Quantification of human activities
The human footprint could reflect human pressure comprehensively and objectively by selecting the spatial factors directly related to human activities (Venter et al., 2016; Woolmer et al., 2008). Based on this, we put forward the HI factor, containing five indexes (population density, grazing density, human access, electrical power infrastructure, and land use/cover), regarded as a vital input variable in our models. The specific calculation and standardized method are as follows.
Population density
Ecological demand is always associated with population density (Liu et al., 2013). The Worldpop program (https://www.worldpop.org/) produces data on population distributions and characteristics at high spatial resolution. We downloaded the population density database and classified greater than 1000 people /km2 as ten scores. For the rest, the score 0-10 were calculated and assigned according to the logarithmic equation (Venter et al., 2016).
Grazing density
We collected the number of cattle and sheep (from the statistical yearbook: https://data.cnki.net/yearbook) and the areas in each county. We transformed (assuming one cattle’s ecological consumption equals five sheep), referring to the literature (Yin et al., 2020), and concluded the grazing density based on the following formula:
\begin{equation} \text{Grazing}_{\text{den}}=\frac{\log x_{i}}{\log x_{\max}}\times 10\nonumber \\ \end{equation}
where Grazingden means the grazing density of each grid,\(x_{i}\) represents the ratio of sheep number to the area in the county where the pixel is located, and \(x_{\max}\) is the largest value in\(x_{i}\).
Human access
Human access means that human activities could enter natural habitats through roads, which may reduce the environmental quality and the number of habitats (Geneletti, 2003). The distance from the road network (obtained from the National Catalogue Service for Geographic Information: www.webmap.cn) was a scoring criterion in our study (see Tab.1).
Electrical power infrastructure
Night light data represents the level of regional socio-economic development and power infrastructure construction (Nordhaus & Chen, 2015), reflecting the ability and intensity of harness nature to some extent. A higher value means more frequent human activities. We scored the processed raster 1-10 by the quantile grading method after the preprocessing of the VIIRS Stray Light Corrected Nighttime Day dataset via Google Earth Engine (GEE).
Land use/cover
Dissimilar land use has diverse effects on ecosystem change processes and the natural environment (Foley et al., 2005). According to the land use classification standards, we downloaded the database of the Globeland30 from the National Catalogue Service for Geographic Information (https://www.webmap.cn/), assigning 10 points to construction land, followed by 7 points to arable land, 3 points to forests and irrigation, 1 point to grassland (taking into account the impact of grazing in mountainous areas), and 0 points to other land attributes (i. e. permanent snow and ice surface).
Finally, we summed all normalized layers with equal weights using the GIS raster calculator to obtain the HI layer. Combined with the above environmental variables, there were a total of 14 variables involved in our models (see detail in Appendix S1.2). All layers were resampled to 1 km and unified the same coordinate system.