Abstract. Accurate and consistent monitoring of anthropogenic combustion is imperative because of its significant health and environmental impacts, especially at city-to-regional scale. Here, we assess the performance of the Copernicus Atmosphere Monitoring Service (CAMS) global prediction system using measurements from aircraft, ground sites, and ships during the Korea United States Air Quality (KORUS-AQ) field study in May to June 2016. Our evaluation focuses on CAMS CO and CO 2 analyses plus two higher resolution forecasts (16-km and 9-km horizontal resolution), to assess their 5 capability in predicting combustion signatures over East Asia. Our results show a slight overestimation of CAMS CO 2 with a mean bias against airborne CO 2 measurements of 2.2, 0.7, and 0.3 ppmv for 16-km and 9-km CO 2 forecasts, and analyses, respectively. The positive CO 2 mean bias in the16-km forecast appears to be consistent across the vertical profile of the measurements. In contrast, we find a moderate underestimation of CAMS CO with an overall bias against airborne CO measurements of -19.2 (16-km), -16.7 (9-km), and -20.7 ppbv (analysis). This negative CO mean bias is mostly seen below 10 750 hPa for all three forecast/analysis configurations. Despite these biases, CAMS show a remarkable agreement with observed enhancement ratios of CO with CO 2 over Seoul metropolitan and over the West Sea, where East Asian outflows were sampled during the study period. More efficient combustion is observed over Seoul (∆CO/∆CO % = 9 ppbv/ppmv) compared to the West Sea (∆CO/∆CO % = 28 ppbv/ppmv). This 'combustion signature contrast' is consistent with previous studies in these two regions. CAMS captured this difference in enhancement ratios (Seoul: 8-12 ppbv/ppmv, West Sea: ~30 15 ppbv/ppmv) regardless of forecast/analysis configurations. The correlation of CAMS CO bias with CO 2 bias is relatively high over these two regions (Seoul: 0.64-0.90, West Sea: ~0.80) suggesting that the contrast captured by CAMS may be dominated by anthropogenic emission ratios used in CAMS. However, CAMS shows poorer performance in terms of capturing local-to-urban CO and CO 2 variability. Along with measurements at ground sites over the Korean peninsula, CAMS produces too high CO and CO 2 concentrations at the surface with steeper vertical gradients (~0.4 ppmv/hPa for CO 2 20 and 3.5 ppbv/hPa for CO) in the morning samples than observed (~0.25 ppmv/hPa for CO 2 and 1.7 ppbv/hPa for CO), suggesting weaker boundary layer mixing in the model. Lastly, we find that the combination of CO analyses (i.e., improved initial condition) and use of finer resolution (9-km vs 16-km) generally produce better forecasts.