Introduction

Forest cover change (FCC) can make dramatic impacts on the environment through the ecological disturbances such as wildfires, anthropogenic deforestation/afforestation, and ecological adaptations to local climate changes1,2,3. Biogeophysical impacts of FCC may originate from different processes. FCC is conventionally considered to take an effect by modulating energy and water fluxes from evapotranspiration4,5. Forest-covered area typically comes with lower surface albedo and higher surface roughness, which may act as the source of perturbation of land-air interaction processes1,6. The impacts of FCC through evapotranspiration and radiation flux changes are especially pronounced over tropical forests with little seasonal variability throughout the year7,8. On the other hand, the boreal forests, as one important biome on earth, cover approximately 30% area of the global forests9. The energy and water fluxes over the boreal forests feature a more distinct seasonality. Both radiation fluxes and evapotranspiration are pronounced in summer, while small in winter over the boreal forest due to the frigid climate and limited solar radiation. Observational evidence regarding the responses of Surface Air Temperature (SAT) to FCC has suggested a weak cooling due to boreal FCC4,10, without any discussion on the physical mechanism behind.

Evapotranspiration and surface albedo are the major controllers of FCC impacts on the local scale11,12. Regarding boreal forests, evapotranspiration barely remains during winter, and surface albedo is strongly influenced by snow. While earlier studies on FCC impacts have primarily focused on local radiative and heat flux effects, recent findings are evident on FCC’s capacity to induce remote global-scale impacts, and the strength of which sometimes outweighs the local impacts13,14. Non-local impacts of FCC typically result from the variations of the large-scale circulation and/or cloud covers15,16,17. Bringing atmospheric circulation into consideration, Earth system models present explicit advantages due to their fully-coupled modeling framework. The Land Use Model Intercomparison Project18 (LUMIP) demonstrates significant large-scale atmospheric circulation anomalies in response to the idealized deforestation scenario, which then brings considerable adjustments to global climate13,19,20.

Under the background of climate change, the pan-arctic, including boreal forests, has been warming at a rate of 2-3 times greater than the global average, known as the arctic amplification. A typical winter SAT pattern over the northern hemisphere is associated with arctic amplification, also known as the Warm Arctic-Cold Eurasia21,22 (WACE). Earlier studies have discussed the effect of Barents-Kara sea ice decline23,24,25 and Eurasia snow cover variabilities26,27,28 on the emergence of WACE, both of which take the effect through stratosphere-troposphere interactions. The surface perturbation triggers upward propagation of planetary waves that weaken the stratospheric polar vortex29,30,31. The weaker polar vortex pilots more cold air mass into the mid-latitudes, making cold anomalies over Eurasia32,33. Whether there is a relationship between the SAT distribution over the arctic-boreal area, as well as the WACE pattern, and the FCC remains an open question to answer.

In this study, we conduct Earth System Model simulations with observed FCC imported to explore its biophysical effect on modulating winter climate over the arctic-boreal area and the potential mechanism behind. Sensitivity experiments are designed with and without the interactive coupling of atmosphere and land component models. We specifically focus on the atmospheric circulation responses to FCC that could, in turn, bring non-local impacts on regional to global climate during winter. The modeled climatic impacts are compared with LUMIP model results that are forced by an idealized deforestation scenario for estimating the impact of deforestation versus afforestation in different regions.

Results

Cold Eurasia resulted from the polar vortex weakening

The global forest cover change from 2000 to 2020 is characterized by a predominant deforestation in tropical forests across Africa, Asia, and South America (Fig. 1). Boreal forest loss occurs in Central Siberia, Alaska, and northern Canada, contrasting with pronounced afforestation in western Europe. Central and eastern Siberia does not exhibit any consistent changes, but a mix of deforestation and afforestation in scattered grids. The Leaf Area Index (LAI) processed from the MODIS satellite data presents a similar distribution of trend from 2000 to 2020 (Supplementary Fig. 1). Afforestation regions spatially correspond to an increasing trend of LAI, and vice versa.

