Phenological Shifts of Vegetation in Seasonally Frozen Ground and Permafrost Zones of the Qinghai–Tibet Plateau

Highlights

What are the main findings?

• A delayed start of the growing season, an advanced end of the growing season, and a shortened length of the growing season were observed during certain sub-periods.
• The end of the growing season showed opposite trends both in permafrost versus seasonally frozen ground regions and between alpine meadow and alpine steppe.

What is the implication of the main finding?

• Vegetation phenology was likely to show new changes amid current climate change.
• When studying changes in vegetation phenology, the impacts of permafrost changes and differences in vegetation types should be considered.

Abstract

Vegetation phenology serves as a crucial indicator reflecting vegetation responses to the growth environment and climate change. Existing studies have demonstrated that in permafrost regions, the impact of frozen soil changes on vegetation phenology is more direct and pronounced compared to climate factors. Amid the slowdown of global warming in the 21st century, permafrost dynamics continued to drive uncertain variations in vegetation phenological stages across the Qinghai–Tibet Plateau (QTP). Using MODIS Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) data during 2001–2024, this study derived vegetation phenological parameters and analyzed their spatiotemporal patterns on the QTP. The results indicate that overall, the start of growing season (SOS) was advanced, the end of growing season (EOS) was delayed, and the length of growing season (LOG) was extended throughout the study period. Additionally, divergent phenological trends were observed across three distinct phases, and regarding frozen soil types, vegetation phenology in permafrost and seasonally frozen ground regions exhibited distinct characteristics. From 2001 to 2024, both permafrost and seasonally frozen ground regions showed an advanced SOS and prolonged LOG, but significant differences were observed in EOS dynamics. For vegetation types, alpine meadow displayed advanced SOS and EOS, alongside an extended LOG. The alpine steppe exhibited advanced SOS and delayed EOS with an extended LOG. Alpine desert displayed SOS advancement and EOS delay, alongside LOG extension. These findings revealed variations in vegetation phenological changes under different frozen soil types and highlighted divergent responses of distinct frozen soil types to climate change. They suggested that the influence of frozen soil types should be considered when investigating vegetation phenological dynamics at the regional scale.

1. Introduction

Plant phenology studies on recurring biological events in plant life cycles [1]. It is widely recognized that climate change affects the vegetation phenology and the condition of the frozen soil [2,3]. The continuous changes in frozen soil brought uncertainties to the changes in vegetation phenology [3,4,5,6]. The Qinghai–Tibet Plateau (QTP), often termed “the Third Pole”, averages over 4000 m in elevation and exhibits unique climatic, geological, and ecological characteristics [5]. Due to its cold environment, a large area of permafrost has formed in the QTP, covering approximately 40% of the plateau’s total area, and seasonally frozen ground accounts for an additional 56% [6].

The vegetation phenology on the QTP changed. The start of the growing season (SOS) was delayed during 2000–2006 [7,8] while it was advanced during 2001–2020 [9,10]. These discrepancies in SOS trend interpretations may stem from data quality issues [10,11,12]. From the perspective of different regions, SOS was significantly delayed in the southwestern plateau, and it was contrasted with the advancement patterns in other regions [10,13]. The end of the growing season (EOS) was significantly delayed in recent decades [9,10]. From a regional perspective, the EOS was delayed in the west and south, while it was advanced in the east [10,14]. In terms of the length of growing season (LOG), it was extended during 1960–2014 [15], and this was predominantly attributed to the advancement of SOS [9,15].

Numerous existing studies suggested that air temperature was the dominant factor controlling changes in the SOS [16]. However, recent studies indicated that changes in frozen soil exerted a more direct and intense impact on vegetation phenology than climatic factors [4,17]. Over recent decades, the plateau has experienced pronounced climate warming [18], driving significant transformations in both permafrost and seasonally frozen ground distributions [19,20]. Permafrost degradation altered soil thermal and hydrological properties, including rising soil temperatures, increased soil moisture in lowland areas, and drained highland soils [21]. These changes influence vegetation coverage [22] and trigger phenological shifts, thereby profoundly impacting alpine ecosystems [21]. Such ecosystem modifications further disrupt soil hydrothermal regimes, establishing feedback mechanisms with permafrost dynamics [22]. For instance, vegetation cover was strongly correlated with the thickness of the active layer [23,24] and served as a critical regulator of soil moisture dynamics [23].

