Reponses of Land Surface Albedo to Global Vegetation Greening: An Analysis Using GLASS Data

Abstract

Global vegetation greening during recent decades has been observed from various remote sensing data. The global and regional climate can be altered by an increase in carbon storage, as well as changes in land surface albedo (LSA) and evaporation. However, the LSA changes induced by global vegetation greening are still not clear, and contrasting responses of LSA to vegetation changes were reported in previous studies. In this study, we analyzed the LSA in response to global vegetation greening using the Global Land Surface Satellite (GLASS) data and a vegetation-induced LSA change model. The results showed that vegetation greening trends could be observed worldwide, which resulted in contrasting LSA responses at regional scales (LSA increased as leaf area index (LAI) increased, or LSA decreased as LAI increased). Moreover, these contrasting LSA responses to global vegetation greening were effectively explained by the albedo difference between a vegetation and soil background. The results provide new insights into the relationship between LSA changes and global vegetation dynamics, and can support recommendations for policies of vegetation protection, and large-scale afforestation and deforestation.

1. Introduction

The global vegetation greening during the past decades has been widely reported based on analysis of long-term data records of satellite observations [1]. As remote sensing can be used as a tool for monitoring the dynamics of terrestrial ecosystems, various satellite-derived vegetation proxies (e.g., normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), and leaf area index (LAI)) have been developed for determining the greening (increasing) and browning (decreasing) trends of global vegetation [2]. Although discrepancies can be found among different proxies and datasets, most of these studies showed significant and widespread greening trends in global terrestrial ecosystems [3]. The most dramatic greening trends can be found in the Northern Hemisphere, such as China, India, Europe, Sahel, North America, Brazil, and Siberia [4]. These vegetation greening trends are associated with climatic and environmental changes, as well as human activities, and can be explained by the CO₂ fertilization effects, followed by nitrogen deposition, climate change, and land cover change [5]. Furthermore, the global and regional climate can also be altered by changes in vegetation, such as mitigation of global warming by increasing of carbon sink, albedo-induced warming, and evaporation-driven cooling [6].

In the process of global vegetation greening, the carbon balance and hydrothermal budget of the terrestrial ecosystem can be altered through biogeochemical and biogeophysical mechanisms, such as changes in radiation and thermal fluxes, and in water vapor. However, the biogeophysical forcings are rarely well evaluated due to the challenges associated with their quantification [7]. Recently, the qualification of radiative and non-radiative forcing of biogeophysical processes has been recognized as an important issue in global climate change studies [8]. Among these biogeophysical processes, the land surface albedo (LSA) is the one of most important variables for evaluating the climatic effect of global vegetation greening [7]. It has been reported that large-scale deforestation and afforestation can result in significant changes in the LSA, and produce non-negligible contributions of radiative forcings to global climate change [9]. In the sixth assessment report (AR6) of the Intergovernmental Panel on Climate Change (IPCC), the radiative forcing induced by historical land use change was −0.20 ± 0.10 W m⁻² from 1750 to 2019, which indicated a cooling effect on the global climate [10].

In previous studies, the responses of LSA to vegetation changes have been investigated using remote sensing data and model simulations at regional and global scales [11,12,13,14,15]. In the modeling studies, the vegetation albedo was usually simulated with the structural parameters of vegetation canopy, albedo of vegetation and underlying surface, as well as other climatic variables (snow cover, snow depth, or temperature), and the major modeling methods of vegetation albedo are shown in

Table 1. Modeling methods of vegetation albedo at global and regional scales.

