We thank the reviewer for pointing out the relevant work by A. Fuller et al. (2025), Using Multitask Machine Learning to Type Clouds and Aerosols from Space-Based Photon-Counting Lidar Measurements. Indeed, we were not aware of the aforementioned paper during the initial submission of our work. We have added a citation and a discussion of this paper to our introduction/literature review. 

While both studies utilize U-Net-based deep learning architectures to enhance Lidar retrievals, our approach differs fundamentally in its training objective and the physical capabilities of the resulting model. A. Fuller et al train their model using CATS L2O operational products, which are derived solely from CATS Lidar measurements. Consequently, their model, while improving resolution and detection of tenuous layers, remains constrained by the physical limitations of Lidar signal attenuation as it cannot be trained to classify atmospheric features where the Lidar signal is fully extinguished. In contrast, our model is trained on a single consistent labeling schema aligned with Cloudnet and PollyXT standards, which combine lidar and radar data. Because the radar penetrates optically thick layers that fully attenuate the Lidar signal, our ground truth contains "invisible" information (from the Lidar's perspective). This allows our model to learn contextual correlations and infer cloud and precipitation properties above the Lidar attenuation limit, effectively mimicking a Lidar-Radar synergy using only Lidar inputs. Thus, we have demonstrated that Cloudnet-like classification can be approximated using only a single lidar system. This represents a methodological and operational advancement beyond previous approaches. Revisions were added to the introduction and discussion sections.

We thank the reviewer for their in-depth comments and feedback as it provided great clarity for additional changes that need to be addressed in our work. Our responses are provided in the attached document 'Final comments RC-2'.