The Fault in Our Forecasts

Detail of a seismogram of the 1906 San Francisco earthquake.

Susan Hough at Asterisk: The reality is that earthquake prediction is hard. “Neither the USGS nor any other scientists have ever predicted a major earthquake,” reads the official statement on the United States Geological Survey’s online FAQ. “We do not know how, and we do not expect to know how any time in the foreseeable future.” Seismologists point to three basic reasons. First, seismology is not chemistry. The Earth itself runs the experiments on its own time. To truly test a proposed prediction method, it must be applied prospectively: predicting the earthquakes that haven’t happened yet. As terrifying as they are, damaging earthquakes do not occur that frequently in any one area — at least not on timescales relevant to individuals. Second, earthquake processes play out far underground, at depths of a few kilometers to hundreds of kilometers beneath the surface. Only a small minority of fault breaks even reach the surface, and even then, observing one after the fact tells us little if anything about the physical processes that govern earthquake nucleation.

It’s impossible to predict when an earthquake will strike. This puts seismologists in a nearly impossible bind: how can they convince the public to take earthquakes seriously without crying wolf?

Short-term weather prediction is possible because we can see weather systems brewing. There’s no way to observe what faults are doing many kilometers beneath our feet. Third, although it’s by no means proven, some evidence suggests that earthquake processes are fundamentally chaotic, the mechanics of earthquake nucleation governed by details we can’t hope to know well enough to predict future behavior of the system. Even if we could peer into the depths of fault zones, it’s possible that earthquakes would remain impossible to predict, just as avalanches remain unpredictable despite the fact that we can see the snow.

Let me stop here to make an important point. Earthquake professionals have made major strides in forecasting the long-term average rate of earthquakes in any given region. Unlike predictions, which involve meaningfully useful bounds in time, space, and magnitude, forecasts are made in terms of probabilities that an earthquake of a certain size will strike on decadal or longer timescales. For example, in 2008, seismologists predicted that there is an estimated 99.7% chance that at least one magnitude 6.7 or greater earthquake will occur in California over the next 30 years. Based on long-term odds, seismologists are known to tell the public that “it’s not a question of if, it’s a question of when” a damaging earthquake will strike.

More here.