Journal of Water Resources Planning and Management

Real-time control of pumps can be an infeasible task in water distribution systems (WDSs) because the calculation to find the optimal pump speeds is resource intensive. The computational need cannot be lowered even with the capabilities of smart water networks when conventional optimization techniques are used. Deep reinforcement learning (DRL) is presented here as a controller of pumps in two WDSs. An agent based on a dueling deep q-network is trained to maintain the pump speeds based on instantaneous nodal pressure data. General optimization techniques (e.g., Nelder–Mead method, differential evolution) serve as baselines. The total efficiency achieved by the DRL agent compared to the best-performing baseline is above 0.98, whereas the speedup is around $2\times$ compared to that. The main contribution of the presented approach is that the agent can run the pumps in real time because it depends only on measurement data. If the WDS is replaced with a hydraulic simulation, the agent still outperforms conventional techniques in search speed.

A novel application of spline curves was developed for tracing average daily trends of water demand. These trends were used as input of a stochastic model to generate synthetic time series considering the number of water users as the main input parameter. Hermite polynomials were used for a piecewise interpolation of some known points of the daily trend which were obtained through reliable equations from the literature, whereas unknown points were deduced based on the mathematical properties of the demand pattern. Daily demand time series were generated for different time resolutions using a Monte Carlo approach based on a mixed probability distribution. Results were compared with real observed demand data to validate the effectiveness of the proposed approach, showing encouraging results.

Traditional economic forecasts had consistently underestimated the hypergrowth of South Florida and its need for freshwater from the 1970s through the 1990s. It was hypothesized that the continued rapid growth of urban and farm water use into the 2000–2030 period would undermine efforts to restore the Everglades. To test this hypothesis, the corresponding author had hybridized two widely used regional economic modeling tools: the static, single-period Impact Analysis for Planning (IMPLAN); and the dynamic multiperiod Regional Economic Modeling Inc. (REMI) to forecast population and water use for the period from 2010–2030. In the present paper, these early forecasts are compared to actual census and water use counts for 2010 and 2015. Some of these early forecasts are found to be surprisingly accurate while others seriously overshoot the mark. The on-target forecasts validate the basic hybrid model while the other “deviant” forecasts measure the success of the radical antidrought policies and the region’s economic reaction to the financial collapse. This retrospective provides a new metric for measuring actual changes in the paradigm of water consumption.

Urbanization is one of the most aggressive forms of land transformation, leading to negative impacts on surrounding aquatic and terrestrial ecosystems. In terms of hydrology, urbanization compromises the ability of the natural landscape to absorb, store, and slowly release water. Watershed management strategies that strategically utilize spatial patterns of natural land cover (e.g., wetlands or forests) have shown initial promise to mitigate flash flood events through the enhancement of natural hydrologic mechanisms. The main objective of this work is to (1) identify spatial configurations of land cover that provide flood regulation ecosystem services; and (2) understand the areal extent of flood regulation services throughout the Blue River Watershed, in the central US states of Kansas and Missouri. We aim to understand the mechanisms that influence the biophysical supply of flood regulation services using a measure of stream flashiness to quantify changes in flood regulation services and landscape metrics to characterize changes in spatial patterns of land cover. Our results show that a landscape configuration of disaggregated wetland patches and a large percentage of low-density grassland patches will increase the provision of flood regulation ecosystem services. These findings are beneficial for watershed managers and/or regional planners who aim to minimize flood impacts through urban development strategies.

Involvement in the operation of a multiobjective and multistakeholder hydropower system requires balancing conflicting objectives and coordinating economic gains to achieve a win–win situation through joint operation. This study developed integrated models of multiobjective joint optimization and synergistic revenue allocation of multistakeholders to resolve conflicts. The multiobjective operation model coupled with the technique for order preference by similarity to ideal solution method determines a noninferior solution for compromising social (water delivery and flood control), and economic (energy production) objectives with the incorporation of the decision makers’ preference. Thereafter, a novel synergistic revenue allocation method based on the total differential equation (SRA-TD) is proposed. This method is used to systematically decompose the contribution of each stakeholder (reservoir) in the synergistic revenue through cooperation, thereby satisfying the principles of efficiency, rationality, and equity. The integrated models are applied to a hypothetical multistakeholder cascade hydropower system. Results are as follows: (1) The compromise solution selected from the multiobjective optimization operation results increases the firm water supply but reduces the energy production revenue without increasing flood storage compared with the benchmark solution (status quo). (2) The SRA-TD method decomposes the synergistic revenue contribution of each reservoir during each period and verifies its physical generation mechanism, thereby satisfying the basic principles of synergistic revenue allocation. (3) Last, the SRA-TD method comprehensively considers the dynamic effects of inflow conditions, reservoir characteristics, and topological structures of a reservoir system. The proposed method provides a viable alternative for promoting joint optimal operations in a multiobjective and multistakeholder cascade reservoir system.

