| 11 Sep 2026
Response of vegetation hydroclimatic stress indicators to global warming and deforestation in the Amazon rainforest
Abstract. The Amazon rainforest (ARF) plays a key role in regulating the global climate, sustaining exceptional biodiversity, and supporting the cultures of indigenous peoples and local communities. The ARF is also considered a tipping element, as deforestation and global warming may impose sufficient stress to trigger a transition to a tropical savanna state. However, the climate and vegetation responses to deforestation and global warming, as well as associated tipping thresholds, remain incompletely understood. Here, we examine the response of four hydroclimatic indicators to deforestation and warming and discuss their implications for forest stability. We performed four idealized ARF deforestation experiments with the Earth System Model MPI-ESM-HR at 2 °C of global warming. These scenarios were compared against two intact-forest scenarios for pre-industrial climate and for 2 °C warming. Our results show that deforestation of 25 % of the forest area can lead to 70 % being under significant hydroclimatic stress. The area under stress increases to 89 % when combining 25 % deforestation and 2 °C warming, exceeding the deforestation-only effects of 50 % deforestation (82 %). Mean Annual Precipitation (MAP) and Dry Season Length (DSL) show the strongest changes, but behave differently: MAP decreases strongly in and near newly deforested areas, while DSL increases across large forest regions downstream of the deforested areas. Although the rainforest remains stable in the model, the enhanced hydroclimatic stress would likely increase tree mortality in the real world if vegetation cannot adapt. Our results hence emphasize the risk of forest dieback at forcing levels that can be reached within the next few decades.
This is an interesting study assessing the stability of Amazon forests from the point of view of hydrologic variables. In general it is well presented and organized. However, conclusions seem too bold in light of the presented results and the methods to obtain such results. I am particularly concerned with the direct connection made between changes in hydrological variables (mean annual precipitation, dry season length, maximum cumulative water deficit, and topsoil (10cm) water availability) and forest stability. Especially because this study does not explore or touch any physiological process of the trees and how they mechanistically respond to reduced water availability. And the literature that uses observational or experimental evidence on that relation between lower water availability and forest stability does not support the assumptions made here. There is surely an effect, but one should consider that forest can, in principle, recover from those droughts. See https://doi.org/10.1038/s43247-026-03887-0 and https://doi.org/10.1038/s41559-025-02702-x .
The fact that convection of is not explicit resolved in the model seems to make a difference here (see https://doi.org/10.1029/2024GL110503). This study reminds us that we do have observational evidence on what happens with precipitation when we clear-cut small parts of the forests. But, for obvious reasons, there is no observational evidence on what happens with precipitation when large chunks (>50%) of the forest are cut. While it is reasonable to consider the vertical flux of moisture from the soil to the atmosphere is reduced when the forest is cut, the reduction of surface roughness could in principle favor the horizontal flux humidity from the Atlantic Ocean. I miss that sort of ponderation in this study, exploring, explaining and evaluating more deeply the mechanisms behind the results. It is well known that the Atlantic Ocean conditions play a key role in the Amazon climate, but there is no explanation on the the ocean is treated in the modeling exercise.
And, important, I am skeptical that comparing the projected modeled changes in the variability of the hydrological variables against the baseline modeled results is the best approach, given the bias of the model. How do the changes in the hydrological variables compare with the OBSERVED variability of such variables? Maybe that would be a more sound analysis. I suggest the authors clarify the what they consider to be “significantly disturbed” in this study, especially in the case of “at least moderate standardized anomalies of at least one of the other three [variables]”, which seems a weak condition to consider a tropical forest as significantly disturbed (and prone to a dieback, as suggested in the discussion).
Lastly, how is CO2 fertilization considered and what role does it play here?
Please see detailed suggestions below.
AbstractP
Why does the forest stand despite the changes in MAP and DSL? Please address thither in the abstract.
Intro
P2L48 - please explain better the study by Esquivel-Muelbert et al.
P2L49 - both the studies of Anderegg et al. and Marengo et al. do not touch the tipping issue. Please bring a bit more impartiality here. The attribution of such climatic change in south Amazon is somewhat inconclusive as of yet. Part of it seems to be caused by deforestation but part not.
P2L55-59 - This is all too biased. Please include other evidence. The works by Nobre et al. 2016 and Lovejoy and Nobre are based on the same evidence - low complexity vegetation modeling.
L59-61 - also the dampening role of two processes not mentioned here: forest hyper (tree) diversity and CO2 fertilization.
Methods
P3L80 This is in contradiction to what is said in line 45. The study cited in L45 revealed that trees can acquire water from depths as much as 10m or 20m in the soil. So how can one evaluate trees hydrological status looking at the first 10cm of soil water availability?
L80 - MCWD stands for maximum cumulative water deficit and not mean climatological water deficit!!!
L99 - ok, but it does not say much about hydraulic status of trees.
L101 - I respectfully disagree. There is no consideration of hydraulic traits which the literature point out as more relevant for tree mortality (e.g. https://doi.org/10.1038/nature15539). Please also consider the newly published https://doi.org/10.1038/s43247-026-03887-0 and https://doi.org/10.1038/s41559-025-02702-x
L113 - please be quantitatively specific on after how many standard deviations you consider the forest is unstable.
L111-115 - When showing the results it would be crucial to differentiate, in a visually explicit way, how much of the forest is considered disturbed due to reason “a” (this is already done in Fig6), due to “b” and especially due to “c”.
L122 - how do you think that having connection explicitly resolved would change the results? See https://doi.org/10.1029/2024GL110503
L166-177 - what about CO2 fertilization, is it on or off in these experiments?Please include that info in the paragraph.
Fig4 - interesting to see how variables get considerable worsened under the 100% deforestation scenario compared to the 75%. I suppose that is because of the differences embedded in the methods of the 100% scenario, like grassland occupying deforested areas.
Fig 5 - I am doubtful about the value of this figure, i.e. analyzing the impact of different deforestation levels by comparing the changes between consecutive scenarios. See, if in one of the scenarios the change was to small, let’s say, 1%, and in the other it was 2%, it is still a small absolute change, but in this figure it would appear as a 100% increase. So it masks the real impact of the scenarios in my opinion. I suggest the authors avoid that type of comparison.
L474 - The fact that forest remained largely stable throughout the course of the simulations is of utmost importance and should be given more prominence along the manuscript, including in the abstract (and maybe in the title). Maybe what authors are calling here “hydroclimatic stress” is not really a stress for plants, as indicated by some observational/experimental evidence (e.g. the studies suggested above and other).
L479 - ok, this explains the issue from the point of view of observations, but it does not explore the model processes that makes the ESM maintain forests in areas with MAP <500mm. Is it a limitation imposed by the PFTs that are considered in the model? If possible please clarify.
L535 - this is contradictory with the results presented here (see L474). Forest remained majorly stable in your results, so how can one conclude that these hydroclimatic stresses [sic] will drive a dieback?
L537 - this conclusion does not seem aligned with the results: there was no considerable dieback in the simulation with +2ºC and 25% deforestation!
L556 - correct reading of Papastefanou et al. study would be that the three major droughts in the 21st century hit, on average, about 50% of the forest area.
L561 - vegetation is showing more resilience than generally addressed in modeling studies. Authors should discuss that as well. Bear in mind that the Amazon has 15,000 different species of trees. Such a diversity could play a dampening role in terms of shifting species dominance without converting the forest into a savanna.
I missed a section of caveats of this modeling exercise…
See also https://doi.org/10.1029/2025WR040039