30-year record of Himalaya mass-wasting reveals landscape perturbations by extreme events

In mountainous environments, quantifying the drivers of mass-wasting is fundamental for understanding landscape evolution and improving hazard management. Here, we quantify the magnitudes of mass-wasting caused by the Asia Summer Monsoon, extreme rainfall, and earthquakes in the Nepal Himalaya. Using a newly compiled 30-year mass-wasting inventory, we establish empirical relationships between monsoon-triggered mass-wasting and monsoon precipitation, before quantifying how other mass-wasting drivers perturb this relationship. We find that perturbations up to 5 times greater than that expected from the monsoon alone are caused by rainfall events with 5-to-30-year return periods and short-term (< 2 year) earthquake-induced landscape preconditioning. In 2015, the landscape preconditioning is strongly controlled by the topographic signature of the Gorkha earthquake, whereby high Peak Ground Accelerations coincident with high excess topography (rock volume above a landscape threshold angle) amplifies landscape damage. Furthermore, earlier earthquakes in 1934, 1988 and 2011 are not found to influence 2015 mass-wasting.

I provide some general comments about what has been undertaken, rather than focussing on detailed line by line comments at this stage.
Methods: • Overall the methods are not well explained (I don't think it would be possible to replicate what has been done based on the detail provided), and much remain ambiguous and undefined. Some of the comments below may arise from this lack of clarity. • The mass wasting time-series is collated by using a pair of images to identify each feature from before and after each monsoon over a 29-year period. By definition, the mapping therefore only intends to capture mass wasting that occurs within the monsoon period only, and not total change from one year to the next. The decision to do this is presumably to isolate the monsoon impact, which is fine, but this has consequences for other analyses that can be conducted with this data. The clarity of this distinction is lost throughout the manuscript which is currently presented as the net contribution of a range of factors that relate both to the monsoon (ASM), but also potentially to the period outside of the monsoon that is not consistently mapped (which potentially includes EQs, road construction and extreme rainfall). As a result, without very careful explanation, the results as presented seem to be quite misleading. There are some important details that arise as a result, which I pick up on below.
• Images have been selected to bracket the monsoon period as tightly as possible, with the date range for pre-and post-monsoon images stated as latest end April and earliest end October respectively. This sampling raises several questions and some implicit assumptions which are not explained in the results that are generated… • On the basis of the description in the manuscript, the mapping period for each feature could cover a shortest period in any year between 30 April and 31 October of 182 days (best case), or a longest period of 550 days (worst case). I note from the supplementary data that the actual mapping period (newest image date minus oldest image date in pair used to identify each feature) has a shortest range of 208 days, and the longest range of 528 days. Critically therefore, some mass wasting features are identified from across a period that is double that of other features. If all mass wasting occurs only in the monsoon this might be OK, but we don't know this and it probably doesn'tspecifically that associated with factors other than the monsoon, so this is a problem. This is an issue that I consider further below. • Moreover, by summarising the average mapping interval by year from the Supp Data, there is considerable systematic variation in the mapping periods between years (minimum average mapping interval = 225 days (2015), maximum = 457 days (2007)), which again is double. The data from some years is therefore implicitly covering a different time period and so may be systematically overor under-representing mass wasting that is unevenly distribution through the year. The explanation of the method does not suggest that the variable mapping period is normalised in the analysis that follows, and even if it was this wouldn't fully account for this inconsistency. • There is also considerable variation in the period over which individual features have been mapped within each individual year of the dataset (e.g. 32-day range between longest and shortest periods in 2006, to a 272-day range in 1995. This is a > 8-fold difference between years, which again raises issues for comparison and for identification of longer-term trends that really need to be considered. • Some important related issues arise here. First, the comparison through time relies on very consistent data, but it remains unclear how this variability in period over which data has been collected is considered, and the manner in which associated temporal uncertainty in the timing of mass wasting is accounted for. Second, there is an implicit assumption here that the majority of the mass wasting under consideration occurs within the monsoon only, and from the above, for some years mass wasting occurring only in ca. 7 months of the year is mapped. Whilst this may well be true for ASM-triggered mass wasting, the same cannot be said for EQs, road construction and potentially extreme rainfall events (see below). How are mass wasting events outside of the monsoon considered and how to they feature in the overall calculations of relative contribution? The manner in which the data appears to have been collated implies that this will, unfortunately, not be possible to resolve….
Completeness of data: • A second significant issue arises from the mapping methodology, and the completeness of the data generated. The numbers of landslides mapped in each year are very low compared to other published datasets for this region and period, particularly for the Gorkha earthquake. Whilst I appreciate the difficulties in comparing datasets from different mapping techniques and imagery, the closest comparable effort appears to be that reported in Regmi et al (2016) which is also based primarily on LandSat data. For a smaller area, they map ca. 4,000 landslides, as compared to only 1,400 landslides in a far bigger area in this paper. • This apparent difference extends to a comparison of the aerial extent of landslides (this paper identifies ca. 17.6 km2 of mass wasting, Roback identifies ca. 87 km2 for 2015 in > 20k events). These differences are surprising given the stated minimum mapped landslide size is small and more comparable to other efforts (300 m2 -although this in itself seems very small, given Landsat pixel sizes….). • Related, I also note the discussion of the power-law fitting, and the lowest size thresholds identified at around 11,000 m2. One implication of this threshold is that below this size, the inventory is incomplete. Therefore, the completeness of the dataset needs to be fully evaluated -as at the moment the manuscript is very unclear on this and there are stark differences to other research. As this completeness is absolutely fundamental to your calculations of relative impacts on mass wasting from different controls, this needs addressing. • As in previous papers, there is a focus here only on new landslides, rather than any form of reactivations. The justification for this is a little vague, but has implications for the results that are generated: "We also removed landslide reactivations from this second "corrected" volume of masswasting. This is because we define earthquake preconditioning, a key process for investigation in this study, as a post-earthquake increase in new ASM-triggered mass-wasting, which would not include reactivations of coseismic or past ASM-triggered deposits." Despite this, it remains true that an EQ might trigger a very small part of a much bigger eventual landslide that fails under rainfall, but this would not be included in this analysis, despite the fact that this event would not have occurred without the earthquake. Without fully evaluating the relative contribution of reactivations versus new landslides, I don't think you can just exclude these, particularly given the ambition here for timedependent susceptibility assessments. From a hazard perspective, this is also problematic -a farmer in Nepal doesn't care if he is hit by a new landslide or a reactivation -it is still happening because of the earthquake and subsequent rainfall… This choice again adds a small but critical subtlety to the simple summary figures that are the headline of this paper, which I feel are problematic without proper explanation.
Non-stationary baseline conditions: • An overarching concern is the very simplified view of the conditions that are potential controls on mass wasting in this part of the Himalayas during the study period. Critically, the baseline conditions are far from stationary. The period considered has experienced considerable social, political and economic change which remains highly challenging to disentangle from impacts on mass wasting. The period covers significant outmigration of working age males -with a well-documented impact on the degradation of agricultural land, the civil conflict -which dramatically impacted farming practise, the changes in the planning process -which influences when and where developments such as road construction takes place, significant urbanisation, plus much much more… The processes upon which you focus (roads, ASM, EQs, extreme rainfall) are playing out upon this backdrop, and I would argue cannot be separated from this without a considerably higher resolution and precision of data. I don't therefore believe that it is possible to 'isolate' the influence of the chosen factors as is suggested here. In the simplest sense this issue is illustrated in the calculation of the long-term average mass wasting rate -which by definition assumes zero change in background conditions by over the 30 years -when these long-term changes may each be a significant and interrelated catalyst in promoting mass wasting.
• Related to this, the word 'correlation' is in my view used incorrectly throughout the manuscriptwithout direct data on the controls rather than just mass wasting on its own, this analysis does not really go beyond identifying 'coincidence'.
Assessment of roads: • The primary impact of roads on mass wasting in the Himalaya is from rural or seasonal roads constructed with minimal engineering input. Rural roads (ca. 2 -3 m in width, unsurfaced and commonly with minimal spectral contrast to the surrounding landscape) are arguably invisible -or at best inconsistently visible -in LandSat imagery (15 m, 30 m). Evidence to the contrary needs to be provided if this is how this has been done… I am therefore unclear on the confidence with which the association between roads and landslides can really be made when the roads must be barely visible in the imagery used here. The method to identify this association is also not well explained so this analysis remains unconvincing. • The only fact presented about roads is the timing of a World Bank funded program -this seems a far too simplistic association to imply in the absence of any data on roads, particularly given the far wider range of donor funded road and rural access programs that have rolled out during this study period, in addition to the wider set of influences on road construction in Nepal (see above, and below). • A key further consideration is that rural road construction in Nepal is widely acknowledged to be a highly seasonal activity, very unevenly distributed through the year. Where the analysis conducted here only samples part of the year (and inconsistently so…), the risk of not capturing the impact of road construction in a uniform manner is therefore high. There are several important related issues for the analysis undertaken: (1) the timing of road construction is closely linked to both the fiscal year and the annual planning cycle in Nepal -with a common rush in road construction towards the end of the former. This research needs to consider how completely road construction impacts are captured if a net impact is to be derived -but I am not clear that the present data can actually do this; (2) The planning process for road construction has changed significantly during the study period considered here -from the 14 step planning process (up until 2017) and the 7 step process thereafter -this significantly influences when roads are built, so there is potential for a shift in bias through the time period considered; (3) related, the budgeting structure and decision making for rural road construction has also changed significantly during this period, with a notable localisation of construction capacity and decision making. A consequence is that when the mapping technique covers the monsoon period only, road construction -and the associated landslides -will potentially not be mapped, which may well be a significant number, and given then the annual date of imagery used in the mapping varies year on year, therefore the period being sampled varies. If road construction was uniformly distributed through the year this might be considered negligible, but it is not, and so this is potentially very important. This variability in road building activity needs to be considered and a demonstration of its role made… Without actual data on road construction, I don't see how this can be achieved, and so this represents a considerable extra effort. • There is also an implicit assumption that road construction has an instantaneous impact on mass wasting and no longer-term legacy effect, which is surprising given the argument for time-dependent landslide susceptibility. So, I don't agree with this statement: "However, as road-associated masswasting is not related to rainfall, a second "corrected" volume of mass-wasting was calculated with this mass-wasting type removed". Road construction can prime a slope to fail if it does not immediately trigger collapse (in the same manner as the EQ preconditioning ideas) -but rainfall remains a key recurrent trigger. In this sense, without more detailed data, it remains very difficult to separate rainfall and road impacts. This really needs thinking through in more detail and supporting with analysis… Consideration of earthquakes: • The selection of potentially landslide triggering earthquakes has a weak justification. The 30 km radius for inclusion is arbitrary, and more fundamentally using epicentre is a very crude approach to identifying which events might be important when we know that the spatial distribution of PGA or PGV is what controls landsliding. It is impossible to associate the pattern of PGA or PGV to the epicentre location alone. If the analysis were to consider for example the extension of the 1g isoseismal into the AOI for all earthquakes during the mapping period, I would not be surprised to see the 1988 and 2011 earthquakes included. This really needs to be considered, and if no impact on mass wasting is observed for these events, then why is this the case? • The paper considers to some extent the ideas around the legacy effects of previous earthquake damage in conditioning subsequent mass wasting, although doesn't really get into much detail on this idea. Research on this topic is in its infancy, but some key issues seem to be overlooked here. The legacy damage effect of earthquakes is not just experienced in the transient response immediately (1 -5 years) after the initial shaking. For example, the Parker et al paper referenced identifies damage legacy effects manifest over intervals of ca. 40 years for Mw7.7 and & 7.1 events in NZ. With this in mind, it is surprising here not to see any consideration of the potential legacy of historic and very much larger earthquakes which have directly impacted the AOI (in addition to 1988AOI (in addition to , 2011AOI (in addition to , also 1934AOI (in addition to , 1833AOI (in addition to , 1505.
Consideration of extreme rainfall: • I'm very surprised given the 30-year period, the extent of the study, and the local climate, that only two extreme rainfall events have been identified as having an influence on mass wasting. This potentially shows the relatively insensitivity of the approach described. I note that the perturbations are identified and then a driver is sought, rather than looking at drivers and trying to identify if a mass wasting response occurred…this seems a slightly circular approach. Given that there is no data on extreme rainfall considered here to assess which events have consequences, and which don't, by only considering two events this analysis does not seem comprehensive enough to really evaluate the net impacts of extreme rainfall events.
Other key comments: • I am uncomfortable about relying on 'using the methods described by Marc et al. 21', -this paper is not a good example of clarity of explanation and I would strongly suggest the methods used here are fully explained to avoid perpetuating ambiguity.
• The LandSat derived background mass wasting rate is compared to the data from Roback to assess the relative impact of the Gorkha earthquake. As far as I can understand from the explanation, this appears problematic because (1) the Roback data is much higher resolution, is collected over an area that is entirely impacted by the earthquake, and is data collected from imagery over < 60 days (these factors all derive a very high event mass wasting rate), but (2) the data that this is compared to presented here is collected over a far bigger area where only a small portion is directly impacted by the Gorkha earthquake, the resolution and completeness is at least an order of magnitude lower, and the time period of mapping is much wider. All of these factors will accentuate the difference between the short and long term mass wasting rates. This just does not appear a valid comparison to make. It is not clear why the data mapped here for 2015 is not used here, or does the relatively limited number of mapped features inhibit this…?
• This point above also raises a question about how to compare rates between the controls identified. For the extreme rainfall, a local average rate is derived and compared to one specific to that event. This seems different for how the comparison is undertaken for Gorkha, where rates for the whole area impacted are considered. As it is known (and shown in Figure 1) that landsliding varies considerably across this area over the long term, how then is the impact of roads handled -is this a deviation in local mass wasting rate for the location of the road only, or a net change in rate for the whole area? When comparing events and controls, how each is treated needs to be consistent, which does not appear to be the case at the moment…Again this has a big influence on the simple summary stats….

