Discrepant trends in global land-surface and air temperatures controlled by vegetation biophysical feedbacks

Satellite-based land surface temperature (Ts) with continuous global coverage is increasingly used as a complementary measure for air temperature (Ta), yet whether they observe similar temporal trends remains unknown. Here, we systematically analyzed the trend of the difference between satellite-based Ts and station-based Ta (Ts–Ta) over 2003–2022. We found the global land warming rate inffered from Ts was on average 42.6% slower than that from Ta (Ts–Ta trend: −0.011 °C yr−1, p = 0.06) during daytime of summer. This slower Ts-based warming was attributed to recent Earth greening, which effectively cooled canopy surface through enhancing evapotranspiration and turbulent heat transfer. However, Ts showed faster warming than Ta during summer nighttime (0.015 °C yr−1, p < 0.01), winter daytime (0.0069 °C yr−1, p = 0.08) and winter nighttime (0.0042 °C yr−1, p = 0.16), when vegetation activity is limited by temperature and solar radiation. Our results indicate potential biases in assessments of atmospheric warming and the vegetation-air temperature feedbacks using satellite-observed surface temperature proxies.


Introduction
The Paris Agreement motivated the long-term goal of 'holding the increase in the global average temperature to well below 2 • C above pre-industrial levels' , to avoid potential detrimental consequences of future warming on human society and ecosystems [1][2][3].The global warming rate is routinely assessed with near-surface air temperature (Ta), for its high relevance to human life, health, and production [4,5].Furthermore, international climate mitigation actions to achieve this goal, such as energy system transformations [6] and ecological restoration [7,8], are almost all formulated around potential effects on Ta.
Ta measurements are routinely available at meteorological stations at 2 m height above the ground surface [9,10].However, the uneven distribution of meteorological stations, especially the scarcity of station observations in remote regions (e.g. the Arctic, deserts, and tropical rainforests), makes it a challenge to map global air temperatures and their longterm dynamics [10,11] Remote sensing derived land surface temperature (Ts), on the other hand, overcomes the limitation of station-based Ta data and has provided temperature measurements with a high spatial resolution, high geolocation accuracy, and continuous global coverage since 2000 [12][13][14].Ts and Ta are tightly coupled from sub-daily [15] to inter-annual timescales [10,13,14], with a temporal F Kan et al correlation usually higher than 0.9.Therefore, Ts is widely used as an alternative of Ta in various aspects of Earth system research, such as the detection of heat waves [16,17] and urban heat island intensity [18][19][20], as well as the biophysical feedbacks of vegetation change on surface temperatures [21,22].
Over vegetated land surfaces, satellite-derived Ts captures the spectral radiance reflected mainly by the vegetation canopy, where the exchanges of energy, water, and momentum between the land surface and the atmosphere occur [9,23,24].Hence, Ts can directly and instantly respond to changes in canopy biophysical properties, such as albedo and transpiration [25,26].By contrast, Ta is driven by the energy partition at the surface-atmosphere interface, as well as warm/cold air mass advection and fine-scale turbulent mixing.These distinct differences in governing processes between Ts and Ta could lead to considerable discrepancies in the values of the two temperature indicators at various spatial scales [10,15,27].For instance, there has been observational evidence that Ts is remarkably higher (>10 • C) than Ta in mid-latitude arid areas, but slightly lower than Ta in tropical rainforests and high latitudes [27].
In response to external perturbations, variations in Ts and Ta also decouple over decadal to longer-term timescales in some locations [28][29][30].For example, a faster warming of Ts than that of Ta was observed over eastern Brazil Cerrado due to the deforestation and desertification in these areas [30].Another station-based study reported an increase in the difference between Ts and Ta in northern China during winter, which was attributed to the increase of snow depth [29].Furthermore, a global-scale modeling study revealed that deforestation had a more substantial local effect on tropical warming and highlatitude cooling when assessed with Ts than with Ta [31].However, observational evidence is lacking in terms of how differently Ts and Ta may change over decadal timescales over broad regional and global scales.Such difference suggested possible uncertainties in incorporating Ts into assessing regional and global warming and its effects.So far, the available satellite Ts observations provide an unprecedented opportunity to compare the trend difference between Ts and Ta (denoted hereafter as Ts-Ta) over the recent two decades.The primary objectives of our study were to (1) examine whether Ts increases at the similar rate as Ta over the past two decades; (2) identify the key surface biophysical processes driving potential differences in the decadal trends of the two temperature measures.

