What Does Ten Thousand Years of Simulated Himalayan Earthquakes Show?

Nepal's earthquake record is long by the standards of most countries and far too short by the standards of the fault underneath it. Written accounts of damaging shaking reach back a few centuries. The Main Himalayan Thrust, the low-angle fault on which India grinds beneath Tibet and which raised the mountains overhead, has been storing and releasing strain for millions of years. Every hazard map in the region has to answer one question above all others: how large can an earthquake here get? Nobody has been watching long enough to know.

Gorkha in 2015 is the reference point most people carry. That magnitude 7.8 rupture killed roughly 9,000 people, and studies afterwards concluded it stopped short of the surface, leaving strain unreleased in the segment to the south. Larger Himalayan earthquakes sit in the historical and geological record. Whether the fault can produce something larger still, in the magnitude 9 class seen on the subduction zones off Sumatra and Japan, is one of the genuinely open questions in continental seismology.

One way to lengthen a short record is to dig trenches across the fault and read the old ruptures preserved in the sediment. Another is to stop waiting and manufacture the record. That is the route taken by Govinda Niroula, Mark Stirling and Jack Williams of the University of Otago in New Zealand, with Matthew Gerstenberger of Earth Sciences New Zealand and Luca Dal Zilio of Nanyang Technological University in Singapore, in a paper published in the open-access journal Seismica. They built 10,000 years of synthetic Himalayan earthquakes, then asked what that fictional history implied for the real ground beneath Nepal.

The engine is RSQSim, a physics-based earthquake simulator already applied to fault systems from California to New Zealand. Instead of drawing earthquakes at random from a statistical distribution, it treats the fault as a mesh of small patches, each governed by rate-and-state friction, the laboratory-derived law describing how a sliding surface's resistance depends on how fast it is moving and how long its two sides have been in contact. Stress accumulates, a patch fails, the failure either spreads to its neighbours or stalls. Let it run long enough and earthquakes emerge by themselves, without anyone deciding in advance how big they should be.

What the researchers fed it were geodetic models of the fault's geometry and slip rate, both built from satellite measurements of how the surface of the Himalaya is moving today. A synthetic catalogue is worth something only if it resembles reality in the places where reality is known, and this one does. It reproduces the observed magnitude-frequency distribution, the sizes of ruptures and their spatial pattern along the fault.

Then comes the result. Geodesy shows that the Main Himalayan Thrust is not uniformly stuck. Some stretches are strongly coupled, accumulating strain that must eventually be released in an earthquake. Others are weakly coupled, creeping quietly and storing very little. In the simulations, those low-coupling zones behaved consistently as rupture barriers. A rupture that began in one locked segment tended to run into a weak patch and stop there rather than carry on into the next segment. Across 10,000 simulated years the largest earthquake reached magnitude 8.9, which is an enormous earthquake and still smaller than the fault's full length would permit.

That measurement deserves care, and the authors give it some. It is a property of a modeled fault under an assumed geometry, an assumed slip rate and an assumed friction law, not a measurement of what the rock will do. Change the assumptions and the number can move. It also lands in an argument that is very much live: a real strand of the published literature holds that the Himalaya can generate magnitude 9 or more, and this work does not close that question. What it offers is a mechanism by which a limit could arise in the first place.

The second half of the study turns the synthetic catalogue into a hazard estimate. Probabilistic seismic hazard assessment, the calculation that sits behind building codes, asks how strongly the ground at a given site is likely to shake within a given stretch of time. Conventional models assemble an inventory of possible earthquake sources and assign each a rate. The team let the simulated catalogue supply the earthquakes instead, then compared the resulting picture of Nepal against a conventional one.

The two disagree, and they disagree lopsidedly. At the 10 percent probability of exceedance in 50 years used in much engineering practice, the simulation-based peak ground acceleration reaches 0.1 g, or 20 percent, above the conventional estimate at one extreme, and 0.6 g, or 70 percent, below it at the other. At the rarer, stronger shaking levels the same shape holds: as much as 30 percent higher, as much as 40 percent lower. Those are the outer edges of the comparison, not its centre. Measured in relative terms, 84 percent of sites differ from the conventional model by less than 60 percent at the 10 percent level, and by less than 30 percent at the two rarer ones. Most places move much less than the headline numbers suggest.

Which of the two is closer to the truth is not something this paper claims to settle, and its authors do not ask anyone to discard the conventional approach. Their stated conclusion is that physics-based simulations offer a valuable complement to conventional source models, particularly for faults like the Main Himalayan Thrust, where the rheology varies along the fault and the earthquake record is thin. Simulators of this type are not themselves new, and the group has described its Himalayan catalogue at meetings before. The peer-reviewed barrier result and the Nepal comparison are the fresh material.

Seismica, where the work appears, is a community-run diamond open-access journal published by McGill University Library, and it posts the peer-review reports alongside each article. Everything in it about maximum magnitude is a statement about a simulation. For Nepal, the practical question is narrower than the magnitude debate. Would a hazard map built from simulated ruptures put the strongest expected shaking somewhere other than the current one does? The site-by-site comparison here says the two agree closely at most places and diverge sharply at a few.