Figures:
• Figure: 1 -no admin boundaries make this difficult to orient. DEM data has no source / attribution. DEM scale is odd -seems to be reporting a new highest place on earth! Surely the max elevation is 8,848 m, not >11,000 m? • Figure: 2 -confidence intervals would be useful here, particularly given the variability on the log axis. • Figure: 3 -it would seem sensible to only show EQs of sufficient magnitude to trigger landsliding (e.g. 6Mw…) not all.
Supplementary data: • The mapped shapefiles would be more widely useful than the *.txt file of summary stats. • I would also like to see the full inventory posted to this repository for wider scrutiny and for use in other research: https://www.sciencebase.gov/catalog/item/586d824ce4b0f5ce109fc9a6 Review of Jones et al "Perturbations to Himalaya mass-wasting by rainfall, earthquakes and road building" Oliver Francis This is an interesting and in-depth study into the mass wasting processes of the Himalaya. The authors use an extensive landslide inventory and precipitation data to estimate the contributions of different processes to mass wasting. The authors produce the first empirical relationship between total monsoon rainfall and mass wasting and use this to identify 4 perturbations to the average mass wasting events. Further they identify mass wasting associated with road building and identify it as a growing contributor to erosion rates. I believe this are novel and important findings which further our understanding of the erosion of the Himalaya.
I have several comments concerning the methodology and findings of the paper and several smaller suggestions for improving the clarity of the manuscript. I will go through the major comments first and provide a list of the smaller comments with line numbers afterwards.

Major comments
The work presented here in this manuscript quantifies the volumes of mass wasting produced by different processes across a 30 year time period and argues these are meaningful for understanding long term erosion rates and forecasting changes due to climate change or road construction. However, without an understanding of the frequency of these processes it is impossible to convert these volumes into rates and put them into context of time scales longer than the 30 years recorded here. I recommend identifying the return period of the extreme rainfall events in 1993 and 2002 and comparing those with the frequency of Gorkha Earthquake in order to fully understand their relative contributions to erosion rates.
Along a similar line I also found the quantification of the contribution of road construction to mass wasting confusing. It is unclear in the manuscript how these mass wasting events are identified, are they simply the volume of material produced during the construction? Or do they include the enhanced mass wasting associated with undercutting of slopes and poor slope reinforcement? Separating these two processes is important for understanding the continued contribution of road construction to mass wasting after the study period. It is also unclear how the recorded areas are converted to volumes; these events are not landslides so using a landslide area -volume scaling is not appropriate. Volumes of material produced by the construction of the roads could be estimated by measuring the width of the roads and the heights of their cuttings into the hillslope. Accurate estimations of the mass wasting produced by construction and post construction along with road mapping would allow forecasting of mass wasting of future construction projects.
Finally, while acknowledging the limits of the format of the journal, I feel that several extra figures (listed below) are required in order to help the reader interpret the findings and the methodology of the paper. Minor comments Line 54: I find the definition of the "background" rate confusing. Is the background rate simply the estimated rates of mass wasting from the monsoons? Or does it also include storm induced mass wasting outside of the monsoon months? Clarifying this definition along with the timing of the monsoon will help to better explain the aim of the paper. Line 64: Cosmogonic should be cosmogenic Line 79: The definition of mass wasting associated with road building is confusing. Is the sediment produced by cutting and constructing the road recorded or is it the enhancement due to undercutting of slopes? The former is a single event while the later is a continuing process. Figures of examples of road associated mass wasting would help with these definitions Line 112: ~ should be > Line 112: The magnitude-frequencies of the mapped mass wasting events should be included as a figure to help the reader understand their statistics better, particularly as their shape is used in the methodology Line 134: How is the strength of a monsoon defined? Strength could be total precipitation or intensity or another measure of precipitation. I would avoid the word strength and be more specific Line 138: I found the current wording of the methodology confusing. As it is currently worded it sounds like a single mean is produced of the monsoon landsliding rather than a yearly estimate. I suggest rephrasing it to: "each of the 29 ASM seasons." Line 153: A figure of the yearly frequency -magnitudes would help the reader interpret and understand these findings themselves. Line 161: If a monsoon can produce mass wasting significantly greater than the estimated value from the normalisation it raises the question of how well the normalisation works. If this storm is as anomalous as is implied by the normalisation the manuscript should justify why it is not included in the normalisation. This would include discussion on how to identify extreme events, particularly as they seem to be exponential rather than normally distributed in nature. Line 167: The use of landslide density is confusing and hides the fact that the 2002 storm was more erosive than the 1997 storm. I suggest presenting all findings in terms of volume rather than area. Line 169: To combine these two events I feel the return time of the rainstorms needs to be understood. Without knowing the return period of the storms, it is unclear whether these events are frequent, and therefore important to the long-term erosion rate or insignificant. Combining them together could make them seem more significant than they are. Line 175: Preconditioning is set up to be an important part of the manuscript however very little time is spent on describing it in this study. A figure showing the described correlation between earthquake magnitude and preconditioning (particularly as no mention of preconditioning prior to 2015 despite multiple >Mw6 earthquakes) would help to give more attention to this process. Line 177: The timing of the earthquake compared to the monsoon is not clear. It should be noted somewhere that the earthquake occurred prior to the monsoon. It is also unclear whether the coseismic landslides were mapped or the Roback 2018 inventory was used. Line 178: It is not clear why preconditioning increases mass wasting by a factor of 4 but is only the same as 2 ASMs. This should be rewritten for clarity Line 187: Reference to figures Line 193: Again, reference to figures Line 196: If this were true surely the uncorrected trends would not decrease through time? They maintain a similar trend to the corrected i.e. decreasing to the background rate which implies that preconditioning is important here too Line 197: The areas shown in figure 4 cannot be compared to volume therefore it is hard to compare with the other perturbations discussed Line 201: It would be useful to know something about the amount of road construction occurring. What is the correlation between rate of road building and mass wasting? This would help clarify the definition of road associated mass wasting and forecasting into the future. Figures of images showing examples of mass wasting associated with road construction will also help to clarify how they are related and defined. Line 202: As area is shown in figure 4 it is now not clear whether volume or area is discussed here. I again recommend only using volume when discussing or show results. Line 207: A figure showing the absolute values of mass wasting would be useful to show how the contribution by each process changes through time. Line 236: You have not identified the controls on these events, are these years of intense monsoon that are poorly predicted by the normalisation method or are they freak storms of abnormal intensity? It is also possible similar storms occurred without causing significant mass wasting, without any analysis on rainfall patterns in this period it is difficult to interpret these events. Without clear definitions of extreme events and their return periods it is difficult to determine whether these events are likely to change through time. Line 261: The timing of the monsoon should be mentioned to help the reader understand why these months were chosen for imaging Line 300: This section should be above the section discussing precipitation as it links directly to the discussion of mass wasting. Line 343: The averaging of the monsoon should be using volume rather than area. Area is not a useful measure of sediment flux due to the relationship between landslide area and volume. Line 370: Here it would be useful to have a figure of the magnitude -frequencies of the years described. The reader can then made informed decisions on the applicability of this methodology. Figure 2: It would be good to have the events discussed in the manuscript highlighted in this figure. Some form of time series of precipitation would also help to understand the extreme rainfall event described. Figure 4: The "Reactivations + Road" line is hard to see. Area is difficult to compare with volume as discussed in the rest of the manuscript. Area is not important for erosion rates and should not be compared. Absolute volume values through time would be much more useful.

Reviewer #2 (Remarks to the Author):
Anonymous Comment(s) 1: My comments below are intended to strengthen this work in future and bring attention to some key issues which I my view should be addressed prior to considering this research for publication, and I note that addressing these to support the same conclusions represents a significant effort. It may well be that some of the suggestions are not feasible, but if this is the case, the aim and conclusions of the paper should be modified to reflect this.
• Overall the methods are not well explained (I don't think it would be possible to replicate what has been done based on the detail provided), and much remain ambiguous and undefined. Some of the comments below may arise from this lack of clarity.

Response(s) 1:
As implied by the reviewer, many of the key issues raised by this review are in part due to misunderstanding of our methods and data completeness. As such, we believe that many of the issues raised, which could have required significant extra effort or not been feasible in the timeframe, are in fact not such big issues at all. Consequently, we believe that the vast majority of our initial aims and conclusions can be supported. The exception to this is our analysis and results pertaining to roads. As we outline in several places below, we have come to the conclusion that, despite our road-tipping data being temporally continuous, we do not have sufficient supporting data to analyse and assess roadtipping impacts in a comparable manner to the ASM, extreme rainfall and earthquakes. As such, we have decided to remove the road-tipping elements of this manuscript, which we believe makes the narrative of the publication more concise and impactful, and allows us to focus on what are, in our opinion, some very novel and impactful results on earthquake preconditioning.

Comment 2:
The mass wasting time-series is collated by using a pair of images to identify each feature from before and after each monsoon over a 29-year period. By definition, the mapping therefore only intends to capture mass wasting that occurs within the monsoon period only, and not total change from one year to the next. The decision to do this is presumably to isolate the monsoon impact, which is fine, but this has consequences for other analyses that can be conducted with this data. The clarity of this distinction is lost throughout the manuscript which is currently presented as the net contribution of a range of factors that relate both to the monsoon (ASM), but also potentially to the period outside of the monsoon that is not consistently mapped (which potentially includes EQs, road construction and extreme rainfall). As a result, without very careful explanation, the results as presented seem to be quite misleading. There are some important details that arise as a result, which I pick up on below.

Response 2:
This is not correct, and is a misunderstanding arising from a lack of clarity on our part. Our mapping is continuous, with the post imagery used to map one year almost always used as the pre-imagery for the next year. As is now described in lines 99 -103: As such, our mapping does not only capture the monsoon, instead capturing all mass-wasting across the mapped period, with each time slice approximately cantered on the monsoon season.

