Design of Length of Water Resource Simulation Experiments

The research procedure was to: (1)Explore the experimental nature of simulation; (2)outline criteria for making sampling decisions about the length of simulation experiments; and (3)develop test statistics for ranking alternative system designs. Lengths of simulation runs required to achieve desired levels of reliability are shown to depend on serial and cross-covariance properties of simulated net benefit time series. Application of the procedures suggested by this study to a set of simulations of alternative plans for the Rio Colorado, Argentina, shows that the longest runs are required for alternatives with largest variance of annual net benefits. Serial correlation tends to increase required run length whereas cross-correlation tends to reduce required run length. In need of further work are: procedures to rank many alternatives having correlated net benefit series; procedures for nonstationary series; procedures to deal with non-normal benefit series; and procedures to account for the consequences of making incorrect choices.