Ambient carbon monoxide measurements from a nondispersive infrared analyser become understandable when we distinguish what the optical cell senses, how the instrument is calibrated, and what an hourly reading can legitimately mean. NDIR is a continuous infrared absorption method for estimating CO in sampled air, not a direct record of where a person travelled or which engine emitted the gas. Its reported value is shaped by calibration, sample delivery and the time interval assigned to it.
A trace rising from 0.2 to 1.8 parts per million may reflect a genuine ambient increase, a calibration exercise, moisture interference, a flow issue or a faulty data transformation. These explanations cannot be decided by the peak height alone. Reliable interpretation begins with gas identity, timestamp, units, operating flags, zero/span performance and sampling circumstances, then proceeds to meteorology and possible sources.
This guide explains infrared absorption, NDIR gas-filter correlation, carbon monoxide versus carbon dioxide, sample conditioning, zero and span challenges, instrument drift, common false readings, averaging and independent checks. Numerical examples are invented teaching cases, not claimed observations. You will learn to translate technical instrument evidence into useful scientific statements while keeping instrument validity separate from source and exposure claims.
Carbon monoxide and the NDIR measurement principle
Carbon monoxide is a molecular gas with infrared absorption features. An NDIR analyser directs infrared radiation through a sample cell and measures how the gas changes the received radiation at appropriate spectral bands. Detector optics, reference arrangements and calibration convert this signal into a CO concentration. The United States reference measurement principle in 40 CFR Part 50 Appendix C describes non-dispersive infrared photometry for continuous ambient CO measurements.
The appendix includes gas-filter correlation as a way to sensitise the photometer specifically to carbon monoxide absorption. A CO-filled filter cell can be compared with an optical path without that filter, helping distinguish characteristic CO absorption. Modern instruments may use related optical arrangements and additional measures to control interference. Read the instrument’s actual manual before assuming a specific gas wheel, filter geometry, optical path length or water trap exists.
An NDIR CO instrument is not the same as a generic carbon dioxide sensor, though both can use infrared absorption. Different gases absorb at different wavelengths, and the optical discrimination and calibration must match the target molecule. An affordable indoor CO₂ monitor cannot be treated as a validated ambient CO analyser merely because both display concentrations in parts per million. Translate gas identity first: one misplaced digit in the chemical formula changes the entire question.
What a concentration number represents
A displayed CO number is an estimate of the gas concentration in air supplied to the measurement cell for a defined interval. It does not report combustion efficiency, total traffic emissions or a commuter’s personal exposure history by itself. Those quantities require other measurements and models. A single instrument can supply strong local evidence without explaining the complete urban transport system around it.
Parts per million for a gas commonly indicates a mixing ratio by volume or amount fraction under appropriate conventions. Carbon monoxide instruments may report in ppm, while datasets sometimes use milligrams per cubic metre. These units should not be compared as though numerically equal. Reference temperature and pressure assumptions matter for gas-to-mass conversion, and official reporting frameworks may specify different reference conditions.
A plot labelled CO=1.0 without units, time basis or validation status is incomplete. Ask whether each point is an instantaneous analyser output, a one-minute mean, an eight-hour running mean or a validated hourly concentration. Also ask whether timestamp labels indicate period starts or ends. Two values displayed at 10:00 may refer to different actual windows; their equal clock labels are not proof of simultaneous sampling.
Infrared absorption made intelligible
Think of light entering a gas cell and encountering molecules that absorb selected infrared wavelengths. More of the target gas generally increases absorption along an appropriate optical path, subject to optical response and calibration. Instruments do not simply divide absorbed light by a universal constant; their circuitry and data processing account for reference signals, calibration curves and design-specific compensation.
A simplified Beer–Lambert teaching relationship is absorbance A = −ln(I/I0), where I is transmitted intensity and I0 is a comparable reference intensity. Suppose I/I0 = 0.90, giving absorbance approximately 0.105. This does not mean the CO concentration is 0.105 ppm. To infer concentration, we need an appropriate path length, molecular absorption response and an instrument calibration. Absorbance is dimensionless; ppm is a different physical quantity.
Infrared absorption is often described using a logarithmic relationship, but actual analyser algorithms may combine filters, chopped beams, gas correlation and electronic transformations. It would be a mistake to reverse-engineer a regulatory concentration from a vague picture of a single lamp and detector. The useful conceptual bridge is that a CO-sensitive optical difference is related to a traceable standard gas concentration through calibration.