Comment(s) 3:
• Images have been selected to bracket the monsoon period as tightly as possible, with the date range for pre-and post-monsoon images stated as latest end April and earliest end October respectively. This sampling raises several questions and some implicit assumptions which are not explained in the results that are generated… • On the basis of the description in the manuscript, the mapping period for each feature could cover a shortest period in any year between 30 April and 31 October of 182 days (best case), or a longest period of 550 days (worst case). I note from the supplementary data that the actual mapping period (newest image date minus oldest image date in pair used to identify each feature) has a shortest range of 208 days, and the longest range of 528 days. Critically therefore, some mass wasting features are identified from across a period that is double that of other features. If all mass wasting occurs only in the monsoon this might be OK, but we don't know this and it probably doesn't -specifically that associated with factors other than the monsoon, so this is a problem. This is an issue that I consider further below.
• Moreover, by summarising the average mapping interval by year from the Supp Data, there is considerable systematic variation in the mapping periods between years (minimum average mapping interval = 225 days (2015), maximum = 457 days (2007)), which again is double. The data from some years is therefore implicitly covering a different time period and so may be systematically over-or under-representing mass wasting that is unevenly distribution through the year. The explanation of the method does not suggest that the variable mapping period is normalised in the analysis that follows, and even if it was this wouldn't fully account for this inconsistency.
• There is also considerable variation in the period over which individual features have been mapped within each individual year of the dataset (e.g. 32-day range between longest and shortest periods in 2006, to a 272-day range in 1995. This is a > 8-fold difference between years, which again raises issues for comparison and for identification of longer-term trends that really need to be considered. • Some important related issues arise here. First, the comparison through time relies on very consistent data, but it remains unclear how this variability in period over which data has been collected is considered, and the manner in which associated temporal uncertainty in the timing of mass wasting is accounted for. Second, there is an implicit assumption here that the majority of the mass wasting under consideration occurs within the monsoon only, and from the above, for some years mass wasting occurring only in ca. 7 months of the year is mapped. Whilst this may well be true for ASM-triggered mass wasting, the same cannot be said for EQs, road construction and potentially extreme rainfall events (see below). How are mass wasting events outside of the monsoon considered and how to they feature in the overall calculations of relative contribution? The manner in which the data appears to have been collated implies that this will, unfortunately, not be possible to resolve….

Response(s) 3:
We have grouped these comments together as they concern the same issue around the mapping of our mass-wasting database. As outlined above, the mapping was continuous and our revised methodology sections now make this clear.
However, we do agree that the impacts of time-slice length could have been considered. Our analysis pertaining to the ASM and extreme rainfall assumes that the majority of rainfall-triggered mass-wasting will occur between May and Sept, when Nepal is known to experience the vast majority of its rainfall. Therefore, as the reviewer themselves outlines, our varying time slice length will not be a problem if it's true that most mass-wasting occurs during the monsoon-period, which is always regularly included within each time slice. Indeed, the mapping period variation will only be a problem if the months October -April see significant mass-wasting. Our assumption that most mass-wasting occurs between May -September is backed up by the literature (see Petley et al., 2007 andStanley et al., 2020) and by sensitivity analysis we have now conducted, which shows no correlation between time-slice length (i.e. number of October -April months included in a time slice) and mass-wasting occurrence (see supp materials; fig. 2).
To further confirm that the number of Oct -April months included will not impact our results, we went back and re-mapped several extra time-slices between October and April to see how many rainfalltriggered events actually occurred in these periods. These extra time slices were: November 1988 to March 1989 (6 events occurred), September 1998 to January 1999 (4 events occurred), November 2005 to April 2006 (9 events occurred), October 1996 to December 1996 (0 events occurred) and October 1997 to November 1997 (0 events occurred). In all of these cases, the number of events to occur was below 10% (in all but the Nov05 -Apr06 case below 5%) of the total mass-wasting of the main time slices these periods were included in. I.e. the mass-wasting in the Oct -April months is well within the 20% assumed mapping error we applied to all of our calculations. As such, in the case of rainfalltriggered events (i.e. the ASM and extreme rainfall) our sensitivity analysis suggests the variable number of Oct -April months included in each time slice will not impact the results, and any variability due to this should be within the 20% error applied to our calculations.
As mentioned above we have removed roads from this analysis.
Discussion / description on the above is now included in the methods section of the manuscript for clarity, lines 487 -507: However, as our time slices had varying lengths, both between time slices and within time slices (as several tiles were required to map the entire study region, and invariably these tiles had different acquisition dates and cloud cover), it is necessary to consider the effect of this on the subsequent analysis. Our analysis of ASM-triggered and extreme rainfall triggered mass-wasting assumes that all mass-wasting was triggered during a given time slice's monsoon season. As these time slices include months outside of the monsoon period, it is possible that some of these rainfall-triggered events did not occur during the monsoon. However, it is known that this region experiences very little rainfalltriggered landsliding outside of the monsoon period 6,42 . Indeed, we find that there is no correlation between number of non-monsoon months within a time slice and number of mass-wasting events mapped (Supplementary materials, Fig. 2). This suggests that, as expected, the majority of rainfall triggered events can be assumed to have occurred during the monsoon season, and thus that variable time slice length will not unduly impact the results. Furthermore, to ensure that this variability in time slice length does not impact the results, we applied a 20% assumed error to all mapped mass-wasting areas. This assumed error should account for any variability in mapped mass-wasting caused by including non-monsoon months, as well as for any inadvertently included mass-wasting events that are attributable to non-rainfall dominated processes such as undercutting by river channels or roads."

Comment 4:
Completeness of data: • A second significant issue arises from the mapping methodology, and the completeness of the data generated. The numbers of landslides mapped in each year are very low compared to other published datasets for this region and period, particularly for the Gorkha earthquake. Whilst I appreciate the difficulties in comparing datasets from different mapping techniques and imagery, the closest comparable effort appears to be that reported in Regmi et al (2016) which is also based primarily on LandSat data. For a smaller area, they map ca. 4,000 landslides, as compared to only 1,400 landslides in a far bigger area in this paper.

Response 4:
We did not map any coseismic mass-wasting, only rainfall-triggered (and previously road-tips, though as mentioned we have now decided not to include these). This is now made clear in the manuscript, lines 103 -105: "The inventory does not include any coseismic mass-wasting, as inventories of coseismic masswasting triggered by the 2015 Gorkha earthquake already exist 28,32,35 ." As such, the comparison of our data to Regmi et al., (2016) is meaningless, as their inventory was of coseismic failures (and used GF-imagery with a resolution of 0.8 m as well as Landsat, meaning they could map at a far higher resolution anyway).
Where coseismic mass-wasting is considered (i.e. from 2015), we exclusively use the Roback inventory.

Comment 5:
• This apparent difference extends to a comparison of the aerial extent of landslides (this paper identifies ca. 17.6 km2 of mass wasting, Roback identifies ca. 87 km2 for 2015 in > 20k events). These differences are surprising given the stated minimum mapped landslide size is small and more comparable to other efforts (300 m2 -although this in itself seems very small, given Landsat pixel sizes….).

Response 5:
As mentioned above, we did not map any coseismic mass-wasting. So the comparison made here is not valid as it is between rainfall-triggered landslides during the 2015 monsoon season and Gorkha coseismic landslides (with the latter coseismic mapped to a much higher resolution). 300 m 2 is not correct, a mistake on our part. It is technically 900 m 2 (i.e. 30 * 30 m), but we round this up to 1000 m 2 in the manuscript. (see line 97).

Comment 6:
• Related, I also note the discussion of the power-law fitting, and the lowest size thresholds identified at around 11,000 m2. One implication of this threshold is that below this size, the inventory is incomplete. Therefore, the completeness of the dataset needs to be fully evaluated -as at the moment the manuscript is very unclear on this and there are stark differences to other research. As this completeness is absolutely fundamental to your calculations of relative impacts on mass wasting from different controls, this needs addressing.

Response 6:
We agree that discussion on inventory completeness is important, and should be given more attention.
As such, on lines 590 -597 we compare the power-law / area-frequency results of our mapping to others: "Note that our x-min values are an order of magnitude larger than the x-min values obtained for the Gorkha coseismic landslide data base of Roback et al. 28 and the monsoon-triggered landslide data of Marc et al. 4 . This is likely a result of the difference in mapping resolution between this study and the others, with the minimum possible size feature that could be mapped by Roback et al. 28 and Marc et al. 4 an order of magnitude smaller than could be mapped here. However, our cut-off x-min values are comparable to similar studies using imagery with 30 -15 m resolution imagery 77 , suggesting that our inventory is as substantially complete as would be expected given the resolution of the satellite imagery." As we could only map events > 1,000 m 2 , when comparing relative impacts of ASM and 2015 coseismic, we now only compare across a comparable size range (i.e. we remove all Roback events below 1,000 m 2 so we are comparing like for like in terms of event sizes). This information is now included in lines 652 -661: "The comparisons between the ASM-triggered, extreme rainfall-triggered and earthquake preconditioning induced controls are straightforward, as all are based on the same resolution data mapped across the same study region. However, the comprehensive inventory of Gorkha coseismic failures presented by Roback et al., 2018 28 was mapped at a higher resolution and not fully coincident with our original mapping extent, though only 171 (0.69%) of the 24,915 events mapped by Roback et al. fall outside of our study region. As such, to comparably estimate the total Gorkha mass-wasting relative to our ASM totals, we remove all landslides from the Roback inventory which fall outside of our mapped region, or have mapped areas outside of the area-frequency range of our inventory (i.e., any coseismic failures with areas < 1,000 m 2 )."

Comment 7:
• As in previous papers, there is a focus here only on new landslides, rather than any form of reactivations. The justification for this is a little vague, but has implications for the results that are generated: "We also removed landslide reactivations from this second "corrected" volume of masswasting. This is because we define earthquake preconditioning, a key process for investigation in this study, as a post-earthquake increase in new ASM-triggered mass-wasting, which would not include reactivations of coseismic or past ASM-triggered deposits." Despite this, it remains true that an EQ might trigger a very small part of a much bigger eventual landslide that fails under rainfall, but this would not be included in this analysis, despite the fact that this event would not have occurred without the earthquake. Without fully evaluating the relative contribution of reactivations versus new landslides, I don't think you can just exclude these, particularly given the ambition here for time-dependent susceptibility assessments. From a hazard perspective, this is also problematic -a farmer in Nepal doesn't care if he is hit by a new landslide or a reactivation -it is still happening because of the earthquake and subsequent rainfall… This choice again adds a small but critical subtlety to the simple summary figures that are the headline of this paper, which I feel are problematic without proper explanation.

Response 7:
We agree that the inclusion (or not) of reactivations is debateable, particularly when it comes to discussions about preconditioning. Indeed, the reason for having the corrected and uncorrected rates was to see the changing impacts of reactivations through time. By excluding reactivations from the uncorrected rates, we did not intend to exclude them fully from the discussion, only to highlight that much of the extra rainfall-triggered mass-wasting post-2015 is due to reactivations rather than new failures, which is an interesting observation in its own right. To make all of this clearer, and to ensure the subtlety of the discussion is preserved, we are now much more careful of our definition of preconditioning, offering two potential definitions on lines 303 -312: We hope that adding this nuance goes some way to clarifying why we separated reactivations and new failures, and hopefully provides much more useful insight into the potential processes and timescales of earthquake preconditioning.

Comment 8:
Non-stationary baseline conditions: • An overarching concern is the very simplified view of the conditions that are potential controls on mass wasting in this part of the Himalayas during the study period. Critically, the baseline conditions are far from stationary. The period considered has experienced considerable social, political and economic change which remains highly challenging to disentangle from impacts on mass wasting. The period covers significant outmigration of working age males -with a well-documented impact on the degradation of agricultural land, the civil conflict -which dramatically impacted farming practise, the changes in the planning process -which influences when and where developments such as road construction takes place, significant urbanisation, plus much much more… The processes upon which you focus (roads, ASM, EQs, extreme rainfall) are playing out upon this backdrop, and I would argue cannot be separated from this without a considerably higher resolution and precision of data. I don't therefore believe that it is possible to 'isolate' the influence of the chosen factors as is suggested here. In the simplest sense this issue is illustrated in the calculation of the long-term average mass wasting rate -which by definition assumes zero change in background conditions by over the 30 years -when these long-term changes may each be a significant and interrelated catalyst in promoting mass wasting.

