Introduction

Vegetation phenology is the study of periodically recurring natural phenomenon and its relationship with cycles of environmental conditions. It reflects the response of the Earth’s biosphere toclimate and the seasonal and interannual variation in the water cycle (Zhang et al. 2003).In the past 15 years, vegetation phenology has played an important role in global change science research(Richardson et al. 2013). The committee on phenology of the International Society for Biometrics(ISB) seeks to strengthen the understanding of the climate change affecting plant dynamics, and one of the three sessions of the meeting in 2011 was entitled “phenology, climate change and adaptation.” However, early studies on vegetation phenology mainly focused on field observation, which limited the spatial scale of vegetation phenology studies. With the development of remote sensing technology and used in vegetation phenology research in recent years, it is possible to study the temporally and spatially dynamic characteristics of vegetation phenology and their relationships to environmental factors at larger spatial scales.

Beginning of the growing season (BGS) is one of the main plant physiological parameters in the field of vegetation phenology around the world (Morisette et al. 2009). At large scales, the spatial pattern of BGS in high-latitude or high-elevation areas shows zonal latitudinal changes. In local areas, BGS is influenced by topographic features (elevation and aspect) that control climatic and environment conditions, resulting in spatial heterogeneity (Ziello et al. 2009; Song et al. 2011; Wang et al. 2011). At present, a large number of international studies have confirmed that temperature and water conditions are the two key physical environmental factors controlling the spatial pattern of BGS. The spatial pattern of BGS shows a significant correlation with the spring average temperature (Chen et al. 2005), and the temperatures in different months exert various influences on BGS along different altitudinal gradients (Čufar et al. 2012). In arid and semi-arid grassland ecosystems, the water supply is the primary environmental factor for determining vegetation activity (Zhang et al. 2005). Compared with the rainfall, the soil moisture controlled by rainfall has more significant positive correlation with the vegetation phenology (Hashimoto et al. 2001). In addition, the vegetation coverage within an ecosystem also affects the vegetation phenology by changing the surface microclimate (Seghieri et al. 2009). Although numerous studies have focused on the spatio-temporal pattern of vegetation BGS and the relationship between one or two influencing factors, there are few studies on the contributions of multiple environmental factors to the spatial pattern of BGS and the spatial relationship between individual factors and vegetation BGS.

A large number of studies have shown that under the background of climate change trends, the BGS in most parts of the world has been advancing at the rate of 2.3–5.2 days/10 years since 1970s (Rosenzweig et al. 2007). However, in addition to the rainfall and temperature influence the dynamic of BGS, it is also affected by other factors, such as radiation(Verger et al. 2013), photoperiod (Körner and Basler 2010), winter chilling (Zhang et al. 2007), snow cover (Wang et al. 2013), and concentration of carbon dioxide (Kimball et al. 2002; Cleland et al. 2006). The BGS in some regions is even significantly affected by the duration of certain phases of the North Atlantic Oscillation Index (Maignan et al.2008). In high-latitude or alpine ecosystems, the timing of melting and the temperature after the melting phase are considered the key factors controlling the dynamic trend of BGS under climate change trend (Borner et al. 2008; Steltzer et al. 2009). In contrast, in arid and semi-arid areas where the soil moisture plays an essential role in triggering the vegetation BGS (Bisigato et al. 2013). It is unclear how multiple factors of environmental conditions influence the vegetation phenology at different spatial scales, especially such as local surface temperatures and humidity (Morisette et al. 2009). Hence, the quantitative analysis of the spatio-temporal dynamic variation in the relationships between BGS, temperature, and soil moisture remains an active topic of research (Peñuelas et al. 2009).

The study of vegetation phenology is simple on the transect scale, but vegetation phenology is also affected by other natural and man-made pressures in the spatial scale, such as vegetation degradation, grazing disturbance, etc. (Alessandri et al. 2007; Goheen et al. 2007).Vegetation degradation can have a direct or indirect negative effect on vegetation phenology, especially in the vegetation ecosystem limited by soil moisture. One of the characteristics of vegetation degradation is the decrease of vegetation height and the reduction of above ground biomass, and these phenomena indirectly affect vegetation phenology by changing near-surface microclimate (Moriseete et al. 2009). Chen et al. (2011b) advocated vegetation coverage decrease after vegetation degradation would reduce LST by increasing the surface albedo, for the final performance of the BGS delay. Climate change has been considered as a key factor affecting BGS in arid and semi-arid areas for a long time, ignoring the importance of vegetation degradation for BGS (Bisigato et al. 2013), but this problem is considered as a new direction for the study of plant phenology in the future(Peñuelas et al. 2009).

