Dear authors, 

I will start by saying that I have really enjoyed reading your work. 

I think that from a technical perspective, it is impeccable. You used LASSO and you designed an interesting experiment where you test the susceptibility of various sites and compare it with a model that features all the sites. The validation tools you presents are appropriate, which is uncommon in the landslide susceptibility literature with the exception of some cases. For instance, you used a spatial cross-validation structure, you used cutoff dependent and independent performance metrics, you introduced the concept of the Brier score (at least I did not know it, I do not know if it has been used in other landslide papers). 

So, from that perspective I have no doubt the paper should be published right away. 

I have only some minor comments. In terms of literature review, some of the operations you perform have been introduced in certain papers that have become quite relevant with time and yet I did not see them mentioned. This is particularly true for the spatial cross-validation and for the choice of a balanced dataset. 

Figure-wise, everything is very clear and neat. I have suggested combining figures 6 and 7 in the attached pdf. 

The only thing I struggle a bit with is the absence of a susceptibility map. Now, I do understand the complexity this will require, but it would be nice to take at least one site as a reference and then estimate the susceptibility there from all the models you tested. This will mean transferring the predictive function from other single-site-models and from the all-site-models. Then you could show the difference for each case with respect to a reference model which was generated locally, 

Other that this minor suggestions, it is a very good work there and thanks for the nice reading once more!

Kind regards, 

Luigi Lombardo

The paper represents a very interesting study to statistically evaluate shallow landslide occurrence for 10 different sites across Switzerland, characterized by grassland. Very nice and clear work. The abstract accurately reflects the contents of manuscript. All figures and tables included in the text are necessary and appropriate. The adopted methodology is clear, scientifically solid, and reproducible in other contexts. It should be strongly recommended for publishing.

However, there are some minor points that deserve further explanations:

1) It is not very clear what the authors mean by shallow landslide. In the work of Samarin et al (2020), shallow landslides represent a class of soil erosion (areas with displaced topsoil layers and clear boundaries to the surrounding vegetation). It is possible to insert some pictures of the studied phenomena?

2) The aerial images for the study sites were collected during different years (2013, 2014 and 2015). Which events reference the landslide inventories? Is it possible to include in the text the characteristics of the rainfall events which triggered the shallow landslides?

3) The authors have extracted the centre points of shallow landslide with ArcGIS. Nevertheless, I think that the landslide causal factors are better represented by the landslide source areas.

4) The authors selected a simplified geological data set containing only the three main rock formation classes (igneous, metamorphic, sedimentary). Generally shallow landslides occur in the soil cover. Why the authors have not considered the pedological map (or the soil cover) of the study sites?