the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 06 Jul 2026
Spatial patterns in second-order impacts of human activity on climate, land and water
Abstract. In order to assess the full systemic impact of anthropogenic pressures on climate and land, it is crucial to account for second-order impacts mediated by bio-geophysical processes, which typically display a high spatial heterogeneity. In this study, we systematically compare different clustering approaches to capture uni- and multivariate spatial patterns in second-order impacts of human activity on climate, land, and water. In a first step, we estimate effect sizes based on simulations from a spatial global vegetation model. In a second step, we approach the question of suitable spatial clustering. Following a top-down approach first, we map the global pattern of second-order impacts on common natural partitions of the Earth, like climate- or vegetation-zones. Cluster validity indices reveal a close alignment between the second-order impacts of land use change on climate and a biogeographic classification. In contrast, the second-order impacts of climate- and land use change on surface water runoff are best captured by the Köppen-Geiger climate zones. Following a bottom-up approach, we employ multivariate spatially constrained clustering to derive an integrative global partition. Several patches of tropical rainforest on the Indomalayan islands as well as large areas of warm grasslands in Australia are identified as high-impact clusters. The results of this study should be considered illustrative as they are based on only one dynamical vegetation model. Nevertheless, they emphasize the local nature of second-order impacts and elucidate both the potential and risks of spatial aggregation.
Competing interests: At least one of the (co-)authors serves as editor for the special issue to which this paper belongs.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.- Preprint
(23968 KB) - Metadata XML
-
Supplement
(197 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2026-3452', Anonymous Referee #1, 30 Jul 2026
-
AC1: 'Reply on RC1', Hannah Zoller, 15 Sep 2026
Dear Reviewer,
we would like to thank you for thoroughly reading our manuscript. We were particularly pleased by your assessment of the question we address as important and timely. Following your suggestions, we will switch to the term “indirect impacts” and revise the title, abstract, and introduction to avoid any ambiguity regarding our use of the term. A shorter and more concise Results and Discussion Section will further contribute to the clarity and readability and of the manuscript.
Please find our point-by-point responses below.
Major Comments
- The study does not explicitly represent land-atmosphere interactions. My primary concern is that the manuscript repeatedly attributes its findings to the interplay of land-atmosphere interactions (e.g., Lines 20, 24, 161, 228, and 278), whereas the underlying simulations are performed using the offline land surface model LPJmL. An offline land surface model can only respond to prescribed atmospheric forcing and cannot feed land-surface changes back to the atmosphere. Therefore, the two-way land-atmosphere interactions implied throughout the manuscript are not explicitly represented. Consequently, attributing the simulated patterns to the land-atmosphere interactions is misleading. More importantly, this limitation is not merely a matter of wording, but may affect the validity of the manuscript’s central conclusions. For example, land-cover change modifies surface albedo, evapotranspiration, and surface energy fluxes, which in turn alter atmospheric circulation, precipitation, vegetation growth, runoff, and carbon storage through atmospheric teleconnections. Previous studies have even shown that these atmospheric feedbacks can generate remote impacts comparable to, or larger than, the local responses in some regions. Such processes are entirely absent from the current framework. Therefore, I do not believe that this issue can be adequately addressed by the brief discussion of study limitation in Lines 359-360. Instead, the authors should explicitly acknowledge that their analysis only quantifies land-surface responses under prescribed atmospheric conditions and discuss how the omission of land-atmosphere feedbacks may influence their conclusions.
In alignment with Lade et al. 2021, our results are based on a one-way coupling of LPJmL. This offline approach allows us to neatly isolate particular effects without the need to disentangle overlaying feedback effects. We tried to be very clear about this fact both in the Introduction and in the Methods section, visually supported by the conceptual diagram (Figure 2).
Nevertheless, we agree that in case of the other occurrences of the term “interplay/interaction” in l.161, l.228, Appendix L, a change of wording (e.g., to “effects”) will increase clarity. (Note that in l.20 and l.24, we refer to the interplay between different Earth system processes in a more general sense/in literature, in l.278, we refer to the interplay of effects.)
