When you hear “NDIR methane sensor,” you might expect it to measure reliably in any environment. But we should examine the advantages of the principle separately from the performance needed in the field. Chapter 8 of Technology and Solutions for Demonstrating Carbon Reduction discusses methane sensor calibration, contamination, and environmental effects alongside sensor types. Let’s take one step beyond understanding the principle and look at the actual operating conditions.
What the book suggests distinguishing first
You may wonder whether all past data need to be discarded if an inspection finds a calibration problem. First, look for evidence to assess which period was affected and how. Also consider what can be recovered using inspection and recalibration records. Claims about quantities may need to be limited for periods without sufficient supporting evidence. Keeping records of the calibration chain, standard-gas conditions, and use history helps with these judgments. Records also make it possible to explain what happened if a problem arises.
One error to watch for when setting boundaries is cherry-picking: including only the sources where reductions look favorable and leaving out the rest. This is usually unintentional. It can happen naturally when only the items with available data are included.
Calibration. Record when it was performed, which standard and procedure were used, and the pre-adjustment and post-adjustment conditions separately. The difference between two inspection points alone cannot establish the drift path during the intervening period. Review interim checks, environmental logs, and recalibration data together; if evidence for the affected period or correction is inadequate, reflect that in the result’s uncertainty or restrictions on use.
In mechanically ventilated barns, check actual airflow changes caused by static pressure and operating status, as well as contamination and wear, rather than relying only on the fans’ rated capacity. For natural ventilation, examine wind direction and speed, temperature differences, and the condition of openings. Assess within the measurement model how much uncertainty in estimated ventilation rates affects the emission rate.
Individual-animal methods have different strengths and limitations depending on the purpose. An open-circuit respiration chamber calculates mass balance from incoming and outgoing air and flow rates, and involves animal acclimation and substantial costs. GreenFeed uses concentration and airflow while an animal visits the hood, so how representative visits are across the day matters. The SF6 method places a permeation tube with a known release rate in the rumen and collects breath and eructation gases using an external sampling device. Rather than ranking the methods by a single accuracy score based on their names, compare recovery rate, calibration, sampling, and performance records. A concentration measured at one point in a barn is not the same measurand as these emission rates.
Non-dispersive infrared (NDIR) uses methane’s absorption of infrared light. It does not have oxygen dependence or catalyst poisoning of the kind associated with catalytic combustion, but it is not free from optical contamination or the effects of temperature, pressure, and moisture. Do not assume it is suitable for long-term farm monitoring based on its operating principle alone. For each product, check whether its detection limit, repeatability, and drift are adequate in relation to the expected concentration and the difference between indoor and outdoor concentrations.
What to look at in the field
Picture a barn where a sensor has been operating for a long time. Even if numbers keep appearing on the display, you still need to check the condition of the optical window and the effects of temperature and moisture. Inspection and calibration records are also needed to show how performance was maintained in the meantime. Products based on the same principle may still have different specifications and maintenance procedures.
Set calibration intervals based on the use environment, the manufacturer’s procedures, actual drift records, and required uncertainty. Even equipment that does not support field calibration can be managed using a combination of periodic comparison with a reference instrument, return-to-lab calibration, and checks with test gas. What matters is not a uniform interval, but records that support performance between calibrations and rules for handling data when an anomaly occurs.
When comparing sensors, check measurement range and resolution, cross-sensitivity to other gases, response time, and drift together. Environmental conditions such as temperature, humidity, and dust, as well as calibration requirements, should also be reviewed. If the purpose is to distinguish small concentration differences, examine by product and operating mode whether the device’s display units and uncertainty fit that purpose. The name of a sensor principle cannot stand in for these conditions.
AI can be used to flag candidate anomalies, estimate short gaps under defined conditions, and explore relationships among multiple variables. It can also help organize data, conduct quality checks, and prepare report drafts. Because the purpose and cost of failure differ for each function, they should not all be evaluated using the same performance metric. State what inputs and assumptions produced a result, and distinguish estimated values from directly observed values.
Questions to ask during investment review
If you are reviewing equipment, I recommend checking the manufacturer’s procedures, operating environment, and actual drift records instead of setting the calibration interval as a single number. If an inspection finds an anomaly, also check which period’s data it affects. What matters is not simply whether calibration took place once, but how the data afterward are managed.
First, separate the evaluation data. If readings from adjacent times at the same farm are randomly mixed, performance that has learned recurring patterns at that site may be mistaken for performance at a new farm. Use chronological separation when the goal is to predict future operation, and farm-level separation when the goal is deployment at other farms. Distinguish the validation data used for model selection from the data used for the final performance check, and run preprocessing and gap correction using rules established from the training data.
AI evaluation data should be separated according to the intended use. To predict future operations, consider chronological ordering; to deploy at other farms, consider separating data by farm. Also distinguish the data used to select the model from those reserved for the final performance check. Compare with a simple baseline model, and examine not only average error but also bias by farm and season and the conditions under which the model fails. This is necessary to explain its actual scope of use.
Questions to leave with before closing the book
Once you understand how a sensor works, you can see both its advantages and the weaknesses to check. Connect that understanding with calibration history and field inspections. It will help you explain the conditions under which the readings can be used.
About the author
Bongjun Lee graduated from the Graduate School of Management of Technology (MOT) at Sogang University. He spent seven years at Infitron Co., Ltd. and 20 years at Honeywell, working for a total of 27 years in gas measurement in service engineering, technical sales, and channel management. He is now the founder of AI Safety Korea Co., Ltd., where he is researching livestock methane monitoring and measurement, reporting, and verification (MRV) systems using sensors, AI, and data analytics.