Response 8:
We fully agree that the socio-economic processes described here will have some impact on masswasting (another reason for including the 20% mapping error/uncertainty), and would be a fascinating area of future research. However, accounting for this is currently beyond the scope of this paper, so we have to assume that at a given time, all natural mass-wasting drivers will be impacted similarly by socioeconomics. We appreciate that this is a big assumption, but as we say, disentangling the socioeconomics from the natural processes is far beyond what can be achieved here. Furthermore, it is the road-tipping that is most likely to have been impacted by changing socio-economics, which is another reason why we have decided to no longer consider roads within this analysis.

Comment 9:
• Related to this, the word 'correlation' is in my view used incorrectly throughout the manuscriptwithout direct data on the controls rather than just mass wasting on its own, this analysis does not really go beyond identifying 'coincidence'.

Response 9:
In retrospect, we agree that in some places our use of correlation was inappropriate, and have changed terminology to coincident where necessary (e.g. lines 173-175 now read: "Instead, both perturbations were coincident with "cloud-outburst" extreme rainfall events that are reported in the literature").
We do however continue to use correlation for figure 2, the fit of which we do consider to be correlated.

Comment(s) 10:
Assessment of roads: • The primary impact of roads on mass wasting in the Himalaya is from rural or seasonal roads constructed with minimal engineering input. Rural roads (ca. 2 -3 m in width, unsurfaced and commonly with minimal spectral contrast to the surrounding landscape) are arguably invisible -or at best inconsistently visible -in Landsat imagery (15 m, 30 m). Evidence to the contrary needs to be provided if this is how this has been done… I am therefore unclear on the confidence with which the association between roads and landslides can really be made when the roads must be barely visible in the imagery used here. The method to identify this association is also not well explained so this analysis remains unconvincing.
• The only fact presented about roads is the timing of a World Bank funded program -this seems a far too simplistic association to imply in the absence of any data on roads, particularly given the far wider range of donor funded road and rural access programs that have rolled out during this study period, in addition to the wider set of influences on road construction in Nepal (see above, and below).
• A key further consideration is that rural road construction in Nepal is widely acknowledged to be a highly seasonal activity, very unevenly distributed through the year. Where the analysis conducted here only samples part of the year (and inconsistently so…), the risk of not capturing the impact of road construction in a uniform manner is therefore high. There are several important related issues for the analysis undertaken: (1) the timing of road construction is closely linked to both the fiscal year and the annual planning cycle in Nepal -with a common rush in road construction towards the end of the former. This research needs to consider how completely road construction impacts are captured if a net impact is to be derived -but I am not clear that the present data can actually do this; (2) The planning process for road construction has changed significantly during the study period considered here -from the 14 step planning process (up until 2017) and the 7 step process thereafter -this significantly influences when roads are built, so there is potential for a shift in bias through the time period considered; (3) related, the budgeting structure and decision making for rural road construction has also changed significantly during this period, with a notable localisation of construction capacity and decision making. A consequence is that when the mapping technique covers the monsoon period only, road construction -and the associated landslides -will potentially not be mapped, which may well be a significant number, and given then the annual date of imagery used in the mapping varies year on year, therefore the period being sampled varies. If road construction was uniformly distributed through the year this might be considered negligible, but it is not, and so this is potentially very important. This variability in road building activity needs to be considered and a demonstration of its role made… Without actual data on road construction, I don't see how this can be achieved, and so this represents a considerable extra effort.
• There is also an implicit assumption that road construction has an instantaneous impact on mass wasting and no longer-term legacy effect, which is surprising given the argument for time-dependent landslide susceptibility. So, I don't agree with this statement: "However, as road-associated masswasting is not related to rainfall, a second "corrected" volume of mass-wasting was calculated with this mass-wasting type removed". Road construction can prime a slope to fail if it does not immediately trigger collapse (in the same manner as the EQ preconditioning ideas) -but rainfall remains a key recurrent trigger. In this sense, without more detailed data, it remains very difficult to separate rainfall and road impacts. This really needs thinking through in more detail and supporting with analysis…

Response(s) 10:
Despite the fact that our mapping was continuous (i.e. we will have captured all road-tips above the minimum mappable size threshold), in retrospect we agree with many of the reviewer comments here, and conclude that at the current time we do not have sufficient supporting data to analyse and assess roads in a manner comparable to the other mass-wasting drivers considered. As such, we have made the decision to no longer consider the explicit impacts of roads, and instead streamline the manuscript to focus on rainfall and earthquakes, with particular attention switched to earthquake preconditioning.
The potential impacts of roads on our other mass-wasting drivers is now accounted for in the applied mapping error/uncertainty, which is applied to account for any mapped mass-wasting that is actually primarily triggered by road/channel undercutting rather than rainfall alone.

Comment(s) 11:
Consideration of earthquakes: • The selection of potentially landslide triggering earthquakes has a weak justification. The 30 km radius for inclusion is arbitrary, and more fundamentally using epicentre is a very crude approach to identifying which events might be important when we know that the spatial distribution of PGA or PGV is what controls landsliding. It is impossible to associate the pattern of PGA or PGV to the epicentre location alone. If the analysis were to consider for example the extension of the 1g isoseismals into the AOI for all earthquakes during the mapping period, I would not be surprised to see the 1988 and 2011 earthquakes included. This really needs to be considered, and if no impact on mass wasting is observed for these events, then why is this the case?
• The paper considers to some extent the ideas around the legacy effects of previous earthquake damage in conditioning subsequent mass wasting, although doesn't really get into much detail on this idea.
Research on this topic is in its infancy, but some key issues seem to be overlooked here. The legacy damage effect of earthquakes is not just experienced in the transient response immediately (1 -5 years) after the initial shaking. For example, the Parker et al paper referenced identifies damage legacy effects manifest over intervals of ca. 40 years for Mw7.7 and & 7.1 events in NZ. With this in mind, it is surprising here not to see any consideration of the potential legacy of historic and very much larger earthquakes which have directly impacted the AOI (in addition to 1988, 2011, also 1934, 1833, 1505 and 1255…).

Response(s) 11:
Our revised manuscript includes a substantially expanded section on earthquake preconditioning. As suggested by the reviewer, we now dedicate several paragraphs to the analysis of the potential longer term impacts of older earthquakes in 1934, 1988 and 2011 that were coincident with our study region. This is included in lines 314 -352: "This analysis provides insight into short-term Himalaya preconditioning of the type observed by Marc et al.,20 . However, as already described, preconditioning has also been observed over decadal  (Fig. 6). To further investigate these differences, we again divided our study region into the same grids used to analyse the extreme rainfall (Fig. 5) Overall, we believe that these new results are very novel, particularly for the Himalayas, and provide significant further insight into the processes and timescales of earthquake preconditioning.

Comment(s) 12:
Consideration of extreme rainfall: • I'm very surprised given the 30-year period, the extent of the study, and the local climate, that only two extreme rainfall events have been identified as having an influence on mass wasting. This potentially shows the relatively insensitivity of the approach described. I note that the perturbations are identified and then a driver is sought, rather than looking at drivers and trying to identify if a mass wasting response occurred…this seems a slightly circular approach. Given that there is no data on extreme rainfall considered here to assess which events have consequences, and which don't, by only considering two events this analysis does not seem comprehensive enough to really evaluate the net impacts of extreme rainfall events.

Response(s) 12:
We thank the reviewer for these points. As such, we include significant new analysis and discussion where we define and identify all extreme storm events, and then look at mass-wasting response, as suggested. We do this using pre-existing rainfall indices (derived from the TRMM4B42 product Consequently, on the assumption that any mass-wasting above these estimates in the years known to have extreme rainfall events are due to those events, the total extreme rainfall generated mass-wasting can be estimated from the sum of the differences between the actual and predicted total volumes for those years (1993, 2000 -2004, 2006 Overall, this approach has allowed us to much more comprehensively assess the net impacts of extreme rainfall events, at least for 17 years of our time series, which is still a far longer record than any other studies we have seen.

Comment 13:
• I am uncomfortable about relying on 'using the methods described by Marc et al. 21', -this paper is not a good example of clarity of explanation and I would strongly suggest the methods used here are fully explained to avoid perpetuating ambiguity.

Response 13:
We only used the "methods described by" terminology in the manuscript itself, and do include the full outline of Marcs methods in the methods section on lines 600 -621: "The empirical relationships between ASM precipitation and mass-wasting can be used to predict how much background mass-wasting would have been expected to occur each year based on that year's rainfall. The equations derived for the different measures of mass-wasting volume (Fig. 2) are, where V is volume in m 3 /km 2 and x is grid averaged precipitation in mm: • V = 36.855e 0.00288x for uncorrected total volumes.
• V = 28.083e 0.00298x for corrected total volumes propagation of error to obtain the uncertainties for each measure as displayed in Fig. 3. Furthermore, as in the analysis of Marc et al.,4 , the 2014 Jure landslide was removed from the analysis, as this event is widely considered to have been caused by progressive failure across multiple years, and so is known not to be attributable to any single ASM-season 4 ." Comment 14: • The choice of rainfall product really needs to be justified and evaluated (e.g. see Ullah et al., 2018, Remote Sensing). For example, as the rainfall data has ca. 28 km grid size, how does this compare to storm size and how does it influence total monsoon rainfall calculations in an AOI with a significant precipitation gradient, and how does the averaging in the derivation of this product, and in your subsequent treatment of the data behave over mountain topography?

Response 14:
A full treatment of precipitation gradients over mountainous topography is beyond the scope of the

Response 15:
We did look at the relationship between our rainfall triggered mass-wasting and SASMI, but found no relationship between the two. We suspect this is because SASMI is a very regional index, which simply doesn't capture the localised changes in rainfall observed across our much smaller overall study region.
As mentioned, Petley found an inverse relationship, and whilst he offers some explanation for this, it still highlights the fact that very large extent indices such as this are actually not immediately appropriate for use in smaller sub-regions. However, we do appreciate that our use of the term "strength" could induce confusion. As such, we explicitly define what we mean by ASM-strength, and make it clear that in this case the term is used to mean total May -September rainfall in our study region. This is now included in lines 122 -127: "We use total May -September precipitation as a proxy for local monsoon strength, as typical measures of monsoon strength such as the SASMI 39 are derived over extensive regional scales, and do not accurately capture local changes in monsoon conditions across the study region. As such, from this point forward, the term "ASM strength" is defined here specifically as total May -September precipitation across the mapped region." Comment 16: • The LandSat derived background mass wasting rate is compared to the data from Roback to assess the relative impact of the Gorkha earthquake. As far as I can understand from the explanation, this appears problematic because (1) the Roback data is much higher resolution, is collected over an area that is entirely impacted by the earthquake, and is data collected from imagery over < 60 days (these factors all derive a very high event mass wasting rate), but (2) the data that this is compared to presented here is collected over a far bigger area where only a small portion is directly impacted by the Gorkha earthquake, the resolution and completeness is at least an order of magnitude lower, and the time period of mapping is much wider. All of these factors will accentuate the difference between the short and long term mass wasting rates. This just does not appear a valid comparison to make. It is not clear why the data mapped here for 2015 is not used here, or does the relatively limited number of mapped features inhibit this…?

