This paper uses landscape evolution modeling to investigate the relative influence of seasonal precipitation and vegetation variations on erosion rates across four unique climates. The authors drive their LEM with precipitation data and NDVI-derived vegetation cover to determine, across climates ranging from arid to humid temperate, that precipitation outcompetes vegetation cover as a control on erosion, and that these effects vary significantly across the type of climate. Vegetation effects seem to matter most in middle-moisture climates and least in very arid or very humid climates.

The paper deals with an important issue (influence of vegetation on erosion and how we might separate it from precipitation effects). The study design is simple—this is a good thing—and the well-understood suite of study sites provides a nice starting point for the analysis. I do think though that there is one fairly significant (but fixable!) structural weakness in the analysis, and I see a variety of smaller (also fixable) issues related to the treatment and presentation of data and its derived statistics. I could see this paper ultimately being publishable in ESurf, but it will require substantial revision. Below I detail my main concerns and then move on to smaller line comments.

Main points:

1. I feel that the major flaw in this study is the lack of consideration for time lags and intrinsic time scales (in both models and nature). When sediment is eroded from a particular spot in one of these basins, it takes some amount of time to leave the model domain and therefore register as “eroded,” rather than just aggrading somewhere other than where it started. We have no real reason to expect that this timescale is less than three months (one season). In fact, field studies often show very long time lags between initial erosion of sediment and its ultimate export from a basin (in addition to further complexity like heavy-tailed distributions of travel times). It is therefore not clear to me why a precipitation or vegetation value for a given season should be plotted against the erosion that occurred during that three-month period (e.g. Figure 5 and others). The authors could certainly evaluate this time lag with initial tests of the model, and/or potentially with data from these catchments if it exists, but at the moment there is a fairly restrictive implicit assumption that these catchments export all eroded sediment within three months. At an absolute minimum the paper should have a subsection in the discussion making this clear and discussing any steps the authors have taken to look at these potential time lags and their effects on the results/interpretations. What would make the paper much more publishable is to run some tests to see how long a pulse of sediment derived from say, one very wet season, takes to leave the catchment.
2. Aside from Figure 10, all of the prior figures deal with absolute erosion rates relative to absolute values for rainfall/veg cover. But if we are really concerned about system sensitivity, we should care more about proportional changes in forcing and proportional responses. It feels intuitive that a 10 mm/season change in precip might matter more to an arid landscape (in terms of proportional erosion rate change) than a 100 mm/season change in precip matters to a humid-temperate one. I think the authors could give more insight into the responses of the different watersheds if they also used some sensitivity metrics that evaluate proportional forcing and response, not just absolute forcing and response. Figures 4-9 could benefit from extra panels using such metrics.
3. Methods: I don’t think it is sufficient to cite out the key components of the numerical model dealing with how vegetation influences erosion given that this is the entire point of the paper. I know they have been published before, but I ask the authors to please consider presenting them in brief again. Maybe an appendix would be a good place for them?
4. In general, figure construction could be much-improved in that the figures could have provided more useful information and been made a lot easier to read. There are issues with overlapping data that could have been made partially transparent, certain data series (the arid landscape data) that really should be shown on a zoomed-in inset because their range of variability is so low relative to others, etc. For all figures, there are serious issues with readability for people with color vision impairments. It is extremely simple to remedy this by using different symbols, line styles, or just color selections that are friendly to all (see colorbrewer or similar tools online). There are a lot of people out there with color vision impairment, and they will have no idea what you’re trying to say unless you modify figures 2, 4, 5, 6, 7, 8, 9, and possibly 11. 
5. The statistical treatments presented could also use improvement. Why are we assuming that we are interrogating linear relationships? Do we have theory or data that says all of the relationships examined in this study should be linear? Why do we not see confidence intervals presented to add richness to our interpretations beyond the p-values and correlation coefficients? The statistics here feel like they were done pro forma rather than with thought as to the questions being asked and how the stats might or might not contribute to answering them. Yes, the slopes of these linear fits are convenient proxies for model sensitivity to a particular independent variable, but there are many other metrics for sensitivity that don’t require assuming this one particular relationship between the variables.
6. Code/data availability: It feels problematic that the authors have used open-source software that was given DOIs and made available to the community through the efforts of others, and then not made their own code/data permanently and publicly available. Please help build a community where we archive and share code and data freely by depositing the code, etc that supports this paper in a DOI-stamped repository. I would be surprised if this code/data availability statement is not violating journal policies, funder policies, or both.