Why moisture and carbon dioxide matter
Water vapour and carbon dioxide can alter infrared measurements under certain optical and instrument conditions. The regulatory CO NDIR reference description discusses optical filters, reference cells, dual-beam arrangements, moisture traps and other possible provisions for sensitivity and minimising interference. This does not mean every monitor uses the same hardware or has identical susceptibility. Actual interference performance must be established for the instrument in question.
If a humid evening coincides with a rising CO trace, neither a weather explanation nor a pollution explanation is automatically proven. Examine diagnostic records, inlet condensation, sample conditioning and independent CO measurements. If a moisture trap was serviced around the event, test whether the instrument response changed across the intervention. Weather can also affect genuine pollutant dispersion, which makes moisture-related claims particularly easy to oversimplify.
A faulty sample dryer may change the measurement environment, but it can be equally misleading to assume that every positive excursion during rain is a moisture artifact. Some network designs intentionally manage water through inlet and gas-conditioning strategies, while others use optical interference correction. The correct question is whether the particular system’s verified interference controls operated as intended under those conditions.
Zero challenges and baseline correction
A zero-air challenge checks how the instrument responds when supplied with air intended to contain negligibly small carbon monoxide. If the reported value is significantly nonzero, a baseline offset, zero-source problem, leakage or other issue may be present. A baseline check is evidence about the instrument at the time of the test; it should be interpreted using the site’s approved procedures rather than an invented universal pass/fail number.
In a teaching sequence, zero results are 0.00, 0.01, −0.01, 0.02 and 0.15 ppm. The final check stands out. This pattern suggests investigation, not immediate post hoc subtraction of 0.15 ppm from every earlier ambient observation. Determine whether the zero-air source was appropriate, whether the gas delivery path was stable, and when any change began. The quality decision should document which intervals were affected.
Negative readings near zero are possible after background subtraction even though physical CO concentrations cannot be negative. A recorded −0.02 ppm may reflect instrumental noise or offset rather than negative atmospheric molecules. Unless a documented data procedure requires otherwise, replacing negative readings with zero before averaging can bias low-end statistics upward. Preserve original observations and qualifiers for auditability.
Span checks and instrument response
A span challenge exposes the analyser to a defined nonzero CO concentration to test the continued response of the optical measurement system. It differs from a full recalibration, which establishes or adjusts the response function. An instrument passing a challenge at one concentration provides useful evidence, but that does not guarantee that all other concentrations, transient conditions and sampling periods are correct.
Suppose a nominal 10.00 ppm CO reference produces 9.70 ppm. The signed deviation is −0.30 ppm, and the relative deviation is −0.30/10.00 × 100 = −3.0 percent. This calculation alone cannot establish compliance with a programme’s tolerance. The acceptance criterion, reference uncertainty, gas delivery and measurement stability all matter. Do not borrow a control limit from a different instrument or network.
Now suppose a later 10.00 ppm check gives 10.40 ppm. The signed error is +0.40 ppm, or +4.0 percent. The two checks have different signs, which can matter when investigating calibration drift. A report that reduces both to approximately four percent variation loses information about direction. Keep signed residuals, dates, temperature and maintenance context in a control chart.
A common conceptual error is treating a passing span check as proof that the analyser was perfect throughout the previous week. The check supports a narrower conclusion about performance under its challenge conditions. Sample-line leaks, data acquisition errors and transient interferences can still occur between checks. When the earlier record is questioned, reconstruct the timeline of checks and incidents rather than relying on a single green status icon.
Worked example: distinguish a short peak from a long average
Take a teaching series of eight consecutive one-hour CO values in ppm: 0.3, 0.4, 0.5, 0.7, 1.0, 1.2, 0.6 and 0.3. Their sum is 5.0 ppm-observations and their equally weighted eight-hour average is 5.0/8 = 0.625 ppm, provided all eight hours are eligible. The hourly maximum is 1.2 ppm, which is not the eight-hour mean.
If the 1.2 ppm hour was a calibration-gas interval, it should not silently remain in an ambient average. Depending on the relevant reporting rules, the window may be incomplete and unsuitable for official use. An average of the other seven hours is a defined partial-window statistic but not automatically a valid eight-hour regulatory mean. A simple spreadsheet formula cannot supply missing quality criteria.
Compare two hours with equal means of 1.0 ppm. In one hour all short readings are steady at 1.0; in another half the readings are 0.2 and half are 1.8. The mean is identical, yet the temporal behaviour is not. Report the statistic that answers the question: a validated long-window mean for sustained conditions, or validated short-interval evidence for investigating an event, never one disguised as the other.
