Greening vegetation cools mean and extreme near-surface air temperature in China

Satellite observations have shown evident vegetation greening in China during the last two decades. The biophysical effects of vegetation changes on near-surface air temperature (SAT) remain elusive because prior studies focused on the effects on land surface temperature (LST). SAT is more relevant to climate mitigation and adaptation, as this temperature is experienced by humans. Here, we provide the first observational evidence of the greening effects on SAT and SAT extremes in China during 2001–2018 using the ‘space-for-time’ method. The results show a negative SAT sensitivity to greening (–0.35 °C m2 m–2) over China and a cooling effect of −0.08 °C on SAT driven by vegetation greening during the study period. Such a cooling effect is stronger on high SAT extremes, particularly over arid/semiarid areas, where greening could bring an additional cooling of −0.04 °C on the hottest days. An attribution analysis suggests that the main driving factor for the cooling effect of greening is the evapotranspiration change for arid/semiarid regions and the aerodynamic resistance change for humid regions. This study reveals a considerable climate benefit of greening on SAT, which is more concerned with natural and human system health than the greening effects on LST.


Introduction
Satellite observations have shown a widespread greening trend of vegetation at the global scale since the 1980s [1][2][3].Such a greening trend is driven by rising atmospheric CO 2 concentrations (e.g.CO 2 fertilization), climate change (e.g. the warming climate), and anthropogenic changes in land cover and nitrogen deposition [1][2][3][4].Meanwhile, increasing vegetation can produce biophysical feedbacks to local temperature [5][6][7].On the one hand, vegetation growth can enhance evapotranspiration due to the larger leaf area and aerodynamically rougher surface, leading to a cooling effect.On the other hand, greening vegetation can also cause a warming effect by lowering albedo [6][7][8][9][10].Greening is found to cause a net cooling effect globally, except in a few boreal regions [2,7,11].Therefore, understanding the biophysical feedbacks of greening to climate is critical for regional climate projection, adaptation, and mitigation [9,12].
Many studies have been dedicated to the impact of changes in vegetation cover on climate, such as afforestation/deforestation and crop cultivation [13][14][15].However, it should be noted that greening is not equal to vegetation cover changes, as the former reflects the increase in leaf area, but the latter reflects land cover type conversion.Such dramatic changes in vegetation types such as deforestation only occur in specific regions, while vegetation changes are widespread.Investigating climate effects of greening could be more constructive to develop better climate mitigation strategies for policymakers.
It should be emphasized that, to investigate the temperature effects of greening, earlier observationbased studies mostly used land surface temperature (LST) as the temperature metric because LST is readily available from satellite observations to support regional-and global-scale analyses [16].However, the biophysical effects of greening on near-surface air temperature (SAT) remain elusive.Although climate models can disentangle the greening effects on SAT [17][18][19], the model-based results are inevitably subject to imperfect model representations of biophysical processes of vegetation changes [17,20,21].
Compared to LST, SAT is a more widely used temperature metric in climate change research [17,18].SAT is also more relevant to human, animal, and terrestrial ecosystem health than LST, as this temperature is directly experienced by humans [22][23][24].SAT is closely related to LST but may substantially deviate from LST depending on both the atmospheric (e.g.cloud cover and wind speed) and surface (e.g.vegetation types) conditions [25][26][27][28].Studies targeting the biophysical effects of vegetation have revealed that LST and SAT may show contrasting behaviors in response to dramatic land cover change, such as deforestation [17-19, 23, 29].This result hints that the greening effects on LST shown in previous studies cannot be simply extrapolated to the effects on SAT.
Because temperature commonly follows a normal distribution, a small perturbation to mean temperature may lead to large changes in the tail distribution of temperature, namely, temperature extremes [30][31][32].In particular, extreme hot events (e.g.heatwaves) may cause fatal threats to human health and terrestrial ecosystem functions [33][34][35].Given the biophysical effects of greening on mean LST or SAT shown in previous studies, we infer that greening may also impact hot extremes, but the exact effect has rarely been investigated.Moreover, indices that quantify hot extremes are mostly based on SAT [22,35].This further requires us to use SAT, instead of LST, as the temperature metric to quantify the temperature effect of greening.
In this study, we provide the first observational evidence on the biophysical effects of greening on SAT and hot extremes in China using the 'spacefor-time' method.This method allows the biophysical effects of vegetation to be isolated from observations which are the result of a combination of all influencing factors.We focus on China because of its world-leading greening trend since the 21st century [1].China is home to more than 1.4 billion people, and the dense population is more vulnerable to hot extremes.Therefore, it is necessary and urgent to determine the impact of greening on near-SAT and hot extremes, which has implications for ongoing large-scale vegetation restoration and climate mitigation in this country.