Fig. 1: The forest cover change (%) described by GFW data from 2000 to 2020.
figure 1

Regions with the percentage of forest cover less than 0.5% are left blank.

Full size image

Given the minimal winter solar radiation in the boreal forest area, Arctic land energy fluxes depend primarily on the sensible heat flux and longwave radiation flux from the atmosphere. Our offline simulations with the observed FCC imported show negative energy flux anomalies (combined longwave, sensible, and latent heat) over deforestation regions in winter, implying the reduced land-atmosphere energy transfer, while afforested regions show positive anomalies reflecting enhanced upward energy fluxes. However, the anomalies in heat and energy transfer are limited within very small magnitudes (<1 W/m²; Supplementary Fig. 2), which are approximately an order of magnitude smaller than those during summer34,35,36. Consequently, SAT responses in offline simulations are limited in regions with deforestation. The magnitude of the difference is less than 0.1 °C. Regions with afforestation fail to produce any SAT responses (Supplementary Fig. 3).

On the contrary, online simulations with land and atmosphere coupled reveal significantly amplified winter SAT responses. Interestingly, the strongest temperature anomalies occur in boreal regions rather than tropical deforestation hotspots. The spatial pattern shows a dipole-like pattern with pronounced warming over the central Arctic and Greenland while cooling across Eurasia and parts of North America (Fig. 2). The average cooling over the mid- and high-latitudes 40 °N poleward of Eurasia is 0.83 ± 0.37 °C, with the coldest anomalies in central/eastern Siberia. Such a pattern remains in another set of simulations with the FCC only in the boreal area (40 °N poleward; Supplementary Fig. 4), demonstrating that boreal forest changes alone can drive this hemispheric-scale temperature dipole. These results establish boreal FCC as the dominant driver of divergent SAT responses over the central Arctic and Eurasia.

Fig. 2: The responses of temperature and atmospheric circulation patterns to the FCC in winter.
figure 2

The ensembled anomalies of a winter (DJF) surface air temperature (°C), b Sea Level Pressure (SLP), and c 200-hPa geopotential height (gpm) from 2000 to 2020 in response to the observed FCC, as well as d–f their model spreads calculated as the standard deviation of anomalies from 9 ensemble members. Black dots indicate that the anomalies pass the significance test at a 95% confidence level.

Full size image

The atmospheric circulation responses to FCC exhibit vertically coherent anomalies characteristic of stratosphere-troposphere coupling, with the Sea Level Pressure (SLP) pattern comprising an Arctic high-pressure anomaly flanked by twin low-pressure systems over the North Atlantic and Pacific (Fig. 2b), which presents a configuration resembling a modified negative Arctic Oscillation phase with Eurasian-shifted centers37. The upper-level (200 hPa) response features as a positive geopotential height anomaly in the central Arctic encircled by negative anomaly centers in the mid-latitudes (Fig. 2b).

The above responses to FCC show robustness represented by the model spreads between ensemble members (Fig. 2d–f). Model uncertainties on simulating the SAT response are relatively high over Scandinavia and eastern Siberia, while it limited to less than 0.3 °C over most of the boreal regions of Eurasia, which is about 50% smaller than those over Canada on the North American side (Fig. 2d). The model spreads of SLP and 200-hPa geopotential height are typically less than 1hPa (Fig. 2e) and 10 gpm (Fig. 2f), respectively, for the central Arctic, suggesting that different ensemble members agree on the positive geopotential height anomaly on the high level and high-pressure anomaly at the surface.