The Qinghai-Tibet Plateau is known as the “third pole” and the “Asian water tower.” It is an important region containing typical alpine grassland ecosystem in China. In the last 10 years, research on the phenology of alpine grassland in the Qinghai-Tibet Plateau has become a hot topic at home and abroad (Yu et al. 2010; Shen et al. 2011; Wang et al. 2013). In the early period, considerable research was performed on the spatial pattern and the spatio-temporal variation in vegetation phenology on the Qinghai-Tibet Plateau, while the results of the study on the alpine grassland trends were controversial. Some scholars found that the BGS in the Qinghai-Tibet Plateau alpine grassland showed overall advancing trend (Yu and Xu 2009; Li et al. 2010; Ding et al. 2011; Zhang et al. 2013; Li et al. 2014). There are also some scholars who found that the BGS showed a delaying trend (Yu et al. 2010; Piao et al. 2011). Through choosing different data sources in the experiments, it has been shown that the differences in the results were due to the remote sensing data quality of the western plateau (Zhang et al. 2013). Later research focused on the factors that influence the variation in vegetation phenology of alpine grassland ecosystem. Although the factors affecting the BGS in alpine grassland ecosystem are complicated (Chen et al. 2011b), the temperature and water conditions are two key factors influencing the BGS, especially in alpine ecosystems restricted by soil moisture in arid and semi-arid areas (Liu et al. 2014a; Zhou et al. 2016). However, numerous studies have focused on observational data at meteorological sites, ignoring the studies on the relationship between BGS and the spatio-temporal variation in hydrologic and thermal conditions at the regional spatial scale. This style of research is not conducive to the analysis of the contributions of the hydrologic and thermal conditions to the changes in BGS at the regional scale and does not contribute to better understanding the potential influencing factors of the spatio-temporal variation in the BGS.

Qinghai Lake watershed is located on the Qinghai-Tibet Plateau in sensitive semi-arid climate zone of Asia. The region has unique alpine semi-arid ecosystem characteristics and is a complex ecosphere, including natural, social, and economic aspects. It has attracted the attention of a wide range of domestic and foreign researchers and has become the site of experimentation where scientific community study climate change, the ecological environment, and other issues on the Qinghai-Tibet Plateau (Hao 2008). First of all, the study clarifies the contributions of elevation, aspect, mean spring LST, soil moisture, and maximum vegetation coverage during the year to the spatial pattern of BGS and analyzes the relationships between these key factors and the spatial pattern of BGS. Then, we studied the variation characteristics of the spatio-temporal pattern of BGS in Qinghai Lake watershed based on BGS inversion data using phenology models from 2000 to 2016, and separately analyzed how spatio-temporal variation of LST, soil moisture, and vegetation degradation affect BGS. This work was performed to further understand the effects of dynamic spatial changes in potential impact factors on alpine grassland ecosystem BGS in the special ecological environment of the Qinghai Lake watershed.

Materials and methods

Study area

Qinghai Lake is the largest inland saltwater lake in China, located in the northeast part of the Qinghai-Tibet Plateau (99°36′–100°16′ E, 36°32′–37°15′ N) at an average elevation of 3195 m (Fig. 1). The Qinghai Lake watershed covers an area of 29,700 km2 and is typical of arid and semi-arid climatic zones. The average annual temperature in the watershed ranges from − 0.8 to 1.1 °C, and the annual precipitation ranges from 324.5 to 412.8 mm, with a decreasing trend from northeast to southwest. The terrain slopes from northwest to southeast, and the landforms are complex, with multiple alpine distributions. Mountainous areas occupy 68.6% of the total watershed. The main rivers in the watershed are the Buha, Shaliu, Haergai, and Daotang. The Buha is the largest river that flows into Qinghai Lake and represents the entire western portion of the watershed, accounting for 48.76% of the total annual runoff in the watershed. Grassland and water are the two major land cover types in the watershed, accounting for 70 and 16% of the area, respectively.