Observational datasets
Monthly leaf area index (LAI) and albedo are from the global land surface satellite (GLASS) datasets at a 0.05 • × 0.05 • spatial resolution [16].These products have high quality and accuracy with longterm temporal coverage and high spatial continuity without missing pixels.LST and latent heat flux (LH) are from the moderate-resolution imaging spectroradiometer (MODIS) products, i.e.MOD11C3 version 6 for LST [36] and MOD16A2 version 6 for LH [37]; the LST and LH datasets both have a monthly temporal resolution.Land cover type is obtained from the MODIS product (MCD12C1, version 6.1) and updated annually [38].All MODIS products provide data at a 0.05 • × 0.05 • spatial resolution.Monthly sensible heat flux (SH) is from the FLUXCOM dataset [39], which merges the flux measurement from FLUXNET towers and remote sensing data through machine learning, and is provided at a spatial resolution of 0.0833 To reduce the uncertainty arising from the observational data of SAT, two independent gridded SAT datasets are used in this study.One dataset is the China meteorological forcing dataset (CMFD), a spatial-temporal continuous gridded dataset developed specifically for China [40].The CMFD dataset is produced through the fusion of remote sensing products, reanalysis datasets and in situ station data.The native spatial resolution is 0.1 • × 0.1 • for CMFD, and the 3 hour temporal resolution allows us to determine extreme heat events.CMFD also provides surface downward shortwave/longwave radiation fluxes and precipitation data at the same spatial and temporal resolutions.The other SAT dataset is obtained from Hooker et al [41] at a spatial resolution of 0.05 • × 0.05 • and a monthly timescale, hereinafter referred to as H18.This dataset was developed based on satellite remote sensing and weather stations and then extended to the global grid scale through geographic and climatic similarities.