Modeling Method	Reference
${\alpha }_{bs}={\alpha }_c\left[1-\exp \left(\frac{-\omega \beta L_{sai}}{\mu {\alpha }_c}\right)\right]+{\alpha }_{g}exp\left(-\left(1+\frac{0.5}{\mu }\right)L_{sai}\right),$ ${\alpha }_{ws}={\alpha }_c\left[1-\exp \left(\frac{-2\omega \beta L_{sai}}{{\alpha }_c}\right)\right]+{\alpha }_{g}exp\left(-2L_{sai}\right)$, where ${\alpha }_{bs}$ and ${\alpha }_{ws}$ are the black-sky and white-sky albedo, respectively; ${\alpha }_c$ and ${\alpha }_g$ are the albedo of canopy and underlying surface, respectively; $\omega \beta$ is the upward scattering fraction, $\mu$ is the cosine of solar zenith angle, and $L_{sai}$ is leaf and stem area index (LSAI).	[21]
$\alpha ={\alpha }_{soil}·e^{-0.5·LAI}+{\alpha }_{canopy}·\left(1-e^{-0.5·LAI}\right)$, where $\alpha$ is the LSA of a pixel, $fpar$ is the fraction of absorbed photosynthetically active radiation (FPAR), and ${\alpha }_{soil}$ and ${\alpha }_{canopy}$ are the albedo of soil and vegetation canopy, respectively.	[22]
$\alpha =k_1+k_2\left(1-e^{-LAI}\right)+k_3\tanh \left(\frac{d}{k_4}\right)\left(e^{-k_{5}LAI}+\left[1-\frac{1}{1+e^{-k_6{T}^{Max}}}\right]\right)$, where $\alpha$, $LAI,$ $d$ and $T^{Max}$ are LSA, leaf area index (LAI), snow depth, and maximum daily temperature, respectively, and $k_1~k_6$ are coefficients of the regression model.	[23]

. Most of the studies based on remote sensing data reported that the LSA decreased with an increase in vegetation, or increased with a decrease in vegetation, especially when afforestation or deforestation occurred [9,16]. Li et al. [17] found that the global LSA showed a significant decreasing trend from 2002 to 2016, which was explained by the increase in vegetation and decrease in snow cover [18]. Zhai et al. analyzed the LSA trend in the Loess Plateau from 2000 to 2010, and found that the maximum reduction in LSA could exceed 0.050 during the peak season, which was attributed to the vegetation restoration and “Green for Grain” project [19]. However, Han et al. [20] reported an unexpected increase in LSA due to forest greening in China, and found contrasting responses of LSA to the vegetation index for forestry and grass lands. Therefore, the LSA changes induced by global vegetation greening are still not clear, and need to be systematically evaluated using satellite-derived datasets.

In this study, we analyzed the responses of LSA to global vegetation greening using the Global Land Surface Satellite (GLASS) data. First, the annual mean and trends of LSA and LAI from 2000 to 2019 were evaluated. Then, the contrasting responses of LSA to LAI changes were explored using a vegetation-induced LSA change model and albedo differences between vegetation and soil backgrounds. Finally, the LSA changes from 2000 to 2019 were simulated using the vegetation-induced LSA change model. The findings of this study can improve our understanding of the biogeophysical effects of global vegetation greening, and provide a useful reference for determining the influence of global vegetation greening on surface radiation budgets and global climate change.

2. Materials and Methods

2.1. Data

In this study, we used the global LSA, LAI, and clumping index (CI) data to investigate the responses of LSA to global vegetation greening during the recent decades. The GLASS LSA and LAI datasets version 4.0 [24,25] from 2000 to 2019 were employed for representing the spatiotemporal dynamics of the global LSA and LAI data. The GLASS LSA dataset version 4.0 was generated from the Advanced Very High Resolution Radiometer (AVHRR) and the Moderate Resolution Imaging Spectroradiometer (MODIS) data using the direct estimation algorithm [26,27] and statistical temporal filter [28], which provided global LSA data with a temporal resolution of 8 days, and spatial resolutions of 1 km and 0.05° [29]. The GLASS LAI dataset version 4.0 was generated from the AVHRR and MODIS data using the general regression neural networks (GRNN) method [30], with temporal and spatial resolutions the same as that of the GLASS LSA dataset. Compared to similar satellite-derived datasets, the GLASS LSA and LAI datasets are more spatiotemporally continuous, without data gaps caused by cloud coverage and obscuration [31]. To simplify the process of analysis, only the shortwave white-sky albedo and LAI climate modeling grid (CMG) datasets from 2000 to 2019, with a spatial resolution of 0.05°, and a temporal resolution of 8 days were used in this study. The monthly mean LSA and LAI data were then derived by averaging the original data within a month, and the corresponding relationships between date of year (DOY) and months are shown in

Table 2. The corresponding relationships between DOY and months.