Units and a carefully bounded conversion
At approximately 25°C and one atmosphere, an ideal-gas conversion can be expressed as CO mg/m³ ≈ ppm × 28.01/24.45. Under those conditions, 2 ppm corresponds to about 2.29 mg/m³. Carbon monoxide’s molecular weight is approximately 28.01 g/mol. The result is conditioned on molar volume; it is not universally exact across temperature and pressure environments.
For instance, 0.5 ppm converts to approximately 0.573 mg/m³ under that reference convention. Reporting 0.573 as though the analyser itself measured three-decimal accuracy would confuse arithmetic with instrumental precision. Use a sensible number of significant figures and state the reference conditions. If a dataset uses a different standard temperature, obtain its specified conversion rather than copying a value from a teaching sheet.
Mass concentration, mixing ratio and emissions rate are separate quantities. A monitor showing 1 mg/m³ does not tell you the total kilograms of CO released by a road during one hour. Emissions require source activity, exhaust or flow information and appropriate mass-balance assumptions. A concentration is a property of sampled air at a place and time, shaped by release, dispersion and chemistry.
A cautious language model for monitoring claims
A poor summary says the infrared detector proved vehicle emissions were responsible for an alarming rise. A better summary says the station’s validated hourly CO estimate increased from 0.4 to 1.2 ppm, identifies the date and conditions, and describes any supporting evidence without asserting a unique source. If there was no formal causal assessment, explicitly state that the trace cannot by itself identify one.
The related SO₂ fluorescence guide uses emitted light rather than CO’s infrared absorption principle. The ozone photometry guide demonstrates why averaging windows and optical methods must be described precisely. Their similarities are instructive, but neither should be treated as a substitute CO method.
Build a CO instrument audit before interpreting a surprising graph
Begin with the complete timeline. Write down the nominal sampling location, start and end of the reported interval, instrument mode, original value, unit and validation code. Add zero and span responses before and after the event, any alarm, flow or moisture diagnostic, and the maintenance history. An unexplained peak deserves attention, but a peak occurring during a calibration challenge is not necessarily ambient CO. The audit trail prevents a picturesque chart from overruling basic operating facts.
Consider a hypothetical urban monitor with CO near 0.4 ppm through the morning, then a 2.2 ppm reading lasting one minute. An instrument log shows a scheduled 2 ppm gas challenge at the same time. That is strong reason to investigate whether the apparent ambient spike was calibration gas. The conclusion needs the actual mode timing and data-acquisition flags. Merely noticing a similar concentration does not independently prove the source of the signal.
A separate record shows a smooth rise from 0.3 to 1.3 ppm over an hour, with stable diagnostic flow and passing surrounding zero/span checks. Nearby stations show smaller but aligned changes and wind speed decreases. A real atmospheric accumulation hypothesis becomes plausible, though differing station positions and averaging intervals limit direct comparison. A careful analyst reports a supported increase and the contextual evidence, without claiming an individual exhaust pipe has been identified.
What if the rise coincides with condensation on the inlet tubing, a reported sample-pump problem and erratic baseline checks? Those are measurement-quality warnings. The event could still occur in polluted air, but the instrument cannot independently establish the true magnitude for the affected intervals. Use the monitoring programme’s invalidation or qualification process. A meaningful result can be that reliable quantitative determination was not possible, rather than a guessed replacement concentration.
CO, CO₂ and a source-attribution trap
Carbon monoxide is associated with incomplete combustion in many settings, but a CO concentration alone does not reveal the efficiency of a particular engine or appliance. Ambient levels reflect emissions, transport, mixing and removal. Meteorological conditions can make several unrelated sources’ plumes overlap. To claim a traffic contribution, use properly timed traffic context, appropriate background comparisons, meteorology and potentially additional tracers, not one isolated NDIR curve.
Carbon dioxide may also change in combustion-rich air, but CO and CO₂ have different emissions behaviour and atmospheric backgrounds. Their ratio is not a universal fingerprint. A source’s combustion state, dilution, mixing with background air and multiple contributing sources can affect it. Without a specific validated model and measurements, never translate a casual CO/CO₂ comparison into a numerical combustion-efficiency estimate.
Suppose CO increases from 0.5 to 1.0 ppm while CO₂ changes from 430 to 450 ppm. An analyst might calculate a difference ratio of 0.5 ppm CO per 20 ppm CO₂. That ratio describes paired changes for the selected intervals, not necessarily emissions from one source. Background variation and independent uncertainty matter. If the values are not co-located or synchronised, even the ratio may be a misleading construction.