Line comments:

Abstract: the last sentence has a low end of 5%, but just above it says 6.5%. Are these different quantities being reported?

34: changes play a crucial role. Also the grammar here is odd because seasonality is a noun. Maybe just seasonal?

39: Or in this case, “plantscape evolution modeling” !!! (just a joke)

45: If this paper is worth mentioning, you should state its main conclusion rather than just its topic

47: pluralization mismatch—proofread for clarity

53: do you mean a reduction in sensitivity of soil loss potential to storm frequency?

54: again here a very vague statement about what past workers did. If it’s worth mentioning, surely it has some relevant conclusion. Also consider restructuring the sentence because differences in vegetation don’t drive erosion and sedimentation.

56: unclosed (

65: could you state the direction of this effect? I can assume, but haven’t read the paper

66: is this species richness or some other metric?

71-72: “…seasonality in precipitation and vegetation cover conspire to influence…”

74: I suppose it could work either way, but I would have reached for “transience” rather than “transients”

75: “across the extreme…” or “spanning the extreme…”

77: hypothesis 1 sounds strangely tautological. Why not just say “1) P is the first-order driver of seasonal erosion rates.” That implies that everything else is of low significance.

121-122: This is fine, but it might be worth adding one sentence to emphasize the limitations of NDVI, chiefly that it saturates out once the ground is basically shrub-covered and so it couldn’t tell you much about different plant communities for associated erosion-relevant properties like rooting depth, etc.

133: Landlab I think(?) is typically written with only “Land” capitalized

136: typo /

137: is this total relief in the whole catchment? Just want to be clear.

140-143: True, but this is not the only reason you could have initial transience. LEMs  (and the systems they represent) have inherent timescales, and you can’t know a priori whether the state of the system as captured by the SRTM DEM and your various other inputs is at a full equilibrium condition. Surely it is less likely to be in equilibrium with respect to all relevant forcings than to be in some stage of transient response.

151: “summing up” can just be “summing”

157: yes and this is fine, but again there are lots of erosion-relevant properties of vegetation that NDVI is NOT good at measuring. It would be good to be up front about this in the writing.

160: could you provide a few words on the resampling method?

Figure 2: I wonder about the utility of plotting these things against each other given that we surely expect lags between the water balance and the vegetation response? 182-193 accurately describes the figures, but does not lead us to any real insights about the system. What do you want readers to take away from this figure? Some other points about this figure: Please check that these color schemes are friendly to all (i.e. are color-impaired friendly). I doubt they are, given the use of red and green. An easy remedy would be to use varying marker styles (squares, triangles, stars) in addition to colors. Also panel c would be better if the markers had some transparency so the blue didn’t hide the red and the red didn’t hide the black. Separate issue: it is not clear to me why we would necessarily expect these relationships to be linear and therefore why one should use a linear regression here to assess correlation. Finally, why don’t you label the different data series by their climate rather than by the study area names? Any given reader will only care about the former.

215: This description is, to the extent that I understand these things, not quite right: Landlab is a modeling toolkit, not a model in and of itself (I think of it as a toolbox holding many models, each of which is a possible tool to use). SPACE is one of many models that operate within the landlab toolkit or framework.

222: Was there actually model calibration in the true sense? My read is that you chose values that are appropriate for the study basins, which is 100% fine, but is not the same thing as running an actual calibration exercise.

222-223: Revise this sentence for grammar

223: does “erosion” mean erodibility? Save for diffusion/diffusivity. Also lithology is not a model parameter; there are parameters that incorporate the effects of lithology. But these have specific and more descriptive names that we should use.

229-230: re-read this sentence for grammatical consistency

242-243: delete “initial.” It’s just the uplift rate for the whole simulation, which is fine.

277: I guess it’s ok if you want to keep it, but I think the units of veg cover are pretty clear so I don’t know how important it is to write [-] after every use. It’s common colloquially to just speak of proportions with no units.

Figure 4: Modify for color vision impairment viewing; this will be very simple.

Figure 5: same as figure 4. Also the arid data should really be shown on a zoomed-in inset or sub-panel. Just because the arid landscape doesn’t show the same absolute response does not mean that it can’t have equal or greater proportional sensitivity than another climate. We can’t evaluate much about this because the black dots are all jammed together in the corner.

Figures 6 and 7: same comments about figure readability and construction as 4 and 5.

