Large storage operations under climate change: expanding uncertainties and evolving tradeoffs

The Red River basin (figure 1) is a transnational basin covering an area of 169,000 km² between Vietnam (51.3%), China (48%), and Laos (0.7%). The three main tributaries of the Red River (Da, Lo, and Thao) rise in the northern part of the basin and join before reaching the large flood plain in the delta region, which extends over 21,000 km². The basin is characterized by a tropical monsoon climate with two well distinct seasons: the wet season, from May to October, and the dry season from November to April. The average flow in the Red River measured at Son Tay varies from 8,000 m³/s during the monsoon peak to 1,500 m³/s in the dry months.

Zoom In Zoom Out Reset image size

Three main competing sectors drive the current system operations: hydropower production, flood control, and water supply. The first aims at maximizing the energy production, which replaces the profit maximization due to the lack of an energy market. The second sector aims at protecting the city of Hanoi, which is periodically affected by catastrophic floods events during the monsoon season (Kort and Booji 2007). The last sector aims at providing adequate water supply to a variety of water related activities in the Red River delta, where agriculture accounts for 58% of the total water demand and involves around 50% of the local workers in 501,000 ha of cultivated fields fed by a dense artificial canals network. After the Mekong River delta, the Red River delta is the second largest rice production area of the entire Vietnam, which is, in turn, the second-largest rice exporter in the world (Yu et al 2010). All these sectors are expected to be affected by climate change (Arndt et al 2015), thus calling for a better understanding of the climate change impacts on both the overall and the inter-sector system performance. In particular, we measure the system performance in terms of (i) daily average hydropower production(J^hyd (GWh day^-1)); (ii) daily average flood damage, quantified by a dimensionless nonlinear function of the water level in the city of Hanoi (J^flood (-)); (iii) daily average squared deficit with respect to the total water demand of the Red River delta (J^supply (m³ s^-1)²).

We adopt a three-step methodology (figure 2) for assessing the impact of climate change on the multiple sectors affected by the storage operations and for identifying their operational adaptive capacity. Core of the procedure is an integrated model of the Red River system combining conceptual and data-driven models to represent catchments, reservoirs, power plants, river stretches and the delta. All these components have been calibrated and extensively validated against observational data. Their accuracy in reproducing the historical data, measured in terms of coefficient of determination, is reported in table 1, while further details about model building and performance are provided in the supplementary material. The methodology is as follows:

• i.  
  The integrated model of the system is used to design the optimal operations of Son La, Hoa Binh, and Tuyen Quang reservoirs under historical conditions (black lines in figure 2) via evolutionary multi-objective direct policy search (Giuliani et al 2015). This step yields a set of Pareto optimal solutions, representing different tradeoffs between hydropower production, flood control, and water supply, which provide the water authorities with a rich context for supporting the identification of candidate compromise alternatives.
• ii.  
  The set of solutions obtained at the previous point is re-evaluated via simulation under multiple climate projections and different future time horizons to discover the main system's vulnerabilities if no adaptation is implemented (blue lines in figure 2). In addition, this step provides a better understanding of the risk of failure across sectors by exploring the evolution in time of the system tradeoffs and by quantifying the performance uncertainty associated to different climate scenarios.
• iii.  
  The adaptation to the new climate is explored by re-optimizing the operations of Son La, Hoa Binh, and Tuyen Quang reservoirs based on the projected climate change scenarios (red lines in figure 2). This step hence identifies the operational adaptive capacity of the existing reservoirs for mitigating the adverse climate change impacts on the system performance.

Zoom In Zoom Out Reset image size

This work focus on the official climate projections adopted by the Vietnamese authorities, which consider the HadCM3 General Circulation Model (Gordon et al 2000) forced with the A1B emission scenario (IPCC 2000). To study the sensitivity of the system performance with respect to small variations in the climate scenario, we consider a 5-members perturbed physics ensemble of projected temperature and precipitation, provided by the Vietnam Institute of Meteorology, Hydrology and Environment (IMHEN). This ensemble is obtained by using different parameterizations of the HadCM3 General Circulation Model and comprises a baseline scenario (Q0) obtained with the standard model's parameterization; two scenarios using alternative model's parameters that result in smaller and larger temperature increases (Q3 and Q13, respectively); and other two scenarios obtained by model's parameterizations that induce negative and positive precipitation changes, leading to drier (Q10) and wetter (Q11) conditions, respectively (see the supplementary material for further information).

