Response to Reviewer 1 main comment about separation of local and non-local effects, and potential for underestimating the local effect of deforestation on near-surface climate:

We thank the reviewer for his/her comments and appreciate he/she finds our study interesting. We’d like to point out that the moving window regression approach, which our study builds on and develops further, has been applied and evaluated by several previous studies. A very recent study (Jaeger et al., 2025, 10.1029/2024JD042698) proposed an alternative third approach to moving window regression and the chessboard approach and confirms that both are generally valid. However, there has been some evidence that the regression approach underestimates the local effects to some extent, which, in our original submission, we acknowledge in the Limitations section (Section 5). Previous studies using a similar approach based on linear regression indeed pointed this out, when comparing this method to the factorial experiment approach (Lejeune et al., 2018) and the look-up table approach (Winckler et al., 2019 (Fig. S8)). In order to quantify the discrepancies between the different methods for separating local and non-local effects directly for our simulations, we have performed an additional analysis (see attachment, Fig. R1-R3) comparing the temperature sensitivities to deforestation in MPI-ESM as calculated in our study based on the deforest-glob simulation (via linear regression) and in the study of Winckler et al. (2019), where deforestation is performed in every one or two out of four grid cells (chessboard pattern). While the deforestation patterns in these simulations are different, the magnitude of deforestation is similar and therefore it is possible to approximately compare the two methods – the one based on linear regression and the chessboard pattern method (Fig. R1). This analysis showed that the differences in temperature sensitivities between the two methods vary in geographic space and, importantly, show no systematic bias towards underestimation of the local effect (Fig. R2).

It should be kept in mind that this comparison was done only with one model (due to data availability, this comparison was possible only with the MPI model) and that the chessboard pattern method comes with its own set of uncertainties (Winckler et al., 2019). Importantly, while we cannot exclude contamination of the local effects by non-local cooling, these results suggest no systematic underestimation of the local effect as suggested by previous studies (Lejeune et al., 2018; Winckler et al. 2019). Ultimately, modellers should view our results as a first evaluation and are encouraged to perform offline simulations to further test their model’s skill in replicating observationally based temperature (plus albedo- and latent heat-) sensitivities to deforestation.

Further, in this comparison (Fig. R2), we did not apply the scaling of the temperature sensitivities to 100% deforestation (see Section 2.4 in original manuscript, lines 203-210), in order to keep the magnitude of deforestation similar to the one in the experiments of Winckler et al. (2019). Applying the scaling does lead to amplification of the differences in temperature sensitivities (resulting from the two different methods for extracting the local effects), which are more notable over the boreal region in Eurasia and the central parts of the Amazon and Congo basin (Fig. R3). However, since the same scaling approach has been applied to the observational datasets, a comparison between models and observations is still meaningful and the validity of the emergent constraints is not affected.

To respond to the specific example given by the Reviewer, where the forest cover loss for all pixels within the moving window is 50%, we would like to point out that, indeed, this pattern of deforestation has prevented us from applying the classical linear regression approach by Lejeune et al. (2018). To avoid the issue, in our proposed methodology we have taken (instead of change) the values of temperature (T) and tree cover (F), which are different in the deforest-glob and piControl simulations. And, then we have trained a linear regression on the pairs of T and F from both simulations.

To reflect on this Reviewer’s main comment and our additional analyses, we intend to add the following text in the revised manuscript in the first paragraph under Limitations (Section 5): “An additional analysis comparing the temperature sensitivities to deforestation based on the linear regression method applied in this study and the chessboard pattern deforestation experiments of Winckler et al. (2019) did not reveal evidence of systematic underestimation of the local effects (see Supplement). However, the scaling to 100% could lead to overestimation of the local effects (compared to other methods) possibly because the relationship between surface temperature and tree cover change is not strictly linear as albedo and evapotranspiration effects vary as a function of initial tree cover and across biomes (Bonan, 2008; Alibakhshi et al., 2019).”.

Regarding the section on non-local effects, we have elaborated in the specific comments on the importance of considering the non-local effects. In this section, we show a way forward how to include in the comparison models that overestimate the albedo-induced cooling.

Response to specific comments from Reviewer 1:

Lines 14: to be fixed in the revised manuscript

15: suggested improvement: “[…] via an emergent constraint (i.e. resulting from the linear relationships within the model ensemble)”

55-57: suggested improvement: “But large-scale deforestation in these experiments also triggers strong changes in advection of heat and moisture, as well as in atmospheric and ocean circulation, which influence regions that have not undergone deforestation (Winckler et al., 2017; Portmann et al., 2022).”

75: to be removed to avoid confusion: “[…] temperature changes.”

91-92: to be removed to avoid confusion: “The objective is to provide emergent constraints…”; clarification of ”emergent constraint” added (see 15)

164-166: suggested improvement: “The change in forest fraction is 100% and is considered only for pixels, where the respective vegetation cover types are actually present and/or could potentially occur as defined by MODIS-derived land cover maps for 2005 and Köppen-Geiger climate zone maps for the 20^th century.”

