**4. Discussion**

*4.1. Controls on the EOS and EOS Turning Points*

This study is the first to demonstrate pixel-scale spatial heterogeneity of the EOS turning points and explain the turning point controls. The results indicate that the joint effects of climate variables and human activities are the main controls of the EOS turning points. The response of the EOS to environmental changes is complex. Some previous studies indicated that temperature plays a crucial role in EOS regulation [49] however, we show that the temperature control over the EOS is regulated by precipitation and insolation in the meadow and grassland ecotones. The cause of the turning points in most subregions is the abrupt change of temperature and precipitation. The results also reveal that insolation contributed considerably to the EOS changes, which is consistent with some previous reports that the EOS and its relation with GPP is mainly limited by insolation [50]. Other studies have reported that meadow shrinkage, decreased land cover, land albedo changes, and permafrost and seasonal frozen soil dynamics intersect with climate change, which alters the EOS trends [51].

Grazing is the most important human activity that affects grassland dynamics on the Qinghai-Tibetan Plateau [52]. The spatial heterogeneity of community increases, community function alteration, and biodiversity loss are considered to be some of the key disturbances that result in grassland degradation [53,54]. The pika population could also increase the effects of animal distribution on vegetation [55]. Overgrazing reduces the vegetation biomass and height, and restricts the regrowth ability of grassland. Our analysis shows that grazing activities in Qinghai notably decreased around 1998, coinciding with the implementation of national conservation policies (e.g., ecological compensation, restoration of degraded grassland). Grazing in Tibet was not active before 1995 and then rapidly increased, however grazing decreased after 2005 due to the late implementation of ecological conservation projects. The primary industry (mostly agriculture and animal husbandry) increased by nearly a factor of five in 1996–2015 compared with that in 1982–1994, which is also consistent with the EOS change turning points. The tertiary industry in Qinghai and Tibet quickly increased after the turning points, which indirectly reflects the intensification of human activities on the Qinghai-Tibetan Plateau.

#### *4.2. Ecological Significance of the EOS and Its Turning Points*

Phenological changes have grea<sup>t</sup> effects on the structure and function of ecosystems. At the community level, various species have different phenological responses to climate change, whereas the EOS can lead to a change in the competition for light and water conditions [17,56]. Moreover, plant species changes in the community introduced by the EOS can lead to phenological mismatches; for example, the period of high consumer demand for a resource does not match with the period of resource abundance [57]. At the ecosystem level, phenological grassland changes can modify certain land surface parameters (e.g., albedo, sensible heat flux, evaporation, boundary layer conductivity), which affects the regional carbon and water cycles [58]. For example, a later EOS may promote GPP and cause plants to close their stomates and increase water use efficiency if a soil deficit exists [59]. Moreover, the delayed EOS may also increase transpiration and partly offset the GPP, therefore leading to closer relationships between the net ecosystem productivities and EOS changes [60].

The existence of turning points indicates that the EOS trend over long-time periods does not remain unchanged, and the rates of EOS changes differ before and after these points. This observation has several advantages in ecosystem-related studies. First, climatic controls on the EOS in the Qinghai-Tibetan Plateau intersect with each other and follow non-linear relationships with the EOS. An analysis of the EOS before and after the turning points therefore helpful to evaluate the climatic driving mechanisms of the EOS. Second, the detection of spatial heterogeneity of the turning points is helpful for evaluating the large-scale implementation effects of ecological conservation projects. Third, an analysis of the turning points of the EOS relationships with ecosystem functions and services provide important guidelines for fine ecology planning and the development of protection policies.

#### *4.3. Uncertainties, Challenges, and Future Directions*

The uncertainties in this study arise from three aspects. First, although the EOS trends are consistent with the findings of MODIS NDVI and SPOT NDVI, some design shortcomings in the AVHRR sensor may potentially introduce noise into the GIMMS 3g NDVI dataset. Second, the human activities are difficult to quantify for lack of grazing data (intensity and boundary) and statistic data on the county levels for a long time. Third, there is a limited number of phenological stations on the Qinghai-Tibetan Plateau, and most are distributed in the east, which thus does not represent the EOS changes of the entire plateau. The results of the EOS extraction are not fully calibrated by observations owing to limited data availability.

We recommended the following perspectives for future studies. First, extreme climate events (e.g., cold, frost, drought) may have a more direct effect on vegetation phenology than gradual changes in mean climatic conditions [27,28]. Non-structural carbohydrate storage in plants is helpful to avoid damage caused by extreme events [61]. However, extreme climate conditions with variable frequencies and intensities in different seasons on the Qinghai-Tibetan Plateau require rigorous quantification. Second, although many studies have quantified the effects of climate variables in different seasons, spring phenology, growth season length, and human disturbances on the EOS changes, the joint contribution of these variables is low and the control mechanisms of the EOS and its turning points remain poorly understood. The strengthening and development of phenological observations stations are therefore necessary to explain the mechanism of phenology changes in the Qinghai-Tibetan Plateau. Third, ecosystem models are essential tools for simulating the carbon cycle in both historic and future climate scenarios however, their accuracies remain limited by the understanding of the EOS [62]. More reasonable algorithms and reliable observations are required to calibrate the ecosystem models, which will ultimately provide a new research direction but presently faces serious challenges.
