Reliable emission monitoring is essential for effective environmental regulation and the operation of carbon markets. However, high-frequency CO2 data from Continuous Emission Monitoring Systems (CEMS) and material-based monitoring often contain inconsistencies arising from operational variability, sensor drift, and data-processing errors. This study develops a transparent statistical framework to screen the quality of CO2 emission data by integrating CEMS measurements with material-based estimates in a coal-fired power plant. A correlation ratio between the two monitoring approaches is used as a process-level indicator, and four statistical tests, Mann–Whitney U, Bootstrap, Levene, and Dip tests, are applied to detect distributional deviations associated with anomalous behavior. Using one year of high-resolution data, we evaluate the influence of reference dataset size, anomaly magnitude, and anomaly duration on detection performance. The results show that approximately 700 reference samples are sufficient to establish a stable baseline. Anomalies corresponding to daily emission deviations of about 4% or higher, when sustained over several days, can be reliably identified as anomalous at the monthly scale. A composite risk score is further developed to support monthly data screening and risk-based verification. The proposed framework provides a practical tool to improve the reliability of emission data and supports more transparent and efficient environmental monitoring and regulatory oversight.
Jia et al. (Sat,) studied this question.