185: to be fixed in the revised manuscript

186: It is not expected that including pixels with less than 10% forest cover change will alter the results substantially, as in most models the forest cover change is much larger than 10% (see attachment, Fig. R4). We have provided an additional analysis comparing the surface temperature sensitivity with and without the 10% threshold for the MPI model (see attachment, Fig. 5). The MPI model was selected based on the availability of more pixels with less than 10% forest cover change in comparison to other models. The results with using the threshold are more robust, especially for the boreal forest, as evidenced by the number of statistically significant pixels.

192: The information is already (partially) provided in 211-214. Suggested improvement: “In order to evaluate the consistency of seasonal surface temperature responses between the ESMs and the observation-based estimates, we extracted the mean values of local surface temperature responses to deforestation also at monthly timescale for broad latitudinal regions: boreal (from 50° N to 90° N), temperate (from 23° N to 50° N) and tropical (from 23° S to 23° N).”

203: See comment to 192

215-230: Explanation of the emergent constraint concept to be moved to Introduction.

223-230: suggested additions/changes: “By knowing the local effects from observations, and the emerging relationship between local and total effects from model experiments, we could use that relationship to constrain the total effects based on the observed local effects. Thus, by using the local sensitivities derived from in-situ and satellite-based data, we are potentially able to provide emergent constraints for surface temperature, albedo and latent heat flux. In addition, we use the relationship between local surface temperature and albedo, and between local surface temperature and latent heat flux, to explore further the emergent constraints.”

227: The possibility to compare local effects in models with observations has been confirmed by a number of studies (e.g., Bright et al., 2019; Winckler at al., 2019a; Chen&Dirmeyer, 2020). The issue with the possible overestimation of local effects is discussed in the first paragraph. To avoid a misunderstanding about the term “comparable”, we have changed the sentence from “[…] as local surface temperature sensitivities from models and observations are comparable” to “[…] as local surface temperature sensitivities from models and observations can be brought into a meaningful relationship”.

229: Here we refer to the total effects.

249-253: To be moved to Limitations in the revised manuscript.

278: Here we refer to local surface temperature.

280: Fig. A1 to be moved to the main text in the revised manuscript.

288: To be changed to “defined by the standard deviation of the MODIS observations"

304: Section 3.4 to be merged to Section 3.3

316: While in this section we report also on non-local effects, which are not the primary focus of the study, we also evaluate the local effects during summer. This evaluation is particularly important because by considering only summer months we are able to exclude the effects of the Siberian cold bias still observed in many CMIP6 models (Portal et al., 2023). It is also relevant in light of the discussion on the contamination of local effects by non-local effects. In the summer, the non-local cooling is weaker (due to the absence of the snow albedo effect), thus giving possibly clearer view on the local effects.

337: We refer to observational constraints as a broader term, encompassing emergent constraints. When non-local or total effects are referred to, we use specifically the term (observational) emergent constraints, as these effects can be evaluated only by considering the emergent linear relationship between the different models. “Emergent constraints” and “observational emergent constraints” are synonymous. Clarification to be added under Section 2.5.

337: suggested addition at line 350: “Because of the emergent constraint relationship between the local and total surface temperature effects, this overestimation is valid also for the overall response to deforestation, thus showing that most models exhibit too strong cooling in comparison to observations (Fig. A1). For the tropics, however, approximately half of the ESMs show realistic total surface temperature response, as defined by the emergent constraint based on MODIS data, with fewer models being within the margins defined by the FLUXNET-based dataset (Fig. A1).”

360: see above

378: see above

396: see comment to 316

412-496: see general comments

Fig. 3: to be updated in revised manuscript

Fig 4: to be updated in revised manuscript

Fig 5: to be updated in revised manuscript

Fig 6: to be updated in revised manuscript

Fig. A5 and A6: to be updated in revised manuscript

References

Alibakhshi, S., Naimi, B., Hovi, A., Crowther, T. W., & Rautiainen, M. (2020). Quantitative analysis of the links between forest structure and land surface albedo on a global scale. Remote Sensing of Environment, 246, 111854.

Bonan, G. B. (2008). Forests and climate change: forcings, feedbacks, and the climate benefits of forests. Science, 320(5882), 1444-1449.

Bright, R. M., Davin, E., O’Halloran, T., Pongratz, J., Zhao, K., & Cescatti, A. (2017). Local temperature response to land cover and management change driven by non-radiative processes. Nature Climate Change, 7(4), 296-302.

Jäger, F., Schwaab, J., Bukenberger, M., De Hertog, S. J., & Seneviratne, S. I. (2025). Spectral decomposition and signal separation of climate responses to land cover changes. Journal of Geophysical Research: Atmospheres, 130(5), e2024JD042698.

Lejeune, Q., Davin, E. L., Gudmundsson, L., Winckler, J., & Seneviratne, S. I. (2018). Historical deforestation locally increased the intensity of hot days in northern mid-latitudes. Nature Climate Change, 8(5), 386-390.

Winckler, J., Lejeune, Q., Reick, C. H., & Pongratz, J. (2019). Nonlocal effects dominate the global mean surface temperature response to the biogeophysical effects of deforestation. Geophysical Research Letters, 46(2), 745-755.