We notice that the SAT responses to the FCC have some similarities to the WACE pattern. The WACE pattern is known to result from the weakened polar vortex and negative phase of AO at the surface38,39, which also coincides with the atmospheric circulation responses shown in Fig. 2b, c. The negative SAT anomalies over Eurasia associated with the FCC in this study have a similar magnitude as the typical WACE pattern, while the warm temperature anomaly over the Arctic is not as strong. One obvious difference lies over the Ural Mountains on land and the Barents and Kara Seas, over which few anomalies are shown in this study. The difference could be explained by the absence of sea ice decline according to the experiment design. Since sea ice decline leaves more open water area with sea surface temperature warmer than air, increases in the upward sensible and latent heat flux are expected during winter40. Such a heat source results in a warm anomaly over the surrounding area, contributing to abnormal warming over the Barents and Kara Seas23. The polar vortex weakening and the corresponding WACE formation in winter are the consequences of surface perturbation that stimulates troposphere-stratosphere interaction, calling for the exploration of whether the FCC can trigger polar vortex weakening.

The vertical structure of atmospheric anomalies provides robust evidence for polar vortex weakening. The meridional-mean geopotential height anomalies display a characteristic pattern featuring positive anomalies in high latitudes, contrasting with negative anomalies in low latitudes in the upper troposphere and lower stratosphere (Fig. 3a). This structure corresponds to significant deceleration of zonal westerly winds, particularly in the stratosphere, confirming a weakening of the polar vortex (Fig. 3b). The thermal structure shows Arctic warming extending from the surface to the stratosphere with maximum warming near the surface and at the 100 hPa level, while the mid-latitudes experience persistent tropospheric cooling (Fig. 3c). These coordinated changes, including stratospheric warming, westerly wind deceleration, upper-level height anomalies, and surface climate responses, demonstrate that boreal forest cover change triggers dynamical pathways similar to sudden stratospheric warming events41,42. The spatial consistency between these atmospheric responses and observed surface temperature patterns confirms the chain of causality from forest cover changes through vortex weakening to surface climate impacts.

Fig. 3: The vertical structures of the atmospheric responses to the FCC.
figure 3

The meridionally-averaged anomaly ensembles in a Geopotential height (gpm), b Zonal wind (m/s), and c Temperature in the northern hemisphere. Black dots indicate the difference passes the significance test at a 95% confidence level.

Full size image

Evidence has pointed to the weakening of the polar vortex during winter in response to the FCC, resulting in cold SAT anomalies over Eurasia43. To further analyze the planetary wave activity as a potential driver of polar vortex weakening, we address Eliassen-Palm flux44 (E-P flux hereafter) analysis to quantify the upward propagation of Rossby waves. Diagnosed by the E-P flux anomaly between the simulations with and without FCC, latitudes of 30–60 °N have seen a strong upward propagation of the stationary planetary wave from the surface and a significant E-P flux convergence in the upper troposphere (Fig. 4), indicating an important role of the midlatitude wave forcing. This range of latitudes corresponds well to the regions with the most FCC in the northern hemisphere.

Fig. 4: The vertical coupling of planetary waves in response to the FCC from November to February.
figure 4

a The anomalous E-P flux (vectors; m3s−2) and its divergence (shaded; m2s−2) in November, b–d Same as a, but for December, January, and February, respectively. Pink dots indicate the anomaly that passes the significance test at a 95% confidence level. E-P flux anomalies less than 0.5 × 104 m3s−2 are discarded from the plot.

Full size image

In November and January, upward propagation of planetary waves is characterized by a clear convergence of the E-P flux in the Arctic stratosphere, contributing to stratospheric polar vortex weakening (Fig. 4a, c). In December and February, downward movements of E-P flux convergence anomalies are observed in the polar stratosphere, accompanied by the suppression of stratospheric planetary waves (Fig. 4b, d). These characteristics are consistent with the wave-mean flow interaction during the recovery stage of weak polar vortex events45. It shows the intraseasonal alternation of vertical wave couplings, with November and January serving as two critical periods of upward propagation of planetary waves, which can effectively perturb the polar vortex.