Furthermore, we will extend the corresponding paragraph in the Discussion, highlighting potential changes that the inclusion of feedback mechanisms could entail. - The definition and scope of “second-order impacts” require clarification. The terminology “second-order impact" is introduced very early in the manuscript (Line 20), yet its definition remains ambiguous throughout the paper. First, I am not convinced that this relatively uncommon terminology is preferable to the more widely used term indirect impacts, unless the authors intend to distinguish a more specific concept. Second, the manuscript never clearly defines what processes are included within second-order impact and, equally importantly, which processes are intentionally excluded. For example, based on the modelling framework, atmospheric feedbacks discussed in Comment 1 are clearly beyond the scope of the study. Such boundaries should be explicitly stated when the term is first introduced to avoid confusion.
We chose the term “second-order impacts” to distinguish from higher-order impacts created by feedbacks. However, we are comfortable to change to the more widely used term “indirect impacts”. To avoid ambiguity, we will add the following definition to the first paragraph of the Introduction: “[…] leading to so-called indirect impacts. For example, while we consider the decrease in natural vegetation as the direct impact of deforestation, the effects of this decrease on carbon storage and water availability are considered indirect impacts. Hence, direct impacts involve human pressure and one Earth system process, while indirect impacts involve human pressure and two Earth system processes. In particular, we refer to impacts as one-way cause-and effect relations, not including feedback mechanisms.“ - Justification for the selected indicators. The authors use carbon storage density, surface runoff, and natural vegetation cover as representative indicators of climate, water, and land impacts, respectively. However, the rationale for selecting these three variables is not sufficiently explained. For example, from a land-surface perspective, variables such as evapotranspiration or soil moisture may more directly characterize hydrological impacts than runoff, while GPP could arguably be a more representative measure of ecosystem functioning than vegetation cover. The manuscript would benefit from explaining the criteria used to select these indicators and discussing whether the major conclusions remain robust when alternative variables are considered.
The selected indicators align closely with variables representing human pressures (e.g., clearing land, extracting water), which becomes particularly important in case the results get implemented in an impact framework like the Earth System Impact Metric (ESI, Lade et al. 2021). We will clarify this rationale in the manuscript. - The Title and Abstract do not adequately reflect the main findings. The main contribution of this study is not only the investigation of the spatial patterns in second-order impacts, but also the comparison between top-down and bottom-up clustering approaches, with the latter potentially demonstrating superior performance. However, the Abstract devotes relatively little attention to these central findings and instead remains rather general. A similar issue exists in the title, which does not reflect the methodological comparison that forms the core contribution of the study. As a result, even after reading the title and Abstract multiple times, it remains difficult to immediately grasp the primary objectives and conclusions of the manuscript. I recommend revising both sections to better emphasize the novelty and principal findings.
We would like to point out that while the bottom-up approach is intuitively exceeding the top-down approach in clustering performance, it is inferior in explainability and communicability. Hence, while highlighting the limitations of the more established top-down approach, the results of our manuscript suggest to chose a clustering methodology depending on the context and scope of a study.
Nevertheless, we agree that the comparison of the two clustering approaches provides an essential part of our study and hence, we will highlight this aspect more directly in the title and in the abstract. We suggest “Spatial patterns in indirect effects of human activity on climate, land, and water: comparing top-down and bottom-up clustering approaches“ as a new title. - Consistency among different impact estimation methods. The manuscript adopts different methodologies to estimate the strength of second-order impacts for different components. For example, the effects on natural vegetation cover are quantified using one methodology, whereas climate-related impacts are estimated using another. The authors should clarify whether these different methods produce quantitatively comparable estimates. If the results are intended to be directly compared, the rationale for using different estimation methods should be justified. Alternatively, the authors should discuss the potential biases introduced by methodological inconsistency.
Our study is based on the modelling framework used in Lade et al. 2021, whose rationale is based on the fact that atmospheric CO2 is driving the model, while land use change can be switched on and off in LPJmL. Aligning with their methodology allowed for a direct comparison their vegetation-based clustering approach.