Figures 8 and 9: same concerns as above

Figure 10: Why is this rendered as a line graph? These data are not connected in any sort of sequential relationship (except to the extent that precip varies, but if you wanted to plot that you’d put precip on the x-axis). A bar graph would be better, or a stacked bar graph such that all three plots could be condensed into one. Again here I recommend not using site names but their climate labels.

392-394: Not the best example of agreement with your findings. Rainfall intensity in a plot experiment is not the same as total seasonal precip, so this is a bit of a stretch to use as a direct parallel. The others cited here make sense though.

Figure 11: You’re probably sick of hearing it, but please just use different markers or line styles or something to distinguish these series.

442-443: This sentence is vague. How do the dynamics change as a result of changes in veg?

471: “subject to”

486: All of these limitations are fine, but the biggest limitation of the study is that there is no mechanism to consider transient dynamics, for example that sediment eroded off a hillslope in March might take until December to move out of the model domain.

536: see main point above.

This paper investigates a fundamental question-- sensitivty to of erosion to climate across a large eco-climatic and geomorphic gradient. This problem is especially hard to address over long time scales where topographiy reflect the long term meory of the regional climate. The auuthors should be commented for tackling such a difficult problem. The paper can be reconsidered after major revisisons. 

Regional geomorphic context: More information is needed on geomorphology and the erosion history of the study sites. How different are geomorphologies of these different sites. A slope-area plot for each catchment would be telling. Are there landslides or dry soil/rock slides in these watersheds.  What is the erosional history of this region, is the landscape in steady-state or dynamic equilibrium with uplift and climate. What periods do the modern uplift rates cover? Do the millennial erosion rates bracket the uplift rates. Ideally, in order to study the role of climate on erosion rates wouldn’t you need to calibrate your model for each site to targeted erosion rates. That means that in every site your erodibility etc parameter would need to vary in relation to your topography.  The key is that, as the authors would appreciate, the ecoclimatic and tectonic legacy of the system is engraved in the soil/erodibility and topographic properties, if you use the topo from today then you will need to do a calibration to the rest of the parameter space of your model. At least that is my opinion, but I appreciate a justification. Also the “detrending” concept is not well explained in the paper, actually what the authors did was not presented.  

Climate forcing: how can you simulate this model at seasonal time steps (lines 224-225), especially in arid and semiarid climates, where the only way to get erosive runoff would be to capture high-intensity storms. Your seasonal rainfall model will not capture the role of extreme events. Furthermore, how do you solve equation 1 to obtain runoff in seasonal time scales. How do monthly simulations even give you runoff in the drier sites. I know ET and P are inputs, but the infiltration equation won’t produce any runoff unless you have intense storms-- a seasonal time step would only produce drizzles. Am I missing a detail somewhere? In addition, how can we assure that this 20 years of observed weather properly captures the statistical properties of the regional climate. A better approach would be to train a stochastic weather generator and run it for 1000 years without cycling the same rainfall.

Equation (1), please explain how water balance was resolved. Normally ET should be resolved in the model with given P, but my sense is that ET is an external input to the model. If that is so, then P should be used from the same data set. I think reanalysis is a model product, so you need to be consistent because in dry landscapes you may end up with a greater ET than P because of data set differences. Also why not use runoff from reanalysis too if ET is from reanalysis.

Vegetation input in models: Figure 2a was ndvi data converted veg cover fraction? This was not told in the text.  (a) Why is there a generally negative relationship between Seasonal P and fractional veg cover. In the wet site there seems to be two separate data group both of which show slightly negative r. (b) there is so much variability in seasonal ET and veg cover, how is this relationship built in the model? Data from NA shows a strong negative sign between ET and veg cover, how is that possible. I doubt the accuracy of the Reanalysis data, is there an independent way to check this. Similarly, ET-P relationship does not make any sense at the landscape scale. I read the argument for some of these as the steep gradients of temperature, precip, solar radiation, and that make sense, if you introduce the spatial variability vegetation you will need that for hydrology as well as vegetation is variable because of spatial precip and ET etc. That is also true for edibility and threshold related to veg. Does the erosion threshold vary with veg cover.

Fgiure 2 can be augmented with a Budyko curve using the same data sets. The reason is that, it is hard to imagine to have such large carry over mechanisms of soil moisture in these environments to yield a negative relationship between seasonal P and ET. if carry over is neglected Figure 2 c is not physically possible-- zero precip leads to the highest ET? Am I missing the point?