Previous research has established that upward-propagating planetary waves typically influence surface temperatures in the following several weeks through polar vortex weakening46. Notably, our observed E-P flux anomalies bear a strong resemblance to those by Han and Sun47, who demonstrated that November snow cover anomalies can generate similar E-P flux patterns and subsequently dominate Eurasian winter temperature variability. This compelling similarity has led us to further examine snow dynamics in November.

Snow dynamics variation weakens the polar vortex

Analysis focusing on the snow conditions in November suggests that FCC drives significant modifications, with afforested regions in western Europe experiencing dramatic snowfall reductions due to positive temperature anomalies (Fig. 5a), causing precipitation to fall primarily as rain rather than snow, resulting in 13.6 ± 3.4% decreases in ground snow cover in northwest Europe (Fig. 5b). Such an anomaly pattern in snowfall resembles the zonal dipole mode (the first EOF mode of snow cover distribution) previously linked to Eurasian winter cooling48. These changes further affect snow persistence, with afforested areas showing snow cover persistence shortened by 4.9 ± 1.83 days on average, while deforested regions in central/eastern Siberia exhibit comparable lengthening (Fig. 5c). The snow cover dynamics change eventually drives the variation of surface albedo, with 3.1 ± 1.1% decreases in afforested zones and 5.3 ± 1.8% increases in deforested areas in Eurasia (Fig. 5d). The reduced snow cover and lower surface albedo induced by forest cover change (FCC) drive significant increases in upward longwave radiation and turbulent heat fluxes from the surface to the atmosphere. This process is amplified through strengthened snow-albedo feedback49 (SAF), where both greater forest cover and enhanced snowmelt work synergistically to reduce surface albedo and elevate surface air temperatures. Our online simulations demonstrate this effect most prominently in western Europe and Scandinavia, where afforestation intensifies the SAF by up to 1.15%/ °C (Fig. 5e). The regional impact becomes particularly evident in November, when northwestern Europe (including Scandinavia) exhibits substantial warming of 1.99 ± 0.92 °C (Supplementary Fig. 5). On the other hand, the response of total precipitation shows a different spatial pattern (Supplementary Fig. 6). In this way, the FCC-induced snow cover variation can potentially result in thermodynamic perturbations that stimulate upward propagation of planetary Rossby waves.

Fig. 5: The spatial patterns of snow dynamics responses to the FCC in November.
figure 5

The anomalies of the a snow fall from atmosphere (mm of snow water equivalent), b snow cover (%), c snow persistence (day) as the consecutive length with snow cover on the ground, d surface albedo (%), and e snow-albedo feedback (%/ °C). Black dots indicate that the difference passes the significance test at a 95% confidence level.

Full size image

Note that the reduced snow cover and corresponding albedo difference may also be the consequence of the increased SAT in November over western Europe. We further investigate the offline simulation results with no feedback process between atmosphere and land to verify if the FCC itself can lead to the difference in snow cover. As a result, forest cover percentage shows a strong linear correlation with snow interception by canopy, demonstrating that increased forest cover leads to greater snow accumulation on canopies rather than on the ground. Offline simulation results suggest 1% increase in forest cover produces a 0.75% rise in canopy-intercepted snow on average for the area covered by boreal forests (Fig. 6a). Snow on the canopy melts or sublimates at a higher rate compared to the snow on the ground50. Offline simulations quantify that every 1% increase in boreal forest cover corresponds to a 0.29% increase in snowmelt in November without any interference from atmospheric temperature (Fig. 6b). These results establish a direct mechanistic pathway through which FCC alters surface energy budgets—not only by changing albedo but also by modifying snow storage and melt timing through canopy interception processes, with particularly strong impacts during the vulnerable shoulder season when temperature fluctuations frequently cross the melting point.

Fig. 6: The relationship of forest cover and snow accumulation.
figure 6

The changes in a canopy intercepted snow (%) and b ratio of snow melted (%) in response to forest cover change (%). Lines indicate the linear regression with functions and a 95% confidence level delimited by the shaded area.

