*2.2. Data Sources*

NDVI time series have been widely used as an important proxy for quantifying vegetation photosynthetic activity. The MODIS NDVI products (MOD13A2) from 2001 to 2020 with a 1 km spatial resolution and 16-day time step were used in this study. The NDVI data were obtained from NASA (https://lpdaac.usgs.gov, accessed on 20 November 2021) and were preprocessed using the MODIS reprojection tool (MRT) for band extraction and

mosaic, format, and projection conversion. We removed pixels with average annual NDVI values (2001–2020) < 0.1 to prevent the interference of nonvegetation signals [18,29].

The ground-based phenology data were collected from the vegetation phenological observation datasets at Haibei station from 2006 to 2015, which were provided by the Chinese Ecosystem Research Network (CERN) (http://www.cnern.org.cn, accessed on 17 December 2021). In addition, the phenology data of Sidalong station, Liancheng station, Xiyinghe station, and Suganhu station from 2020 were obtained from the National Tibetan Plateau Data Center (https://data.tpdc.ac.cn/en/, accessed on 17 December 2021).

Monthly temperature and precipitation data from 2001 to 2020 were obtained from the National Tibetan Plateau Data Center (http://data.tpdc.ac.cn, accessed on 22 April 2021). These datasets were spatially downscaled from CRU TS v4.02 with WorldClim datasets based on the delta downscaling method and were evaluated using the data of 496 national weather stations across China. The evaluation indicated that the downscaled dataset is reliable for investigations related to climate change across China [30]. The monthly surface soil moisture data (0–7 cm) from 2001 to 2020 were obtained from ERA5-Land and used to represent the water availability indicator to evaluate the water content impacts on vegetation phenology. ERA5-Land provides a soil moisture reanalysis dataset of 0.1◦ × 0.1◦ from 1950 to the present. The monthly soil moisture data from ERA5-Land were resampled to the same resolution as the vegetation phenology data using a bilinear interpolation algorithm. The digital elevation model at a spatial resolution of 1 km was obtained from the Resource and Environmental Science and Data Center (http://www.rsdc.cn/, accessed on 17 December 2021).
