**4. Discussion**

#### *4.1. The Spatial Heterogeneity of Vegetation Phenology in the Qilian Mountains*

The vegetation phenology in the QLMs showed significant spatial heterogeneity. In general, the SOS was later in the central region and earlier in the eastern and western regions of the QLMs, and the EOS exhibited the opposite trend in terms of its spatial distribution. These results are consistent with the results reported by Qi et al. [27] and Sun et al. [35] but inconsistent with the results reported by Qiao et al. [36], who observed that the multiyear mean SOS was gradually delayed from southeast to northwest and that the multiyear mean EOS gradually advanced from southeast to northwest in the QLMs. The main reasons for the differences in the results are the different temporal and spatial resolutions of the remote sensing data. Different remote sensing data sources and data time series may obtain different vegetation phenology results [37,38]. The AVHRR and MODIS datasets have a consistently high temporal resolution time series of data and are widely used for phenology studies [39]. MODIS data (1 km) have a higher resolution than AVHRR data (8 km) and can extract more detailed spatial phenological signals for vegetation types, particularly in heterogeneous areas [40]. The vegetation phenology of the QLMs was characterized by an advanced SOS, delayed EOS, and extended LOS during the period from 2001 to 2020, which are consistent with recent results on the QTP [37,38,41] and on the QLMs [27,35].

In our study, the results showed that the SOS gradually delayed, the EOS gradually advanced, and the LOS gradually shortened with increasing altitude. No significant correlation was found between the SOS and altitude, but a significant negative correlation was found between both the EOS and LOS and altitude. These results are consistent with recent findings on the QTP [2]. The SOS change may have almost nothing to do with altitude [2]. At higher altitudes, there is relatively low air temperature, which is not beneficial for delaying leaf senescence, and the EOS advances to avoid harm from frost and has a shorter LOS [42].

#### *4.2. Response of Vegetation Phenology to Different Driving Factors*

Vegetation phenology responses to different driving factors are complex and variable. In our study, we found that the SOS was negatively correlated with spring temperature and spring soil moisture in most regions of the study area, implying that the advanced SOS could be associated with a warmer spring air temperature and higher soil moisture. In our study, the spring temperature had a stronger influence on the SOS, and most studies have also reported that higher temperatures were the main factor associated with an earlier SOS around the world over the last several decades [18,43–45].

The impacts of driving factors on vegetation phenology were varied in different elevation zones. For SOS, spring temperature seemed to be the main factor limiting vegetation growth. The QLMs are located in high altitude areas with low temperatures (the annual mean temperature in most areas is below 0 ◦C); with an increase in altitude, the temperature gradually decreased (Table 2). Vegetation needs a certain amount of cumulative temperature to green up, so the early stages of vegetation growth are more affected by temperature in relatively cold regions [5,46]. The autumn soil moisture was the main limiting factor at lower elevations (<3500 m a.s.l.), and autumn temperature was the main limiting factor at higher elevations (>3500 m a.s.l.). These results are consistent with the research of Peng et al. [47], which demonstrated that soil moisture was the major limiting factor for the radial growth of Qinghai spruce at the lower elevations of the central QLMs and that temperature was the major limiting factor for radial growth of Qinghai spruce at higher elevations. These results also sugges<sup>t</sup> that vegetation managemen<sup>t</sup> must take elevation differences into account when facing the challenges of climate change. From Table 2, we can see that the annual average soil moisture at lower elevations (0.31 <sup>m</sup>3·m<sup>−</sup>3) was less than that at higher elevations (0.34 <sup>m</sup>3·m<sup>−</sup>3), so this is one possible reason that the autumn soil moisture had a stronger influence on the EOS in lower elevation zones. The EOS showed a significant negative correlation with summer soil moisture in approximately 30.06% of the pixels in the lowest elevation zones (<3000 m a.s.l.). Peng et al. [47] also found that, during the summer at lower elevations, soil moisture is the most important factor limiting xylem cell differentiation based on the Vaganov–Shashkin model.