Full size image

The above statistics, based on offline simulations, are further verified by observational evidence. Analysis of the observed snow water equivalent (accumulated from September to November as snowfall minus snowmelt) reveals distinct snow cover patterns between afforested and deforested regions under equivalent snow accumulation (Fig. 7a). While snow cover increases linearly with accumulation in both cases, afforestation is consistently associated with reduced snow coverage for any given accumulation amount. For instance, 1.5 mm of snow produces 83.2% coverage in deforested areas but only 74.6% in afforested zones (Fig. 7a). This difference stems primarily from enhanced canopy snow interception in forested areas, where intercepted snow experiences faster melt due to greater exposure to atmospheric conditions. The resulting snow accumulation differences directly impact surface albedo, with C3S surface albedo data showing afforested regions exhibit a slower albedo increase with snow cover (0.028%/% in afforested areas versus 0.073%/% in deforested areas; Fig. 7b). This reduced snow-albedo sensitivity in afforested areas reflects the combined effects of enhanced canopy snow interception reducing ground accumulation, and faster melt of intercepted snow due to lack of ground insulation50. Consistent observational patterns across both snow cover and surface albedo validate the modeled interactions, particularly in November when the snow-albedo feedback amplifies subsequent winter climatic responses.

Fig. 7: The relationship of snow accumulation and responses in surface albedo.
figure 7

The scatter of a snow cover (%) and accumulated snow (mm of snow water equivalent), as well as b snow cover and surface albedo (%) in November for deforestation (brown dots and lines) and afforestation (green dots and lines) regions. Lines indicate the linear regression with functions and a 95% confidence level delimited by the shaded area.

Full size image

Model results and observational evidence together reveal a coherent mechanism by which boreal FCC since 2000 has contributed to colder Eurasian winters. This chain of process starts from the shoulder season of November, when afforestation in western Europe alters snow accumulation dynamics through intercepting more snow on the canopy. Snow redistribution involving the FCC-enhanced SAF due to faster melt of intercepted snow. The SAF further decreases snow cover and increases SAT. The surface perturbations trigger upward propagation of planetary waves that weaken the polar vortex through the wave-mean flow interaction. As a result, pronounced geopotential height rise and westerly wind deceleration are observed in the polar stratosphere. The stratospheric disturbances then propagate downward to the troposphere, inducing a negative phase of the Arctic Oscillation that generates cold anomalies over Eurasia. This mechanism explains how regional FCC can generate hemispheric-scale winter climate impacts through stratosphere-troposphere coupling, with November snow-canopy-atmosphere interactions serving as the critical initiation point for the entire cascade of dynamical events.

The temperature anomalies induced by FCC through polar vortex weakening exhibit distinct characteristics compared to the well-documented WACE pattern, with the Eurasian cooling confined to higher latitudes. This contrasts sharply with the North American sector, where cold anomalies extend remarkably southward to 30 °N, covering most of Canada and the continental United States. When compared to the second EOF mode of Northern Hemisphere winter temperature variability51, the FCC-induced pattern shows a similar spatial pattern over Eurasia but with opposite distribution over North America. This hemispheric asymmetry implies that the polar vortex response from FCC involves not only weakening but also distinct meridional stretching toward Eurasia. The different temperature responses between continents probably result from the spatial heterogeneity of FCC and the atmospheric wave propagation characteristics. These findings highlight the need for future research to separately examine Eurasian and North American FCC impacts, particularly regarding how their distinct forest-snow-atmosphere couplings generate such asymmetric winter temperature responses through modified vortex dynamics.

Polar vortex weakening absent in LUMIP models

While our study incorporating observed FCC demonstrates clear polar vortex weakening and associated cold anomaly in Eurasia, the idealized global deforestation scenario of LUMIP fails to reproduce a similar response (Fig. 8). Analysis of the six participating models reveals no consistent SAT anomaly pattern, four of which show concurrent Eurasian and Arctic cooling, and the remaining two, CESM2 and IPSL-CM6A-LR, even produce the pattern of cold Arctic and warm Eurasia. The magnitudes of temperature responses also differ substantially from model to model.

