**3. Results**

#### *3.1. EOS Spatial Distribution and Variation Characteristics*

The obtained EOS presented high spatial heterogeneity across the grassland of the Qinghai-Tibetan Plateau during 1982–2015 (Figure 2). The EOS results extracted using the HANTS and Polyfit methods were not consistent, but their spatial distribution trends were similar (Figure 2a,b). The mean multi-year EOS began on the 291th day of the year (end of September) and spanned nearly one month from the southeast to the northwest (Figures 2 and 3a). The EOS started early (around the 277th day) in subregion IX, which has the highest elevation and lowest AGDD0 values, and started late (around the 300th day) in subregions II and III, which are characterized by relatively warm-humid conditions. In the central Qinghai-Tibetan Plateau (subregion V), the EOS occurred on the 295th day. In subregion I, the EOS was early in the west and late in the east with a mean EOS on the 292th day. The spatial heterogeneity variations were significantly controlled by the MI (EOS = 16.55 × MI + 287.28, R2adj = 0.20 \*\*), with an early EOS in the drought subregions (IV, VII, VIII, and IX) and late EOS in the relatively humid subregions (II, III, and VI). The EOS spatial heterogeneity was essentially insensitive to AGDD0.

The mean EOS on the Qinghai-Tibetan Plateau exhibited a slow delayed trend with an average rate of 0.08 days/year. The EOS results extracted using the HANTS and Polyfit methods presented similar patterns (Figure 2d,e). Using these two fitting methods, 60.2% of the study area presented delay trends (27.8% area is significant), while 39.8% of the study area presented advance trends (13.4% area is significant). The EOS trends differed between the nine subregions during 1982–2015 (Figure 3b), showing a delay in the northwest and an advance in the southeast. Subregions I and IX showed significantly delayed trends with more than 0.20 days/year. The EOS of subregion II, with a main land use type of wetland, was also delayed by a rate of 0.12 days/year. The EOS in subregion VIII, which is characterized by alpine, cold, and dry climatic conditions, presented a negative trend with the fastest variation rate ( −0.12 days/year) compared with the other subregions. The EOS of subregion IV showed an advanced trend in the north but delayed trend in the south, with a mean EOS trend of 0.02 days/year. The EOS in subregions III and V showed slight advanced trends of –0.02 and −0.01 days/year, respectively. Subregions VI and VII both presented a slightly delayed trend with an average of approximately 0.04 days/year.

#### *3.2. Detection of EOS Turning Points in the Subregions*

The EOS changed over time and presented delayed trends during 1982–2015, but the rates of change were not fixed in each subregion over different periods, and notable turning points were observed (48.2% is significant) (Figure 4c). For example, in subregion I, the turning point occurs in the year 1994, for which the EOS was delayed before 1994 and slightly advanced after 1994 (Figure 4f). Similarly, subregion II showed a delayed EOS before 2002 and then a slightly advanced EOS after 2002. In the remaining subregions (III, IV, VI, and IX), the change trends were similar and the turning point year was 1994, where the EOS was delayed prior to 1994, suddenly advanced in 1995, and then maintained the previous change trend until 2015. The turning point trends in subregions V, VII, and VIII occurred in 1994, 1994, and 1999 respectively, but were not significant. These results demonstrate that the EOS changes clearly exhibit turning points and a wide range of EOS change trends with significant spatial heterogeneity on the Qinghai-Tibetan Plateau. The pattern of EOS turning points extracted by HANTS and Polyfit (Figure 4a,b,d,e) have a small difference in subregion I and VI.

**Figure 2.** Distributions of the end of season (EOS) (calendar day) and their change trends (days/year). Multi-year means of the EOS obtained using the (**a**) HANTS method and (**b**) Polyfit method. (**c**) Average values of the HANTS and Polyfit methods. Trends of the EOS obtained using the (**d**) HANTS methods and (**e**) Polyfit method. (**f**) Average values of the EOS trends obtained using the HANTS and Polyfit methods.

