Long-range earthquake forecasting based on a single predictor

Summary. For a long-term predictor from which a joint distribution of earthquake occurrence time and magnitude has been obtained, and also a record of past successes, false alarms and failures, Bayesian statistical methods yield predictive information of the kind needed as a basis for decision-making on precautionary measures. The information is presented in terms of risk refinement, intensity probability and success probability. After the event the relative likelihood that a prediction was a success or failure can be estimated. Comparisons can also be made of the performance of different forecasting models. The application of these methods is illustrated by an example based on the proposed swarm-magnitude predictor.