Month	January	February	March	April	May	June	July	August	September	October	November	December
Date of year (DOY)	1	33	65	97	129	153	185	217	249	281	305	337
	9	41	73	105	137	161	193	225	257	289	313	345
	17	49	81	113	145	169	201	233	265	297	321	353
	25	57	89	121	153	177	209	241	273	305	329	361

. In addition, the global foliage CI product (500 m) derived from MODIS bidirectional reflectance distribution factor (BRDF) data was also used for describing the clumping effect of the vegetation canopy [32]. To maintain a consistent spatial resolution with the global LSA and LAI datasets used in this study, the global foliage CI data were aggregated into a spatial resolution of 0.05°, and assumed to be invariant over time from 2000 to 2019.

2.2. Methods

In this study, we analyzed the responses of LSA to global vegetation greening using a vegetation-induced LSA change model, trend analysis, and annual-mean calculations. The flowchart for analyzing the responses of LSA to global vegetation greening is shown in Figure 1.

2.2.1. Vegetation-Induced LSA Change Model

The variations of LSA are determined by the changes of land surface reflectance characteristics, and have connections with various climatic and environmental factors, such as the greenness of vegetation, LAI, soil moisture, surface roughness, and fraction of snow cover [16,33,34]. To determine the influences of global vegetation greening, a vegetation-induced LSA change model was needed for this purpose. If we assume that the land surfaces are covered with a vegetation canopy (e.g., forests, shrubs, grass, and crops) and soil backgrounds (or underlying surface of vegetation canopy, e.g., bare soil, sands and desert, inland water, and soil covered by litter or snow), and the nonlinear mixture effect is ignored, the LSA can be expressed as linear combinations of the albedo values of the vegetation and soil backgrounds [22]:

$$\alpha =fpar\cdot {\alpha }_{\mathrm{veg}}+\left(1-fpar\right)\cdot {\alpha }_{\mathrm{soil}}$$

(1)

where $\alpha$ is the LSA, $fpar$ is the fraction of absorbed photosynthetically active radiation (FPAR), and ${\alpha }_{\mathrm{veg}}$ and ${\alpha }_{\mathrm{soil}}$ are the albedo values of the vegetation and soil backgrounds, respectively. It should be noted that the soil background used in this study not only included bare soil, but also included the sands/desert, inland water, and soil covered by litter or snow.

According to the Beer–Lambert law and radiative transfer theory of a vegetation canopy, the relationship between the FPAR and LAI can be expressed as follows [22,35]:

$$fpar=1-e^{-\gamma \cdot \mathsf{\Omega }\cdot \mathrm{LAI}} ,$$

(2)

where $\gamma$ is the foliage CI, $\mathsf{\Omega }$ is the G-function, which stands for the gap probability of vegetation canopy at a given beam direction, and LAI is the LAI of the vegetation canopy. The G-function is a function of the leaf angle distribution (LAD), and when the leaves of the canopy are assumed to be uniformly distributed, the value of the G-function can be set to 0.5. In the previous studies, the CI was usually not considered and assumed to be 1, which may result in errors in simulations. In this study, a global foliage CI map was used for representing the clumping effect of the vegetation canopy.

According to Equations (1) and (2), the relationship between the LSA and LAI can be further expressed as:

$$\alpha =(1-e^{-\gamma \cdot \mathsf{\Omega }\cdot \mathrm{LAI}})\left({\alpha }_{\mathrm{veg}}-{\alpha }_{\mathrm{soil}}\right)+{\alpha }_{\mathrm{soil}}$$

(3)

