5.1. Sensitivity Analysis for Calibration and Validation of the Numerical Model
A river system model was created in the HEC-RAS according to
Figure 3. It includes Jhelum River and its tributaries. The river x-sections helped to create the bathymetry and bank geometry. Water and sediment discharges (divided into sediment fractions of sand, silt and clay) on daily basis were applied to each river at its upper most x-section as upstream boundary conditions. The dam face formed the model end section on the downstream side where the daily reservoir level in a.s.l. was applied as the downstream boundary condition.
Sensitivity analysis for the selection of certain parameters like roughness coefficients, computational time step, sediment transport functions, and fall velocity was performed to simulate bathymetric changes in the reservoir. After importing all the input data to the HEC-RAS model, the model was run for different Manning’s values going from 0.02 to 0.04 with the incremented by 0.005 and by keeping a computational time step of 12 h with the model default sediment parameters (the Ackers and White method as the sediment transport function, Exner 5 as the bed sorting method, and Report 12 as the fall velocity method). It is found that low values of
n lead to high erosion and cause high sediment transport rates because of the higher flow velocities. The simulated bed profile for Manning’s value of 0.03 was found to agree closely with the observed bed level (
Figure 9), and statistical analysis also showed that the value of
n = 0.03 is the most suitable one because it has a minimum RMSE value and NSE, R
2 value is close to 1 as shown in
Table 2. Thus,
n = 0.03 was adopted as the project default value. Further, it has also been observed that the increasing value of
n from 0.02 to 0.03 leads to the computed bed profile closer to the measured profile, and its further increased value (
n = 0.035) shows more deviation from the observed. Thus, it is observed that a nonlinear relation exists between reservoir sedimentation and the Manning’s roughness coefficient.
Further, to analyze the effect of time step, the model was run for the computational increment as 24, 12, 9, 6, and 3 h, by keeping other parameters same as default. It is observed that the computed results are sensitive to the varying computational increment. The smaller time step gives more stability to the numerical model, but it increases the run time. Therefore, it is an important model parameter that directly influences the model’s efficiency. The comparison of the observed and computed results for the year 2010 and 2014 are shown in
Figure 10, and from statistical analysis (
Table 2), it is observed that similar results were found for the computational increment equal or less than 6 h. Hence, in the present study, 6 h is selected as computational increment time step as the project default value for further analysis.
Furthermore, the model was calibrated on observed longitudinal bed profile data and delta advance rate, from 2005 to 2010, by selecting one by one all the sediment transport functions and different fall velocity methods available in HEC-RAS software by keeping the 6-h computational increment time and the Manning’s roughness coefficient value as 0.03. The results showed that the Ackers and White sediment transport function combined with the Van Rijn fall velocity method was found to be the best among the lot. All other transport functions underestimated the location of the delta pivot point, but overall, a good agreement was found between the observed and simulated longitudinal bed profiles, as shown in
Figure 11. The original ground level (OGL) of the reservoir in 1967 is also shown in
Figure 11.
All functions underestimated the deposition close to the dam face; the earthquakes of 2005 and 2008 could be the reason behind the movement of the foreset slope. The result of the calibration in 2010 is shown in
Figure 12.
Model Validation
The numerical model was validated for the years 2014 and 2017 by comparing the computed and observed longitudinal bed profiles of the reservoir, as illustrated in
Figure 13. From
Figure 13, it is revealed that the topset slope closely matches the topset slope of the observed delta profile. A fairly good agreement was found between the simulated and observed measurement of the bed profile exclusively on the far and near of the topset slope and on the pivot point of the sediment delta. However, minor discrepancies have been observed on locations where the river channel is narrow and on the river bends. This is due to the presence of 3D effects, i.e., helical flow due to secondary currents, which can only be resolved by a 3D flow and sediment model [
18]. Further, the model underestimated the sediment deposition close to the dam embankment. Overall, the results of the calibration and validation of the numerical model indicate that the model was able to compute the bed levels with reasonable accuracy, thus pointing to its suitability for the reservoir sedimentation studies.