Fig. 1
figure 1

Location of the Qinghai Lake watershed and its land cover types in 2012

Full size image

Data sources

Land cover data of Qinghai Lake watershed were derived from updated data of land use in 2012 by the land and resources administration department of Qinghai province. The remote sensing data were obtained from the National Aeronautics and Space Administration (NASA) website (http://ladsweb.nascom.nasa.gov/data/). The data are the MOD13Q1 16-day synthetic enhanced vegetation index (EVI) from the MODIS sensor aboard the Terra and Aqua satellites and the MOD11 8-day synthetic LST data products, with spatial resolutions of 250 m and 1000 m, respectively. The data cover the time span of 2000–2016.The surface temperature data are selected from the surface temperature mean data from 15 April to18 June of each year during the study period to represent the spring data. Before the application of all the data, the MODIS Reprojection Tool software was used to cover all images to the projection conversion batch. Then, using Erdas software to stitch and cut the two images at the same time in the study area, the EVI and LST data for the required range were obtained. EVI enhances the vegetation signal and corrects the soil background and the aerosol scattering effects through adding the blue band, which directly reflects the degree of lushness of the surface vegetation (Gao et al.2000). The maximum EVI (EVImax) during the year represents the largest vegetation coverage during the year. Vegetation degradation caused by the decrease of vegetation activity in long time series can be reflected by the dynamic change of vegetation index. EVImax in every year during 2000–2016 were used to reflect the trend of vegetation degradation in the Qinghai Lake watershed in this study. In this paper, we use the maximum value composite (MVC) method to obtain EVImax of each year during the study period and use raster calculations to obtain the average EVImax data for the different years (Holben 1986).

The digital elevation model (DEM) data had a spatial resolution of 30 m and were processed by the first version (V1) of ASTER GDEM. The data were obtained from the Geospatial Data Cloud site, Computer Network Information Center, Chinese Academy of Sciences (http://www.gscloud.cn).The aspect data were produced from DEM data in the ArcGIS10.2 environment and described in accordance with the clockwise direction, with angles ranging between 0 (due north) and 360 degrees (due north).

Spatial pattern of 0–10 cm volumetric soil water content in past springs during 2000 to 2016 in this study was obtained using the temperature-vegetation drought index (TVDI) model inversion (Sandholt et al. 2002). In order to correct soil moisture retrieval data for past springs, we use the linear relational expression (y = − 0.5980x + 0.6789 r 2 = 0.6541), which is based on 30 surface soil moisture data points in the watershed measured from May 15 to 20, 2014, and the inverted soil moisture content for the same period and position. The 30 sample points were evenly arranged in the Qinghai Lake watershed and spanned different elevations. The soil samples were collected using a 100 cm3cutting ring and were dried at 105 °C for 24 h in an oven.

Methods

Phenology model

In the ArcGIS environment, we individually standardized the sequential sequence of EVI data then used the asymmetric Gaussian function fitting method to reconstruct the normalized EVI data in time series. The threshold method was used to extract the BGS during the study period (White et al. 1997).With reference to the study of the BGS in Qinghai Lake watershed, the threshold of the BGS was set to 0.1 (Li et al. 2014). The vegetation phenological calculations are based on January 1 as the beginning of the year as the starting point for standardization, i.e., January 1 is the first day, January 2 is the second day, and so on, to determine the BGS. The threshold method is as follows:

$$ {\mathsf{EVI}}_{\mathsf{Ratio}}=\frac{\mathsf{EVI}-{\mathsf{EVI}}_{\mathsf{min}}}{{\mathsf{EVI}}_{\mathsf{max}}-{\mathsf{EVI}}_{\mathsf{min}}} $$
(1)

In the formula, EVIRatio is output ratio (the range is 0 to 1), EVI is a raster enhanced vegetation index values, EVImin is the minimum of the year, and EVImax is the maximum of the year.

Temperature-vegetation drought index model

There is a negative correlation between vegetation index and LST based on remote sensing information, and if the coverage of the vegetation contains the bare land to full coverage, soil moisture contains extremely arid to very moist, the scatter diagram with EVI and LST are horizontal and vertical coordinates in this region respectively shows a triangular feature space. The dry surface of triangle is connected with the maximum LST (Tsmax) under different vegetation index, which represents the upper limit of drought in arid area, and the drought index is 1. On the contrary, the minimum LST (Tsmin) of different vegetation indices is linked together to form a triangular wet edge, representing the wettest area in the area and the drought index being zero. Sandholt et al. (2002) combined the interpretation of the ecological features of the feature space and put forward the temperature-vegetation drought index (TVDI) to estimate the soil moisture content of the surface soil.