The Everglades is a large, complex, and highly managed ecosystem, and its natural hydrologic properties, water quality, soils, flora, and fauna have been significantly altered by the Central and Southern Florida Project. The multi-billion-dollar Comprehensive Everglades Restoration Plan (CERP) was authorized in 2000 to restore and protect the Everglades ecosystem. However, the progress in implementing CERP projects has been slow because of their size, complexity, and long-term nature. In 2016, a review of CERP recommended updating it, and in 2018, a mid-course assessment was recommended to incorporate new scientific knowledge. This study began a reassessment of the CERP by evaluating the relative importance of a range of factors, including operations, water storage, removal of barriers to water flow, and seepage control. The purpose was to identify those elements most affecting hydrologic and ecological responses in different regions of the Everglades. The South Florida Water Management Model, a regional hydrologic model, was used as a primary tool to simulate the restoration scenarios over a 36-year period (1965–2000). All scenarios provided substantial improvements to the hydrologic system when compared to the existing conditions. However, restoration benefits of these scenarios were unequally distributed across the Everglades landscape. The scenario that expanded the surface storage in the Everglades Agricultural Area (EAA) to $2.5\mathrm{billion}{\mathrm{m}}^3$ reduced flood control discharges to the northern estuaries by 66% relative to the existing conditions, which was 8% higher than CERP. However, the scenario with CERP projects significantly increased flows to Everglades National Park. The findings of this study are relevant to Everglades restoration especially in light of recent and emerging science on storage feasibility and climate change.

Achieving a water, food, and energy (WFE) nexus balance through policy interventions is challenging in a transboundary river basin because of the dynamic nature and intersectoral complexity that may cross borders. The Nile basin is shared by a number of riparian countries and is currently experiencing rapid population and economic growth. This has sparked new developments to meet the growing water, food, and energy demands, alleviate poverty, and improve the livelihood in the basin. Such developments could result in basinwide cooperation or trigger conflicts among the riparian countries. A system dynamics model was developed for the entire Nile basin and integrated with the food and energy sectors in Egypt to investigate the future of the WFE nexus with and without the Grand Ethiopian Renaissance Dam (GERD) during filling and subsequent operation using basinwide stochastically generated flows. Different filling rates from 10% to 100% of the average monthly flow are considered during the filling process. Results suggest that the GERD filling and operation would affect the WFE nexus in Egypt, with the impact likely to be significant if the filling process occurred during a dry period. Food production from irrigated agriculture would be reduced by 9%–19% during filling and by about 4% during GERD operation compared with the case without it. The irrigation water supply and hydropower generation in Sudan will be reduced during the filling phase of the GERD, but this is expected to be improved during the dam operation phase as a result of the regulation afforded by the GERD. Ethiopian hydropower generation is expected to be boosted by the GERD during the filling and operation of the dam, adding an average of $\mathrm{15,000}\mathrm{GWh}/\mathrm{year}$ once GERD comes online. Lastly, the results reveal the urgency of cooperation and coordination among the riparian countries to minimize the regional risks and maximize the regional rewards associated with the GERD.

Water distribution networks (WDNs), as vital systems of any city, are vulnerable to the intrusion of contamination, and the security of WDNs is of great importance. A promising approach for the protection of WDNs against such threats is to deploy contaminant detection sensors in the network. One of the modeling strategies to identify an appropriate configuration of sensors is the expectation-based model. In this model, the location of sensors is determined so that the expected impact of contamination events is minimized. In this work, we consider two existing expectation-based models that have been presented in the literature. Although both models identified the same optimal sensor placements, choosing the better model in practical implementation may not be clear to practitioners, who must consider complexity and resolution times. This work first answers this question and proves that both models have the same quality—and hence, in terms of solution time, it makes no difference which model is used. As the second contribution of this work, the expectation-based model is extended to incorporate worst-case, value-at-risk (VaR), and conditional VaR (CVaR) measures. Computational results compare the damage of risk-based models with real-world WDNs, and indicate that the CVaR-based model may be an excellent approach to address risk measures in this problem. The CVaR-based model optimizes the CVaR measure and, at the same time, does not cause a significant increase in the optimal value of other risk measures.

