**3. Results**

#### *3.1. Performance Evaluation of the GLDAS-NOAH Data*

Due to the strong heterogeneity of the underlying factors over the YZR basin, such as the vegetation, soil type, and elevation, etc., it is essential to evaluate the performance of the GLDAS-NOAH data at both the site scale and the river basin scale. In this study, the observed monthly precipitation and temperature data from 1982–2015 from the twenty meteorological gauging stations covering the upper, middle, and lower reaches of the YZR basin (Figure 1) and extracted model outputs of corresponding grids from the GLDAS-NOAH dataset were utilized to conduct the performance evaluation.

#### 3.1.1. Precipitation Performance at Site Scale

The correlation coefficients between the gridded monthly GLDAS-NOAH precipitation and the corresponding observed monthly precipitation showed high consistency (*R* > 0.80), except the Bomi station, which had a relatively lower *R* value of 0.57 (Table 2). This can be partly attributed to the dramatic topographic variations around the Bomi station ranging from 3100 m to 5000 m, whereas the elevation of the gauging station was 1300 m, which may have insufficiently represented the regional precipitation within the GLADS-NOAH grid area of approximately 625 km<sup>2</sup> (0.25◦ × 0.25◦). Lv et al. [61] demonstrated a similar conclusion in the study on the performance evaluation of the TRMM satellite precipitation data in the YZR basin. As shown in Table 2, the *MB* between the GLDAS-NOAH outputs and in-situ measurements ranged from −49.95 mm to 22.99 mm (accounting for 0.07%–17.47% of annual precipitation) and the *RMSE* ranged from 21.23 mm to 81.45 mm (accounting for 4.20%–29.44% of annual precipitation), which may be attributed to the mismatch of the point and grid scale.


**Table 2.** Statistical indicators of precipitation between the GLDAS-NOAH and gauging stations.


**Table 2.** *Cont*.

#### 3.1.2. Temperature Performance at Site Scale

As shown in Table 3, the GLDAS-NOAH data and in-situ measurements showed a high agreement, i.e., *R* ≥ 0.81, −0.63 ≤ *NSE* ≤ 0.97, −1.01 ◦C ≤ *MB* ≤ 6.85 ◦C, and *RMSE* ≤ 7.25 ◦C. In the Bomi station, the GLDAS-NOAH surface air temperature showed a larger discrepancy, i.e., the absolute value of *MB* = 6.85 ◦C and ≤4.59 ◦C for other stations, and *RMSE* = 7.25 ◦C and ≤5.37 ◦C for other stations. Similar to the performance of precipitation, the worse performance of temperature could be partly attributed to the dramatic topographic variations around the Bomi station, which may have caused uncertainties in the temperature estimation. Given that the monthly *NSE* > 0.5 and monthly *R* > 0.77 implied a good simulation, the high *R* and *NSE* from all stations indicates the high representation of the GLDAS-NOAH surface air temperature over the YZR basin.

**Table 3.** Statistical indicators of temperature between the GLDAS-NOAH and gauging stations.


#### 3.1.3. Spatio-Temporal Patterns at River Basin Scale

In this study, the monthly precipitation and temperature in the entire YZR basin were calculated from 1982–2015 to further investigate the correspondence between the measured and GLDAS-NOAH data. As shown in Figure 2, the *R* values of precipitation and temperature of the two datasets reached 0.97 and 0.99 respectively, meaning there were good consistencies between the temporal variation patterns of precipitation and temperature, while the GLDAS-NOAH precipitation was larger than the measured precipitation, and the GLDAS-NOAH temperature was smaller than the measured data. Although the GLDAS-NOAH overestimates and underestimates the precipitation and temperature respectively, such inconsistency is not the key issue in this study because of the following two reasons. Firstly, the observed data were at the point scale, while GLDAS-NOAH data represented the average performance at the pixel scale, approximately an area of 25 km × 25 km. With the high divergence of underlying factors, such as vegetation, soil type, and elevation, etc. over the YZR basin, it is rather hard for the scarce observed data to represent the pixel average features. Th gauge-based precipitation data showed that the difference in annual precipitation between Jiangzi and Shigatse was as much as 40%, while their distance was only 80km [64]. Secondly, this study focused on the spatio-temporal variation trends rather than absolute magnitudes. That is, the systematic overestimation or underestimation could be reasonably eliminated on a tendency or relationship analysis, if the variation patterns of the GLDAS-NOAH data fitted well with those of the in-situ data.