Portal, A., d'Andrea, F., Davini, P., Hamouda, M. E., & Pasquero, C. (2023). Atmospheric response to cold wintertime Tibetan Plateau conditions over eastern Asia in climate models. Weather and Climate Dynamics, 4(3), 809-822.

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

Response to Reviewer 1 main comment about separation of local and non-local effects, and potential for underestimating the local effect of deforestation on near-surface climate:

We thank the reviewer for his/her comments and appreciate he/she finds our study interesting. We’d like to point out that the moving window regression approach, which our study builds on and develops further, has been applied and evaluated by several previous studies. A very recent study (Jaeger et al., 2025, 10.1029/2024JD042698) proposed an alternative third approach to moving window regression and the chessboard approach and confirms that both are generally valid. However, there has been some evidence that the regression approach underestimates the local effects to some extent, which, in our original submission, we acknowledge in the Limitations section (Section 5). Previous studies using a similar approach based on linear regression indeed pointed this out, when comparing this method to the factorial experiment approach (Lejeune et al., 2018) and the look-up table approach (Winckler et al., 2019 (Fig. S8)). In order to quantify the discrepancies between the different methods for separating local and non-local effects directly for our simulations, we have performed an additional analysis (see attachment, Fig. R1-R3) comparing the temperature sensitivities to deforestation in MPI-ESM as calculated in our study based on the deforest-glob simulation (via linear regression) and in the study of Winckler et al. (2019), where deforestation is performed in every one or two out of four grid cells (chessboard pattern). While the deforestation patterns in these simulations are different, the magnitude of deforestation is similar and therefore it is possible to approximately compare the two methods – the one based on linear regression and the chessboard pattern method (Fig. R1). This analysis showed that the differences in temperature sensitivities between the two methods vary in geographic space and, importantly, show no systematic bias towards underestimation of the local effect (Fig. R2).

It should be kept in mind that this comparison was done only with one model (due to data availability, this comparison was possible only with the MPI model) and that the chessboard pattern method comes with its own set of uncertainties (Winckler et al., 2019). Importantly, while we cannot exclude contamination of the local effects by non-local cooling, these results suggest no systematic underestimation of the local effect as suggested by previous studies (Lejeune et al., 2018; Winckler et al. 2019). Ultimately, modellers should view our results as a first evaluation and are encouraged to perform offline simulations to further test their model’s skill in replicating observationally based temperature (plus albedo- and latent heat-) sensitivities to deforestation.

Further, in this comparison (Fig. R2), we did not apply the scaling of the temperature sensitivities to 100% deforestation (see Section 2.4 in original manuscript, lines 203-210), in order to keep the magnitude of deforestation similar to the one in the experiments of Winckler et al. (2019). Applying the scaling does lead to amplification of the differences in temperature sensitivities (resulting from the two different methods for extracting the local effects), which are more notable over the boreal region in Eurasia and the central parts of the Amazon and Congo basin (Fig. R3). However, since the same scaling approach has been applied to the observational datasets, a comparison between models and observations is still meaningful and the validity of the emergent constraints is not affected.

To respond to the specific example given by the Reviewer, where the forest cover loss for all pixels within the moving window is 50%, we would like to point out that, indeed, this pattern of deforestation has prevented us from applying the classical linear regression approach by Lejeune et al. (2018). To avoid the issue, in our proposed methodology we have taken (instead of change) the values of temperature (T) and tree cover (F), which are different in the deforest-glob and piControl simulations. And, then we have trained a linear regression on the pairs of T and F from both simulations.

To reflect on this Reviewer’s main comment and our additional analyses, we intend to add the following text in the revised manuscript in the first paragraph under Limitations (Section 5): “An additional analysis comparing the temperature sensitivities to deforestation based on the linear regression method applied in this study and the chessboard pattern deforestation experiments of Winckler et al. (2019) did not reveal evidence of systematic underestimation of the local effects (see Supplement). However, the scaling to 100% could lead to overestimation of the local effects (compared to other methods) possibly because the relationship between surface temperature and tree cover change is not strictly linear as albedo and evapotranspiration effects vary as a function of initial tree cover and across biomes (Bonan, 2008; Alibakhshi et al., 2019).”.

Regarding the section on non-local effects, we have elaborated in the specific comments on the importance of considering the non-local effects. In this section, we show a way forward how to include in the comparison models that overestimate the albedo-induced cooling.

Response to specific comments from Reviewer 1:

Lines 14: to be fixed in the revised manuscript

15: suggested improvement: “[…] via an emergent constraint (i.e. resulting from the linear relationships within the model ensemble)”

55-57: suggested improvement: “But large-scale deforestation in these experiments also triggers strong changes in advection of heat and moisture, as well as in atmospheric and ocean circulation, which influence regions that have not undergone deforestation (Winckler et al., 2017; Portmann et al., 2022).”

75: to be removed to avoid confusion: “[…] temperature changes.”

91-92: to be removed to avoid confusion: “The objective is to provide emergent constraints…”; clarification of ”emergent constraint” added (see 15)

164-166: suggested improvement: “The change in forest fraction is 100% and is considered only for pixels, where the respective vegetation cover types are actually present and/or could potentially occur as defined by MODIS-derived land cover maps for 2005 and Köppen-Geiger climate zone maps for the 20^th century.”

