The 2025 Mw7.7 Mandalay, Myanmar, earthquake reveals a complex earthquake cycle with clustering and variable segmentation on the Sagaing Fault

Authors: Solene L. Antoine, Rajani Shrestha, Chris Milliner, Kyungjae Im, Chris Rollins, Kang Wang, Kejie Chen, Jean-Philippe Avouac

Abstract

Large earthquakes often occur on faults that were known to have produced destructive events in the past. However, anticipating the characteristics of these earthquakes and their impacts remains a great challenge. The 2025 Mw7.7 Mandalay earthquake was produced by rupture of an unusually long stretch of the Sagaing Fault in Myanmar, primarily a section that had not broken since 1839 and was considered a zone of high hazard, as well as sections that experienced more recent earthquakes. These observations challenge the usual approaches used in seismic hazard studies to evaluate how a fault is spatially divided and strain is built up and released over time. We show that physics-based simulations of earthquake sequences can provide an alternative approach.

The March 28, 2025, Mw7.7 Mandalay earthquake occurred at 20 am UTC (1:20 pm local time) (1) and was followed by a Mw6.7 aftershock 1 h later (2). With maximum intensities greater than 9 (1), it is the most strongly and extensively felt and damaging earthquake recorded in this region during the instrumental period. Each time such a destructive earthquake happens, questions arise regarding whether such an event was to be expected and whether some of its characteristics could have been anticipated. For example, the 2023 Mw7.6 and Mw7.8 Kahramanmaraş, Türkiye, earthquake sequence is an example of fault rupture known to have produced destructive earthquakes in the past (3). However, the supershear nature of these events could not have been anticipated from the complex and curved geometry of the ruptured faults (4), questioning the view that a supershear rupture requires a simple fault geometry (5–7). In this case, spatial variations of the stress field helped steer the supershear ruptures (8). This example highlights that anticipating the location and magnitude of future events and some of the characteristics determining their impacts, like the stress drop and rupture speed, remains a daunting challenge.

The 2025 Mw7.7 Mandalay earthquake is another example of a large earthquake that accompanied a supershear rupture (1) along a well-known active fault, the Sagaing Fault, which accommodates part of the northward motion of India relative to Sunda (9) (Fig. 1A). In that case, the chance of a large event on that particular fault segment had been anticipated from the long-lasting “seismic gap” observed since the last Mw>7 event in 1839. The location and extent of this historical event are not well known, and it could have reached Mw7.9 based on scaling relations between length and magnitude (10). The Meiktila segment (Fig. 2A) was therefore identified as a region of elevated seismic hazard (10, 11). The fact that this fault could produce supershear ruptures was also anticipated in view of its continuous and rather straight geometry (6, 7). Over the last 200 y, the Sagaing Fault produced another eight Mw>7.0 earthquakes (10, 12), in accordance with its great length and rather fast slip rate of 16 to 24 mm/yr, and geodetic observation of interseismic locking from the surface to a depth of 10 to 16 km (13–16). The largest recorded event is a Mw7.7 in 1946 (10, 12). The geometry of the >1,200 km-long Sagaing Fault (Fig. 2A) is well documented from its geomorphic expression (10) and active seismicity (Fig. 2B) (12), and its slip rate and locking depth from geodetic measurements (13–17).

Building on these previous observations, we raise the following 1) how the slip and rupture extent observed during the 2025 Mandalay earthquake relate to the slip deficit accumulated due to interseismic locking and previous historical ruptures, 2) whether the Mw7.7 magnitude observed in 2025 is representative of the maximum earthquake magnitude on the Sagaing Fault and what is the return period of such events, 3) whether the rupture extent, slip magnitude, and timing of the Mandalay earthquake could have been anticipated based on simple modelling of the seismic cycle. We address these questions by documenting the surface deformation caused by the 2025 Mandalay earthquake from cross-correlation of Sentinel-2 optical and Sentinel-1 SAR images and a derived finite-fault slip model. We then analyze the seismic cycle on the Sagaing Fault through moment budget analysis and quasidynamic earthquake cycle simulations.

Observations on the 2025 Earthquake

Optical and Radar Remote Sensing Observations of Surface Deformation.