A useful question is whether the evidence establishes co-occurrence, transport direction or source identity. Co-occurrence needs matched valid time series. Transport inference adds wind and site geometry. Source identity requires independent discriminatory information. Naming these levels explicitly prevents explanatory language from outrunning the measurements. It also helps students transfer the same logic to noisy groundwater, climate and laboratory data.
The calibration-gas delivery path can create its own error
A reference cylinder label gives a stated concentration under defined conditions, but the concentration actually reaching the analyser depends on the delivery system. A leak, contaminated regulator, incorrect valve selection or dilution setting can compromise the challenge. When span response appears wrong, check the entire pathway before assuming the optical bench is defective. The reference needs appropriate traceability and uncertainty, not just a plausible printed number.
In a simple dilution example, imagine combining a 20 ppm CO standard stream at 1 litre per minute with verified zero air at 4 litres per minute, assuming ideal mixing and appropriately matched flow units. The nominal resulting concentration is 20 × 1/(1+4) = 4 ppm. This is a mass-balance teaching calculation, not a claim that an actual calibrator’s flows were accurate or leak-free.
If the zero-air flow inadvertently falls to 3 litres per minute while the standard stream stays at 1 litre per minute, the nominal mixture becomes 20 × 1/(1+3) = 5 ppm. An analyser challenged with this mixture might read close to 5 ppm and be operating correctly, even though the operator incorrectly expected 4. The mistake is in delivered concentration rather than the NDIR sensor. This is why flow records, standards and configuration belong to calibration review.
A check should be evaluated only after confirming the delivered challenge gas is appropriate. Otherwise, operators risk altering a correctly performing instrument to agree with an incorrect standard. The result can damage subsequent ambient data. Independent audit procedures are valuable because they can test calibration assumptions rather than merely repeat the instrument’s own routine.
Detecting additive offsets and sensitivity changes
Suppose a zero challenge reports +0.10 ppm while a 10.00 ppm span reports 10.10 ppm. A simplistic two-point model might suggest an additive offset with approximately unchanged slope. Contrast this with a zero of +0.10 and a span of 9.50, for which the response difference is 9.40 for a 10.00 ppm challenge. The second pattern also indicates a change in effective sensitivity under that model.
For the second set, the approximate slope is (9.50−0.10)/(10.00−0) = 0.94 indicated units per ppm, with a 0.10 indicated-unit offset. If an ambient indicator were 1.04, a straightforward inverse model would give (1.04−0.10)/0.94 = 1.00 ppm. The arithmetic is a teaching exercise, not permission to correct official data outside an approved calibration procedure.
One temptation is to subtract the current zero from every value and call the readings repaired. That can fail when sensitivity changes, interference varies with sample composition, or the offset drifted during the period. Recalibrations also have time boundaries. Document the original data, the condition of the instrument, any approved correction and the revised value separately so later users know precisely what changed.
A zero/span pattern can narrow possible problems, but not uniquely diagnose a component. A gas-delivery error may mimic a gain change. Water interference may vary with ambient conditions even when a dry span passes. Data-acquisition software may use an old calibration coefficient despite acceptable optical response. Use repeated checks and independent evidence rather than an instrument-failure story generated from one arithmetic pair.
Data completeness and eight-hour calculations
A CO dataset may contain minutes, hours and running multi-hour summaries. An eight-hour rolling average is not the same as a fixed daily eight-hour block. The maximum daily rolling eight-hour average depends on how all eligible overlapping windows are constructed, which hours are valid and the programme’s completeness conventions. You cannot infer it from the highest hourly concentration multiplied or divided by eight.
Take ten consecutive hourly teaching values in ppm: 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0 and 1.1. The first eight-hour mean is (0.2+0.3+0.4+0.5+0.6+0.7+0.8+0.9)/8 = 0.55 ppm. The next window, hours two through nine, has mean 0.65 ppm; the third, hours three through ten, has mean 0.75 ppm. Rolling windows reveal progression that a single fixed window can hide.
These values are illustrative and assume complete, equal-duration validated hours. If hour five is invalid, a spreadsheet may average the remaining seven and display a number, but the official completeness requirement must decide whether a rolling window is usable. Retain the number of valid contributing hours and the eligibility decision. A visible average without its denominator can be dangerously persuasive.