Fig 5—inverse correlation of erosion with vegetation only makes sense if hillslope diffusion is constant. Given this impressive precip gradient, I suspect hillslope diffusion should also depend on precip. I don’t see a variable hillslope diffusion in the model parameters table.

Effects of seasonal variations in vegetation 1 and precipitation on catchment erosion rates along a climate and ecological gradient: Insights from numerical modelling

by Sharma and Ehlers

An interesting paper. Definitely suitable for ESurf. Three simulation scenarios hold the idea of comparing the role of temporal changes of driving and resisting force in sediment production. The climate range of the study sites bring another level of complexity and explanation. These are cool.

I understood the competition between the driving and resistance forces on sediment yield. In the first two scenarios, one of them keeping constant another force is varied. In the third one, both are changing. Arid, semi-arid, wet seasons are relatively easier to comprehend where are in the first and third regions – very low runoff has little impact and no vegetation growth in arid case, or runoff production does not change vegetation more due to (maybe ecosystem shift from water limitation to energy limitation, space competition etc.) (please see fig 12 of Collins and Bras, 2010-WRR) [Similar to Douglas (1976); Fournier (1960); Wilson (1969) can be found in Walling and Kleo (1979)]. However, the beauty of the study in investigating the case of Mediterranean climate when the mismatch between vegetation growth (summer time) and runoff production (winter).

My questions are:

1. Seasonal timestep confused me a bit. How do you relate the erosion rate to vegetation cover directly? Let me clarify. Wet season may cause greater vegetation cover and may also greater erosion rates. Or vice versa. But when erosion rates are greater than the uplift rate slopes are getting gentler. The following wet or dry season, the erosion rates will be relatively less responsive to the vegetation cover and/or runoff production. To summarize my question:

In this kind of long-time steps (seasonal here), the relationship between erosion rate and vegetation cover may be affected by inherited simulated slope (refers to energy in shear stress formula) values from previous season (model time step). So, I am suspicious that the signal may be blended.

1. One of the sites (Nahuelbuta) has a MAP of 1400 mm. If NDVI is notorious for saturation problem (fig. 16 from Huete et al., 2002 - RemSenEnv), NDVI to vegetation cover conversion maybe affected from this problem. You mentioned NDVI values reach up to 0.8. So, in shear stress partitioning (most probably you are using it based on nv, ns etc. from your Table 1) maybe affected from the NDVI to vegetation cover fraction conversion.

Need clarification:

1. In Figs 5, 7, and 9: There should be 20 filled and 20 hollow circles for each site. Is not it? When I try to count, I counted more than 20 or less. Less may be understandable due to overlap. But why more? Did I miss something there?

I have following minor edits:

L71. …the seasonality in precipitation and vegetation covers ARE…. ‘is’ or ‘are’?

L92. Santa Gracia National Reserve.. Upper case for National Reserve.

L98-9. Use lower case for compass directions.    …in the north… …in the south… L94.

L103. Comma after ‘respectively’.

L107. Capitalize ‘plate’. …the Nazca Plate …   …the South American Plate….

L128. Comma after ‘(section 3.5)’.

L136. Typo. Delete ‘/’.

L139. Word order. Either ‘Considered catchment sizes’… or ‘Investigated catchment sizes’….

L140. Insert comma after ‘La Campana’.

L145. It may be better to include/mention that the spatial resolution is ~111 km at the equator. At your study sites (26°S to 38°S), it is shorter than this value.

L149. Do not capitalize seasons. Also L225.

L149. Do not capitalize ‘austral’.

L150. Word choice. ‘Illustrated’ or ‘given’. Please reconsider.

L319. Comma after ‘(LC)’.

L319. It may be better ‘(Figs. 6 and 7)’ …

L532. If the given ‘porosity’ parameter (0.20) for loam soil, it is a bit lower than the reported values in the literature.

a) You cited Istanbulluoglu and Bras (2006) for infiltration values in Eq.3. However, table 1 of Istanbulluoglu and Bras (2006) used porosity as 0.45 for loam. Moreover, in table 2 of Clapp and Hornberger (1978), they gave 0.451 (0.078) (std dev.) for loam, 0.420 for sandy clay loam.

b) When, I read L170-174, bulk densities vary from 800 kg m^-3 to 1500 kg m^-3. Let’s assume the density of quartz is 2650 kg m^-3. Your density should vary from ~0.43 to ~0.70. Please check.