185: to be fixed in the revised manuscript

186: It is not expected that including pixels with less than 10% forest cover change will alter the results substantially, as in most models the forest cover change is much larger than 10% (see attachment, Fig. R4). We have provided an additional analysis comparing the surface temperature sensitivity with and without the 10% threshold for the MPI model (see attachment, Fig. 5). The MPI model was selected based on the availability of more pixels with less than 10% forest cover change in comparison to other models. The results with using the threshold are more robust, especially for the boreal forest, as evidenced by the number of statistically significant pixels.

192: The information is already (partially) provided in 211-214. Suggested improvement: “In order to evaluate the consistency of seasonal surface temperature responses between the ESMs and the observation-based estimates, we extracted the mean values of local surface temperature responses to deforestation also at monthly timescale for broad latitudinal regions: boreal (from 50° N to 90° N), temperate (from 23° N to 50° N) and tropical (from 23° S to 23° N).”

203: See comment to 192

215-230: Explanation of the emergent constraint concept to be moved to Introduction.

223-230: suggested additions/changes: “By knowing the local effects from observations, and the emerging relationship between local and total effects from model experiments, we could use that relationship to constrain the total effects based on the observed local effects. Thus, by using the local sensitivities derived from in-situ and satellite-based data, we are potentially able to provide emergent constraints for surface temperature, albedo and latent heat flux. In addition, we use the relationship between local surface temperature and albedo, and between local surface temperature and latent heat flux, to explore further the emergent constraints.”

227: The possibility to compare local effects in models with observations has been confirmed by a number of studies (e.g., Bright et al., 2019; Winckler at al., 2019a; Chen&Dirmeyer, 2020). The issue with the possible overestimation of local effects is discussed in the first paragraph. To avoid a misunderstanding about the term “comparable”, we have changed the sentence from “[…] as local surface temperature sensitivities from models and observations are comparable” to “[…] as local surface temperature sensitivities from models and observations can be brought into a meaningful relationship”.

229: Here we refer to the total effects.

249-253: To be moved to Limitations in the revised manuscript.

278: Here we refer to local surface temperature.

280: Fig. A1 to be moved to the main text in the revised manuscript.

288: To be changed to “defined by the standard deviation of the MODIS observations"

304: Section 3.4 to be merged to Section 3.3

316: While in this section we report also on non-local effects, which are not the primary focus of the study, we also evaluate the local effects during summer. This evaluation is particularly important because by considering only summer months we are able to exclude the effects of the Siberian cold bias still observed in many CMIP6 models (Portal et al., 2023). It is also relevant in light of the discussion on the contamination of local effects by non-local effects. In the summer, the non-local cooling is weaker (due to the absence of the snow albedo effect), thus giving possibly clearer view on the local effects.

337: We refer to observational constraints as a broader term, encompassing emergent constraints. When non-local or total effects are referred to, we use specifically the term (observational) emergent constraints, as these effects can be evaluated only by considering the emergent linear relationship between the different models. “Emergent constraints” and “observational emergent constraints” are synonymous. Clarification to be added under Section 2.5.

337: suggested addition at line 350: “Because of the emergent constraint relationship between the local and total surface temperature effects, this overestimation is valid also for the overall response to deforestation, thus showing that most models exhibit too strong cooling in comparison to observations (Fig. A1). For the tropics, however, approximately half of the ESMs show realistic total surface temperature response, as defined by the emergent constraint based on MODIS data, with fewer models being within the margins defined by the FLUXNET-based dataset (Fig. A1).”

360: see above

378: see above

396: see comment to 316

412-496: see general comments

Fig. 3: to be updated in revised manuscript

Fig 4: to be updated in revised manuscript

Fig 5: to be updated in revised manuscript

Fig 6: to be updated in revised manuscript

Fig. A5 and A6: to be updated in revised manuscript

References

Alibakhshi, S., Naimi, B., Hovi, A., Crowther, T. W., & Rautiainen, M. (2020). Quantitative analysis of the links between forest structure and land surface albedo on a global scale. Remote Sensing of Environment, 246, 111854.

Bonan, G. B. (2008). Forests and climate change: forcings, feedbacks, and the climate benefits of forests. Science, 320(5882), 1444-1449.

Bright, R. M., Davin, E., O’Halloran, T., Pongratz, J., Zhao, K., & Cescatti, A. (2017). Local temperature response to land cover and management change driven by non-radiative processes. Nature Climate Change, 7(4), 296-302.

Jäger, F., Schwaab, J., Bukenberger, M., De Hertog, S. J., & Seneviratne, S. I. (2025). Spectral decomposition and signal separation of climate responses to land cover changes. Journal of Geophysical Research: Atmospheres, 130(5), e2024JD042698.

Lejeune, Q., Davin, E. L., Gudmundsson, L., Winckler, J., & Seneviratne, S. I. (2018). Historical deforestation locally increased the intensity of hot days in northern mid-latitudes. Nature Climate Change, 8(5), 386-390.

Winckler, J., Lejeune, Q., Reick, C. H., & Pongratz, J. (2019). Nonlocal effects dominate the global mean surface temperature response to the biogeophysical effects of deforestation. Geophysical Research Letters, 46(2), 745-755.