Also distinguish local daily boundaries from UTC and daylight-saving transitions. A station operating in a jurisdiction with clock changes may have anomalous apparent hour counts on transition days. Software must use explicit timestamps and time zones, rather than treating every day as a naive 24-row spreadsheet. This is a data-management issue that can change which observations belong to a regulatory or analytical period.
Why the highest CO figure can be the wrong headline
A short peak is often visually dramatic, but the relevant question may concern the pattern of sustained exposure or a long-window concentration metric. A minute peak, an hourly average and an eight-hour statistic answer different questions. Selecting whichever produces the largest number without telling the reader the interval turns scientific measurement into rhetoric. The faithful summary names the window and makes no silent substitution.
If a one-minute value is 4.0 ppm while the surrounding 59 minutes are approximately 0.4 ppm, the whole-hour concentration is not 4.0 ppm. Under a deliberately simple equal-minute average, the sum is 4.0+59×0.4=27.6 ppm-minutes, yielding 27.6/60 = 0.46 ppm. Actual equipment response and short-window data validity still matter, but the arithmetic exposes how a brief maximum differs from the mean.
Conversely, an hourly mean of 0.8 ppm does not imply every minute stayed under 0.8. A single long-window statistic can conceal significant variation. Readers interested in event analysis should seek validated high-resolution records, while readers interested in official comparisons should use the relevant validated averaging period. Neither can substitute for the other simply because both are expressed in ppm.
An important follow-up is representativeness. A monitor near a road may describe a spatially specific environment. Another station far from traffic may capture broader background. Disagreement can be scientifically meaningful rather than evidence of defective sensors. The two sites may both be correct for their intended sampling locations and periods.
Distinguish ambient concentrations from personal exposure
Ambient CO monitoring is an environmental observation at a location. Personal exposure depends on time spent indoors, in vehicles, near combustion appliances, at different elevations and in other microenvironments. One station’s eight-hour average cannot reconstruct every person’s inhaled concentration. For an exposure assessment, define the person’s movements, microenvironment concentrations and measurement uncertainty; do not imply a fixed-station record is a personal dosimeter.
A regulatory comparison also requires the correct authority and effective rules. Limits and statistical forms may differ among jurisdictions and purposes. This guide is methodological, so it avoids making a universal claim about a specific safe or unsafe ppm threshold. To evaluate compliance, identify the applicable CO standard, averaging definition, completeness requirement, instrument designation and data validation status.
An indoor CO alarm is not the same as an ambient reference-grade NDIR instrument. Alarms are designed around a specific warning purpose and response behaviour, while ambient stations aim to measure concentrations to prescribed performance standards. Their readouts and threshold settings should not be treated interchangeably. During a suspected acute indoor CO hazard, follow emergency safety guidance rather than attempting to diagnose danger using an educational graph calculation.
Independent transfer problems and checked answers
Problem one: a nominal 5.00 ppm span standard produces 4.80 ppm. The signed error is −0.20 ppm, or (−0.20/5.00)×100 = −4.0 percent. Is it acceptable? Only the monitoring programme’s applicable criterion can decide. Do not import a tolerance from an unrelated instrument or assume that four percent is either universally trivial or unacceptable.
Problem two: six valid ten-minute values are 0.2, 0.2, 0.3, 0.3, 0.6 and 0.8 ppm. They sum to 2.4 and their equal-duration hourly mean is 2.4/6=0.4 ppm. The maximum ten-minute value is 0.8 ppm. A report should preserve that interval distinction rather than labelling the 0.8 result hourly.
Problem three: a CO reading is 3 ppm under approximately 25°C and one atmosphere. The reference conversion is 3×28.01/24.45 ≈ 3.44 mg/m³. The arithmetic should be accompanied by its temperature and pressure assumptions. The instrument did not magically become more precise because the converted value includes two decimals.
Problem four: two monitors show 0.8 ppm at 14:00. One reports the hour from 13:00 to 14:00 and the other from 14:00 to 15:00. Are the concentrations coincident? No, not on the information given. Reconstruct explicit interval boundaries and evaluate clock settings before comparing the numbers or calculating correlation.
Problem five: a zero result changes from 0.00 to 0.12 ppm after equipment service, while a 10.0 ppm span result changes from 10.0 to 10.12 ppm. The pair is consistent with a simple offset change but does not prove that is the only effect. Check the zero source, gas challenge path, sampling conditions and all relevant quality requirements before altering reported data.