$$ \mathsf{TVDI}=\frac{\mathsf{Ts}-{\mathsf{Ts}}_{\mathsf{min}}}{{\mathsf{Ts}}_{\mathsf{max}}-{\mathsf{Ts}}_{\mathsf{min}}} $$
(2)
$$ {\mathsf{Ts}}_{\mathsf{max}}={\mathit{\mathsf{a}}}_{\mathsf{1}}+{\mathit{\mathsf{b}}}_{\mathsf{1}}\mathsf{EVI} $$
(3)
$$ {\mathsf{Ts}}_{\mathsf{min}}={\mathit{\mathsf{a}}}_{\mathsf{2}}+{\mathit{\mathsf{b}}}_{\mathsf{2}}\mathsf{EVI} $$
(4)

Tsmax and Tsmix are obtained by linear fitting of dry boundary and wet edge according to feature space, a 1, b 1, a 2, and b 2 are fitting equation coefficients, respectively. The soil moisture of Qinghai Lake watershed was retrieved by EVI and LST in 2014; the dry side and wet side fitting equation of characteristic space were shown at Fig. 2.

Fig. 2
figure 2

EVI-LST feature space scatter diagram in 2014

Full size image

Redundancy analysis

Redundancy analysis was to evaluate the relationship between one or a set of variables and the other group of multivariate data from a statistical point of view (Ter-Braak and Smilauer, 2002). Its advantage is that it can independently maintain the contribution rate of each environment variable and effectively test the explanatory variable and determine the minimum variable group with the maximum explanatory power for the change of the response variable. The two-dimensional sorting diagram can directly explain the relationship between the response variables and the environment factors. In the two-dimensional sorting chart, the angle between the arrows indicates the correlation among the variables. The smaller the angle is the higher correlation is. If the angle is the same direction, it indicates positive correlation, and the reverse indicates negative correlation. Arrow length of quantitative environment factors represents the degree of influence of environmental factors that affect variables. In this study, Vegan 3.0.2 software was used to analyze the original variables and typical variables of the sample points. Vegan is an external package in the R language and is widely used in community ecology analysis (Oksanen et al. 2010). It can provide all the basic sorting methods and produces beautifully ordered graphs.

Linear regression method

The linear regression trend method is a model of regression analysis of variables that change over time. The model uses the ArcGIS raster calculator function, and the raster cells are used to analyze the changing trend of the variable elements year by year, which reflects the spatio-temporal dynamic characteristics of vegetation and environmental factors. The model is as follows:

$$ \mathit{\mathsf{T}}=\frac{\mathit{\mathsf{n}}\times \sum \limits_{\mathit{\mathsf{y}}=\mathsf{1}}^{\mathit{\mathsf{n}}}\mathit{\mathsf{y}}\times {\mathit{\mathsf{P}}}_{\mathit{\mathsf{y}}}-\sum \limits_{\mathit{\mathsf{y}}=\mathsf{1}}^{\mathit{\mathsf{n}}}\mathit{\mathsf{y}}\sum \limits_{\mathit{\mathsf{y}}=\mathsf{1}}^{\mathit{\mathsf{n}}}{\mathit{\mathsf{P}}}_{\mathit{\mathsf{y}}}}{\mathit{\mathsf{n}}\times \sum \limits_{\mathit{\mathsf{y}}=\mathsf{1}}^{\mathit{\mathsf{n}}}{\mathit{\mathsf{y}}}^{\mathsf{2}}-{\left(\sum \limits_{\mathit{\mathsf{y}}=\mathsf{1}}^{\mathit{\mathsf{n}}}\mathit{\mathsf{y}}\right)}^{\mathsf{2}}}\times \left({\mathit{\mathsf{Y}}}_{\mathit{\mathsf{n}}}-{\mathit{\mathsf{Y}}}_{\mathsf{1}}\right) $$
(5)

Where n is the total number of studies period in the formula, y is the year y in the study period, P y is the variable element data for year y, Y n is the end of the study period, Y 1 is the first year of the study period, and T is the variable for variable elements in the study period (when T > 0, the variable increases (extends), when T < 0, the variable decreases (shortens)).

