the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 17 Mar 2025
Evaluating biogeophysical sensitivities to idealized deforestation in CMIP6 models using observational constraints
Abstract. Forests are an important component in the framework of nature-based solutions for mitigating climate change. However, there are still uncertainties about the biogeophysical effects of forest cover changes affecting heat and water fluxes as captured by Earth System Models (ESMs) simulations and observations. In this study, we investigate the differences in the surface temperature response to idealized, complete deforestation and the temperature sensitivity to percentage change in forest cover in ESMs and observations. In this comparison, the separation between local (at the place of deforestation) and non-local (nearby or distant locations) effects is crucial as observations capture only the former. Here, we propose a modified methodology to separate local and non-local effects in climate models suitable for simulations with linear rate of deforestation. The local sensitivity of a climate variable per unit deforested area is represented by the slope of the linear regression, where tree cover is an explanatory variable. The non-local effect is defined as the difference between the overall change in the respective climate variable and the local effect. Our analysis of eleven ESMs of the Coupled Model Intercomparison Project Phase 6 (CMIP6) that participated in the idealized global deforestation experiment deforest-glob, reveals a coherent local temperature response among climate models characterized by warming in the tropics and cooling in the northern higher latitudes. The temperature response however varies in magnitude, space and time with ESMs showing distinctive seasonal and spatial patterns. A closer look at the albedo response to deforestation across norther latitudes shows an overestimation in the ESMs in comparison to observations that translates via an emergent constraint into an overestimation of the overall simulated cooling effect. The overestimation of the local albedo sensitivity cannot be explained solely by the higher percentage of snow cover in ESMs. In terms of local latent heat flux sensitivity, the ESMs ensemble mean is overestimated for the boreal region, but it is in good agreement with the observational constraint in the temperate forests and the tropics. However, the inter-model spread and the internal model variation in these regions are considerable. ESMs having higher local albedo and latent heat flux sensitivities than the current observational constraints can still exhibit a realistic temperature response due to compensatory effects between the two sensitivities. Non-local effects contribute to consistent cooling throughout the globe, which persists also during the summer when the influence of the overestimated albedo sensitivity over snow is weaker. Having a deeper understanding of how local and non-local biogeophysical effects are represented in ESMs can give us insights into the net climate impact of deforestation and help us improve next generation ESMs.
Competing interests: Vivek Arora and Roland Séférian are members of the editorial board of the Earth System Dynamics journal.
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 preprint. The responsibility to include appropriate place names lies with the authors.- Preprint
(10626 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 04 May 2025)
-
RC1: 'Comment on egusphere-2025-979', Anonymous Referee #1, 24 Apr 2025
reply
This study proposes a new methodology to separate local and non-local biogeophysical effects of deforestation in model experiments. This new methodology consists of calculating a linear regression between moving 5x5 model grids in the control simulation and the same 5x5 model grids in the deforestation simulation using tree cover, latitude, longitude, and elevation as predictor variables. The slope of the tree cover predictor variable is deemed to represent the local sensitivity of the dependent variable of interest (i.e. surface temperature, albedo, or latent heat flux). Using this new methodology on CMIP6 deforestation model experiments, this study derives emerging relationships between local and total surface temperature changes, albedo changes and latent heat flux changes. It also compares local surface temperature changes, albedo changes, and latent heat flux changes between CMIP6 model experiments and observational data from FLUXNET and MODIS to constrain model results. This study finds that the local surface temperature derived from model experiments tends to be lower than the temperature derived from observations. The albedo response is also stronger in most models compared to observations, particularly in areas with high snow cover.
This is an interesting study that provides an alternative way to calculate local effects from model experiments when it is not possible to use the checkerboard method. This provides us with a deeper understanding of the differences in local effects between models and observations. However, I believe the local effects calculated from this new methodology are contaminated by non-local effects. The extent of which is hard to determine. The magnitude of contamination would depend on the range of forest cover change among pixels included in the regression window, with a greater range likely leading to less contamination and a smaller range likely leading to more contamination. As an extreme example, if the forest cover loss is around 50% for all pixels in the window, and all pixels have a very similar surface temperature change of 1 degree between the control and the deforestation simulation, the resulting slope of the tree cover variable will be around 2. In this case, the bulk of the change in temperature would be attributed to local effects. However, this change in temperature of 1 degree encompasses both local effects and non-local effects coming from pixels outside the window. The contamination of local effects by non-local effects (which are generally a cooling) is likely one of the factors why the local temperature effects calculated by the models are significantly below observations in many cases. This should at the very least be discussed more thoroughly in the limitations and mentioned as an additional plausible cause of differences between observations and models when discussing results.
