When can we know how large an earthquake will be? Is the magnitude of an earthquake controlled by the conditions and dynamics at the start of its growth? If so, measurements of the initial seismic waves from an earthquake, and even of the area in which it will occur, could enable early warnings of ground shaking. If not, then the chances of such short-term prediction are low. Writing in Nature, Ide1 compares the onsets of thousands of large earthquakes around Japan with those of nearby small ones. He finds that the onsets of about 20% of large earthquakes are indistinguishable from those of closely located small ones, within the frequency range of the seismic waves that he analysed.
Earthquakes often begin with a short phase of small-amplitude waves only, before growing to the final-sized event2,3. One mechanism that could explain this observation is a cascading failure, in which changes in stress from one randomly failing patch of geological fault cause other patches to fail — like toppling dominoes. In this case, the magnitude of an earthquake is controlled by the dynamic conditions as the event grows, and is impossible to predict until the quake slows or stops.
An alternative possibility involves slow slip — the relative movement of the rocks on either side of a fault. This slip, which is undetectable by seismometers, could gradually accelerate in a limited region of the fault before attaining a critical speed and breaking out to reach the final quake size. If correct, the earthquake magnitude might be determined by the size of the region of preceding slow slip or by the characteristics of the initial waves; and if these properties could be observed and understood, short-term prediction might be possible.
Some studies of seismometer records have found that earthquake magnitude is independent of the first few hundredths of a second4 or longer3. However, these analyses were limited to only a few earthquakes. Other studies5,6 have suggested a dependence of the final quake size on the onset. But these analyses involved indirect parameterizations of the data, and might not accurately account for the energy loss of seismic waves as they travel through Earth7.
Ide compared the onsets of closely located earthquakes of different magnitudes to determine whether these onsets provide any indication of final quake size. He carried out a comprehensive analysis of all of the large earthquakes that were recorded in sufficient detail, between June 2002 and April 2018 along about 1,100 kilometres of the Japan Trench — a subduction zone, in which the Pacific Plate beneath the Pacific Ocean is being pushed under the Okhotsk Plate beneath Japan. He followed a procedure that has been used to identify phenomena known as repeating earthquakes8. These are similarly sized quakes whose seismometer records are so alike that the events must involve repeated, similar movement of the same patches of fault8.
Instead of searching for similarly sized events, Ide started with 1,654 large earthquakes (of magnitude greater than 4.5) and compared them with all of the known small events (of magnitude less than 4) that are positioned closely enough (within about 100 metres) for their locations to be indistinguishable. He then calculated the similarity between the first 0.2 seconds of the seismometer records of these quakes.
Ide discovered 390 pairs of large and small earthquakes that have highly similar beginnings, with the onsets of 200 of the large events being indistinguishable from those of approximately co-located small events. He interpreted this finding to indicate that the onsets of large earthquakes can be identical to those of small ones, and therefore that the initial conditions and dynamics of a quake do not determine its magnitude.
By separating the earthquakes into subduction-type events — similar to the 2011 Tohoku-Oki earthquake (Fig. 1) — and other types, Ide discovered that the subduction-type earthquakes are more likely than the others to have paired events. He also found that the subduction-type pairings can be separated by more than ten years, whereas those of the other types are limited to small earthquakes occurring close in time to the large event.
The author interpreted these results using a model in which an earthquake fault consists of patches that have a range of sizes and relatively constant rupture characteristics. Slip of one such patch might trigger slip of a larger neighbouring patch, and so on. This picture is consistent with numerical models that have been proposed to explain repeating earthquakes9. The similarity of onsets over extended time periods implies that a long-term characteristic structure is present and able to repeatedly host large and small quakes.
By considering co-located earthquakes, in which the seismic waves from both small and large events take the same paths to measuring stations, Ide eliminated bias from waves travelling different paths through Earth3,4. His conclusion that the onset of an earthquake does not control its final size agrees with detailed observations of well-recorded quakes on the San Andreas Fault in Parkfield, California10. It is also consistent with global statistical compilations of large earthquakes5,7,11 that found that all quakes grow at approximately the same rate and start to differ only when the rupture shows signs of slowing.
Most earthquakes around Japan occur offshore or deep underground, and so are not close to seismometers, limiting the spatial coverage and frequency range of recording. Ide’s analysis focuses on high-frequency waves (of 1 hertz and above). It therefore misses any onset differences at the lower frequencies at which most of the energy is released by large quakes. It would also miss any preceding slow slip such as that observed in laboratory experiments and numerical models12 — reliable, consistent observations of this slow slip before actual earthquakes remain elusive.
Another issue is that, even though Ide aimed to compare large and small earthquakes, about 60% have a magnitude difference of less than 1.5, similar to the variation seen in repeating-earthquake sequences8. Only about one-eighth of the paired events have a magnitude difference of more than 2.
For now, earthquake-prone populations must rely on earthquake early-warning systems, which have been long established in Asia and have been introduced in the past few years in California. These warning systems involve estimating earthquake magnitude using near-source seismometers, and sending this information to vulnerable populations ahead of the more slowly travelling, damaging seismic waves13. Ide’s results are a key step towards a better understanding of earthquake initiation — knowledge that could improve the speed and accuracy of these warnings.
Nature 573, 42-43 (2019)