These scenarios are downscaled, first, using the PRECIS regional modeling system (Jones et al 2004) and, then, using the Quantile Mapping statistical downscaling technique (Dèquè et al 2007). The corresponding projected streamflow trajectories in the Da, Thao, Lo, and Gam rivers are obtained by simulating the HBV models of the four sub-catchments forced with the projected trajectories of temperature and precipitation. The control scenario covers the period 1990–2010, while the future scenario spans the entire century. In the following analysis, we focus on two future time periods, namely 2040–2060 to assess medium-term impacts and 2078–2098 for long-term effects.

Figure 3 and table 2 compare the historical streamflow in the four rivers with the simulations over the horizon 2078–2098, as projected by the 5 scenarios. According to these projections, the future streamflow pattern will be characterized by a longer monsoon period (light gray area in figure 3) than the historical one (dark gray area in figure 3), with a general increase in the average annual flow, followed by a shorter but drier period as quantified by the low values of the 10th percentile (see table 2). Under the standard scenario Q0, we expect an average increase of the streamflow (i.e., +33.4% and +17.9% in the median and total annual flow) associated with high variability both in terms of dry and wet conditions (i.e., −29.1% in the 10th percentile and +14.7% in the 90th percentile). Coherently with their definitions, scenario Q10 is the one producing the lowest streamflow, resulting in the smallest average annual flow, while scenario Q11 is producing the highest streamflow peaks, captured by the highest 99th percentile. Finally, Q3 and Q13 show more variable streamflow patterns, with the former more similar to Q10 and the latter to Q0.

Zoom In Zoom Out Reset image size

The aim of this section is the inter-sector analysis of the system performance, measured in terms of hydropower production J^hyd, flood damages J^flood, and water supply deficit J^supply, achievable under historical hydroclimatic conditions by adopting alternative operations of the existing network of reservoirs. The performance of each solution is estimated via simulation of the integrated model of the Red River system over the historical horizon 1990–2010 using the observed streamflow trajectories. This analysis allows quantifying the space for a negotiated agreement representing a fair compromise between the conflicting interests involved. Figure 4(a) illustrates the performance of the designed Pareto optimal operating policies, where hydropower production and flood damages are plotted on the primary axes, while the water supply deficit is represented by the dimension of the circles. The black arrows indicate the directions of increasing preference, with the ideal solution (Utopia point) that would be represented by a small circle in the top-left corner of the figure, while each circle represents a different tradeoff between the three objectives.

Zoom In Zoom Out Reset image size

Results show that a clear conflict exists between the maximization of hydropower production and the prevention of flood damages. The best performance in terms of J^flood (solution PF in the bottom-left corner of the figure) is obtained by maintaining the reservoirs' level as low as possible to buffer the monsoon peak of the inflows (see the solid lines in figure 4b). This strategy negatively impacts on the energy production due to the reduction in the water turbined and the losses in terms of hydraulic head during the dry season, when the reservoir is almost empty. On the contrary, given the dimensions of the reservoirs and the large power capacity installed, the best performance in terms of J^hyd (solution PH in the top-right corner of the figure) keeps the reservoirs at their maximum level and releases the full turbine capacity for the entire year (see the dashed lines in figure 4b). Beside reducing the buffering capacity during the monsoon season, this energy-driven operating policy increases the vulnerability of the water supply in the delta and attains very high values of water supply deficit (i.e., large circles). The conflict between flooding and water supply is instead weaker, because the drawdown of the reservoirs' level during the monsoon season is indirectly making available large water volumes in the Red River delta. However, the temporal difference between the peak of the monsoon and the water demand peak in February, associated to the submersion of the rice fields, requires an ad hoc operations of the reservoirs to attain the minimum values of J^supply (solution PS, associated to the dotted lines in figure 4b), ultimately degrading the performance in terms of J^hyd and J^flood.

Overall, when evaluated over the historical hydroclimatic conditions, the set of Pareto optimal operating policies illustrated in figure 4 provides a rich context for understanding the complex management tradeoffs and dynamics in the Red River system and represents a valuable tool for supporting the identification of candidate compromise solutions. A potentially fair compromise policy (CP) to implement for balancing the three objectives is the one identified by the green cross in figure 4. This solution has been selected according to the criterion of the minimum distance from the Utopia point (Eschenauer et al 1990), which identifies the absolute optima of the three objectives (i.e., a small circle in the top-left corner of the figure). The CP performance is equal to J^hyd = 59 (GWh day^-1), J^flood = 8634.7 (–), J^supply = 55.01 (m³ s^-1)², i.e., better than 49% of the solutions in the Pareto optimal set in terms of hydropower production, 59% in terms of flood damages, and 45% in terms of water supply, thus equitably sharing the benefits within the competing sectors.