Additionally, although the manuscript is generally well-written, it could benefit from some reorganization to improve the flow, some clarification of concepts in certain areas, and consistency in the terms used to avoid confusion, as indicated in the specific comments. Moreover, the section on non-local effects is not directly relevant to the research objectives (comparing the local effects across a range of ESMs and against observations) and could be moved to the supplementary material. The effectiveness of the figures could also be improved, mainly by making the differences between observations and model experiments stand out more, as per the specific comments.
Specific comments:
14: Typo, northern latitudes
15: Unclear what emergent constraint means without having read the paper, suggest to remove “via an emergent constraint” in the abstract, or elaborate further on the concept of emerging constraints
55-57: Large-scale deforestation also triggers changes in ocean circulation which would impact non-local effects (Portmann et al., 2022)
75: Typo, cloud cover instead of forest cover
91-92: Define what observationally based emergent constraints are. Needs more background. See my comments for line 215-230. It is hard to understand what it means and if it is any different than observational constraints or simply emergent constraints?
164-166: Add how the change in forest fraction is represented when using the FLUXNET data to make it consistent with the ESM experiment and the MODIS-based data explanations
185: Add comma after the word pair
186: You need pixels with greater than 10% forest cover change to calculate the local effects, but pixels with zero or minimal forest cover change can help strip out the non-local effects (as changes in those would be attributed mainly to non-local effects). You could therefore calculate the linear regression for all pixels, but only in windows where there is a sufficient number of pixels with greater than 10% forest cover change. It would be interesting to see how this changes the magnitude of the local effects.
192: Consider adding that you then calculate the mean value of the local sensitivity over the period and region of interest for completeness.
203: Consider starting the new subsection here “2.5 Comparing local sensitivities between observations and ESMs – emergent constraints”
215-230: Most of this section should go in the Introduction as it provides background info on the emergent constraint approach and how it can be used, and discusses goals of the study.
223-230: Hard to understand, would benefit from a clearer description of what you are trying to achieve: are you saying that if we know the local effects from observations, and the emerging relationship between local effects and total effects from model experiments, we could use that relationship to constrain the total effects based on the observed local effects? Regarding albedo, if we know the albedo change from observation, and the emerging relationship between albedo and total temperature from model experiments, we could use that relationship to constrain the total temperature change based on the observed albedo change?
227: I don’t think we can say that local temperature sensitivities are comparable between models and observations in boreal and temperate regions based on your data (although they appear to be comparable in the tropics). Most models show local temperature effects significantly lower than observations, which, as discussed earlier, could be due to the incorporation of the non-local cooling effects in the local effects calculated from models.
229: Local surface temperature or total surface temperature?
249-253: Sentence starting with “It has to be noted…” to sentence ending by Fig A4: could go in the limitations section of the Discussion.
278: Clarify local or total surface temperature.
280: The emerging linear relationship between local temperature and total temperature is a key result, I think the figure showing this result (currently A1) should be incorporated in the main text and not the supplementary information.
288: Add “defined by the standard deviation of the MODIS observations” for clarity.
304: Section 3.4 should really be part of section 3.3 and not a separate section.
316: This section on the non-local effects doesn’t add much information that is directly related to the objectives of the study and can deter from the core message. I would move Figure 7 and corresponding information to the supplementary material.
337: Observational constraints, observational emergent constraints, or emergent constraints? Is there a difference between the three?
337: This section would also benefit from a discussion on how we could use the emergent linear relationship between local temperature and total temperature changes to constrain total temperature changes from ESMs based on observed local temperature changes and the limitations/pitfalls of doing so.
360: Observational constraints, observational emergent constraints, or emergent constraints?
378: Observational constraints, observational emergent constraints, or emergent constraints?
396: This section could be removed, since it does not add much new information compared to other studies, and if the results regarding the non-local effects are moved to supplementary material as per prior comment.
412-496: This is a major limitation as discussed in my general comments. Indication that non-local effects contaminate local effects should be mentioned along with its repercussions, including overestimation of the local cooling effect.
Fig 3: Replace Bright et al. (2017) by FLUXNET in the legend to make it clearer where this data come from when looking only at the figure. It is already indicated in the text that the FLUXNET data comes from Bright et al. (2017). Y axis: add “local” temperature change to avoid confusion.