Response 16:
The aim of the relative comparison between the overall 2015 coseismic mass-wasting the mass-wasting typically triggered in an average (unperturbed by extreme storms) ASM seasons. So the fact that Gorkha was mapped over 60 days is irrelevant. The coseismic events in the Roback inventory are predominantly assigned to either the April main shock or the May aftershock, i.e. represent relatively instantaneous mass-wasting that was then mapped over a 60 day period. We simply want to estimate the total coseismic mass-wasting triggered by Gorkha relative to what is triggered across an average 5 month ASM-seasons, so the issues of mapping timescale difference is very much built into the comparison / not relevant.
However, we accept the issues about differences in mapping resolution, and the fact that the Gorkha coseismic inventory includes events at a scale we could not map. As such, as already discussed above, in the revision we only make this comparison for coseismic events above the mappable size range. This should make the comparison more applicable.
Furthermore, we accept that the region impacted by Gorkha is slightly smaller than our overall study region, though still impacted ~50% of it. However, the aim of this comparison is not to just look at the region impacted by an earthquake and see how this compares to the monsoon, but rather to look at a given large region of the Himalayas, and for that region see how much mass-wasting across longer timescales is due to earthquakes vs rainfall. If you only ever looked at the specific regions bounded by high earthquake PGA, you will ignore vast regions that are continuously impacted by rainfall, but never by large magnitude earthquakes, thus underestimating overall impacts of rainfall through time, and overestimating earthquakes. By taking a large region, that is roughly 50% impacted by a single earthquake, we can thus get an appreciation of the relative impacts of earthquakes vs rainfall, which would include the fact that overall earthquake extents are far less than rainfall extents -i.e. the argument about low-frequency, localised high-magnitude events vs high-frequency, large-extent, low magnitude events.
And in response to the final comment, again, we did not map coseismic failures ourselves, and as the Roback inventory is the most complete coseismic dataset for Gorkha this was used.

Comment 17:
• This point above also raises a question about how to compare rates between the controls identified.
For the extreme rainfall, a local average rate is derived and compared to one specific to that event. This seems different for how the comparison is undertaken for Gorkha, where rates for the whole area impacted are considered. As it is known (and shown in Figure 1) that landsliding varies considerably across this area over the long term, how then is the impact of roads handled -is this a deviation in local mass wasting rate for the location of the road only, or a net change in rate for the whole area? When comparing events and controls, how each is treated needs to be consistent, which does not appear to be the case at the moment…Again this has a big influence on the simple summary stats….

Response 17:
We accept that we needed to be clearer with how we compare drivers over different scales. We no longer look at roads. As for the other drivers, we use two consistent scales for comparisons in the revision. One, we look at specific extreme storms / localised earthquake preconditioning using the gridcells in figure 5. Two, when making the overall comparisons between net contributions of each driver, we use the scale of the entire region, estimating the total net contribution of each across our whole mapped area, as outlined on lines 624 -662: "With multiple different mass-wasting drivers isolated, we can estimate the relative contributions of each to overall mass-wasting across our mapped region. Across this region, the average ASM-triggered mass-wasting volume for unperturbed years (1988 -1992, 1994 -1999, 2005, 2007 -2014) is 1.89 x10 7 m 3 . From the equations in fig. 2, we know how much mass-wasting would have been expected based on the ASM-precipitation in each year. As such, on the assumption that any mass-wasting above the expected in years known to observe an extreme event are due to that event, the total mass-wasting attributable to each extreme event type can be estimated from the difference between the expected and actual.
To estimate the total relative mass-wasting due to extreme rainfall events, we first calculate the sum of the differences between expected and actual total mass-wasting volumes for all years known to have experienced a storm (1993, 2000 -2004 and 2006 • Figure: 1 -no admin boundaries make this difficult to orient. DEM data has no source / attribution. DEM scale is odd -seems to be reporting a new highest place on earth! Surely the max elevation is 8,848 m, not >11,000 m?

Response 18:
Added the outlines of the Districts of Nepal to help orient figure. DEM issue has been solved (DEM corrected and filled), with correct elevation now stated. Attribution to JAXA/ALOS now included in figure reference.

Comment 19:
• Figure: 2 -confidence intervals would be useful here, particularly given the variability on the log axis.

Response 19:
We have added error bars to all points to help the reader get an understanding of the uncertainty -this error includes the 20% mapping error applied to all events (intended to account for unseen impacts of undercutting, socio-economics, varying mapping intervals etc.) and the standard deviations in the Larsen parameters used to undertake the area to volume conversions.
Comment 20: • Figure: 3 -it would seem sensible to only show EQs of sufficient magnitude to trigger landsliding (e.g. 6Mw…) not all.

Response 20:
Fair point, now only show the main earthquakes in 2015, 2011 and 1988 (the only ones above Mw 6.0).

Comment 21:
Supplementary data: • The mapped shapefiles would be more widely useful than the *.txt file of summary stats.
• I would also like to see the full inventory posted to this repository for wider scrutiny and for use in other research: https://www.sciencebase.gov/catalog/item/586d824ce4b0f5ce109fc9a6

Response 21:
We said this to the editor in the initial cover letter, but may not have reached the reviewers: we fully commit to publishing the full dataset as shapefiles shortly after the publication of the paper.

Reviewer #3 (Remarks to the Author):
Oliver Francis

Comment 22:
The work presented here in this manuscript quantifies the volumes of mass wasting produced by different processes across a 30 year time period and argues these are meaningful for understanding long term erosion rates and forecasting changes due to climate change or road construction. However, without an understanding of the frequency of these processes it is impossible to convert these volumes into rates and put them into context of time scales longer than the 30 years recorded here. I recommend identifying the return period of the extreme rainfall events in 1993 and 2002 and comparing those with the frequency of Gorkha Earthquake in order to fully understand their relative contributions to erosion rates.

Response 22:
We agree that an analysis of event frequency is necessary to facilitate a full understanding of the relative long-term contributions of the events we have identified, and have made this clear in the manuscript on lines 185 -188: "Second, what is the return period of these events? This latter point is particularly important, as quantifying the expected frequencies of extreme rainfall events is vital for hazard management, as well as for gaining a better understanding of longer-term mass-wasting contributions." As outlined to reviewer 2 in comment/response 12, we have then substantially expanded our section on extreme rainfall, where we now specifically we specifically define extreme storms based on the 2002 event (lines 200 -212), before using a new dataset of extreme rainfall data (lines 190 -198) to identify how many extreme storm events similar to 2002 have occurred (lines 214 -226), and thus estimate recurrence intervals (lines 263 -270). We also compare these to the estimated recurrence intervals of earthquakes of magnitude comparable to Gorkha (lines 418 -422).

Comment 23:
Along a similar line I also found the quantification of the contribution of road construction to mass wasting confusing. It is unclear in the manuscript how these mass wasting events are identified, are they simply the volume of material produced during the construction? Or do they include the enhanced mass wasting associated with undercutting of slopes and poor slope reinforcement? Separating these two processes is important for understanding the continued contribution of road construction to mass wasting after the study period. It is also unclear how the recorded areas are converted to volumes; these events are not landslides so using a landslide area -volume scaling is not appropriate. Volumes of material produced by the construction of the roads could be estimated by measuring the width of the roads and the heights of their cuttings into the hillslope. Accurate estimations of the mass wasting produced by construction and post construction along with road mapping would allow forecasting of mass wasting of future construction projects.

Response 23:
We agree that separating road-tips (which is exclusively material excavated by during road construction and dumped onto hillslopes) and mass-wasting enhanced by road undercutting. The events we mapped were exclusively road-tips and did not include any enhanced monsoon-triggered mass-wasting due to road undercutting. This is because identifying which monsoon-triggered mass-wasting events were influenced by roads, and to what degree a given road actually influenced a given mass-wasting, would require far higher resolution road-data than is available for much of our time series. We account for the influence of road undercutting on our monsoon-triggered events by implementing a 20% error on all areas, where part of this error is designed to account for any mass-wasting events mapped as monsoontriggered, that were actually influenced by undercutting by rivers or roads. We had been using Larsen's shallowest soil landslide type parameters for calculating the volumes of these road tips, but in retrospect we agree this is inappropriate. Furthermore, whilst we agree that in theory it could be possible to estimate the volumes of road tips by measuring road widths and cutting heights, this would require DEMs with a resolution of at least 5 m. Unfortunately, we only have access to 30 m DEMs, and whilst 12.5 m DEMs could be purchased, this would still not be sufficient to accurately estimate road widths/heights, which are typically < 10 m. As such, we are unfortunately unable to estimate road volumes at the current time, and consequently have made the decision to remove all work relating to road-tipping from this manuscript. Whilst we think it presents a fascinating area of further study, we simply do not have sufficient data at the current time to properly assess road impacts, so believe that removing roads and focusing on earthquakes vs rainfall will make the manuscript more streamline and impactful.

Comment 24:
Line 54: I find the definition of the "background" rate confusing. Is the background rate simply the estimated rates of mass wasting from the monsoons? Or does it also include storm induced mass wasting outside of the monsoon months? Clarifying this definition along with the timing of the monsoon will help to better explain the aim of the paper.

Comment 24:
The word "background" was intended to refer only to mass-wasting due to the annual monsoon. I.e. to only include regular low-magnitude drivers, and not include low-frequency high-magnitude drivers such as storms and earthquakes. This is now defined / discussed on lines 39 -43 and 61 -64: "Background rates of mass-wasting are driven by tectonic uplift 2,7 and climate [8][9][10][11]

Comment 26:
Line 79: The definition of mass wasting associated with road building is confusing. Is the sediment produced by cutting and constructing the road recorded or is it the enhancement due to undercutting of slopes? The former is a single event while the later is a continuing process. See comment/response 1 and comment/response 23; roads now removed from manuscript due to lack of supporting data.

Comment 28:
Line 112: The magnitude-frequencies of the mapped mass wasting events should be included as a figure to help the reader understand their statistics better, particularly as their shape is used in the methodology.

Response 28:
We have included a new figure (4), which shows the cumulative distribution functions of the areafrequency data for key selections of our inventory: 1993 event, 2002 event, the post-2015 years, and the unperturbed years.

Comment 29:
Line 134: How is the strength of a monsoon defined? Strength could be total precipitation or intensity or another measure of precipitation. I would avoid the word strength and be more specific.

Response 29:
In this case we define strength specifically as the total grid-average monsoon (May -Sept) precipitation across our study region. We have now explicitly defined this in the manuscript, making it clear that  (Fig. 2)."

Comment 30:
Line 138: I found the current wording of the methodology confusing. As it is currently worded it sounds like a single mean is produced of the monsoon landsliding rather than a yearly estimate. I suggest rephrasing it to: "each of the 29 ASM seasons."

Comment 31:
Line 153: A figure of the yearly frequency -magnitudes would help the reader interpret and understand these findings themselves.

Response 31:
We have included a new figure (4), which shows the cumulative distribution functions of the areafrequency data for key selections of our inventory: 1993 event, 2002 event, the post-2015 years, and the unperturbed years. We do not show these for all years, as having 29 separate curves is unreadable, and detracts from the comparison between the main events.

Comment 32:
Line 161: If a monsoon can produce mass wasting significantly greater than the estimated value from the normalisation it raises the question of how well the normalisation works. If this storm is as anomalous as is implied by the normalisation the manuscript should justify why it is not included in the normalisation. This would include discussion on how to identify extreme events, particularly as they seem to be exponential rather than normally distributed in nature.

Response 32:
We have added substantial new discussion r.e. extreme rainfall (see comment/responses 12 and 22).
This includes more explicit discussion of why the extreme rainfall events don't fit the normalisation

Comment 33:
Line 167: The use of landslide density is confusing and hides the fact that the 2002 storm was more erosive than the 1997 storm. I suggest presenting all findings in terms of volume rather than area.

Response 33:
All results now presented with volume, usually as volume/density (m 3 /km 2 ), to account for the varying grid cell sizes used in much of the new analysis.

Comment 34:
Line 169: To combine these two events I feel the return time of the rainstorms needs to be understood.
Without knowing the return period of the storms, it is unclear whether these events are frequent, and therefore important to the long-term erosion rate or insignificant. Combining them together could make them seem more significant than they are.

Comment 35:
Line 175: Preconditioning is set up to be an important part of the manuscript however very little time is spent on describing it in this study. A figure showing the described correlation between earthquake magnitude and preconditioning (particularly as no mention of preconditioning prior to 2015 despite multiple >Mw6 earthquakes) would help to give more attention to this process.

Response 35:
As outlined in comment/response 11, we have added substantially more discussion and analysis on earthquake preconditioning. Due to figure limitations, we decided not to include the suggested figure, though we do add a paragraph to fully describe preconditioning (lines 273 -288), and devote significant space to discussing why earlier M w > 6.0 earthquakes in 1988 and 2011 did not induce the same short term perturbation as seen following 2015 (lines 354 -389).