Satellite-based Ts and station-based Ta
Mathematically, the difference between the linear trends of Ts and Ta (the Ts trend minus the Ta trend) is equal to the linear trend in the difference between Ts and Ta (trend in Ts minus Ta, or Ts-Ta).A positive (negative) Ts-Ta trend indicates that the land surface warms faster (slower) than its surrounding air.In order to analyse the trend difference on seasonal cycle globally, here we defined spring as MAM (March-April-May) in NH (Northern Hemisphere) and SON (September-October-November) in SH (Southern Hemisphere), summer as JJA (June-July-August) in NH and DJF (December-January-February) in SH, autumn as SON in NH and MAM in SH, winter as DJF in NH and JJA in SH.In this study, monthly global Ts maps (0.05 • × 0.05 • , 2003-2022) were derived from the MODIS sensor onboard the Aqua satellite platform (MYD11C3 v6) [32] (text S1).The monthly daytime (maximum) and nighttime (minimum) Ta were obtained from the Climatic Research Unit (CRU) v4.07 data set [33], with a spatial resolution of 0.5 • and covering the 1901-2022 period (text S2).

Observations of environmental factors
The environmental variables used in the study were the Enhanced Vegetation Index (EVI), snow cover, precipitation (Pre), cloud cover, soil moisture (SM), and elevation.The monthly 0.5 • × 0.5 • EVI data were derived from MODIS sensors on board the Terra satellite (MOD13C2 v006) [34].Monthly snow cover data were obtained from ERA5-land reanalysis data with a spatial resolution of 1 • [35].Monthly Pre and cloud cover data were provided by the CRU v4.07 product [33].Surface SM estimates, with a spatial resolution of 0.25 • , were derived from the Global Land Evaporation Amsterdam Model v3.7b [36,37].A 0.5 • × 0.5 • global terrain elevation map was retrieved from the Global Multi-resolution Terrain Elevation (GMTED2010) [38].

Boosted regression trees (BRTs)
We applied the BRT model (text S3) to quantify the relative contributions and partial dependency of environmental factors toward the inconsistent warming between Ts and Ta.The approach was conducted separately with combinations of three latitudinal groups and two seasonal groups.The latitudinal groups were northern extratropics (>23.5 • N), tropics (23.5 • N-23.5 • S), and southern extratropics (<23.5 • S); the seasonal groups were summer and winter for northen and southern extratropics and annual mean for tropics.In these experiments, we included the elevation and trends in other environmental factors (EVI, Pre, snow cover, cloud cover, and SM) as predictors, and the trends in Ts-Ta as the response, using all available 0.5 • × 0.5 • land gird cells.

Quantifying the effects of various biophysical process
Based on the surface energy balance equation, changes in Ts-Ta can be analytically expressed as a F Kan et al function of a series of biophysical attributes including albedo (α), evapotranspiration (ET or λE), surface incoming shortwave radiation (S in ), air longwave radiative emissivity (ε a ), aerodynamic resistance (r a ) and ground flux (G) (details in texts S4 and S5).Given that the aerodynamic datasets used for deriving r a were updated only up to 2017 [39] and the spatial resolution of radiation datasets derived from CERES was 1 • [40], the attribution to biophysical attributes was conducted only for the period of 2003-2017 at 1 • × 1 • gird cells.
However, the real-world conditions limit the use of analytical model of Ts-Ta trend.For example, the increase in soil evaporation (a key ET component) in a warmer climate will not influence the canopy temperature.Hence, we decomposed Ts-Ta trend (∆Y) into additive contributions of the five observed biophysical processes using a multi-variate linear regression (equation ( 1 where dY dt (or dX dt ) represents the linear trend of Y (or X) from 2003 to 2017.∂Y ∂X represents the sensitivity of Y to a biophysical factor X (α, ET, S in , εa, ra or G).The sensitivities were calculated as the regression coefficients of a multiple linear regression analysis performed with the trend in Ts-Ta against the trends of all listed explanatory variables for 2003-2017.The trend in Ts-Ta was decomposed into contributions of each variable X (∆Y X ), which was denoted as the product of the sensitivity against variable X ( ∂Y ∂X ) and the concurrent trend in X ( dX dt ).ϵ represents the discrepancy between observed and predicted trends, which indicates the unexpained variations of the Ts-Ta trend by the considered biophysical factors.This approach was applied to each grid cell in a 3 × 3 (3 • × 3 • ) running window; the total contribution of each factor to the trend of Ts-Ta was calculated by averaging the decomposed contribution of factors (∆Y X ) across all global (or zonal) vegetated land surfaces.