**Figure 3.** Bar graphs of the (**a**) mean EOS values and (**b**) their trends. "\*\*" indicates *p* < 0.01 and "\*" indicates *p* < 0.05.

**Figure 4.** Turning point distributions and variations for subregions. (**<sup>a</sup>**–**<sup>c</sup>**) EOS turning points distributions with results of HANTS, Polyfit, and their mean values. The numbers represent the modes of the turning years in each subregion. (**d**–**f**) Scatter plots and results of piecewise regressions with the results of HANTS, Polyfit, and their mean values. The vertical dashed lines represent the turning points in the different subregions.

#### *3.3. EOS Variations Controlled by Climatic Variables before and after Turning Points*

The EOS changes exhibited close relationships with the climatic variables, but the dominant climatic variable differed in each subregion before and after its associated turning point (Figure 5). Temperature was the dominant control over the EOS changes in most subregions (I, II, IV, VI, VII, VIII, and IX) before and after the turning point year. In contrast, subregion III showed that the EOS was mainly controlled by the precipitation. Central subregion VII showed that the EOS was jointly controlled by the effects of temperature and precipitation. The area where the EOS changes was controlled by temperature covered the largest proportion, followed by precipitation and insolation (Figure 5d). The results indicate that the proportions controlled by each climate variable changed before and after the turning point years. For example, the EOS in subregion V was controlled by precipitation before the turning point, which then switched to temperature (Figure A2). The EOS of only approximately 40% of the area in subregion VI was significantly controlled by temperature prior to the turning point, which thereafter increased to 70%. Contribution of climates to EOS variation are similar with the fitting results of HANTs and Polyfit (Figure A3).

**Figure 5.** Relative influence of different climate variables (temperature, precipitation, and insolation) on EOS changes before (**a**) and after (**b**) the turning point, and over the entire study period (**c**) with the EOS means values of HANTS and Polyfit methods. (**d**) Area proportions controlled by the different dominant climate variables.

#### *3.4. Controls on the EOS Turning Points*

The changes of the annual EOS turning points are consistent with the turning points of the climate variables in most subregions (Table 2). In subregions I and II, the year of the EOS turning point coincides with the year of the insolation and precipitation turning points, respectively. Furthermore, the years of the EOS and temperature turning points are consistent in subregions III–IX. The major determining climatic variable for the EOS turning points is precipitation, followed by temperature and insolation. The relationship with the EOS turning point and insolation is generally weak (R<sup>2</sup> < 0.05).

**Table 2.** Correlation coefficients and P values between the turning points of the EOS and the turning points of climate variables.


The relationship between the EOS and human activities was studied at the province level owing to limited statistical data in certain counties and subregions. The economic data show a consistent turning point with the EOS. Before the turning point year (~1996 for Qinghai and ~1995 for Tibet), Qinghai maintained a large amount of sheep, which reflected high grazing activity, and the economic development was slow with low production in the primary, secondary, and tertiary sectors. However, after the turning point year, the grazing intensity decreased and reached a stable change rate, whereas the economy developed rapidly, especially in the secondary sector. For Tibet, the grazing intensity was small before the turning point year but showed a rapid growth rate after the turning point. Similar to Qinghai, Tibet experienced fast economic growth after the turning point, especially in the tertiary sector.

At the province level, the annual EOS was closely related to climate, human activities, and their intersections (Table 3). For Qinghai, a combination of the turning points of climate and human activities can explain 78.86% of the EOS turning points changes, with climate

independently accounting for 40.22% and human activities accounting for 10.45%. The intersections of climate and human activities can explain 28.19% of the EOS variation in Qinghai. In Tibet, the EOS change due to climate (66.17%) was larger than that in Qinghai and the effect of human activities (6.8%) was weaker. The climate and human activities intersections in Tibet (9.98%) were also smaller than in Qinghai.

**Table 3.** Independent contributions of the turning points of climate, human activities, and their intersections to the annual turning points variations of EOS at the province level.