If we assume that the ${\alpha }_{\mathrm{veg}}$ and ${\alpha }_{\mathrm{soil}}$ do not vary significantly over time, and the observations of monthly LSA and LAI from 2000 to 2019 can be obtained from the GLASS datasets, the values of ${\alpha }_{\mathrm{veg}}$ and ${\alpha }_{\mathrm{soil}}$ for a given pixel can be obtained by the linear least squares regression method. In this linear model, the monthly LSA ($\alpha$) is the dependent variable, the monthly FPAR ($1-e^{-\gamma \cdot \mathsf{\Omega }\cdot \mathrm{LAI}}$) is the independent variable, the albedo difference between vegetation and soil backgrounds (${\alpha }_{\mathrm{veg}}-{\alpha }_{\mathrm{soil}}$) is the slope of linear equation, and the albedo of soil backgrounds (${\alpha }_{\mathrm{soil}}$) is the intercept of linear equation. In this study, all valid monthly mean LSA and LAI data from 2000 to 2019 were used when there were at least three pairs of LSA and LAI data, and the absolute change of the FPAR was larger than 0.1. If the FPAR for a given pixel was always less than 0.1, the pixel was not covered by vegetation, and therefore the ${\alpha }_{\mathrm{soil}}$ was set to the mean value of the LSA, and ${\alpha }_{\mathrm{veg}}$ was set to 0.0. If the FPAR for a given pixel was always larger than 0.8, the albedo of the soil background could not be properly interpreted, and therefore the ${\alpha }_{\mathrm{soil}}$ was set to no data, and ${\alpha }_{\mathrm{veg}}$ was set to the mean value of the LSA. If the derived ${\alpha }_{\mathrm{veg}}$ was negative, the value of ${\alpha }_{\mathrm{veg}}$ was set to no data. The flowchart for obtaining the albedo of vegetation and soil background is shown in Figure 2.

When the LAI changes ($∆\mathrm{LAI}$), the corresponding responses of LSA ($∆\alpha$) can be derived as:

$$∆\alpha =(1-e^{-\gamma \cdot \mathsf{\Omega }\cdot ∆\mathrm{LAI}})\left({\alpha }_{\mathrm{veg}}-{\alpha }_{\mathrm{soil}}\right)+{\alpha }_{\mathrm{soil}}$$

(4)

2.2.2. Calculation of Annual Mean Values

In this study, the annual means of the LAI, LSA, and FPAR were calculated. Because the areas and solar radiation varied in different pixels, the weights were considered in the calculation of annual mean values. The area- and radiation-weighted annual means of the LAI, LSA, and FPAR were calculated as [11]:

$$\overline{LAI}=\frac{{\sum }_{i=1}^{12}{\sum }_{j=1}^n{A}_{j}LA{I}_{i, j}}{{\sum }_{i=1}^{12}{\sum }_{j=1}^n{A}_j}$$

(5)

$$\overline{\alpha }=\frac{{\sum }_{i=1}^{12}{\sum }_{j=1}^n{A}_j{F}_{i,j}{\alpha }_{i, j}}{{\sum }_{i=1}^{12}{\sum }_{j=1}^n{A}_j{F}_{i,j}}$$

(6)

$$\overline{FPAR}=\frac{{\sum }_{i=1}^{12}{\sum }_{j=1}^n{A}_{j}FPA{R}_{i, j}}{{\sum }_{i=1}^{12}{\sum }_{j=1}^n{A}_j}$$

(7)

where $\overline{LAI}$, $\overline{\alpha }$, and $\overline{FPAR}$ are the annual means of the LAI, LSA, and FPAR, respectively; $i$ stands for the number of each month (from 1 to 12); $j$ stands for the number of each pixel (from 1 to $n$ (total number of pixels)); $A_j$ is the area of pixel $j$; and $F_{i,j}$ is the downward solar radiation of pixel $j$ and month $i$. In this study, the monthly weights of downward surface solar radiation were retrieved from the atmospheric radiative kernel derived using the off-line Community Earth System Model-Community Atmosphere Model version 5 (CESM-CAM5) [36].

