Journal of Water Resources Planning and Management

This study presents different operation techniques for real-time reservoir regulation on the basis of two hydrologic models and two optimization methods: the first-order autoregressive (AR) model, the GR3 conceptual rainfall-runoff model, the stretched-thread (ST) method, and dynamic programming (DP). With these elements, three reservoir-operation techniques can be designed by combining one hydrological model with one optimization method, namely GR3 with ST, AR with ST, and AR with DP (stochastic DP). The last possibility (GR3 + DP) is not computable. From the efficiency viewpoint, the techniques for a daily reservoir regulation are compared using a 3-yr recorded data series and then a generated 100-yr data series. The comparisons show surprisingly a favorable efficiency for the technique based on the ST method in the Bar-sur-Seine reservoir upstream from Paris. The study confirms the value of simple optimization methods such as ST and the applicability of scenarios methods in real-time reservoir operation. It emphasizes that the basis of a good reservoir-operation system is to view forecast and decision making as a whole unit.

Water quality as well as quantity can be important in determining water-supply reliability. Saline ground water emissions in the upper watersheds of the Brazos and other major river basins in Texas and neighboring states contribute large salt loads to the rivers. Water management and use is severely constrained by the natural salt pollution. A modeling and analysis approach, using a generalized river/reservoir-system simulation model called RESSALT, was applied in an evaluation of the water-supply capabilities of a system of 12 reservoirs in the Brazos River Basin. The general evaluation methodology is also applicable to other river basins where salinity is of concern. Simulations with the RESSALT model combine a specified water-use scenario and associated system operating rules with sequences of adjusted historical monthly stream flows, evaporation rates, and salt loads for a simulation period of many years. Alternative water-use scenarios represent water use during a specified past year, projected future use, and hypothetical yields. Relationships between water demands, allowable salt concentrations, and reliabilities are developed for alternative modeling assumptions and management strategies including multiple-reservoir release policies and impoundment of runoff from primary salt-source areas.

A new method for calculation of frequency-equivalent storms and hydrographs is presented. This procedure is an extension of the frequency-equivalent nonlinear hydrograph (FENLH) methodology. Streamflow data records for a total of 129 basins in the state of Kansas have been used. A new storm criterion for generating frequency-equivalent storms was selected. Peak flow rates, hydrograph volumes, and peak reduction factors, produced by the storm criterion, were related to the basin physiographic and climatic data. The combination of index flood and multiple regression techniques were used to obtain predictive equations for peak flow rates and hydrograph volumes for six different return intervals. The use of the FENLH model as a prototype and the selection of the criterion utilized in this study to determine frequency-equivalent conditions, provided storm dimensions that appeared realistic. The accuracy of the results obtained indicated that the methodology presented in this study can be applied to other areas.

An evaluation of time-series of streamflow at 166 basins in the northeastern United States reveals that this region is remarkably homogeneous in terms of the year-to-year variability and persistence of average annual streamflow. Goodness-of-fit tests reveal that the 166 time-series of annual streamflow tested were well approximated by a lag-one normally distributed autoregressive process with a fixed lag-one correlation and a fixed coefficient of variation. Computer experiments revealed that the observed variability about these fixed values could easily arise from sampling alone, due to the varied and short records available. Recent research on the behavior of water supply systems reveals that simple analytic models can describe the general relationships among reservoir storage, yield, reliability, and resilience for systems dominated by overyear storage requirements. An analytic storage model is combined with the regional model of annual streamflow, resulting in general relations among storage, reliability, yield, and resilience useful for water supply systems in the Northeast. An example using the water supply systems for New York City, Boston, Providence, Rhode Island, and Springfield, Massachusetts, documents how the proposed methodology may be used to compare the behavior of four very different systems.

A personal computer (PC) based decision-support system (DSS) can provide real-time information for improved flood prevention and control in the Trinity River basin in Texas. The DSS components (1) retrieve, process, and file rainfall and streamflow data; (2) estimate basin average rainfall, update model parameters, and forecast runoff; and (3) simulate reservoir operation in order to forecast regulated flows basinwide. The DSS uses custom software for retrieving and processing the data and an industry-standard database-management system (DBMS) for filing the data. It uses specialized versions of the well-known programs HEC-1 and HEC-5 for forecasting runoff and simulating reservoir operation, respectively. The DSS integrates these components with a specialized program manager. Although not yet used as designed, potential improved flood prevention and control with the DSS are demonstrated with historical storm data.

A new type of estuary-management model based on discrete-time stochastic linear quadratic optimal control is presented. It is a feedback-control model that enables decision makers to determine the upstream reservoir releases during a time interval after the salinity and nutrient levels are observed at specified locations in the estuary at the beginning of the time interval. The optimal upstream reservoir releases are determined so that the salinity and nutrient levels at these locations are as close as possible to the prescribed levels for the remaining time intervals in the sense of statistical expectation. The ungauged inflows, precipitation, and evaporation are incorporated into the model as random variables. The control vector for the estuarine system consists of the freshwater inflows into the estuary, and the state vector contains the salinity and nutrient levels at specified locations for measurement in the estuary. The dynamic-programming principle is used to analytically derive the feedback-control law that expresses the control vector as a linear function of the state vector. The parameter matrix in the system equation is recursively updated by recursive least squares. Numerical examples are performed for the Lavaca-Tres Palacios Estuary in Texas for the purposes of illustrating the viability of this methodology.

In recent years, considerable progress has been made in the development of fast network flow algorithms aiming at solving linear programming problems with network substructures. Recognizing the fact that the basic structure of a water-supply system is a network, network algorithms have been used for system operation and management. However, to date, most algorithms are designed to solve transshipment problems in a pure network setting with total demand being equal to total supply. The non–network-type constraints and variables are precluded from the network models. Consequently, network models are used to perform optimization only for the network portion of the water-supply system under certain overall operational guidelines. In this study, an algorithm, EMNET, is introduced for solving the regional-water-supply-system optimization that corresponds to a generalized network problem with additional non–network-type constraints and non–network-type variables. The multiperiod, multiobjective optimization model for the regional water-supply system of the Metropolitan Water District of Southern California is used for case study. The results show that, depending on the network substructure in the model, EMNET is 11–117 times faster than standard linear programming codes, such as MINOS.