Fig 4: Use a different color for the MODIS data point (observation) to make it stand out more compared to model data. Y axis: add “local” temperature change to avoid confusion. Replace observational constraints with “observational emergent constraints” in the caption to make it consistent across all figures.
Fig 5: Same comments as Fig 4. Consider using the same color for the MODIS data point in plots c-d-e and for the MODIS lines in plots a-b, and same color for the CMIP6 data points in plots c-d-e and for the CMIP6 lines in plots a-b for consistency.
Fig 6: Same comments as Fig 4.
Fig. A5 and A6: Same comments as Fig 4, as applicable.
References
Portmann, R., Beyerle, U., Davin, E., Fischer, E. M., De Hertog, S., & Schemm, S. (2022). Global forestation and deforestation affect remote climate via adjusted atmosphere and ocean circulation. Nature Communications, 13(1), 5569. https://doi.org/10.1038/s41467-022-33279-9
Citation: https://doi.org/10.5194/egusphere-2025-979-RC1 -
RC2: 'Comment on egusphere-2025-979', Anonymous Referee #2, 24 Apr 2025
reply
Review of “Evaluating biogeophysical sensitivities to idealized deforestation in CMIP6 models using observational constraints” by Mileva et al.
The work of Mileva et al. explores and quantifies relationships between changes to key surface energy balance variables and forest cover change in several independent observational datasets as well as in ten distinct ESMs. From this, the authors identify robust statistical relationships between surface albedo change and local temperature change, and between local temperature change and total temperature change “emerging” from the ESMs. These findings are important when viewed in the context of similar relationships seen in the observational records since this provides insights critical to the goal of model development/improvement. The paper is well-written with logical organization, and the work has been carried out thoroughly and carefully. The methods are sufficiently described and documented, and the study’s limitations are made clear.
I only have a few comments (or rather, suggestions) for improving the methodological clarity and depth of the discussion. The first is related to the methods. I think it should be made clearer up-front about the limitations of the comparisons being made between the ESM results and the observation-based datasets. “Apples-to-apples” comparisons are never really being made. Perhaps adding a table resembling the following is most efficient at clearly communicating important differences among the datasets being compared in the study.
Timestamp
Conditional
Aggregation
Physics of ∆LST
ESMs
3-hourly
None
True monthly/seasonal/annual means
Coupled
FLUXNET-based
Monthly
None
True monthly/seasonal/annual means
Uncoupled
MODIS
Local noon (albedo); 13:30 (LST)
Clear skies
Snapshot monthly/seasonal/annual clear-sky means
Coupled
Regarding the discussion, I wonder if the paper would be strengthened by expanding on the application benefits of the emergent constraint found in this study as it relates to model development. For example, how, specifically, would it contribute to a “deeper understanding of how local and non-local biogeophysical effects are represented in ESMs” as is stated in the Abstract? The authors demonstrate important offsetting biases in a few of the models surrounding the mechanisms governing the local surface temperature response, and thus there is a risk that such an emergent constraint might not lead to any meaningful model improvement.
Additionally, I feel it could also be beneficial to add some discussion surrounding the robustness of the finding of the linear relationship between the local and the total surface temperature change due to deforestation, and whether the authors think this relationship would hold in the case of more realistic (or real world) patterns and scales of deforestation.
Detailed comments
P5, L142-143: Since this sentence is lumped into the paragraph describing the remote sensing datasets, clarify that “all” here includes the FLUXNET-based dataset presented in the previous paragraph.
P6, L147: “using an averaging filter”. What is this (or how is it different than averaging)?
Section 3.1 and Figure 2: The authors acknowledge the clear-sky bias limitations of observations based on satellite remote sensing (P3, L73), so I don’t understand why the models’ all-sky responses are being compared to the MODIS-based response? Wouldn’t it make more sense to compare to the FLUXNET-based dataset here which the authors acknowledge (P3, L75) overcomes this important limitation?
Figure 2: I don’t see any stippling on any of the figure panels.
P17, L388: I’m confused by this sentence, as these two models seem to be among those with the weakest latent heat flux sensitivity?
Figure A3: Consider adding the observations here as in Fig. 3 so it becomes easier for the reader to more easily benchmark the individual models.
Citation: https://doi.org/10.5194/egusphere-2025-979-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 99 | 24 | 5 | 128 | 4 | 4 |
- HTML: 99
- PDF: 24
- XML: 5
- Total: 128
- BibTeX: 4
- EndNote: 4
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1