the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 31 Mar 2026
Erosion and dispersal processes drive vegetation trajectories in a highly erosive badland catchment
Abstract. Badlands are among the most erosive places on earth, providing large sediments fluxes to rivers and oceans. In these environments, erosion is strongly controlled by vegetation, whose cover and composition both vary in space and time. Quantifying vegetation dynamics and their drivers is therefore essential to understand and predict badland erosion. Here, we use time series of high-resolution aerial and Landsat satellite imageries to reconstruct 40 years of vegetation change in a highly erosive badland catchment of French south-western Alps. Vegetation cover increased from 38.7 % to 46.2 % of the studied area), primarily through the colonization of bare surfaces by young pines. Spatial patterns of colonization and extinction are driven both by geomorphic factors, such as slope stability and local erosion rate, and by ecological factors, such as grain dispersal. Remotely sensed greening trends are correlated with climate, initial vegetation type and colonization intensity, suggesting that both climate changes and ecological succession are contributing to the greening of badlands. Quantifying these spatial and temporal trends reveals that vegetation dynamics are tightly coupled with erosion, as they both control and respond to erosion patterns.
- Preprint
(1501 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-834', Anonymous Referee #1, 13 May 2026
-
RC2: 'Comment on egusphere-2026-834', Anonymous Referee #2, 27 May 2026
General:
This paper explores the combination of high-resolution, low-frequency data with low-resolution, high-frequency data on vegetation class and NDVI to explore vegetation growth/die-off patterns in a badland complex and their relation to erosion. The research is exciting and timely and its rationale is made clear for a general ESurf readership; the discussion of the results is thoughtful and well-contextualised. Nonetheless, I felt the methods and results need clarification and the figures need polishing; in particular, there were frequent discrepancies between the text and figures. I have made numerous comments below in this vein and therefore suggest minor revisions before publication.
Throughout:
There are grammatical / typographical errors. A few examples (not all): superscripts appear to have been omitted; some references need formatting; units for distance switch between m and cm; figure numbering in text doesn’t match captions.
Intro:
Anderson et al. 2020 is not in the reference list.
51-55: There may be some relevant examples from the New Zealand literature here concerning vegetation controls on badland (I believe this is equivalent to what they’d call “gully complex”) erosion. Mike Marden’s work is probably the place to start, e.g. https://doi.org/10.33494/nzjfs522022x226x, https://doi.org/10.1002/rra.882.
Materials and methods
82 “Highest recorded worldwide” – do the authors have a reference for this statement?
93 “An extended study area…” – what assumptions allow this larger area to act as a proxy for the study catchment? I feel more detail is needed to justify this decision.
Figure 1: What is the underlying image in the top-left panel? How does the 8 km2 Landsat analysis area extend around this catchment – is it marked on this panel? I note that vegetation outside of the catchment (particularly east and north) is more dense and complete – how does this affect its ability to act as a proxy for the Laval catchment? Please add subplot labels. Please remove the spell-check underlines from the bottom-left panel. The caption requires some elaboration – e.g. why were these two photos selected? What are their locations and viewfields? These could be marked in the bottom-right map to orient the reader.
85 “Two areas”: the first is well defined, but I was expecting a second catchment to be defined after. Is the “second” area in fact the “extended study area” included in the Landsat analysis? I recommend clarifying this point.
102: Could the two rain gauges be named on Figure 1, or identified somehow with annotations? This would help the reader know if “Laval rain gauge” refers to the one in the headwaters or outlet.
103-105: Is the Plateau station the same as the Draix station? I recommend clarifying and/or annotating the map.
111: I suggest distinguishing between “high frequency” and “high resolution” throughout. Readers might assume that, given the Landsat data / other rasters used here, “high resolution” refers to spatial rather than temporal resolution.
112 “complemented” how? A quick note on any gap-filling procedure is required.
116: I realize the extended details are in Klotz et al., but a quick note on the meaning of “automatic samplers” (what are they sampling – bedload??) would help here.
137-145: Could have slightly less detail here. I feel e.g. that it is sufficient to say “Landsat 5-8” and “June-September” etc.
147: “NDVImax” in which context: annual? Monthly? The reader needs more context here (I don’t know that NDVImax was defined until much later in the paper). In general, I think this paragraph needs more clear explanation for a geomorphological audience; at present I don’t think the description favours reproducibility. For instance, how exactly is phenology modelled on a pixelwise basis? How exactly are NDVImax estimates “adjusted”? L150 says this method “partially” corrects for sampling bias resulting from increasing Landsat frequency – so how did the authors resolve (or even measure) any remaining bias? Hence I feel more detail would help readers understand the method here.
