Testing earthquake forecasts using reliability diagrams



Reliability diagrams are a standard forecast evaluation tool for comparing forecasted probabilities of binary events with their observed frequencies. To make this comparison for rare events, forecasts of similar probability are grouped together. Here, we demonstrate the utility of reliability diagrams for earthquake forecast evaluation with an application to a modified version of the Holliday et al. time-dependent earthquake forecasting method. We quantify reliability with a χ2 statistic and find that the forecasting method is consistent with reliability; that is, forecast probabilities match rates of occurrence to within the expected uncertainties.