General comments

This paper combined gridded atmospheric data from HAR v2 with 87 landslides records from GLC and GFLD to evaluate the critical condition that triggered landslides in Kyrgyzstan and Tajikistan. The results show the significant role of snowmelt in landslide triggering processes. The optimal thresholds of landslide for the sum of rainfall and snowmelt were assessed as well. Overall speaking, this paper is well organized and the research results are valuable for understanding the landslide occurrence caused by mixed triggering factors. However, some issues in my opinion need further clarification prior to it can be considered for publication. The first issue is the limitation of landslide inventory, the landslide record extracted from GLC and GFLD mainly based on media report (more concentrated in the populated area) and mixed-up new landslides and recurrent landslides. This limitation will directly affect the representative of obtained thresholds in the entire study area. In addition, the critical thresholds for different landslide type (new landslide and recurrent landslide) should be discussed separately. Thus more solid and spatial representative landslide inventory that can cover all study area obtained from remote sensing identification is necessary. The second issue is how to prove the spatial resolution (10 km) of the gridded atmospheric data from HAR v2 is sufficient to precisely reflect the rainfall or snowmelt condition in each landslide site. The third, the surface runoff, subsurface runoff and ground water caused by rainfall and snowmelt have different time lag characteristic, why these two factors can directly be summed up? Please plot the temporal hydrograph of rainfall and snowmelt respectively, make sure the individual time lag then proceed further calculation. Except the current thresholds evaluated for the mixed-up landslide occurrence, I suggest the authors further discussing the occurrence thresholds for different landslide types (new landslide and recurrent landslide), and the relation with landslide scale (area) in order to strengthen the contribution of this paper.

Specific comments

1. P. 5, Ln 117-118, “….landslide event occurred during or one day after a rainfall event….” Why limit to only one day, please give more explanation or references.
2. P. 6, Ln 144, “…events were randomly split into k folds with k=8…” Why select k=8?
3. P. 7, Ln 161-162, “…Nine landslides events did not occur within any rainfall event, snowmelt event or rainfall + snowmelt event….” This “not detected” uncertainty comes from landslide inventory or observed atmospheric data need more explanation.
4. P. 7, Figure 2, what are the subtitle (DJF, MAM, JJA, SON) on the left top means?
5. P. 9, Figure 4, since daily-based atmospheric data were used in this study, and the exact date of landslide occurrence is available. Please select some landslide cases to plot the event-based temporal process of rainfall and snowmelt.
6. P. 10, Ln 184-185, “…Predictive performance is better when using the entire period than just using the UTL period…” Why? The rainfall properties calculated in the period up to the day of landslide occurrence should have more direct contribution to landslide initiation.
7. P. 16, Figure 7, numbers of historical landslides located in the zone with low to moderate susceptibility. Please give more explanation.

Technical corrections

1. P. 11, Ln 212, “… from median report…..” should be “media”.

The paper is well written and structured. There are a couple of things that need to be further specified. 

I would suggest using different terms instead of "climatic disposition/climatic disposition maps". Perhaps "climatic predisposition to landslides/climatic predisposition maps" better explain the concept.

Then, it could be useful to describe the limitations in the use of gridded products in detecting atmospheric triggering conditions.

About section 2.1.2 Atmospheric data, I recommend including a more detailed description on how the amount of snowmelt has been calculated using the Surface Energy Balance (SEB) and how the thickness of the existing snow is calculated/taken into account in the snowmelt evaluation.

Section 3.2 presents Cumulative distribution function (CDF) curves of mean intensity (Imean), maximum intensity (Imax), and accumulated amount (Q) of non-landslide triggering event (NLTE), landslide-triggering entire event (LTE entire), and landslide-triggering up-to-landslide event (LTE UTL) for rainfall, snowmelt, and rainfall+snowmelt. What is the probability density function used to calculate the CDF? Please describe.

Please check line 177-178, the colours of the poins, probably, have been inverted.        

About figure 5, can you please comment on the fact that for very small values (0,1) of snowmelt for Imax, Imean, Q, there is around 60% probability of having LTE and UTL. It could be useful to exclude from the analyse the rainfall and snowmelt events with Imean = 0, Imax = 0, and Q = 0.

About the indicators used to evaluate the thresholds performance, it could be interesting to calculate the efficiency index (TP+TN/TP+FP+FN+TN; please check literature in this regard). It can be seen as a single indicator for quantifying the correct perditions over the total. It could be used in replacement of HR and FAR and compared with the AUC.

In section 4. Discussion, could be useful to comment also about the lack of information on the landslide types. Also this aspect increases the uncertainty of the results.