Estimation of biophysical sensitivity of temperature to greening
To isolate the effect of greening, we apply a widely used 'space-for-time' method to calculate the biophysical sensitivity of temperature (including LST and SAT) to LAI [5,42,43].The advantage of this method is that it excludes the impact of natural climate variability or the long-term warming trend on vegetation growth.Specifically, for a given target pixel, the sensitivity of a certain variable ( dVar bio dLAI ) is computed from all available samples of the comparison between the target and nearby pixels in a spatial moving window.The background climate in the moving window is assumed to be essentially the same.In this study, the moving window is set to 0.5 • × 0.5 • (i.e. 10 × 10 pixels in H18 and 5 × 5 pixels in CMFD) according to previous studies [43,44].For data products with different spatial resolutions, we used linear interpolation to make them consistent with the spatial resolution of the SAT products.Then, we can obtain dVar bio dLAI for the target pixel as follows: where x and y indicate the LAI and variable (e.g.SAT and LST), respectively, and i and j are the geolocations of the target and nearby pixels within the moving window, respectively.We adopt the median value to avoid the impact of statistical outliers of the samples.The available samples in a moving window are selected by three criteria: the LAI difference is larger than 0.1 m 2 m −2 , the elevation difference is less than 100 m, and there is no difference in the majority land cover type.We calculate sensitivity only when the percentage of valid samples is higher than 20%.
A sensitivity value can be calculated for each month and finally averaged over the growing season (April to September) for each pixel.Finally, we aggregate dVar bio dLAI to 1 • × 1 • to smooth spatial heterogeneity.Since the accuracy of climate datasets (e.g.SAT) is relatively lower over the Tibetan Plateau, we exclude pixels with an average altitude above 4000 meters.
The uncertainty of sensitivity is usually estimated through the ensemble members in model experiments [7,45,46].As the space-for-time method excludes the impact of natural climate variability to vegetation change, we consider each individual growing season as the 'ensemble members' in this study.We computed the sensitivity dVar bio dLAI in each growing season during the study periods (that is, 14 values in H18 from 2003-2016 and 18 values in CMFD from 2001-2018).The uncertainty of the sensitivity is then estimated by the standard deviation of dVar bio dLAI calculated in different years.
The sensitivity represents the potential change in a variable due to one LAI unit increase.The change in a variable caused by the actual LAI change can be estimated as follows: where dVar bio dLAI is the time average of sensitivities of every year in a certain pixel.∆LAI is the actual LAI change in this pixel during 2001-2018.The trend of growing season LAI in this pixel is obtained through linear regression method.Then ∆LAI is estimated by multiplying the linear trend with time (18 years).
To characterize the spatial distribution of the temperature sensitivity to vegetation change, we conduct the analysis separately in arid and humid regions.The aridity index (AI) is used to measure moisture availability and is defined as the ratio of potential evapotranspiration to precipitation [47], expressed as follows: where PET is the annual potential evapotranspiration (mm yr −1 ), and P is the annual precipitation (mm yr −1 ).To calculate AI, PET is from the Climatic Research Unit Time Series (CRU TS; v.4.06) dataset [48] at a spatial resolution of 0.5 • × 0.5 • .P is from CMFD and is further resampled to grids of 0.5 • × 0.5 • to match the PET data in spatial resolution.In this study, arid and semiarid areas are classified as areas with AI values less than 0.65 [47,49,50].

Decomposition of biophysical sensitivity
We perform an attribution analysis on the SAT sensitivity to LAI ( dSAT bio dLAI ) based on the surface energy balance [7], expressed as follows: where f is an energy redistribution factor and is given by: where DSW is the downward shortwave radiation, DLW is the downward longwave radiation, LH is the LH flux, r a is the aerodynamic resistance, ρ is the density of air (1.21 kg•m −3 ), C p is the constant-pressure specific heat capacity of air (1,013 J kg −1 K −1 ), σ is the Stephan-Boltzmann constant (5.67 × 10 −8 W m 2 K −4 ), ε s is the land surface emissivity, and ε a is the effective atmospheric emissivity.
Therefore, the SAT sensitivity to LAI can be decomposed into the sensitivities of albedo ( dα bio dLAI ), LH flux ( dLH bio dLAI ), downward shortwave radiation ( dDSW bio dLAI ), downward longwave radiation ( dDLW bio dLAI ) and aerodynamic resistance ( dr abio dLAI ) to LAI.All these sensitivities can be estimated using the same 'spacefor-time' method (i.e.equation ( 1)).More details for deducing equation ( 4) can be found in the supplementary information.

