**3. Results**

#### *3.1. The Verification of Vegetation Phenological Results*

For this study, a regional plant phenological dataset was developed based on data acquired from 2001 to 2018. Figure 4 shows that the remote sensing monitoring data of SOS (R<sup>2</sup> = 0.84, *p* < 0.01), EOS (R<sup>2</sup> = 0.72, *p* < 0.01), and LOS (R<sup>2</sup> = 0.86, *p* < 0.01) have strong similarity with the phenological observation data. Specifically, the times of SOS monitored by remote sensing and observed by phenological stations are distributed near a straight line (*Y* = *X*). However, the times of EOS and LOS observed by remote sensing and phenological stations are generally distributed above the straight line (*Y* = *X*). This showed that the time product of SOS is highly consistent with the values observed at phenological stations, but the time product of EOS is delayed when compared with that of phenological stations; in addition, the LOS product of remote sensing monitoring is longer than observed at the phenological station (Figure 4).

**Figure 4.** The comparison between remote sensing monitoring data (RSMD) and phenological observation data (POD): (**a**) start of the growing season (SOS); (**b**) end of the growing season (EOS); (**c**) length of the growing season (LOS).

#### *3.2. Spatiotemporal Pattern of Plant Phenology*

During the study period, the spatiotemporal trends and standard deviations of SOS, EOS, and LOS had a heterogeneous geographical distribution from 2001 to 2018. The spatial distribution of the multiyear mean SOS primarily occurred between day 100 and 150, and the multiyear average SOS arrived before day 100 in the low-elevation river valley areas of the Yellow and Lancang river basins and appeared after day 150 in some highelevation or high-latitude areas of the Yangtze River Basin (Figure 5a,d). Similarly, the high value (>16 day/year) of standard deviation for SOS principally occurred in the Lancang River Basin and the southwestern part of the Yangtze River Basin, with the lowest value (<8 day/year) in the center of the Yangtze River Basin and the southeastern part of the Yellow River Basin (Figure 5g). We also found that the Yellow River Basin had the earliest SOS, and the time is in advance (Figure 5j). Furthermore, the spatial distribution of the multiyear average EOS was mainly observed from day 265 to 283, the multiyear average EOS arrived before day 265 in the northeast of the Yellow River Basin, and appeared after day 280 in the center of the Yangtze River Basin (Figure 5b,e). The high value of the standard deviation of EOS was mainly in the Yangtze and Lancang river basins (Figure 5h). In addition, we also compared the temporal trend of EOS in different basins; the earliest EOS was in the Lancang River Basin and the latest in the Yangtze River Basin (Figure 5k). Last, the spatial distribution of the multiyear average LOS was mainly between day 120 and 160, while the multiyear average LOS was longer than day 150 in some areas of the Yellow and Lancang river basins (Figure 5c,f). The high value of the standard deviation of LOS was mainly distributed in the Lancang and Yangtze river basins (Figure 5i). Furthermore,

we also found that the Lancang River Basin had the longest LOS, which is becoming longer over time (Figure 5l).

**Figure 5.** Spatiotemporal patterns of vegetation phenology: (**<sup>a</sup>**–**<sup>c</sup>**) the spatial pattern of a multi-year average of the start (SOS), end (EOS), and length (LOS) of the growing season on the Three-River Headwaters from 2001 to 2018; (**d**–**f**) timefrequency distribution of SOS, EOS, and LOS, respectively; (**g**–**i**) standard deviation for the SOS, EOS, and LOS, respectively; (**j**–**l**) temporal variation characteristics of vegetation phenology of (A) Yangtze, (B) Yellow, and (C) Lancang river basins in SOS, EOS, and LOS, respectively. The different letters above the box plots indicate significant differences among different basins at *p* < 0.05. The green boxplots indicate the overall distribution characteristics of SOS, EOS, and LOS values in different basins. The yellow boxplots indicate the overall distribution characteristics of the trend of SOS, EOS, and LOS values in different basins.

For this study, SOS, EOS, and LOS have different distributions at different elevations, slopes, and aspects in the THRH (Figure 6). Specifically, the SOS generally showed an upwards (0.001 day/m, R<sup>2</sup> = 0.17, *p* > 0.01) trend with an increase in elevation (Figure 6a). This phenomenon indicates that with an increase in elevation, the SOS is delayed. In contrast, EOS and LOS decreased (0.002 day/m, R<sup>2</sup> = 0.34, *p* > 0.01 and 0.003 day/m, R<sup>2</sup> = 0.84, *p* < 0.01, respectively) as elevation increased, which represents that the time of EOS and the LOS advance and shorten with an increase in elevation, respectively (Figure 6b–c). Furthermore, SOS and EOS decreased significantly (0.32 day/◦, R<sup>2</sup> = 0.93, *p* < 0.01 and 1 day/◦, R<sup>2</sup> = 0.85, *p* < 0.01, respectively) with an increase in slope (Figure 6d–e). This indicates that the time of SOS and EOS advance with an increase in slope. However, the relationship between LOS and slope was the opposite of that between SOS or EOS and slope. The LOS was prolonged (0.2 day/◦, R<sup>2</sup> = 0.94, *p* < 0.01) with an increase in slope (Figure 6f). Last, we find the north-facing slopes had the lowest value of SOS but had the highest value of EOS and LOS. The results showed that the times of SOS, EOS, and LOS were the earliest, latest, and longest, respectively, on the north slope.