**Figure 2.** GLDAS-NOAH and measured monthly precipitation and temperature from 1982 to 2015.

In order to identify the spatial performance between the observed and the GLDAS-NOAH data, especially the vertical zonality characteristics, 20 meteorological stations and the corresponding grids were divided into four bands by elevation, i.e., 0–3500 m, 3500–4000 m, 4000–4500 m, and 4500–5000 m. As shown in Table 4 and Figure 3, the precipitation and temperature from the GLDAS-NOAH and in-situ data both decreased with the increase in elevation, implying that the GLDAS-NOAH data could represent the climate characteristics of the vertical zonality in the YZR basin. In terms of precipitation, the mean values of the observed data in 3500–4000 m and 4000–4500 m were 457.6 and 500.4 mm respectively, while those of the GLDAS-NOAH data were 640.4 mm and 619.7 mm respectively. This discrepancy was owing to the large *MB* values of the Lhunze and Nyemo stations as shown in Table 2. However, the variation ranges of the GLDAS-NOAH and in-situ precipitation at the four elevation bands were similar. Compared to the precipitation, the vertical variation characteristic of the GLDAS-NOAH temperature at the four elevation bands was much closer to that of the observed data. Although the observed temperature in the first elevation band seemed much bigger than the GLDAS-NOAH data, which was mainly ascribed to only three gauging stations in this elevation band, the GLDAS-NOAH temperature at the other three elevation bands exhibited a high agreemen<sup>t</sup> with the observed data, regardless of whether it was from the perspective of the mean values or from the perspective of the variation ranges. Similar results of the GLDAS performance in the YZR basin and QTP were demonstrated by Zhang et al. [65] and Zhang et al. [66].


**Table 4.** Elevations of the 20 gauging stations.

**Figure 3.** Spatial performance of the measured and GLDAS-NOAH precipitation (left) and temperature (right).

#### *3.2. Transition Characteristics of the Dry-Wet Regime*

The SPEI values at four time scales of 1-month, 3-month, 6-month, and 12-month were calculated from 1982 to 2015. SPEI 1 represented that when calculating monthly SPEI, the water deficit condition was taken into account only within the current month. SPEI 3 represented that the water deficit conditions of both the first two months and the current month were considered, and similar representations were used for SPEI 6 and SPEI 12. As shown in Figures 4 and 5, the temporal variations of the SPEI at the four time scales representing the dry-wet conditions in the YZR basin showed that SPEI 12 exhibited the highest agreemen<sup>t</sup> with the NDVI at both the annual and growing seasonal (from May to September) scales (*R* ≥ 0.6), indicating that the highest dependency of the present dry-wet condition was on that of the preceding 12 months. In this study, SPEI 12 was used as the indicator to analyze the spatio-temporal variations of the dry-wet conditions at the annual and growing season scales in the YZR basin.

**Figure 4.** Changes of SPEI at different time scales.

**Figure 5.** Changes of the annual (**left**) and growing season (**right**) SPEI.

As shown in Figure 5, in terms of the annual time scale, the SPEI showed a significantly increasing trend with a rate of 0.07/decade from 1982 to 2015 ( *P* < 0.1, *n* = 34), implying an overall wetting tendency in the YZR basin. However, Wang (2016) and Liu (2015) demonstrated that the precipitation, temperature, and potential evapotranspiration, which were closely related to the dry-wet conditions in the YZR basin, all changed significantly in the late 1990s [67,68]. In order to detect whether there was a transition from wet to dry during the 1990s in the YZR basin, the year of 2000 was taken as the turning point to investigate the variation characteristics of SPEI. Interestingly, the two periods divided by 2000 exhibited the opposite changing trends, i.e., the wetting period was from 1982 to 1999 with a slightly increasing rate of 0.225/decade ( *P* > 0.1, *n* = 18), and the drying period was from 2000 to 2015 with a significantly decreasing rate of 0.25/decade ( *P* < 0.1, *n* = 16). Compared to the variation trend of the annual SPEI, the growing season SPEI showed a relatively non-significantly increasing trend with a rate of 0.053/decade from 1982 to 2015 ( *P* > 0.1, *n* = 34). The opposite changing trend before and after 2000 was also exhibited at the growing season scale, where there was a slightly increasing trend with a rate of 0.16/decade ( *P* > 0.1, *n* = 18) from 1982 to 1999 and a slightly decreasing trend with a rate of 0.21/decade ( *P* > 0.1, *n* = 16) from 2000 to 2015.