The N–S and nearly horizontal orientation of the surface displacements, and the large displacement gradients in the near-fault domain make this earthquake ill-suited for SAR interferometry. Surface ruptures and ground displacements can alternatively be documented from cross-correlation of optical (18), and SAR amplitude images (19). Here, we perform cross-correlation on Sentinel-2 optical, and Sentinel-1 SAR images (SI Appendix, Text S1 and Table S1). Postearthquake images were acquired several days after the event, so the measured surface displacements include that of the Mw~6.7 aftershock. The results reveal highly localized right-lateral deformation along a prominent 510 km-long surface rupture which is remarkably continuous and straight (Fig. 1 B and C). The rupture follows very closely the main fault strand mapped by Wang et al. (10), and none of the subsidiary faults were ruptured in that event (Fig. 2A). East–West (E–W) displacements are minor (SI Appendix, Fig. S1 A and D), which is consistent with the purely strike-slip focal mechanism determined by the USGS (Fig. 1A). We measure the horizontal components of fault slip parallel and perpendicular to the fault trace using 1 km-wide swath profiles placed every 1 km along the fault trace (SI Appendix, Text S1 and Fig. S2). The surface slip distribution (Fig. 2C, black curve) shows little variability along strike with a mode value of 3.8 ± 0.004 m, an average of 3.3 ± 0.004 m, and a peak value of 5.7 ± 0.1 m in the epicentral area. The mean value is lower than the mode value because it includes the regions where fault slip tapers down. Fault-normal horizontal slip is negligible (Fig. 2C, red curve).

Finite-Fault Slip Model.

We combine the surface displacements measured from the optical images and SAR amplitude images (including the range direction sensitive to vertical displacements) to determine the three-dimensional (3-D) surface displacement field (SI Appendix, Text S2 and Fig. S1 C–E) and fault slip at depth (Fig. 2D; and see SI Appendix, Text S3). To determine the slip distribution at depth, we consider a 3-D fault model that follows, at the surface, the trace of the rupture observed in the imagery (Fig. 1B) and is subdivided into nine planar segments (SI Appendix, Fig. S4). The dip angle of each segment is determined using a nonlinear Bayesian inversion of the 3-D displacement maps (20). It steepens from 68°E to 82°W southward (SI Appendix, Fig. S3). This fault geometry is supported by previous GNSS and seismicity measurements that suggest a 60 to 70 eastward dipping geometry north of the Meiktila segment (12, 13, 15), and a more vertical geometry along the Meiktila segment (16). The best-fitting slip model (Fig. 2D) obtained through standard linear least-squares inversion of the surface displacements (SI Appendix, Text S3 and Figs. S4–S7) is characterized by a relatively homogeneous slip distribution. Slip tapers off at a depth between 14 km in the North and 10 km in the South (Fig. 2 D and E), right above the locking depth determined from GNSS measurements of interseismic strain (13–16) (dashed blue box in Fig. 2D). The total moment computed from the slip potency and assuming a shear modulus of 30 GPa corresponds to a moment magnitude of 7.74. Note that our source model includes deformation due to the Mw6.7 aftershock. This value is consistent with the seismological estimate of the moment released by the mainshock and the Mw6.7 aftershock of 4.7681e+20 N-m (1), corresponding to a moment magnitude of 7.71. The mean stress drop calculated from our finite slip model using Okada’s Green functions (21) and slip-weighting (22) is 4.7 MPa.

Observations and Modeling of Seismic Cycle on the Sagaing Fault.

Seismogenic potential and moment budget analysis.

The known seismicity is most probably too short to represent the full range of possible ruptures on the Sagaing Fault. One might therefore wonder how frequently earthquakes similar to the 2025 event might return and, given that the fault is nearly twice as long as the segment that ruptured in that event, whether even larger ruptures might occasionally occur. To answer these questions, we first compare the moment released by earthquakes over the last two centuries, including the 2025 Mandalay earthquake, with the moment deficit accumulated over the interseismic period. We consider the magnitudes reported by Wang et al. (10) or, alternatively, those of the ISC-GEM catalog (23, 24), and a fault slip rate of 16 to 24 mm/yr with a locking depth of 10 to 16 km (15) (SI Appendix, Text S4). We find that the moment budget on the Sagaing Fault is approximately conserved over that period (Fig. 3B). The fault segment to the north ruptured in a sequence of Mw > 6.5 events between 1946 to 1956 and had a Mw > 6.5 event as recent as 2012. The segment to the south hosted two Mw > 6.5 events between 1929 and 1930. As a result of these events, the interval between 1929 to 1956 seems to be a period of enhanced seismicity (a “cluster”) when the seismic moment release catches up with the moment deficit (Fig. 3B).

