The ongoing collision of India with Asia is partly accommodated by slip on the Main Himalayan Thrust (MHT). The 25 April 2015, Mw 7.8 Gorkha earthquake is the most recent major event to rupture the MHT, which dips gently northward beneath central Nepal. Although the geology of the range has been studied for decades, fundamental aspects of its deep structure remain disputed. Here, we develop a structural cross section and a three-dimensional, geologically informed model of the MHT that are consistent with seismic observations from the Gorkha earthquake. A comparison of our model to a detailed slip inversion data set shows that the slip patch closely matches an oval-shaped, gently dipping fault surface bounded on all sides by steeper ramps. The Gorkha earthquake rupture seems to have been limited by the geometry of that fault segment. This is a significant step forward in understanding the deep geometry of the MHT and its effect on earthquake nucleation and propagation. Published models of fault locking do not correlate with the slip patch or our fault model in the vicinity of the earthquake, further suggesting that fault geometry was the primary control on this event. Our result emphasizes the importance of adequately constraining subsurface fault geometry in megathrusts in order to better assess the sizes and locations of future earthquakes.
Many structural cross sections of the Himalaya have been developed over the past several decades, based on geology, topography, and seismicity (e.g., Le Fort, 1975; Stocklin, 1980; Schelling and Arita, 1991; Pandey et al., 1995; Pearson and DeCelles, 2005; Khanal and Robinson, 2013). They usually include an active thrust (Main Frontal Thrust [MFT]), exposed along the southern edge of the Sub-Himalayan foothills, which merges at depth with a fault (Main Himalayan Thrust [MHT]) that dips gently to the north beneath the Lesser Himalaya. This “décollement,” or bed-parallel fault, in turn steepens downward onto a ramp that dives beneath the Higher Himalaya before flattening again northward under the Tethys Himalaya of southern Tibet. This ramp is necessary to produce the large Gorkha-Pokhara Anticlinorium (GPA), which extends through most of central Nepal, as well as the High Himalaya. Despite their overall similarity, the cross sections differ in significant features, such as the location, depth, and size of the ramp, the existence of other ramps, and the total amount of shortening. In addition, the sections drawn through the Kathmandu Klippe (Pearson and DeCelles, 2005; Khanal and Robinson, 2013) are not fully consistent with basic parameters of the Gorkha earthquake obtained from the global centroid-moment-tensor solution (GCMT; Ekström et al., 2012), such as the ∼12 km depth and ∼7° dip of the fault. These fault plane parameters are consistent with a joint slip inversion data set based on GPS, teleseismic, and interferometric synthetic aperture radar (InSAR) data, which indicate that the rupture extended ∼140 km along strike and ∼50 km downdip (Avouac et al., 2015). Such a rupture would be consistent with slip on a décollement.
STRUCTURAL CROSS SECTION
We present a cross section based on quantitative models of fault-related folding (assuming deformation by flexural slip and conservation of line length, thickness, and area; Suppe, 1983) that incorporates the regional geology (Stocklin, 1980; Geologic Map of Nepal [www.dmgnepal.gov.np/geology-of-nepal]; Pearson and DeCelles, 2005; Khanal and Robinson, 2013; Fig. DR1 in the GSA Data Repository1) and the fault orientation and depth from the main-shock GCMT solution. We propose that not one, but two active blind ramps bound the fault that slipped in the earthquake (Fig. 1). A deep, active, northern ramp is required to produce the GPA. A middle ramp to the south is required by the stratigraphy exposed in the hanging walls of the MFT and Main Boundary Thrust (MBT).
Both the MFT and MBT sole into the MHT. They are also both parallel to the rocks exposed in their hanging walls, which therefore reflect the stratigraphic levels to which they sole. However, these levels are different: The rocks above the MFT are mid-Miocene Lower Siwalik units, while those above the MBT are the Proterozoic Dandagaon Formation (Pearson and DeCelles, 2005; Khanal and Robinson, 2013). Based on this difference, we infer that the MBT, which formed first, rose from a deeper décollement within the Lesser Himalayan Sequence. When the fault system broke forward, it moved onto an upper décollement at the base of the Lower Siwalik units. This décollement truncated the frontal thrust, with the upper part of the MBT in its hanging wall. As slip continued south on faults in front of the MBT, the upper part of the MBT was progressively displaced, leaving the lower part behind as the middle ramp of the active fault.
We use this dual-ramp interpretation to build the cross section. Our section shows deformation postdating the largely ductile strain preserved in the Greater Himalayan Sequence (e.g., Le Fort, 1975). The earliest stage of deformation (Figs. 1A and 1B) occurs on the thrust termed the Ramgarh Thrust (Pearson and DeCelles, 2005) or brittle Main Central Thrust (MCT; Searle et al., 2008), which underlies the greenschist-facies rocks at the base of the Greater Himalayan Sequence. The distinction affects pre–13 Ma shortening estimates, but it does not change our structural cross section. We refer to this fault as the MCT herein.