Differences in vegetation types also lead to differences in vegetation phenology. The QTP is predominantly characterized by three primary vegetation types, which are alpine meadow, alpine steppe, and alpine desert [24]. It was currently believed that the alpine desert had the earliest SOS, while the alpine steppe had the latest [25]. However, the specific changes in their phenology in recent decades remained unknown [25]. Regarding the EOS, there were significant controversies over the changes in the three types of vegetation [25,26,27].

In summary, the respective phenological changes in vegetation in different frozen ground remained vague. Moreover, due to the slowing warming trend over the QTP in recent decades [28] and the identification of a climate change turning point in the 1990s in existing studies [7], new changes in vegetation phenology may occur. This study will utilize the vegetation and frozen soil data to explore the phenological changes on the QTP under climate change [6,29], while highlighting the distinct responses of different permafrost types to climate change. To ensure the consistency and continuity of the data and avoid the uncertainties introduced by multi-source data fusion, this study simulates the spatiotemporal changes in vegetation phenology on the QTP from 2001 to 2024 based on the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI) of MODIS13A2. The research objectives are to explore (1) the spatiotemporal variation in alpine grassland vegetation phenology on the QTP over the past 24 years, and (2) the phenological characteristics among vegetation types and frozen soils.

2. Study Area

The QTP is situated in western China between 26°00′–40°12′N and 67°41′–104°41′E, encompassing a total area of 2.57 million km² [30] with an average elevation exceeding 4000 m [31]. The region features a low annual average temperature, with low precipitation that is concentrated in the period from June to September [32,33]. Characterized by cryospheric dominance, the QTP contains extensive frozen ground covering approximately 96% of its total area, comprising 42% permafrost and 58% seasonally frozen ground (Figure 1a) [6]. In the QTP, the main vegetation types on the frozen soil are alpine meadow, alpine steppe, and alpine desert [29]. Among them, the alpine meadow is mainly distributed in the eastern part, the alpine desert is primarily in the northern part, and the alpine steppe is mainly in the southwestern part of the QTP (Figure 1b). Vegetation types differ between permafrost and seasonally frozen ground regions. In seasonally frozen ground, alpine meadow dominates (60.5%), followed by alpine steppe (39.5%) [34]. In permafrost regions, alpine meadow and alpine steppe account for 31.9% and 32.9%, respectively, with alpine desert comprising 35.2% [34].

3. Data and Methods

3.1. Data

The MODIS MOD13A2 datasets from 2001 to 2024 were obtained from NASA’s AppEEARS platform (https://appeears.earthdatacloud.nasa.gov/, accessed on 1 May 2025). Permafrost distribution data were derived from Zou et al. [6]. Vegetation classification was derived from the 30-m resolution Tibetan Plateau Vegetation Maps (1990–2020) [29]. The original vegetation data underwent cross-validation with the vegetation atlas of Li et al. [35], merging alpine vegetation and alpine desert categories while retaining the alpine desert nomenclature. Alpine meadow, alpine steppe, and alpine desert collectively account for 99% of the plateau’s total vegetated area. Evergreen vegetation was excluded through field investigations and reference to relevant publications. Firstly, vegetated regions with an average NDVI exceeding 0.1 during July to September [13,36], and those with an average EVI ranging from 0.07 to 0.8 within the same time period were retained [37,38]. Subsequently, areas where the difference between summer and winter average NDVI (or EVI) values was less than 0.1 were identified and excluded from the study scope [13]. This process effectively eliminated evergreen vegetation from the analysis. Due to the unavailability of 2024 precipitation and temperature data, this study utilized corresponding datasets spanning the period from 2001 to 2023 [39,40].

3.2. Method

Clouds and poor atmospheric conditions often affect the value of NDVI (or EVI) [13]. To distinguish abnormal points resulting from noise and remove their impacts [13,36], this study employed the Savitzky–Golay algorithm (SG) to replace them with corresponding data from the fitted smooth curve of the NDVI and EVI data [41]. The formula is as follows.

$$Y_j={\displaystyle \frac{{\sum }_{i=-m}^{i=m}{C}_i{Y}_{j+1}}{N}}$$

(1)

$Y_j$ is the composition sequence data, and $Y_{j+1}$ represents the original NDVI. $C_i$ is the filter coefficient, and $N$ is the width of the sliding window. This method has been validated in previous studies [13,36]. Then, this study used the dynamic threshold method to invert the vegetation parameters [41,42,43]. The fitting function is the sixth-degree polynomial [7]. Through repeated experiments and by drawing on the experience of predecessors, we set the optimal threshold as the following indicators. For NDVI, the SOS of vegetation is defined as the point when the left-half amplitude of the fitting function increases by 20%, and the EOS of vegetation is defined as the point when the right-half amplitude of the fitting function decreases to 60% [8,44]. For EVI, the SOS of vegetation is defined as the point at which the left-half amplitude of the fitting function increases by 20%, and the EOS of vegetation is defined as the point at which the right-half amplitude of the fitting function decreases to 50% [37]. The LOG is the difference between the SOS and the EOS.