**Figure 6.** The relationship between different terrain factors and the start (SOS), end (EOS), and length (LOS) of the growing season: distribution and change characteristics at different elevations (**<sup>a</sup>**–**<sup>c</sup>**), slopes (**d**–**f**), and aspects (**g**–**i**).

#### *3.3. Linking Climatic and Soil Factors to Plant Phenology*

The correlation coefficients between plant phenology connected to the principal climate characteristics along with soil physical and chemical factors were significant at *p* < 0.01 (Table 1). In the Yangtze River Basin, our results show that the SOS was positively correlated with monthly mean shortwave radiation (MMR; 0.73\*\*), pH (0.50\*\*), and total phosphorus (TK) (0.44\*\*) but negatively correlated with monthly mean precipitation (MMP; −0.68\*\*), available nitrogen (AN; −0.39\*\*), monthly mean relative humidity (MMH; −0.38\*\*), and monthly mean soil moisture (MMSM; −0.37\*\*). Furthermore, the correlation coefficients between EOS and MMP, pH, TK, and MMR were −0.45\*\*, 0.40\*\*, 0.41\*\*, and 0.44\*\*. Finally, we found that LOS was significantly negatively correlated with pH (−0.46\*\*) and TK (−0.37\*\*), but LOS was significantly positively correlated with MMR (0.53\*\*), MMH (0.52\*\*), and AN (0.38\*\*) during the growing season.

In the Yellow River Basin, significant positive relationships were observed between SOS and monthly mean temperature (MMT; 0.50\*\*), monthly mean soil temperature (MMST; 0.48\*\*), MMR (0.50\*\*), and MMSM (0.31\*\*). However, the EOS was significantly negatively correlated with MMSM (−0.39\*\*) and AN (−0.32\*\*) and significantly positively correlated with pH (0.37\*\*). In addition, we found that LOS was significantly negatively correlated with MMR (−0.55\*\*), MMT (−0.46\*\*), and MMST (−0.43\*\*).

In the Lancang River Basin, the results indicated that there were significant positive correlations between the SOS and AK (0.50\*\*), MMT (0.65\*\*), MMST (0.55\*\*), MMR (0.53\*\*), and MMSM (0.43\*\*). In addition, we found that the EOS was significantly negatively correlated with MMST (−0.41\*\*), MMT (−0.43\*\*), MMR (−0.36\*\*), and MMP (−0.33\*\*). Meanwhile, we also found that the correlation coefficients between LOS and MMH, MMT, MMST, MMR, and AK were 0.41\*\*, −0.69\*\*, −0.65\*\*, −0.58\*\*, and −0.50\*\* (Table 1).



MMR, MMP, MMT, MMH, MMSM, and MMST indicate monthly mean shortwave radiation, precipitation, temperature, relative humidity, soil moisture, and soil temperature, respectively. AK, AN, AP, BD, CEC, pH, POR, SOM, TK, TN, and TP represent available K, alkali-hydrolysable N, available P, bulk density, cation exchange capacity, pH (H2O), porosity, soil organic matter, total K, total N, and total P, respectively. \*\* and \* indicate significance coefficients of less than *p* < 0.01 and *p* < 0.05, respectively. a,b and c represent the Yangtze River Basin, Yellow River Basin, and Lancang River Basin, respectively.

The mechanisms involved in patterns in the length of the plant growing season in different basins were explored using SEM. In general, the effect of soil factors on LOS is greater than that of climate factors in the Yangtze River Basin. Specifically, AP, pH, and TN had a significant effect on the LOS (*p* < 0.01), with scores of 0.30, −0.65, and −0.77, respectively. However, the impact scores of MMR and MMH on LOS were only 0.35 and 0.33 (Figure 7a). Path analyses identified that climate factors, as a key functional indicator of the LOS in the Yellow River Basin, had either direct or indirect effects via edaphic factors. Specifically, the MMR (scored at −0.55), MMT (scored at −0.30), and MMST (scored at 0.54) had significant effects on the LOS (Figure 7b). Furthermore, in the Lancang River Basin, the effects of each variable on LOS were different (ranging from −0.52 to 0.25), which suggests that the LOS might be co-determined by both the soil and climatic factors (Figure 7c). This assumption was confirmed in that soil factors were significantly affected by climatic factors. Specifically, the AK (scored at 0.41), AP (scored at 0.27), and AN (scored at 0.32) were significantly (*p* < 0.01) influenced by MMT. Furthermore, AK and AP had a significant interaction (scored at 0.58).

**Figure 7.** Mechanisms involved in the patterns of the length of the plant growing season in different basins. Structural equation modeling (SEM) was used to analyze the total effects of variables. The black and red solid lines represent positive and negative standardized SEM coefficients, respectively, while the line thickness indicates the magnitude of these coefficients for the Yangtze (**a**), Yellow (**b**), and Lancang (**c**) river basins, respectively. MMR, MMT, MMH, and MMST represent monthly mean shortwave radiation, temperature, relative humidity, and soil temperature, respectively. AN, AP, pH, TN, BD, POR, and AK represent alkali-hydrolysable N, available P, pH (H2O), total N, bulk density, porosity, and available K, respectively.