Portal, A., d'Andrea, F., Davini, P., Hamouda, M. E., & Pasquero, C. (2023). Atmospheric response to cold wintertime Tibetan Plateau conditions over eastern Asia in climate models. Weather and Climate Dynamics, 4(3), 809-822.

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

We thank the reviewer for his/her comments and appreciate he/she finds our study interesting.

Response to Reviewer 2 first main comment about improving the methodological clarity:

In order to highlight the differences between the various datasets, Table A1 will be updated as follow:

and the following table will be added:

Table XX. Overview of climate variables datasets. “True” values are those that account for differences in the sun zenith angle.

Additionally, the following text will be added to Section 2.1: “All models couple land, atmosphere and ocean in terms of momentum, matter and energy (Lawrance et al., 2016).”

Addition to section 2.3: “The differences between the various datasets in terms of temporal resolution and continuity, consideration of cloud coverage, aggregation methods and consideration of land/ocean/atmosphere interactions are summarized in Table XX.”

Response to Reviewer 2 second main comment about improving the discussion on emergent constraints:

Many processes relevant for the climate cannot be explicitly represented by mathematical expressions, which require the parameterization of climate models. The development of parameterizations and the final parameter adjustment (or ‘model tuning’) rely heavily on observations and are two areas where the availability of emergent constraints could lead to new developments and improvements. Suggested addition to the revised manuscript: “As models are usually evaluated based on how well they reproduce a subset of past observations, their ability to predict future climate is more uncertain (Flato et al., 2013). By using emergent constraints, modelling centres are potentially able to improve the parameterization and tuning of ESMs, so that they are better adapted to simulate future climate without being overfitted to historical data.”

Response to Reviewer 2 third main comment about separation of local and non-local effects:

See detailed responses to Reviewer 1 main comment.

Further, the magnitude of deforestation in the deforest-glob simulation amounts to 20 million km² and  is therefore similar to historical deforestation from 800 to 2015 (~22 million km²) (Boysen et al., 2020). The overestimation of albedo sensitivities over snow and the difficulties of models to represent turbulent heat fluxes, as found in our study, have been confirmed also in the study of Luo et al. (2023), which is based on historical land use and land cover changes and therefore represents more realistic patterns of deforestation. Taking into account also the similar local surface temperature response that can be observed in our study and the study of Luo et al. (2023) and the similar magnitude of deforestation, one can postulate that the surface temperature emergent constraint would hold true also under more realistic conditions. Testing the emergent constraint in different experiments and multi-model ensembles is an important next step towards confirming its robustness.

Response to specific comments of Reviewer 2:

P5, L142-143: This sentence refers only to the MODIS products. Suggested change: “All MODIS-based datasets were reprojected to the WGS84 coordinate system and resampled to 0.05°.”

P6, L147: To avoid confusion, “filter” is removed: “[…] the pixels at the original resolution of 30 m were resampled to 0.05° using averaging”

Section 3.1 and Figure 2: The FLUXNET-based dataset only provides changes in surface temperature but not in albedo and latent heat flux, which are available from MODIS observations and are essential for explaining the temperature response. Therefore, it was important to include the MODIS datasets from the beginning of the analysis. Here, we also wanted to present an updated version of local surface temperature effects as presented in the study of Alkama&Cescatti (2016), which is also based on MODIS data, by using the newest generation of these products (v061), which have undergone various calibration improvements. Detailed maps of the FLUXNET-based temperature responses are available in Bright et al. (2017).

Figure 2: All figures are to be provided at 330 dpi resolution, so that the stippling is visible (minimum required is 300 dpi).

P17, L388: Here we refer mainly to Fig. 6a, where MPI and IPSL are among the models with the highest latent heat flux sensitivity (together with UKESM and GISS) and thus further away from the observational constraint for latent heat flux, depicted on the x-axis.

Figure A3: to be updated in revised manuscript

References

Alibakhshi, S., Naimi, B., Hovi, A., Crowther, T. W., & Rautiainen, M. (2020). Quantitative analysis of the links between forest structure and land surface albedo on a global scale. Remote Sensing of Environment, 246, 111854.

Alkama, R., & Cescatti, A. (2016). Biophysical climate impacts of recent changes in global forest cover. Science, 351(6273), 600-604.

Bonan, G. B. (2008). Forests and climate change: forcings, feedbacks, and the climate benefits of forests. Science, 320(5882), 1444-1449.

Boysen, L., Brovkin, V., Pongratz, J., Lawrence, D., Lawrence, P., Vuichard, N., ... & Lo, M. H. (2020). Global climate response to idealized deforestation in CMIP6 models. Biogeosciences Discussions, 2020, 1-35.

Bright, R. M., Davin, E., O’Halloran, T., Pongratz, J., Zhao, K., & Cescatti, A. (2017). Local temperature response to land cover and management change driven by non-radiative processes. Nature Climate Change, 7(4), 296-302.

Flato, G., Marotzke, J., Abiodun, B., Braconnot, P., Chou, S. C., Collins, W., ... & Rummukainen, M. (2014). Evaluation of climate models. In Climate change 2013: the physical science basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change (pp. 741-866). Cambridge University Press.