2.2.3. Trend Analysis

To evaluate the changes of the LAI, LSA, and FPAR, a Theil–Sen trend analysis was performed. The trends of the LAI, LSA, and FPAR for a given pixel or region from 2000 to 2019 can be obtained by [37]:

$$T=\mathrm{Median}\left(\frac{X_j-X_i}{j-i}\right) \forall j>i$$

(8)

where $T$ is the trend derived by the Theil–Sen trend analysis, $X$ is the LAI, LSA, or FPAR data, $i$ and $j$ are the data indices ($j>i)$, and the total number of data points is $n$. When the trend is positive ($T>0$), the variable shows an increasing trend; when the trend is negative ($T<0$), the variable shows a decreasing trend. Herein, the trend analysis results were examined by Mann–Kendal statistics, and only the trend values at a significance level of 0.05 (95% confidence) were retained in the results.

3. Results

3.1. Annual Mean and Trends of LSA, LAI, and FPAR

The spatial distributions of the annual mean LSA, LAI, CI, and FPAR from 2000 to 2019 are shown in Figure 3. The LSA exhibited higher albedo values (0.3–0.8) over the perennial or seasonal snow-covered areas and deserts, and lower albedo values (0.0–0.2) over the vegetated areas and inland water, which showed a negative correlation with the LAI and FPAR. The spatial patterns of the LAI and FPAR were quite similar, and showed higher values (4.0–6.0 for LAI, and 0.6–1.0 for FPAR) over tropical regions, and gradually decreased as the latitude increased. The CI showed higher values (0.8–1.0, indicating a nearly random distribution) over the sparsely vegetated areas, and lower values (0.4–0.8, indicating clumped distribution) over the densely vegetated areas. It should be noted that the FPAR data were derived from the GLASS LAI data, CI map, and the vegetation-induced LSA change model, and the spatial distribution of the FPAR was slightly different than that of the LAI due to the clumping effect.

The spatial distribution of trends of the LSA, LAI, and FPAR from 2000 to 2019 are shown in Figure 4. Significantly decreasing trends of the LSA were found in Alaska and northwestern Canada, Siberia, the Loess Plateau, and Antarctica, and significantly increasing trends of the LSA were found in southeastern China, Kazakhstan, Saudi Arabia, Sahel, and Brazil. In contrast, the LAI showed significantly increasing trends in most of the vegetated areas globally, especially in southeastern China, the Loess Plateau, India, Europe, sub-Saharan Africa, Siberia, North America, and Brazil. Only a few areas, such as northern Brazil, Kazakhstan, and parts of sub-Saharan Africa, showed significantly decreasing trends in the LAI due to land cover and land use changes. The spatial pattern of the FPAR trend was quite similar to that of the LAI trend, which showed significant greening trends globally. It is interesting that contrasting responses of the LSA to the LAI and FPAR changes were found at regional scales. For example, the LSA of southeastern China increased significantly as the LAI and FPAR increased, whereas the LSA of Siberia decreased significantly as the LAI and FPAR increased, which needed to be explained by the vegetation-induced LSA change model.

3.2. Albedo Differences between Vegetation and Soil Backgrounds

To determine the responses of LSA to global vegetation greening, the relationships between monthly LSA and FPAR were investigated pixel by pixel, and contrasting responses of LSA to FPAR changes can be found at different geographic locations. To illustrate this phenomenon, two pixels located in southeastern China (117.2° E, 28.0° N) and Siberia (95.5° E, 53.3° N) were selected for this purpose, and the scatter plots between the monthly LSA and FPAR at these two pixels are shown in Figure 5. The scatter in this figure stands for the monthly LSA against the corresponding monthly FPAR at pixel scale. The monthly LSA and LAI datasets from 2000 to 2019 were employed in this study, with a total number of 240 scatters if all data are available. According to Equation (3), the intercept of the regression line is soil albedo (${\alpha }_{\mathrm{soil}}$), and the slope of the regression line is the albedo difference between the vegetation and soil background (${\alpha }_{\mathrm{veg}}-{\alpha }_{\mathrm{soil}}$). When the albedo difference is positive, the LSA increases as the FPAR and LAI increase; when the albedo difference is negative, the LSA decreases with the FPAR and LAI increases. If we use this method to explore the relationship between the LSA and FPAR pixel by pixel, the spatial distributions of the responses of LSA to FPAR and LAI changes can be obtained.