The statistical analysis was performed to evaluate the calibration and validation of sediment transport model efficiency. The results show that the model performed well in both phases of calibration and validation by considering the combination of the Ackers and White equation as the transport function, the van Rijn equation as the fall velocity method, a Manning’s roughness coefficient of 0.03, and a computational increment of 6 h. The minimum RMSE (calibration 2.18 and validation 3.18) with NSE (0.96, 0.90), R
2 (0.97, 0.93), and PBIAS is close to the perfect values of 1 and 0%, respectively, as shown in
Table 2.
The model efficiency was also assessed by comparing the sediment deposition in different pockets of the Mangla Reservoir as given in
Table 3. It has been observed that the computed volume of deposited sediments in different reservoir pockets are quite close to the observed amount as reported by the dam authorities.
At the majority of the x-section locations, a good agreement between the observed and simulated bed levels was noted, although the results were poor in the lower Jhelum River reach. This is because the said reach is sandwiched between the upper Jhelum and the lower Jhelum reach. It is a short river reach, which connects the two reaches transversally. The channel width is narrow and marked by the presence of bends where a complex 2D or even 3D flow prevails, hence, the wide gap between the observed and simulated sediment results in
Table 3.
However, bed levels near the dam are not well simulated by the numerical model. In the period 2005–2017, the bed levels simulated with HEC-RAS do not follow the increase shown in the surveys (
Figure 12 and
Figure 13). One probable explanation could be the devastating earthquakes of October 2005, 2008, 2011, 2013, 2014, and 2015 (having magnitudes of 7.6, 6.4, 7.2, 7.7, 4.5, and 6.3, respectively) in northern Pakistan, which may have caused landslide/subsidence of the foreset slope of the sediment delta. It is clear that such phenomena cannot be modeled by HEC-RAS.
In
Figure 13, it can be seen that, in the year 2014, the bed levels near to the dam embankment are higher than the tunnel intake levels. The fact that the tunnels still function and have not been choked by the higher flow velocity is the eroding of the sediment deposits. These localized phenomena, of course, cannot be replicated by a 1D model, as they occur due to lateral variation of the flow velocity, which requires at least a 2D model for its description.
5.2. Impact of Future Reservoir Operating Scenarios
The reservoir operation depends upon a number of factors, i.e., inflows and its seasonal variation, downstream water level, MOL, sediment movement pattern and rate, flow rate change due to variable climate, trap efficiency, live and dead storage, etc. Clearly, determining the optimum reservoir operation strategy is by no means evident. Thus, we adopted a heuristic approach that allows us to select the best strategy and at the same time explains the trade-offs between the main factors by developing a set of plausible operating scenarios as listed in
Table 4.
The results for Scenario S1 with projected future discharges from RCP4.5 and the simulation outcomes for the years 2025, 2035, 2045, and 2055 are shown in
Figure 14. As the future projections for the Mangla Basin predict an increase in flows, Scenario 1, which keeps the MOL equal to the existing 319 m a.s.l., leads to rapid sedimentation. As a result, the delta moves rapidly toward the dam face. In this scenario, the bed levels close to the main embankment aggraded rapidly and reached the level 320 m a.s.l. in the year 2045, thus, rapidly filling the dead storage space.
Figure 15 represents the pivot point location and the bed profiles after 30 years for the five reservoir operation scenarios.
This shows that the raise in MOL slows down the rate of delta movement towards the turbine intakes as more and more sediments are deposited along the topset slope and in the upstream river reaches. Furthermore, it is clearly observed that by increasing the minimum reservoir level, the live storage capacity decreases gradually and the dead storage remains intact. In Scenarios S3 and S4, the movement of the delta reduced significantly as the high water level slowed down the velocity in areas far upstream from the dam, thus promoting deposition at those locations.
Table 5 shows the mean advancement rate of the Mangla delta toward the dam face.
