see attached file

This paper presents an analysis of surface-based RGB video camera images to detect contrails over London during 5 different days. The advantage of using the ground observations is that narrower and fainter contrails not detected from most satellite imagery can be observed and tracked through the early stages of development providing data for comparison with contrail prediction/diagnostic models. In this paper, the approach is limited to otherwise clear skies and comparisons are made to contrails predicted by widely used CoCiP model informed by adjusted ERA 5 reanalysis input. From the 283 flights, the paper finds that in 69% of the cases, the CoCiP correctly predicts the presence or absence of contrails with a low false alarm rate complemented by a much higher false negative rate resulting in a definitive underestimate of contrail formation by the model. However, the false negatives are more prominent for short-lived contrails and at higher temperatures (lower altitudes), where the ambient temperature and Tsac are close. The CoCiP appears to perform best for persistent contrails at low temperatures. Further comparisons show that the CoCiP tends to underestimate the contrail width by 15%, on average. Errors in formation and width by the model are presumed to be due mainly to assumptions in the model and to temperature and humidity uncertainties in the adjusted ERA 5 profiles, specifically sub-grid scale variability. Future work is proposed to expand the network of cameras, apply the analysis methodology to many more flights, and combine ground and satellite measurements to fully track persistent contrails. 

General Comments

This is essentially a demonstration study that shows how ground-based cameras can be used to assess contrail formation and provide some guidance for modeling the same. It is very limited in actual sampling (only 14 hours in 5 days) and in the range of atmospheric conditions. Satellite analyses indicate that only 15% of persistent contrails occur in otherwise clear skies [Bedka et al., GRL, 2013]. While it proposes to expand the use of cameras in future studies, no way forward past the clear-sky limitation is proffered. Perhaps, that is not a problem as it could be valuable for examining contrails for that clear-sky portion of the natural cloudiness spectrum. While the sampling limitations are mentioned at the end, they should be emphasized more both for the cloud conditions and the number of cases. I would recommend publication after that large concern and others highlighted below are addressed.

Specific comments

Line 103.  How were the ERA5 humidity fields corrected? Does the iAGOS dataset correspond to the same times or is it an average correction applied generally?  That needs to be fleshed out, even if it is in one of the references. What were the magnitudes of the alterations?

Fig. 2 Were the wide contrails in top right and lower right of (a) and (b) not picked up by CoCiP? If so, it should be noted in the text.

Line 202. Perhaps, the sub grid scale variability effect could be estimated by assessing the frequency of correct predictions within the same grid box.

Line 205.  It would be useful to examine the sensitivity of the CoCiP predictions to perturbations in the ERA 5 humidity fields in this paper. That may help determine how far off the RHi is I the ERA5, providing evidence for future corrections and possibly improve the CoCiP accuracy for these cases.

Line 205. Would we expect the humidity fields to be more accurate over this domain than say the middle of the Atlantic, since much of the area includes surface sites where radiosonde profiles taken? If so, would we expect lower detection accuracies in other areas where radiosonde profiles are not available for assimilation into the ERA5?  Or, does the assimilation process damp out the impact of the more accurate data? 

Table 1. How independent are the samples from the 1,619 unique waypoints? I would expect many of those points to be in the same air mass on a given day. For example, the accuracy is very low on 14 Jan compared to the other days.

Fig. 6. Y-axis should be False negative rate? Same in the caption.

Lines 254-259 & Fig. 7. There should be some discussion of the distribution of the points rather than just stating the average difference. In Fig. 7b, there is a significant number of samples near the measured zero-line that have much more spreading than in the prediction, while the others are clustered at or above the agreement line.  Can you shed some light on this? One particular day or type of contrail?

Line 275-279 & Fig. 8b. It is stated that contrails predicted on 5 Nov “appear to show a reasonable agreement.” That may be stretching it a bit. There is one contrail that agrees near the bottom center and another predicted contrail that occurs within an observed one, but it is not clear which observed contrail the top-right prediction is supposed to be matched up with. Moreover, at least half of the observed contrails are not even predicted. Different wording may be more appropriate.

Conclusions. Please note the discussion in the general comments about future use of this approach. Also, it would be useful in the future to add a cloud lidar to the analysis to enable evaluation of the some of the error sources that contribute to disagreements in this paper.