In order to explore the spatial evolution characteristics of the dry-wet regime in the YZR basin, the SPEI was divided into three categories according to Table 1, i.e., SPEI < −0.5 (Dry), −0.5 < SPEI < 0.5 (Normal), and SPEI > 0.5 (Wet). Figure 6 shows the area ratios occupied by the di fferent ranges of SPEI from 1982 to 2015 in the YZR basin at both the annual and growing season scales. The proportions of the annual variation pattern of the dry, wet, and normal areas were consistent with those of the growing season. Unlike the temporal opposite variation trend that occurred before and after 2000, the spatial reversal phenomenon of the dry-wet regime occurred in the three-year wet period from 1999 to 2001. Before this wet period, the proportion of wet areas showed an increasing trend, while the proportion of dry areas declined, causing a wetting condition of the YZR basin. The dry-wet regime reversed after this period, which experienced a drying period as implied by the decrease in wet areas and the increase in dry areas.

**Figure 6.** Area proportions of the dry, wet, and normal areas indicated by the annual **(left**) and growing season (**right**) SPEI in di fferent ranges.

The determination of the turning point of the dry-wet regime in the YZR basin was of grea<sup>t</sup> importance to conduct further investigation in this study. As the most direct evidence for the dry-wet condition at the river basin scale, runo ff has been widely used to represent the dry-wet characteristics [69–71]. The Nuxia hydrological station located downstream of the YZR basin, controls about 80% of the drainage areas of the YZR basin, and its long-term variation of runo ff could be used to e ffectively indicate the dry-wet transition in the YZR basin. In this study, time series of runo ff from 1982 to 2015 at the Nuxia hydrological station were adopted to further detect the transition point of the dry-wet regime by using the Mann-Kendall nonparametric test. The results of the Mann-Kendall significance test showed that the runo ff showed a significantly increasing trend from 1982 to 1999 (*Zc* = 1.89, *P* < 0.1), while the runo ff exhibited a significantly decreasing trend ( *Zc* = −1.76, *P* < 0.1) from 2000 to 2015. Meanwhile, there was no point detected from 1982 to 2015 that could divide the time

series into two parts with a significantly increasing and decreasing trend respectively. Combining with the SPEI spatio-temporal variations, it could be concluded that there was a transition of the dry-wet regime in the YZR basin, which occurred at the year 2000.

To investigate the spatial variation of the dry-wet regime and given the high spatial heterogeneity of the climate conditions in the YZR basin, the spatial distributions of the mean annual and growing season SPEI before and after 2000 were interpolated by using the Kriging method (Figure 7). It could be seen that the annual spatial distribution of the SPEI showed a high consistency with that of the growing season during both periods (1982–1999 and 2000–2015). Before 2000, the dry areas were mainly located in the eastern upstream and midstream regions of the study area, while the western upstream and southeastern downstream regions were relatively wet. Similar to the temporal reversal phenomenon at the year of 2000, the spatial pattern of the SPEI also displayed a reversal phenomenon. The annual and growing season spatial distribution of the SPEI indicated that the spatial distribution of the dry-wet regime in the YZR basin before and after 2000 was opposite. According to Figures 5 and 7, the overall wetting tendency of the YZR basin was mainly attributed to the remarkable wetting trend of the midstream region over the past 34 years, while the prominent trend of drying from 2000 to 2015 in the basin was ascribed to the fact that the western upstream and southeastern downstream turned drier.

**Figure 7.** Annual (**a, b**) and growing season (**c, d**) spatial distributions of SPEI.

To further explore the mechanism of the spatial reversal of the dry-wet regime in the YZR basin around 2000, the annual and growing season *slopes* of the SPEI from 1982 to 2015 with the significance test were analyzed at the pixel scale (Figure 8). It could be found that the western upstream and southeastern downstream regions in the YZR basin presented a gradually drying trend over the past 34 years, while the eastern upstream and midstream regions became wetter, which was consistent with characteristics shown in Figure 7. According to Figure 8a, the mean annual SPEI exhibited an upward trend (*slope* > 0) accounting for 55.07% of the total basin area, indicating a humidification process, whereas an opposite trend (*slope* < 0) occurred in the remainder of the basin. As shown in Table 5, the results of the significance test at the pixel scale indicated that the areas where the annual SPEI showed an extremely significant decrease and a significantly decreasing trend accounted for 10.9% and 2.3% of the total basin respectively, while the areas with an extremely significant increase and a significantly increasing trend of the annual SPEI accounted for 18.05% and 8.25% respectively, inducing an overall wetting process from 1982 to 2015 in the YZR basin. Similar to the annual SPEI, the growing season SPEI indicated a wetting trend (*slope* > 0) across 55.54% of the basin area, while a drying trend (*slope* < 0) occurred in the remaining areas. The results of the significance test for the

growing season SPEI at the pixel scale indicated that the area proportions with an extremely significant decrease and a significant decrease in SPEI were 8.7% and 2.03% respectively, while the areas with an extremely significant increase and a significant increase in SPEI occupied 15.21% and 8.87% of the basin respectively.