Comment 36:
Line 177: The timing of the earthquake compared to the monsoon is not clear. It should be noted somewhere that the earthquake occurred prior to the monsoon. It is also unclear whether the coseismic landslides were mapped or the Roback 2018 inventory was used.

Response 36:
Timing of Gorkha relative to the monsoon now included (line 281-282). Clarified that we did not map any coseismic events (lines 103 -105), and that the pre-existing Roback inventory was used to look at the coseismic events (lines 402 -405)

Comment 37:
Line 178: It is not clear why preconditioning increases mass wasting by a factor of 4 but is only the same as 2 ASMs. This should be rewritten for clarity

Response 37:
The ASM-normalised mass-wasting rate is strongly dependent on the rainfall observed in that year. So as the 2015 rainfall was very low, 2015 had a factor of 4 perturbation relative to what would have been expected given the rainfall, which equates to approx. 2 ASMs in absolute terms. To avoid this confusion, the section that included line 178 has been completely re-written, so that the normalised mass-wasting rates and ASM-equivalents are now in completely different sections (the former in lines 290 -300, and the latter in lines 391 -405).

Comment 38:
Line 187: Reference to figures Line 193: Again, reference to figures

Comment 39:
Line 196: If this were true surely the uncorrected trends would not decrease through time? They maintain a similar trend to the corrected i.e. decreasing to the background rate which implies that preconditioning is important here too

Comment 39:
This discussion hinges on what is considered to be preconditioning. If preconditioning is only considered to be new failures, then we see that the corrected rates fall immediately back to near the normal. And yes, the uncorrected rates also decay back to the normal in a few years (due to less and less reactivations), but by the strictest definition this still wouldn't be considered earthquake preconditioning, instead being due to the natural process of material slowly being transported off hillslope and/or revegetation stabilising previously unstable coseismic deposits. However, we accept this is an area that is open for debate, and highly dependent on definitions, so we have added several lines explicitly defining different potential types of preconditioning, and how our results relate to each (lines 304 -312).

Comment(s) 40:
Line 197 of mass wasting associated with road construction will also help to clarify how they are related and defined.
Line 202: As area is shown in figure 4 it is now not clear whether volume or area is discussed here. I again recommend only using volume when discussing or show results.

Comment 41:
Line 207: A figure showing the absolute values of mass wasting would be useful to show how the contribution by each process changes through time.

Response 41:
Figure 4 now removed, so comparison between reactivation and road processes is no longer valid.

Comment 42:
Line 236: You have not identified the controls on these events, are these years of intense monsoon that are poorly predicted by the normalisation method or are they freak storms of abnormal intensity? It is also possible similar storms occurred without causing significant mass wasting, without any analysis on rainfall patterns in this period it is difficult to interpret these events. Without clear definitions of extreme events and their return periods it is difficult to determine whether these events are likely to change through time.

Response 42:
We add several sentences to explicitly discuss the questions raised here (lines 169 -175), concluding that it is the latter case, with these being freak storms of abnormal intensity. We clearly define what we mean by an extreme storm (lines 200 -212) and identify other storms that occurred without mass wasting (lines 214 -226). This allows a more quantitative assessment of how often these events are likely to occur (e.g. recurrence intervals; lines 263 -270).

Comment 43:
Line 261: The timing of the monsoon should be mentioned to help the reader understand why these months were chosen for imaging

Comment 44:
Line 300: This section should be above the section discussing precipitation as it links directly to the discussion of mass wasting.

Response 44:
Moved above precipitation discussion (now line 509)

Comment 45:
Line 343: The averaging of the monsoon should be using volume rather than area. Area is not a useful measure of sediment flux due to the relationship between landslide area and volume.

Comment 46:
Line 370: Here it would be useful to have a figure of the magnitude -frequencies of the years described.
The reader can then made informed decisions on the applicability of this methodology.

Comment 47:
Figure 2: It would be good to have the events discussed in the manuscript highlighted in this figure.
Some form of time series of precipitation would also help to understand the extreme rainfall event described.

Response 47:
Added monthly rainfall time-series to figure. Added labels for the main perturbing events.

Comment 48:
Figure 4: The "Reactivations + Road" line is hard to see. Area is difficult to compare with volume as discussed in the rest of the manuscript. Area is not important for erosion rates and should not be compared. Absolute volume values through time would be much more useful.

Response 48:
Figure now removed.

REVIEWER COMMENTS
Reviewer #3 (Remarks to the Author): Review of revised Jones et al 2020 Nature Communications paper Oliver Francis This revised manuscript has tackled the major comments I raised in the previous round of reviews. The manuscript quantifies the impact of the Asia Monsoon on the Himalaya via an in-depth study of a 30 year long mass wasting inventory. Quantifying the mass wasting associated with the Monsoon has allowed the authors to identify the contribution of extreme rainfall and earthquake preconditioning to erosion rates -the first study to do so in the Himalaya. Alongside this they have identified a potential topographic signature of long-term earthquake preconditioning in the highest and steepest parts of the mountain range. I have some minor comments and suggestions for the authors, primarily on clarifying some of the terminology and definitions they have used. I have summarised the main points in the paragraphs below and listed the other points with line number at the end of this document. I found the terminology used to describe the various mass wasting volumes (Total mass wasting volume, Corrected mass wasting volume, and Scar volume) used in this paper confusing. In the methodology section (lines 513 -514) Scar area (or volume) is described as the result of a correction of the total landslide polygons to remove the influence of long runout landslides. This leads to confusion when discussing the Corrected mass wasting volume, which is just the total mass wasting volume minus the mass wasting associated with reactivating previous deposits. I suggest renaming the "Corrected mass wasting" to a term which closer describes the process by which it is produced, perhaps "New mass wasting" or "non-reactivating mass wasting". I also would like to see some uncertainty analysis undertaken on the definition of an extreme rainfall event. Currently the authors use a threshold of 1.8x the mean annual 24-hour total of precipitation of a cell to define an extreme event based upon the 2002 rainfall event. This threshold could be too high for other unrecognised events which trigger abnormal volumes of landsliding. A possible method of quantifying the uncertainty in the precipitation threshold used could be via analysing the mass wasting inventory. A threshold for an extreme mass wasting event could be defined and compared to the precipitation threshold required to produce it, any unpredicted extreme mass wasting events can then be quantified and used as uncertainty for the precipitation threshold. Line 632: A complete list of the perturbed years rather than a hyphen would be clearer (2000,2001,2002,2003,2004) Figure 1: Does this figure include the Coseismic mass wasting? Figure 3: The left Y axis (ASM Normalised Mass wasting rate) could start at 0 and the Earthquake Magnitudes could be displayed on the left for clarity Figure 4: The symbols in the legend should be larger to see the colours easier

Review of Jones et al,
The author present a new long term, multitemporal catalogue of landsliding in a substantial fraction of the Nepal Himalayas. Specifically they map about 30 years of annually resolved landsliding, that they use to discuss how monsoon strength, extreme rainfall and earthquake drives the regional landsliding. The methodology follow previous works (Marc et al 2015(Marc et al , 2019 in attempting to calibrate a relation between monsoon strength (estimated by satellite) and total landslide volumes, and then quantify deviation from this relationship to highlight the influence of earthquake (by focussing on the monsoon after the Gorkha earthquake) and extreme rainfall. After these analysis the author attempt to do estimate the relative erosion caused by the extreme rainfall and earthquake trigger relative to the annual monsoon.
I note that the manuscript has been drastically changed based on previous referees comments, so that one control effect (road) has been removed, while entirely new analysis on extreme rainfall and the role of past earthquake have been added. I appreciate clarification has been added about the methods and implications, but still I have several questions that may be important both on the old and new analysis.
As of now the key result is the identification of an empirical relationship between monsoon rainfall and annual landsliding. This allows to better constrain the impact of the earthquake, although the authors do not add much on this from The control for extreme rainfall and is likely an interesting result but is needing some clarification/extra analysis to be entirely convincing and quantitative. A key issue is that the author use diverse satellite rainfall (with severe limitations in steep topography), one for the long term monsoon strength and one for the extreme rainfall. The budgeting of the different erosion factor is rather weak in my opinion and also likely need some clarification/updated analysis.
So overall, I think the work is a nice prolongation of Marc et al 2015/2019 on its first part (landslide mapping, monsoon forcing, EQ perturbation) although it cannot be published in this current form as substantial revisions and clarifications are needed. Then the paper is spreading in several directions with the newly added analysis (on extreme rainfall and EQ budget) and may need either substantial work to be at the level of the existing literature or need to be left for later.
My recommendation would be to focus on the new results that should be very valuable to the earth surface/natural hazard community (the MIL/ERIL scaling and the EQ preconditioning as a function of PGA/topo) and leave for future work claims about the sediment budget (between MIL/ERIL/EQIL) that are currently quite poorly analyzed.
Below I detail the major issues I have, a few minor issues and a list of line by line comment/suggestions. In this review I used frequently the term MIL, EQIL and ERIL for Monsoon, Extreme Rainfall and EQ induced landslides, respectively.
Sincerely, do not hesitate to contact me if needed

Major issue 1) Comparison of MIL and ERIL :
I think the approach of the author (triggered by the revisions and not planned) is problematic for several reasons.
1.1) The authors use Persiann to define the background MIL (in the form of a relation between total landsliding and the total rainfall over 5 months), because of its long term availability and stability. But then given that it is consider inadequate to capture extreme rainfall they use TRMM to constrain anomalous years.
I do believe that 1993 and 2002 are likely due to additional extreme events, as reported in the literature and as captured by TRMM for the 2002 events. However, I have many doubts about their following analysis.
Thus several suggestions come to me : Why to use non-overlapping product with different rainfall retrieval methods… At this point using Aphrodite raingage product (Yatagai et al 2012) may be better : It was shown by Andermann et al 2011 to be likely the most reliable rainfall product across the Nepal Himalayas. So potentially it could solve the author problem, by giving them a prediction as Vtot = Rmsn + Rextreme, from 1988 to 2015. Indeed with 0.25° daily resolution from 1951 to 2015 Aphrodite has the potential to capture daily extreme and long term trends. (True in this case the "postseismic" in 2016-2018 should be calibrated with another dataset… This seems less problematic to me as better satellite data exist toward the present (and could be blend with Aphrodite maybe) and as to me the true novelty of the work is not in the post seismic calibration as much as in the meteorological calibration including MIL and ERIL) I show below that the data on Landslide scar volume provided by the author does match monsoon rainfall for most year (especially after removing the largest landslide see comment below). 1993, 2002 and 2015 are outlier. Some other years too. On top of that, I show below rainfall maps from Aphrodite for the 2002 and 1993 event date given by the author which confirm that these events were captured by Aphrodite.
Rainfall map in mm for July 20+July 20 of 1993, from Aphrodite product. Both days had several gridcell >100 mm/day. Extent roughly similar to Fig 1 and 5 from the authors.
Same as above for the 23 of July 2002. Several gridcell have > 250 mm/day. Anyway, Aphrodite seems to be the best, unique product with which the author could solve the MIL and ERIL respective influence, without introducing bias from different sensor and algorithm. It could also allow a unique treatment of extreme from 1988 to 2015, based on gridded daily rainfall.

1.
2) It is not clearly explained how the authors jump from a whole area/ whole monsoon landslide mean to an analysis that relies on localized grids (for the ERIL and also EQ long-term preconditionning analysis). If they compare the landslide density in one given gridcell in one year to the 29 long term mean, or even the density predicted in average for this year it seems inadequate !! Because it is clear that the density varies a lot spatially during their 29 period, so they would need to account for that.
Could not the author show that the deviation from Monsoon forcing (

2) ERIL quantification and their long-term budget:
The definition of extreme rainfall and their recurrence time is fairly naive and with some inconsistencies. In part because they use short time series (Aphrodite would help), but not only. For example, why do the author keep a mixed threshold of 200 mm and 80 % increase (L212) ? Also a % increase relative to the mean is not telling you how much "extreme" is an event without knowing its standard deviation… An Z-score or anomaly (i.e. [X -mean(X)]/Sigma(X) ) would be the qualitative minimum... But more generally the standard way to look at extreme value is through the (Generalized) Extreme Value theory (e.g., Overeem et al 2009, Saito et al 2012, Marc et al 2019b, or through the recent framework of the Metastatistical Extreme Value distribution (see Zorzetto et al 2015). In both framework in each grid cell the author could have fit a GEV or MEV distribution to the daily rainfall data of Aphrodite over the last 30 years… With this the author could define quantitative recurrence interval for diverse daily intensity across their study area and thus "how extreme" were the event in 1993 and 2002. I think these quantification are needed to be consistent with the literature on extreme rainfall and to adequately estimate their budget relative to background MIL (especially through their recurrence time that may not be the same for 1993 and 2002).