Observed annual and seasonal trends of Ts-Ta
We first examined the long-term trend in the difference between satellite-based Ts and station-based Ta for their overlapping period of 2003-2022.We considered both daytime (Ts: 1:30 PM, Ta: maximum, approximately 2:00 PM) and nighttime (Ts: 1:30 AM, Ta: minumum, approximately 6:00 AM) differences.To minimize the uncentainties arising from the time gap between 1:30 AM and 6:00 AM, we calculated the difference between Ta at 1:30 AM and daily minimum Ta using ERA5-Land reanalysis data (figure S1), and then added this difference to the minimum Ta from CRU. Results suggested highly different signs and magnitudes in the trends of Ts-Ta between daytime and nighttime (figure 1).Globally, the annual mean daytime Ts-Ta presented an insignificant negative trend (-0.0015 • C yr −1 , p = 0.704), whereas there was a significant positive trend in nighttime Ts-Ta (0.0116 • C yr −1 , p < 0.01).Hence, the land surface warmed slower than its surrounding air for daytime but much faster for nighttime.Furthermore, the trend in Ts-Ta was −5.9% (daytime) and 55.0% (nighttime) of the contemporary trends in Ta, respectively.This finding suggested substantial biases when using Ts for the estimation of global air temperature changes.
The seasonal investigation also revealed asymmetric trends of Ts-Ta both within diurnal and annual cycles.For daytime, the Ts-Ta trend displayed a progressive shift from negative in summer to positive in winter; whereas for nighttime, the trend was constantly positive throughout the four seasons (figure 1).The day-night contrast in their Ts-Ta trends was strongest during summer, when the trends in daytime (-0.011 • C yr −1 , p = 0.06) and nighttime (0.015 • C yr −1 , p < 0.01) Ts-Ta were quantitatively similar but opposite in sign (figure 1).Hence, the summer warming based on Ts was much smaller (−42.6%)than that based on Ta for daytime, but was much larger (64.9%) for nighttime (figure 1).During winter, the trend discrepancies between Ts and Ta were consistent for daytime (0.0069 • C yr −1 , p = 0.08) and nighttime (0.0042 • C yr −1 , p = 0.16); the warming rate determined by Ts was 21.4% and 14.8% greater than that by Ta in daytime and nighttime, respectively.The contrasting seasonal patterns in daytime versus nighttime Ts-Ta trends implied a potential role of terrestrial vegetation in alleviating the differences in Ts-and Ta-based warming, considering more vigorous vegetation activities in summer (at noon) with the abundant solar radiation.
The four major climatic zones based on the Köppen-Geiger climate classification all exhibited pronounced day-night contrast (figure S2).During daytime, arid and boreal regions have observed strong negative Ts-Ta trends in summer, and temperate regions have observed moderately negative Ts-Ta trends throughout the year.During nighttime, all climate zones excluding boreal regions in winter have observed significantly positive Ts-Ta trends, with the most pronuonced signal detetcted for arid regions.The day-night contrast in Ts-Ta trends was robustly detected in arid and temperate regions within the 0-35 • N latitudinal band, whereas for other regions, this contrast can be detected only for certain seasons (figure S3).For example, in boreal regions (35-72 • N), daytime and nighttime Ts-Ts trends showed opposite signs in summer, with consistent signs or no detectable changes in other seasons.We next investigated the spatial patterns of Ts-Ta in summer and winter-the two seasons with the most extreme Ts-Ta trends as revealed in figure 1regarding both its multi-year mean and long-term trend (figure 2).In general, the mean and trend of Ts-Ta were highly variable both spatially and temporally because of the heterogeneity in land cover types and environmental drivers.Daytime Ts-Ta exhibited a weakening seasonality from the high latitudes to tropics.In the high latitudes with a generally negative mean daytime Ts-Ta during winter, Ts increased faster than Ta, except for eastern North America and eastern Siberia.In the mid-low-latitude arid or semi-arid regions, Ts increased slower than Ta during summer, attenuating the positive contrasts between Ts and Ta in these regions.During winter, however, the regional dominance of negative Ts-Ta trend changed toward more positive, amplifying the surface-air temperature contrast.In the tropical rainforests, Ts was consistently lower than Ta throughout the year, which remained largely unchanged except for eastern Amazon and the Congo basin.We detected a faster regional warming trend with Ts than with Ta in these tropical regions, where intense deforestation occurred over the past 20 years [41,42].This finding suggested that previously reported warming effect of tropical deforestation [43,44] with Ts might be overestimated.On the other hand, compared with the daytime changes in Ts-Ta, we found a weak seasonality of nighttime Ts-Ta.During nighttime, the land surface was cooler but warmed faster than the air for a large fraction of the global land surface (61.8% for summer and 48.6% for winter).
We also tested the robustness of our results by including MODIS Terra LST product (MOD21C3 v061) [45] and the reanalysis-based Ta from the ERA5 database [35] (figure S4).Ta at 10:30 AM

