The Sensitivity of Vegetation Dynamics to Climate Change across the Tibetan Plateau

: Vegetation dynamics are key processes which present the ecology system’s response to climate change. However, vegetation sensitivity to climate change remains controversial. This study redeﬁned vegetation sensitivity to precipitation (VSP) and vegetation sensitivity to temperature (VST) by the coefﬁcient of determination (R 2 ) obtained by a linear regression analysis between climate and the normalized difference vegetation index (NDVI), as well as by using an analysis of variance to explore the signiﬁcant differences between them in different seasons from 1982 to 2013, and exploring the general changed rules of VSP/VST on a timescale. Moreover, the variations in VSP and VST across the Tibetan Plateau were plotted by regression analysis. Finally, we used structural equation modeling (SEM) to verify the hypothesis that the respondence of VSP and VST to the NDVI was regulated by the hydrothermal conditions. Our results showed that: (1) the annual VSP increased in both spring and winter (R 2 = 0.32, p < 0.001; R 2 = 0.25, p < 0.001, respectively), while the annual VST decreased in summer (R 2 = 0.21, p < 0.001); (2) the threshold conditions of seasonal VSP and seasonal VST were captured in the 4–12 mm range (monthly precipitation) and at 0 ◦ C (monthly average temperature), respectively; (3) the SEM demonstrated that climate change has signiﬁcant direct effects on VSP only in spring and winter and on VST only in summer (path coefﬁcient of − 0.554, 0.478, and − 0.428, respectively). In summary, our ﬁndings highlighted that climate change under these threshold conditions would lead to a variation in the sensitivity of the NDVI to seasonal precipitation and temperature.


Introduction
The vegetation community is the main component of ecosystems [1]. As far as we know, vegetation dynamics, taken as a barometer of global ecological change, are mainly affected by climatic factors [2]. Climate change affects vegetation dynamics by altering vegetation phenology, plant community composition, and biogeochemical cycles [3][4][5]. Therefore, for sustainable development, it is crucial to assess how the vegetation dynamics respond to climate change [6][7][8].
Current climate change is characterized as a continuous rise in temperature and frequent extreme precipitation, and the varying rules of vegetation productivity are mainly attributed to the temporal distribution of precipitation and temperature [9][10][11] because of the excellent match between soil nutrient availability and the temporal variability of the hydrothermal environment [12][13][14][15]. Previous studies have demonstrated that the increased annual precipitation significantly limited the growth of shrub grassland in the Tibetan Plateau [16]. Meanwhile, studies in Inner Mongolia have shown that the seasonal distribution of precipitation has directly led to the inter-annual fluctuation in grassland community productivity [17]. However, beyond that, vegetation dynamics are also governed by temperature change [18]; to a certain extent, the increased mean annual temperature could satisfy the heat supply required by the normal growth of plants, and it has also changed the microclimate environment of the plant community, thus directly or indirectly affecting the growth and development of plants and biomass production [19][20][21].
On a large scale, satellite observations have been applied to the quantitative description of vegetation growth patterns already [22,23], and the normalized difference vegetation index (NDVI) based on the global vegetation index has been routinely used to monitor vegetation growth at a global scale [24]. The peak NDVI value synthetically reflects the photosynthetic activity of plants under current environmental conditions [25,26]. Hence, the NDVI could be used to explore the response of vegetation sensitivity to the spatial pattern of hydrothermal conditions and inter-annual variations on a spatiotemporal scale [27]. For example, Piao et al. [28] considered that the NDVI in arid and semi-arid areas of Western China showed an upward trend due to the change in climate from warm-dry to warm-wet during 1982 to 1999. Similarly, from 1982 to 2011, the average annual NDVI of China also increased by about 0.0006 [29]. However, another study showed a continuous downward trend of NDVI with climate change in Northeast China [30]. Others have noted that the responses of vegetation dynamics to climate change are related to the vegetation region [31].
The Tibetan Plateau is highly sensitive to climate change [32][33][34]. Numerous studies have noted the significant influence of the inter-annual variability of precipitation on community composition and that vegetation production is sensitive to the timing and size of precipitation inputs [35][36][37]; increased precipitation variance will also affect the aboveground net primary production (ANPP). In contrast, Hsu et al. [1] hold that the ANPP is approximately 40 times more sensitive to mean precipitation than to the inter-annual variance in precipitation. As for the variation in temperature, Fu et al. [38] found that the sensitivity of phenology to warming has declined significantly as temperatures have increased over the past three decades, and it may be beneficial to the vegetation as it reduces the risk of late spring frost damage by avoiding the premature unfolding of leaves. All of these studies have attempted to decompose and assess the sensitivity of vegetation productivity to changes in climate variables [39], improving our understanding of climatevegetation relationships; nevertheless, the annual variation in vegetation sensitivity to temperature and precipitation, as well as the relationships between the NDVI and seasonal patterns of temperature and precipitation [40,41] have not been quantified.
Therefore, to better manage grassland ecosystems under global climate change, longterm time series of grassland NDVI and climate  were used to analyze the dynamic effect of climate on the NDVI in the Tibetan Plateau and hypothesize that the respondence of VSP and VST to the NDVI is regulated by the hydrothermal conditions. Specifically, the objectives of our study were to: (1) reveal the temporal variation in vegetation sensitivity to precipitation (VSP) and temperature (VST); (2) quantify the hydrothermal threshold conditions between vegetation sensitivity and climate change.