Biophysical sensitivity of LST and SAT
Figures 1(a) and (b) display the biophysical sensitivity of SAT to LAI during the growing season in China from the H18 and CMFD datasets, respectively.Note that the H18 and CMFD datasets have different coverage times, i.e. 2003-2016 for H18 and 2001-2018 for CMFD.However, such a difference has little influence on the results, as the interannual variation in the sensitivity is negligibly small.The two datasets consistently show negative sensitivities of SAT to LAI changes in most regions of eastern China, suggesting the cooling effect of greening.Both datasets demonstrate that in the northern part of eastern China, such as inner Mongolia, the negative sensitivity is strongest and could exceed −1 • C m 2 m −2 .Figure 1(e) illustrates the biophysical sensitivity of LST to LAI during the growing season.The LST sensitivity is close to the SAT sensitivity in the spatial pattern but larger than the SAT sensitivity in magnitude.This suggests that the biophysical impacts of vegetation changes on SAT, in addition to LST, are also detectable from observations.
The H18 and MODIS datasets also suggest stronger sensitivity in arid/semiarid regions of China, which have been recognized as regions of strong land−atmosphere coupling [49,51,52].Figure 1(f) illustrates the regional mean sensitivity for arid/semiarid and humid regions.Both the MODIS LST and the H18 SAT show more negative sensitivity to greening in arid/semiarid regions.For the LST sensitivity from MODIS, the regional mean value is −1.1 ± 0.13 • C m 2 m −2 in arid/semiarid regions and −0.30 ± 0.03 • C m 2 m −2 in humid regions.Previous study estimated LST sensitivity as −0.5 to −0.8 • C m 2 m −2 in China using a global land surface model [45].The sensitivities estimated from our observational-based method show similar magnitude, which enhances the robustness of our results.For the SAT sensitivity from H18, the regional mean value is −0.54 ± 0.05 • C m 2 m −2 in arid/semiarid regions and −0.25 ± 0.02 • C m 2 m −2 in humid regions.In contrast, the SAT sensitivity calculated from CMFD is overall larger in magnitude in humid regions than in arid/semiarid regions, as there is also a higher SAT sensitivity in southwestern China (figure 1(c)).
Along with the impact of aridity, we further explore the biophysical sensitivity of different vegetation types by aggregating all IGBP vegetation types into four board types, including forest, other wooden vegetation (OWV), grassland, and cropland [42,45] (supplementary table S1).Both datasets show consistent results, suggesting that SAT shows higher sensitivity to vegetation change in grasslands in arid/semiarid regions than in humid regions.However, the results from the CMFD dataset show a higher SAT sensitivity than H18 in forest and OWV areas.In addition, there is a high degree of uncertainty about sensitivity in forest and OWV areas in arid/semiarid regions (figures 1(b) and (d)).This may provide possible explanation for the different results of the two datasets in the arid/semiarid and humid regions, as most forests are located in humid areas.
We further explore the impacts of actual vegetation changes on SAT and LST based on the estimated sensitivity.As shown in figure 2(a), China shows an overall increasing trend in LAI during 2000-2018.During this period, most greening regions show concomitant decreases in both SAT and LST (figures 2(a), (c) and (e)).SAT decreased by −0.06 • C and −0.08 • C due to greening during 2001-2018 from H18 and CMFD datasets, respectively.As a comparison, previous studies using models estimated that greening has induced SAT decreasing at −0.01 to −0.08 • C per decade in China [7,46,53].The changes in SAT estimated from both datasets are generally consistent in magnitude with the previous results.In addition to prior studies on large scale, we provide a detailed spatial distribution of greening-induced SAT change in China.
The areas of larger cooling trends are not necessarily the areas of higher SAT sensitivity to LAI, as the temperature change is also determined by the magnitude of actual vegetation changes.A larger decrease in SAT and LST (exceeding 0.4 • C) is found in northern, central, and southwestern China, where LAI exhibited a larger increasing trend (figure 2(f)).A few arid/semiarid regions (e.g.Loss Plateau in central China) also show larger decreases in LST and SAT due to the relatively higher LST and SAT sensitivities.The results from both datasets show that SAT decreases more in arid/semiarid areas than humid areas where the vegetation type is grassland and cropland.The strongest cooling in the arid/semiarid areas occurs in grasslands.The greatest decrease in SAT in humid regions occurs in OWV areas.In terms of the forest, the results of the two datasets differ considerably and have relatively high uncertainty (figures 2(b) and (d)).
As shown in figure 2(f), both the MODIS LST and the H18 SAT display a stronger cooling trend in arid/semiarid areas despite the smaller LAI changes.
Nevertheless, this result remains uncertain, as the CMFD dataset suggests that the SAT decrease is overall larger in humid regions than in arid/semiarid regions.The uncertainty may be addressed by the inconsistency of the results from the two datasets in the forest area.