Environmental drivers of the divergent temperature trends
To understand the mechanisms of observed spatial variations in the differences between Ts and Ta, we employed a machine learning model that related the trend in Ts-Ta to a set of environmental factors including elevation, trends of the EVI, Pre, snow cover, cloud cover, and SM (figure 3; see Methods).We identified an apparent inconsistency in the dominant driver of the daytime Ts-Ta trend across both seasons and latitudinal zones (figure 3(a)).For example, the increasing EVI was the dominant driver for the Ts-Ta trend in the tropics, southern latitudes and during summer in the northern latitudes, while the decreasing snow cover dominated the Ts-Ta trend during winter in the northern latitudes (figures 3(a), S5(a) and S6(c)).In all cases, except for winter in the northern latitudes, a higher EVI (related to increased vegetation cover and/or enhanced vegetation growth) tended to attenuate the Ts-Ta trend (figure S7(a)).This attenuation could be attributed to the cooling of vegetation canopy surface through evapotranspiration [46] and turbulent heat transfer [21,47], leading to slower warming of the land cover surface than the surrounding air [27].In cold seasons, the presence of snow cover with high albedo (reduced shortwave radiation absorption) and thermal emissivity (increased outward longwave radiation) resulted in slower warming in Ts rather than Ta (figure S8(c)).Other environmental factors played minor roles in regulating the Ts-Ta trend.For example, increased SM slowed warming at the land surface more than in the air in all cases, because wetter soils facilitated evapotranspiration and heat convection.Increased cloud cover amplified surface warming more than the air during both summer and winter in northern latitudes by enhancing the downward longwave radiation; however, it attenuated surface-air warming discrepancies by reducing the downward shortwave radiation reaching the surface in the tropics and southern latitudes.
In the attribution of the nighttime Ts-Ta trend, we additionally included daytime Ts-Ta (as a proxy for daytime heat storage) as a potential factor [48].We found that the trend in nighttime Ts-Ta was determined by combined influences of various factors including daytime Ts-Ta.In general, the land surface absorbed and stored solar energy during the day, which was then released during the night.Hence, faster warming in Ts than Ta during daytime would increase the amount of stored energy and heat, which, when released during nighttime, could warm the surface more than the air in northern latitudes, particularly during winter (figures 3(b) and S8(g)).In both NH and SH, the nighttime trend in Ts-Ta was less sensitive to its daytime trend in summer than in winter, presumably because vegetation dissipated the absorbed solar energy more effectively in summer.Other environmental drivers, such as elevation, Pre and cloudiness, also exerted considerable influence on the nighttime Ts-Ta trend.For example, Pre negatively affected the nighttime Ts-Ta trend in NH, because increased water supply for plant transpiration could reduce the solar energy absorbed by the land during daytime.Nevertheless, Pre positively affected the nighttime Ts-Ta trend in the tropics and NH, because increased rainfall moistened the air and enhanced downward longwave radiation emitted by the atmosphere during nighttime.