Correlation coefficient model

The Pearson correlation coefficient is a coefficient that expresses the degree of straightness between two variables. This method can analyze the correlation between two time series of variable data in the ArcGIS runtime environment at the pixel scale (Chen et al. 2008). The closer the correlation coefficient is to 1 or − 1, the stronger the correlation is; the closer the correlation coefficient is to 0, the weaker the correlation is. When the correlation coefficient is positive, there is a positive correlation between the two variables, and vice versa. In general, absolute values of the correlation coefficient of 0.8 ~ 1.0 are considered extremely strongly correlated, 0.6 ~ 0.8 are considered strongly correlated, 0.4 ~ 0.6 are considered moderately correlated, 0.2 ~ 0.4 are considered weakly correlated, and 0.0 ~ 0.2 are considered extremely weakly or not correlated. The Pearson correlation coefficient model is as follows:

$$ {\mathit{\mathsf{R}}}_{\mathsf{GE}}=\frac{\sum \limits_{\mathit{\mathsf{i}}=\mathsf{1}}^{\mathit{\mathsf{n}}}\left({\mathit{\mathsf{P}}}_{\mathit{\mathsf{i}}-}\overline{\mathit{\mathsf{P}}}\right)\left({\mathit{\mathsf{E}}}_{\mathit{\mathsf{i}}}-\overline{\mathit{\mathsf{E}}}\right)}{\sqrt{\sum \limits_{\mathit{\mathsf{i}}=\mathsf{1}}^{\mathit{\mathsf{n}}}{\left({\mathit{\mathsf{P}}}_{\mathit{\mathsf{i}}-}\overline{\mathit{\mathsf{P}}}\right)}^{\mathsf{2}}\sum \limits_{\mathit{\mathsf{i}}=\mathsf{1}}^{\mathit{\mathsf{n}}}{\left({\mathit{\mathsf{E}}}_{\mathit{\mathsf{i}}-}\overline{\mathit{\mathsf{E}}}\right)}^{\mathsf{2}}}} $$
(6)

In the formula, R GE is the correlation coefficient of two variable data, P i is variable data in i year, P is the average value of the study period, E i is another variable data in i year, and E is the average of the variables for the study period.

Results

Spatial pattern and spatio-temporal variation of BGS

Alpine grassland ecosystem multi-year average BGS in the watershed shows a horizontal trend of delay from southeast to northwest, but local areas exhibiting spatial heterogeneity are mainly distributed in the western and northern regions of the Qinghai Lake watershed (Fig. 3). The alpine grassland in the Qinghai Lake watershed began to enter the BGS in middle to late April (days 100 ~ 120), and the average BGS is day 144.4. The period from the mid-May to early June (days 130–160) is the peak of the vegetation experiencing the BGS in the watershed, and 85.9% of the BGS dates were concentrated in this period. In early June (days 150–160), 95.0% of the vegetation in the watershed entered the BGS, and the BGS of the whole watershed vegetation lasted approximately 60 days. The southern shore of Qinghai Lake, the Daotang river sub-watershed and the Buha River downstream into lake are the regions where the vegetation entered the BGS the earliest in the watershed. The BGS began to expand to the surrounding areas over time. In the western and northern parts of the mountainous areas, the BGS of vegetation shows a non-zonal vertical pattern. The vegetation in the northwestern part of the watershed is the last to enter the BGS, occurring primarily in mid-June (days 160–170). The BGS of the farmland on the northern and southern shores of the Qinghai Lake, which grows rape (Brassica campestris L.), is concentrated in early June (days 150–160).

Fig. 3
figure 3

Spatial pattern and dynamic changes of alpine grassland BGS in the Qinghai Lake watershed during 2000–2016

Full size image

The average value of BGS in the Qinghai Lake watershed was 2.2 days ahead (SD = 11.5391) during 2000–2016 according to the raster data statistics in Fig. 3b, and the rate was 1.4 days/10 years. The BGS trend showed obvious spatial heterogeneity. The BGS in the northwestern portion of the watershed, the western side of Qinghai Lake, and the eastern part of the Daotang river sub-watershed showed obvious delays of more than 10 days, and these areas account for 16.0% of the total area of vegetation cover. The BGS in the north-central region of the watershed and the northeastern and southern areas of Qinghai Lake showed an advanced trend about 6.0 days ahead of schedule, accounted for 25.1% of the vegetation area in the watershed. Because of the impact of land use change, the BGS of areas that had been converted from grassland to shrub land was significantly earlier approximately 10 ~ 20 days.