Problem six: an eight-hour series has seven valid hourly values, all 0.5 ppm, plus one invalid period. The average of available hours is 0.5 ppm, but the full eight-hour regulatory statistic cannot be assumed valid without a completeness rule. Distinguish a mathematical mean from a qualifying official result, and preserve the coverage note.
Problem seven: a monitor’s CO curve and a road’s traffic count rise simultaneously. Does that prove vehicles generated the whole concentration? No. A plausible source relationship is not the same as quantified attribution. Compare spatial background, wind, time alignment, alternative sources and instrument quality to develop the hypothesis further.
Problem eight: a zero-corrected sample appears twice in a CSV, with identical timestamp and value. Should both rows be included in an hourly mean? Not without identifying the duplication. A duplicated record changes a naïve equal-row average by effectively increasing its weight. Use a unique time index or other source identifier, verify the correct reading and document how the duplicate was resolved.
Data engineering mistakes that change monitoring meaning
A dashboard may store CO in ppm but export a column labelled ppb. If a value of 0.5 ppm is displayed as 0.5 ppb without conversion, the dashboard understates the nominal concentration by a factor of one thousand. Conversely, 500 ppb equals 0.5 ppm. Unit checks should be part of every data pipeline test, especially when combining data from instruments designed for different concentration ranges.
Rounding can also distort low-end patterns. Ten readings of 0.04 ppm rounded individually to one decimal place become ten zeros, and averaging them yields zero. Their unrounded mean was 0.04 ppm. Calculate aggregates at appropriate internal precision, then round the final display sensibly. A zero produced by a display format does not necessarily mean the sampled air contained no CO.
If a data system repeats the last valid measurement during a communications outage, a long artificial plateau may appear. Compare original instrument timestamps, transmission timestamps, validity flags and record counts. An atmospheric explanation is not needed for an acquisition artefact. But a flat line alone is not enough to prove the artefact; confirm with system diagnostics and independent records.
Common questions with precise answers
Why can NDIR instruments disagree during humid weather? Designs differ in optical discrimination, sample treatment, calibration and environmental response. Investigate each model’s certified performance and actual conditions. Do not declare all moisture-related variation interference, because weather also changes dispersion and actual CO concentrations. Comparisons must use aligned time windows and independent quality checks.
Does an NDIR result directly measure a vehicle’s exhaust emissions rate? No. It estimates the concentration in the air being sampled. Emissions rate requires information about exhaust flow or source activity and a suitable mass-balance approach. Atmospheric dispersion mixes source emissions with background air before a monitor observes the result.
Why do low-level CO readings sometimes become negative? A processed signal can cross below zero because of baseline correction and noise; physical CO concentration cannot. Preserve the numerical observation and qualifiers according to the applicable data-quality process. Automatically setting negative values to zero can systematically alter low-concentration averages.
Does successful calibration make all earlier data exact? No. Calibration relates instrument response to standards under specified conditions; routine and independent checks provide evidence for stability. Interference, sample leaks, timing errors and unrecorded faults can still affect results. A trustworthy scientific record keeps the chain of evidence and makes uncertainty visible.
Why is the eight-hour rolling average not just the mean of today’s first eight hours? A rolling calculation shifts its window through eligible hours. Its maximum can occur later, and completeness rules can alter which windows count. Define each start and end time explicitly before calculating or comparing the results.
How should I explain an unexplained short CO peak? State the observed value, unit, averaging period, station, validity status and available contextual evidence. If equipment and transport checks cannot discriminate a cause, say attribution remains unresolved. A precise admission of uncertainty is preferable to a polished invented source story.
Transfer the skill from instruments to technical reading
Learning to interpret CO monitoring is also an exercise in scientific language. The words measurement, estimate, source, exposure, calibration, concentration and emission each refer to different concepts. Confusing them invites conclusions that sound persuasive but are not supported. Read the original method, then trace how a signal became a reported figure before extending the claim beyond the instrument.
A valuable counterpart is SO₂ ultraviolet fluorescence, whose physical sensing principle is different despite comparable zero/span vocabulary. You can also compare optical PM₂.₅ sensors to see why instrument-specific interference matters. Transfer the questions, not the numerical constants or assumptions.
A strong final answer to a real measurement question includes a reliable concentration estimate, its exact sampling period, known instrument limitations, independent corroboration where available and a clear statement of what cannot be inferred. That is the standard for translating a technical trace responsibly into a public explanation. It protects both scientific meaning and the reader’s ability to make a sound decision.
Institutional sources
40 CFR Part 50 Appendix C — Carbon monoxide measurement by non-dispersive infrared photometry.