Concerning the limitations of this approach, we mention that the different methodologies might be partly responsible for certain differences in the spatial layout of the two effect types (l.163-164).
As correctly pointed out, the two estimation methods do not create quantitatively comparable results. Accordingly, we do not compare the absolute size of climate-induced effects and land use change-induced effects. We will clarify this fact in the manuscript.
Minor Comments
- Line 52: I do not consider a spatial resolution of 0.5° to constitute “high spatial resolution”. This description should be moderated. We will remove that term.
- Line 71: Please explain why only the last 30 years were selected for the analysis. The differential quotient approach, in contrast to the linear regression approach, requires states of variables rather than time courses. Following Lade et al. 2021, we use averages over the last 30 years of the simulation as representatives of the recent state under different assumptions regarding human-driven land use change. The choice of 30 years aligns with the established use of 30-year climatological averages and reduces interannual variability.
- Both the Results and Discussion are considerably longer than necessary. I recommend reducing repetitive descriptions, removing some regional examples, and emphasizing the overarching conclusions. In addition, the Results should explain why the observed patterns emerge rather than only describing them. For example, in Lines 195-203, why does natural vegetation exert a negative influence on runoff primarily in Asian tropical rainforests? Does this relationship not exist in tropical rainforests elsewhere? We will follow your suggestion, aiming for a shorter and more concise Results and Discussion Section. This procedure will be facilitated by the inclusion of a table on main mechanisms underlying the positive and negative effect sizes (see Minor Comment 6).
Concerning the negative effect of land use change on water runoff: as mentioned in l.197-198, four out of the five clusters exhibiting the strongest negative impact are rainforests (Asian, African, North American, Oceanian), hence, this phenomenon is not restricted to the Asian rainforest. In order to avoid such misunderstandings, we will follow your suggestion and focus on the main findings rather than individual examples. - Since the bottom-up clustering consistently outperforms the top-down approach in most evaluation metrics, whether the extensive discussion of the top-down results is necessary. The manuscript would benefit from placing greater emphasis on the more successful clustering framework. As mentioned above, we do not consider the bottom-up clustering approach as superior in all aspects. It outperforms the top-down clustering with respect to typical clustering validity indices, but the resulting clusters are less intuitive and hence, the results are more difficult to communicate. Besides, most existing studies are based on a top-down approach and our study aims to build awareness for the caveats that this approach entails, independently of the chosen natural partition. We will emphasize these aspects in the Conclusions.
- Under the bottom-up approach, different impact pathways (e.g., climate -> vegetation versus vegetation -> water) would likely produce substantially different spatial clusters. If so, how comparable are the resulting cluster systems? This issue deserves further discussion. Yes, based on the maps presented in Figure 4, we expect a univariate clustering to substantially differ depending on the impact pathway defining the feature space. This expectation is further supported by the large differences in clustering performance of top-down partitions depending on the impact pathway (Figure 6). In the manuscript, we use multivariate clustering as a tool to find integrative partitions under these exact circumstances. And still, as Figure K2 demonstrates, there is no “one-size-fits-all”-clustering. We are happy to highlight this fact in the Conclusions.
- Besides Figure 8, I strongly recommend adding a summary figure or table explaining the dominant mechanisms responsible for the identified high positive and high negative impact clusters. This would substantially improve readability and reduce the need for lengthy textual descriptions. We will include a little table summarizing the main mechanisms leading to positive/negative effect size for the four impact pathways. By placing it at the beginning of Section 3, it will allow for more conciseness throughout the whole Results and Discussion section. Nevertheless, we would like to point out that the underlying mechanisms can depend on the location being considered and hence, further explanations in the text cannot be fully avoided.
- Line 393: Please explain why a decrease in natural vegetation cover leads to increased evapotranspiration in these regions. This result appears counterintuitive and requires further explanation. Removing vegetation can lead to a more exposed soil and thereby to an increase in soil evaporation. On the other hand, crop transpiration might be higher than transpiration on formerly largely barren soil, like e.g., in parts of central Australia. We will add these explanations to the manuscript.
