Warning Systems in Risk Management

A method is presented here that allows probabilistic evaluation and optimization of warning systems, and comparison of their performance and cost-effectiveness with those of other means of risk management. The model includes an assessment of the signals, and of human response, given the memory that people have kept of the quality of previous alerts. The trade-off between the rate of false alerts and the length of the lead time is studied to account for the long-term effects of "crying wolf" and the effectiveness of emergency actions. An explicit formulation of the system's benefits, including inputs from a signal model, a response model, and a consequence model, is given to allow optimization of the warning threshold and of the system's sensitivity.