The very first reservoir used by humankind must have been controversial. How much water will the reservoir provide? How much water can be used now, and how much must be saved for later use?
Real-time data networks reporting meteorologic data, snow conditions, streamflows and reservoir levels, combined with hydrologic simulation modeling, provide new tools for answering these questions. The figures and examples of watershed and reservoir behavior in these notes were created by the Seattle Forecast/Analysis model (Seafm), a system developed by Hydrocomp for the Seattle Water Department. Seafm has been used by Seattle Water since 1991.
Now, thirty years after the computer revolution, and approaching the millennium, the information that can be at hand for reservoirs includes Figure 1, the Deficit (2) Exceedance Probability in the current drawdown period, Figure 2, the near term future Flood Frequency, and Figure 3, the future Reservoir Inflow Exceedance Probability. These figures, and similar figures for hydro power production, snowpacks,aquifers, and water deliveries are created interactively on demand on graphical workstations by continuous simulation of hydrologic processes. (3) They show future conditions in reservoirs given current reservoir conditions and scheduled reservoir releases, and they can be created every week, or every day, as conditions change in a reservoir and watershed. They are the most recent addition to the centuries old technology for building and operating reservoirs.
The analysis that creates Figures 1 to 3 is detailed modeling of infiltration, actual evapotranspiration, surface runoff, interflow, snow accumulation and melt, etc. (Figure 4).
The data required for continuous simulation modeling are watershed characteristics (represented by model parameters), watershed state (the current snowpack, soil moisture, and aquifer levels in the watershed), and meteorologic records (Figure 5).
Model parameters for watershed characteristics are found from physical data (vegetation and soils maps, topographic data and field measurements), and by calibrating the modeling system to reproduce recorded historic streamflows that are tributary to the reservoir. Watershed state or "initial condition" is observed (e. g. reservoir levels), and is continuously calculated. Watershed variables that are not easily observed are found by modeling, by running the modeling system from a point a year or more in the past up to the current time.
The meteorologic records needed for modeling are provided by alternate futures. The weather ten days in the future often bears little relation to ten day meteorologic forecasts. (4) Historic weather provides samples of alternate futures, weather that might occur at the current time of year. The alternate futures are used in continuous modeling to create equally likely time series of streamflows, reservoir levels, power production, etc. These time series are summarized statistically to create Figures 1, 2, 3, etc.
If we have historic recorded streamflows, or reservoir inflow records, why bother creating "alternate future" flow records from meteorologic data? Future streamflows are a function of future weather and the current watershed state. Historical streamflows result from a full spectrum of watershed states. They are not aware of, and are not conditioned by the current watershed state, and they cannot be used for near term future steamflows. The effects of watershed state are illustrated by Figures 6 and 7.
Figure 6 shows exceedance probability for future
reservoir inflows into
Modeling results for future reservoir inflows, flood frequency, power production, etc. can be used directly for decisions on reservoir operations, or they can be used as data for formal optimization methods.
Flood control operations in reservoirs should logically be based on the near term flood frequency (Figure 2), which is dependent on time of year and on the current watershed state. To account for near term flood frequency, flood control space in reservoirs is often made seasonally dependent, but seasonal dependence is a partial solution. When near term flood frequency is available, operators can plan for the range of peak flows or runoff volumes that will occur. Flood control operations that anticipate a higher range of peak flows or a lower range of peak flows than will occur forego opportunities to reduce flood damages. (7)
Future deficit probability (Figure 1) informs reservoir operations. Benefits for reservoir design are equally pervasive, but are more subtle. If a reservoir is operated conservatively, to provide less than its historic "safe yield", (9) deficits are not considered. Water is always released to meet current demand. If a reservoir provides water in excess of its safe yield, risk in operations must be considered.
A high risk reservoir operation ignores deficits. A low risk operation uses future deficit probabilities to limit reservoir releases in some (not all) drawdown seasons (Figure 8). In a high risk operation deficits are infrequent but are very severe - water deliveries may fall to only a few percent of normal. In a low risk operation deficits are more frequent but are manageable - water deliveries may be 80 to 90 percent of normal. A typical relationship between deficit frequency and deficit severity is sketched in Figure 9.
Figure 10 is an extension of Figure 9. It may not be intuitive, so definitions of "risk" and "mean water deliveries" are needed. With Seafm's capabilities we can operate a reservoir to meet normal water deliveries when the risk of a future deficit remains below a selected risk criteria, but we will reduce water deliveries when the risk of a future deficit exceeds this criteria. Reducing water deliveries in a drawdown season reduces the mean of water deliveries for all historic drawdown seasons. If the criteria that triggers rationing is "low risk", then water might be rationed in almost all drawdown seasons. If the criteria that triggers rationing is "high risk", then water might be rationed only once or twice historically.