The variation of the hydrologic regime due to the climate change scenarios illustrated in figure 3 is likely to alter the medium and long term performance of the history-based policies illustrated in figure 4 as well as their tradeoffs between the operating objectives. To assess the vulnerabilities under projected climate of these solutions, we re-evaluated them via simulation over the period 2078–2098 under the five scenarios described in section 2.3. Figure 5 shows that climate change is expected to produce a general degradation of the system performance in all the sectors. The major concern is associated to the projected increase in the flood damages, which are on the average increasing (the circles move to the right) from 35% to 520%. In fact, the duration of the monsoon is expected to increase, with higher streamflow in September and October (figure 3). The history-based policies, however, tend to keep the reservoirs' level high in this period because the flood risk is supposed to be low and, consequently, they fail in controlling floods in Hanoi when exposed to this non-stationary alteration of the hydrologic regime. Ensuring flood protection under the projected scenarios results to be particularly challenging due to the projected increase in the Thao River streamflow during the monsoon season (figure 3), which is critical as this river is not regulated and directly contributes to the floods in Hanoi. Further expanding the flood buffering capacity of the system by damming also the Thao River might be an adaptation option to consider.

Zoom In Zoom Out Reset image size

Our results also show negative climate change impacts on water supply vulnerability in all the scenarios (colored circles are larger than the black ones), with the values of J^supply that increase from 15% to 160% due to the expected decrease of the streamflow during the dry season. Given the importance of the Red River delta in terms of rice production, this climate change impact may represent a significant challenge for the future national food security, potentially forcing rice import from China and Thailand. Finally, the smaller variability in the average water availability (see the average annual flow in table 2) produces less pronounced impacts on the hydropower production, with variations in the values of J^hyd from −7% to +5%. Although the two drier scenarios (i.e., Q10 and Q3) induce modest reductions in hydropower production, the other three scenarios (i.e., Q0, Q11, and Q13) allow attaining even better performance in this objective. Yet, we expect increasing climate change impacts on hydropower when an energy market will be introduced in Vietnam (Nguyen 2007). In fact, depending on the energy price patterns, some periods of the year will be more valuable than others in terms of maximizing hydropower revenue, possibly inducing relevant losses if the energy price was high in winter, which is projected to be drier than now.

The analysis of the system performance illustrated in figure 5 allows exploring the full range of potential benefits and risk induced by the changing climate. This contributes in understanding how the uncertainty in the climate scenarios is transferred, and potentially amplified, when evaluated in terms of the operating objectives affected by alternative storage operations, ultimately altering the equity and fairness of the solutions. Under the baseline scenario (Q0), the performance of the candidate Compromise Policy identified in figure 4 is projected to degrade from +170% to +320% in terms of flood damages, and to change in water supply deficit and hydropower production from −63% to +3680% and from −3% to +7%, respectively, depending on the future operating policy adopted. The predicted performance changes under Q10 are more negative for hydropower production (i.e., from −16% to −3%) and water supply deficit (i.e., from −52% to +4100%), with less pronounced impacts on flood damages (i.e., from +20% to +177%). On the contrary, Q11 suggests that the long-term degradation of performance in terms of flood damages with respect to CP can vary from +337% to +1138%, which is 10 times greater than under scenario Q10. Finally, the impacts predicted by Q3 and Q13 are similar to the ones under Q10 and Q0, respectively.

The results in figure 5 also shows that the variability associated to the five scenarios affects the evolution of the system tradeoffs by modifying the shape of the tradeoff curves. The conflict between J^hyd and J^flood against J^supply is exacerbated under the projected climate, with most of the colored circles in figure 5 that are larger than the ones under historical conditions. Moreover, the clear tradeoff between hydropower and flooding illustrated in figure 4 maintains the same shape only under Q10, though translated toward the bottom-right part of the objective space. This tradeoff curve becomes instead more vertical under Q0, Q13, and Q3, meaning that a small improvement in J^flood induces a large loss in J^hyd. Under Q11, it instead evolves toward a horizontal slope, where a small increase in hydropower production produces large flood damages. This evolution of the tradeoffs towards more pronounced conflicts, combined with the uncertainty in the long-term policy performance, hampers the identification of negotiated compromise solutions.

The results discussed so far refer to a business-as-usual situation where, despite the evident changes in the hydroclimatic conditions, no adaptation measures are implemented. More realistically, the operations of the system will be adapted to the new conditions. To better understand the system vulnerabilities at different future time horizons and to assess the potential adaptive capacity of the system achievable by modifying the operations of the three reservoirs, we analyze the policy performance attained under the assumption of full adaptation to the projected future, i.e. we assume that the system operator recognizes the change and optimally re-operates the system based on the new hydrological conditions. The analysis is focused on scenarios Q10 and Q11, as these latter are the ones showing the largest impacts on the performance of the baseline operating policies.

