Figure 7
From: The Limits of Earthquake Early Warning Accuracy and Best Alerting Strategy
Alert outcomes as a function of different alerting thresholds. For our four damage thresholds of (from top to bottom) 2%g, 5%g, 10%g, and 20%g in Fig. 6, we categorize each earthquake into correct, missed, and false alerts (following the definitions in Fig. 1C) for different choices of alerting threshold shown on the top x-axis. In Figs 4–6, S1–S2, we assumed the alerting threshold was equal to the damage threshold, or equivalently that the probability of exceeding the damage threshold was 0.5. In this figure, we have relaxed that assumption, with the bottom x-scale defining the alerting threshold in terms of the probability of exceeding the damage threshold. The top x-scale defines the alert threshold as the equivalent minimum expected median ground motion that would trigger an alert. We plot the histograms in units of how many occurrences of each outcome is expected per decade given typical California seismicity rates. The blue line shows the number of times per decade that the observed ground motion exceeded the damage threshold; these events can only result in missed or caught alerts, although the relative proportion of each changes with the alerting threshold. If the user only wishes to take action when there is a high probability that the ground motion will exceed the damage threshold (right-most bars in each subplot), then there will be few false alerts but most damaging earthquakes will be missed. If instead, the user wishes to make sure that an alert is always received before damaging ground motion is experienced, the alerting probability can be set low (left-most bars in each subplot). This will result in few missed alerts at the cost of many false alerts since most of the earthquakes for which the user takes action will not be damaging. Letters a, b, and c in (D) indicate examples discussed in text.