To analyze the spatiotemporal variation in vegetation phenology on the QTP from 2001 to 2024, this study used the Theil–Sen median slope estimation and Mann–Kendall non-parametric test. The Theil–Sen median is a stable trend analysis approach used for trend determination, aiming to avoid the influence of errors [45]. This approach is often employed for trend analysis of long-term time series data [44]. The formula is as follows.

$$S=median\left({\displaystyle \frac{y_i-y_j}{x_i-x_j}}\right)$$

(2)

The S represents the Theil–Sen slope, and the positive value indicates the trend of postponement or lengthening, while the negative value indicates the trend of advancement or shortening.

Mann–Kendall is a non-parametric statistical test, which is usually used in combination with the Theil–Sen median slope estimation to determine the statistical significance of the obtained trend [46]. It does not require time series data that follows a normal distribution or a linear trend, and is not affected by missing values and outliers [47]. The formula is as follows.

$$Z=\left\{\begin{array}{c}{\displaystyle \frac{S-1}{\sqrt{Var\left(S\right)}}} , S>0\\ 0 , S=0\\ {\displaystyle \frac{S+1}{\sqrt{Var\left(S\right)}}} , S>0\end{array}\right.$$

(3)

$$S={\Sigma }_{\mathrm{i}=1}^{n-1}{\Sigma }_{j=\mathrm{i}+1}^n\left\{\begin{array}{c} 1, {\mathrm{y}}_{\mathrm{j}}-{\mathrm{y}}_{\mathrm{i}}>0\\ {0,\mathrm{y}}_{\mathrm{j}}-{\mathrm{y}}_{\mathrm{i}}=0\\ {-1,\mathrm{y}}_{\mathrm{j}}-{\mathrm{y}}_{\mathrm{i}}<0\end{array}\right.$$

(4)

y_i and y_j are the vegetation phenology information in the i and j years, respectively, and n represents the length of the time series. The range of Z value is (–∞, +∞) and the significance level of 0.05 was used. Meanwhile, this study used partial correlation analysis to examine the partial correlations between climatic factors and phenological indices. Finally, in this study, the maximum value composite method was used to calculate each raster cell’s maximum NDVI (NDVI_max), and the Spearman correlation analysis was adopted to study the correlation between NDVI_max and different phenology parameters.

4. Results

4.1. Spatiotemporal Pattern of Vegetation Phenological Indicators

The spatial distribution of vegetation phenology on the QTP was analyzed by calculating the mean values of vegetation phenology for each pixel across 24 years of results. Comparative analysis revealed consistent results between EVI and NDVI. From a temporal perspective, SOS was advanced, EOS was delayed, and LOG was prolonged during 2001–2024. From a spatial perspective, the SOS occurred earliest in eastern regions. It progressively delayed towards the southwestern areas (Figure 2a). The EOS showed the earliest occurrence in central regions. It had a gradual postponement towards southeastern areas (Figure 2b). The LOG exhibited maximum duration in eastern regions. It gradually shortened westward (Figure 2c). The results based on EVI are similar (Figure S1).

From the perspective of the inter-annual variation in the SOS from 2001 to 2024, the SOS showed an advancing trend (Figure 3). The results of EVI were also similar (Figure S2). Still, the southwest had a significant delaying trend (Figure 4a). Across different periods, the SOS was divided into three phases: 2001–2006, 2007–2018, and 2019–2024 (Figure 3). In the eastern region, the SOS had a significant advancement before 2006, changed to a slight advancing trend from 2006 to 2018, and then turned into a delaying trend after 2019 (Figure 4b–d). In the central region, there was a strong advancing trend before 2006, which changed to a slight delaying trend from 2006 to 2018 and then turned into an advancing trend after 2019 (Figure 4b–d). In the southwestern region, there was a relatively strong delaying trend before 2006, which changed to a weaker delaying trend from 2006 to 2024 (Figure 4b–d).