Greening effects on hot temperature extremes
To further explore the biophysical effects of greening on hot extremes, we investigate the SAT sensitivity to LAI and the SAT changes driven by actual LAI changes as a function of SAT during the growing season.In this section, only the SAT from the CMFD dataset is used for analysis, as CMFD provides SAT on the daily and sub-daily scales.We first sort the daily SAT of the growing season in all years and then divide them into 10 bins according to the corresponding percentiles.Then, the SAT sensitivity to LAI and the SAT changes driven by LAI changes are calculated within each SAT bin.
As illustrated in figure 3(a), the mean SAT is generally more sensitive to vegetation change on hotter days over China.Greening could potentially cause a more intensive decrease in SAT on days with higher mean SAT.Specifically, the difference in the SAT sensitivity between the 90-100th percentile and 0-10th percentile of SAT reaches −0.10 • C m 2 m −2 .This phenomenon is more pronounced in arid/semiarid areas (figure 3(c)) but is absent in humid areas (figure 3(e)).For example, the SAT sensitivity within the 90-100th percentile of SAT in arid/semiarid areas is −0.49 which is two times larger than that within the 0-10th percentile of SAT.The higher SAT sensitivity brings an additional cooling of approximately 0.043 • C (figure 3(c)), suggesting an enhanced vegetation cooling effect of greening on hotter days in arid/semiarid areas of China.
We further perform the same analysis on daily maximum temperature (TX).Similar to the results of SAT, the TX sensitivity to LAI is stronger at higher percentiles over China, especially in arid/semiarid areas (figures 3(b) and (d)).Moreover, TX is more sensitive than SAT to vegetation change in arid/semiarid areas.A comparison of figures 3(c) and (d) shows that the TX sensitivity is slightly higher than the SAT sensitivity within each percentile for arid/semiarid areas (e.g.−0.51 • C m 2 m −2 for TX and −0.49• C m 2 m −2 for SAT at the 90-100th percentile).The results indicate that greening in arid/semiarid areas has an intensified cooling effect on extreme high temperatures compared to average temperatures.

Biophysical drivers of SAT sensitivity
To unveil the mechanisms for the SAT sensitivity to LAI, we attribute the SAT sensitivity to the nonradiative effects related to changes in LH and r a , the radiative effects related to albedo changes and the indirect climate feedback related to changes in DSW and DLW (see section 2.3).Each contributor is calculated independent to SAT.Each contributor is calculated from the corresponding dataset using the 'space-fortime' method, and the original datasets (including albedo, LH, SH, DSW and DLW) are independent of the SAT.As shown in figures 4(a) and (b), the sum of all contributors matches well with the sensitivity directly calculated from the SAT of both datasets, validating the decomposition results.For example, the rebuild sensitivity is −0.32 ± 0.02 • C m 2 m −2 As mentioned in the previous section, the SAT sensitivity is higher in arid/semiarid areas.This is mostly explained by the larger magnitude of the LH-related effect (figures 4(c) and (d)).The negative sensitivity induced by increasing LH could reach −0.60 • C m 2 m −2 in arid/semiarid regions, whereas the r a -induced sensitivity is almost constant with the regional average.In contrast, both radiative and nonradiative effects are smaller in magnitude in humid areas, with a larger contribution from the r a -related effect (figures 4(e) and (f)).The sensitivity induced by r a is approximately −0.07 • C m 2 m −2 in humid regions, contributing over 60% of the rebuild sensitivity.This result is in line with a previous study showing that the cooling effect of vegetation gradually decreases with increasing LAI [42].In addition, the effect of greening on r a is mainly caused by changes in vegetation structure and morphology, which vary little with temperature.This may explain why the cooling effect of vegetation in humid areas does not change significantly at higher temperatures.
Past studies of the climate effects of global greening using models have come to similar conclusions: in the mid-latitudes of the northern hemisphere, the nonradiative cooling effects of vegetation are stronger than the radiative warming effects [7,42,45,46,52,53].Different decomposition analyses based on the surface energy balance also show that this nonradiative cooling effect on SAT is mainly caused increases in LH and decreases in r a , although the relative magnitude of the two contributors remains controversial [7,45,46,53].Overall, compared with the results from the controlled experiments of models, the sensitivities estimated by the 'space-for-time' method show similar magnitudes but with detailed spatial distributions in China.Our results provide observational evidence for this conclusion and indicate that the main contributing terms may not be the same in arid and humid regions of China.
It should be noted that, as shown in section 3.1, the H18 and CMFD datasets show some contrasting results when arid/semiarid and humid regions are differentiated.Although the average results in China are similar in both datasets, the rebuild sensitivity from the CMFD dataset is overestimated in arid/semiarid areas but underestimated in humid areas (figures 4(b), (d) and (f)).Thus, the difference between arid/semiarid and humid regions derived from a single SAT dataset shown in sections 3.2 and 3.3 should be interpreted with caution.

