**3. Results**

#### *3.1. Reservoir Impact Evaluation on Lhasa River Flow*

#### 3.1.1. Double-Mass Curve Analysis of Lhasa River Flow

To reckon LR discharge change under reservoir influence, DMC analysis, along with regression lines for two time spans, was carried out to better understand the hydrological phenomena and the likely change years in the time series. The double-mass curves for annually recorded rainfall and discharge, following the work of Searcy and Hardison (1960) [75], were individually applied for the time spans of 1956–2016 and 2000–2016 (Figures 3 and 4) respectively. The application of individual cumulative mass curves for two time spans was done with the aim of developing a more valid and reliable impact assessment in terms of change in the hydrological time series of the LR.

The DMC analysis of LR discharge from 1956 to 2016 revealed a nearly proportional behavior of the rainfall in correspondence to the measured LR discharge. However, we saw certain years of change along the time series that served as break points in the pattern of high and low flows in the LR discharge. The years for change are highlighted and indicated in Figure 3, where the pattern of streamflow breaks to differ from the preceding years. The result of the long-term DMC analysis was the identification of the change years, of which the years 2007 and 2013 were of particular significance for the current study. These two years marked the operation of the two major reservoirs (Zhikong and Pangduo, respectively) that were considered for the current study. The impact of chosen reservoir functioning on the hydrological behavior of the LR manifested itself in the long-term DMC analysis.

To further understand the phenomena of reservoir influence on LR discharge, doublemass analysis was applied to the time series from 2000 to 2016 (Figure 4), and we saw three identified change points in the time series during these years.

**Figure 3.** Double-mass curve for cumulative rainfall and cumulative discharge of the Lhasa River for the time span of 1956–2016. The years for change in hydrological time series are highlighted and supported by text.

**Figure 4.** Double-mass curve for the cumulative rainfall and cumulative discharge of the Lhasa River for the time span of 2000–2016. The years for change in hydrological time series are supported by the text.

The year 2003 showed a change, as the maximum rainfall was recorded in this year and produced the simultaneously highest discharge during the year for the chosen study time period. The next identified change year was 2007, which deviated from the streak of data points along the regression line. This was the time when one of the selected reservoirs in the study was built on the LR. The Zhikong hydropower station was completed in 2006 and started functioning in September 2007. The most prominent break point in the hydrologic time series was identified in 2013 when the second major reservoir started operating on the LR, i.e., the Pangduo hydropower station, where the data points deviated from the regression line and indicated a peculiar hydrological behavior in the LRB. Yet again, the hydraulic interventions demonstrated themselves in the form of change points across the study time span of hydrologic time series.

#### 3.1.2. Variation Assessment of Lhasa River Streamflow under Reservoir Operations

Figure 5 shows the inter-annual variation of the hydro-meteorological behavior of the LRB along the two time phases. The CV for the hydro-meteorological phenomena from 1956 to 1999 and from 2000 to 2016 exposed an aggravated variability in the latter time span compared to the previous time span. The CV of 24% for LRB precipitation during 1956–1999 was lowered to 20% in the second time span of 2000–2016; however, the CV values for both time spans were relatively closer, which means that the change in the pattern of precipitation advanced with a greater pace in the study time span compared to the former long time span of 1956–1999. Similarly, for the annual temperature, the CVs of 8% and 6% for the time spans of 1956–1999 and 2000–2016, respectively, were again closer values and revealed a faster temperature change in the LRB during 2000–2016.

**Figure 5.** Changes of annual river discharge (at the Lhasa hydrometric station), annual mean temperature, and annual recorded precipitation for the Lhasa River Basin from 1956 to 1999 and from 2000 to 2016 (study time period with reservoirs functioning in the study area).

Since it is a typical QTP catchment, the LRB is prone to complex climate change phenomena [25]. These climate variables are closely associated with the hydrologic cycle. Particularly for the LRB, the LR discharge is furnished by the precipitation [25], and

variability in rainfall poses a direct influence on the hydrological behavior of the LRB. Temperature is also an important feature in determining hydrological phenomena because it asserts its influence in the form of evapotranspiration, and, thus, variations in the temperature of the LRB may have a potential impact on the water resources in the area.

The CV values for LR discharge revealed an increased variability in 2000–2016 compared to 1956–1999. The CV of 25% for 1956–1999 copiously increased to 34% during the study time span of 2000–2016. With rainfall being the determining factor for discharge in the LRB, we saw that during the years of 1956–1999, the CV for rainfall and LR discharge were very close at 24% and 25%, respectively, thus indicating a close correspondence among them. Conversely, a large difference in the variability of rainfall and LR discharge was unveiled during the study time span of 2000–2016. This signified that during this time span, apart from the climatological justification, some other factor proclaimed its influence on the LR's hydrological dynamics.

The LR was subjected to some major hydraulic interventions during the years of 2000–2016, and the increased discharge alteration can be well-attributed to the reservoir operations in the LRB during this time period. The Zhikong and Pangduo reservoirs became operational in 2006 and 2013, respectively, establishing a clearly visible modification in the LR's hydrological regime, as presented in Figure 5. The LRB is experiencing an aggravated climatological variation accompanied with human interferences, resulting in a substantially altered hydrological phenomena in the area that warrants better planning and managemen<sup>t</sup> practices in future.