157: Which “high resolution vegetation maps”? The 0.5 m ones described earlier, in their pre-binarization state? A little more clarity/consistent terminology is warranted.
176: “C2M algorithm” - citation?
182 : “the distance associated with the point cloud” – does this refer to the vertical distance between the two point clouds after their offset was applied? Please clarify.
183 “Annual erosion” – suggest “mean annual erosion”.
188: “local erosion rate” – from the DoD? If so, this could be made clearer.
189: Ellipsis feels inappropriate here.
198: “depend on” – suggest “vary with” to avoid conflating correlation and causality.
Results:
Figure 2: I suggest scaling the distance between columns in (c) to approximate the time interval between years – e.g. the final 3 columns should be closer together. Suggest annotating (a) and (b) with the year.
I note that L 208 says pine forest was 17.3% in 1982 but Fig 2 c shows 18.4%. The paper needs a careful check to ensure details are consistent across text and figures.
214: “Fig. 6” – does this refer to Fig. 3? There are numerous other mismatches of this type; again, a careful check for consistency between text and figures is needed.
Figure 3: What is the red heptagon in panel (b)? Panels a/b need labelling. Areas that “have been removed” need delineating – e.g. is there a minimum red patch size that was removed? Were all red patches removed, or only those outside the catchment boundary? Could the caption explain why both panels are needed, not just a larger view of panel (a)? It might be a good idea to remind the reader in the caption (or the paragraph above) of the duration over which the trends were measured.
Figure 4: Remember units for the y-axis label. Remember to label panels a/b. Could the legend for (b) have more detail about the direction of the transition (e.g. bare-veg vs veg-bare)? I know this is mentioned in the text, but adding it to the legend will help time-scarce readers. I think the legend of (a) needs more detail: e.g. “Density” might be updated to “% of pixels purely in this class” or similar – if I am interpreting this wrong, you’ll hopefully see why more precision is needed! Also, what are the black vs grey markers in (a)? Aside from these details, this figure is informative and eye-catching, and I commend the authors on the great data visualisation.
240: “temporal dynamic of NDVImax, averaged across all pure pixels of each type” – I am confused here: does it mean the trend is extracted pixel-wise for each pixel of e.g. class “Pine”, and then these trends are averaged, and then the trend is compared to the trend in e.g. temperature? Or does it mean that, for each year, the pixelwise NDVImax is averaged across all “Pine” pixels, and this value is then compared to e.g. mean annual temperature for the same year? Clarification will prevent readers having the same confusion; moreover, I suggest moving such details to the methods.
245: The legend of Table 1 goes some way to clarifying my confusion above, but introduces a new confusion: “annual maximum NDVI for each vegetation type” doesn’t mention any averaging. What about “annual maximum NDVI, averaged across each vegetation type”?
Figure 5: Legend text hard to read. Suggest updating caption to remove mention of vegetation “types”, given that this figure is based on the binary data.
257: Slopes were expressed as proportions earlier; suggest using consistent units.
260: How can a slope be higher than 300%? Also, looking at the graph, I think this should say “more than 0.25” instead of “more than 0.5”? As with the earlier figures, please ensure the text matches the values in the figure.
261: Earlier in the paragraph, the first tick-mark value on the y-axis was rounded down to 0.1. Here, 0.12 is used. Please be consistent with rounding across all sub-plots and the text describing them. Similarly, I was confused by “then reaches 0.5 when 4% of the neighborhood…” because, as far as I can tell from Figure 6b, colonisation probability reaches 0.5 when about 20% of the neighborhood is vegetated. Again, please ensure text accurately describes the figures.
274-278 (correlation tests): It would be good to have table or plot in the supplement to show what the authors mean by e.g. “strongly associated” in L 277.
Figure 6: Remember to label the subpanels.
Discussion :
In figure 2 and the relevant results text, it is clear that pine forest has almost doubled in relative presence across the study period. What do the authors feel is the mechanism for this increase? Is this to do with the specific physical traits of the pine species that allows them to outcompete the deciduous forest when there is grassland/bare soil to be colonized? These questions arose when I read the results and I felt like they could be explored in more detail in the discussion.
I felt that the negative greening trends in Figure 3 and the pattern of erosion they reveal (or not) could have been better contextualised against the literature from badlands in other humid and/or rapidly eroding landscapes, e.g. Kumar et al. 2026 (https://doi.org/10.1029/2025JF008720), Taylor et al. 2018 (https://doi.org/10.1016/j.geomorph.2017.10.007), Fuller et al. 2020 (https://doi.org/10.1002/esp.5010), Leenman and Tunnicliffe, 2018 (https://doi.org/10.1130/B31849.1).
