General Comments:

The main topic discussed in this manuscript is the uncertainty on the hazard estimation of hydrologically-triggered landslides. It presents a new model at global scale (PHELS), that estimates the daily hazard of hydrologically-triggered landslides at a coarse resolution (36 km) at the same time that it estimates its uncertainity by generating ensemble simulations. The paper is focused on the analysis of the performance of the temporal component (hydrological predictors) of the hazard estimation, as the static part (landslide susceptibility) is based in a paper already published (Felsberg et al., 2022).

The manuscript analyses the potential of three main hydrological predictor variables: the daily rainfall, the 7-day antecedent rainfall index (ARI7) and the root-zone soil moisture content (rzmc), although it does not go into detail on the uncertainty on the obtention of these values. Specially the rzmc is a factor that is very sensitive to the input parametres of the Catchment Land Surface Model (CLSM), as for exemple the soil porosity. Although my expertise is not in data-driven models, I assume that the results could also be affected by this sensitivity, therefore I’d recomment that authors acknowledge that some uncertainty could be induced by the source of the hydrological predictors.

The topic is interesting and novel, since as the authors point out, the literature on quantification of hazard uncertainty is very scarce, and only some attempts to quantify uncertainty of susceptibility or rainfall thresholds uncertainty have been published. In general, I think the authors should further emphasize the main advantages and limitations of PHELS compared to other models, such as the ones that don’t provide uncertainty.

The paper is very well written, clear and, even if in some parts some clarification may help (at least for the non-experts in the topic), it is in general easy to read.

The conclusions are consistent with the evidence and arguments presented. They address the main questions proposed.

The Figures and Tables are in general clear, and helpful to follow the paper.

Specific comments:

L29: binary approach: for the landslide hazard assessment or for the empirical temporal probability?

L34-40: do all these refer to the root zone or some include also lower layers of the soil?

L99: CLSM- what is the source of the inputs of the model (e.g.: soil porosity)? Also, in what units is rzmc?

L125: as the temporally dynamic soil moisture...or also ARI7?

Table 1: I am not sure if ARI7&rzmc were not tested because conceptually they both represent the same parameter (soil moisture)? If this is the case, I am not sure this is correct, as ARI7 may imply infiltration of water to lower levels (and consequent instabilization of the slope due to the water table rise) while rmzc is only for the upper layer of the soil.

L186: As I understand, PHELS is trained with all the LSE for 2007-2020, and tested with the same data? Do you think this could induce some misleading results in the evaluation?

Figure 4: What about the high H values in 2016? All the predictors show high H around 2016-02 but no LSE is recorded. Would they be false alarms or a limitation of the GLC?

Figure 6: This is an interesting Figure!

Figure 8a: Again, going back to the rzmc absolute values. I can see values around 0.5 m3/m3 in SE Asia, that would correspond to soil porosity of 0.5 (considering that the soil is fully saturated). These are very high values for porosity, only typical for some sort of coarse sand or silt, but not common. I have noticed this in SMAP-L4 products derived from CLSM, and in my opinion is a limitation that the use of global models have. I would only make a point here to say that absolute values of rzmc may be overestimated due to this

L332: I think this could be because ARI7 is actually giving a larger picture of the mechanical process going on in the slopes and closely related to the instability process.  So, I agree that soil moisture of the upper layer is not always a good indicator. 

Conclusions: as I said earlier, in the discussion/conclusions I miss some emphases on the advantages and limitations of PHELS over other models, i.e: the applicability of such a model