Study Area
The Tibetan Plateau (26 • 00 -39 • 47 N, 73 • 19 -104 • 47 E), located in southwestern China (Figure 1), is the highest and most extensive plateau in the world, with an average altitude exceeding 4000 m [42]. The typical continental plateau climate is characterized by low temperatures and limited precipitation; the winter is long, cold, and windy, whereas the summer is cold and rainy, with frequent hail [43]. The temperature and precipitation have distinct regional distribution patterns in this area, with a mean annual air temperature (MAT) ranging from −15 to 20 • C and mean annual precipitation (MAP) ranging from 50 to 700 mm from the northwest to the southeast [44]. More than 63.5% of the area of the Tibetan Plateau is covered by alpine desert steppe, alpine steppe, and alpine meadow. Alpine grassland is the dominant ecosystem on the plateau, and the vegetation is alpine meadow (dominated by Kobresia and Poa species) and alpine steppe (dominated by Stipa purpurea and Stipa subsessiliflora) [45]. The other vegetation types on the Tibetan Plateau include forest, shrub, and wetland.
Atmosphere 2022, 13, x FOR PEER REVIEW 3 of 14 the summer is cold and rainy, with frequent hail [43]. The temperature and precipitation have distinct regional distribution patterns in this area, with a mean annual air temperature (MAT) ranging from −15 to 20 °C and mean annual precipitation (MAP) ranging from 50 to 700 mm from the northwest to the southeast [44]. More than 63.5% of the area of the Tibetan Plateau is covered by alpine desert steppe, alpine steppe, and alpine meadow. Alpine grassland is the dominant ecosystem on the plateau, and the vegetation is alpine meadow (dominated by Kobresia and Poa species) and alpine steppe (dominated by Stipa purpurea and Stipa subsessiliflora) [45]. The other vegetation types on the Tibetan Plateau include forest, shrub, and wetland.

Normalized Difference Vegetation Index Data
The longest time series (1982-2013) GIMMS NDVI3g dataset, with a spatial resolution of 0.083°, were compiled by merging segments (data strips) during a half-month period using the maximum value composites (MVC) method. It was obtained from the AVHRR instrument onboard the NOAA satellite series 7, 9, 11, 14, 16, and 17 [26] and downloaded from ECOCAST (https://ecocast.arc.nasa.gov (accessed on 6 February 2022)). These data were corrected, including calibration, view geometry, and volcanic aerosols, and verified with a stable desert control point [46]. Meanwhile, the annual 24 NDVI raster database was processed in ArcGIS 10.2 (ESRI, Inc., Redlands, CA, USA) to obtain the maximum NDVI value every year.