Luo, X., Ge, J., Guo, W., Cao, Y., Liu, Y., Chen, C., & Yang, L. (2023). An evaluation of CMIP6 models in representing the biophysical effects of deforestation with satellite‐based observations. Journal of Geophysical Research: Atmospheres, 128(12), e2022JD038198.

Winckler, J., Lejeune, Q., Reick, C. H., & Pongratz, J. (2019). Nonlocal effects dominate the global mean surface temperature response to the biogeophysical effects of deforestation. Geophysical Research Letters, 46(2), 745-755.

Response to Reviewer 1 main comment about separation of local and non-local effects, and potential for underestimating the local effect of deforestation on near-surface climate:

We thank the reviewer for his/her comments and appreciate he/she finds our study interesting. We’d like to point out that the moving window regression approach, which our study builds on and develops further, has been applied and evaluated by several previous studies. A very recent study (Jaeger et al., 2025, 10.1029/2024JD042698) proposed an alternative third approach to moving window regression and the chessboard approach and confirms that both are generally valid. However, there has been some evidence that the regression approach underestimates the local effects to some extent, which, in our original submission, we acknowledge in the Limitations section (Section 5). Previous studies using a similar approach based on linear regression indeed pointed this out, when comparing this method to the factorial experiment approach (Lejeune et al., 2018) and the look-up table approach (Winckler et al., 2019 (Fig. S8)). In order to quantify the discrepancies between the different methods for separating local and non-local effects directly for our simulations, we have performed an additional analysis (see attachment, Fig. R1-R3) comparing the temperature sensitivities to deforestation in MPI-ESM as calculated in our study based on the deforest-glob simulation (via linear regression) and in the study of Winckler et al. (2019), where deforestation is performed in every one or two out of four grid cells (chessboard pattern). While the deforestation patterns in these simulations are different, the magnitude of deforestation is similar and therefore it is possible to approximately compare the two methods – the one based on linear regression and the chessboard pattern method (Fig. R1). This analysis showed that the differences in temperature sensitivities between the two methods vary in geographic space and, importantly, show no systematic bias towards underestimation of the local effect (Fig. R2).

It should be kept in mind that this comparison was done only with one model (due to data availability, this comparison was possible only with the MPI model) and that the chessboard pattern method comes with its own set of uncertainties (Winckler et al., 2019). Importantly, while we cannot exclude contamination of the local effects by non-local cooling, these results suggest no systematic underestimation of the local effect as suggested by previous studies (Lejeune et al., 2018; Winckler et al. 2019). Ultimately, modellers should view our results as a first evaluation and are encouraged to perform offline simulations to further test their model’s skill in replicating observationally based temperature (plus albedo- and latent heat-) sensitivities to deforestation.

Further, in this comparison (Fig. R2), we did not apply the scaling of the temperature sensitivities to 100% deforestation (see Section 2.4 in original manuscript, lines 203-210), in order to keep the magnitude of deforestation similar to the one in the experiments of Winckler et al. (2019). Applying the scaling does lead to amplification of the differences in temperature sensitivities (resulting from the two different methods for extracting the local effects), which are more notable over the boreal region in Eurasia and the central parts of the Amazon and Congo basin (Fig. R3). However, since the same scaling approach has been applied to the observational datasets, a comparison between models and observations is still meaningful and the validity of the emergent constraints is not affected.

To respond to the specific example given by the Reviewer, where the forest cover loss for all pixels within the moving window is 50%, we would like to point out that, indeed, this pattern of deforestation has prevented us from applying the classical linear regression approach by Lejeune et al. (2018). To avoid the issue, in our proposed methodology we have taken (instead of change) the values of temperature (T) and tree cover (F), which are different in the deforest-glob and piControl simulations. And, then we have trained a linear regression on the pairs of T and F from both simulations.

To reflect on this Reviewer’s main comment and our additional analyses, we intend to add the following text in the revised manuscript in the first paragraph under Limitations (Section 5): “An additional analysis comparing the temperature sensitivities to deforestation based on the linear regression method applied in this study and the chessboard pattern deforestation experiments of Winckler et al. (2019) did not reveal evidence of systematic underestimation of the local effects (see Supplement). However, the scaling to 100% could lead to overestimation of the local effects (compared to other methods) possibly because the relationship between surface temperature and tree cover change is not strictly linear as albedo and evapotranspiration effects vary as a function of initial tree cover and across biomes (Bonan, 2008; Alibakhshi et al., 2019).”.

Regarding the section on non-local effects, we have elaborated in the specific comments on the importance of considering the non-local effects. In this section, we show a way forward how to include in the comparison models that overestimate the albedo-induced cooling.

Response to specific comments from Reviewer 1:

Lines 14: to be fixed in the revised manuscript

15: suggested improvement: “[…] via an emergent constraint (i.e. resulting from the linear relationships within the model ensemble)”

55-57: suggested improvement: “But large-scale deforestation in these experiments also triggers strong changes in advection of heat and moisture, as well as in atmospheric and ocean circulation, which influence regions that have not undergone deforestation (Winckler et al., 2017; Portmann et al., 2022).”

75: to be removed to avoid confusion: “[…] temperature changes.”

