**3. Results**

#### *3.1. Spatial Pattern of Snow Seasonality over the QLMA*

The spatial heterogeneity pattern of snow seasonality metrics can be observed in the QLMA (Figure 5). The earliest snowfall occurs in the central area within the QLMA (before October, DOY < 30). This region is also the one with the last snowmelt (after April, DOY > 210), resulting in the longest snow season. For the western and northern regions, the first snowfall occurs weeks later (before December, DOY < 90), and snowmelt occurs late (after April, DOY > 210), the length of the snow season is shorter than in the center. In the northern part of Qinghai Lake and the eastern and southern fringes of the study area, the first snowfall occurs the latest (after December, DOY > 120), the snowmelt occurs earliest (before February, DOY < 150), and the snow season is the shortest. The spatial pattern of SCD is similar to that of SSL, but the spatial differences are not significant because of the widespread presence of intermittent snow. The SCD is less than 30 d in most of the study area, especially in the southeast, but higher in the central and western parts.

**Figure 5.** Mean of snow seasonality metrics in QLMA from 2002 to 2020. (**a**) represents the FSD, (**b**) represents the LSD, (**c**) represents the SSL, and (**d**) represents the SCD.

#### *3.2. Land Surface Phenology among Different Vegetation Types*

Figure 6 shows the calculated LSP metrics for the different vegetation types. LSP metrics for all vegetation except desert varied with elevation: SOS delays (Figure 6a), EOS advances (Figure 6b) and LOS shortens (Figure 6c) as elevation rises. There are no significant trends in SOS or LOS with elevation for desert, but EOS advances with elevation.

Figure 6d shows the average LSP metrics across elevation. Desert and alpine vegetation have the earliest growing seasons starting in late April. The growing season of forest begins in mid-May, while shrub, grass and cultivated vegetation start growing in late May. The EOS of different vegetation types are relatively close. The EOS of vegetation types other than alpine vegetation generally appear in early September, with desert and forest a few days earlier. The EOS of alpine vegetation is the earliest, occurring around mid-August. Desert has the longest growing season lengths (SSL = 120), SSL in forest (SSL = 105) and alpine vegetation (SSL = 101) are shorter. Shrub, grass and cultivated vegetation have the shortest growing seasons of about 95 d.

**Figure 6.** The calculated LSP metrics for the different vegetation types change with elevation. (**a**) represents the SOS, (**b**) represents the EOS and (**c**) represents the LOS. (**d**) shows the average of all elevations. The bar graph represents the average SOS and EOS of six different vegetation types, and the arrows inside the bar indicate their LOS. (**e**) shows the histogram of pixel numbers of different elevation gradients.

#### *3.3. Spatial Pattern of Land Surface Phenology over the QLMA*

The spatial pattern of the mean values of land surface phenology metrics in the QLMA area is shown in Figure 7. Many western areas were filtered out because of the lack of seasonal vegetation. The earliest start of the growing season is in the western and northern margins, usually before 1st May, followed by the eastern areas, where the growing season starts within about a month (before 1st June). Vegetation in the central part of the study area has the latest start of the growing season, occurring after 1st June, and in a few areas even later (after 15th June) (Figure 7a). The growing season for most of the vegetation in the study area ends between 15th August and 15th September. Growing seasons in the northern part of Qinghai Lake and the eastern edge of the study area end half a month later. The vegetation with the latest end of the growing season is located in the western region, occurring after 1st October (Figure 7b). The spatial distribution pattern of LOS is similar to that of SOS, with the northern edge and the sporadic areas in west having the longest vegetation growing season, which exceeds 150 d at most. From the eastern area toward the center, the vegetation growing season gradually shortens from up to 120 d to up to 90 d (Figure 7c).

**Figure 7.** Mean of land surface phenology metrics in QLMA from 2003 to 2021. (**a**) represents the SOS, (**b**) represents the EOS and (**c**) represents the LOS.

#### *3.4. Spatial Pattern of the Correlation between Snow Seasonality and Land Surface* Phenology Metrics

Different snow seasonality metrics have different strengths and directions of influence on different LSP metrics, and there is spatial variation in this correlation (Figure 8). Overall, the snow seasonality metrics have a similar impact on EOS and LOS, in contrast to SOS. FSD shows a significant positive correlation with SOS in the southern part of Qinghai Lake and a mostly nonsignificant negative correlation with the central part. FSD shows mainly negative correlation with EOS and LOS in the study area, especially in the central region, where the negative correlation is significant. EOS and LOS in the eastern part of the study area show a nonsignificant positive correlation with FSD. The effect of LSD on LSP is different from that of FSD. LSD shows a significant negative correlation with SOS in the western part of the study area and the southern part of Qinghai Lake. Both EOS and LOS in the central and western parts of the study area show positive correlations with LSD, where LOS in the western part shows a significant positive correlation with LSD. EOS and LOS in the southeast of the study area, however, show a negative but insignificant correlation with LSD. The spatial pattern of SSL and SCD effects on LSP is similar to that of LSD, compared to the more significant correlation of SCD with LSP.