The relationship between BGS spatial pattern and environmental factors

Based on the ArcGIS environment, 100 samples were randomly selected in the study area, and the minimum distance allowed between the random points was set to 5 km. Data on the BGS, elevation, aspect, soil moisture, LST, and mean EVImax over the years were collected for the random samples. According to Fig. 4, five environmental factors could explain 94.18% of BGS spatial pattern characteristic. The mean spring LST (r 2 = 0.3921), elevation (r 2 = 0.4437), and 0–10 cm soil moisture (r 2 = 0.3437) are the main environmental factors controlling the multi-year average BGS in the watershed. In addition, elevation, soil moisture, and temperature exert contrasting effects on the BGS, and the position of LST in controlling spatial pattern of BGS is better than soil moisture. The effect of aspect and mean EVImax over the years on the overall pattern of BGS is not obvious.

Fig. 4
figure 4

RDA results between BGS and environment factors. El represents elevation, As represents aspect, T represents mean spring LST, M represents 0-10 cm soil moisture, and E represents the mean EVImax over the years

Full size image

The results of single factor analysis show that the spatial pattern of multi-year average BGS with elevation and LST separately exhibit quadratic parabolic patterns (Fig. 5), the watershed BGS is first delayed then advances with the increase of elevation and temperature. There are turning points in the relationships between BGS, elevation, and temperature in areas where the elevation is approximately 4200 m and the mean spring LST is about 8.2 °C. There is a significant negative linear correlation between soil moisture and the spatial pattern of BGS and shows significantly delayed with increasing soil moisture.

Fig. 5
figure 5

Correlation of BGS with elevation, land surface temperature and 0-10cm soil moisture of watershed sample points

Full size image

The relationship between BGS spatio-temporal variation and environmental factors

The average LST in spring decreased by 0.39 °C (SD = 0.8212) at a rate of 0.23 °C/10 years in the watershed (except Qinghai Lake) from 2000 to 2016 showed significant spatial heterogeneity (Fig. 6a). The portion of the watershed that experienced the LST decrease of more than 0.5 °C represents 43.42% of the total area, and 21.42% of the regional LST decreases were more than 1.0 °C and were mainly concentrated in the western and north-central portions of the watershed and along southern shore of Qinghai Lake. The area where LST obviously increased is concentrated, and the area with temperature increases greater than 0.5 °C accounts for 14.79% of the total area and is mainly distributed along the western and northern shores of Qinghai Lake and at high elevations around the watershed. Although the LST in the spring in some areas of Qinghai Lake increased and decreased significantly, the average correlation coefficient between LST and the spatio-temporal dynamic variation of BGS was only 0.0064 (SD = 0.190), which was not an obvious correlation (Fig. 6b).The results show that the change in LST in Qinghai Lake watershed was not the primary driver of the dynamic changes in BGS in the watershed.

Fig. 6
figure 6

Spatio-temporal variation of LST in spring and its relationship with BGS during 2000–2016

Full size image

In the alpine grassland ecosystem, the soil moisture at 0–10 cm decreased by 4.08% (SD = 0.0755) in spring. The soil moisture of the surface soil showed a decreasing trend at a rate of 2.55% per 10 years and showed spatial heterogeneity in terms of the spatio-temporal dynamic changes (Fig. 7a).The area of significantly reduced soil moisture is mainly distributed along the southern shore of Qinghai Lake and in the northern part of the watershed. The area in which the soil moisture decreased by more than 10% accounted for 20.92% of the total area. In the northern part of the watershed, the western high-elevation area and the northern and eastern shores of Qinghai Lake, the soil moisture showed a slight increase trend. The area in which the soil moisture increased by more than 10% accounted for only 2.17% of the region. The soil moisture content exhibited an overall positive correlation with the dynamic changes in BGS, with an average correlation coefficient of 0.1532 (SD = 0.2657) (Fig. 7b). 26.34% of the alpine grassland BGS showed a weak correlation with soil moisture, 15.39% showed a moderate positive correlation, and 3.44% showed a strong correlation. The regions with moderate positive correlations and above were concentrated in the western and northern parts of the Qinghai Lake watershed. The regions in which alpine grassland BGS and soil moisture showed weak negative correlations and above accounted for only 10.27% of the total area. The results show that the soil moisture plays an important role in the spatio-temporal changes in BGS.