#### *3.2. MK-S Trend Analysis on Lhasa Streamflow under Reservoir Influence*

Investigations of the trends in the time series of hydrological data were found to be an imperative means for the detection and understanding of changes in a rainfall–runoff process. Their results are exploitable in water managemen<sup>t</sup> planning and flood-protection. Climatic changes, together with a different type and stage of human impact, are considered to be the main causes of rainfall–runoff changes [76]. In the current study, an MK-S test was applied to determine the direction and magnitude of the trend of hydro-meteorological phenomena of the LRB; the findings are presented as Table 1.

**Table 1.** Trend analysis on hydro-meteorological variables of the Lhasa River Basin. MK–τ represents Mann–Kendall's trend at *p* = 0.05 (bold values are significant at *p*-value), and S represents the Sen's slope estimator for change. The negative sign indicates a decrease.


Dams influence variations in river discharge, particularly over seasonal time scales [77,78]. The seasonal variation in LR discharge is presented in Table 2, where maximum variation is shown to be have been experienced by the high flow months of the wet monsoonal season from June to October with a CV value of 62%, followed by the spring season from March to May with a CV value of 56%. The dry winter season from November to February was foundtoexperiencetheminimumvariabilitywithaCVof47%.

## *3.3. Lhasa River Streamflow Simulation and Prediction*

3.3.1. SWAT Modeling of Lhasa River Flow under Reservoir Influence

In the current study, the SWAT model identified nine parameters sensitive to the runoff generation phenomena of the LRB. The sensitivity, ranges, and optimum values of the selected parameters for the study (as identified by SUFI-CUP) are presented in Table 3. The model ranked SOL\_BD, EPCO, GW\_REVAP, ESCO, and GW\_DELAY as the most influential parameters in controlling the runoff phenomena in the LRB. This indicated

that the LR discharge is predominantly controlled by the soil physical characteristics, evapotranspiration, and ground water processes in the LRB. This was supported by the previously discussed seasonal MK-S trend results for LR discharge, which also indicated a strong association of evapotranspiration phenomena and ground water movement in the LRB in the runoff generation process, particularly in the dry winter season.

**Table 2.** The change and trend on seasonal Lhasa River discharge for the time period of 2000–2016. CV stands for coefficient of variation, MK–τ represents Mann–Kendall's trend at *p* = 0.05 (bold values are significant at *p*), and S represents the Sen's slope estimator for change in LR discharge (m<sup>3</sup> s<sup>−</sup><sup>1</sup> month−1). The negative sign indicates a decrease.


**Table 3.** Sensitivity of selected parameters in influencing Lhasa River flow.


"r" denotes the relative method, and "v" denotes the replace method.

The performance of the SWAT model in simulating LR discharge under the chosen reservoirs' influence for the time span of 2000–2016 is presented in Figure 6a. A comparison of observed and simulated LR discharge is shown in Figure 6b. The simulated hydrological time series corresponded appreciably well to the observed data series and regularly fluctuated with the precipitation pattern. The high peaks were very well captured by the SWAT model most of the time, particularly during the calibration years (2005–2010), with a few being under-estimated. For the validation years (2011–2016), the model again managed to capture the high peaks, but some peaks were under-estimated. The lower flow was consistently under-estimated by the model. A similar weakness of the SWAT model in capturing the low flows of the LR was reported in [25]. Overall, the model performed well in simulating the LR streamflow by conforming to the work of Moriasi et al. (2007) [79], where the modeling performance was acceptable if R<sup>2</sup> > 0.5, NSE > 0.5, and PBIAS < ±25%. The performance of the SWAT model during calibration and validation is presented in Table 4. However, the comparison between observed and simulated hydrological data series revealed an R<sup>2</sup> value of 0.75 (Figure 6b), and majority of the values were enclosed by

the 95% prediction and confidence interval. Few of the high flow values were dispersed because they were under-estimated by the model. This confirmed the competency of the SWAT model in simulating the LR discharge under the reservoir operations selected for the current study.

**Figure 6.** (**a**) SWAT simulation of Lhasa River discharge recorded at the Lhasa hydrometric station for time span of 2005–2016. (**b**) Comparison of observed and SWAT-simulated Lhasa River discharge from 2005 to 2016.

The association of observed and SWAT-simulated LR discharge was verified by the correlation tests presented in Table 5. All the correlation coefficients produced high values and thus proved that the SWAT-simulated results could be used to predict the future LR discharge from 2017–2025.

**Table 4.** Performance of the SWAT model in simulation of Lhasa River flow under reservoir operations. p-factor: percentage of data that is enclosed by the 95PPU band; r-factor: the average width of the 95PPU band divided by the standard deviation of the measured variable (from 0 to <sup>∞</sup>, with 0 showing perfect match).


**Table 5.** Statistical correlation of observed and SWAT-simulated Lhasa River discharge.


Bold values are significant at *p* = 0.05.