91-92: to be removed to avoid confusion: “The objective is to provide emergent constraints…”; clarification of ”emergent constraint” added (see 15)

164-166: suggested improvement: “The change in forest fraction is 100% and is considered only for pixels, where the respective vegetation cover types are actually present and/or could potentially occur as defined by MODIS-derived land cover maps for 2005 and Köppen-Geiger climate zone maps for the 20^th century.”

185: to be fixed in the revised manuscript

186: It is not expected that including pixels with less than 10% forest cover change will alter the results substantially, as in most models the forest cover change is much larger than 10% (see attachment, Fig. R4). We have provided an additional analysis comparing the surface temperature sensitivity with and without the 10% threshold for the MPI model (see attachment, Fig. 5). The MPI model was selected based on the availability of more pixels with less than 10% forest cover change in comparison to other models. The results with using the threshold are more robust, especially for the boreal forest, as evidenced by the number of statistically significant pixels.

192: The information is already (partially) provided in 211-214. Suggested improvement: “In order to evaluate the consistency of seasonal surface temperature responses between the ESMs and the observation-based estimates, we extracted the mean values of local surface temperature responses to deforestation also at monthly timescale for broad latitudinal regions: boreal (from 50° N to 90° N), temperate (from 23° N to 50° N) and tropical (from 23° S to 23° N).”

203: See comment to 192

215-230: Explanation of the emergent constraint concept to be moved to Introduction.

223-230: suggested additions/changes: “By knowing the local effects from observations, and the emerging relationship between local and total effects from model experiments, we could use that relationship to constrain the total effects based on the observed local effects. Thus, by using the local sensitivities derived from in-situ and satellite-based data, we are potentially able to provide emergent constraints for surface temperature, albedo and latent heat flux. In addition, we use the relationship between local surface temperature and albedo, and between local surface temperature and latent heat flux, to explore further the emergent constraints.”

227: The possibility to compare local effects in models with observations has been confirmed by a number of studies (e.g., Bright et al., 2019; Winckler at al., 2019a; Chen&Dirmeyer, 2020). The issue with the possible overestimation of local effects is discussed in the first paragraph. To avoid a misunderstanding about the term “comparable”, we have changed the sentence from “[…] as local surface temperature sensitivities from models and observations are comparable” to “[…] as local surface temperature sensitivities from models and observations can be brought into a meaningful relationship”.

229: Here we refer to the total effects.

249-253: To be moved to Limitations in the revised manuscript.

278: Here we refer to local surface temperature.

280: Fig. A1 to be moved to the main text in the revised manuscript.

288: To be changed to “defined by the standard deviation of the MODIS observations"

304: Section 3.4 to be merged to Section 3.3

316: While in this section we report also on non-local effects, which are not the primary focus of the study, we also evaluate the local effects during summer. This evaluation is particularly important because by considering only summer months we are able to exclude the effects of the Siberian cold bias still observed in many CMIP6 models (Portal et al., 2023). It is also relevant in light of the discussion on the contamination of local effects by non-local effects. In the summer, the non-local cooling is weaker (due to the absence of the snow albedo effect), thus giving possibly clearer view on the local effects.

337: We refer to observational constraints as a broader term, encompassing emergent constraints. When non-local or total effects are referred to, we use specifically the term (observational) emergent constraints, as these effects can be evaluated only by considering the emergent linear relationship between the different models. “Emergent constraints” and “observational emergent constraints” are synonymous. Clarification to be added under Section 2.5.

337: suggested addition at line 350: “Because of the emergent constraint relationship between the local and total surface temperature effects, this overestimation is valid also for the overall response to deforestation, thus showing that most models exhibit too strong cooling in comparison to observations (Fig. A1). For the tropics, however, approximately half of the ESMs show realistic total surface temperature response, as defined by the emergent constraint based on MODIS data, with fewer models being within the margins defined by the FLUXNET-based dataset (Fig. A1).”

360: see above

378: see above

396: see comment to 316

412-496: see general comments

Fig. 3: to be updated in revised manuscript

Fig 4: to be updated in revised manuscript

Fig 5: to be updated in revised manuscript

Fig 6: to be updated in revised manuscript

Fig. A5 and A6: to be updated in revised manuscript

References

Alibakhshi, S., Naimi, B., Hovi, A., Crowther, T. W., & Rautiainen, M. (2020). Quantitative analysis of the links between forest structure and land surface albedo on a global scale. Remote Sensing of Environment, 246, 111854.

Bonan, G. B. (2008). Forests and climate change: forcings, feedbacks, and the climate benefits of forests. Science, 320(5882), 1444-1449.

Bright, R. M., Davin, E., O’Halloran, T., Pongratz, J., Zhao, K., & Cescatti, A. (2017). Local temperature response to land cover and management change driven by non-radiative processes. Nature Climate Change, 7(4), 296-302.

Jäger, F., Schwaab, J., Bukenberger, M., De Hertog, S. J., & Seneviratne, S. I. (2025). Spectral decomposition and signal separation of climate responses to land cover changes. Journal of Geophysical Research: Atmospheres, 130(5), e2024JD042698.

