OCR Text |
Show the extra pumpage was determined relative to the required lift at neighboring nodes, progressively moving away from the well of interest. The solutions were rather interesting and to some extent counter- intuitive. No agencies pumped in excess of their demand, once the water rights constraints were relaxed. Contrary to expectation, the pumping pattern did not change with the new specification of lifts as a function of well demand. The well nodes located in the lower areas of the valley, where aquifer conditions are close to artesian, persisted as the most economic for supply. Retrospection revealed that this was logical. The aquifer properties ( e. g. Transmissivity) are good in this area, the existing drawdown is small, aquifer heads are close to the surface, and relative to surface demand points at higher elevations, the lift at these points is comparable to that at the well nodes adjacent to the higher demand points. A higher degree of confidence may thus be associated with the solutions in chapters 3 and 4. A final note with respect to these results is that they are based only on a set of 47 well nodes selected for optimization. If a finer discretization of the aquifer, and a greater number of candidate well nodes are selected, the improvement of the optimal solutions over the existing groundwater yields is likely to be even more dramatic. Computationally, the transient groundwater management model formulated is quite demanding. The formulation presented and its implementation, compares very well in efficiency relative to those presented in the literature. However, the model solution times and costs on a super- minicomputer preclude a large set of parametric applications with a monthly or seasonal time scale. The unit response formulation leads to a fairly dense, but structured constraint coefficient matrix. Decomposition of the problem mathematically may be feasible and should be pursued. Of greater interest, given the advent of parallel processing architecture in computing machines, is the development of specialized optimization algorithms that would focus on a high degree of parallel computing and vectorization techniques to solve the problem. The mathematical structure of the unit response formulation appears to lend itself very well to such a task. Similar comments apply to the solution of the surface water and conjunctive use models presented. A current difficulty with such a development is a lack of ready access to parallel computing hardware, and a lack of widespread understanding in the Civil Engineering field of the algorithmic structures needed for efficient and reliable parallel computing. 8.2 Surface Water Models Given the continuing controversies in Utah regarding the development of reservoirs versus exploitation of groundwater, an objective of this work was to construct an efficient model for screening candidate reservoir sites for development. Multiple yields, yield reliability, consideration of hydropower generation, flood control and recreation, and the embodiment of a high degree of detail as to the physical, hydrologic and water rights peculiarities of the river basin was considered essential. An initial formulation of such a model was developed with the support of the Utah Division of Water Resources in conjunction with the efforts of this study. The Bear River applications described in chapter 5 were part of this effort. It is the contention of this author that the reservoir screening model postulated in chapter 5, and its extensions in chapters 6 and 7 are rather unique and innovative, relative to the recent systems analytic literature on the subject. Classical linear programming and dynamic programming formulations are similar to the water supply formulation for conjunctive use of surface and groundwater presented in chapter 2. They typically suffer from the curse of dimensionality - rapid growth in mathematical problem size as the number of sites and/ or the length of record increases. The practicality of applying such models to even a moderate set of candidate sites is dubious. It was noted that the rapid growth in problem size in these models stems largely from the month by month mass balance, and flow preservation constraints. Further, with the large problem 178 |