*3.1. Model Assessment and Key Environmental Variables*

The mean AUC of three species in training and testing all exceeded 0.9 in current and future modeling. The model for the geographic distribution prediction performed excellently and had high accuracy.

We selected the variables whose contribution rate for three species is more than 0.1 for analysis. The internal jackknife test of the MaxEnt model for environmental variables' importance showed that aspect was the most critical factor determining the distribution of the three species. Aspect contributed 33.9% to model output for *Picea crassifolia*, 51.7% for *Sabina przewalskii* and 56.1% for *Potentilla parvifolia* (Table 2). In addition, elevation contributed 20.2% to model output for *Picea crassifolia*, 26% for *Sabina przewalskii* and 15.9% for *Potentilla parvifolia*. The following factors were precipitation of driest month (Bio14: 23.1% for *Picea crassifolia*, 3.4% for *Sabina przewalskii* and 3.5% for *Potentilla parvifolia*), annual precipitation (Bio12: 10.7% for *Picea crassifolia*, 4.9% for *Potentilla parvifolia*) and temperature seasonality (Bio4: 6% for *Picea crassifolia*). The total contributions of three subsoil variables (S\_CEC\_CLAY, S\_BULK\_DEN, S\_CEC\_SOIL) did not exceed 3% (Table 2). The results indicated that subsoil conditions had very limited impacts on the potential distribution of *Picea crassifolia*, *Sabina przewalskii* and *Potentilla parvifolia*. The cumulative percentage of aspect, elevation, Bio14, Bio12, Bio4, S\_CEC\_CLAY, S\_BULK\_DEN, and S\_CEC\_SOIL was 95.6% for *Picea crassifolia*, 82.5% for *Sabina przewalskii* and 83.5% for *Potentilla parvifolia,* respectively.
