Water science finds itself at an interesting and critical crossroads. Sophisticated atmospheric modeling, remote sensing, and Internet‐based exchange of data enable exciting new synergies to develop among scientists, policy‐makers, and the private sector. Paradoxically we find it evermore difficult to validate products from these high‐technology tools and to exploit their full potential due to a severe and sustained decline in available hydrologic data sets.

In order to limit the probability that a water resource system will fail when it is operated in accordance with its optimum operating policy the latter must be derived subject to relevant chance constraints. When a stochastic dynamic program is used, certain of these constraints can be handled in the same manner as deterministic constraints, whereas others can be applied indirectly by imposing a penalty for failure, the optimum value of which can be found by an iterative search. This search and a precise evaluation of the response of the system may be carried out within the dynamic program by means of minor additions to its basic algorithm. The application of this procedure to four systems demonstrates its power and illustrates the manner in which systems respond to the imposition of constraints on failure.

Stochastic dynamic programing can be used to derive optimum operating policies that maximize the expected net dollar benefits for a water resource system. It is shown that these policies may give rise to significant probabilities of system failure. Such probabilities can be estimated by system simulation. A procedure is proposed in which the combined use of dynamic programing and simulation allows an optimum operating policy to be derived that will maximize the expected net benefit and yet not violate constraints on the probability of failure. Limitations to failure are applied directly by imposing a fixed penalty on the system whenever it fails, thus inducing amendments in the optimum policies derived. Further developments of the technique are foreshadowed.

No abstract