| 11 Jun 2026
Investigating aerosol-cloud interactions by fusing DOE ARM observations with ship tracks: A methodology
Abstract. Aerosol-cloud interactions (ACIs) remain one of the largest uncertainties in Earth’s climate system, partly because large-scale meteorology can independently influence both aerosols and clouds, complicating causal attribution. To leverage the rich ground-based measurements from Atmospheric Radiation Measurement (ARM) campaigns and improve aerosol effect attribution, we develop an approach to identify ship-emission-influenced observation cases through source tracing. The method detects local peaks in cloud condensation nuclei number concentrations (NCCN) and applies 24 h backward trajectories to determine whether the sampled air masses intersect ship emissions in the past day. Applying this framework to the ARM Eastern North Atlantic (ENA) observations in 2023 yields several dozen ship influenced cases, including a stratocumulus case in which two ship plumes contribute to a pronounced NCCN enhancement. An increase in cloud fraction and liquid water path, along with a 1 h delay in precipitation, is observed by the comprehensive ARM measurements at the time of the NCCN spike. As large-scale meteorological conditions remain steady, the cloud responses are more likely an aerosol-driven signal rather than meteorology-mediated covariability. Preliminary application to the Marine ARM GPCI Investigation of Clouds (MAGIC) campaign is also discussed. This framework provides a basis for building a multi-year, multi-site library of ship-emission-influenced cloud measurements, offering improved observational constraints for ACI research and model evaluation.
Review for “Investigating aerosol-cloud interactions by fusing DOE ARM observations with ship tracks: A methodology”
The manuscript by Zhang et al. proposes a methodology for detecting clouds affected by ship emissions using ground-based observations. The authors first identify local maxima in CCN number concentration (NCCN) from ARM measurements. They then calculate 24-hour backward trajectories originating from the ARM site at the times corresponding to these NCCN maxima to determine whether they intersect the path of a ship. If an intersection is found, the observed NCCN enhancement is attributed to ship emissions. If clouds are simultaneously observed, the properties of these potentially ship-polluted clouds are subsequently analyzed. Overall, the manuscript is well written, easy to follow, and the proposed methodology is scientifically sound. However, there are several challenges that should be acknowledged and discussed before the manuscript can be considered for publication.
One important question concerns whether the clouds sampled during the CCN peaks have actually interacted with the transported ship-emitted aerosols. It is possible that the clouds simply advected into the field of view of the ground-based sensors from elsewhere, while the enhanced CCN remained confined below the cloud layer. In the example case study, the backward trajectories are initialized at altitudes of only 100, 200, and 300 m above the surface, whereas the cloud base is located at approximately 1 km altitude (Figs. 6 and 7). Furthermore, the reported updraft velocities appear relatively weak, raising questions about whether the transported aerosols could realistically reach cloud base within the required timescale. The authors should discuss this potential limitation and provide justification for assuming that the observed clouds were indeed influenced by the detected ship-emitted aerosols.
In addition, how the authors would use the resulting dataset could be added to the discussion. Given the continuous dispersion and dilution of ship emissions, it is challenging to unambiguously distinguish clouds that are affected by ship emissions from those that are not. Consequently, the conventional approach of comparing polluted and unpolluted cloud populations to quantify cloud susceptibility, and ultimately estimate the effective radiative forcing due to aerosol–cloud interactions (ERFaci), does not appear to be directly applicable using this source-tracing methodology. The authors should comment on the intended applications of the dataset and discuss its limitations in the context of aerosol–cloud interaction studies.
Finally, Section 3.2 appears incomplete, as it lacks a discussion comparable to that presented in Section 3.1. I recommend renaming the section to "Challenges in obtaining ship-emission-affected cases during MAGIC", which would better reflect its current content.
Other minor comments:
1. Lines 25-26: I would particularly recommend refraining from relying on scientific findings from the distant past, as these estimates were based on older models that have since undergone substantial modifications and improvements. For example: “A 4% increase in stratocumulus cloud coverage can offset the warming induced by a doubling of atmospheric carbon dioxide concentrations (Randall et al., 1984).”
2. Lines 57-59: Satellite retrievals of Nd and LWP are also subject to sampling biases arising from retrieval failures when the liquid cloud effective radius is larger than 30 µm, which may introduce artifacts into the derived correlations and potentially lead to misleading physical interpretations (Cho et al., 2015; Choudhury and Goren, 2025).
References:
Choudhury, G., & Goren, T. (2025). Sampling bias from satellite retrieval failures of cloud properties and its implications for aerosol-cloud interactions. Geophysical Research Letters, 52, e2025GL115429. https://doi.org/10.1029/2025GL115429
Cho, H.-M., Zhang, Z., Meyer, K., Lebsock, M., Platnick, S., Ackerman, A. S., et al. (2015). Frequency and causes of failed modis cloud property retrievals for liquid phase clouds over global oceans. Journal of Geophysical Research: Atmospheres, 120(9), 4132–4154. https://doi.org/10.1002/2015JD023161
3. Lines 119-127: Could the authors please clarify in the manuscript why and how the “Virtual Ship Track” dataset was used in the study, when already a high resolution ship emission dataset is available? Virtual ship track data is tied to MODIS overpass time, and may therefore lead to much less number of intersections with the ENA site.
4. Figure 2: I do not fully understand why the ship emission locations marked in the two panels of Fig. 2 are so different. If the emission data is an instantaneous snapshot, for every hour, it should just be one pixel for every ship location. What do these lines represent and why is panel (a) not an extension of panel (b), since (b) is 4 hours behind (a)?
5. Line 201: Could the increase in low cloud cover and liquid water path (LWP) be due to the movement of clouds with high LWP from other regions above the ARM site? Can the authors comment on whether the clouds were stagnant or being transported from elsewhere during the peak in CCN concentrations.
6. Figure 6: I recommend removing the fill color in the pink bar for indicating the Nccn peak. It is interfering with the color keys beneath, confusing its interpretation. I would recommend only highlighting the boundaries of the pink bar.
7. Line 237: On what basis do the authors state that “the boundary layer is topped by stratocumulus”? Please clarify.
8. Line 239: I suggest using “cloud radar reflectivity” instead of cloud reflectivity.
9. Line 288: The term “ship-track” could also mean bright linear cloud features observed in satellite images. I would rather use “ship ground track”.