When a sensor graph shows a gap, it is tempting to fill it in quickly. But I suggest that we also look at why the gap appeared. That is the topic of carbon data quality in Chapter 8 of Technology and Solutions for Demonstrating Carbon Reduction. Missing sensor data may be a simple inconvenience, but they may also signal something related to the result, such as a fault or a pause in feeding.
A distinction to make first as you read the book
It helps to treat the method used to fill missing data as part of the data record. Document the length of the interval estimated and the rules used. It can help to set processing criteria in advance so that, after seeing the results, you do not choose only the method that is favorable. Changing an observed value is not inherently wrong. But the reason and scope of the change need to be traceable. Separating original data from imputed data also helps others review the results.
Reporting (R) preserves context and calculations. It documents the entire process: which rules were applied to source data and how the conclusion was reached. This includes how missing values were filled, the criteria used to exclude outliers, which model and version were used, and how uncertainty was estimated. A report’s quality is judged by reproducibility, not by how impressive its conclusion sounds.
Season. If intake, feed composition, and ventilation conditions differ between summer and winter, a simple before-and-after difference will mix in seasonal effects. Do not conclude that the difference is entirely a technology effect or entirely a seasonal effect; assess whether the comparison design and adjustments are valid.
Quality control (QC) consists of checks for consistency, completeness, and errors during data collection and estimation. Quality assurance (QA) includes reviews of whether the system and results meet defined quality objectives. Prevention at the planning stage, checks during implementation, and review of results should complement one another. Specific roles and procedures should follow the applicable guidance; do not narrow QA to prevention alone or QC to corrections after the fact alone.
Concentration is the proportion of a gas in air, while an emissions rate is the mass emitted per unit of time. To measure or estimate emissions, connect concentration with flow rate and a measurement model suited to the purpose. Even if concentration falls, an increase in air movement means you cannot conclude that the mass emissions rate changed in the same direction. When converting emissions rates over time into total emissions, you also need to account for the measurement period and how missing intervals are handled.
If the time reference has not been confirmed, first estimate the size of the error and delay and assess their impact on the analysis. An error that is small for daily aggregation may be significant when analyzing events at a minute-level resolution. If the data cannot be recovered, exclude them from combined analyses at that time scale or limit the scope of interpretation, and document the reason.
AI can be used to identify candidate anomalies, estimate short missing intervals under defined conditions, and explore relationships among multiple variables. It can also help organize data, perform quality checks, and draft reports. The purpose and cost of failure differ across these functions, so do not assess them all using the same performance metric. Disclose the inputs and assumptions used to produce a result, and distinguish estimates from directly observed values.
What to look for in the field
Think about a brief communications outage and an equipment failure. Both may appear as gaps on the screen, but that does not mean they should be filled in the same way. Examine the reason and duration of the missing observations and the operating conditions at the time. If values have been imputed, distinguish them from directly measured values.
AI can organize observations and estimate missing values. Its value does not lie in hiding that a value is estimated; it lies in making clear what data and assumptions produced a value that can be trusted, and within what scope.
Separate evaluation data first. If data from adjacent times at the same farm are randomly mixed, performance that has memorized recurring patterns at that site may be mistaken for performance at a new farm. If the purpose is to predict future operation, separate by time order; if the purpose is deployment at other farms, separate by farm. Distinguish validation data used for model selection from data used for the final performance check, and apply preprocessing and missing-data imputation using rules established from the training data.
Finally, consider privacy and data rights. Farm data may be linked to information about animals, production performance, and finances. If such data are reused for finance, insurance, or marketing beyond the purpose of emissions verification, farms may lose trust. As mentioned in Chapter 1, losing trust also means losing data quality. Purpose limitation, separate consent, differentiated access rights, and a process for raising objections matter as much as technical security. Chapter 12’s contract discussion and Recommendation 12 in the policy proposals develop this design further.
Questions to ask in an investment review
If you receive data, ask about the missing-data rate and the rules used to exclude or impute values. It also matters how much the conclusions change when those rules are applied. It is easy to overlook whether the clocks on sensors, feeders, and ventilation controllers are aligned. Combining data from different times as if they described the same situation can change the interpretation.
If a control group is difficult to establish, repeated baseline measurements can help identify variability and time trends, but they should not be treated as having removed all confounding. Even when a change is smaller than the natural variation in individual observations, the average effect can be estimated with enough independent repetitions. Conversely, an apparently large difference may reflect seasonal or growth effects. Present the effect estimate and confidence interval alongside the limitations of causal interpretation.
Record results that are lower than expected or show no clear effect as well. Including observations that differed from expectations and a review of possible causes in the report can help prevent the same problem from recurring in follow-up trials.
A question to leave with before closing the book
Sometimes a record that explains gaps and the reasons for handling them is more useful than a smooth graph. Do not hide missing intervals; document what data were used to fill them and the scope of that imputation. That explanation becomes the starting point for the next review.
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.