With respect to the EOS, an overall delay trend was detected (Figure 3). The results of EVI were also similar (Figure S2). The northeastern and southwestern regions had significant delays, and there was a slight advancing trend in the central areas (Figure 4e). From a perspective of different periods, the EOS was partitioned into three periods, including 2001–2009, 2010–2015, and 2016–2024 (Figure 3). In the eastern regions, EOS initially showed a delaying trend, then shifted to an advancing trend, and reverted to an advancing trend again after 2016 (Figure 4f–h). In the central regions, EOS exhibited an advancing trend from 2001 to 2024 (Figure 4f–h). In the southwestern areas, a delayed trend prevailed before 2009, which shifted to a stronger advancing trend after 2010. After 2016, EOS exhibited a delayed trend (Figure 4f–h).

An overall lengthening trend of the LOG was observed (Figure 3). The results of EVI were also similar (Figure S2). In the southwestern regions, there were shortening trends, while in the central and northeastern areas, a lengthening trend was shown (Figure 4i). The LOG was divided into two periods, including 2001–2018 and 2019–2024 (Figure 3). For the LOG, the central and eastern regions exhibited a significant lengthening trend before 2018, which was shortening after 2018 (Figure 4j,k). The southwestern areas exhibited a shortening trend before 2018, which continued after 2019 (Figure 4j,k).

4.2. Phenological Characteristics Across Different Frozen Soil Regions

An advancing trend in SOS was revealed across both seasonally frozen ground and permafrost from 2001 to 2024 (Figure 5a). Despite the overall advancing trend of SOS, phased analyses indicated a shift from advancement to delay over time. Specifically, during 2001–2006 SOS was advanced in both seasonally frozen ground and in permafrost. However, from 2007 to 2018, the SOS was delayed in seasonally frozen ground and permafrost. Notably, between 2019 and 2024, it was delayed in seasonally frozen ground while it was advanced in permafrost (Figure 5a). Throughout all time periods, the SOS showed a negative partial correlation with temperature and a positive partial correlation with precipitation (

Table 1. Partial correlation analysis between climate factors and SOS from 2001 to 2023.

Periods	Precipitation	Temperature
2001–2023	0.378	−0.151
2001–2006	0.365	−0.195 *
2007–2018	0.375 *	−0.132 *
2019–2023	0.396	−0.148

Note: * Indicates significance level (p < 0.05).

).

In terms of EOS, from 2001 to 2024, seasonally frozen ground exhibited a delaying trend while permafrost exhibited an advancing trend (Figure 5b). Between 2001 and 2009, although EOS exhibited a delayed trend in seasonally frozen ground, it showed an advancing trend in permafrost. From 2010 to 2015, EOS in seasonally frozen ground shifted to advance, while permafrost continued to advance. Between 2016 and 2024, it advanced in both seasonally frozen ground and permafrost (Figure 5b). Throughout all time periods, the EOS showed a positive partial correlation with temperature and a negative partial correlation with precipitation (

Table 2. Partial correlation analysis between climate factors and EOS from 2001 to 2023.

Periods	Precipitation	Temperature
2001–2023	−0.286	0.353
2001–2009	−0.281	0.333
2010–2015	−0.309 *	0.394 *
2016–2023	−0.277	0.348

Note: * Indicates significance level (p < 0.05).

).

LOG showed a prolonged trend in both permafrost and seasonally frozen ground from 2001 to 2024 (Figure 5c). During 2001–2018, the LOG was prolonged in seasonally frozen ground and permafrost. However, from 2019 to 2024, LOG shortened sharply in seasonally frozen ground and permafrost (Figure 5c).

4.3. Phenological Characteristics Among Different Vegetation Types

The phenological changes in different vegetation types exhibited variations. Based on SOS, all three types of vegetation were advanced during 2001–2024. Between 2001 and 2006, all three types of vegetation were advanced. From 2007 to 2018, they were shifted to delay. Between 2019 and 2024, SOS for the alpine meadow and alpine steppe was delayed, while the alpine desert was advanced (Figure 6a).

Regarding EOS, the alpine steppe and alpine desert were delayed, while the alpine meadow was advanced. During 2001–2009, the alpine steppe was delayed, while the alpine meadow and alpine desert were advanced. From 2010 to 2015, the alpine meadow and alpine steppe were advanced, while the alpine desert was delayed. Between 2016 and 2024, the three types of vegetation were advanced (Figure 6b).