3.3.2. Seasonal ARIMA Application for Predicting Hydrological Regime of Lhasa River Basin under Reservoir Operations

While making use of the observed LR hydrological time series to identify the future trend of LR streamflow under reservoir operations for the years 2017–2025, the SARIMA model (1, 0, 0) (2, 1, 2)12 was found to be the optimum combination for forecasting of observed streamflow under the cumulative impact of reservoirs by justifying the performance evaluation criteria presented in Table 6 for attaining the lowest AIC and BIC values, a lower RMSE value of 0.29 m3/s, and a MAPE value of only 4.02%—values which confirmed the validity of the model. The SARIMA model was validated for the years of 2013–2016. SARIMA produced closely corresponding predicted values for LR streamflow during the validation time span, with its correlation coefficient of R<sup>2</sup> = 0.80 revealing an efficient model that is capable of predicting the future discharge for the LR. The forecasted monthly LR discharge was seen to follow a decreasing trend during the time period of 2017–2025 under reservoir influence (Figure 7a).

**Table 6.** Performance of SARIMA model in predicting Lhasa River streamflow from 2017 to 2025 using observed and simulated hydrological time series.


The SARIMA model (1, 0, 0) (2, 1, 0)12 was found to be the optimum combination for forecasting of SWAT-simulated streamflow under the cumulative impact of reservoirs. The SARIMA model produced correlation coefficient of R<sup>2</sup> = 0.88 for the validation years from 2013 to 2016 for SWAT-simulated and forecasted LR discharge with a relatively higher MAPE value of 31.09% (Table 6) for the simulation-based forecasted LR discharge. The predicted discharge using the SWAT-simulated hydrological time series likewise showed a decreasing discharge.

**Figure 7.** (**a**) Forecasted monthly Lhasa River streamflow for time span of 2013–2025 using the observed hydrological time series from 2005 to 2016. SARIMA model validation years from 2013 to 2016 are marked. (**b**) Forecasted monthly Lhasa River streamflow for time span of 2013–2025 using SWAT-simulated hydrological time series from 2005 to 2016. SARIMA model validation years from 2013 to 2016 are marked.

The comparison of observation-based and simulation-based LR discharge presented in Figure 8a showed a very close correspondence between both hydrological time series with an R<sup>2</sup> of 0.90. However, the simulation-based forecasted LR discharge was higher for high flow months in future. In advancing through the years from 2017 to 2025, the difference in the high peaks was seen to be increasing among the observation and simulation-based forecasted LR discharge, as presented in Figure 8b. However, both hydrological data series were shown to experience a decrease in the future years researched in the study.

**Figure 8.** (**a**) Comparison between observation-based and SWAT-simulation-based Lhasa River forecasted flow from 2017 to 2025. (**b**) Scatter plot of observation-based and simulation-based forecasted monthly Lhasa River discharge for 2017–2025.

To corroborate the association of both forecasted hydrological time series, statistical correlational tests used in the study produced values of ≥0.80 and are presented in Table 7. This testifies to the credibility of the approach used in the current study and shows that simulation-based future LR discharge can be a replacement to observation-based discharge and be utilized for further analyses regarding water resource management, planning, distribution, hydropower generation, irrigation scheduling, and reservoir operational procedures in the LRB. This can prove to be an aid in overcoming the hydrological data scarcity issue because the LRB is a quintessential basin of the QTP with barely observed data [25].

**Table 7.** Statistical correlation between forecasted observation-based and simulation-based Lhasa River discharge, 2017–2025.


Bold values are significant at *p* = 0.05.

A flow–duration curve offers a practical approach for studying the flow characteristics of streams and for examining the association of one basin with another. A flow–duration curve is a cumulative frequency curve that shows the percent of time during which specified discharges were equaled or exceeded in a given period. A rather easier conception of the flow–duration curve is that it is a streamflow data demonstration combining the flow characteristics of a stream throughout the ranges of discharge in one curve [80]. The flow–duration curves for the observed LR discharge and forecasted observation-based and simulation-based LR discharge are presented in Figure 9.

**Figure 9.** Flow–duration curves for the monthly observed, forecasted observation-based, and SWATsimulation-based Lhasa River discharge.

We saw that the SWAT-simulation-based predicted LR discharge produced a steeper sloped curve following the similar high and low flow pattern of the observed LR discharge. However, the simulation-based predicted low flows dropped drastically through the years taken for the study. On the contrary, the forecasted observation-based LR discharge revealed a flat sloped curve with a remarkably low peak events for the coming years. The authors of [25] also revealed a significantly decreased LR streamflow in future, though under the impact of climate change, and they attributed the decrease to the temperature change. Our study also revealed a considerable decrease in the future LR discharge under reservoir influence. The LRB is the largest inhabited QTP basin experiencing aggravated climate change and human interference impacts on its rainfall-dominated runoff generation mechanism. This suggests a call for better and strategic water resource managemen<sup>t</sup> in the LRB. The findings from the current study can be used by water resource managers and hydropower engineers to develop flow–duration curves for the hydropower plants considered in the study by using their turbine capacity to estimate the required and available flow for producing power in future years. The study can also be replicated in basins with similar characteristic around the globe.