3) EQ contribution to long-term budget:
I think the conclusions of the author (EQIL is negligible relative to rainfall) is an inference based on several mistakes done while comparing MIL and EQIL. I suspect that unless extending substantially the paper (with new methods and discussion), the authors may rather want to drop this part and limit themselves to rainfall control and preconditionning.
3.1) Clearly the long term budget of erosion due to EQ is not only due to the recurrence of one type of EQ, but to every EQ causing substantial ground shaking (so from Mw ~6 to 9 in the Himalayas, as paleo landslide data suggest (Schwanghart 2016). The approach of suming the impact of various EQ size is classic (Keefer 1994) and was applied in Nepal with substantially more care (Marc et al 2019), so if the author want to bring anything useful they cannot just show the impact of Gorkha and its recurrence time.
Further, the author neglect the aspect that MIL and EQIL may also have a different potential to trigger very large/ deep landslides.
3.2) EQ affected area vs monsoon affected area : Given your zone of interest is much larger than the EQ affected zone (despite imagery very few landslides occurring the Siwaliks and in the most eastern part of your mapping areas, where your fig 1 show a lot of the total monsoon induced mapping.) you are bound to underestimate the ratio EQIL/MIL … Possibly by almost a factor 2.

4) EQ long term preconditionning
Here my main interrogation is about the method : L331 : We then plot these summed PGAs against each grid cell's 2015 monsoon season percentage change in mass-wasting volume density (m3/km2) ( Fig. 7a -b)." I am not sure what the authors did, as for the approach to spatialize ERIL (see comment 1.2). L216 they say they compare in each cell the total landsliding normalized by mean landsliding. Does it mean they do not account for a meteorological forcing, and the fact that strong monsoon should cause above average landsliding ? Or they also find a scaling at the 0.25° scale ? This should be clarified ! Also I think the author approach of coupling PGA and Excess topography (which essentially gives how frequent (and steep) are slopes in each gridcells) may be quite promising but I still think several clarification are needed : Fig 7 (and S1) should plot the Excess mass wasting VS Excess Topography (NOT the opposite as it is now), so that knowing PGA and Topography, mass wasting can be predicted. This will change the R2 and I recommend the author to check if the best relationship is linear or not (log log plot) and recommend that they check if summing PGA above a threshold (as often PGA<= 0.1-0.2 g do not cause landslide and may not cause preconditionning, see Meunier et al 2007, Yuan et al 2013 improve their scaling. Also it is not clear what they mean by weighted (From the values, it seems they simply did the product of PGA and topography). If so they should rather say that. Last the author should comment on 1) potential cross-correlation between PGA and Topo from diverse EQ, 2) The influence of various threshold angle (the one they used is not even given in the methods !).

L96 visual inspection ?
Results L128: "22 years display moderate to strong exponential relationships between mass-wasting volume per unit area and total grid-averaged precipitation, with R2 values for these years of 0.61 -0.66 for" I think the wording is slightly ambiguous: what about something like : we found that mass-wasting volume per unit area is increasing total grid-averaged precipitation L298: In 2017 you state landslide was perturbed by 1.5-1.7 but with their uncertainties (on the annual volume) they are largely overlapping with the monsoon scaling 1 sigma… L304 -L308: I agree with the author that both reactivation and new landslides are important for he postseismic evolution of hazard but do not think they should discuss the "definition" of preconditionning. They preconditionning (as it name suggest) means the "conditions" of the landscape as changed so that future landslides may be more of less likely in (part of) the landscape. If it is due to strong shaking induced damage, there should be more NEW landslides. The part due to coseismic landslide reactivation ( L320: "less well understood" … Not sure the short term preconditionning I better understood, but it is more frequently observed. So I would rather say that. L321-322 : You suggest only large EQ can mobilize deep landslides but this is not so simple when you look at data either in Nepal or elsewhere, where very deep landslides can also be mobilized during rainfall (eg Morakot or Talas typhoon in Taiwan/Japan, see Marc et al 2018 ) L325-327 : Why do you consider all these EQ for their "preconditionning impact" and not for their coseismic impact and thus for the long-term budget of your conclusions ? Also you are contradicting yourself : you said long preconditionning may affect EQIL so you should look at Gorkha coseismic landslides (NOT the MIL) and see if they are better predicted when also considering past EQ shaking, like what they did in Parker et al., 2015.
L345 : you report R2 = 0.34 in text but R2= 0.377 in Figure S1. Also this figure should plot Excess mass wasting VS Excess Topography (NOT the opposite as it is now) and then show the R2 … Methods L465 : it can be pansharpened with pan chromatic imagery at 15m resolution (This is not a technically a resampling as it is an extra layer of different data.) L474: "All types of rainfall triggered mass wasting" Unclear and the reference is not helping. Do you mean not only landslides but also say, rilling, debris flow ? Or do you mean mapped landslides were not differentiated by types ? Or something else ?
L481: "fully within existing mass-wasting" is a bit confusing; Do you mean that you mapped textural changes within active landslides, for which the responsible process is likely unclear and A-V scaling unconstrained. Or do you refer to the scarring of a revegetated previously mapped landslide ? In this second case it has more chance to be a proper landslide, and a reactivation too, but I would phrase: "scarring/vegetation disturbance within the boundary of previously mapped landslides were considered reactivation." something like that. See Major comment fro the required clarification between Reactivation and Remobilization.
L499 : I am not sure this prove that the difference in mapping duration is not affecting your result, given obviously the primary control will be ASM rainfall. I appreciate your effort but if you want to support this claim I think the figure should show that there is no correlation between the non monsoonal number of months (ideally the non monsoonal amount of rainfall in case some year had extra monsoonal substantial precipitation) and the deviation from your landslide volume / Monsoon rainfall relationship.
L507 : or roads ? From the rebuttal I had the impression roads related slides where mapped separately and then not included in this version, it is not reflected in this sentence.
L537 : "it is the only accessible precipitation product with a spatial resolution of at least 0.25o by 0.25o that spans our required time period of 1988 -2018" This is simply not true. At least the two following product MSWEP (Beck et al 2017) and Aphrodite (Yatagai 2012) cover the period and have the resolution you want. I do not think you have any reason to use Persiann, which by only using infrared measurement is severely limited (see conclusions in Beck et al 2017).
L559: Ok but others product are also made for that, possibly with less issues (for daily rainfall) see Yatagai 2012, Beck et al 2017 L619-621 : You argue you can remove the Jure landslide because of progressive failure. I agree but it may not be the only large landslides that may be decoupled from the monsoon. In Marc et al 2019, several other landslides were shown to be anomalous relative to the monsoon in which they occurred, and this was for only a few years…. So clearly your analysis may have several years disturbed by exceptional landslides, and removing only Jur may not be adequate…. In the figure above I show the simplest correction that is to remove the largest scar volume in each year… In Marc et al 2019 a threshold for the ratio between the largest and second largest scar area seemed to be able to discriminate suspicious slides, and could be reused here.  Review of 30-year record of Himalaya mass-wasting reveals landscape perturbations by extreme events

Oliver Francis
My area of expertise is the remote sensing of landsliding from satellite imagery and the impact of earthquakes on landsliding and erosion rates. Therefore, I feel I have the expertise to review this paper.
The manuscript by Jones et al. attempts to identify a relationship between the strength of annual monsoon rainfall and mass wasting volume in the Himalaya and determine factors which can perturb this relationship. Their analysis has revealed that intense localised rainstorms can produce above average mass wasting for a given monsoon strength. These events should be considered separately from the overall trend of mass wasting and monsoon strength to ensure the best fit. These events could reduce the mass wasting of the next large event as the landscape needs to recharge before its slopes can fail en masse again. By contrast earthquakes can precondition areas to fail at higher rates than expected, though this effect seems to be short lived. Areas of high excess topography showed the greatest amount of preconditioning following the 2015 Gorkha Earthquake, however it is unclear whether this pattern holds for other earthquakes.
In my review of a previous version of this manuscript I raised some minor concerns concerning the use of some terminology and how extreme rainfall events were determined in the study. I believe my previous comments have largely been addressed in this version of the manuscript.
I believe this is good well written manuscript and I found it to be interesting and thought provoking. I believe it is worthy of publication in Nature Communications. However, I have a few comments which I feel need to be addressed before it is ready for publication. I have summarised my thoughts on the manuscript below with specific line numbers highlighted where appropriate.

General comments
The idea that a landscape recovery time exists for the recurrence of landslides in a particular area is very interesting but feels a little underexplored here. The authors suggest that the ability of a hillslope to respond to a perturbation depends on the history of the area, in this case the time since the last major storm. If their hypothesis is correct you would expect to see a decrease in the mass wasting of the local area for several years regardless of whether another large storm occurs or not. In other words, is there a reduction in the mass wasting volume in the year 2003 or 1994 (or even in 2005 in the area which was affected by the extreme rainfall) in the local area of the previous extreme rainfall events? If there is a reduction this would further support the proposed hypothesis.
It is not completely clear why the authors chose to focus on excess topography as a measure of landslide susceptibility to combine with PGA, particularly as it seems to work as an indicator for only one of the earthquakes studied. Previous studies (Meunier et al. 2008;Marc et al. 2016) have indicated that topographic slope of an area is a reasonable approximation of the susceptibility of slope to coseismic landsliding and therefore seismic damage of an area. A comparison between the impact of excess topography and slope on the prediction of postseismic landslide preconditioning would support the author's choice of metric.
I feel the manuscript could use some additional figures to help the reader better understand the relationships between PGA, excess topography and mass wasting volumes. I recommend the authors consider producing a map showing the epicentres of the earthquakes being analysed along with the excess topography and PGA of the earthquakes highlighted. Currently the reader has to solely rely on the author's description of the area which does not feel sufficient as the author proposes several reasons for the null results of several analyses. A map or another visual representation of the area will enable the reader to decide for themselves the likely reasons behind the presented results.

