Why do the largest continental earthquakes nucleate on branch faults?

Nearly every earth scientist would agree with these three statements: Earthquakes occur on faults, large earthquakes occur on large faults, and large earthquakes start on large faults. But in recent work, we find that when it comes to the largest continental earthquakes, that last statement is untrue. And reconciling why it's untrue may resolve a longstanding riddle about rupture of these massive - but surprisingly weak and moveable - faults that lace the continents and have caused terrible human losses over the past century. The San Andreas in California, the North and East Anatolian faults in Turkiye, the Alpine Fault in New Zealand, and the Philippine Fault in the Philippines are some of the best known of these "continental transform" faults.

In a pair of papers published this month in Seismological Research Letters (Bird and Stein, 2024; Stein and Bird, 2024), we argue that ruptures on branch - or "splay" - faults enable transform faults to unzip; they are the fuse that lights the fire.

The evidence for branch fault rupture

Before about the year 2000, seismic monitoring was inadequate for researchers to be confident about whether events ruptured on or off the main faults. Thus, the evidence for our reappraisal of continental earthquake nucleation rests on a small sample: the five great strike-slip quakes that have struck in the past 25 years, which occurred in Turkiye, New Zealand, Alaska, Tibet, and China.

But for these five quakes, which have magnitudes 7.8 and larger, the case is clear. The slip rate on the branch faults is a fraction of that on the main faults, and many of the branch faults move, or slip, in different directions than the main fault. So, they are minor players rather than segments of the main faults. All but one of the five largely transform (major strike-slip) faults ruptured from one end to the other, in what seismologists term a unilateral rupture

For the most recent of these - Turkiye's 2023 Kahramanmaraş shock that ultimately killed up to 50,000 people - the branch fault lies in the center. The result was a bilateral rupture because the fault unzipped in both directions (top panel below). One could certainly argue that for New Zealand's Kaikōura shock, most of the fault ruptures were branches. However, for China's Wenchuan shock, which killed 80,000 people, the branch fault is buried and so its trace is not visible at the Earth's surface. But all of these cases have one remarkable characteristic in common: Their ruptures did not begin on the straight, simple, continuous part of the fault.

How strong is the evidence for this hypothesis?

It is far from proven. For one thing, when we drop down just one-tenth of a magnitude unit to 7.7 (that's 70% of the energy of a magnitude 7.8 shock, and 50% of the energy of a magnitude 7.9 shock), only about half of the events appear to have initially ruptured on a branch fault. And the three most recent magnitude 7.6-7.7 events do not appear to have ruptured on a branch fault: the 1999 Izmit (Turkiye), 2013 Balochistan (Pakistan), and the 2023 Elbistan (Turkiye) quakes. That could be because as events get smaller, it's harder to tell if branch faults were involved. That explanation is consistent with our finding that among 100 magnitude 6.5 and larger continental transform earthquakes since 1977, two-thirds ruptured unilaterally (Bird and Stein, 2024).

Because unilateral ruptures are perhaps more likely to involve a branch fault off the end of the rupture, this suggests that there might be no actual magnitude threshold for the occurrence of branch fault nucleation. Alternatively, the lack of evidence for magnitude 7.7 and smaller branch fault nucleation could mean that ruptures that do not begin on branches fizzle out earlier and end up being smaller. Or, finally, it could mean our sample is too small, and we are over-interpreting our results, which is always a temptation.

So, do all branch faults trigger a transform fault earthquake? The answer is no. In our survey, only about 25-50% do. But we do not contend that every branch fault rupture must trigger a great quake. Rather, we argue the converse, that branch fault nucleation might be a prerequisite for a great continental quake.

Can we look further back in time to expand the sample? With some trepidation, yes: We judge that the 1939 magnitude 7.9 Erzincan quake on the North Anatolian Fault probably ruptured on a branch fault. The 1906 magnitude 7.9 San Francisco earthquake nucleated where the San Andreas and San Gregorio faults coalesce - a site of many mapped branch faults. So, for 1906 we can neither eliminate the possibility of branch fault rupture, nor can we prove it occurred.

Why great continental quakes might require branch fault nucleation

A mature transform fault's damage zone, with tens to hundreds of kilometers of cumulative slip, distinguishes it from other continental faults. The damage zone is a band that is hundreds of meters wide and composed of rocks that have been pulverized and sheared in countless earthquakes over the eons. According to theoretical and laboratory studies (e.g., Noda et al., 2011), once a rupture enters the transform fault from the branch fault, flash heating given off by the shearing process causes a thin band of rock to melt in the fault zone. Together, heat from the shearing and melt cause the pore fluid pressure to spike, making the fault slippery and enabling it to rupture.