In terms of LOG, alpine meadow, alpine steppe, and alpine desert were prolonged during 2001–2024. During 2001–2018, the LOG for all three vegetation types was extended. Between 2019 and 2024, the LOG for the alpine meadow and the alpine steppe was shortened, while the alpine desert was extended (Figure 6c).

4.4. Analysis of Phenological Change Among Different Frozen Soils and Various Vegetation Types

Alpine meadow and alpine steppe are distributed in both permafrost and seasonally frozen ground, while alpine desert is only in permafrost. This study studied the changes in the phenology of alpine meadow and alpine steppe in permafrost and seasonally frozen ground.

For SOS, the changing trends of alpine meadow in permafrost and seasonally frozen ground were advanced, except that from 2007 to 2018 in both frozen soil and from 2019 to 2024 in seasonally frozen ground (Figure 7a,b). The alpine steppe was delayed in seasonally frozen ground while advancing in permafrost from 2001 to 2024. In other periods, its changing trends were similar in permafrost and seasonally frozen ground.

For EOS, the alpine meadow exhibited an advancing trend in permafrost and seasonally frozen ground, except for the period from 2001 to 2009 in seasonally frozen ground (Figure 7c,d). In permafrost, the alpine steppe showed an advancing trend in all different periods, while its changing trends in seasonally frozen ground varied across periods. Due to the significant delay from 2001 to 2009, although it exhibited an advancing trend in the subsequent years, it still showed a significant delay in the entire period from 2001 to 2024.

For LOG, the changing trends of alpine meadow and alpine steppe in permafrost and seasonally frozen ground were similar (Figure 7e,f). Due to their extension from 2001 to 2018, even though they shortened by a relatively large margin from 2019 to 2024, their overall trend from 2001 to 2024 still showed extension.

4.5. Variation Characteristics of Vegetation Greenness

The trends of NDVI_max differed across different time periods. From 2001 to 2024, NDVI_max increased by 4 × 10⁻⁴/year, with decreases in the southwest region (Figure 8a). From 2001 to 2018, NDVI_max increased by 9 × 10⁻⁴/year and it decreased sharply from 2019 to 2024 (−11 × 10⁻⁴/year) (Figure 8b,c). From the perspective of different frozen soil regions, NDVI_max increased by 6 × 10⁻⁴/year in the permafrost and by 3 × 10⁻⁴/year in the seasonally frozen ground.

5. Discussion

5.1. Vegetation Phenological Changes

This study found that from 2001 to 2024, SOS was advanced, EOS was delayed, and LOG was prolonged, which was consistent with previous research [9,10]. From a spatial perspective, the SOS was advanced in the east and delayed in the southwest, which was consistent with former studies [13,48]. Regarding EOS, the delayed trends in the southwest align with those noted by Liu et al. [26], where LOG was also observed.

SOS exhibited an advanced trend before 2006, which was consistent with Zhang et al. [49]. This study found that SOS was partially negatively correlated with pre-season temperature (Table 1). Therefore, warmer springs and winters were likely the main forces driving the SOS advancement [49]. Since 2007, SOS has begun to delay, with intensified delays post-2018. The delayed trend of SOS could be explained by the increase in precipitation [50] (Table 1). Since the SOS began to show a continuously increasing trend of delay in 2019, the advancing trend of the SOS on the QTP from 2001 to 2024 was already smaller than that from 2001 to 2020 [9,15]. EOS showed a delaying trend before 2009. After 2009, it turned into an advanced trend. This finding was consistent with Li et al. [51]. This advanced trend could be explained by the intricate interaction between the decreasing precipitation and increasing temperature during this period [51] (Table 2). Since 2016, EOS has presented an advanced trend. This finding matched the previous conclusion, and it could be attributed to a drier August and colder September [52]. Previous studies have shown that changes in EOS were primarily influenced by temperature, precipitation, and sunshine duration [10,53,54,55]. Additionally, some studies have demonstrated that spring phenology has a greater impact than climate change in controlling the interannual variability of autumn phenology on the QTP [56,57]. Therefore, more research efforts were required to explore the causes of autumn phenological changes. The changing trends of SOS and EOS determined the changing trend of LOG. LOG has been significantly shortened since 2019, which has resulted in the overall trend of LOG from 2001 to 2024 being much smaller than that from 2001 to 2018. Vegetation-specific analyses further highlight discrepancies. The alpine steppe exhibited SOS delay, EOS delay, and LOG extension, contrasting previous findings (SOS delay, EOS advancement, LOG shortening from 2001 to 2019) [58], likely due to continued EOS delays post-2019. Alpine meadow showed SOS advancement, EOS advancement, and LOG extension. The alpine desert displayed SOS advancement, EOS delay, and LOG extension.