Summary and discussion
Previous studies have suggested the cooling effects of vegetation on LST in China [52,54].In this study, we reveal an overall cooling effect of greening on near-SAT over China in the 21st century.The negative sensitivity of SAT to greening is observed throughout China except on the Tibetan Plateau (−0.38 ± 0.03 • C m 2 m −2 ), with higher SAT sensitivities in arid/semiarid areas in northern China (−0.54 ± 0.03 • C m 2 m −2 ).Owing to the most significant LAI increase, the cooling effect of greening on air temperature is larger in arid/semiarid regions.Moreover, we find that greening can cause a higher cooling effect on extremely high SAT than on mean SAT in arid/semiarid regions of China.Our results illuminate that the physical mechanisms of SAT response to LAI change are not the same in regions with different degrees of aridity.SAT shows less negative sensitivity to greening in humid areas, and the main driver is increasing surface roughness.The higher sensitivity of arid/semiarid regions is mainly attributed to nonradiative cooling effects, especially through the increased LH flux.In summary, this study provides the first observational evidence of the biophysical impacts of large-scale vegetation greening on SAT and SAT extremes in China, which reveals the potential climate benefits of greening due to the close relationship of SAT with humans and ecosystems.
As mentioned before, our results indicate that greening may induce stronger cooling effects on hot extremes in arid/semiarid regions.High-temperature extremes are one of the most influential disasters and have pervasive impacts on ecosystems, infrastructure, and human health [33,35,55].On the one hand, the arid/semiarid region covers the densely populated Beijing-Tianjin-Hebei urban agglomeration, where people are highly vulnerable to extreme events.On the other hand, the ecosystem in semiarid regions of China is fragile, and frequent and intense hot extreme events will have catastrophic impacts on local ecosystem functions [49,56,57].Warming in arid/semiarid regions is reported to be a dominant contributor to continental warming [58].Stronger negative sensitivity in arid/semiarid regions means that greening in these regions is expected to bring more temperature benefits to natural and human systems.In addition, the cooling effect of greening on hot extremes identified in this study provides new insights into the climate benefits of vegetation greening.
Our attribution analysis show that greening cools the local and regional climate more efficiently through enhanced evapotranspiration, especially in arid/semiarid regions, which has implications for natural climate mitigation.For example, vegetation restoration and conservation can be prioritized in arid/semiarid regions to provide more climate mitigation potential for vegetation through biophysical processes.Meanwhile, vegetation restoration and conservation in arid/semiarid regions can help to combat land degradation and desertification.However, the cooling effect of greening is achieved at the expense of consuming more terrestrial water in the form of evapotranspiration [59][60][61].Thus, water resource security also needs to be considered when large-scale revegetation (e.g.ecological programs) is conducted in semiarid regions.
Nevertheless, there are still some shortcomings in our study.It should be noted that the 'space-for-time' method used in this study assumes that neighboring grids share the same climatic background, and therefore excludes the atmospheric feedbacks of vegetation to temperature [62].Thus, our results only show the local effect of greening on SAT.However, greening vegetation in a certain pixel may cause non-local changes in circulation patterns, humidity and cloudiness that can feedback on local SAT [23,63].The limitation of this method has less impact when studying LST, which is logically more relevant to local surface properties.For SAT, the combined of local and nonlocal effects need to be explored in future, as SAT is susceptible to temperature advection [24,29,64].
To enhance the robustness of the results, we used two independent SAT datasets (H18 and CMFD) to evaluate the sensitivity.However, the results from these two sets of data products are not entirely the same.The sensitivity estimated from H18 is higher in arid/semiarid areas than in humid areas, while the sensitivity from CMFD behaves the opposite.We found that the disparity between the two datasets occurs mainly in the forest type.Differences in the production procedures of these two datasets may have contributed to the difference in results.H18 establishes statistical methods to predict SAT by MODIS LST [41].Therefore, the results of H18 SAT and MODIS LST have similar spatial patterns.The CMFD dataset, on the other hand, incorporates a large amount of station data in China based on reanalysis and remote sensing products [40].More in-situ station data may make CMFD different from H18 and then lead to inconsistent estimated sensitivity.Thus, more spatiotemporally continuous data may be needed in the future for further analysis.

