Review of manuscript “Benchmarking tree species classification from proximally-sensed laser scanning data: introducing the FOR-species20k dataset

This research article represents an important step towards filling the gap in automated tree species classification from proximally-sensed laser scanning data. The authors compiled the FOR-species20k benchmark dataset, an open collection of over 20,000 individual tree point clouds, mostly from Europe (but also North America and Oceania). The final dataset resulted in 33 species of trees, capturing significant variations in tree heights and crown architecture. The dataset was then used for benchmarking seven different deep-learning methods, including both methods operating directly on 3D point clouds and on 2D images obtained projecting point clouds. To do so, the authors launched a data science competition. 2D image-based models had a higher overall accuracy than 3D models and the top-scoring method turned out to be DetailView.

Overall, I liked the article and I found it well written for the most part. Particularly, I would like to commend the authors for their commitment to open science, especially for the public release of the FOR-species20k dataset.

Even though the scope of the article does not perfectly align with my expertise, I was able to generally understand the methods and the workflow. I recommend that additional reviewers with deep expertise in laser scanning and DL methods provide further technical insights. Instead, I would like to focus on the form and readability of the article.

I have some remarks about that. In general, I think the introduction could be better written.

Although it is thorough and informative, I believe it could benefit from additional structure and conciseness. For example, from line 50 to line 69 are reported some advantages and limitations of the use of laser scanning for tree species classification, and then, from line 70 to 120 are listed all the key trends and gaps resulting from a literature review. I would first talk about the pros of using laser scanning for tree species classification and then include all the limits of the technique in the bulleted list (which I still appreciated because it helps keeping things in order). Reordering the content to discuss the benefits of laser scanning followed by its limitations could improve the logical flow, making it easier for readers to follow the argumentation, particularly those who are not experts in this field, like me.

The introduction is also a bit lengthy. I would try to make it shorter, perhaps by thickening the information presented in the bulleted list. Also, for example, the information reported from line 125 to line 135 are a bit redundant with what has been already said.

Additionally, in line 137 I would mention the name of the dataset that has been built.

About the materials section, I think that the information reported in lines 147-154 are a bit redundant with what is later said. For example, it is mentioned twice that the data comes mainly from Europe but also from North America and Oceania (more specifically, in lines 148-149 and in lines 156-157). I would put all the information contained in lines 147-154 in the data origin sub-section, obviously removing the repetitions.

Thank you for addressing all my comments and improving the readability of the article. I have no additional suggestions and I recommend the publication of the manuscript in its current form.