Proportional-integral (PI) control, as one of the most popular classic control methods, has been applied widely to the real-world practice of canal automatic control. The performance of a PI controller largely depends on two key parameters, namely the proportional constant $K_p$ and the integrational time constant $T_i$. Rather than tuning these parameters empirically or in terms of the canal morphology, this study proposes a linear quadratic regulator (LQR) to determine their optimal values. The proposed LQR utilizes an integrator delay model to represent the hydrodynamics of open canals in order to minimize changes in water levels and flow rates. In addition, the weights for the optimization objective in the LQR are determined by an optimized quadratic performance indicators estimate (OQPIE), using the precalculated nondimensional integrated square of error and nondimensional integrated absolute discharge change as well as inherent designed parameters, which potentially impact the stability of system states. In this way, the LQR can fit various canal automation applications, especially for low-gradient canals. The optimal PI controller was tested on two different-scaled canals. Results showed that the objective was met satisfactorily, and stability can be reached in hours.

The configuration of a burst detection network is fundamental to safely monitoring a water distribution network (WDN). A pipe burst event can be detected only when the effects of the burst are distinguishable from background noise in the monitoring system. A new method of optimizing pressure sensor placement is proposed to monitor pipes in cases in which burst events are likely to have severe effects. An objective function for monitoring water leakage was defined. A genetic algorithm was used to find the optimal sensor placement in the pipe monitoring network and the most economical quantity of sensors. The method was illustrated with three case studies. In each case, the optimal solution evenly distributed the sensors across the WDN, and the number of sensors was determined based on the marginal gain of an additional sensor. In the optimal configuration, monitored pipes were identified clearly.

Uncertainty inherent in precipitation predictions from general circulation model (GCMs) may lead urban drainage systems to be underdesigned (or overdesigned) in the future. This issue can be mitigated with the use of risk analysis models. In this study, a decision-making tool, developed based on six models (minimin, minimax, expected value, Hurwicz, Savage, and scenario-based multiobjective robust optimization), was used to select GCM/representative concentration pathways (RCP) scenarios that would lead to robust designs of an urban drainage system located in Fortaleza, Brazil. The implementation costs of the studied drainage system were estimated using runoff derived from rainfall predictions from six GCMs and two RCPs. After applying the proposed decision-making tool, three GCM/RCP scenarios were selected for yielding the most resilient and reliable designs. The range of feasible GCM/RCP scenarios reflects the level of optimism or pessimism held by a decision maker. We strongly recommend that this method be incorporated in urban drainage system design in order to help municipal planners make better decisions in view of climate change uncertainty.

This study presents a methodology for optimizing the unit commitment (UC) of the Xiluodu and Xiangjiaba cascaded hydropower plants that rank as the second- and third-largest hydropower plants in China. This UC problem exhibits high complexity due to the integrated consideration of hydraulic coupling between plants, head-sensitive forbidden zones, and other hydropower system nonlinearities. Hydraulic coupling is considered through a group of nonlinear tailrace level curves that are represented using binary 0–1 variables. The interval segmentation is used to determine the binary value by quickly identifying the forebay level range of the immediate downstream plant. The irregular forbidden zones are divided into several quadrangles to approximately describe the upper and lower boundaries using linear functions of the net head. The described forbidden zone boundaries are further integrated with power generation limitations to determine discontinuous safe operating zones. With the aid of polynomial approximations of other nonlinearities, the presented problem is finally formulated as a mixed-integer nonlinear programming (MINLP) model and addressed by an optimization method based on the branch-and-bound technique. The practical behavior of the proposed methodology is demonstrated by real-world data. Moreover, the MINLP model is compared with a commonly used mixed-integer linear programming model and shows improvements in solution accuracy and computational efficiency.

The value of multiple transient tests under the same experimental conditions for the purpose of leakage detection is assessed. The gain in signal-to-noise ratio and leakage detection accuracy due to the multiple tests are derived theoretically and evaluated experimentally. The role of the multiple measurements in suppressing interferences and thus minimizing false detection is proven using experimental data. A theoretical extension of existing leakage detection method that was derived for single measurements is tested. From a practical perspective, this research quantitatively shows that the multitest strategy allows detecting smaller leaks in noisier environments, and/or using smaller-amplitude waves.