5.2. Comparison of Vegetation Phenology Between Permafrost and Seasonally Frozen Soil Regions

Research on the relationship between changes in frozen soil and vegetation phenology remains limited on the QTP [56,57]. This study reveals broadly similar overall phenological trends between seasonally frozen ground and permafrost regions from 2001 to 2024. The SOS advanced in both types of frozen ground. The EOS advanced in permafrost but delayed in seasonally frozen ground. The LOG was extended in both cases. However, the rates of these changes differed. It was likely due to varying vegetation responses to frozen soil dynamics and differences in vegetation composition between permafrost and seasonally frozen ground. Specifically, SOS showed a larger advancement in permafrost than in seasonally frozen ground.

During the thawing period, permafrost exhibited higher soil moisture in the surface layer compared to the lower layer [56]. Seasonally frozen ground shows the opposite pattern. Their surface layer moisture is lower relative to the lower layer [56]. Different types of vegetation also responded differently to moisture and other factors. The SOS of alpine meadows is more strongly correlated with spring temperature. The SOS of the alpine steppe is more closely linked to spring moisture [58]. In seasonally frozen ground, the proportion of alpine meadows was significantly higher than that of alpine steppe. In permafrost, the areas occupied by alpine meadow and alpine steppe are approximately equal. This difference in vegetation composition, combined with the more favorable moisture environment in permafrost during thawing, explained the more significant SOS advancement in permafrost regions. It is attributed to the stronger dependence of steppe phenology on moisture and the superior moisture conditions in permafrost.

For EOS, this study identified divergent changes in alpine steppe phenology between permafrost and seasonally frozen ground regions. Alpine steppe in seasonally frozen ground regions was predominantly distributed in areas with relatively higher soil moisture. In permafrost regions, it was concentrated in areas with relatively lower soil moisture [34]. Additionally, the EOS exhibited a non-significant positive correlation with soil moisture content [59]. This indicated that the difference in humidity between the two frozen soil types may lead to differences in EOS trends. Other factors, such as insolation, land cover changes, albedo variations, meadow shrinkage, and interactions between frozen soil dynamics and climate change, also contribute to EOS changes [60,61,62]. However, their different impacts on specific vegetation species remain unclear.

In terms of phased trends, SOS shifts showed consistency between the two frozen ground types. They transitioned from advancement to delay around 2006. They experienced continuous delays after 2019. However, EOS trends diverged between frozen ground regions. Permafrost exhibited a continuous advancing trend. Seasonally frozen ground transitioned to advancement after 2010. For LOG, both frozen ground types followed the same phased pattern. They shifted from extension to shortening in 2018. Currently, research on the drivers behind these segmented phenological changes is relatively scarce. This underscores the need for further investigation in the future.

5.3. The Correlation Between the Variation in Vegetation Greenness and Frozen Ground

NDVI_max exhibited strong consistency with vegetation coverage [59]. During 2001–2024, NDVI_max showed an increasing trend across the QTP, aligning with a previous study [60]. However, a significant decreasing trend was observed in the southwestern region (Figure 8a), consistent with Anniwaer et al. [61]. NDVI_max increased in the permafrost, and Natali et al. attributed enhanced vegetation productivity in permafrost to SOS advancement, EOS delay, and LOG extension caused by permafrost degradation [62]. Wang et al. suggested that permafrost degradation increases surface soil moisture availability, facilitating water infiltration into deeper soil layers [4], and elevated root-zone soil moisture promotes greening [63]. Notably, NDVI_max increases in seasonally frozen ground were smaller than in permafrost, indicating weaker greening increase trends in the former.