Table 6 shows the total volume lost as percentage of gross storage capacity (including the upraised Mangla Dam capacity) for the analyzed scenarios using future discharges derived from RCP4.5 and 8.5, which varies from 30.98% to 32.75%. The results reveal that there is a maximum difference of 1.77% in the total storage lost among different operational scenarios. It also shows the impact between the two sets of future discharges (RCP4.5 and 8.5) on reservoir sedimentation up to the year 2065 and has minor influence in view of comparative volume lost (i.e., <1%). Therefore, further results for the RCP4.5 are discussed because RCP8.5 has similar impact as RCP4.5 on Mangla Reservoir sedimentation.
The results indicate that the total volume loss because of sedimentation is around 3.0 BCM in GSC for all the scenarios. In view of these results for all operational scenarios, sediment-trapping capacity is high, but there is a small difference among all of them because of the vast reservoir area and capacity. However, the storage loss by sedimentation in the main Mangla pocket close to the dam has a greater impact, as compared to the total volume lost due to sedimentation in the reservoir, as illustrated in
Figure 16. Scenarios S1 and S2 having a minimum reservoir level show maximum deposition in the main Mangla pocket as compared to other scenarios until 2045. Afterward, a gradual fall has been observed because of increase sediment outflows, as shown in
Figure 17. Further, Scenarios S3, S4, and S5 have progressively increased sediment deposition in the main pocket (
Figure 16) with the lowest sediment outflows (
Figure 17) and minimum delta advancement rate with the highest reduced level of sediment bed at the pivot point (
Figure 15) as compared to Scenarios S1 and S2. This reveals that the rising drawdown level of the reservoir reduced the sediment delta advancement rate and kept the pivot point far away from the dam face for a longer period, but it has also increased the upstream bed level because of high sedimentation. However, keeping the minimum drawdown level in the reservoir triggers high advancement of sediment deposits toward the dam face.
Outflowing sand particles significantly contribute to the wear and tear of the turbine blades and at the outlet surfaces.
Table 7 provides information on the proportion of sand outflows and inflows for four scenarios.
The sand outflows increase with time as lower drawdown levels are maintained and decreases with higher drawdown levels. Simulation S1 has the least drawdown level among the four simulations, demonstrating this problem to the fullest. Initially, the sand outflows are small during the decade 2035–2045 i.e., 0.005 million tons/year but after about a decade as the availability of sand increases, the outflows increase by about 14% to reach 9 million tons/year (
Table 7). Further, no sand outflows were observed through the turbine intakes in Simulation S4 having a minimum reservoir level of 329 m a.s.l.; in Simulation S5, every year, the minimum drawdown level of the reservoir is increased by 0.75 m. Thus, the objective of deterring the volume loss by operating the reservoir at MOL may contravene the objective of passing less concentrated sediment water through turbine intake to prevent wear and tear and vice versa.
The low drawdown level also increases the sand particles proportion in water. When this sand-laden water passes through the hydropower tunnels, it causes abrasion of the mechanical surfaces exposed to high-speed flow, e.g., to turbine blades. Generally, this phenomenon occurs for the particle size exceeding 0.1 mm but may also happen for smaller particles in case of high operation heads and angular quartz particles [
6]. As a result, it may cause clogging or blocking of the outlets. The risk is higher if the dam is situated in a seismically active zone, as Mangla is, due to liquefaction of the accumulated sediments.
The composition of the sediment outflow through the reservoir vary from the low flow season to the high flow season. During the low flow season (February and March), the outflow consists of coarse silt and sand, while in the high flow season (May and June), the composition changes to clay and fine silts. Maximum sediment outflows of coarse silt and sand from the reservoir therefore do not occur at the same time as the peak flows in the river. Hence, this coarser material is deposited in the river downstream of the dam and subsequently re-scoured once the higher flows are released from the dam later in the year. This depositing and reworking of the sediment in the river could cause additional and significant environmental damage. Furthermore, these highly concentrated sediment outflows are also very harmful to biota in downstream river reaches. Recent case studies have estimated detailed biological and physical alterations, reductions of benthic organisms due to bed siltation, and severe fish mortality [
68].