We next build a long-term magnitude-frequency distribution (MFD) enforcing conservation of moment (25) and assuming a power-law, “Gutenberg–Richter (GR),” cumulative distribution truncated at a maximum magnitude Mmax (26) (SI Appendix, Text S4). The model parameters are the b-value, representing the slope of the Gutenberg–Richter MFD law, and Mmax and its associated return period (Fig. 3C). We use a range of b values, moment buildup rates, locking depths, and seismic moment release fractions to build a suite of seismicity models (27) (SI Appendix, Text S4). We determine the most likely model parameters in view of the seismicity recorded in the ISC-GEM catalog (1904 to 2020) using a Bayesian inversion scheme (27) (SI Appendix, Text S4). We consider only crustal seismicity (depth ≤ 40 km) within 60 km from the fault along its 1,235 km-long trace (Fig. 3A). The most recent version of the ISC-GEM catalog lists 129 events between 1906 and 2019 (114 y) within this region. This choice excludes the Mw7.88 event in 1912 (ISC-GEM catalog) which occurred on the secondary Kyaukkyan fault (10). Though, considering a wider zone including this event has a negligible impact on the moment budget analysis (SI Appendix, Fig. S8). The analysis takes into account that the magnitude of completeness, Mc~, of the catalog has considerably varied with time, and returns a relatively standard b-value of 1.04 (Fig. 3 C, Inset; and see SI Appendix, Text S4). We obtain a probability distribution of the maximum magnitude that peaks at Mw7.9 (Fig. 3C). This shows that moment conservation does not require the Sagaing Fault to produce events significantly larger than the Mandalay earthquake. The return period of a Mw7.9 event is estimated to be 250 y. Our analysis, however, does not preclude larger events as the maximum magnitude has a relatively broad probability distribution. The number of Mw~> 6.5 earthquakes on the Sagaing Fault over the last century also exceeds the number of events expected from the long-term model, suggesting a more pronounced clustering than expected for a memoryless process. In this study, clustering refers to successive ruptures of the same fault section recurring more frequently than expected from a homogenous Poisson process, in accordance with the usage of that notion in paleoseismic studies (28, 29).

Quasi-dynamic seismic cycle modeling.

Even minor geometric changes can in principle result in barriers to the propagation of earthquakes, and influence the spatiotemporal pattern of ruptures on a fault (30). To test whether that hypothesis may hold for the Sagaing Fault, we run simulations to evaluate whether the observed rupture pattern on this fault can arise from its geometry using the 3-D quasi-dynamic earthquake simulator Quake-DFN (31) (Fig. 4; and SI Appendix, Text S5). The model represents the entire Sagaing Fault including the splay fault system to the north where greater influence from the geometry should be expected. The model considers faults governed by rate-and-state friction embedded in an elastic half-space with homogeneous initial stress and remote loading (SI Appendix, Text S5 and Fig. S10). The parameters are adjusted to obtain seismic ruptures with slip comparable to that of the 2025 event (Figs. 2C and 4C). Rate and state friction is probably not appropriate to account for enhanced weakening at seismic slip velocities (32). Nonetheless, we obtain realistic coseismic slip with a value of a-b of about −0.003, of the same order of magnitude as values measured in the laboratory (33). By contrast, to limit the computational cost, we choose a Dc value of 3 cm. Although this value is much larger than the typical laboratory measurements of <0.1 mm, it is widely recognized that on rough natural fault Dc is likely orders of magnitude greater than on very smooth laboratory faults (34, 35). As a result, the nucleation size is relatively large, and our simulations can not produce earthquakes with magnitude lower than Mw~6.5. The ratio of the nucleation size to the cell size (between 3 and 8 at the center of the fault) is large enough to ensure that the simulations are well resolved numerically (36) but not large enough to get significant self-organized complexity (37).

With these parameters, a planar fault produces essentially a periodic pattern with Mw7.8 full ruptures returning every 140 y (SI Appendix, Fig. S11). Simulations run with the nonplanar geometry (Fig. 4A), however, result in partial ruptures and complex seismic cycles that bear similarities with observations on the Sagaing Fault, including occasional Mw7.7 ruptures reaching 500 km of rupture length alternating with sequences of more partial ruptures (Fig. 4 B–D). The model does not produce longer ruptures with these parameters because stress variations resulting from the nonunique fault strike create heterogeneities that are able to arrest ruptures rather than running from one end to the other of the 1,235 km-long fault model. Once a rupture is arrested, stress barriers become self-sustained. We also notice that, although the fault geometry is nonvarying during the simulations, the extent and location of partial ruptures vary substantially, with successive ruptures of the same fault portion arresting at different points. Indeed, the simulation shows that branch junctions and major fault bends are causes of segmentation, yet in a not systematic way. In particular, we observe that the northern region characterized by the splay fault system produces smaller and more frequent earthquakes, as suggested in the historical seismicity (Fig. 3A) and expected for a more complex fault system. Events rupturing the relatively straight and smooth central segment, corresponding to the Meiktila seismic gap, return relatively regularly in the model, and most often break the entire length of the segment. However, the ruptures extend to variable distances north and south of this region. These larger events reach magnitudes slightly above 7.7, and Mw>7.5 events return irregularly. Their mean return period is 141 y, with a SD of 40 y. The mean return period of Mw~>7.5 events is consistent with the historical record and comparable with the value of 96 y predicted by most likely MDF derived from moment conservation (Fig. 3C). Therefore, the model succeeds in producing periods of clustered partial ruptures, as observed between 1929 and 1956, and occasional larger ruptures like the Mandalay earthquake. It however fails to produce the relationship between the ruptures of 1956 and 2025 as we never observe enhanced slip along sections that ruptured in relatively recent earthquakes in the simulation.