Meteorological Records
Daily climate data  were collected from the Meteorology Information Centre of the Chinese National Bureau of Meteorology (http://data.cma.cn(accessed on 26 February 2022)) [26]. The primary climatic elements of temperature and precipitation were included, and the daily temperature and precipitation were processed to obtain the climate factors every year (annual mean temperature and annual mean precipitation) and for spring (March-May), summer (June-August), autumn (September-November), and winter (December-February). Moreover, the climatic raster database was spatially interpolated using ANUSPLIN 4.2 (Centre for Resource and Environmental Studies, Australian National University, Canberra) [47,48], and resampled with a resolution of 8 km. ANUSPLIN is a suite of FORTRAN programs that employs thin plate smoothing splines

Normalized Difference Vegetation Index Data
The longest time series (1982-2013) GIMMS NDVI3g dataset, with a spatial resolution of 0.083 • , were compiled by merging segments (data strips) during a half-month period using the maximum value composites (MVC) method. It was obtained from the AVHRR instrument onboard the NOAA satellite series 7,9,11,14,16, and 17 [26] and downloaded from ECOCAST (https://ecocast.arc.nasa.gov (accessed on 6 February 2022)). These data were corrected, including calibration, view geometry, and volcanic aerosols, and verified with a stable desert control point [46]. Meanwhile, the annual 24 NDVI raster database was processed in ArcGIS 10.2 (ESRI, Inc., Redlands, CA, USA) to obtain the maximum NDVI value every year.

Meteorological Records
Daily climate data (1982-2013) were collected from the Meteorology Information Centre of the Chinese National Bureau of Meteorology (http://data.cma.cn (accessed on 26 February 2022)) [26]. The primary climatic elements of temperature and precipitation were included, and the daily temperature and precipitation were processed to obtain the climate factors every year (annual mean temperature and annual mean precipitation) and for spring (March-May), summer (June-August), autumn (September-November), and winter (December-February). Moreover, the climatic raster database was spatially interpolated using ANUSPLIN 4.2 (Centre for Resource and Environmental Studies, Australian National University, Canberra) [47,48], and resampled with a resolution of 8 km. ANUSPLIN is a suite of FORTRAN programs that employs thin plate smoothing splines fitted to develop continuous climatic surfaces with elevation and noisy weather station data [49].

Calculation of Vegetation Sensitivity (R 2 )
Over the Tibetan Plateau, fishnet points (30 rows × 60 columns) were established with ArcGIS 10.2 software (ESRI, Inc., Redlands, CA, USA), and non-vegetated land such as lakes and snowy mountains were removed by the digital land-use dataset (vector) at a scale of 1:100,000 developed by the Resources and Environment Data Centre, Chinese Academy of Sciences (CAS) [50,51]. Meanwhile, the values of temperature, precipitation, and the NDVI in these fishnet points were extracted from the climate and NDVI raster points, respectively. Then, a linear regression analysis of the climate and NDVI was used to reflect the sensitivity (R 2 ) of the climate on the NDVI [52]. The sensitivity of the climate on the NDVI can be defined as Equation (1); a similar method was used to obtain the sensitivity between seasonal climate and the NDVI: where y i is the actual value of the NDVI and Y i is the fitted value of the NDVI.

Sensitivity Analysis
We used an analysis of variance (ANOVA) in R (Psych package) to explore the significant differences in VSP (and VST) in different seasons collected from 1982 to 2013. Then, the changed rules of VSP (and VST) in different seasons on a timescale were analyzed. Meanwhile, over the Tibetan Plateau, a regression analysis was performed between seasonal VSP/VST and seasonal mean precipitation/temperature from 1982 to 2013. Finally, we introduced a dummy variable (an increasing time sequence) for climate change referring to the annual increased precipitation and temperature and used structural equation modeling (SEM) to verify the hypothesis that the variations in VSP and VST are regulated by certain hydrothermal conditions. Structural equation modeling (SEM) is a multivariate technique that involves computer algorithms and statistics; it has been used in recent studies to explicitly evaluate the causal relationships among multiple interacting variables [53]. We used it to test the direct and indirect effects on the variations in VSP and VST and to describe the hypothetical causal relationships [54]. By selecting the appropriate variables and models based on certain statistical criteria [55], the standard estimate results express the influence on the seasonal VSP (and VST) using a path coefficient generated by the AMOS statistical tool (17.0.2, Amos Development Corporation, Crawfordville, FL, USA).