Biophysical processes underlying the diverging temperature trends
Our analysis of environmental factors identified EVI as a key driver of the inconsistent warming between Ts and Ta, which involves a series of biophysical processes in the surface energy budget.To further understand how vegetation modulates surface warming, we decomposed the observed trend in Ts-Ta to contributions from different surface biophysical attributes (i.e.α, ET, S in , εa, ra, G) using multivariate linear regression.Together, these processes adequately reproduced the trend in Ts-Ta during 2003-2017 and its spatial variations (figures 4 and S11).It is noteworthy the changes of time period of Ts and Ta introduced difference to the results in figure 2. For example, in Cerrado of Brazil, the remarkably positive Ts-Ta trend during 2003-2017 was attenuated when during an extended period of 2003-2022 (figures S10 and S4(i), (j)), because the deforestation in this region has decelerated in recent years.Below we elaborated in detail the biophysical attributes for the inconsistency in daytime warming because changes in solar shortwave radiation and plant transpiration were almost negligible during nighttime.
During the summer, ET-related changes in the Ts-Ta trend were the most significant among all the surface biophysical processes considered, driving slower land surface warming than air warming with a mean difference of −0.0102 • C yr −1 (72.34% of the overall trend; figure 4(a)).The dominance of ETrelated effects spread over middle latitudes (20-60 • N and 20-40 • S), especially in global greening hotspots or regions with intensive land management practices (figures 4(c), S9(c) and S12(a)).For example, in north China where large-scale ecological restoration projects were implemented [49], and in central USA where agricultural expansion and intensification occurred [50].The contributions of ET became weak during the cold-season dormancy (figure 4(b)), with an overall positive impact (0.0019 • C yr −1 ) on the Ts-Ta trend, mainly in the tropical latitudes (15 • N-15 • S).This amplifying effect of ET on the Ts-Ta trend prevailed over the regions with a pronounced browning signal (figures 4(d), S10(c) and S12(b)), such as the arc of deforestation in Amazon [51,52] and the eastern Africa with reoccurring droughts [53].The contrast between summer and winter reinforced the critical role of ET in regulating daytime temperature changes.This seasonality was consistent with our result demonstrating EVI changes as the dominant enviromnental driver in summer (figure 3).Given that plant transpiration accounts for nearly 2/3 of terrestrial ET [54][55][56][57] and also represents the primary contributor to observed ET changes (figure S13), our results provide compelling evidence that vegetationdriven ET changes controlled the surface-air warming divergence.The other turbulent process ra, which regulated sensible and/or latent heat transfer from the land surface to the atmosphere through aerodynamic resistance [21], had limited impact on daytime Ts-Ta trend in both summer (−0.0002 • C yr −1 ) and summer (0.0005 • C yr −1 ).
Biophysical processes controlling the net solar radiation available at the surface-through alterations to α, S in , and εa-overall had a weak impact globally (−0.0027 • C yr −1 ) during summer, together contributing 18.8% to the slower Ts-based warming compared with Ta-based warming (figure 4(a)).Among these factors, S in caused a sizable decrease in Ts-Ta (−0.0045 • C yr −1 ) in summer, which was partly offset by εa (0.0003 • C yr −1 ) and α (0.0015 • C yr −1 ).During winter, however, these radiative processes contributed to 95.9% of the higher Ts-based warming rate, compared with Tabased warming (0.0094 • C yr −1 ) (figure 4(b)).These processes combined served as the primary driver for contributing to the accelerated warming of Ts relative to Ta during winter (α: 0.

