accuracy requirements here are usually looser in relative terms but stricter in stability: a greenhouse dosing system that drifts by 10% wastes gas continuously, and a fermentation monitor that drifts invalidates a batch record. This is where the dual-wavelength reference channel earns its cost, and where recalibration intervals measured in years rather than months change the operating model.
Cross-sensitivity deserves a check in this group of applications. Water vapour is the usual complication - fermentation headspace and greenhouse air are both wet - and the practical answers are a heated or hydrophobic filter, a compensated optical scheme, or both.
Accuracy requirements here are usually looser in relative terms but stricter in stability: a greenhouse dosing system that drifts by 10% wastes gas continuously, and a fermentation monitor that drifts invalidates a batch record. This is where the dual-wavelength reference channel earns its cost, and where recalibration intervals measured in years rather than months change the operating model.
Cross-sensitivity deserves a check in this group of applications. Water vapour is the usual complication - fermentation headspace and greenhouse air are both wet - and the practical answers are a heated or hydrophobic filter, a compensated optical scheme, or both.
Calibration and verification in the field
An optical CO2 sensor does not consume anything, so its calibration question is different from an electrochemical cell: not "has it been used up" but "has the optical path changed". Three checks cover it.
Zero in clean outdoor air. Outdoor concentration is stable and known within a few tens of ppm, which makes fresh air a usable reference for a ventilation-grade sensor. For percent-range process sensors, nitrogen is the proper zero gas.
Span with certified gas. One point near the top of the working range is enough for a linear optical sensor. Choose the point where the decision is made: 1 000 ppm for ventilation control, 1% vol for a cellar safety monitor, 50% vol for fermentation.
Automatic baseline correction - with care. Many indoor sensors assume the lowest reading over a period equals outdoor air and re-zero themselves accordingly. In an office that empties every night this works. In a room that is occupied continuously, or in a greenhouse dosed with CO2, the assumption is false and the algorithm walks the calibration away from reality. Check whether the function can be disabled before specifying the sensor for a continuous process.
Practical interval: for a dual-wavelength sensor in a clean environment, annual verification is usually sufficient, with a visual check of the optical windows at the same time. In dusty or wet installations the filter, not the sensor, sets the service schedule.
Zero in clean outdoor air. Outdoor concentration is stable and known within a few tens of ppm, which makes fresh air a usable reference for a ventilation-grade sensor. For percent-range process sensors, nitrogen is the proper zero gas.
Span with certified gas. One point near the top of the working range is enough for a linear optical sensor. Choose the point where the decision is made: 1 000 ppm for ventilation control, 1% vol for a cellar safety monitor, 50% vol for fermentation.
Automatic baseline correction - with care. Many indoor sensors assume the lowest reading over a period equals outdoor air and re-zero themselves accordingly. In an office that empties every night this works. In a room that is occupied continuously, or in a greenhouse dosed with CO2, the assumption is false and the algorithm walks the calibration away from reality. Check whether the function can be disabled before specifying the sensor for a continuous process.
Practical interval: for a dual-wavelength sensor in a clean environment, annual verification is usually sufficient, with a visual check of the optical windows at the same time. In dusty or wet installations the filter, not the sensor, sets the service schedule.
Sensor technologies for methane, side by side
The unit decides the range; the range narrows the technology. Three families are used for methane in industrial instruments, and each fails in a different way.
Catalytic bead (pellistor). Burns the gas on a heated catalyst and reads %LEL. Cheap, sensitive to nearly all hydrocarbons, and the only practical option for hydrogen. Requires oxygen, is poisoned permanently by silicones and sulphur compounds, cannot resolve ppm concentrations, and drifts enough to need calibration several times a year.
NDIR (optical). Measures infrared absorption and reads ppm, %LEL or %vol depending on the optical path. Needs no oxygen, cannot be poisoned, tolerates over-range exposure without damage, and holds calibration for years. Blind to hydrogen and acetylene, and sensitive to condensation on the windows if the design has no compensation.
Semiconductor (metal-oxide). Low cost and very sensitive at low concentrations, which makes it attractive on paper for leak search. In practice its output depends heavily on humidity and temperature, it responds to almost any reducing gas, and it drifts - acceptable for a consumer alarm, not for a measurement that has to stand up in a compliance record.
For the three questions at the top of this article the mapping is simple. "Is there a leak?" - ppm-range NDIR. "Is this atmosphere about to explode?" - %LEL, optical where contaminants or low oxygen are possible, catalytic where hydrogen is the target. "How much gas is in this pipe?" - %vol NDIR, because nothing else stays linear to 100%.
Catalytic bead (pellistor). Burns the gas on a heated catalyst and reads %LEL. Cheap, sensitive to nearly all hydrocarbons, and the only practical option for hydrogen. Requires oxygen, is poisoned permanently by silicones and sulphur compounds, cannot resolve ppm concentrations, and drifts enough to need calibration several times a year.
NDIR (optical). Measures infrared absorption and reads ppm, %LEL or %vol depending on the optical path. Needs no oxygen, cannot be poisoned, tolerates over-range exposure without damage, and holds calibration for years. Blind to hydrogen and acetylene, and sensitive to condensation on the windows if the design has no compensation.
Semiconductor (metal-oxide). Low cost and very sensitive at low concentrations, which makes it attractive on paper for leak search. In practice its output depends heavily on humidity and temperature, it responds to almost any reducing gas, and it drifts - acceptable for a consumer alarm, not for a measurement that has to stand up in a compliance record.
For the three questions at the top of this article the mapping is simple. "Is there a leak?" - ppm-range NDIR. "Is this atmosphere about to explode?" - %LEL, optical where contaminants or low oxygen are possible, catalytic where hydrogen is the target. "How much gas is in this pipe?" - %vol NDIR, because nothing else stays linear to 100%.
Integration: what to look for beyond the range
Digital output. A sensor that returns a linearised, temperature-compensated concentration over a serial interface removes an entire analogue design problem from your board.
Warm-up and duty cycling. For battery devices, how fast the sensor can wake, measure and sleep matters more than its steady-state current.
Temperature range and compensation. Cold storage and outdoor enclosures push sensors outside laboratory conditions; specifications valid from –40 °C are not universal.
Certification. If the installation is in a hazardous area, the sensor needs an Ex component certificate - intrinsically safe designs based on optical radiation protection are what allow an infrared emitter inside a certified instrument.
Filter and ingress protection. Dust and condensation are the two field killers of optical windows, and a removable filter that does not require a metal-ceramic sinter simplifies both maintenance and certification.