Lejeune, Q., Davin, E. L., Gudmundsson, L., Winckler, J., & Seneviratne, S. I. (2018). Historical deforestation locally increased the intensity of hot days in northern mid-latitudes. Nature Climate Change, 8(5), 386-390.

Winckler, J., Lejeune, Q., Reick, C. H., & Pongratz, J. (2019). Nonlocal effects dominate the global mean surface temperature response to the biogeophysical effects of deforestation. Geophysical Research Letters, 46(2), 745-755.

Portal, A., d'Andrea, F., Davini, P., Hamouda, M. E., & Pasquero, C. (2023). Atmospheric response to cold wintertime Tibetan Plateau conditions over eastern Asia in climate models. Weather and Climate Dynamics, 4(3), 809-822.

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

We thank the reviewer for his/her comments and appreciate he/she finds our study interesting.

Response to Reviewer 3 main comment about separation of local and non-local effects, and potential for underestimating the local effect of deforestation on near-surface climate:

See detailed responses to Reviewer 1 main comment.

In addition, it should be noted that the key idea of the separation method(s) is that the non-local effects are part of the background climate that is similar in close-by grid cells.  Deforestation outside the moving window undoubtedly affects the climate variable within, but not in a systematic way (e.g. as a function of tree cover change within the moving window) and therefore related effects can be excluded. 

Response to specific comments of Reviewer 3:

Lines 130 – 133: The local surface temperature response of the two transitions – grassland to evergreen needleleaved forest (ENF) and grassland to deciduous broadleaved forest (DBF), is almost identical (Bright et al., 2017, Fig. 3 d,e). Choosing the ENF instead of the DBF transition will, therefore, not influence the validity of our results for the temperate region. To reflect on this comment, the following text will be added to the revised manuscript: “While broadleaved forests in Europe were replaced by coniferous forests throughout the last centuries and may not be the dominant forest type in Europe anymore (Naudts et al., 2016), the local surface temperature response of the two types of forest conversion is very similar (Bright et al., 2017) and therefore the choice of a specific forest transition does not influence the observational constraint.”. In order to justify the use of the DBF transition (instead of ENF), we will add also a reference to the study of Ma et al. (2023), according to which in the temperate region 28.8% of the forests are deciduous, and 23.1% are coniferous. We will also provide in the Supplement an additional plot (similar to Fig. 3b in the original manuscript), showing both transitions.

Figure 1: The green area is defined by the slope of the multiple linear regression and is represented as a plane in the multidimensional space, defined by the predictor variables (tree cover, latitude, longitude and elevation).

Lines 277 – 279: The emergent constraint relies mostly on the consistent behaviour of the ESMs (e.g., decrease of temperature with increasing albedo), but also on the presence of variation/errors in the models (Williamson et al., 2021). For example, if all ESMs exhibit the same error or the spread between models is removed (e.g., due to model improvements), the emergent constraint could disappear (Hall et al., 2019). Suggested improvement: Our previous findings showed that across the ESMs an approximately linear relationship between temperature and albedo or latent heat flux changes can be found. These linear relationships open up the possibility of providing observation-based emergent constraints.

Lines 318 – 320: The main idea here was to provide a plausible ensemble mean by excluding outlier models. The excluded models are MIROC, CanESM, CNRM and UKESM (Fig. 7). Suggested improvement: “In this analysis, we concentrate on models having plausible local albedo and latent heat flux sensitivities, defined here as being within two standard deviations of the model ensemble mean, which leads to the exclusion of four out of the eleven ESMs…”

References

Alibakhshi, S., Naimi, B., Hovi, A., Crowther, T. W., & Rautiainen, M. (2020). Quantitative analysis of the links between forest structure and land surface albedo on a global scale. Remote Sensing of Environment, 246, 111854.

Bonan, G. B. (2008). Forests and climate change: forcings, feedbacks, and the climate benefits of forests. Science, 320(5882), 1444-1449.

Bright, R. M., Davin, E., O’Halloran, T., Pongratz, J., Zhao, K., & Cescatti, A. (2017). Local temperature response to land cover and management change driven by non-radiative processes. Nature Climate Change, 7(4), 296-302.

Hall, A., Cox, P., Huntingford, C., & Klein, S. (2019). Progressing emergent constraints on future climate change. Nature Climate Change, 9(4), 269-278.

Ma, H., Crowther, T. W., Mo, L., Maynard, D. S., Renner, S. S., Van den Hoogen, J., ... & Parada-Gutierrez, A. (2023). The global biogeography of tree leaf form and habit. Nature Plants, 9(11), 1795-1809.

Naudts, K., Chen, Y., McGrath, M. J., Ryder, J., Valade, A., Otto, J., & Luyssaert, S. (2016). Europe’s forest management did not mitigate climate warming. Science, 351(6273), 597-600.

Williamson, M. S., Thackeray, C. W., Cox, P. M., Hall, A., Huntingford, C., & Nijsse, F. J. (2021). Emergent constraints on climate sensitivities. Reviews of Modern Physics, 93(2), 025004.

Winckler, J., Lejeune, Q., Reick, C. H., & Pongratz, J. (2019). Nonlocal effects dominate the global mean surface temperature response to the biogeophysical effects of deforestation. Geophysical Research Letters, 46(2), 745-755.