- The study does not explicitly represent land-atmosphere interactions. My primary concern is that the manuscript repeatedly attributes its findings to the interplay of land-atmosphere interactions (e.g., Lines 20, 24, 161, 228, and 278), whereas the underlying simulations are performed using the offline land surface model LPJmL. An offline land surface model can only respond to prescribed atmospheric forcing and cannot feed land-surface changes back to the atmosphere. Therefore, the two-way land-atmosphere interactions implied throughout the manuscript are not explicitly represented. Consequently, attributing the simulated patterns to the land-atmosphere interactions is misleading. More importantly, this limitation is not merely a matter of wording, but may affect the validity of the manuscript’s central conclusions. For example, land-cover change modifies surface albedo, evapotranspiration, and surface energy fluxes, which in turn alter atmospheric circulation, precipitation, vegetation growth, runoff, and carbon storage through atmospheric teleconnections. Previous studies have even shown that these atmospheric feedbacks can generate remote impacts comparable to, or larger than, the local responses in some regions. Such processes are entirely absent from the current framework. Therefore, I do not believe that this issue can be adequately addressed by the brief discussion of study limitation in Lines 359-360. Instead, the authors should explicitly acknowledge that their analysis only quantifies land-surface responses under prescribed atmospheric conditions and discuss how the omission of land-atmosphere feedbacks may influence their conclusions.
-
AC1: 'Reply on RC1', Hannah Zoller, 15 Sep 2026
-
RC2: 'Comment on egusphere-2026-3452', Fabian Stenzel, 05 Aug 2026
Dear authors,
I appreciate the chance to read your present study ("Spatial patterns in second-order impacts of human activity on climate, land and water"), in which you derive 4 grid-cell based interaction strengths between climate and land-use change as drivers, and affecting runoff, carbon storage, and vegetation structure, based on existing simulations from the DGVM LPJmL. To cluster the gridded results for larger units you then apply two classes of clustering approaches. The first (“top-down”) is based on predefined spatial units such as continents or biomes, and the second (“bottom-up”) is based on analyzing the interaction strength maps with a clustering algorithm. You find that the “bottom-up” approach generally gives much better results and then use it, finding 19 clusters with strong effects in at least 2 of the 4 interaction dimensions, and 3 “high impact clusters” where even 3 out of 4 dimensions show strong effects.
The paper is situated in the analysis of regionally resolved interactions between different Earth system processes, and thus especially relevant for the planetary boundaries and tipping points frameworks, as well as more broadly studying Earth system interactions.
The paper generally reads well, and I support the publication in ESD after reviewing some aspects of it. My major comments refer to the details of the selection and description of the LPJmL simulations and their discussion, as well as the balance between different parts of the manuscript, that partially have been also raised by an anonymous reviewer (R1) before me.
1. Please give more details for the LPJmL simulations you used and discuss their limitations:
- Which specific LPJmL version did you use?
- Which land-use input did you use?
- Please mention, how you calculate the carbon storage density. I suppose it should be the sum of vegetation, soil, and litter carbon.
- Please note that using monthly climate inputs (that internally have to be distributed over the days in a month) leads to ambiguity, that can be prevented by using (more up-to-date) daily data e.g. from ISIMIP (GSWP3-W5E5) or TRENDY (CRU-JRA). This is especially relevant, since you compare the differences between runs. I am sure you have checked, that the seed with which monthly values are distributed to days is the same between runs, otherwise you have a climate effect, where you don’t expect one.
- LPJmL4.X.Y as described by Schaphoff et al. 2018 has 11 PFTs. Your text in (p.6 line 99) could be mistaken so that LPJmL4 would have 12. I believe you use barren land as land cover type 12, but strictly speaking this is not a PFT. Please also add how you calculate it, because you need separate grid cell fractions for CFTs for this (barren = 1 – sum(PFT fpc) – sum(CFT fpc).