The proportion of significant positive and significant negative correlations indicates the main direction of influence of snow seasonality metrics on LSP metrics (Table 1). All the 12 correlations have a relatively obvious directionality, meaning that there is no positive correlation with the same proportion of negative correlations. Among them, positive correlations dominate in FSD\_SOS, LSE\_EOS, LSD\_LOS, SSL\_L0S and SCD\_LOS. The proportion of insignificant positive correlations is about 7% more than the proportion of insignificant negative correlations, and the proportion of significant positive correlations exceeds the proportion of significant negative correlations by twice. However, negative correlation is the major of FSD\_LOS, SSL\_SOS and SCD\_SOS. The proportion of insignificant positive correlations is approximately 6% less than the proportion of insignificant negative correlations, and the proportion of significant positive correlations is less than half of the proportion of significant negative correlations.

**Figure 8.** Correlation between different land surface phenology metrics and different snow seasonality metrics. "Significant" means *p* < 0.1.

**Table 1.** Area of significant correlation between snow season metrics and phenology metrics (%), SN for significant negative (*p* < 0.1) correlation and NN for nonsignificant negative (*p* > 0.1, r < −0.2) correlation. SP for significant positive (*p* < 0.1) correlation and NP for nonsignificant positive (*p* > 0.1, r > 0.2) correlation.


#### *3.5. Elevation-Dependent Correlation between Snow Seasonality and Land Surface* Phenology Metrics

We further investigated the phenological response along the elevation gradient (Figure 9). FSD is significantly negatively correlated with SOS and positively correlated with LOS at low elevation, and the correlation gradually decreases as the elevation increases to 3500 m. The direction of the effect of FSD above 3500 m changes. The correlation with SOS turns from a nonsignificant positive correlation to a significant positive correlation by degrees, and the correlation with LOS changes to a progressively increasing negative correlation. The correlation between FSD and EOS is weak and always fluctuates around 0. There is no shift in the correlation between LSD and LSP metrics with rising elevation. Regardless of the elevation, LSD is always nearly significantly positively correlated with EOS and LOS. LSD is significantly negatively correlated with SOS at low elevation, and the correlation and significance decrease with rising elevation, but there is an increasing trend above 3900 m. The correlation of SSL and SCD with LSP metrics has similar characteristics

with elevation. Below 3300 m, they show a nonsignificant weak correlation with LSP metrics. The correlations gradually increase with elevation, and approach stability at around 3500 m. Above 3500 m, SSL and SCD show a significant negative correlation with SOS and a significant positive correlation with EOS and LOS, especially LOS. The effect of SCD is more obvious than SSL.

**Figure 9.** Left panel shows the land surface phenology metrics response to different snow seasonality metrics at different elevation. (**a**) means FSD, (**b**) means LSD, (**c**) means SSL and (**d**) means SCD. Right panel shows the vegetation constitute at different elevations.

Vegetation constitute varies at different elevations. Forest and cultivated vegetation are largely absent above 3000 m. Shrub increases rapidly from 4.95% to a maximum of 27.56% above 2800 m, with most of its distribution concentrated between 2900 m and 3700 m. Grass is the dominant vegetation type in most areas, usually accounting for more than 60% of the vegetation.

#### *3.6. Interspecific Variation in the Response of Land Surface Phenology*

To characterize differences in the responses of land surface phenology to snow seasonality, we compared the responses of six vegetation phenological characteristics to snow (Figure 10). The elevation interval with the highest concentration of vegetation distribution is selected: 2600–3500 m for forest and cultivated vegetation, 3100–4000 m for grass, shrub and desert, and 3600–4200 m for alpine vegetation. FSD is significantly negatively correlated with SOS of forest and LOS of alpine vegetation, and significantly negatively correlated with EOS of shrub. EOS of forest, shrub and grass is significantly positively correlated with LSD, as is LOS of shrub and grass. SSL and SCD affect vegetation in almost the same direction. SSL is significantly correlated with EOS for more vegetation types, while SCD is significantly correlated with SOS and LOS of vegetation. The SOS of forest is significantly positively correlated with SCD, while the SOS of desert is significantly negatively correlated with SCD. EOS of shrub and grassland is more sensitive to SSL and is significantly positively correlated. LOS of alpine vegetation is significantly positively correlated with both SSL and SCD, while LOS of desert is significantly positively correlated with SCD only.

**Figure 10.** Correlation between vegetation phenological metrics of eight types of vegetation and snow phenological metrics. \* *p* < 0.1, \*\* *p* < 0.05, \*\*\* *p* < 0.01.