Relationships of NDVI with MAT and MAP
The regression analysis demonstrated that the NDVI over the Tibetan Plateau was positively correlated with the MAP and MAT. A similar phenomenon was found from 1982 to 2013 (Table 1) where the NDVI exhibited positive correlations with MAP (and MAT). In addition, the VSP ranged from 0.51 to 0.71 (Table 1), which was higher than the VST (0.20-0.29), indicating that the vegetation was more sensitive to MAP than MAT over the Tibetan Plateau.

Seasonal Sensitivity between NDVI and Climate Records
VSP (and VST) showed remarkable differences among the different seasons (Table 1 and Figure 2). The maximum VSP was presented in summer, followed by autumn, winter, and spring. All of the VSP values in summer, autumn, and winter were greater than 0.5, while it was less than 0.25 in spring. Meanwhile, the peak value of VST was found in spring, followed by winter, summer, and autumn.   According to Figure 3, the sensitivity of the NDVI to MAT and MAP fluctuated from 1982 to 2013 and showed non-significant changes ( Figure 3A,B). However, both showed seasonal variations; specifically, VSP in spring (y = 0.003x − 6.37, R 2 = 0.32, p < 0.001) and winter (y = 0.007x − 13.68, R 2 = 0.25, p < 0.001) increased with the time series (from 1982 to 2013) ( Figure 3C). VST presented a significant decreasing trend from 1982 to 2013 only in summer (y = −0.002x + 3.75, R 2 = 0.21, p < 0.001) ( Figure 3D). In addition, both VSP and VST exhibited non-significant variations in other seasons from 1982 to 2013.   On the spatial scale, the regression analysis showed a quadratic curve relationship between the precipitation and VSP (y = −0.0002x 2 + 0.02x + 0.21, R 2 = 0.47, p < 0.001) ( Figure 4A), indicating that when the values of precipitation are approximately 48 mm (summer and autumn) the VSP will be maximized, but the change rate is minimal. This differs from the relationship between temperature and VST, where VST decreased significantly with the increased temperature (y = 0.03 + 0.33 1+e x+0. 67 1.83 , R 2 = 0.95, p < 0.001). Interestingly, the change rate of VST sharply decreased when it exceeded the critical threshold (0 • C, summer) ( Figure 4B). 4A), indicating that when the values of precipitation are approximately 48 mm (summer and autumn) the VSP will be maximized, but the change rate is minimal. This differs from the relationship between temperature and VST, where VST decreased significantly with the increased temperature (y = 0.03 + . . . , R 2 = 0.95, p < 0.001). Interestingly, the change rate of VST sharply decreased when it exceeded the critical threshold (0 °C, summer) ( Figure 4B).