**Figure 8.** Annual (**a, b**) and growing season (**c, d**) variation trends of SPEI with the significance test.


**Table 5.** Area ratios of the SPEI occupied by different trends.

Comprehensively taking the results of the temporal variations (Figure 5), spatial distribution characteristics (Figure 7), and trend analysis with the significance test (Figure 8) of the annual and growing season SPEI into account, the annual spatio-temporal variation characteristics of the SPEI were consistent with those of the growth season, which both showed a reversed phenomenon before and after 2000. In terms of the temporal variation, the YZR basin presented a wetting trend before 2000 and a drying trend after 2000, while from the perspective of the spatial pattern, the arid areas became wetter and humid areas became drier.

#### *3.3. Spatio-Temporal Characteristics of Vegetation*

Regarding 2000 as a turning point indicated by the transition of the dry-wet regime in the YZR basin, temporal variations of the annual and growing season NDVI were analyzed (Figure 9). The mean annual NDVI was 0.27, which fluctuated between 0.25 and 0.28 over the past 34 years and significantly increased at a rate of 0.002/decade (*P* < 0.1, *n* = 34), implying a gradual improvement of the vegetation cover in the YZR basin. However, the NDVI before and after 2000 showed a completely opposite tendency, that is, the NDVI increased significantly at a rate of 0.01/decade before 2000 (*P* < 0.1, *n* = 18), and decreased significantly at a rate of 0.006/decade after 2000 (*P* < 0.1, *n* = 16). With respect to the variations of the growing season NDVI, the mean NDVI was 0.34, which fluctuated between 0.31 and 0.35 during the past 34 years and increased non-significantly at a rate of 0.002/decade (*P* > 0.1, *n* = 34). Similar to the changes of the annual NDVI, the growing season NDVI before 2000 showed a significantly increasing trend at a rate of 0.015/decade (*P* < 0.1, *n* = 18), and a significantly decreasing trend at a rate of 0.009/decade after 2000 (*P* < 0.1, *n* = 16). To sum up, the NDVI-indicated vegetation in the YZR basin gradually improved from the perspective of the annual and growing season NDVI variations from 1982 to 2015. However, since the beginning of the 21st century, the vegetation cover has decreased noticeably which corresponds with the simultaneously drying tendency in the whole basin, demonstrating that the improvement of the vegetation cover before 2000 was mainly induced by the gradual wetting of the basin while the degradation of the vegetation cover was attributed to the drying of the basin after 2000.

**Figure 9.** Changes of the annual (**left**) and growing season (**right**) NDVI.

The mean annual and growing season spatial distributions of the NDVI from 1982 to 2015 are portrayed in Figure 10. For the mean annual NDVI, the areas where the NDVI values ranged from 0.1 to 0.3 accounted for approximately 61.08% of the YZR basin, and were mainly located in the upper and middle reaches and high-altitude areas of the downstream regions. Only 5.59% of the area had an NDVI value above 0.7, and were largely concentrated in the midstream and southeastern downstream regions. From the perspective of the whole basin, the mean annual NDVI values gradually increased from northwest to southeast, implying a consistent improvement of the vegetation cover. The growing season vegetation cover indicated by the growing season NDVI exhibited a similar spatial distribution of the mean annual vegetation cover. The spatial variations of the NDVI-indicated vegetation in this study were similar to the vegetation cover dynamic monitoring results in the YZR basin reported by Jiang et al. [45].

**Figure 10.** Annual (**left**) and growing season (**right**) spatial distributions of NDVI.

The spatial characteristics of the variation trends for the mean annual and growing season NDVI indicated by the NDVI *slope* in the YZR basin from 1982 to 2015 are depicted in Figure 11a,c respectively. In terms of the mean annual NDVI, approximately 59.4% of the NDVI *slope* in the basin was greater than 0, denoting an increasing trend of the vegetation cover, while the *slope* of the NDVI in the remaining areas was less than 0, implying a degradation of the vegetation cover. As shown in Figure 11b, d and Table 6, the results of the significance test showed that the areas with an extremely significant decrease and significant decrease of NDVI accounted for 7.3% and 4.84% of the whole study area respectively, while the areas with a non-significant increase and non-significant decrease of NDVI took up 31.24% and 27.86% respectively, and the areas with an extremely significant increase and significant increase of NDVI mainly located in the middle reaches occupied 22.76% and 6.1% respectively. The growing season NDVI showed a similar variation characteristic with that of the annual NDVI, i.e., the areas with increased vegetation cover had a proportion of 57.13% while the NDVI *slope* of the other areas was less than 0, indicating the degraded vegetation cover.