The spatial distributions of the albedo values of the vegetation and soil backgrounds derived from the vegetation-induced model are shown in Figure 6. The spatial pattern of the albedo values of vegetation was similar to that of the LSA, which showed higher albedo values over snow-covered areas and deserts, and lower albedo values over vegetated areas. Higher vegetation albedo values were found in Western Europe, southeastern China, southeastern United States, sub-Saharan Africa, South America, and coastal regions of Australia, and much lower values were found in Eastern Europe, Siberia, northeastern China, northern areas of North America, and the southern edge of the Sahel Desert. The albedo difference between the vegetation and soil background (${\alpha }_{\mathrm{veg}}-{\alpha }_{\mathrm{soil}}$) is a critical variable for determining the responses of the LSA to vegetation changes. When the albedo difference is positive, the albedo of the vegetation is greater than the albedo of the soil background, and the LSA increases with vegetation greening; in contrast, when the albedo difference is negative, the albedo of the vegetation is lower than the albedo of the soil background, and the LSA decreases with vegetation greening. From the albedo differences derived in this study (Figure 7), we found that positive albedo differences were found in southeastern China, Western Europe, sub-Saharan Africa, Brazil, and the southeastern United States. In these areas, the LSA increased with vegetation greening, which can be seen from the spatial distribution of the LSA trend from 2000 to 2019 (Figure 4a). Negative albedo differences were found in Siberia, Eastern Europe, the northern part of North America, and the southern edge of the Sahel Desert. In these areas, decreasing LSA trends were found as the vegetation greening increased, which showed a contrasting response of the LSA compared to the areas with positive albedo differences. The spatial distribution of the coefficients of determination are shown in Figure 8. It can be seen that the vegetation-induced LSA change model performs well with respect to most vegetation-covered areas.

3.3. Responses of LSA to Vegetation Changes

The spatial distribution of the LSA response categories from 2000 to 2019 is shown in Figure 9. The category in which the LSA increased as the LAI increased was found in southeastern China, southern Brazil, India, and Western Europe. The category in which the LSA decreased as the LAI decreased was seldom observed. The category in which the LSA increased as the LAI decreased was found in eastern Brazil and the southern edge of the Sahel Desert, which indicated that deforestation or land cover changes occurred. The category in which the LSA decreased as the LAI increased was found in the Loess Plateau, Siberia, Alaska and northwestern Canada, Pakistan, southeastern United States, northern Brazil, and Eastern Europe. Most of these changes were explained by the albedo differences between the vegetation and soil background, as well as the vegetation-induced LSA change model. The only exception is the southeastern United States, in which the response type was not well explained by the albedo differences, mainly because the change in the LSA in this area was small, and more easily affected by other non-vegetation factors.

The temporal variations of LSA, LAI, and LSA simulated by the vegetation-induced LSA change model in six regions are shown in Figure 10. Southeastern China (Figure 10a) and southern Brazil (Figure 10f) were in category where the LSA increased as the LAI increased; the Loess Plateau (Figure 10b), Siberia (Figure 10c), and Alaska and northwestern Canada (Figure 10d) were the category where the LSA decreased as the LAI increased; and the southern edge of the Sahel Desert (Figure 10e) was in the category where the LSA increased as the LAI decreased. For most of the vegetated areas, the vegetation-induced LSA change model reproduced the LSA trends from 2000 to 2019, although the non-vegetation factors, such as variations of snow cover and soil moisture, land cover changes, and changes in vegetation and soil background albedo values could not be accurately represented by this model.

4. Discussion

In this study, the spatial distribution of contrasting responses of LSA to global vegetation greening were found and explored using GLASS data. In previous studies, the LSA of vegetation was considered to be much darker than that of barren land and desert, and it was usually assumed that the LSA would decrease as the LAI and FPAR increased [20]. This is mainly because most of these studies focused on the LSA change induced by land cover changes, such as deforestation and urbanization, in which the LSA would increase with these land cover changes [8,9,16,38]. However, Han et al. [20] reported an unexcepted increase in LSA due to forest greening in China, and contrasting responses of the LSA to the EVI were also found in forest and grassland greening. The results of this study confirmed this phenomenon and extended the analysis worldwide using the long-term GLASS LSA and LAI data. The spatial distribution of the albedo differences between the vegetation and soil background, as well as the vegetation and soil albedo climatology, were also obtained by this study.

