5 Discussion

5.1 Impact mechanism of reservoir

One of the main functions of the reservoir is to regulate runoff to reduce the influence of floods. Usually, a part of the reservoir capacity is discharged before the wet season arrives, and some of the peak flow is intercepted during the wet season. This regulation will obviously reduce flood peaks and increase baseflow during dry season, but will not directly affect the total runoff. In this study, the average streamflow of the downstream Zhenxi station decreased by about 14% after the construction of the reservoir, which may be mainly due to the increased direct consumption of water surface evaporation and indirect consumption of drinking water, farmland irrigation, etc. (T. Maavara, Lauerwald, Regnier, & Van Cappellen, 2017; Vorosmarty et al., 1997). In other words, the regulation of runoff by the reservoir directly leads to the reduction of peak flow and the increase of baseflow, and the increased direct and indirect consumption of reservoir storage leads to the reduction of average or total streamflow. Some studies have well illustrated the above point of view: Piman, Lennaerts, and Southalack (2013) showed that hydropower dams in the tributaries of the Mekong River will lead to a 25% decrease in the streamflow during the wet‑season and an increase of 95% during the dry‑season; In addition, Piman, Cochrane, and Arias (2016) also pointed out that the impact of hydropower dams on the downstream daily average streamflow is greater, with the maximum daily streamflow decreasing by 36% and the minimum increasing by 168%. Vorosmarty et al. (1997) pointed out that the loss of Lake Nasser on the Nile River caused by evaporation and leakage accounts for about 13% of its inflow every year, Lake Kariba on the Zambezi River loses about 20% of its inflow, while the smaller Tiga Reservoir loses about 26% of its upstream inflow. Shiklomanov (2000) pointed out that the reservoir‑based evaporation would increase the consumption of the global river runoff by about 5%, which would be larger if the domestic, industrial and agricultural water consumption were taken into account.
As mentioned in Section 4.2, after the construction of the reservoir, the recession coefficient a of downstream Zhenxi station is significantly reduced, while b is significantly increased, which leads to the change of the master recession curve and S-Q relationship. Why is there such a change? We try to find the answer from the scatter plot of log (|dQ /dt |) vs. log (Q ). By comparing the log (|dQ /dt |) vs. log (Q ) points distribution characteristics before and after the construction of the reservoir at Zhenxi station, we find that the scatter points at the low flow stage after the construction of the reservoir are significantly shifted to the right (Fig. 6a). This also results in a counterclockwise rotation of the linear regression line, which eventually increases the slope (b ) and decreases the intercept (a ). This shift indicates that after the construction of the reservoir, the downstream baseflow (Q ) increased during the dry‑season, but the recession rate (dQ/dt) did not change significantly, which can be understood as the addition of a constant flow (Qcon ) in the original recession process. To verify this assumption, we add a constant value (1m3/s) to the master recession curve (MRC) of Zhenxi station before the construction of the reservoir, and then draw the log (|dQ /dt |) vs. log (Q ) scatter plots (Fig. 6b). It is found that the scatter points in the low flow stage are also shifted to the right after increasing the constant value of 1m3/s to the MRC. The recession coefficienta calculated by the linear regression method is reduced to 0.01, and b is increased to 1.22, which are basically the same as the calculated values after the construction of the reservoir. In addition, it can be seen from Table 2 that after the construction of the reservoir, the Qmin of Zhenxi station increased from 0.29m3/s to 1.38m3/s, increased by 1.09m3/s, and the 7Qmin increased by 7.03m3/s, which should not be a coincidence. That is to say, after the construction of the reservoir, the variable water storage of the basin is increased, and a relatively constant flow (about 1m3/s) is added to the baseflow recession process of downstream Zhenxi station, which causes the log (|dQ /dt |) vs. log (Q ) scatter points to shift to the right in the late recession stage, and finally causes the recession coefficient a to decrease and b to increase. This relatively constant flow may come from the continuous and stable drainage of the reservoir, the continuous leakage, and the continuous return flow of domestic, agricultural and industrial used reservoir water. Specific sources still require detailed field investigations.
Figure 6. -dQ/dt vs. Q scatter plots in double logarithmic coordinate system, with the Vog extraction method. The lines in the figure are linear regression lines. (a) Observed values before and after the construction of the reservoir, (b) calculated values before and after adding a constant flow.

5.2 Master recession analysis considering constant addition flow

It can be seen from Fig. 6b that after adding a constant flow, the log (|dQ /dt |) vs. log (Q ) scatter points no longer satisfies the linear relationship, and the slope gradually becomes larger in the late stage of the recession. The recession coefficients calculated by the traditional linear parameterization method reflect the average linear approximation of this nonlinear relationship. Therefore, a new master recession analysis method should be adopted to analyze the baseflow recession process with a constant addition flow. For example, we can deduct the constant flow (Qcon ) before drawing the log (|dQ /dt |) vs. log (Q ) scatter plot for parameterization (i.e. all measured recession streamflow minusQcon ), and then add Qconto the master recession curve (MRC). At this time, the equation of the MRC becomes\(Q_{(t+t)}=Q_{\text{con}}+\left[a\left(b-1\right)t+{(Q_{t}-Q_{\text{con}})}^{1-b}\right]^{\frac{1}{1-b}}\), where a and b in the formula are obtained by linear regression of the scattered points after subtractingQcon . After subtractingQcon = 1m3/s from the recession streamflow of Zhenxi station from 2009 to 2013, the recession coefficients a and b are 0.022 and 1.10 respectively, which are close to the analysis results of 1982 to 1986. Finally, the MRC considering constant addition flow is drawn (Fig. 7). After comparison, it can be found that the MRC obtained by the average linear approximation will overestimate the streamflow in the middle recession and underestimate the streamflow in the later recession, but the deviation is not obvious as a whole. That is to say, in Sect. 4.2, the recession analysis results obtained by linear approximation can reflect the overall change of the recession characteristics before and after the construction of the reservoir, but there will be some deviations in some specific stages. The master recession analysis method proposed in this paper can also be used in basins with inter‑basin water transfer behavior. When there is stable inflow, Qcon is positive, and when there is stable outflow, Qconis negative.
Figure 7. Master recession curves with and without constant addition flow, semi logarithmic coordinate.