Conclusions and discussion
In summary, our analysis revealed inconsistent trends between Ts derived from satellite observations and Ta derived from surface meteorological stations.We found that Ts observations produced slower warming during daytime but faster warming during nighttime than Ta, particularly in growing seasons.These contrasting behaviors on diurnal and seasonal timing were related to the greater sensitivity of Ts to Earth's greening related to either direct (i.e.afforestation) or indirect (rising CO2) human activities [49,59,60].In particular, we found that the changing vegetation biophysical attributes, mainly ET, had central roles in the daytime decoupling of warming between the land surface and the air (figure 4).During warm-season daytime, Earth greening tended to cool the land (canopy) surface more strongly than air, mainly by elevating transpiration rates.However, during dormant season or nighttime when vegetation activity is limited by low temperatures and availability of solar radiation, the very weak transpiration led to opposite behaviors of the Ts-Ta trend (figure 1).The descrepancies of Ts-Ta trend between daytime and nighttime were strongest in summer.However, this day-night contrast diverged across climate zones during winter.Moving from warm ecosystems to cold ecosystems, the winter Ts-Ta trend shifted from a simialr pattern to summer to an opposite pattern to summer.This geographical pattern of seasonal transition is likely because the diurnal cycle of surface temperature in cold ecosystems is controlled mainly by snow cover, in contrast to warm ecosystems controlled mainly by vegetation activities [61][62][63].
Our quantification of the discrepancies in surface-air warming has certain limitations that should be acknowledged.First, considerable uncertainties exist in the station-based Ta data over remote regions, such as the Arctic, drylands and mountainous areas, due to the paucity of available meterological stations.Given the low signal-to-noise ratio, the pronounced contrast of Ts-and Ta-based warming rates observed, for example, in summer of arid regions, should be interpreted with caution.Second, meteorological stations are typically situated on unchanged landscapes, which means that values of locations undergoing land cover changes were extrapolated from values of unchanged locations.This extrapolation could, to a certain degree, dilute the signal of temperature change related to land cover changes.Third, satellite-observed Ts captures the spectral radiance at diverse heights, e.g. from the bare land surface in deserts to canopies exceeding 10 m in tropical forests.The varying vertical heights of Ts relative to Ta could underlie their differing trends across space.Fourth, the physiological response of vegetation to rising CO2, including partial stomatal closure and enhanced water use efficiency, is challenging to detect in observational records.This physiological forcing-induced reduction in ET partly offset the increase in ET caused by CO2 fertilization and longer growing season, with the potential to amplify surface warming [64].Fifth, our assessment of the Ts-Ta trend based on linear methods may not capture potential nonlinearity of temperature changes in the real world, especially for locations intervened strongly by human activities.For example, the recent deceleration of deforestation in eastern Brazil has slowed down the pronounced positive trend in Ts-Ta observed before 2018.Last, although our empirical models based on priori knowledge of environmental drivers and biophysical parameters could well reproduce the spatial pattern of Ts-Ta trend and are insensitive to model parameter adjustments, they lack the capability to definitively establish causality between the Ts-Ta trend and associated environmental drivers.Another issue is that empirical models could not fully disentangle the intricate interplay among various factors, such as the ET-water vapor-cloud-radiation feedback loop.To tackle these challenges, we encourage the use of coupled land-atmosphere models to verify our observation-based signal, for example, by comparing paired numerical simulations with and without inclusion of vegetation forcing (e.g.LAI and land cover changes).
While the identified divergence between Ts and Ta trends in our work called for caution in applying satellite-based Ts in climate change detection and impact assessment.We found that Ts cannot be regarded as a reliable alternative measure for Ta in some circumstances.For example, in summer daytime, the global warming rate assessed with the Ts was almost 40% lower than that assessed with Ta; hence, the use of satellite-based Ts could strongly underestimate the rate of near-surface air warming.Particular cautions should be exercised in investigating vegetation biophysical feedbacks on global warming using Ts from remote sensing observations as well.Satellitedriven assessments that compared the Ts of forest canopy and neighboring no-forested open lands have shown that forest cover gain would cool the surface in tropical regions but warm the surface in cold regions [22,65,66].However, our results demonstrated that the use of Ts might overestimate the effect of greening to mitigate air warming in tropical regions and amplifying air warming in the Arctic because of the stronger sensitivity of Ts to ET-related cooling effect and α-related warming effect.This has implications for climate impact assessment of large-scale afforestation/reforestation practices that are viewed as key natural solutions to combat climate change.We argue that the biophysical effect of tree planting to cool near-surface air may not be as strong as suggested by previous Ts-based assessments.
Remotely sensed Ta and ground-based Ts are both indispensable temperature measures in use in climate change research.On the one hand, they respond F Kan et al interactively to anthropogenic radiative forcings and show similarity in many ways.On the other hand, they also provide independent and complementary information of the Earth system about warming, which has not been well recognized in climate change and risk assessment.Compared with Ta relevant to heat stress and thermal exposure of humans [67], Ts is a more appropriate variable for examining changes in ecosystem biophysical attributes as its higher sensitivity to vegetation activity changes.Our results showed that canopy surface temperature would not increase in a manner similar to the surrounding air temperature in warm seasons.This is physically controlled by land-atmosphere interactions, but can also be interpreted as a self-adapting measure of vegetation to provide a buffer against surrounding air warming.Nevertheless, it remains unclear whether this response of vegetation to rising temperatures will continue with future climate change, particularly under growing abiotic stresses (e.g.human activities, wild fires, and droughts).Satellite measurements of canopy surface temperature (Ts), which responds to environmental factors almost simultaneously, can potentially provide real-time monitoring of ecosystem heat stress and an early warning of heat damage.

