Estimating Loss-of-Life Due to Dam Failures: Pitfalls, Fallacies and Recommendations

There are a number of applications in which it is useful, if not necessary, to assess the potential for loss-of-life in the event of dam failure. Applications include classification of a dam's hazard class, probabilistic risk assessments, and emergency planning. In certain applications, such as probabilistic risk analyses for dams, it is necessary to estimate the actual number who may lose their life. It is intuitive that the problem of estimating the what fraction of the population-at-risk is inherently random due to such factors as time of day, dissemination of warnings to those downstream (with/without the benefit of emergency action plan). Such assessments are further complicated by the limited data that is available to construct models for making such predictions. As part of recent study, the authors were required to estimate the potential for loss-of-life in the event of a dam failure and to assess the uncertainties in such estimates. Based on this case study, a measure of the natural variability and the uncertainty in model predictions were quantified. The results were surprising in a number of respects. In this paper we present results for a number of case studies, lessons and insights from these applications.

In this paper we identify and demonstrate the difficulties and biases associated with empirical models that have been developed to estimate the loss-of-life in the event of dam failure. One model in-particular developed by DeKay & McClelland (D&M) (1993), which is an extension of earlier work by the U.S. Bureau of Redamation, is an empirical approach to estimate the fraction of population-at-risk that would lose their life in the event of a dam failure. On close examination of this model and the data on which it is based, a measure of the randomness in the potential for loss of life, and more importantly the uncertainty in the model predictions is developed. 14 pp., 7 references.