Thank you for the opportunity to review this short-form manuscript. The authors design a set of straightforward experiments that include testing the efficacy of using 0- to 15-day antecedent soil moisture information from a modeled global reanalysis data product, in conjunction with rainfall data, to identify the triggering conditions for landslides using machine learning. The Results and Conclusion are intuitive in that antecedent soil moisture improves model performance, with the benefit decreasing somewhat with increased lag. This deprecation in model performance seems minor for a lag that is equivalent to the latency of the modeled soil moisture data product (~5 days). Although I appreciate the streamlined presentation of this study, I think it would be helpful for readers to see more text related to (1) the kind of landslides this study is relevant to, (2) why the spatiotemporal resolution of the modeled soil moisture data product is appropriate for the landslide type(s) considered here, and (3) a deeper interpretation of the Results. Regarding #3 - What are the rainfall depth/duration characteristics and the antecedent soil moisture levels that the best-performing model settles on? And do these characteristics make sense relative to the landslide type(s) and/or any previously published regional thresholds? The objective of this study is crystal clear, but the authors may consider questions like these to expand the relevance of their work for the broader scientific community.

Sincerely,

Matthew A. Thomas

Other Notes:

LN 20: Comma needed in “4862”?

LN 30: May consider highlighting that ANNs have also proven successful for forecasting subsurface hydrologic response for landslide-prone hillslopes. https://doi.org/10.1029/2020GL088731

LN 34-41: Is it worth mentioning that virtually all of these kinds of rainfall and soil moisture products (e.g., NASA GPM and SMAP) have some kind of latency?

LN 57-58: “On the other side,” may be unnecessary text.

LN 133-134: What are the implications for assuming landslide timing as the end of the day?

Thank you for the opportunity to review this  manuscript.

Please see the attached pdf file

This is an interesting study emphasizing the utility of ERA5 soil moisture data for improving landslide forecasting, despite the coarse resolution (9x9km) and considerable latency (5d) of the product. Overall, the results are intuitive and confirm expectations based on previous research, namely that ERA5, which reflects more than just antecedent rainfall, will improve landslide forecasting when paired with rainfall data. This is useful despite these limitations with resolution and latency of ERA5 and I agree with reviewers #1 and #2 that this will ultimately contribute to the literature and be appreciated by readers of NHESS. The explicit evaluation of latency impacts is interesting, though from the perspective of implementation for early warning it’s unclear why a hypothetical latency of 15d is useful since the actual product currently has the fixed delays of 5d, particularly when other more practical questions about the broader utility of ERA5 could be investigated (see below).

As pointed out by both reviewers, the paper is indeed lacking on discussion and broader implications, so overall, I found the analysis somewhat narrow in terms of the scope of hypotheses tested.  As Reviewer #1 noted in his paper for the San Francisco Bay Area, California (Thomas, Collins, and Mirus, WRR, 2019), SMAP data is useful, but in-situ soil moisture sensors have the capacity to improve over the general limitations of satellite soil moisture and rainfall data, even though the latter is theoretically available everywhere on the globe. As we further note in our recent perspective (Mirus, Bogaard, Greco and Stahli, NHESS, 2025), hillslope monitoring stations are advantageous for improving forecasts, but are difficult to maintain and come with other representativeness issues. So, we suggested more rigorous comparison of in-situ sensors from hillslope locations with satellite data for landslide prone areas would shed more light on the utility of satellite data for landslide forecasting. The current study misses this opportunity. Are there any in-situ hydrological data available anywhere in your study area to expand the value and impact of your study? I realize that the Contra Costa and Alameda counties from Matt’s paper are only ~5,000 km2 whereas Sicily is closer to 26,000 km2, so I’m curious if you’d find over this larger scale that the satellite soil moisture, despite the lags, is still more useful than in-situ sensors for spatially explicit landslide forecasting?

The study is fine otherwise ad comparable to a technical note or methodological contribution. Again, I would urge the authors to dig deeper in their analyses, as all three reviewers suggest, to enhance the impact and utility of this work to inform landslide early warning systems in Italy and worldwide.

Ben

The paper is presented in the format of a short communication or technical note aimed at providing insights into the added value associated with the use of volumetric soil water content as provided by ERA5-Land, as an additional proxy for event identification. I agree with the points raised by the other reviewers: I appreciate the clarity and simplicity of the approach. All steps are straightforward and easy to understand; nonetheless, I also share some concerns (and it would therefore be useful to provide further information regarding):

- what is the current early warning system used in Sicily for landslide forecasting, and what is the added value compared to this benchmark? (In this regard, if the system is based solely on triggering rainfall, it would be very interesting to understand the added value of using soil moisture as an event discriminator);

- what are the characteristics of the movements in the area? It is obviously well known that the added value of using volumetric water content as a proxy depends on the characteristics of the event—events in finer-grained soils are more influenced by antecedent rainfall, while with thinner soil layers or higher permeability, the importance of triggering precipitation increases.

With that in mind, is it possible to identify a sort of zonation to understand in which areas there might be added value in using reanalysis-based volumetric soil water content as a proxy? If possible, a spatial representation of the information could be very helpful.

- while I understand the rationale for using information with up to a 15-day delay in the first part of the analysis, from the introduction of the ANN approaches onward, I find it less useful and would limit all representations to 5 days before the present time; it is likely that in upcoming releases (ERA6 is expected in December 2025 with 14 km resolution) the delay will decrease and not increase.

- although it is well known to most, for completeness I would include a brief description of the ERA5-Land reanalysis (it would also be helpful to emphasize that 9 km is the resolution of the land module only, while the atmospheric part is a statistical interpolation from the parent model "ERA5"). For accuracy, I would avoid referring to ERA5-Land as "products" or "dataset" and instead use the term "reanalysis outputs."

Additionally, I note:

- in the abstract, avoid or explain any acronyms used;

- there's a typo in Figure 1.