- With LPJmLv4 instead of v5, you are missing N limitation, thus overestimating the CO2 fertilization effect.
- The sensitivity analysis using 6 more (GCM based) climate input datasets could be integrated better. Currently in the Methods (p. 5 lines 91-92) it is not clear that this is only for sensitivity tests and that the results won’t go into the main analysis.
2. Details of the clustering approaches.
What exactly are the inputs to the “bottom-up” approach? As far as I understood it, the final effect maps are also going in, so naturally much better clusters are derived than for “top-down”. Is that a fair comparison between the two cluster algorithms?
Like this it appears that the result is clear before hand, and only serves as an argument to focus on the bottom-up clustering results. This is not a problem per se, but could be stated more clearly. I would additionally suggest to shorten the analysis of the “top-down” part, if it is then discarded anyways.
At the same time, the policy implications of using dynamic and thus hard to name and locate clusters could be discussed in more detail.
3. Definition of second-order impacts
Like reviewer R1, I also believe that your current definition of “second-order impacts” based on direct and indirect effects is not sufficiently clear. The example you give rather confuses me, as I would say deforestation IS the vegetation cover change, thus zero-order, and everything that follows next would be direct or first-order impact (like runoff/ET changes).
Maybe the notion of local vs. remote effects can help you to improve the definition? I admit that for remote effects besides river discharge, LPJmL offline is not the right tool.
4. Justification of using LPJmL offline
Unlike reviewer R1, I believe that it is still worthwhile doing the analysis with LPJmL. Since the authors are using historical land-use and weather patterns, even if not directly coupled, the climate inputs includes land-atmosphere interactions to the historical land-use changes.
This is however only true for the factual wLUC simulation, and not the counterfactual noLUC simulation that uses the same climate inputs. This limitation needs to be stated clearly.
5. Definition of impacts on climate change
You chose carbon storage density changes as a metric for climate change. Since grid-cells in LPJmL are not the same size, carbon storage density changes in gC/m2 does not impact atmospheric CO2 concentrations in the same way everywhere. Secondly, the translation from carbon storage decline to atmospheric CO2 increase is certainly not linear. These limitations needs to stated more clearly.
I hope my comments (please see some additional inline comments below) can help you to further improve your study.
Best regards,
Fabian Stenzel
Potential conflict of interest: I am affiliated with SRC as a guest researcher and work in a common project together with study co-author Ingo Fetzer.
-------------------------
Additional comments:
p.2 line 36: If you want to compare with different hierarchical clustering approach, aimed at reducing computational costs please see Ehret et al. 2020 (https://doi.org/10.5194/hess-24-4389-2020) or Priesner et al. 2026 (https://doi.org/10.1140/epjs/s11734-026-02526-1) who additionally compare different distance metrics.
p.4 lines 78-80: The way you define it, you are rather talking about small changes in the land use input, rather than “noise” (numerical non-deterministic behavior?) in the model.
p.8 lines 161-162: And certainly also the different methodology, by regressing values for each year instead of a 30 year average, right?
p.10 lines 191-192: Can you really draw this conclusion from the Figure (and what does “best explain” mean – “highest RS”, or per no. of clusters so “most top-left”)? It is hard to see (you also highlighted “non climate” partitions), maybe plot only 3 colors for different symbols, red for climate and blue for PFT, and orange for biome to confirm this.
p.13 lines 266-267: I don’t find this very surprising, as each PFTs parameters describe a direct relationship with climate, especially in the temperature limits and water/radiation requirements. It would be interesting to single out the climate from the CO2 effect, by increasing CO2 regularly but keeping the climate as 1901-1930 and vice versa.
Fig. 6: Some combinations are hard to distinguish, e.g. same symbol and similar color as in “climates P” and “continents biomes”. In panel B the names are offset from the boxes (which I believe) they belong to. There is no clear method for when to draw boxes, right? It appears a bit arbitrary. In general everything is very small, even when zooming in.