Climate Change and its Effect on Seasonal VSP and VST
As illustrated in Figure 5

Climate Change and Its Effect on Seasonal VSP and VST
As illustrated in Figure 5 The SEM explained 30.6% of the variation in VSP in spring (PSp), 22.9% of the variation in VSP in winter (PWi), and 28.3% of the variation in VST in summer (TSu) across the Tibetan Plateau ( Figure 6). Table 2 shows a summary of the direct, indirect, and total effects of the variables. Dramatic climate changes were strongly associated with increased PSp and PWi, which indicated that PSp and PWi could be well explained by annual increasing precipitation and temperature (climate change) with R-squared values of 0.306 and 0.229, respectively. Despite significant bivariate relationships among MAP, MAT, and The SEM explained 30.6% of the variation in VSP in spring (PSp), 22.9% of the variation in VSP in winter (PWi), and 28.3% of the variation in VST in summer (TSu) across the Tibetan Plateau ( Figure 6). Table 2 shows a summary of the direct, indirect, and total effects of the variables. Dramatic climate changes were strongly associated with increased PSp and PWi, which indicated that PSp and PWi could be well explained by annual increasing precipitation and temperature (climate change) with R-squared values of 0.306 and 0.229, respectively. Despite significant bivariate relationships among MAP, MAT, and PSp or PWi being found, the results mostly demonstrated the indirect positive effects on PSp or PWi. The rank of total effects on them, exhibiting the same decreasing order, was as follows: climate change, MAT, and MAP (Table 2). Meanwhile, negative effects of climate change on TSu were observed, which tells us that TSu might be well illuminated by climate change (R 2 = 0.283). In addition, MAP and MAT only had indirect negative effects on TSu through climate change. The total effects on TSu were decreased in order of climate change, MAT, and MAP (Table 2). Note: CC, MAP, and MAT represent climate change, mean annual precipitation, and mean annual temperature, respectively; PSp, PSu, PAu, and PWi represent vegetation sensitivity to precipitation in spring, summer, autumn, and winter, respectively; TSp, TSu, TAu, and TWi represent vegetation sensitivity to temperature in spring, summer, autumn, and winter, respectively. *** correlation is significant at the 0.001 level; ** correlation is significant at the 0.01 level.
Atmosphere 2022, 13, 1112 9 of 14 Atmosphere 2022, 13, x FOR PEER REVIEW 9 of 14 Figure 6. SEM was used to analyze the direct and indirect effects among variables in the Tibetan Plateau. The standardized total coefficients are listed on each path. The thickness of the solid arrows reflects the magnitude of the standardized SEM coefficients, the black solid line represents the positive effect, and the red solid line represents the negative effect. MAP and MAT represent the mean annual precipitation and mean annual temperature, respectively; PSp, PSu, PAu, and PWi represent vegetation sensitivity to precipitation in spring, summer, autumn, and winter, respectively; and TSp, TSu, TAu, and TWi represent vegetation sensitivity to temperature in spring, summer, autumn, and winter, respectively. Note: CC, MAP, and MAT represent climate change, mean annual precipitation, and mean annual temperature, respectively; PSp, PSu, PAu, and PWi represent vegetation sensitivity to precipitation in spring, summer, autumn, and winter, respectively; TSp, TSu, TAu, and TWi represent vegetation sensitivity to temperature in spring, summer, autumn, and winter, respectively. *** correlation is significant at the 0.001 level; ** correlation is significant at the 0.01 level.