On reading Figure 4, I became curious about the fact that “pure” pine pixels show stronger greening trends than other vegetation. To what extent is this related to the age cycle of these pines – do these data record the maturing of trees that were dispersed/self-sown shortly before the satellite era? Or is this greening the result of increasing (a) density or (b) health of trees with no change in mean age? I feel the authors could discuss the potential mechanism driving this “greening” in more detail.
I had similar questions on reading L285-297 : are the authors claiming the greening in “pure” pixels results from increasing plant density, or healthier plants? I’m not sure what “enhanced growth” means in physical terms. Moreover, it’s not clear whether the hypothesised driving mechanism of shorter frost periods relates to new colonisation, “enhanced growth” in “pure” vegetation pixels, or both – I think the authors could clarify this point in this paragraph.
I was fascinated to read in line 265 that “all bare surface were colonized” at high elevations. I felt that this point was not explored thoroughly enough in the discussion – the authors mention lower gradients on the ridgelines, but is there also reduced competition for sunlight? Or, is there some neighborhood effect here given that the other side of the ridge is generally forested, so that pixels along this ridgeline are in the sweet spot of high vegetated neighbourhood and reasonably low slope? I would suggest a little more discussion of this finding.
297: “higher success of recent colonisation” – I couldn’t see any data to back this up.
314: “Consistent with the trends seen in areas with the highest drainage areas” – Suggest to be more specific here about the nature of the trend/pattern.
320-1: “slope threshold therefore corresponds to a stability threshold” – this could very easily be tested by comparing the slope to the erosion/deposition depth at each pixel. “unconsolidated material” – is the badlands comprised only of unconsolidated material? The introduction describes them as developing on “weak sedimentary rocks” which I think of as “consolidated”. The authors might want to specify here if they are only referring to talus downslope of an initial failure that loosened material.
334-335: “both studies analysed existing plant cover rather than dynamic pattern[s]” – could the authors explain why existing plant cover would have a higher slope threshold? The logic here is not clear to me.
337-8: “slightly lower critical slopes on the SE aspect” – do the authors have a plot to back up this statement?
I found the discussion in 383-393 very insightful. Perhaps the authors might highlight on Figure 7 where they feel their study catchment currently is in the cycle, and comment on how this autogenic cycling may be a third driver of the observed greening trend described in L 298-309?
412-416: This paragraph feels a bit out of place, and might be best reduced to a sentence for the conclusion?
436-460: I think this section could be condensed to be more concise; the level of detail is occasionally superfluous.
One thing that I had hoped the authors would reflect on in the discussion are the downstream implications of their results. For instance, knowing what we know now about patterns of vegetation establishment in the badlands in space and time, can we infer anything about the resultant control of vegetation on erosion and ultimately for sediment flux into the main outlet channel?
Conclusions:
465: Can we say “in total biomass”? I don’t remember the authors commenting on biomass at all; suggest rewording to fit with the rest of the paper.
467: What are “slowly eroding, even depositing environments”? I struggled to imagine the physical processes described here and feel this phrase needs clarifying.
A small number of statements (e.g. 468-469) need careful checking for whether they are a true factual statement or an interpretation of the results, in which case the language needs toning down with phrases like “appear to”.
469-470: “depend on vegetation type and ecological succession” – can the authors be more specific here?
Misc: I could not review the vegetation maps, erosion map, and greening map.
Citation: https://doi.org/10.5194/egusphere-2026-834-RC2 -
EC1: 'AE Comment on egusphere-2026-834', Daniel Parsons, 28 May 2026
Thank you for submitting your manuscript to Earth Surface Dynamics. As you can see we have had two reviews and I thank them for their careful, constructive and detailed assessments of the paper as part of the open review process. The reviews identify a number of strengths in the manuscript, including the timely focus on vegetation–erosion interactions in badland landscapes, the potentially valuable combination of remote-sensing, topographic and erosion-rate datasets, and the broader relevance of the work for understanding coupled biogeomorphic change.
Having considered the reviews and the discussion to date, I am inviting you to submit a major revision.
This decision reflects the fact that, while the manuscript has clear potential, the reviews raise several substantive issues that need to be addressed before the paper can be considered further for publication. These concerns are not limited to presentation or minor clarification. Rather, they relate to the framing of the contribution, the connection between vegetation change and geomorphic process, the clarity of the erosion-rate methodology, the consistency between figures and text, and the strength of some of the interpretations.
