A Performance Evaluation Framework for Continuous Methane Monitoring: Effects of Event Association and Metric Choice

Continuous monitoring (CM) methane leak detection and quantification (LDAQ) systems provide high-time-resolution information on intermittent and persistent emissions, but the reported performance of these systems depends on how monitoring outputs are evaluated. Reported detection events must first be associated with actual emission events before detection classification, detection limit, localization accuracy, and quantification error can be interpreted. Here, we develop and apply a performance evaluation framework for CM methane LDAQ systems using a three-month single-blind controlled-release dataset. The framework defines temporal association, spatial association, concurrent emissions, and excess reported detections, and introduces complementary event-level metrics for quantification agreement, reporting frequency, temporal precision, temporal coverage, and start-time identification. We compared the resulting performance interpretation with two existing protocols when applicable. For the same actual emission and reported detection events, different event association rules change true positive (TP), false positive, and false negative classifications, leading to different 90% detection limit values. Quantification interpretation also depends on metric choice: mean relative error can suggest overestimation even when most individual TP events are underestimated, whereas the bounded Q metric provides a complementary fold-error summary. Event-level metrics further show that TP events differ in temporal coverage and start-time identification. These results demonstrate that CM performance should be interpreted together with the evaluation rules and metric definitions.