Table 8 gives information regarding outflows of sand fraction, which is pertinent to the above-mentioned risks.
The observed sediment deposition rate at the Mangla Reservoir from 2005–2017 was about 0.023 BCM per year, while rates of deposition for future years under impact of climate change with different reservoir operational rules are show in
Table 9. The sediment deposition rate increases from 0.023 BCM to 0.032 BCM per year due to the increase in summer precipitation and discharges due to climate change, which causes more weathering and erosion in the Mangla catchment area. It also indicates that the deposition rate decreases gradually with the passage of time because total sediments outflows are increasing, as shown in
Figure 17.
To summarize, the minimum reservoir operation level has a significant impact on reservoir sedimentation. The results showed that, by keeping the reservoir MOL lower, a higher volume of water can be made availability for useful purposes e.g., irrigation and hydropower. However, it leads to a rapid increase of the bed levels near to the dam (Scenarios 1 and 2). Additionally, low MOL enhances the sediment outflows, which increases the risk of turbine abrasion and tunnel choking. An opposite measure is to keep the MOL high, which has the disadvantage of reducing water availability, but on a positive note, it slows down the delta advance (Scenarios 3 and 4). A further negative effect of high MOL is the increased foreset slope of the sediment delta, which might collapse, thus leading to mass movements of the bed load into the power tunnel(s). Considering all these factors, a strategy (Scenario S5) that leads to a gradual increase in MOL of the reservoir with time is the optimal strategy because it slowly introduces the change on yearly basis, arresting the delta advancement, preserving the live storage, and keeping the sand outflows within their reasonable limits.
5.3. General Lessons Learnt from the Present Study
The present study results can be generalized to multi-purpose reservoirs situated in water stressed, arid, or semi-arid zones. The reservoir can have multiple objectives e.g., irrigation supplies, hydropower, flood control, etc. It is also well known that streams in arid zones carry far more sediment than temperate regions that arrive in great quantities during flood seasons. Thus, reservoir sediment management has a very vital role to play in prolonging the life of a reservoir.
From the reservoir operations viewpoint, it is of paramount importance to ensure sustainable use of the reservoir by reducing sedimentation and not allowing sediment to approach the power tunnels or water intakes too closely. This would prevent the clogging of the latter. The threat is all the more palpable in a seismically active zone where sand liquefaction can cause highly turbid water to invade the installations, leading to its shut-off and degradation.
The sediment management is greatly dependent upon the reservoir operations policy, which fixes the reservoir minimum water level. If the water level is low, more turbid water can be made to pass through the tunnels, thus promoting flushing of the sediment. However, it has the inconvenience of allowing the sediment to deposit more closely to the tunnels/intakes and degrading the turbines. The other option is to keep the reservoir water level high, which may not be possible due to end-users being dependent on the water, or, the reservoir has to be depleted to a certain level before the arrival of a flood season to store a flood peak. Further, the high water level can cause sediment mounds to form far from the intakes, but this also increases the susceptibility to slope failure or slumping. The MOL may be set low in initial years after the commissioning of the reservoir, but it may have to be raised gradually to reduce the sediment accumulation in later years.
The suggestions that can be advanced based on the present case study recognize the fact that flows may need to be passed through the installations housing hydraulic machinery; however, at the design stage, it would be prudent to reserve some low-level outlets only for flushing and not to install turbines on those. A maximum “shut off” sediment load has to be defined for an installation at which it must be immediately shut down to protect the installation. To avoid the risk of installation clogging during an earthquake or slumping, a “deep water suction dredging system” of adequate capacity should be procured in advance to clear the blockage as soon as possible. Further, regular repair of exposed hydraulic surfaces should be undertaken to repair the damage done by silt-laden water. The turbines selected should be able to withstand the high sediment load. Further, options may be explored to protect the machinery through specialized coatings e.g., ceramics and polyurethane. Furthermore, outflows near to the MOL should be strictly monitored to assess its effect on the downstream fish and benthic organisms, and measures may be taken to protect the vulnerable species during the occurrence of such flows.