**Figure 11.** Annual (**<sup>a</sup>**, **b**) and growing season (**c**, **d**) spatial variations of NDVI with the significance test.


**Table 6.** Area ratios of NDVI with different change trends.

In terms of the spatial variation characteristics for both the mean annual and growing season NDVI, it could be found that the vegetation cover upstream of the YZR basin seldom changed, which could be attributed to the specific land cover types in the upper reaches including the Gobi Desert, glaciers, and plateau meadows, which were less affected by climate change. However, in the midstream region, except for the high-altitude areas at the edge of the basin, the vegetation cover showed a dramatical upward tendency, while in the downstream region, except for the high-altitude areas such as Bomi, the vegetation cover presented a tendency of extremely significant decrease, which were consistent with the results obtained by Lv et al. [72]. Combined with the transition characteristics of the dry-wet regime in the YZR basin, it could be revealed that the vegetation cover increased in the midstream and eastern upstream regions where the climate became wetter, while the vegetation cover in the downstream and western upstream regions decreased where the climate turned drier.

#### *3.4. Response of the Vegetation to the Dry-Wet Transition*

Based on the analysis results of the spatio-temporal evolution characteristics of the SPEI and NDVI, it could be seen that the vegetation cover was closely related to the dry-wet regime in the YZR basin. To further explore the influence of the dry-wet conditions on vegetation cover, a correlation analysis between the SPEI and NDVI for both annual and growing season from 1982 to 2015 at the pixel scale was conducted. As shown in Figure 12, about 71.57% of the basin area showed a positive correlation between the mean annual SPEI and NDVI, while the remaining areas exhibited no or negative correlations. The results of the significance test as shown in Table 7 demonstrated that the areas where the NDVI was extremely significantly negatively correlated and significantly negatively correlated with the SPEI accounted for 0.85% and 0.95% respectively, and were mainly located in the eastern midstream and northwestern downstream regions. The areas showing a non-significantly positive correlation and non-significantly negative correlation between the NDVI and SPEI took up 56.28% and 26.37% of the basin area respectively, and were mainly concentrated in the western upstream and part of the midstream regions. In addition, the areas where the NDVI was extremely significantly positive and significantly positively correlated with the SPEI occupied 10.28% and 5.27% of the basin respectively, and were mostly concentrated in the junctions of the middle and upper reaches and the southeastern downstream regions. As for the correlation analysis between the growing season NDVI and SPEI, approximately 65.96% of the total area presented a positive correlation. The results of the significance test indicated that the areas showing an extremely significantly negative correlation and a significantly negative correlation between the NDVI and SPEI took up 1.36% and 1.55% respectively, while the areas where the NDVI showed a non-significantly positive correlation and a non-significantly negative correlation with the SPEI accounted for 49.62% and 30.98% respectively. The extremely significantly positive and significantly positive correlation between the NDVI and SPEI accounted for 10.73% and 5.8% of the total area respectively.

**Figure 12.** Annual (**<sup>a</sup>**, **b**) and growing season (**<sup>c</sup>**, **d**) correlation analysis between the SPEI and NDVI with the significance test.

˄ ˅ By combining the temporal variations (Figures 5 and 9) with the spatial distributions of the variation trends (Figures 8 and 11) for the SPEI and NDVI, it was unequivocal that the spatio-temporal variation characteristics of the SPEI were consistent with those of the NDVI in the YZR basin, implying the important role of the dry-wet conditions on the vegetation dynamic variations. In terms of the temporal variation, before 2000, the YZR basin exhibited a wetting tendency and simultaneously the vegetation cover increased, while a tendency of drying was presented after 2000, and the vegetation cover consistently decreased. Such a synchronization phenomenon was also revealed from the perspective of the spatial distribution where the western upstream and southeastern downstream

regions showed a drying trend with decreasing vegetation cover, while the midstream and eastern upstream regions displayed a wetting trend with increasing vegetation cover.


**Table 7.** Area ratios of NDVI and SPEI in different correlation degrees.