Fig. A8: Labels appear erroneous. “Boreal broadleaved evergreen” does not exist in LPJmL. “Temperate needleleaved evergreen” is missing.
Fig. D1A: The dashed boxes are off.
Citation: https://doi.org/10.5194/egusphere-2026-3452-RC2 -
AC2: 'Reply on RC2', Hannah Zoller, 15 Sep 2026
Dear Reviewer,
we would like to thank you for thoroughly reading our manuscript. We were pleased to learn that you are generally supporting the publication of the manuscript in ESD. Following your suggestions for improvement, we will increase the level of details regarding the LPJmL simulations. Moreover, we will refine the Results and Discussion Section to improve the balance between different sections.
Please find our point-by-point responses below.
- Please give more details for the LPJmL simulations you used and discuss their limitations:
Which specific LPJmL version did you use? Our results are based on the modelling output by Lade et al. 2021, who used LPJmL 4.0.002. We will add the version to the manuscript.
Which land-use input did you use? The land-use change implementation used by Lade et al. is based on historical CFT allocation. We will add further details on the land use change implementation in Section 2.1 Estimating the strength of indirect impacts.
Please mention, how you calculate the carbon storage density. I suppose it should be the sum of vegetation, soil, and litter carbon. We use solely vegetation carbon as carbon storage density, in alignment with Lade et al. 2021. We will add this detail to the manuscript.
Please note that using monthly climate inputs (that internally have to be distributed over the days in a month) leads to ambiguity, that can be prevented by using (more up-to-date) daily data e.g., from ISIMIP (GSWP3-W5E5) or TRENDY (CRU-JRA). This is especially relevant, since you compare the differences between runs. I am sure you have checked, that the seed with which monthly values are distributed to days is the same between runs, otherwise you have a climate effect, where you don’t expect one. Thank you very much for this remark. Our simulations are based on the modelling output by Lade et al. 2021, who used a fixed seed across runs to distribute monthly values to days.
LPJmL4.X.Y as described by Schaphoff et al. 2018 has 11 PFTs. Your text in (p.6 line 99) could be mistaken so that LPJmL4 would have 12. I believe you use barren land as land cover type 12, but strictly speaking this is not a PFT. Please also add how you calculate it, because you need separate grid cell fractions for CFTs for this (barren = 1 – sum(PFT fpc) – sum(CFT fpc). Grid cells were classified as barren land whenever the other eleven plant functional types together cover less than 50% of the total grid cell area. We will add this explanation to the manuscript and further clarify that LPJmL only distinguishes eleven PFTs. As described on p.6, l.100 the Earth partition you are referring to is based on the LPJmL run without human land use, so there is no cropland. In particular, barren = 1-sum(PFT fpc).
With LPJmLv4 instead of v5, you are missing N limitation, thus overestimating the CO2 fertilization effect. Thank you very much for pointing out this potential caveat. We will mention it in Section 3.5 Limitations of the Study.
The sensitivity analysis using 6 more (GCM based) climate input datasets could be integrated better. Currently in the Methods (p. 5 lines 91-92) it is not clear that this is only for sensitivity tests and that the results won’t go into the main analysis. We will clarify the fact that the runs driven by the six additional climate models serve as sensitivity analysis only.
- Details of the clustering approaches.
What exactly are the inputs to the “bottom-up” approach? As far as I understood it, the final effect maps are also going in, so naturally much better clusters are derived than for “top-down”. Is that a fair comparison between the two cluster algorithms? Like this it appears that the result is clear before hand, and only serves as an argument to focus on the bottom-up clustering results. This is not a problem per se, but could be stated more clearly. I would additionally suggest to shorten the analysis of the “top-down” part, if it is then discarded anyways.
At the same time, the policy implications of using dynamic and thus hard to name and locate clusters could be discussed in more detail.
