Important non-local effects of deforestation on cloud cover changes in CMIP6 models

We analyzed the idealized global deforestation (i.e. deforest-glob) simulations from climate models participated in the CMIP6 (table S1). The deforest-glob experiment was designed to investigate the biophysical role of land cover change on climate and to inter-compare modeled biogeochemical response to deforestation (Lawrence et al 2016). As CMIP6 models have different forest cover distributions in the pre-industrial control (i.e. piControl) simulation, the selection of grid cells for deforestation is based on the fractional forest cover in a given model's piControl simulation. In deforest-glob, 20 million square kilometers of forested area (covered by trees) is converted to natural unmanaged grasslands over the top 30% of land grid cells in terms of their area of tree cover (figure S1) over a period of 50 years with a fixed rate of 400 000 km²yr⁻¹ followed by at least 30 years of constant land cover. It is worth noting that although all CMIP6 models follow this deforestation protocol, a few models (e.g. UKESM1-0-LL) also have vegetation change after year 50 due to their model structure (Boysen et al 2020). CO₂ concentration, land-use and land management and other forcings are kept constant at their pre-industrial levels. More detailed information on the deforest-glob experiment can be found in Lawrence et al (2016).

We compared the last 30 years of the piControl and deforest-glob simulations to derive the mean response to deforestation. The statistical significance level of the difference was estimated based on a two-tailed modified Student's t test (Zwiers and von Storch 1995). We focused on the three-month period of June, July and August (JJA), when the cloud differences maximize between forests and grasslands (Duveiller et al 2021, Xu et al 2022). The cloud differences also maximize in the tropical Southern Hemisphere during boreal summer (not shown).

Deforestation affects the climate locally at deforested boxes (termed 'local effects') by altering land surface properties (e.g. albedo, roughness, evapotranspiration and leaf area index), and also influences both deforested and un-deforested grid boxes by changing advection of heat and moisture and atmospheric circulation as well (termed 'non-local effects', Pongratz et al 2021). Separating the local versus non-local effects in the model simulations is not a trivial task and different approaches have been proposed. Prior methods used to extract the local and non-local effects and assess their relative roles include: (a) a checkerboard approach of altered and un-altered grid boxes (Winckler et al 2017); (b) a comparison of coupled and offline simulations to isolate atmospheric feedbacks (Chen and Dirmeyer 2020); and (c) a series of numerical simulations to examine the far-reaching teleconnections (e.g. the global responses to boreal, temperate, and tropical deforestation, Devaraju et al 2018). This separation is essential, since coupled models capture the total (local plus non-local) climate response to deforestation. As the CMIP6 models do not provide the similar deforestation scenarios in the offline experiments or specific experiments (e.g. latitudinal band), we used the first method to estimate the local and non-local effects on global clouds. It is assumed that the non-local effect is similar over the un-deforested and near-by deforested boxes as done in many other studies (e.g. Zhou et al 2007, Pongratz et al 2021). We spatially interpolate the non-local signal to the neighboring deforested regions to obtain a global map of the non-local effect. The local effect thus can be obtained by subtracting the interpolated non-local effect from the total effect. As we need to fill in the deforested values with interpolated values while leaving the original values unchanged, we extrapolate the non-local signal in the core deforested boxes surrounded by deforested boxes by solving the Poisson's equation over the original values beyond data boundaries. It is worth noting that using a chessboard-like pattern (i.e. one out of two deforested boxes, the other for un-deforested boxes) may reduce the horizontal interpolation errors in the best possible way, as the local effects only rely on interpolation from directly adjacent un-deforested boxes. However, we used the idealized deforestation scenarios in this study (see section 4). Winckler et al (2017) have compared the idealized sparse and extensive deforestation simulations and found that the local effects using this checkerboard approach do not differ substantially. Therefore, we decided to use this approach to disentangle the local and non-local effects. More detailed descriptions and limitations of this approach can be found in Winckler et al (2017).