In a second stage (Figs. 1B and 1C), a new, deeper décollement level developed, and a new ramp broke forward within the rocks of the Lesser Himalayan Sequence; this geometry is necessary in the reconstruction because the original MCT décollement is now refolded, eroded, and exposed both continuously at the base of the High Himalaya, and as klippen to the south.
A third stage of deep ramp formation (Figs. 1D and 1E) is required to have produced the steep dips (up to 70°; Fig. DR1) on either side of the GPA through refolding, and to raise the rocks in the hinge of the GPA enough to be exposed at the surface. The development of the upper décollement level at this stage resulted in the formation of the middle ramp.
Currently (Fig. 1F), the active deep ramp to the north represents the latest and deepest of a series of three deep ramps that have broken forward over time, producing a refolded, or imbricated duplex. The middle and upper décollements are connected by the active middle ramp. The topographic signature of the middle ramp can be seen north of the Kathmandu Valley, where the Sheopuri Lekh range stands ∼1200 m above the valley floor (Fig. DR2). The final section is consistent with the exposed geology (Fig. 2).
THREE-DIMENSIONAL FAULT MODEL
(1) The MFT is located at the topographic break at the front of the system (Fig. DR3).
(2) The middle ramp can be traced along strike using the MBT as a proxy (Fig. DR3), because in our model, the distance between the top of the middle ramp and the base of the MBT is determined by the total slip on all faults south of the MBT. We used a value of 21 km for this distance. This amount is within the range of shortening estimates for central and eastern Nepal found by previous authors (15–40 km; Schelling and Arita, 1991; Schelling et al., 1991; Mugnier et al., 1999; Hirschmiller et al., 2014). This range is in part due to differences in balancing methods and depth to décollement used in different studies. However, when the same techniques and assumptions are used, the shortening amount does not vary significantly along strike, at least in central Nepal (Hirschmiller et al., 2014), suggesting that while the total amount is not well constrained, the slip past the MBT remains relatively constant along strike. This allows us to use the MBT as a proxy for the shape of the middle ramp.
(3) The top of the deep ramp can be traced along strike using the axis of the GPA as a proxy (Fig. DR3).
The resulting regional MHT model is shown in Figures 3 and 4 and in Figure DR4. A digital file with the fault geometry for all of Nepal is available in the Data Repository. Within the region of the Gorkha earthquake, the two ramps are separated by a middle décollement. However, to the east and west, the ramps approach each other, merging to yield a single large ramp with no middle décollement; we call these junctions “pinch points” (bold lines in Fig. 4B), as the middle décollement pinches out. This model geometry is the consequence of the curvature of the MBT and the GPA. Toward western Nepal, the MBT and GPA, and hence the middle ramp and deep ramp, diverge once more, and the middle décollement is again present.
COMPARISON OF THE FAULT MODEL WITH EARTHQUAKE SLIP AND SEISMICITY
Strikingly, when we compare our fault model to the Gorkha earthquake coseismic slip inversion data set (updated version of Avouac et al.  model; Wei, 2016, personal commun.), we see that the location and shape of slip (>1 m) match closely with the location and shape of an isolated patch of middle décollement bounded on all sides by ramps (Fig. 3). This indicates that not only is the main active fault composed of several discrete segments with different dips, but that this geometry controlled the shape and size of the main-shock rupture. Thus, subsurface fault geometry may play an important part in defining rupture patches elsewhere. Beyond the Gorkha rupture zone, we observe no other pinch points in the Nepal Himalaya—the middle décollement extends for over 400 km westward, while the single larger ramp (middle ramp + deep ramp) dominates to the east (Fig. 4). In far western Nepal, the deep ramp has an abrupt shift to the north. We propose that this may be a lateral break in the MHT that may act as a barrier to slip in some earthquakes.
A comparison of our model to relocated earthquakes (Mw ≥ 2) from A.D. 1995 to 2001 (Ader et al., 2012) shows that the lateral break matches a change in the distribution of microseismicity (Fig. 4C; Fig. DR5), which has been suggested to be associated with a change from a single ramp to an active duplex structure (Harvey et al., 2015). In eastern Nepal, the microseismicity, commonly associated with the deep ramp, outlines the northeastern edge of the slip patch.