Figure 1 .
Figure 1.Spatial distribution of biophysical sensitivity of temperature to greening in the growing season over the study period in China.(a) Biophysical sensitivity of surface air temperature (SAT) from the H18 dataset to the leaf area index (LAI).The violet line is the contour with an arid index of 0.65, which differentiates between the arid/semiarid and humid regions.The gray background with red lines as the boundary is the arid/semiarid area.(b) Sensitivity of SAT from H18 datasets on different vegetation types in arid/semiarid and humid regions, respectively.OWV: other wooden vegetation.(c) Same as (a) but for SAT from the CMFD dataset.(d) Same as (b) but for SAT from the CMFD dataset.(e) Same as (a) but for land surface temperature (LST) from the MODIS dataset.(f) Sensitivity of LST and SAT to LAI in arid/semiarid and humid areas over China, respectively.

Figure 2 .
Figure 2. Spatial distribution of the temperature induced by greening and the change in leaf area index (LAI) in the growing season over the study period in China.(a) Spatial map of the change in surface air temperature (SAT) from the H18 dataset induced by LAI change.The violet line is the contour with an arid index of 0.65, which differentiates between the arid/semiarid and humid regions.The gray background with violet lines as the boundary is the arid/semiarid area.(b) SAT changes induced by greening on different vegetation types in arid/semiarid and humid regions, respectively.OWV: other wooden vegetation.(c) Same as (a) but for surface air temperature (SAT) from the CMFD dataset.(d) Same as (b) but for SAT from the CMFD dataset.(e) Same as (a) but for land surface temperature (LST) from the MODIS dataset.(f) Temperature changes induced by greening and LAI changes in arid/semiarid and humid areas over China, respectively.(g) Spatial map of actual change in LAI during 2001-2018.

Figure 3 .
Figure 3. Biophysical sensitivity of daily mean surface air temperature (SAT) and daily maximum air temperature (TX) to greening and corresponding changes induced by variations in leaf area index (LAI) at different temperature percentiles.(a) Average biophysical sensitivity (blue bar) of SAT to LAI and variation in SAT induced by LAI change (red bar) as a function of percentiles of SAT over China.(b) Same as (a), but the y-axis denotes the sensitivity and variation in TX.(c) Same as (a) but for the average of arid/semiarid areas in China.(d) Same as (c), but for the sensitivity and variation of TX.(e) Same as (a) but for the average of humid areas in China.(f) Same as (e), but for the sensitivity and variation of TX.

Figure 4 .
Figure 4. Biophysical sensitivity of surface air temperature (SAT) to leaf area index (LAI) from the H18 and CMFD datasets ( dSATbio dLAI ) attributed to different surface processes over China.(a) dSATbio dLAI associated with changes in surface albedo, downward shortwave radiation (DSW), latent heat (LH), aerodynamic resistance (ra) and downward longwave radiation (DLW).The rebuild term is the sum of all previous contributors.(b) Same as (a) but from CMFD dataset.(c) Same as (a) but for arid/semiarid areas of China.(d) Same as (c) but from CMFD dataset.(e) Same as (a) but for the humid area of China.(f) Same as (e) but from CMFD dataset.