Discussion and Implications

Exceptional Characteristics of the 2025 Rupture.

The 2025 Mandalay earthquake represents the longest, 510 km, documented continental rupture ever (38), exceeding the 450-km length of the 2001 Mw7.9 Kokoxili rupture (39). Longer ruptures have only been observed on subduction megathrusts, which extend to much greater depths of 40 to 50 km, as for example the Mw9.1 Sumatra (>1,700 km) (40) and the Mw9.5 1960 Chile earthquake (1,000 km) (41). The length of the 2025 Mandalay rupture is about five times greater than that expected from the global scaling properties of strike-slip events (42, 43). Following these scaling laws, a 510-km length should correspond to a Mw8.2 event with a mean slip of 9 m, and a moment of 2.7 × 10^21^ Nm that is more than five times the observed moment. Similarly, an average and mode slip of 3.3 to 3.8 m, as measured for the 2025 event, should correspond to a Mw7.3 to 7.4 earthquake. This demonstrates the difficulty of accurately determining magnitudes from anticipated rupture lengths and average offsets. Moreover, the use of offsets measured at a single paleoseismic site may show variations from event to event, but these differences may be a function of more than just the magnitude of the earthquake ruptures, including local fault geometry and surface lithology (44, 45).

The exceptional rupture length of the 2025 event was likely enabled by the simple geometry of that particular section of the Sagaing Fault which lacked any geometrical barrier adequate to impede rupture. This simple geometry results from the long-term fault structural smoothing associated with ~460 km of accumulated slip over the past ~13 Myr (46, 47). Few continental faults have accumulated such large offsets. The San Andreas (>440 km) (48) and the Alpine and Wairau (430 to 480 km) (49) plate boundary faults are other exceptional examples of smooth faults that accumulated large slip. The Altyn Tagh Fault, an intraplate strike-slip fault which has accumulated several hundreds of kilometers since the onset of the India-Asia collision (50), also falls in that category. These so-called mature faults have long been identified as being favorable to hosting large (51, 52) and eventually supershear ruptures (5–7) as happened with the Mandalay earthquake. The supershear velocity of the 2025 event could also have been favored by the contrast of elastic properties (53) between the Cenozoic Central Myanmar sedimentary Basin (MCB) to the West, and the Mogok Metamorphic Belt (MMB) constituted of high-grade metamorphic rocks to the East (54). The southward supershear propagation of the 2025 rupture is consistent with the theory which predicts a preferred propagation in the direction of motion of the stiffer block in the case of a supershear pulse (53).

The large (510 km/10 km) length/width aspect ratio of the 2025 Mandalay rupture is also the largest ever documented for a strike-slip rupture. Notably, slip did not reach deeper than the geodetically inferred locking depth of the Sagaing Fault. However, it has been argued that very long strike-slip ruptures might extend to depths below the locking depth due to enhanced dynamic weakening, and that this process would result in seismic quiescence at these depths during the interseismic period (55). This model is contrasting with the seismic activity at depths down to ~20 km revealed by local seismicity monitoring between 2016 and 2018 over the area that ruptured in 2025 (12). This seismicity is consistent with a rupture confined to the locked area. If dynamic weakening played a role in the development of the Mandalay earthquake, it did not result in a seismic rupture extending beyond the locking depth.

Our study shows that slip at the surface is not lower than at depth, revealing no Shallow Slip Deficit (SSD) (56) (Fig. 2E). SSD is often argued to be the result of dynamic off-fault elastoplastic deformation (57) and is commonly observed for continental ruptures (58). The absence of SSD for this event is consistent with the observation that large (>Mw7.5) surface-breaking events tend to have maximum slip at the surface (58). It is also consistent with the view that the rupture of smooth, mature faults are not associated with much off-fault deformation (OFD) (59).