Fig. 8: The surface air temperature (°C) anomaly in response to the prescribed forest cover change between deforest-globe and pi-control experiments in the LUMIP.
figure 8

a The surface air temperature anomaly (shaded) between the deforest-globe and pi-control simulations present in the BCC-CSM-MR model. b–f Same as a but for CESM2, CMCC-ESM2, IPSL-CM6A-LR, NorESM2-LR, and MPI-ESM1-2-LR, respectively. Black dots indicate that the difference passes the significance test at a 95% confidence level.

Full size image

Corresponding to the different responses on SAT in LUMIP models, responses of atmospheric circulations, including polar vortex responses, substantially differ from our simulations with observationally constrained FCC. While our study demonstrates coherent stratosphere-troposphere coupling through vertically propagating anomalies, the LUMIP models do not show any robust signals of polar vortex weakening represented in fields of either geopotential height or zonal wind (Fig. 9). Four models (BCC-CSM2-MR, CMCC-ESM2, IPSL-CM6A-LR, and MPI-ESM1-2-LR) respond as limited high-pressure anomalies confined to the upper stratosphere, failing to extend downwards to the troposphere. Consequently, the warming effects remain trapped in the stratosphere, while surface conditions maintain cold anomalies throughout. The zonal-mean cross sections confirm that none of the six models reproduces the characteristic dipole structure of geopotential height and zonal wind between mid- and high-latitudes.

Fig. 9: The meridional cross section of the anomalies of geopotential height (gpm), zonal wind (m/s), and temperature (°C) of the northern hemisphere between globe-deforest and pi-control experiments of LUMIP.
figure 9

Black dots indicate that the difference passes the significance test at a 95% confidence level.

Full size image

Similar to previous studies, we also find substantial disagreement of SAT between LUMIP models in response to the idealized deforestation scenario, which is considerably greater than that between the ensemble members of our CESM2 experiments20,52. The greater uncertainties could result from different causes. First, the LUMIP simulations are conducted in a fully-coupled model configuration, while our experiment is configured with a prescribed SST field that is non-interactive with the atmosphere. The involvement of the feedback process between ocean, land, and atmosphere could substantially enhance the model uncertainty. Moreover, LUMIP models have slightly different deforestation patterns (Supplementary Figure 7) due to the experiment design18,52. Magnified by the feedback between land and atmosphere, biases in responses from different FCC patterns could be magnified substantially.

More importantly, the absence of both the afforestation in the idealized FCC scenario and the cold SAT anomalies in Eurasia in the LUMIP models indirectly verifies that the afforestation over the west Europe could potentially be the more important contributor to the cold winters in Eurasia compared to the FCC in other boreal regions. Our experiments with observed FCC underscore that a realistic spatial pattern of both forest loss and gain is necessary, as afforestation and deforestation exert asymmetric influences on winter atmospheric dynamics through their differential effects on snow cover and surface energy budgets.

Discussion

This study demonstrates that the boreal forest cover change (FCC) from 2000 to 2020 induces winter cooling over Eurasia via polar vortex weakening through model sensitivity experiments. We propose a novel mechanism through which FCC can have an impact on winter climate via the inhibition of snow accumulation on the ground and the reduction of surface albedo. The snow cover anomalies mostly happened in the afforestation area in western Europe, subsequently enhancing upward propagation of planetary waves, resulting in polar vortex weakening in the stratosphere. The weakened polar vortex then propagates downward, leading to the surface circulation pattern similar to the negative phase of the Arctic Oscillation that causes warm SAT anomalies in the Arctic and cold anomalies over Eurasia.