FFigure 1 .
Figure 1.Trends in global mean Ts-Ta.Histograms showing a linear trend in annual and seasonal (spring, summer, autumn, winter) mean Ts-Ta during daytime and nighttime, observed over the 2003-2022 period.The pink pentagrams represent the Ts-Ta trend as a percentage of the Ta trend in all cases, whether they were significant (filled) or not (hollow).Percentage are not shown for nighttime autumn because the Ta trend was negligible, resulting in sufficiently large ratio that exceeded the right-y-axis range.Error bars indicate the uncertainty ranges (1-standard deviation).* * p < 0.05; * * p < 0.1.

( 1 :
30 PM) and 10:30 PM (1:30 AM) from ERA5 were obtained to match the overpass time of Terra (Aqua).During winter, the three dataset groups (T Aqua s distribution of the Ts-Ta trends for both daytime and nighttime.However, during summer, compared with T Aqua s − T CRU a , T Aqua s − T ERA5 a observed at 1:30 PM showed a consistent pattern, whereas T Terra s − T ERA5 a observed at 10:30 AM showed a contrasting pattern.This result indicateed that varying mechanisms drived the surface-air warming divergence at 1:30 PM and 10:30 AM, with stronger surface cooling at noon due to the more vigorous plant transpiration than that in the morning.

Figure 2 .
Figure 2. Spatial pattern of the multi-year mean and long-term trends of Ts-Ta.(a), (c), (e), annual, summer, and winter mean daytime values, respectively.(b), (d), (f), annual, summer, and winter mean nighttime values, respectively.Horizontal axis of the color legend shows the long-term trend, where positive (negative) values represent faster (slower) increases in Ts than Ta.The vertical axis of the color legend shows the multi-year mean, where positive values represent a higher Ts than Ta; negative values represent a lower Ts than Ta.Greenland and Antarctica are excluded from the analysis.

Figure 3 .
Figure 3. Environmental controls on the spatial variations of the Ts-Ta trend.The relative importance of six environmental factors (Enhanced vegetation index, EVI; precipitation, Pre; snow cover; cloud cover; soil moisture, SM; elevation for the spatial variations of trends in daytime (a) and nighttime (b) Ts-Ta.The variable importance (percentage of explained variance) was assessed separately for different latitudinal bands (>23.5 • N, north; 23.5 • N-23.5 • S, tropics; <23.5 • S, south) and seasons (summer and winter for north and south, annual mean for tropics), based on the boosted regression tree (BRT) model.For the nighttime Ts-Ta trend, the corresponding daytime value, representing the changes in daytime energy storage, was included as the seventh explanatory variable.Grid size was determined by the importance of each factor; grid shapes indicate the positive (rectangle) or negative (ellipse) sensitivity of the Ts-Ta trend on each factor.Details of the seasonal definitions can be found in figure 1.
0039 • C yr −1 , S in : 0.0015 • C yr −1 , εa: 0.0039 • C yr −1 ).The αrelated effect was dominant over middle latitudes (25-45 • N) (figures 4(d) and S10(a)) because of the decreased snow cover over North America and Europe [58] (figure S12(d)) and afforestation over temperate regions during cold seasons [48] (figure S12(b)).The εa-related effect prevailed in northern high latitudes (50-70 • N) (figures 4(d) and S10(d)), where the increased εa associated with the larger cloud cover or greater atmospheric water vapor enhanced the amount of downward longwave radiation warming the land surface.Positive S in -related changes were detected over low latitudes (10 • N-30 • S) (figures 4(d) and S10(b)) because a increase in S in would lead to more amount of shortwave solar radiation available to warm the land surface.Apart from the turbulent and radiative processes, ground heat flux exerted a weak but nonnegligible impact on the surface-air warming decoupling.The transfer of surface absorbed energy to deep soils through G mitigated Ts warming more than Ta in both summer and winter (summer: −0.0013 • C yr −1 , winter: −0.0022 • C yr −1 ).