Once the fault begins to unzip, the damage zone also acts as a seismic "wave guide" by trapping the waves shed off the rupture tip, which promotes the continued propagation of the rupture. So, the damage zones of transform faults become pathways for dynamic rupture. By dynamic rupture, we mean that once the fault starts to slip - once it overcomes the initial fault friction - slipping along the rest of the fault is easier. Think of your stuck sock drawer that refuses to slide back in; that is the static stress you must overcome. But once you give it a shove, the drawer slams back into the dresser without any resistance; that is the dynamic rupture with greatly reduced friction.

Unlike the branch fault, the transform fault is far from static failure. Instead, when the branch fault rupture slams into the main fault, it causes the main fault to heat up suddenly and pressurizes fluids that are concentrated along the fault, triggering the main fault rupture.

How do we know mature transform faults are generally far from failure?

Extensive laboratory and theoretical work suggest that transform faults fail at very low levels of shear stress (e.g., Noda et al., 2011). Another indication that they fail at low stress levels is that models of tectonic plate motion, in which all large faults are included, cannot get the transform faults to slip at all unless their friction coefficients are set to very low values (less than 0.1, rather than the 0.8 or so that would be expected for most faults) (Bird and Stein, 2024). Yet another indication is seen when one simply plots seismicity in California; the sites of its great historic transform events (in 1857, 1872, and 1906) are all "seismicity holes" today, up to 168 years after those ruptures. By "seismicity hole" we mean that few earthquakes occur at these sites (purple highlighted faults in Figure 3). This observation suggests that, unlike the surrounding regions, these ruptured sections of the San Andreas and Owens Valley transform faults remain far from failure.

Is supershear rupture of the branch fault the key?

Parts of some ruptures have been found to propagate at supershear velocities (above the shear-wave velocity), which means that the rupture tip outpaces the most powerful, energetic seismic waves that it excites as it ruptures. This is closely analogous to an airplane breaking the sound barrier, except that the speed of seismic waves in Earth's crust is ten times greater than the speed of sound waves in air. But in both cases, a "Mach cone" is formed that creates the equivalent of a sonic (or seismic) boom (Figure 4).

Depending on the angle between the branch and main fault, the fracture energy and strength of the faults, the Mach wave from a supershear branch fault's rupture could strike a large section of the main fault at once, enabling the rupture to jump onto the main fault. The main fault would continue to rupture at supershear speeds, which would promote an expanding rupture and, ultimately, result in a higher earthquake magnitude.

But not so fast. There is only weak evidence that the branch fault ruptures were supershear. For the 2023 Kahramanmaraş earthquake, two studies found that the branch fault rupture progressed at supershear speed (Abdelmeguid et al., 2023; Ren et al., 2024), yet two others found that it ruptured at sub-shear speeds - and so it would not produce a Mach cone (Jia et al., 2023; Melgar et al., 2023). For the other great quakes, the lack of seismic stations close to the branch fault makes the presence of supershear early in the rupture process unclear. So, we cannot definitively ascribe great transform ruptures to supershear branch fault nucleation; instead, this remains a tantalizing possibility that likely will be resolved only when the next continental transform fault ruptures.

Why does this debate matter?

The strongest conclusion of our pair of studies is that for the largest continental transform earthquakes, we shouldn't expect a simple rupture will start and then grow on the transform fault. In the western U.S., as well as Japan, Taiwan, and parts of Turkiye and Mexico, earthquake early warning systems have been established to provide several seconds of warning before people experience strong shaking. These systems detect the quake as close to its source as possible, and then send this information to surrounding populations via cell phones.

If a branch fault ruptures in a magnitude 6 shock near a major transform fault, several options exist for earthquake early warning systems. In one scenario, a warning is sent for shaking expected for a magnitude 6 event. In a second scenario, that warning could include the possibility of much larger shaking from an ensuing great event that propagates along the larger fault.

Beyond this, if a branch fault ruptures near a major transform fault, could an earthquake early warning algorithm preferentially search for contagion - indications that the shaking amplitude or duration is growing beyond what was initially expected? We leave it to the builders of earthquake early warning algorithms (e.g., Allen and Melgar, 2019) to assess the feasibility and advisability of these measures, but there is a possibility that this discovery could be harnessed to save lives.