Yes, it is indeed intuitively clear that the bottom-up clustering approach yields better results with respect to cluster validity indices than the top-down approach. At the same time, as you are mentioning, the resulting clusters are harder to interpret and thereby communicate, which is particularly important if a corresponding study is addressing a non-academic community. Hence, the main message of our manuscript is not the superiority of the bottom-up clustering approach, but a comparison of the two approaches from different perspectives, building awareness for the caveats of both of them. Since this point is apparently not clear in the current version of the manuscript (compare Reviewer 1, Minor Comment 4), we will clarify it both in the Introduction and in the Conclusions.- Definition of second-order impacts
Like reviewer R1, I also believe that your current definition of “second-order impacts” based on direct and indirect effects is not sufficiently clear. The example you give rather confuses me, as I would say deforestation IS the vegetation cover change, thus zero-order, and everything that follows next would be direct or first-order impact (like runoff/ET changes).
Maybe the notion of local vs. remote effects can help you to improve the definition? I admit that for remote effects besides river discharge, LPJmL offline is not the right tool.
In order to avoid confusion, we will follow the suggestion of Reviewer 1 and use the more established term “indirect impact” instead of the term “second-order impact” (compare Reviewer 1, Major Comment 2). In particular, we will add the following definition to the first paragraph of the Introduction: “[…] leading to so-called indirect impacts. For example, while we consider the decrease in natural vegetation as a direct impact of deforestation, the effects of this decrease on carbon storage and water availability are considered indirect impacts. Hence, direct impacts involve human pressure and one Earth system processes, while indirect impacts involve human pressure and two Earth system processes. In particular, we refer to impacts as one-way cause-and effect relations, not including feedback mechanisms.“- Justification of using LPJmL offline
Unlike reviewer R1, I believe that it is still worthwhile doing the analysis with LPJmL. Since the authors are using historical land-use and weather patterns, even if not directly coupled, the climate inputs include land-atmosphere interactions to the historical land-use changes.
This is however only true for the factual wLUC simulation, and not the counterfactual noLUC simulation that uses the same climate inputs. This limitation needs to be stated clearly.
As mentioned in the responses to Reviewer 1, Lade et al. deliberately chose an offline approach to be able to isolate particular effects without the need to disentangle overlaying feedback effects (compare Major Comment 1). This scope is not conflicting with the imprint of feedback effects in the historical climate data since the response variable in the estimation of LUC→ CC is the modelled carbon storage density, and not atmospheric CO2.
We will still point the reader to the imprint of feedback effects in Section 2.1 Estimating the strength of indirect impacts.- Definition of impacts on climate change
You chose carbon storage density changes as a metric for climate change. Since grid-cells in LPJmL are not the same size, carbon storage density changes in gC/m2 does not impact atmospheric CO2 concentrations in the same way everywhere. Secondly, the translation from carbon storage decline to atmospheric CO2 increase is certainly not linear. These limitations need to be stated more clearly.
We would like to point out that carbon storage density is only used in the estimation of the effect LUC → CC. The change in natural vegetation cover is not scaled by grid cell area as well and multiplying both and numerator and denominator in the differential quotient would result in the same effect size. Nevertheless, we agree that area is a crucial factor when it comes to the absolute direct and indirect impact of human disturbances and corresponding applications of our results should take it into account.
The fact that carbon storage decline translates non-linearly to atmospheric CO2 increase will be stated in Section 2.1 Estimation the strength of indirect impacts.Additional comments:
p.2 line 36: If you want to compare with different hierarchical clustering approach, aimed at reducing computational costs please see Ehret et al. 2020 (https://doi.org/10.5194/hess-24-4389-2020) or Priesner et al. 2026 (https://doi.org/10.1140/epjs/s11734-026-02526-1) who additionally compare different distance metrics. Thank you very much pointing us towards these two articles. Especially the Priesner et al. article fits very well into our argumentation and we will refer to it in the Introduction.
p.4 lines 78-80: The way you define it, you are rather talking about small changes in the land use input, rather than “noise” (numerical non-deterministic behavior?) in the model. Yes, we agree and will update the terminology accordingly.