A central issue is the need to clarify the manuscript’s contribution to Earth Surface Dynamics. At present, much of the analysis is focused on vegetation classification, NDVI trends and greening. These elements are valuable, but the revised manuscript needs to make clearer how they advance understanding of geomorphic processes, erosion dynamics, sediment production or landscape evolution. If the paper’s main contribution is vegetation dynamics, then the connection to geomorphology needs to be strengthened. If, instead, the intention is to demonstrate vegetation–erosion feedbacks, the evidential basis for those feedbacks needs to be set out more explicitly and cautiously.
In particular, please revisit the way greening is interpreted. NDVI greening may reflect changes in vegetation vigour, canopy density, vegetation type, phenology or other ecological processes, but these do not necessarily map directly onto root cohesion, bare-ground reduction, sediment availability or erosion resistance. The revised manuscript should explain what greening is being taken to represent in this study, why that interpretation is justified, and how it relates to erosion processes. Where the link is indirect or inferential, the language should be moderated accordingly.
The reviewers also raise an important concern about the timing of the datasets. The principal vegetation transitions appear to occur largely between 1994 and 2012, whereas the erosion rates are measured later, between 2015 and 2021. This temporal mismatch needs to be addressed directly. Please clarify whether the erosion data are being used to explain earlier vegetation changes, to characterise the contemporary geomorphic setting, or to support a broader hypothesis about vegetation–erosion interactions. If the data do not permit a direct causal interpretation, please revise the framing and conclusions so that this limitation is clear.
The lidar differencing and erosion-rate methodology also require further explanation. Please provide a clearer account of the “best alignment” procedure, the assumptions involved, and how these assumptions affect the derived erosion rates. In particular, please address the concern that an assumption of spatially uniform erosion appears difficult to reconcile with later results suggesting spatial variability in erosion as a function of slope. Please also resolve the inconsistency identified by the reviewers regarding whether vegetation filtering of the lidar data was undertaken by the data producer or by the authors. The revised manuscript should make the processing workflow sufficiently transparent for readers to evaluate the robustness of the erosion-rate estimates.
Figure 6 and the associated interpretation require substantial revision. Both reviews raise concerns about whether the transition-probability relationships shown in this figure are consistent with the mapped vegetation transitions elsewhere in the manuscript. For example, the apparent probability of colonisation across a range of slopes seems difficult to reconcile with the limited spatial extent of colonisation shown in the maps. Similarly, the relationship between extinction probability and drainage area needs clearer spatial support. Please check the calculations carefully, clarify exactly what the probabilities represent, and consider adding additional visualisations that show where these transition classes occur in the catchment. If the current analysis does not support the stated conclusions, the interpretation should be revised.
More generally, please present the variability in the binned analyses more fully. Averages alone make it difficult to assess whether the observed relationships are robust, noisy, threshold-like, or influenced by a small number of observations. Please consider adding box plots or equivalent summaries showing medians, quartiles and ranges for each bin, as suggested in the review. This would make the results more transparent and would help readers evaluate the strength of the reported topographic controls.
The figures, captions and text also need a systematic consistency check. The reviews identify several discrepancies between numerical values in the text and figures, mismatched figure references, missing subplot labels, unclear legends, inconsistent units and insufficiently explained masks or removed areas. These issues currently make the manuscript harder to follow and reduce confidence in the results. Please undertake a comprehensive audit of all figures, captions, references to figures, units, numerical values and panel labels. Where maps are central to the interpretation, please ensure they are available, legible and adequately explained.
The discussion would benefit from a clearer distinction between results, interpretations and hypotheses. Several mechanisms are discussed, including pine expansion, slope thresholds, drainage-area effects, neighborhood vegetation effects, elevation, frost-period change, ecological succession and autogenic badland cycling. These are all potentially interesting, but the revised manuscript should indicate more clearly which mechanisms are directly supported by the data and which are plausible explanations requiring further testing. Statements around stability thresholds, biomass, vegetation dependence and feedback cycles should be checked carefully and, where necessary, softened.
I also encourage you to strengthen the geomorphological context of the paper. The reviewers identify relevant literature on vegetation controls in badlands, gully complexes and rapidly eroding landscapes, including work from New Zealand and other humid, high-erosion settings. Engaging with this literature would help clarify how this study contributes to existing understanding of vegetation–erosion interactions and landscape evolution. In particular, please consider whether the results have implications for sediment production, sediment delivery to the outlet, or downstream sediment flux. Even if these implications cannot be quantified fully, they would help strengthen the relevance of the study for the journal.
Please provide a detailed point-by-point response to all reviewer comments when submitting the revised manuscript. Given the open review format, I would encourage you to make the response as transparent and specific as possible, indicating where changes have been made and explaining clearly where you have chosen a different interpretation from that suggested by a reviewer.