Figure 6 contrasts the performance of the history-based solutions over the period 1990–2010 (black circles) first with their re-evaluation in the medium term (i.e., 2040–2060, shown in grey) and long term (i.e., 2078–2098, shown in orange and blue for Q10 and Q11 scenarios, respectively). In addition, the figure shows the performance of fully adapted operating policies over the period 2078–2098 (red circles). Again, the green crosses identify the performance of the candidate Compromise Policy, selected according to the criterion of the minimum distance from the Utopia point in the history-based Pareto optimal set evaluated over the historical period (CP_h), along with its re-evaluation in the medium (CP_Q10M, CP_Q11M) and long (CP_Q10L, CP_Q11L) term. Using the same criterion, we also selected an Adapted Compromise Policy (ACP_Q10, ACP_Q11) within the fully adapted Pareto optimal set.

Zoom In Zoom Out Reset image size

Results suggest that the impacts on the policy performance are characterized by a non-stationary and non-monotonic trend. Over the period 2040–2060 under Q10 (figure 6a), we observe significant reductions in hydropower production and larger water supply deficit, while lower flood damages. The difference in the performance of CP_Q10M with respect to CP_h (measured in terms of Δ J^hyd,Δ J^flood , and Δ J^supply ) are indeed equal to −12%, −28%, and +280%, respectively. On the long term, characterized by a relative increase in average annual water availability combined with large streamflow reductions in the dry period (see table 2), the history-based operating policy amplifies climate change impacts and produces definitely larger flood damages and larger water supply deficit (+71% and +153% under CP_Q10L with respect to CP_h), along with a 7% reduction in hydropower production. The adaptation of the operating policies to the changed climate can contribute in reducing these negative impacts. Results in figure 6(a) show that the performance potentially achievable over the period 2078–2098 under full adaptation (red circles) allows bouncing back to adequate levels of performance in all the three objectives. The degradation of performance adopting the compromise policy ACP_Q10 with respect to the historical performance of CP_h is indeed reduced to -5% in terms of hydropower production, +17% in flood damages, and +9% in water supply deficit, corresponding to a difference in the same objectives relative to policy CP_Q10L equal to +2%, -31%, and -57%.

Similar results are obtained on the Q11 scenario (figure 6b), with the non-stationary and non-monotonic climate trend associated to decreasing hydropower production and increasing water supply deficit on the medium term, followed by an average long term increase of 20% in the annual streamflow, combined with a +34.9% in the peaks, which is projected to produce huge flood damages if no adaptation options are implemented. Under this wet scenario, the projected impacts on the performance of CP_Q11M and CP_Q11L with respect to CP_h are equal to −7%, −12%, +112% in the medium term and +2%, +508%, +25% in the long term. However, results show that the performance of the Pareto optimal policies assuming full adaptation successfully exploit this additional water availability to attain better performance in hydropower (i.e., the maximum production increases from 60.62 GWh day^-1 in the historical set to 62.73 GWh day^-1 in the fully adapted set) and water supply deficit (i.e., the minimum deficit decreases from 18.55 (m³ s^-1)² to 17.64 (m³ s^-1)²). Relatively to CP_Q11L, the fully adapted ACP_Q11 attains +3% in hydropower production and -60% in flood damages, with a small increase in the water supply deficit (+4%). Overall, these results demonstrate that the modification of the reservoirs operations represents an effective adaptation measure for reducing the adverse climate change impacts.

Figure 7 illustrates how the benefits discussed in the previous section can be obtained by relatively simple adaptation of the baseline reservoirs operations to the changed climate conditions. The adaptation to Q10 consists in larger drawdown of Son La and Hoa Binh during the winter period, followed by the possibility of maintaining a higher average storage during the monsoon season, which is predicted to be drier than historically, thus allowing higher hydropower production. The adaptation on Tuyen Quang is more evident and suggests maintaining a higher storage during the first months of the year to better ensure water supply to the Red River delta, while keeping a larger flood pool from August to November to buffer the longer monsoon season, which is particularly evident in the projected streamflow of the Gam River (figure 3).

Zoom In Zoom Out Reset image size

In the case of Q11, effective adaptation is obtained by balancing the storages in Son La and Hoa Binh in order to improve the hydropower production by increasing the hydraulic head at the Son La power plant without degrading flood control. This strategy is acceptable in terms of flood control because, despite Q11 is the wettest scenario, the projected streamflows in the Da River consists in a longer monsoon season with a reduced peak. The fall peaks predicted both in the Lo and the Gam rivers suggest, instead, to reduce the storage in the Tuyen Quang reservoir to increase the flood protection with respect to the historical operations.