Figure 3 shows the spatial distributions of the start and end dates, expressed as day of year (DOY), and the length of growing season averaged for the multi-year data from 2000 to 2014. The overall SOS was delayed from southeast to northwest (Fig. 3a). The growing season begins between April 30 and May 20 (120–140 DOY) in the eastern and central regions; while it is between May 20 and June 20 (140–170 DOY) in the northwest region. However, more than 84% of the region began to grow in the period of 120–160 DOY with a regional mean date of 147 ± 10 DOY (15–30 May), the mean end date was 285 ± 6.5 DOY (Fig. 3b), and the average LOS was 140 ± 14 days (Fig. 3c). In the trend analysis, no significant changes were found in the EOS. The changes of LOS were mainly affected by SOS, so we mainly focused on the changes of SOS in the following results.
This headwater area experienced warming-wetting climate changes during 2000–2014 (Fig. 4). The ATP showed a significant increasing trend at a rate of 73.21 mm decade−1 (R2 = 0.383, p < 0.05), while the annual mean air temperature significantly increased at a rate of 0.86°C decade−1 (R2 = 0.500, p < 0.05). Responding to the climate changes, the SOS advanced significantly at a mean rate of 1.03 day yr−1 (R2 = 0.335, p < 0.05) in the TRSR during 2000–2014. The SOS of alpine meadow advanced 0.63 day yr−1 (R2 = 0.254, p < 0.05) while the alpine steppe advanced 0.65 day yr−1 (R2 = 0.526, p < 0.05; Fig. 5). The SOS of the alpine meadow occurred mostly between the end of May and beginning of June (140–155 DOY) during 2000–2007, and advanced significantly to mid-May during 2008–2009 (135–140 DOY). The areas where the SOS advanced accounted for 80% of the region, and this advance was greater in the eastern and southern valley areas (Fig. 6).
Figure 4. Interannual variations in (a) annual mean precipitation and (b) annual mean temperature (AMT) in the TRSR during 2000–2014.
Figure 5. Interannual variations of the whole-region SOS, alpine meadow SOS, and alpine steppe SOS in the TRSR during 2000–2014.
Figure 7 shows the headwater boundary and main grassland types (grassland type data were provided by the National Earth System Science Data Center, China) in the TRSR. The main grassland types in the Yellow River and the Yangtze River sources were alpine meadow and alpine steppe, but in the Lancang River source, it was mainly alpine meadow.
Analysis of the correlations between the precipitation, temperature, and SOS of the main grassland types revealed that the impacts of precipitation and temperature on the spring vegetation phenology were basically consistent in the three different headwater regions (Fig. 8). However, in the time series, the changes in precipitation and temperature affected the SOS differently. The precipitation during the preceding year, particularly in October, was significantly negatively correlated with the SOS, while the precipitation after November of the preceding year gradually transitioned from a negative correlation to a positive correlation. The precipitation approximately 30 days ahead of the SOS was slightly negatively correlated with the SOS. Water required for the SOS in the TRSR is derived mainly from melting snow and ice, and the spring precipitation (i.e., the cumulative precipitation in the previous winter) plays an important role. Spring temperatures were negatively correlated with the SOS, especially in January and April (Fig. 8).
Figure 8. Seasonal variations of the correlation coefficient between SOS and temperature (red line) and that between SOS and precipitation (blue line) for the main grassland types: (a, c, e) alpine meadow and (b, d) alpine steppe, in the (a, b) Yangtze River, (c, d) Yellow River, and (e) Lancang River source regions.
Figure 9 shows the mean and 95% confidence interval of the SOS trend in each 20-mm precipitation interval in the whole TRSR and in three temperature zones. First, the variance of the SOS trend in the colder area was very different from those in the warmer or median-temperature areas. The whole-region interannual changes in the SOS (Fig. 9a) were consistent with those in the lower-temperature area (Fig. 9b). Under different precipitation amounts, the interannual changes in the SOS generally displayed a bimodal pattern. In the zone receiving 160–240 mm of precipitation, the SOS gradually advanced. In the 240–320-mm zone, the interannual advance of the SOS weakened drastically with the increase in precipitation. Meanwhile, in the 320–520-mm zone, the interannual advance of the SOS continuously increased. In the 520–660-mm zone, the interannual advance showed a single peak; particularly at 660-mm, the interannual advancement reached the maximum. With precipitation exceeding 660 mm, the interannual advance gradually weakened. The interannual advance of the vegetation growing season is more obvious in the areas with more precipitation. However, the appearance of the second peak indicates that in the areas with lower precipitation, precipitation was not the dominant factor affecting the advance of the SOS.
Figure 9. Annual trends of SOS (a) for different precipitation amounts across the whole TRSR and (b) in different temperature zones.
The pattern of changes in the interannual advance of the SOS with the changes in precipitation in the lower-temperature zone was generally consistent with the pattern in the entire region, which was significantly different from the changes in the median- and higher-temperature zones, although the changes in the latter two zones were consistent with each other (Fig. 9b). In the 360–660-mm zone, the interannual advance of the onset of the vegetation growing season was relatively stable in the median- and higher-temperature zones. In the 660–800-mm zone, the weakening of the interannual advance of the onset of the vegetation growing season occurred in all three temperature zones. However, this weakening of the advance of the SOS gradually enhanced with the increase in temperature.