I would be pleased to consider a substantially revised version of the manuscript. However, the revision will need to go beyond typographical correction and figure polishing. It should directly address the conceptual, methodological and interpretive issues raised in the reviews, and should make clear whether the manuscript is primarily documenting vegetation dynamics or whether it can robustly support conclusions about vegetation–erosion feedbacks in badland landscape evolution.
Kind regards, Dan
Citation: https://doi.org/10.5194/egusphere-2026-834-EC1 -
EC2: 'Comment on egusphere-2026-834', Daniel Parsons, 03 Jun 2026
Look forward to your response. I hope you find the reviews and my summary useful.
Thanks
DP
Citation: https://doi.org/10.5194/egusphere-2026-834-EC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 279 | 149 | 20 | 448 | 24 | 35 |
- HTML: 279
- PDF: 149
- XML: 20
- Total: 448
- BibTeX: 24
- EndNote: 35
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This paper attempts to quantify the topographic and erosion-rate controls on vegetation changes in a badland drainage basin. Based on the data, feedback mechanisms involving vegetation changes and geomorphic processes are proposed.
Concerns:
1) Suitability for the journal: the analysis is heavily focused on vegetation dynamics. Four of the five figures showing results (i.e., Figs. 2-5 out of Figs. 2-6) are focused on different remote-sensing methods for quantifying vegetation change. As such, the paper would be more suitable for Biogeosciences or a similar journal that focuses on vegetation. A lidar difference map is discussed but isn’t presented and so cannot be assessed or reviewed.
2) The paper discusses how differencing of lidar surveys conducted between 2015 and 2021 were adjusted for the “best alignment” and then combined with measured sediment losses to obtain erosion rates. The paper states that erosion rates obtained in this way assume uniform erosion across the study area (line 174). That contradicts Figure 6, which shows that slope and erosion rates are positively correlated. The authors should explain more convincingly how their "best alignment" procedure and and the assumptions inherent in that procedure may or may not be consistent with other results they present. Also, the paper presents contradictory statements about who filtered vegetation from the lidar. On lines 164-165 it states that the producer did this filtering while line 170 states that the authors did this.
3) Mismatch in time scale between vegetation changes and measured erosion rates: Vegetation changes occurred almost entirely as a transition of grassland to forest between 1994 and 2012 (Fig. 2c). The erosion rates were measured later (2015-2021). So, there is no overlap in time between the vegetation changes and the erosion rates.
4) Comments regarding Figure 6: C1: The results presented in the figure lead the authors to conclude that slope (line 311) and drainage area (line 314) are the main controls on spatial trends in vegetation change. Figure 6d (there are no letters labeling the panels of this figure, so I am just guessing that this is supposed to be panel d) shows that the transition probability to extinction approaches 1 and colonization approaches 0 as drainage area approaches 10^5-10^6 m^2. Figure 5 does not seem to show this. Areas of extinction tend to occur throughout the study area – they are not clustered in valley bottoms with drainage areas 10^5-10^6 m^2 as far as I can tell. If they are, please visualize this clearly. C2: Please help me understand how the transition probability for colonization of areas with 25% slope could be 0.5 (meaning non-colonized areas with 25% slope have a 50% probability of becoming colonized, and, as Fig. 6a shows, with all areas of slopes between 10% and 60% having transition probabilities of at least 0.3) when colonization is, in fact, extremely rare (i.e., the dark green pixels in Fig. 5 seem to make up less than 1% of the study area). This is another example in which the results presented in Figure 6 just do not seem to align with the trends (or lack thereof) in Figure 5
5) Variability in the data about mean trends: please present box plots (min, max, median, and quartiles) for each bin, rather than just the average) in Fig. 6 so that we can see how much variability is present.
6) Lines 289-290 suggest that increased veg cover and/or greening is happening worldwide. In North America, the primary temporal trend is increasing tree mortality due to an increase in wildfire size/frequency/intensity and pest infestations, both driven in part by greenhouse warming (https://doi.org/10.1016/j.foreco.2009.09.001).
7) Why focus on greening? I can understand how changes in vegetation type (Fig. 5) may be influenced by, and in turn influence, erosion rates. But I don’t see how greening (which comprises a large portion of this study) is related to erosion rates (which are primarily related to vegetation through vegetation type or percent bare ground) because greening can occur without any change in root cohesion or other aspect of vegetation that tends to be more directly related to erosion. This concern dovetails with point 1 on how this is primarily a vegetation study with limited connection to geomorphology or landscape evolution.