**Table 2.** The annual average soil moisture and temperature from 2001 to 2020 in four elevation zones.

At the landscape level, the SOS was negatively correlated with spring temperature in most regions with different vegetation types. More specifically, 42.36%, 23.31%, 31.54%, and 24.83% of the areas of broadleaf forests, needleleaf forests, shrublands, and meadows, respectively, showed a significantly negative correlation between SOS and spring temperature (Figure 8a). This is because the broadleaf forests, needleleaf forests, shrublands, and meadows are mainly located in semi-arid regions (more than 78% of these are located in the semi-arid region) where the climate is relatively humid compared with arid regions; however, the temperature is low in the study area, and higher temperature in spring could decrease the damage from frost and promote spring thawing [5]. The spring soil moisture had a stronger influence on the SOS of deserts (Figure 8c). This is because about 90.37% of deserts are located in arid areas with limited soil water conditions (Table 3). The soil water is an indispensable intermediary used to ensure nutrient substance transport, which is likely to be the main reason for the negative correlation between soil moisture and SOS for deserts. Additionally, there are many shallow-rooted plants in deserts, and these shallow-rooted plants are more sensitive to soil moisture changes than other plants [48,49].

A negative correlation was observed between summer soil moisture and EOS, but a positive correlation was observed between autumn soil moisture and EOS for most vegetation types. This negative correlation shifted to a positive correlation from summer to autumn, indicating that the summer and autumn soil moisture had a grea<sup>t</sup> influence on the EOS, but the correlation was the opposite in these two seasons. The main reason for this differential response is that the precipitation in QLMs is mainly concentrated in summer [25], and too much moisture prevents vegetation growth because a high soil moisture can limit the absorption of soil nutrients by vegetation [50]. Ren et al. found that the precipitation played a more important role than temperature in the interannual variation of the SOS and EOS in Inner Mongolia [19]. However, compared with temperature and soil moisture, precipitation had a relatively limited impact on the EOS in the QLMs. This is because precipitation may a have lagged effect on vegetation phenology, meaning that soil moisture is a more straightforward driving factor for vegetation phenology than precipitation and has a number of sources in the QLMs, including precipitation, snowmelt, surface runoff, and groundwater.

**Table 3.** The annual average soil moisture and temperature from 2001 to 2020 for different vegetation types.


#### *4.3. Limitations and Future Work*

It should be noted that there may be some limitations to our current study. The number of ground observations of vegetation phenology is insufficient, especially in the central and western parts of the QLMs because of a lack of phenological observation networks. At the same time, the existing observation stations have a relatively short historical record. Digital cameras have been shown to be valuable tools to validate the phenology derived from satellite imagery at a low cost [40] because of their high temporal and spatial resolutions. In future, automated digital cameras are promising for providing consistent and continuous monitoring of vegetation growth at local and regional scales [51,52].

The vegetation phenology results calculated from remote sensing data may contain some uncertainties that are due to the inaccuracy of satellite data. NDVI data have been widely used for phenology characterization because they are simple to measure for most optical sensors [53,54]. However, because NDVI data are sensitive to the soil background and are easily saturated in high vegetation coverage areas [55], the applications of NDVI data may have some limitations. Considering the sparse vegetation in the western part of the QLMs, the modified vegetation index, such as the soil-adjusted vegetation index (SAVI), may be appropriate for detecting vegetation growth changes because of its ability to minimize the effects of the soil background. In the future, collective analyses of multiple VIs (such as the land phenology index, enhanced vegetation index, and perpendicular vegetation index) may improve the accuracy of phenology estimation [56–58]. The spatial resolution of ERA5-Land soil moisture data is relatively low, which may hide some spatial details of soil moisture parameters. But the high-resolution soil datasets are difficult to obtain for a large study area [59]. Future studies should integrate a series of soil moisture datasets at a higher resolution to further discuss the response relationship between vegetation phenology and soil moisture. Vegetation phenology is also influenced by other factors,

such as radiation, soil nutrients, climate extremes, and human activities, so more attention should be paid to exploring the phenology variations in response to these driving factors in future work. The vegetation phenology response to driving factors may be nonlinear, and the interactions between climatic factors have critical role in vegetation phenology, so other methods like the GeoDetector model can be used to detect the contribution of driving factors to vegetation phenology and the interactions between driving factors. Our findings sugges<sup>t</sup> that the variation in soil moisture should be considered in future studies on climate warming and the environmental effects of phenology in water-limited areas.