Comparison of 2025 Rupture with Past Ruptures on the Sagaing Fault.

We analyze whether the 2025 rupture pattern complies with the fault segmentation proposed by Wang et al. (10) (triangles in Fig. 2A), and whether it was influenced by the rupture history of the fault (Fig. 2B). To illustrate this discussion, we calculated the slip deficit accumulated since the last rupture (dashed grey lines in Fig. 2C). While slip on Meiktila segment conforms the slip-predictable model (60) (Fig. 2C), we do not observe lower slip on the more recently ruptured segments as would be expected from the model (Fig. 2C). As a result, the extent of the 2025 rupture relates only loosely to past ruptures. In fact, the northern end of the 2025 rupture tails off in the section that ruptured during the 1946 Mw7.7 earthquake, and again during a Mw6.8 event in 2012. The stress drop due to these previous events therefore seems to have acted as a stress barrier. However, the 2025 rupture overlaps by 30 km with the 1946 rupture. The 2025 event also reruptured sections that had experienced recent earthquakes where little slip was expected based on the slip-predicable model. Most notably, the maximum slip of 5.7 m is observed along a section that had ruptured in a Mw7.1 event in 1956. To the south, the relationship with previous ruptures is even more complicated. The 2025 event was able to rupture 50 to 100 km into the section that had ruptured during the Mw7.3 earthquake of 1930, rerupturing on the way a segment that ruptured in a smaller event in 1929. Our observations show that slip due to the 2025 rupture complies, at a first order, with that expected from a slip-predictable only along the Meiktila seismic gap (Fig. 2C) where the fault is very smooth and straight, but departs significantly from the prediction elsewhere. The fault geometry is probably a key factor that determines where conformity with the slip-predictable model is verified or not. The discrepancy with the expectation from the slip-predictable model implies that it can take several successive ruptures for coseismic slip to balance the deficit of slip due to fault locking in the interseismic period, as is also observed in the quasidynamic simulation (Fig. 4E).

The 2025 Event Part of a Multicentury Cycle.

The history of successive fault ruptures on the Sagaing Fault shows that spatial segmentation is variable, as reported on a number of continental faults (61) and subduction megathrust (62–64). For example, the Imperial Valley fault is a clear example where the same portion of the fault ruptured in 1940 and again in 1979 as part of a larger event (61). We also observed that the release of seismic moment on the Sagaing Fault is quite irregular over time, with the last century that appears to be a period of enhanced seismicity compared to the long-term average required to balance the moment budget (Fig. 3c). Such clustering of repeating large events is commonly observed in paleoseismic records and referred to as “supercycles” (28, 29, 62, 64). Our simulations (Fig. 4) show that such a supercycle can simply emerge from the influence of the along-strike fault geometry on seismic ruptures, the planar-fault reference model creating only periodic ruptures. Other factors could also play a role such as spatial heterogeneities and temporal evolutions in fault strength and friction (65), self-sustained complexity of the seismic cycle when the nucleation length is much smaller than the faut size (37), and interactions between faults (66). However, it is interesting to note that these factors are not needed to obtain a seismic cycle with realistically complexity.

Conclusions

The 2025 Mw7.7 Mandalay earthquake is a unique example of extremely long (510 km) and narrow (10 km) rupture which did not produce any significant shallow slip deficit. These features, along with the sustained supershear velocity, are most likely related to the smooth and straight geometry of this section of Sagaing Fault owed to its structural maturity (>400 km). The Mw7.7 magnitude of this event is close to Mw7.9 maximum magnitude needed for moment conservation. This estimate is comparable to the maximum magnitude estimated by Wang et al. (10) based on their proposed segmentation of the Sagaing Fault. However, the consistency is coincidental since the 2025 rupture complies with neither the proposed segmentation nor the scaling laws that were used to infer the possible maximum magnitudes. In fact, the distribution of past ruptures on the Sagaing Fault does not reflect a persistent segmentation of the fault, and stress barriers resulting from past events could partially be overcome during the 2025 event. This weak segmentation might also be favored by the high structural maturity of this fault. As a result, the seismic cycle on the Sagaing Fault is rather the 2025 event ruptured both a 186-y seismic gap and sections which had ruptured during a cluster of smaller ruptures between 1929 and 1956. This supercyle pattern makes it challenging to forecast the time, rupture extent, and magnitude of future earthquakes. However, the fact that such a behavior is reproduced in the Quake-DFN simulations suggests that data assimilation into this or other earthquake simulators (67) could provide a way for time-dependent hazard assessment in the future.

Supplementary Material