The ATN and the summer mean NDVI each showed a strong dependence on the elevation level, and in both cases, they were more controlled by the dependence of precipitation on the topography than that of air temperature over the TRSR (Fig. 10a). The AMT linearly decreased by a rate of 4.6°C km−1 with increasing elevation. But the precipitation showed a unimodal curve with the maximum around the 4000-m elevation. According to the segmented regression, three break-points were found, and the changes of precipitation could be classified into four elevation ranges: < 3093, 3093–3929, 3929–5001, and > 5001 m. In the region with elevations below 3093 m, ATP increased by 28.14 mm km−1 with rise in elevation, and by 265.85 mm in the 3093–3929-m zone, while it decreased by 197.06 mm km−1 in the 3929–5001-m zone, and showed an increase of 109.76 mm km−1 at elevations over 5001 m (Table 1).
S0 S1 S2 S3 P1 P2 P3 SOS 33.86 −17.67 5.95 −9.82 3.09 (2.99, 3.10) 3.67 (3.64, 3.70) 5.19 (5.15, 5.23) ATN −1.85 16.45 −10.30 −5.89 2.96 (2.90, 3.02) 3.67 (3.63, 3.71) 5.00 (4.80, 5.20) AMN −0.02 0.75 −0.38 −0.17 2.98 (2.92, 3.04) 3.65 (3.60, 3.70) 5.48 (5.33, 5.63) ATP 28.14 265.85 −197.06 109.76 3.09 (3.04, 3.14) 3.93 (3.90, 3.96) 5.00 (4.96, 5.04) AMT −4.63* SOS_trend −0.71 0.26 −0.69 1.72 2.66 (2.55, 2.77) 3.88 (3.78, 3.98) 5.24 (5.20, 5.28) ATP_trend 10.54 35.10 −32.65 1.07 3.05 (2.94, 3.16) 3.91 (3.60, 4.22) 5.17 (5.00, 5.34) AMT_trend 0.08 0.66 0.08 0.34 3.08 (3.04, 3.12) 3.92 (3.88, 3.96) 5.03 (4.93, 5.13)
Table 1. The regressed slopes (S) and the estimated elevation (km) break-points (P) and their 95% confidence intervals from the segmented regression for SOS (day km−1), annual total precipitation (ATP; mm km−1), annual mean air temperature (AMT; °C km−1), annual total NDVI (ATN), annual summer mean NDVI (AMN), SOS_trend (day yr−1 km−1), ATP_trend (mm decade−1 km−1), and AMT_trend (°C decade−1 km−1), with elevation (km) over the TRSR, Qinghai Province, China. The symbol * represents the linear regression without break-points
Figure 10. (a) Variations in climatic factors and NDVI with elevation and (b) changes of the SOS with elevation.
Following the precipitation and temperature changes along the elevation gradient, both the ATN and the AMN showed almost the same patterns of change along the increasing elevation level (Fig. 10a). The changes of NDVI can be explained well by the climatic factors, and the regressed slopes and estimated elevation break-points can be found in Table 1.
The changes of the SOS were more interesting in the area with elevations around 3095 m, which probably represented ridges for the SOS trend (Fig. 10b). In the 2000–3095-m zone, the SOS was delayed 36.64 day km−1 increase in elevation (R2 = 0.97, p < 0.05), whereas in the 3095–5800-m zone, the SOS advancement showed some fluctuations (Fig. 10b). The regressed slopes and the estimated elevation break-points can be found in Table 1.
The trends of AMT, ATP, and SOS also showed strong dependence on elevation in the study region (Fig. 11a), and we estimated their elevation break-points (Table 1). Although the AMT was still rising in the whole area, the rate of the rise varied depending on the elevation levels, with break-points at 3080, 3920, and 5030 m. Over the area with elevations lower than 3080 m, the climate warming trend was slow, with a speed of around 0.2°C decade−1, but the speed showed an almost linear increase at the rate of 0.08°C decade−1 km−1. Meanwhile, in the 3080–3920-m zone, the rate was 0.66°C decade−1 km−1; and in the 3920–5030-m zone, the rate was weakened to 0.08°C decade−1 km−1. Above 5030 m, the rate was 0.34°C decade−1 km−1. However, the trend of precipitation changes showed a unimodal pattern along the increasing elevation gradient, with the wetting trend increasing as elevation increased, reaching a maximum at elevation 3920 m, and then the wetting speed decreased with the increasing elevation.
Figure 11. (a) Interannual changes of the climatic factors and the SOS at different elevation levels, and (b) a comparison of interannual changes of the SOS in the TRSR and the TP at different elevation levels. The interannual changes of the SOS in the TP were obtained from Qiu et al. (2017).
The interannual trend of the SOS was averaged at 50-m elevation intervals, and the SOS advanced between the 2000- and 6000-m elevations, although it varied with different elevation levels (Figs. 11a, b). Three clear boundaries were present in the TRSR with 2660-, 3880-, and 5240-m elevation contours. In the 2000–2660-m zone, the SOS continuously advanced at a rate of 0.71 day yr−1 km−1. In the 2660–3880-m zone, the relative delay shifted at a rate of 0.26 day yr−1 km−1. In the 3880–5240-m zone, the SOS advanced greatly, at a rate of 0.69 day yr−1 km−1. In the zone above 5240 m, the SOS was delayed greatly, at a rate of 1.72 day yr−1 km−1 (Table 1). In the high elevation/low temperature area, the annual advance trend of vegetation growth was more sensitive to the temperature increase than that in the low elevation/high temperature area, and the annual advancement in days of SOS was larger than that in the low elevation/high temperature area.
Elevation plays an important role in the redistribution of temperature and precipitation, which were both negatively correlated with the SOS, although their effects varied with elevation (Fig. 12). In the zone below 3000 m, precipitation was the dominant factor affecting the SOS, and the negative effects of temperature increased with increasing elevation. In the 3000–4400-m zone, precipitation remained the dominant factor affecting the start of the vegetation growing season. In the 4400–5500-m zone, temperature was the dominant factor. At higher elevations, the effects of both precipitation and temperature gradually weakened.