*2.3. Satellite Data*

#### 2.3.1. GIMMS NDVI Product

To quantify the variation of vegetation dynamics at regional scales, we used the Global Inventory Modeling and Mapping Studies (GIMMS) NDVI3g product derived from the Advanced Very High Resolution Radiometer (AVHRR) sensor National Oceanic and Atmospheric Administration (NOAA) polar satellite series with a spatial resolution of 8 km and a 15-day interval [41,42]. The GIMMS NDVI product has already been corrected to minimize the effects of clouds and aerosols using the maximum value composite (MVC) method. Previous studies have demonstrated that this dataset can reflect the real response of vegetation to climate change and provides more accuracy when evaluating the long-term trends of vegetation activity [43]. In this study, we extracted the subset of coverage in the TRHR from the global bimonthly NDVI for the period 1982–2015 and resampled the bimonthly NDVI of the study area to a daily value with a resolution of 0.1◦ × 0.1◦.

## 2.3.2. ET Product

Considering that the MOD16 ET product is missing in the TRHR, we used the ET product produced by the modified satellite-based Priestley–Taylor algorithm (Appendix A) driven by net radiation (Rn), air temperature (Ta), diurnal temperature range (DT), and the NDVI [44]. This product has been validated at 16 eddy covariance (EC) flux tower sites, and performed better than MODIS ET products at a regional scale, with a higher squared correlation coe fficient (R2) and a lower root mean square error (RMSE) [45]. The modified satellite-based Priestley–Taylor (MS-PT) product has provided more reliable and long-term spatiotemporal variations of the ET estimations of China [46].

## 2.3.3. DEM Data

We used the global digital elevation model (DEM) data with a spatial resolution of 250 m acquired from 90 m Shuttle Radar Topography Mission (SRTM) images (version 004) (http://srtm.csi.cgiar.org/) in Geo-TIFF format.

#### 2.3.4. Land Cover Data

The GlobeLand30 product developed by the National Geomatics Center of China (NGCC) provides detailed land cover information about a global coverage of high-resolution imagery at 30 m for the years 2000 and 2010 [47]. It is generated from the Thematic Mapper (TM), Enhanced Thematic Mapper plus (ETM+) of America Land Resources Satellite (Landsat) and the multispectral images of the China Environmental Disaster Alleviation Satellite (HJ-1) developed by integrating the pixel-object knowledge-based approach with other auxiliary datasets. This dataset is freely available and consists of 10 land cover types, including forest, grassland, shrubland, wetland, water bodies, tundra, bare land, artificial surfaces, cultivated land, permanent snow, and ice, with an overall accuracy of 80.33% [48].