Discussion
This study redefined the sensitivity index between climate and the NDVI by the coefficient of determination (R 2 ) obtained by a linear regression analysis. It found that vegetation sensitivity to precipitation (VSP) increased in spring and winter, while vegetation sensitivity to temperature (VST) decreased in summer. In fact, R 2 is often used to study the influence of climate on vegetation dynamics. For example, it has been used to analyze the effects between the NDVI and climate, and reflect the effect of the climate on net primary production [52,55]. In addition, compared with the previous calculation method of VSP and VST by least squares [1,38], the spatial heterogeneity of VSP and VST was reduced in the updated method because it fragmented the space by dividing the Tibetan Plateau into many small plots, on average, to discuss the sensitivity of vegetation to precipitation and temperature. On a timescale, the findings showed that there were no significant variations in the sensitivity of the NDVI to MAT and MAP from 1982 to 2013. Similarly, this phenomenon has been mentioned in previous studies, such as Hsu et al. [1] who found that changes in the inter-annual variability of precipitation had negligible effects on the mean ANPP. On the contrary, Li et al. [55] highlighted that the NDVI is adaptable to the significant increase in temperature but is sensitive to the decrease in precipitation on the Inner Mongolia Plateau, possibly due to regional differences. In addition, VST and VSP in the different seasons showed significant dynamic change at the scale of space-time (Figures 3 and 4).
More precisely, the VSP in spring and winter showed a significant increase from 1982 to 2013 ( Figure 3C). The single-hump relationship between VSP and precipitation ( Figure 4A) demonstrated that VSP ascended first and then descended with increased precipitation. Interestingly, the precipitation in spring and winter was distributed only at the edge of the hump, whereas the precipitation in summer and autumn was distributed at the hump. That is, if there is an inter-annual change in precipitation, there will be a significant fluctuation in VSP in spring and winter, whereas the VSP in summer and autumn will fluctuate less ( Figure 4A). Numerous studies have also indicated that vegetation is sensitive to precipitation in spring and winter [39]. However, the single-hump relationship between VSP and precipitation in those studies was an inverted hump, which may be caused by the quality of the remote-sensing image and different data-processing methods [56][57][58][59]. In fact, it is easy to understand in terms of the supply-demand relationship between water resources and plants, especially in dry and cold environments (e.g., spring and winter), where precipitation is unquestionably the main limiting factor for plant growth [60][61][62], and a slight increase in precipitation will lead to a significant change in VSP. Nevertheless, increasing precipitation does not necessarily promote growth in VSP when precipitation reaches a certain amount (e.g., summer and autumn). In contrast, the sensitivity of vegetation to precipitation showed a slightly decreased trend because of the negative effect of soil leaching on the positive relationship between precipitation and vegetation [63,64].
For VST, a sigmoidal model was found between VST and temperature ( Figure 4B). Overall, VST decreased with increased temperature, but the change rate of VST was far different across the temperature gradient. When the temperature was around 0 • C (summer), the change rate of VST was the largest; a small fluctuation of temperature would lead to a plummet or soar in VST. However, when the temperature was outside that range (spring, winter, and autumn), there would be little fluctuation in VST with a variation in temperature. The reason is probably that summer is the pivotal growing season for alpine plants on the Tibetan Plateau [65,66], and strong evidence has demonstrated that warming induced an earlier start of the growing season [67][68][69], delayed the end of the growing season [70][71][72], and enhanced biomass accumulation [73][74][75][76][77][78]. In addition, the fluctuation in temperature at 0 • C, which led to a large amount of water to condense; water condensation might significantly promote the growth of aboveground biomass, reduce the root-to-shoot ratio, and enhance the accumulation of photosynthetic products in whole plant leaves [79,80]. It was only the uncoordinated changes between the rapidly increased aboveground biomass and the slight fluctuation in temperature that led to the extreme decline in VST in summer. In summary, variations around the critical threshold for the hydrothermal environment (monthly precipitation: 4-12 mm; monthly average temperature: 0 • C) resulted in variations in seasonal VST and seasonal VSP. However, the lagged response period of vegetation to hydrothermal environments is generally only two months [81,82]; consequently, when the average temperature and precipitation span a long time, although the average temperature and precipitation are near the threshold conditions, their variation will not cause significant dynamics in VSP and VST. That is why the annual VSP and VST during 1982-2013 did not change significantly with the MAP and MAT, varying around 25-34 mm and −2.5-0.0 • C, respectively.
Additionally, the result of SEM also proved that climate change can only regulate VSP and VST under limited hydrothermal conditions. It showed that climate change had positive effects on PSp and PWi (path coefficients = 0.554 and 0.478, respectively), and a negative effect on TSu with a path coefficient of −0.428 ( Figure 6), but for any other dependent variable the effect of climate change on them is not significant ( Table 2). It means that with the exception of the monthly precipitation range of 4-12 mm and monthly average temperatures around 0 • C, annual increasing precipitation and temperature under other conditions do not cause significant changes in the VSP and VST. Even if there is an interaction between precipitation and temperature [60,[83][84][85] and impaction from other climate factors, such as solar radiation and carbon dioxide concentrations [72,[86][87][88][89], it will not change as climate change refers to an increasing time sequence.

Conclusions
Climate change will alter the sensitivity of the NDVI to seasonal precipitation and seasonal temperature. Our analysis quantified these changes and indicated that the sensitivity of the NDVI to seasonal precipitation showed a significant increase in spring and winter, while there was a significant decreasing trend in the sensitivity of the NDVI to seasonal temperature in summer. Our sensitivity analysis also quantified the critical threshold conditions of these changes, and pointed out that the threshold conditions of seasonal VSP and seasonal VST were captured in the 4-12 mm range (monthly precipitation) and 0 • C (monthly average temperature), respectively. This study highlighted that climate change under these threshold conditions would lead to a variation in the sensitivity of the NDVI to seasonal precipitation and seasonal temperature. Therefore, more attention should be paid to the mechanism of plant physiological changes under hydrothermal threshold conditions in future studies.