The sensitivity experiments establish that land-atmosphere coupling is essential for reproducing these winter climate impacts, revealing that atmospheric circulation responses dominate FCC effects during cold seasons. Previous studies have agreed on the importance of the non-local impacts of FCC, and the cloud and radiation effects are paid specific attention to in influencing SAT during summer20,53. In winter, however, the effect of clouds becomes less pronounced due to the lack of solar radiation. The total cloud cover responses to the FCC do not present a spatial correspondence to those of SAT anomalies (Supplementary Fig. 8). We therefore speculate that troposphere-stratosphere interaction could be the primary contributor to the FCC impacts during winter.

This modeling study recognizes certain inherent limitations while affirming the robustness of its central findings regarding forest cover change impacts. Our study explores the polar vortex weakening based on the coupling of the atmosphere and land component models. The specific focus on the land-air interaction helps to isolate the direct influence of FCC on atmospheric circulations. However, the FCC impact through oceanic feedbacks has a comparable or even greater effect compared to those via land-air interactions. For example, an analysis regarding LUMIP models has confirmed the impact of deforestation on the slowdown of Atlantic Meridional Overturning Circulation54,55 (AMOC) that could bring substantial impacts to global climate14. In addition to our study suggesting polar vortex weakening due to FCC through land-air interactions, further studies could involve oceanic processes in response to the historical FCC for a more comprehensive exploration of the climatic impact of forest cover change.

Our proposed mechanism points to the key effect of afforestation in the process of snow accumulation inhibition and SAF enhancement, and afforestation is extensively spread across western Europe. It lacks direct evidence through the analyses in this study on whether it is only Western Europe with the most extensive afforestation that has a positive effect in stimulating the upward propagation of planetary waves. What the E-P flux diagnosis represents is the zonally averaged conditions of planetary wave propagation. Aside from the afforestation in western Europe, central Siberia shows deforestation and an increase in surface albedo after snowfall in November, which could also contribute to the planetary wave propagations. This study confirms the effect of FCC conditions in the boreal area as a whole in weakening the polar vortex, while exploring the regional contribution of FCC requires a specific configuration of model sensitivity experiments in further studies.

The experimental approach necessarily simplifies real-world forest transitions by linearly interpolating between 2000 and 2020 forest cover snapshots, therefore failing to capture either any climatic impacts of the temporal heterogeneity of forest cover change, including the forest recovery after deforestation. The CESM2 configuration’s year-end vegetation updates introduce technical artifacts—likely explaining our unique January planetary wave signals (Fig. 5c) absent in observational studies—yet crucially, these implementation details do not compromise the core physical mechanism. The proposed SAF pathway remains valid because its key November processes precede the model’s vegetation updates and involve fundamental physics. These findings thus establish afforestation as a meaningful contributor to winter climate variability, though future work with more realistic vegetation transition schemes could better constrain the timing and magnitude of effects.

Compared to the observed Eurasian snow cover variability known to influence polar vortex strength in the previous study47, our study complements that climate change and FCC could act synergistically to amplify snow cover reductions in Eurasia and corresponding climatic impacts. Specifically, while background warming drives initial snow loss, afforestation intensifies this trend through biophysical processes, making FCC an amplifier in the process of snow-albedo feedback that plays a critical role in the climatic variability in boreal winters. The highlights of this study expand the understanding of land cover-climate interactions by emphasizing the climatic effect of FCC through dynamic atmospheric circulation effects in cold seasons when radiative and evaporative effects of vegetation are minimized. For current Earth system model studies, the importance of land use/cover change may have been less emphasized. Future climate mechanism research and climate change projections must therefore integrate both dynamic and thermodynamic effects to avoid underestimating or misattributing regional climate changes associated with large-scale land use/cover changes.

