Emissions and Leakage Detection
Industrial facilities measure emissions to identify releases that require inspection, repair or immediate control. However, a high sensor reading does not by itself reveal the source, release rate or consequence. Wind may carry a plume away from one instrument and towards another, while start-up, shutdown or maintenance may change the expected reading.
Marketways develops detection systems that interpret measurements within their physical and operating context. The analysis establishes what the monitoring network can observe, tests competing source explanations and directs limited response capacity towards events that warrant action.
Define the decision before selecting the detection method
An emissions programme may need to detect a release, locate its source, estimate its magnitude or verify that a repair worked. These decisions require different evidence and levels of precision.
Marketways begins by defining the event, response and time available. A safety-critical release may require an immediate conservative alarm. A maintenance programme may tolerate more investigation if the system can identify the likely component and quantify the loss.
This distinction prevents one model score from being used for several incompatible decisions. It also defines the performance measures that matter to management.
Map what the monitoring network can observe
A monitoring network observes only the compounds, locations, release rates and periods covered by its instruments. Therefore, the absence of an alert proves little when the network was offline, the plume passed between sensors or the release remained below the detection limit.
Marketways maps sensor position, sampling frequency, downtime, calibration history and detection limits. We compare this coverage with equipment, pipelines, vents and other plausible sources.
The resulting observability map separates a normal reading from an unobserved condition. It also shows where an additional sensor, mobile inspection or alternative technology could materially improve detection.
Establish a baseline for each operating state
Start-up, shutdown, maintenance and changes in production can alter expected emissions. Weather can also dilute, concentrate or redirect a plume before a sensor records it. A single fixed threshold may consequently create false alarms in one state and miss a material release in another.
Marketways links measurements with production, equipment state, maintenance and meteorological data. We estimate the expected range for comparable conditions and retain the operating state beside every anomaly.
An alert then represents a departure from a relevant baseline. It does not merely show that one reading exceeded an arbitrary number.
Separate detection, localisation and quantification
Detecting an unusual concentration answers whether a release may have occurred. It does not necessarily identify the source or quantify the release rate. Research based on controlled releases shows that systems can detect an event while assigning it to an adjacent or incorrect location.
Marketways treats these outputs separately. Detection may combine temporal anomalies across sensors. Localisation tests which candidate source and wind path best explain the observations. Quantification adds dispersion, flow or engineering evidence and reports uncertainty.
This separation prevents a reliable alert from creating false confidence in a weaker source or magnitude estimate. Management can act on what the evidence supports and investigate what remains uncertain.
Test the data used to label past events
Verified releases provide valuable examples, but they are not a neutral record of every event. Investigators usually examine cases that an earlier alarm, inspection or complaint selected. Historical data can therefore underrepresent small, intermittent or poorly covered releases.
Marketways examines how each event was discovered and confirmed. Where appropriate, we supplement historical cases with controlled releases, engineering simulations and investigator-designed scenarios. We also preserve normal operating periods that resemble an event.
These tests reveal whether the model has learned the release mechanism or merely repeated the selection pattern of the previous monitoring system.
Reconcile evidence from different measurement scales
Point sensors, optical surveys, mobile inspections, aerial measurements and satellites observe different spatial and temporal scales. No single technology provides perfect sensitivity, coverage, frequency, source attribution and all-weather performance.
Marketways assigns each source of evidence a defined role. A broad-area observation may identify a site for investigation, while component inspection can locate the equipment that requires repair. Independent measurements can also test whether source-level estimates reconcile with site-level totals.
The combined system uses agreement to strengthen a finding and disagreement to trigger investigation. It does not average incompatible measurements into one apparently precise number.
Prioritise events through consequence and response capacity
Not every anomaly requires the same response. Management must consider the estimated magnitude, duration, location, affected receptors, safety consequence and time available to intervene. Access, shutdown requirements and the number of qualified investigators also constrain the response.
Marketways converts these considerations into explicit triage rules. High-consequence or rapidly escalating cases receive immediate attention. Lower-confidence cases may require corroborating evidence or a scheduled inspection.
The priority is an operational recommendation, not a regulatory or safety determination. Authorised specialists retain responsibility for those decisions.
Use Agentic AI to assemble evidence and coordinate approved action
An investigator may need sensor traces, weather data, equipment state, maintenance history and previous inspection findings before deciding what to inspect. Retrieving this evidence manually can delay the response and produce inconsistent case files.
An Agentic AI workflow can gather permitted records, construct the chronology, test defined source hypotheses and route the case to the responsible team. It can also monitor inspection, repair and verification tasks under approved procedures.
The workflow must expose its sources, missing data and tool actions. It must not certify compliance, close a safety event or change safety-critical controls without authorised review.
Verify performance through controlled and operational evidence
Overall accuracy can conceal the events that matter. A system may appear accurate because releases are rare, yet miss a large share of verified events or overwhelm investigators with false alerts.
Marketways evaluates detection probability, false-alert rate, localisation error, quantification bias, warning lead time and inspection yield. We separate results by release size, weather, sensor coverage and operating state. Controlled tests establish known conditions, while operational review shows whether performance transfers to the facility.
Expected behaviour is defined before results are examined. This discipline prevents a convenient threshold or metric from being selected after a weak result becomes visible.
Confirm that intervention reduces exposure
The purpose of detection is to support timely control. More alerts create no value if teams cannot locate the source, complete the repair or verify that the release has stopped.
Marketways links every confirmed case with inspection, repair and subsequent measurement. We track time to confirmation, time to intervention, verified reduction, recurrence and unresolved cases. A repaired component is resurveyed because completion of the work order does not prove that the release ended.
These measures show whether the programme reduces exposure rather than merely increasing monitoring activity.
A compressor-site example shows why context matters
Suppose a continuous monitor records a methane increase near a compressor site. A fixed threshold produces an alarm, but the reading coincides with changing wind and a planned operating event.
Marketways would first check whether the sensor network could distinguish the candidate sources under those conditions. The analysis would compare the timing with equipment state, maintenance activity and measurements from other locations. If the evidence supports a release but cannot localise it reliably, the workflow would request an approved field inspection rather than name a component with false precision.
After repair, independent measurement would confirm whether emissions returned to the expected range. The case would then improve future detection tests without being treated as a complete record of every possible event.
How Marketways develops an emissions detection system
Marketways defines the management decision, maps observability and builds operating-state baselines before selecting a model. We separate detection, localisation and quantification, test historical selection effects and reconcile measurements across scales.
The engagement may combine Risk Detection, Predictive Maintenance and Reliability and Anomaly and Novelty Detection. Management receives transparent triage rules, evaluation results, investigation workflows and measures that connect alerts with verified intervention.
Related guides and evidence
References
- UNEP: Oil and Gas Methane Partnership 2.0
- US EPA: Next Generation Emission Measurement research
- US EPA: Control Techniques Guidelines for the Oil and Natural Gas Industry
- Single-blind detection, localisation and quantification of methane emissions
- A survey on gas leak detection and localisation techniques