Fig. 7
figure 7

Spatio-temporal variation of soil moisture in spring and its relationship with BGS during 2000–2016

Full size image

Vegetation dynamics showed obvious spatial heterogeneity under the interference of multiple factors such as climate change and human activities (Fig. 8a). EVImax showed a significant increasing trend in western region indicating that the vegetation growth tends to flourish, especially in the middle reaches of the Buha River. EVImax showed a large decreasing trend in the middle and northern part of the watershed, this means vegetation degradation trend was obvious. The decreasing trend of EVImax in the area adjacent to northern and east coast of Qinghai Lake was especially obvious, because of the vegetation cover type changed under the implementation of returning farmland to grassland or forest. Spatio-temporal pattern between BGS and EVImax showed a positive correlation at watershed scale, and average correlation coefficient is 0.1148 (SD = 0.2565). 25.34% of the alpine grassland BGS showed a weak correlation with EVImax, 11.49% showed a moderate positive correlation, and 2.15% showed a strong correlation (Fig. 8b). This shows that vegetation degradation in Qinghai Lake watershed degradation affects the change of BGS, vegetation degradation trend would lead to BGS in advance, but lush vegetation trend would result of BGS delay.

Fig. 8
figure 8

Spatio-temporal variation of EVImax and its relationship with BGS during 2000–2016

Full size image

Discussion

Effect of environment factors on BGS spatial pattern

Qinghai Lake watershed on the Qinghai-Tibet Plateau has experienced a particularly significant warm and humid climate change trend (Liu et al. 2008). The annual average temperature has increased at a rate of 0.28 °C/10 years (Chen et al. 2011a), but the spring average temperature has decreased, due to the Qinghai Lake watershed is located on the northeastern edge of the Qinghai-Tibet Plateau and is affected by the area near northeastern China, where the average temperature in the spring has obviously decreased (Duan and Xiao 2015). On the other hand, Qinghai Lake watershed is an important pasture and tourist area in northwest China. Grassland degradation in some areas has been caused by overgrazing (Li et al. 2016), and the grassland was reclaimed for farming in order to meet the need for agricultural tourism, as rape sightseeing is quite important (Jiang et al. 2016). Overgrazing and farming directly lead to an increase in the area of bare surfaces (Harris 2010), increased surface reflectance, decreased contributions to near-surface latent heat exchange, and decreased surface temperatures.

Vegetation is the key link in the material and energy exchanges among the soil, water, and atmosphere, and its spatial pattern of BGS reflects the spatial pattern of environmental conditions. The responses of BGS to temperature and moisture are not isolated but interact and affect each other. The elevation directly determines the spatial variation in temperature and the characteristics of the local precipitation pattern, and precipitation is the key factor controlling soil moisture. Therefore, elevation, LST, and soil moisture play the same important role in controlling the pattern of multi-year average BGS. The terrain of the Qinghai Lake watershed is high in the northwest and low in the southeast, and the annual average precipitation decreases from northeast to northwest. These variation results in the spatial pattern of alpine grassland BGS in the watershed, which is characterized by zonal horizontal changes from southeast to northwest. Meanwhile, the BGS also has a close relationship with elevation, what is observed in alpine grasslands throughout the whole Qinghai-Tibet Plateau (Ding et al. 2012). The vertical variation in the climate in the western mountainous region of the watershed determines the non-zonal vertical phenology of the vegetation, and the spatial heterogeneity of the alpine grassland BGS is localized.

At elevations above 4200 m, the BGS is no longer delayed with elevation increased, and this phenomenon also exists in other parts of the Tibetan Plateau (Shen et al. 2014). The main reason is that the elevation of 4200 m is the annual snow line position in the Qinghai Lake watershed, the alpine grassland ecosystem begin to transition into alpine stony sparse vegetation ecosystems at this altitude, and the vegetation in the different vegetation ecosystems has different responses to changes in environmental factors (Piao et al. 2006). The initial growth of vegetation growing above this elevation needs lower temperatures and accumulated temperatures than those required by the alpine grassland (Piao et al.2011; Fu et al. 2014). Thus, the temperature is no longer the key factor for the BGS; soil moisture and air humidity are more important for BGS.