This study found that NDVI_max trends on the QTP from 2001 to 2024 correlated with vegetation phenological shifts, likely because SOS and EOS changes drove LOG variations, and prolonged LOG increased the days of assimilation, thereby enhancing NDVI_max [64,65]. This study used the Spearman correlation analysis, and this is a non-parametric test. It is simple and distribution-free [66]. Specifically, NDVI_max trends show a positive correlation with SOS trends (Figure 9a), a negative correlation with EOS trends (Figure 9b), and a negative correlation with LOG trends (Figure 9c). These patterns may arise from warming-induced SOS advancement, which promotes rapid early-stage vegetation growth [16]. However, accelerated aboveground biomass accumulation also intensified evapotranspiration and respiration, exacerbating vegetation drought [67]. Without sufficient water replenishment, this can trigger abscisic acid accumulation, chlorophyll degradation, or leaf damage, ultimately hastening senescence and advancing EOS [68].

5.4. Limits and Future Works

This study utilized data sources with relatively coarse spatial and temporal resolutions, and a tiny fraction of the results had small values. But according to previous studies [69,70], vegetation phenology variation trends are not significantly affected by differences in temporal resolution. Therefore, these values are still reliable. Also, a study suggested that the coarse resolution may have masked some subtle changes [71]. Thus, in future research, we will adopt data with higher spatial and temporal resolutions, and compare and validate the results obtained with those derived from coarse-resolution data. Meanwhile, we also need to obtain as much in situ observation data as possible and compare it with the results derived from remote sensing data.

Due to the complexity of the mechanism by which frozen soil changes affect vegetation activities, such as the transition from freezing to thawing, which can influence vegetation phenology by regulating soil moisture dynamics and nutrient availability [4,25]. Therefore, this study did not investigate the influencing factors of phenological changes in the growth of different vegetation types on various frozen soil types, and this can be explored in future work.

6. Conclusions

This study examines the spatiotemporal trends of vegetation phenology on the QTP over 24 years (2001–2024), with a specific focus on phenological variations across different frozen soils and vegetation types. The results indicate that the overall phenological trends observed during the past 24 years were consistent with previous findings, characterized by an advanced SOS, a delayed EOS, and an extended LOG. However, distinct phase-specific shifts in trends were identified, including a delayed SOS, an advanced EOS, and a shortened LOG during certain sub-periods. This indicates that the vegetation phenology may show different changing trends in the future compared with previous ones.

Phenological changes exhibited notable differences between permafrost and seasonally frozen ground regions. In permafrost regions, both SOS and EOS were advanced, accompanied by a prolonged LOG. In contrast, seasonally frozen ground regions showed an advanced SOS, a delayed EOS, and an extended LOG. Furthermore, vegetation phenology in frozen soils displayed variation during sub-periods. In permafrost regions, the SOS exhibited a pattern of advancement, followed by delay, and subsequent advancement in the third sub-periods, while the EOS advanced consistently. This resulted in an initial prolongation and subsequent shortening of the LOG. In seasonally frozen ground regions, the SOS showed a trend of advancement, followed by two consecutive delays, while the EOS exhibited a pattern of delay, advancement, and further advancement. Consequently, the LOG was first prolonged and then shortened.

Significant disparities in phenological changes were observed among different vegetation types. Alpine meadow and alpine steppe shared similar SOS trends, characterized by an initial advancement, followed by a delay, and a subsequent continuous delay. In contrast, the alpine desert exhibited an SOS pattern of advancement, followed by delay, and then renewed advancement. Regarding EOS trends, the alpine meadow showed consistent advancement, while the alpine steppe displayed a pattern of delay and two consecutive advancements. The alpine desert was advanced, delayed, and advanced. For the LOG, alpine meadows and alpine steppe showed an initial prolongation followed by shortening, whereas the alpine desert experienced consistent prolongation.

Notably, the same vegetation type exhibited divergent phenological changes in permafrost and seasonally frozen ground regions, particularly during specific sub-periods. For example, during 2019–2024, the SOS of the alpine meadow was advanced in permafrost but delayed in seasonally frozen ground regions. Similarly, during 2001–2009, the EOS of alpine meadow was advanced in permafrost regions, but it was delayed in seasonally frozen ground regions. Additionally, during 2001–2006, the EOS of the alpine steppe advanced in permafrost regions but was delayed in seasonally frozen ground regions. Our research findings indicate that the phenology of different vegetation types varies across various types of frozen ground, providing insight for further studies on the response of vegetation to frozen soils. The processes and mechanisms underlying the impacts of environmental changes and permafrost dynamics on vegetation phenology are intricate and multifaceted, representing a key focus for future investigations. Moreover, disparities in spatial and temporal resolution may introduce uncertainties into phenological simulation, thereby underscoring the need for further research to address these potential sources of variability.