Methods

Data sources

FCC in this study is prescribed through the forest cover data from Global Forest Watch56 (GFW) with satellite-derived forest cover snapshots in the years 2000 and 2020. The original 1km-resolution binary forest cover data (forest/no forest pixels) were processed by aggregating pixel counts within each model grid to calculate the area weight of forest cover for model input. We calculate the gridded forest cover in percentage for each year from 2000 to 2020 by assuming a linear rate of forest cover change as the model input data. Our sensitivity experiments are designed as follows. The control simulation is with forest cover fixed in 2000, and the sensitivity experiment has the observed FCC imported. The offline and online configurations are applied to each pair of sensitivity experiments, making four simulations in total for one group of experiments.

In addition, we applied snow dynamics variables from multiple reanalysis data and observations. The ERA5 assimilates snow observations heavily from both weather stations and satellite measurements57. We employ the snowfall data in the form of snow water equivalent from ERA5. Snow cover data applied from the MODIS satellite observations58. In addition, we employ 10-day average surface albedo data processed by Copernicus Climate Change Service59 (C3S) from an ensemble of satellite measurements of SPOT-VGT and PROBA-VGT. The observational datasets have been gridded for the convenience of analysis.

Model configuration

We employ the latest version of the Community Earth System Model60 (CESM2) as the modeling platform. We configure both offline and online simulations to quantify non-local impacts through land-air interaction. The offline simulations exclusively activate the land model component of Community Land Model61 (CLM5) forced by the ERA5 reanalysis data57. For online simulations, we configure a two-way coupling between the land (CLM5) and the atmosphere component model of the Community Atmospheric Model (CAM6) within CESM2. In both cases, the ocean component remains as the data model, with online simulations incorporating climatological Sea Surface Temperature (SST) and sea ice data from the Hadley Centre62 as the oceanic boundary condition. This experimental design eliminates extra variability through ocean-air interaction that could hinder the FCC responses.

The land and atmospheric component models in CESM2 are configured with a global coverage with 2-degree resolution (144 × 96 points). The simulation period is from 2000 to 2020, before which a 100-year spin-up is conducted with the SST climatology in 1980—2000 and forest cover condition fixed in the year 2000. The 100-year length for spin-up is long enough in this study, as it does not involve any biogeochemistry.

The GFW forest cover data does not contain any forest type information, so we borrow the forest type distribution directly from the CLM5. Specifically, the FCC occurs between forest and grassland, being consistent with the LUMIP simulations. In this way, tree removal from deforestation is replaced by the same area weight of grassland in each grid. In particular, the boreal forest is replaced by C3 arctic grass, the temperate forest is replaced by C3 grass, and the tropical forest is replaced by C4 grass. Correspondingly, the prescribed afforestation conducts the opposite. The transition of PFT is imported at the end of each year and takes effect in the following year.

To eliminate the interference of internal variability, a total of 9 groups of ensemble simulations are conducted. A random perturbation of air temperature with a magnitude of 10-8 K is added to the initial condition of each ensemble experiment. In comparison, the CESM2 large ensemble project63,64 (CESM2-LE) applies a 10−14 K level of initial disturbance for each ensemble member. We cannot afford as many ensemble members as in CESM2-LE due to our limitation in computational resources, so we properly increase the magnitude of initial perturbations.

The LUMIP data

The LUMIP project facilitates model intercomparison of deforestation impacts through fully coupled simulations of different Earth system models18. Its global deforestation scenario (deforest-globe) branched from piControl simulations prescribes a linear deforestation trend removing 20 million km² of forests over 50 years, targeting the top 30% forest-covered grid points. Unlike our experimental design using observed FCC and prescribed SST boundary conditions, LUMIP’s fully-coupled framework (with active ocean/sea ice components) may introduce additional uncertainties relative to real-world conditions. We analyze the first 50 years of deforest-globe and piControl simulations across six LUMIP models (BCC-CSM2-MR, CESM2, CMCC-ESM2, IPSL-CM6A-LR, NorESM2-LR, and MPI-ESM1-2-LR) to assess inter-model variability stemming from both structural differences and FCC implementation variations.