Deforestation across global land in CMIP6 models typically leads to an increase in surface albedo (figure 1(a)) and a concomitant decrease in evapotranspiration efficiency and roughness length (Davin and de Noblet-ducoudré 2010). These changes may affect the Earth's energy balance (radiative effect) and the partitioning of available net radiation into sensible and latent heat fluxes (non-radiative effect), especially in relation to the cloud formation and rainfall. The CMIP6 models agree that deforestation increases surface albedo over the tropical (e.g. Amazon, Congo and Southeast Asia) and temperate (e.g. North America and Eurasia) regions (figures 1(a) and S1). Since forests are more efficient in exchanging the available energy with the atmosphere through evapotranspiration (Bonan 2008, Runyan et al 2012), deforestation also alters the distribution of clouds (figure 1(b)) and thus influences the albedo at the top of the atmosphere (TOA) (figure 1(c)). The cloud cover changes due to deforestation have a strong latitudinal dependence, with decreased cloudiness in the tropics but with increased cloudiness in the temperate and boreal regions (figure 1(d)).

Zoom In Zoom Out Reset image size

Combined with the increase in surface albedo, the change in cloudiness may increase the TOA albedo and produce net radiative cooling, and vice versa (figure 1(c)). For example, deforestation together with increased cloudiness will reflect more solar radiation back to space (i.e. increased TOA albedo) and thus amplify the radiative cooling effect of the higher surface albedo in the temperate and boreal regions, indicating a cloud feedback there. In the tropics, the decrease in total cloud amount due to deforestation will partially cancel the higher surface albedo of deforested areas and result in a small positive TOA albedo change. In the subtropics except for East Asia, the forest fraction changes and cloud cover responses are small in models (figures 1(b) and S1).

Different types of clouds could have different effects on both surface and TOA energy balance. In general, low-level clouds typically have a cooling effect by strongly reflecting radiation back into space, while high-level clouds tend to have a warming effect (Boucher et al 2013, L'Ecuyer et al 2019). Another feature in climate models is the increased low-level cloud amount in the temperate regions in some CMIP6 models (figure S2). Our findings are consistent with previous notion that deforestation may produce mesoscale circulations that alter clouds and precipitation (Bonan 2008, Duveiller et al 2021).

The separation of local and non-local effects may help explain the biophysical impacts of deforestation on clouds. We first examine the temperature changes (figures 2, S3 and S4). When comparing neighboring forests and grasslands in observations, any non-local signals are minimized or removed, because advection of heat and moisture and atmospheric circulation between the neighboring regions are similar. Thus, the local signals modeled due to deforestation are comparable with that observed (Winckler et al 2017, Chen and Dirmeyer 2020). Please note that this assumption neglects the impacts of possible small regional circulation changes due to small-scale, heterogeneous deforestation, which may also alter clouds and precipitation (Bonan 2008, Khanna et al 2017). In general, some CMIP6 models (e.g. BCC-CSM2-MR, IPSL-CM6A-LR, MPI-ESM1-2-LR) show that deforestation can cause local summer warming over the deforested areas (figure 2), which is consistent with observational findings (figure S5). In contrast, some other models exhibit summer cooling in the temperate regions (figure 2), which may be due to the indirect effects of clouds or additional contributions of roughness length change to surface temperature in models (Devaraju et al 2018, Liu et al 2023).

Zoom In Zoom Out Reset image size

The above evaluation approach can be used for other variables, including clouds. Results show that the multi-model ensemble mean results agree on the sign of local cloud reductions due to deforestation, in particular in the tropics (figure 3(a)). The modeling results are also consistent with previous observational studies in the temperate and boreal regions (Duveiller et al 2021, Xu et al 2022), although the simulated signals are relatively weak. Furthermore, deforestation-induced local cloud changes mainly come from the contribution of low-level clouds (figures S6–8). For the local effects, decreased evapotranspiration associated with deforestation and increased sensible heat fluxes associated with local warming lead to drying of the troposphere and thereby inhibit cloud formation in the tropics. In the middle and high latitudes, changes in cloud cover are less pronounced and their sign depends on the region (figure 3(a)). The cloud cover responses to deforestation differ, depending on changes in sensible heat fluxes at different scales that are associated with convection and atmospheric boundary layer (Bosman et al 2019, Xu et al 2022).