Effect of hydrothermal conditions on spatio-temporal change of BGS

The average BGS in the Qinghai Lake watershed showed an advancing trend during 2000–2016 which was the same as the whole Qinghai-Tibet Plateau based on vegetation phenology data observed by weather stations (Yu and Xu 2009); Li et al. 2010; Ding et al. 2011). However, this trend is not consistent with the results of Shen et al. (2014) who used multi-source remote sensing image data to determine the average BGS trend on the Qinghai-Tibet Plateau during the period of 2000–2011(Shen et al. 2014). The reason may be that they only used the 12-year time series of data to discuss the interannual trend of vegetation phenology on a shorter time scale, and its regularity might not be obvious. The rate of BGStrend in the alpine grassland ecosystem is 1.4 days/10 years, slightly faster than the rate of 1.04 days/10 years for the Qinghai-Tibet Plateau (Zhang et al. 2013). This may also be due to differences in the length of the study period.

The mean spring LST in the Qinghai Lake watershed showed a decreasing trend, while the alpine grassland BGS in the watershed advanced, and there was no significant correlation between each other. This finding is consistent with the results suggesting that the effect of spring LST on the alpine grassland BGS throughout the whole Tibetan Plateau is not obvious (Liu et al. 2014b). In the specific study area, the alpine grassland BGS showed a significant correlation with soil moisture. The main reason is that the annual precipitation in the Qinghai Lake watershed is mainly concentrated in July, the spring precipitation is relatively small, and the climate is relatively dry (Li et al. 2009).Therefore, due to the small supply of water, the environmental stress increased under the springtime arid and semi-arid environmental conditions. Thus, the soil moisture plays a more important role in the spatio-temporal variation of BGS (Cong et al. 2017).

Effect of grassland degradation on spatio-temporal change of BGS

The Qinghai Lake watershed is an important livestock production area in China, and grassland degradation is one of the main forms of ecological environment deterioration. The most obvious and direct changes of grassland degradation is to reduce vegetation average height and coverage of underlying surface, or the litter decreased in early BGS. Different characteristics of underlying surface can improve the surface runoff, rainfall infiltration, and evapotranspiration of soil moisture, ultimately change the near-surface soil micro environment, thus affecting the vegetation BGS (Deutsch et al. 2010). Difference of soil nutrient content under different degradation gradients would also change the vegetation BGS. With the aggravation of grassland degradation in alpine grassland ecosystem, the contents of organic carbon, total phosphorus, and available nitrogen in soil would decrease gradually (Li et al. 2008; Lin et al. 2007), and the main plant of Cyperaceae, Gramineae, and some ruderal showed BGS in advance (Ye et al. 2014). This is consistent with the conclusions that grassland degradation would lead to BGS in advanced in this study.

Vegetation degradation is not only manifested by the decrease of vegetation biomass and coverage, but also by the change of dominant species in vegetation communities that result of the disturbance of heavy grazing for a long time (Tong et al. 2004). The phenological characteristics of different plant species in alpine grassland ecosystem are different, and their responses to climate change are also different. Whether spatio-temporal pattern of BGS abnormal regional is related to species succession under grassland degradation still needs further research.

Conclusions

The complex terrain conditions and unique ecological environment of the Qinghai Lake watershed result in complex factors that influence the alpine grassland BGS. The spatial pattern of alpine grassland multi-year average BGS in the watershed from 2000 to 2016 is affected by elevation, LST, and soil moisture and is increasingly delayed from southeast to northwest. Alpine grassland BGS shows spatial heterogeneity in the local region because of the effects of large differences in altitude and climatic factors. The correlations among aspect, vegetation coverage, and BGS are relatively small. Since the alpine grassland ecosystem near the elevation of approximately 4200 m begins to transition into alpine stony sparse vegetation, the correlations between BGS, elevation, and temperature change. These relationships no longer show that the delay of BGS with the increasing elevation and decreasing temperature. Regardless, there is always a significant negative correlation with soil moisture.

Under the background of climate change, the spatio-temporal dynamics of alpine grassland ecosystem BGS in the watershed showed obvious spatial heterogeneity, but the overall trend was an advance. However, the interannual variability in LST in spring had no obvious effect on the BGS of the alpine grassland. In the local study area, the 0–10 cm soil moisture content in the spring controls the spatio-temporal dynamic changes in alpine grassland BGS in the watershed, indicating that the soil moisture in arid and semi-arid areas plays an important role in controlling the BGS in alpine areas. In recent years, the human activities and overgrazing in the Qinghai Lake watershed have led to the degradation of the alpine grassland. Grassland degradation trend caused BGS in advance through change in the surface microclimate and soil physical and chemical properties. However, this part of the research remains insufficient and would become the focus of the next step. The results would be of great significance for further understanding the response of vegetation phenology to climate change.