p.8 lines 161-162: And certainly also the different methodology, by regressing values for each year instead of a 30-year average, right? Yes, indeed. We will reformulate the corresponding sentence in l.164 to clarify that it refers to the phenomenon of the patchy pattern as well.
p.10 lines 191-192: Can you really draw this conclusion from the Figure (and what does “best explain” mean – “highest RS”, or per no. of clusters so “most top-left”)? It is hard to see (you also highlighted “non climate” partitions), maybe plot only 3 colors for different symbols, red for climate and blue for PFT, and orange for biome to confirm this. By “best explained/represented” we mean that the corresponding partitions stand out with respect to their RS index compared to partitions of similar size (number of clusters). As being explained in the caption of Figure 6, we marked all partitions that fulfil this criterium. While there is a clear dominance of e.g., climate-based partitions in the case CC → RO, the dominating classification criterion is not as clear in other cases. Since the proposed coloring scheme would be difficult due to partitions combining classification criteria, we highlighted the dominating classification criterion in bold letters. We will clarify both the procedure and interpretation in the text.
p.13 lines 266-267: I don’t find this very surprising, as each PFTs parameters describe a direct relationship with climate, especially in the temperature limits and water/radiation requirements. It would be interesting to single out the climate from the CO2 effect, by increasing CO2 regularly but keeping the climate as 1901-1930 and vice versa. We assume you are referring to lines 263-265. While we agree that the effect of climate change on vegetation cover is a very direct one, we do not see how this explains the fact that the bottom-up RS index perform best for the effect LUC → RO, while the bottom-up BR index performs best for the effect CC → LUC. We agree that isolating the effect of changing CO2 values could yield interesting insights, as studies like Piao et al. 2007 demonstrated.
Fig. 6: Some combinations are hard to distinguish, e.g. same symbol and similar color as in “climates P” and “continents biomes”. In panel B the names are offset from the boxes (which I believe) they belong to. There is no clear method for when to draw boxes, right? It appears a bit arbitrary. In general everything is very small, even when zooming. Yes, in Figure 6B, the names are offset from the boxes in order to increase readability. We will try to find an allocation which is visually clearer. The boxes were drawn to highlight partitions, which stand out in RS compared to partitions of similar size. We will further clarify this aspect in the manuscript. Overall, we will replot the figure in order to avoid same symbol/similar colour- combinations and check if increasing symbol size etc. increases readability of the plot.
Fig. A8: Labels appear erroneous. “Boreal broadleaved evergreen” does not exist in LPJmL. “Temperate needleleaved evergreen” is missing. Thank you very much for pointing out this mistake. “Boreal broadleaved evergreen” should be “boreal broadleaved summergreen” and “temperate needleleaved“ should be “temperate needleleaved evergreen”. We will correct this mistake both in the supplementary code and in the figure.
Fig. D1A: The dashed boxes are off. We will correct this figure.
-
AC2: 'Reply on RC2', Hannah Zoller, 15 Sep 2026
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 147 | 88 | 19 | 254 | 16 | 11 | 10 |
- HTML: 147
- PDF: 88
- XML: 19
- Total: 254
- Supplement: 16
- BibTeX: 11
- EndNote: 10
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Zoller et al. investigate an important and timely question: how strong and spatially heterogeneous the indirect impacts (so-called second-order impact) of human activities on climate, land, and water are, and how these impacts can be effectively classified into spatial clusters. The authors first quantify these impacts using the LPJmL land surface model and then compare two clustering approaches: a conventional top-down approach based on predefined geographical or ecological units (e.g., continents and plant functional types) and a bottom-up approach based on spatially constrained clustering. Through multiple evaluation metrics, the authors conclude that the bottom-up clustering generally outperforms top-down approach and identify several global hotspots of high-impact clusters.
This is an ambitious and comprehensive study that addresses a challenging problem. However, the manuscript is also technically complex, and after carefully evaluating the methodology and interpretation, I have several fundamental concerns regarding the reliability of the conclusions. I hope the following comments will help improve the manuscript.
Major Comments
Minor Comments