Zoom In Zoom Out Reset image size

In addition to the local effects, the models can simulate deforestation-induced changes over or outside the deforested regions (e.g. via atmospheric teleconnections). In general, there is a non-local cloud enhancement in the temperate and boreal regions (figure 3(b)). Deforestation-induced temperature decreases and relative humidity increases in the middle and high latitudes would lead to relatively wetting of the lower atmosphere and an increase in cloud cover (Laguë and Swann 2016, Liu et al 2023). Note that the regions with significant cloud cover changes do not necessarily coincide geographically with large forest changes. The deforestation effects can propagate to regions that are remote from the deforested areas through the modulation of larger-scale circulations and moisture transport (Swann et al 2012, Devaraju et al 2015). Therefore, the large-scale atmospheric circulation adjustments due to deforestation can lead to remote changes in precipitation and cloud cover in the middle and high latitudes (Portmann et al 2022). Since the multi-model ensemble mean results exhibit a strong increase in clouds in the extratropical regions, the local effect (i.e. cloud inhabitation) is overwhelmed by the non-local effect over these regions (figures 3(a)–(c)). In the tropics, the signals are weak (figure 3(b)), indicating that models show inconsistencies in simulating the non-local effects (figure S9).

Our analysis of cloud distribution shows that deforestation has opposite effects on clouds between the tropical and extratropical regions in CMIP6 models. However, these opposite changes contradict the recent observational results (Xu et al 2022), in particular in the temperate and boreal regions (Duveiller et al 2021). Specifically, observational studies show reduced clouds in the temperate and boreal regions due to deforestation (Duveiller et al 2021, Xu et al 2022), but CMIP6 modeling results show opposite changes. There are some differences in the spatial and temporal deforestation patterns in observations and models. First, the impacts of forest on clouds are scale-dependent in observations (Li et al 2016, Li and Wang 2019). For example, small-scale (e.g. kilometer-scale) deforestation could trigger local thermal circulations that impact clouds and rainfall (Khanna et al 2017). The patchy, heterogeneous deforestation patterns drive changes in mesoscale circulation from forests into the atmosphere that can enhance rainfall and cloudiness over nearby deforested areas. However, the response between forest loss and rainfall enhancement may reverse when there is large-scale clearing (Leite-Filho et al 2021). Second, GCM studies have uncertainties typically limited by coarse spatial resolution and unrealistic representation of deforestation patterns. It is therefore more suitable for assessing regional to large-scale deforestation impact.

To examine the sensitivity of cloud responses due to the temporal variations in spatial extent of deforestation, we also analyzed the time series of cloud cover and rainfall changes in models (figures 4 and S10) as a function of linearly forest area loss (figure S11). As a consequence of linear forest loss, the multi-model ensemble mean results exhibit a strong linear local reduction in cloud cover over the globe and Northern Hemisphere boreal latitudes (figures 4(a) and (b)). The models have broadly reached the equilibrium by the end of the deforestation period (e.g. over a period of first 50 years with a linear loss of forest area). In addition to the large-scale changes, regional cloud responses to different amounts of deforestation are somewhat different (figures 4(c)–(f)). Over South America and Central Africa, the models simulate a strong linear decrease in cloud cover over time. In the boreal regions, the models show wide spread in simulating the cloud changes over North America and Eurasia (figures 4(e) and (f)). We also examine the rainfall responses to the linear deforestation and found a negative response of rainfall to the linearly fractional cover of forest loss (figure S10).

Zoom In Zoom Out Reset image size