HOW DEEP UNCERTAINTIES LIMIT NATURAL HAZARD ASSESSMENTS

In Shakespeare's Henry IV, Glendower says "I can call spirits from the vasty deep," and Hotspur replies "Why, so can I, or so can any man; but will they come when you do call for them?"

Scientists assessing natural hazards face this issue. They can make detailed hazard assessments, but the earth often acts differently. Insight into why this occurs comes from a recent approach in risk analysis, which separates uncertainties in predicting future events into shallow versus deep uncertainties. Shallow uncertainties arise when the probabilities of outcomes are reasonably well known. In contrast, deep uncertainties arise when the probabilities of outcomes are poorly known, unknown, or unknowable. In such situations, past events may give little insight into future ones. We can develop models based on the past performance, but would place little confidence in them.

For example, earthquake hazard assessments involve multiple deep uncertainties. One is assessing earthquake recurrence in time. Many seismic hazard maps use time-independent models, whereas others favor models in which the probability of a large earthquake decreases after one occurs and then rises. Distinguishing between these from the available earthquake record is difficult or impossible. A second source of deep uncertainty involves where to expect earthquakes. On plate boundaries, we expect all parts of the boundary to slip eventually, although attempts to forecast the timing of major slip events have been unsuccessful. Within plates, however, it is hard to forecast where large earthquakes will occur. A third source of deep uncertainty is how large an earthquake to expect. We do not know whether the largest known earthquake i is the largest that happens there, or just the largest observed to date. In many areas, such as the Eastern US, simulations show that our short earthquake record is likely to have missed the largest earthquakes.

These deep uncertainties result from our limited understanding of how earthquakes vary in time, space, and size. Thus although improved data and models can improve hazard models, we probably face fundamental limits as to what we can say about future earthquakes. These uncertainties should be recognized, incorporated into hazard maps, and factored into formulating hazard policy that balances the costs and benefits of mitigation options.