Minor comments
Line 68: Authors state the Gorkha earthquake caused a post seismic perturbation of mass wasting before casting doubt on the idea in line 74.
Line 121 onwards: It is not clear what total precipitation > 25mm means. Is this the sum of the precipitation of days which had total rainfall values above 25mm?
Line 195: Authors could be clearer about whether landslides were reported to be associated with these rainfall events.
Line 211: A simple definition of Z score either here on in the methods would ensure the reader understands the use of this metric.
Line 223: Dismisses the role of rainfall in the 1989 perturbation but in line 355 the authors change their minds to suggest rainfall may have played a part.
Line 226: This section is a little imprecise in the description of the gridded Z scores. Line 226 states that the 2004 monsoon has 11 days of Z scores above 12 but does not state whether this is for particular cells or the entire region. Similarly line 231 states the extreme rainfall occurred after the 15th of June but it is not clear if the authors are discussing all 8 of the days. Line 253 discusses the return period of Z scores but it is not clear whether the return periods relate to the entire area or particular grid cells.
Line 237: The comparison between 2 groups of cells at a single time does not seem strong enough to determine whether the previous 2002 event had a long-lasting effect on the area. Instead, the authors could use their % change based analysis used in the 'Impact of earthquakes' section to compare the cells. This would remove any stochastic differences which maybe present when comparing the groups for a single time slice.
Line 319: Are the grid cells used in this analysis the same as the APHRODITE cells used before?
Line 371: This section of analysis would be greatly improved with a map view of the area. Currently the readers have little information of where the epicentres of the earthquakes are in relation to each other until the end of the paragraph.
Line 419: This section does not discuss the other result of the manuscript, that large events can also reduce the likelihood of another extreme mass wasting event occurring.
Line 451: Are landslides from 2011 -2012 mapped in the pre-monsoon satellite image of 2013 and thus discounted from the inventory?
Line 478: What is the difference between a remobilisation and a reactivation? In the rest of the manuscript these terms seem to be used fairly interchangeably Line 509: When coseismic landslides are identified and removed are the remobilisations within them still considered for the next time slice?
Line 581: Words missing or out of order within the brackets at the end of the line.
Line 586: This last section is a little unclear. From what I understand the main product used is PERSIANN-CDR as this covers the entire period. The APHRODITE product is used to validate the PERSIANN-CDR analysis and determine the Z scores of the daily precipitation. This last section suggests that the PERSIANN-CDR product is only used for the post 2015 rainfall events. Figure 4: If the authors decide to put a map of excess topography or earthquake PGA in the main text, figure 4 seems like a good candidate to be put into the supplemental to make room if the number of figures becomes an issue.
References used in this report: On this second reading, I think that the work by Jones et al., has gained in focus and robustness. So I congratulate the authors for the efforts they put in addressing all my comments and performing substantial amount of new analysis. The monsoon relationship is I believe more clearly exposed and demonstrated, as well as the importance of extreme rainfall triggering. The role of excess topography seems quite supported by the data. I think overall the dataset and analysis of Jones et al., will be well received by the geomorphological and natural hazard communities and should have broad impact.
I think the work is nearly ready for publication, and only have a number of minor comments for a few places where some clarifications or nuances could be added. My ultimate main concern is about the extreme rainfall analysis that gives somewhat inconsistent results, probably because of its naive approach (zeta score). I would recommend that the authors acknowledge a bit more the limit of their current approach and dataset or else back it up with some modest refinement of their analysis, and finally update the abstract and conclusions on this question (see my suggestion below).
Below are Line by line minor comments. Congratulation again to the authors for this hard work and interesting results.

Odin Marc
Line By Line Comment (Number refer to marked change manuscript) : L244 :the extremity ? Can you say that ? What about writing : it is necessary to define how extreme the 1993 and 2002 cloud outburst events were ? (the I realize it looks quite a bit like a previous sentence a few line above) .
L262 : "Correlate" is a bit vague and suggest you "correlated" two dataset which I think you did not... coincide may be a better term here.
l267 : Nb of days above the Z score anomaly Interesting, but beware that 3 consecutive days with high zeta score or 3 distributed days with high zeta score is very different the former being rarer (longer recurrence time) and likely more efficient at triggering landsliding … So in case you see that in some years, sequences of anomalies you should mention it, and re assess the return period of such events (by doing Z score of for the total rainfall over X days (X depending on the sequence they find..) and not daily … ) Further it seems to me that this mechanism is directly contradicted by the observations of post seismic landsliding: In the framework where landsliding is limited by the available soil to fail, the coseismic massive amount of failure should exhaust the slopes which we would then expect to have a lesser response to future rainfaill You (and I and others in previous works) have shown the opposite, with more landsliding immediately after the EQ, and then no less in the following years ...
So, to me this explanation seems quite shaky, and inaccuracy or mislocation of the extreme rainfall seems more likely, and should probably be mentionned as an alternative (and if possible evaluated)… Especially you cite a report for 500 mm in a few days over a given area in 1993, how does that compare to Aphrodite 93 event ?
L283-285 : I am skeptical about this statement and also have to say it is not very clear to what recovery timescales the author refer ...10-100 years ?
L287 : Again I would put coincide instead of Correlate. Also more fundamentally, you need to be careful because the rainfall Zeta score are quite poorly consistent with the magnitude of the extra landsliding : 93 is the strongest deviation but one of the weakest rainfall anomaly, while 2004 had no landslide anomaly but a strong rainfall anomaly … TO me the easiest explanations (that are not really mentioned for now) are 1) inaccurate/mislocated rainfall measurement or 2) inadequate quantification of the extreme . For 1) you can check if strong anomaly are just outside of your zone … For 2) you can do the Zscore analysis on 2, 3 or 5 days bins that may be more relevant time scale for widespread landsliding (different year may have been triggered by different timescales of intense rainfall)… If it works such index could enhance FIG, if it does not you can just mention that these longer timescales do not help to explain landsliding ...

L293-294 :
As of now return time are poorly convincing … The 93 landsliding seems to have been a 1/40 yr event, but with the current methods of the author it is a 4yr return rainfall … Reversely for 2004 ... Maybe accounting for temporal cluster of the extreme daily Zscore is the key to this mismatch … If not the authors should at least highlight this inconsistency, and maybe state that "landslide anomalies have 20 yr return (2 in 40 yrs) and relevant rainfall return may span between 5 and 30 yrs ".
General comment/perspective on Paragraph about ERIL : Overall, reading this new addition I wondered why the authors did not apply the same approach for ERIL and EQ preconditioning : They have grids with the average monsoon landsliding from all unexceptional years, they can compute the % change or ratio of the landsliding during the extreme rainfall year and correlate it to the rainfall Z score (no need of threshold) or Z score * ET ?? I guess it is ok if the author consider this as too much, but it seems to me that they already have all the cards in hand and could (potentially) transform their qualitative map into a plot similar to I think the work is nearly ready for publication, and only have a number of minor comments for a few places where some clarifications or nuances could be added. My ultimate main concern is about the extreme rainfall analysis that gives somewhat inconsistent results, probably because of its naive approach (zeta score). I would recommend that the authors acknowledge a bit more the limit of their current approach and dataset or else back it up with some modest refinement of their analysis, and finally update the abstract and conclusions on this question (see my suggestion below).

Odin Marc
Line By Line Comment (Number refer to marked change manuscript): Comment 1: L244 :the extremity ? Can you say that ? What about writing : it is necessary to define how extreme the 1993 and 2002 cloud outburst events were ? (the I realize it looks quite a bit like a previous sentence a few line above) .
Response 1: Phrasing changed as suggested on line 212.
Comment 2: L262 : "Correlate" is a bit vague and suggest you "correlated" two dataset which I think you did not... coincide may be a better term here.
Response 2: Changed to coincide as suggested, line 232.
Comment 3: l267 : Nb of days above the Z score anomaly Interesting, but beware that 3 consecutive days with high zeta score or 3 distributed days with high zeta score is very different the former being rarer (longer recurrence time) and likely more efficient at triggering landsliding … So in case you see that in some years, sequences of anomalies you should mention it, and re assess the return period of such events (by doing Z score of for the total rainfall over X days (X depending on the sequence they find..) and not daily … ) Response 3: We have now added a new paragraph that considers consecutive vs distributed high rainfall days on lines 244 -250. In summary, this paragraph highlights that most of the other perturbations did not coincide with consecutive high rainfall days, indeed, only 2002 experienced just two consecutive high rainfall days. As such, the lack of high rainfall consecutive days cannot explain why 2004 doesn't observe a perturbation, and as we don't see consecutive high rainfall days, we do not re-asses the return periods. However, on lines 306 -308 we highlight that considering consecutive vs distributed rainfall should be considered an important area of future work.
Comment 4: L268-273 : The structure is a bit heavy given you are gone abandon one option What about saying : Given the 2004 extreme anomalies occurred on XX(give dates) it is unlikely that the lack of landslide response could be attributed to an incompletely saturated hillslopes (ref, ref) Response 4: The structure of this section has been changed as suggested on lines 239 -242. Response 5: We have decided to remove the Gruber permafrost data as it does not contribute to any of the results / discussion points.
Comment 6: L280-282 : Careful here, you do not state it but this mechanism (a sort of negative preconditioning) is typical of soil landslide. So do you think your mapping (and especially volume estimate) is mainly driven by slides that depends on the presence of soil, or more bedrock landslides ? On similar data I usually found the large landslides (presumably bedrock) where Further it seems to me that this mechanism is directly contradicted by the observations of post seismic landsliding: In the framework where landsliding is limited by the available soil to fail, the coseismic massive amount of failure should exhaust the slopes which we would then expect to have a lesser response to future rainfaill You (and I and others in previous works) have shown the opposite, with more landsliding immediately after the EQ, and then no less in the following years ...
So, to me this explanation seems quite shaky, and inaccuracy or mislocation of the extreme rainfall seems more likely, and should probably be mentionned as an alternative (and if possible evaluated)… Especially you cite a report for 500 mm in a few days over a given area in 1993, how does that compare to Aphrodite 93 event ?
Response 6: We have now fully re-written this section, lines 244 -308 to highlight the points raised by the reviewer, in particular to make the point that the idea of landscape recovery (negative preconditioning) is contradicted by other observations and to discuss issues relating to inaccuracy and missallocation of rainfall. Furthermore, on lines 300 -303 we compare the 1993 Aphrodite rainfall total to that reported in the literature to illustrate that there are inaccuracies in the extreme rainfall data that could explain why not all perturbations fully correlate with extreme rainfall.
Comment 7: L283-285 : I am skeptical about this statement and also have to say it is not very clear to what recovery timescales the author refer ...10-100 years ?
Response 7: This line has now been removed.
Comment 8: L287 : Again I would put coincide instead of Correlate. Also more fundamentally, you need to be careful because the rainfall Zeta score are quite poorly consistent with the magnitude of the extra landsliding : 93 is the strongest deviation but one of the weakest rainfall anomaly, while 2004 had no landslide anomaly but a strong rainfall anomaly … TO me the easiest explanations (that are not really mentioned for now) are 1) inaccurate/mislocated rainfall measurement or 2) inadequate quantification of the extreme . For 1) you can check if strong anomaly are just outside of your zone … For 2) you can do the Z-score analysis on 2, 3 or 5 days bins that may be more relevant time scale for widespread landsliding (different year may have been triggered by different timescales of intense rainfall)… If it works such index could enhance FIG, if it does not you can just mention that these longer timescales do not help to explain landsliding ...
Response 8: Correlate changed to coincide. Furthermore, lines 244 -250, 275 -283 and 285 -308 have been added to mention some of the issues mentioned here including misallocation of rainfall and the potential of using consecutive days to more robustly quantify the extreme. Again, it should be noted that with the exception of 1 grid cell in 2002, we did not observe any consecutive days with high Z-scores for any of the perturbations so do not further consider this issue here, though we do mention it as a potentially interesting area of future study (lines 306 -308).
Comment 9: L293-294 : As of now return time are poorly convincing … The 93 landsliding seems to have been a 1/40 yr event, but with the current methods of the author it is a 4yr return rainfall … Reversely for 2004 ... Maybe accounting for temporal cluster of the extreme daily Zscore is the key to this mismatch … If not the authors should at least highlight this inconsistency, and maybe state that "landslide anomalies have 20 yr return (2 in 40 yrs) and relevant rainfall return may span between 5 and 30 yrs ".
Response 9: Lines 289 -294 have been re-written as suggested to say that landslide anomalies have return periods of ~15 years (2 in 30 years) and that the return periods of rainfall events capable of causing these anomalies are 5 -30 years.
Comment 10: General comment/perspective on Paragraph about ERIL : Overall, reading this new addition I wondered why the authors did not apply the same approach for ERIL and EQ preconditioning : They have grids with the average monsoon landsliding from all unexceptional years, they can compute the % change or ratio of the landsliding during the extreme rainfall year and correlate it to the rainfall Z score (no need of threshold) or Z score * ET ?? I guess it is ok if the author consider this as too much, but it seems to me that they already have all the cards in hand and could (potentially) transform their qualitative map into a plot similar to Fig 7 but with an index relative extreme rainfall (Landslide anomaly vs Rainfall anomaly in brief !) it would be much more striking and exciting … Response 10: We thank the reviewer for this idea, and whilst we consider this extra work to be beyond the scope of this paper, it is undoubtedly something we will look into for future publications.
Comment 11: L301 : Ref 3 is ok here but Marc et al 2015 is more relevant and should be added here.