The LSA changes induced by global vegetation greening were effectively explained by the vegetation-induced LSA change model and the albedo differences between the vegetation and soil background. The spatial distribution of the four LSA response categories from 2000 to 2019 were derived using the GLASS data, and most of these LSA changes over vegetated areas were simulated by the vegetation-induced LSA change model.

The results of this study indicate that the LSA is altered by global vegetation greening, and the radiative forcing induced by the LSA changes cannot be neglected. In the future, the warming and cooling effects caused by the vegetation-induced LSA changes should be considered in global climate change and carbon neutrality studies.

The value of LSA is largely determined by the geophysical parameters of land surfaces (e.g., land cover types, vegetation greenness, vegetation structure, soil surface roughness, soil moisture, fraction of snow cover, snow grain size, soot concentrations of snow, and snow depth). For most circumstances, the properties of snow, vegetation, and soil are the most important driving factors of LSA, and the importance of these driving factors varies with geographical locations [39]. However, the LSA can also be significantly influenced by many indirect factors, such as air temperature, precipitation, human activities, and atmospheric circulation. Therefore, it is important to model the relationship between LSA and these indirect factors, which can be used to project the variations of LSA in future scenarios.

Although the vegetation-induced LSA change model and albedo differences were used for explaining the contrasting responses of the LSA to global vegetation greening, there are still several issues that need to be addressed in the future:

(1)

The vegetation-induced LSA change model used in this study is relatively simple, and can be improved by consideration of the temporal cycle of vegetation and soil background albedo values, as well as the effect of snow cover. In this study, the vegetation and soil background albedo values were assumed to be temporally invariant, which is not consistent with the true situation, and can result in an estimation error of the albedo difference between vegetation and soil background. Although the effect of snow cover can be simply presented by the variation of soil background albedo values, the role of snow cover in LSA changes, such as the vegetation masking effect on snow cover [40], still need to be explored in the future.

(2)

The spatiotemporal trends of the LSA derived in this study may not be perfectly consistent with former studies at regional scales [17,18,31], mainly because of the differences in temporal ranges and datasets. Our previous [18] study suggested that the temporal span is critically important for analyzing LSA trends, and the trend results can be significantly affected by the temporal spans and selection of the datasets. Therefore, longer temporal spans and more robust LSA datasets are still needed for this purpose [41].

(3)

In this study, the responses of LSA to global vegetation greening were investigated using the long-term remote sensing datasets and a vegetation-induced LSA change model. Although the accuracies of these satellite-derived datasets have been validated and evaluated by various in situ observations [29,30], the findings of these studies still need to be further validated with ground measurements in the future.

(4)

In addition, the vegetation-induced LSA change model can be used for projecting future LSA changes in different shared socioeconomic pathway (SSP) and representative concentration pathway (RCP) scenarios, and the climatic effect of vegetation-induced LSA changes can also be evaluated.

5. Conclusions

In this study, the responses of LSA to global vegetation greening were analyzed using GLASS data and a vegetation-induced LSA change model. The main findings of this study are: (1) Contrasting responses of LSA to global vegetation greening were discovered by this study, and the spatial distribution of different LSA response types was determined; and (2) the LSA changes over most of the vegetated areas were effectively explained by the vegetation-induced LSA change model and albedo differences between the vegetation and soil background. (3) The LAI trends at regional scales can be reproduced by the vegetation-induced LSA change model, although the temporal cycle of vegetation and soil background albedo values, and the influences of snow cover, need to be carefully considered in future studies. The results of this study provide new insights for the relationship between LSA changes and global vegetation dynamics, and can support recommendations of policies for vegetation protection, and large-scale afforestation and deforestation.