Hello to everyone,  

sorry if my review has been duplicated. I logged in from the university computer and did not realise Hakan Tanyas had left his username and password saved. So, if you see the other comment, it is not Hakan but rather me, Luigi Lombardo.

To start the discussion on the right foot, I will begin by saying that I enjoyed reading your document.  I know all of you for your scientific contributions, and when I received the reviewing request from Paolo, I expected a pleasant reading. This is also what I can confirm after going through your document.

As an invited perspective, you have touched upon most of the topics one would want a reference document to refer to and get an idea of what four well-established scientists think of the state-of-the-art in LEWS.   To be fair, I have no substantial complaints to raise. The text is well written.

Not being formally a research paper, there is no science or experimental design to criticize, and in the end, this is your perspective, so even on the content, it is not easy to say you are wrong here or there, or re-run some analyses, or any of the other typical comments/concerns one would raise in a standard scientific contribution.  In short, the manuscript is good as is, to the point of being likely publishable in its current form.  

But, as I was asked to provide feedback on this type of contribution (because on the science, there is little to nothing to object to), what I could comment on would be more on the literature aspects.  In fact, this is essentially a review paper with a personal touch, and on the review aspects, I will make a few points that I feel could have been improved here and there.  

The first comment, in this sense, has to do with the (allow me to say it) autoreferential flavour given to the manuscript in certain parts.  I have counted 22 self-citations to Ben Mirus. I am generally fine with large numbers of self-citations on a topic that almost belongs to the cited scientist. And, I am willing to be lenient in this case because it is your own perspective. This being said, 22 citations is a really large number, especially on a topic that is so general.

I kindly invite the authors to read the text once again and check certain sentences. Outside very specific examples, literally, anyone could have been referenced. I am aware I am being picky, but this is genuinely what I thought as I was recognizing the references one after the other across the text. This a genuine suggestion; there is no need to leave a sour taste on the reader. You have produced a very nice text; just cut the autoreferential side of it.

For instance, the number of self-citations related to Thom Bogaard (7), Roberto Greco (8) or Manfred Stahli (4) is much more justifiable (if not due).

Aside from this aspect, the other comment I could make, and I am very aware of being biased in this case, is related to the sections on "Extrapolating across spatial and temporal scales" and/or "Towards improved landslide forecasting models". Please do not interpret this comment as my request for any personal citation, but I would have expected to read a bit more on your take on large-scale LEWS. In this sense, I am a firm believer that space-time statistics (or ML) will constitute the way to go in the future, being a framework able to integrate the rainfall signal and also geological, topographic and land-use proxies. These are all elements that reflect your idea of incorporating soil hydrology in the prediction, albeit indirectly.

In this sense, taking aside any of the space-time contributions I authored, I would really make a case here for the work of Stefan Steger (Adopting the margin of stability for space–time landslide prediction–A data-driven approach for generating spatial dynamic thresholds  AND/OR Deciphering seasonal effects of triggering and preparatory precipitation for improved shallow landslide prediction using generalized additive mixed models). I feel this topic is something very difficult to leave unaccounted for. What Stefan shows is the power of space-time models, being the only data-driven option where the rainfall signal does not need to be separated from the underlying landscape characteristics. As such, this is also the first modelling architecture where one does not need to have a rainfall threshold but rather a unified probability threshold.

In some sense, and albeit in a less fancy modelling way, this is also what Stanley et al. (2012) try to do (Stanley, T. A., Kirschbaum, D. B., Benz, G., Emberson, R. A., Amatya, P. M., Medwedeff, W., & Clark, M. K. (2021). Data-driven landslide nowcasting at the global scale. Frontiers in Earth Science, 9, 640043.).

Along the same lines, I would have expected the work of Pudasaini to deserve a mention here. For instance, Pudasaini and Krautblatter (2021) propose an unprecedented level of mathematical formalism that certainly can be bridged to your perspectives. In the end, what Pudasaini published constitutes the highest level of physics-based modeling in the literature and could be easily mentioned in the section "Limits of process understanding". Together with it, one could extend the considerations, although not strictly for LEWS, to the implementations this type of modeling requires and why they are not supporting LEWS yet.

Pudasaini, S.P. and Krautblatter, M., 2021. The mechanics of landslide mobility with erosion. Nature communications, 12(1), p.6793.

Overall, this is a good document as is, even without my comments. But if you would want to welcome them, I would definitely suggest cutting some unnecessary references and expanding on others to widen your perspective.

This being said, I am aware this is your perspective and not mine, so I will leave it up to you. In any case, thanks for the interesting reading. Cheers,

LL

This invited commentary highlights the crucial role of hydrological information in landslide early warning systems. It emphasizes that reliance solely on rainfall data is insufficient; instead, hydrological data from slopes susceptible to landslides, especially subsurface hydrological responses, is more critical.

Overall, I agree with and commend the authors' viewpoints. The content and direction proposed in the article can enhance our understanding of the hydrological conditions and mechanisms leading to landslides and potentially improve the performance of landslide early warning systems. The article is worthy of publication. Here are a few suggestions for the authors' consideration to refine the manuscript:

1. Lines 298-232: The authors mention that "the relative change in hillslope wetness conditions often provides the most informative variables, rather than the specific values/precise values of hydrologic variables." While the variation is significant and relates to initial conditions before rainfall or the initial saturation of the slope, it is more crucial whether the hydrological metrics exceed tolerable levels. For instance, the value of pore water pressure, regardless of its change magnitude, if it surpasses the slope's tolerance or historical highs, the likelihood of triggering a landslide is high (e.g., Fig. 1).

1. Format of Fig. 1: This figure presents hydrological responses at two nearby points, but the data is combined into a single plot, making it challenging to distinguish. I suggest dividing it into three subplots: (a) rainfall information, (b) hydrological response at SP1, and (c) hydrological response at SP2. In these subplots, data on pore water pressure and soil moisture/saturation should be represented on different Y-axes. This would help readers better understand the authors' statements.

1. Information in Fig. 1 and Lines 139-145**: Considering the pore water pressure at SP1 and SP2, both increase noticeably during rainfall peaks. However, SP2 appears to have greater and more prolonged saturation. Landslide occurrence does not necessarily require prolonged saturation; extensive saturation in usually unsaturated areas can often drive slope stability to historical lows. Thus, the hydrological responses at SP1 and SP2 can have different applications in landslide early warning systems. As SP2 is located downslope from SP1, SP2’s data can be used to assess long-term slope storage capacity, while SP1’s data can serve as an early or urgent warning. When the pore water pressure at SP1 reaches higher values, indicating significant subsurface saturation potentially extending upslope.

1. Information in Fig. 2: The authors have consolidated three formats of hydrometeorological thresholds for landslide initiation and suggest that the red general threshold in the figure is the expected one. While I understand the differences among these types, I wonder why the red general threshold is considered more universal or broad. The meaning of the red general threshold is actually similar to the intensity-duration threshold approach. Please provide a more detailed explanation of why the general threshold curve differs from the other two. Additionally, the hydrometeorological threshold curve, also referred to as the critical line, varies with climatic regions or administrative practices. Even within the same country, different areas may have other critical lines. Perhaps Section 2.2 could emphasize that the critical line varies by location, aligning with the recommendation in Lines 328-332 to "Determine regional controls on landslide hydroclimatology."

Wei-Li Liang

National Taiwan university