GCMT focal mechanisms are available for six thrust-mechanism aftershocks following the Gorkha earthquake (Mw 5.1–7.3; Fig. 4A). If we assume that these aftershocks occurred on the MHT, then we can compare the nodal plane dips with the local dip of our modeled MHT. The north-dipping nodal planes of four of the aftershocks dip 23°–26° and line up along the middle ramp, and a fifth, located close to the eastern pinch point, dips 14°, all of which are consistent with our model. The largest aftershock (Mw 7.3) is ambiguous: Reported depths range from 12 km to 28 km (National Earthquake Information Center, 2015). This earthquake could have occurred on the deep décollement or on an unidentified fault in its hanging wall.
LONG-TERM CONTROL OF FAULT GEOMETRY ON SEISMICITY
Geodetic studies have shown that the MHT is locked, and they have constrained the downdip extent of coupling (Ader et al., 2012; Stevens and Avouac, 2015). When we compare the downdip transition zone of these coupling models to our fault geometry, we note that it follows our deep ramp in western Nepal, but not in the Gorkha earthquake region (Fig. 4B; Fig. DR6). It seems that although the coupling pattern is largely a consequence of fault depth and geometry, it is also complicated by other factors. Within the coupling transition zone, we expect to see a combination of coseismic and interseismic slip. A comparison of the coseismic slip model of Avouac et al. (2015) and the slip model of Elliott et al. (2016), which includes deformation up to 1 June 2015, shows that both have similar updip boundaries, but the longer-period observation includes significantly more slip on the deep ramp (Fig. 4A). The key observation is that neither the coupling nor any previously proposed fault geometries (e.g., Elliott et al., 2016) can explain the shape and abrupt gradients in slip observed in the Gorkha earthquake, whereas the changes in fault geometry can.
The active fault geometry varies at the time scale of millions of years (Figs. 1A–1D); however, the geometric barriers should have remained roughly constant for the last million years (Figs. 1D and 1E). The paleoseismological record is limited, but the last earthquake that occurred in the region (1833) seems to have been controlled by the fault geometry. The rupture area and size (Mw7.5–7.9) of the A.D. 1833 earthquake have been estimated using intensity reports (Bilham, 1995; Szeliga et al., 2010) and closely resemble those of the Gorkha earthquake.
On the upper décollement, neither coseismic nor measurable afterslip occurred following the Gorkha earthquake (Grandin et al., 2015; Lindsey et al., 2015; Elliott et al., 2016), leaving open the question of how convergence is accommodated on this part of the fault. Studies suggest that in very large historical events, the MHT ruptured coseismically to the surface (e.g., A.D. 1255, 1505, 1934; Nakata et al., 1998; Sapkota et al., 2012; Bollinger et al., 2014), and that the majority of Himalayan convergence reaches the frontal fault as slip (Lavé and Avouac, 2000). Thus, we expect that the upper décollement will produce large slip events. Great earthquakes likely initiate at the base of the seismogenic zone and rupture through geometric barriers to the surface. Smaller earthquakes, however, seem to break geometry-controlled patches, and some of these may be limited to the upper décollement.
Empirical relationships can be used to estimate earthquake magnitude based on slip area (Hanks and Bakun, 2008). The magnitudes empirically inferred for both the Gorkha and the 1934 earthquakes based on their reported slip areas match their instrumentally inferred magnitudes (Mw 7.8 and Mw 8.4, respectively; Chen and Molnar, 1977; Sapkota et al., 2012; Avouac et al., 2015). This suggests that this method is appropriate, so we used it to examine the areas of discrete fault segments in our model. The area of the upper décollement south of the Gorkha earthquake (Fig. 4A) would correspond to Mw 8.0, much larger than any known earthquake there.
West of the Gorkha region, the fault area consisting of both the middle and upper décollements could produce an earthquake up to Mw 9.2, although the middle ramp and potential lateral break could serve as geometric rupture barriers, limiting earthquake size. Scarce historical records are not sufficient to constrain possible scenarios. In eastern Nepal, trenching studies have defined the extent of surface rupture for the 1255 and 1934 earthquakes (Fig. 4A; Nakata et al., 1998; Sapkota et al., 2012; Bollinger et al., 2014). Based on the reported extent of the 1255 earthquake, we can infer a magnitude up to Mw 8.7 for this event. In our fault model, there are no geometric barriers or pinch points in this region that could have limited earthquake rupture. The difference in lateral extent of the 1255 and 1934 earthquakes is a reminder that other types of slip barriers also limit rupture areas.
We produced a new cross section of the central Nepal Himalaya, using constraints from surface geology and the 2015 Gorkha earthquake. Our section requires a middle ramp, separating a middle and upper décollement, in addition to the deep ramp traditionally understood to be responsible for the growth of the Gorkha-Pokhara Anticlinorium. We extended the cross section into a 3-D model of the MHT for all of Nepal using surface geology. By comparing our model with an independent slip inversion data set, we show that variations in fault orientation at depth were likely responsible for limiting the size of the Gorkha earthquake.
Geometric boundaries such as bends and step-overs in map view have been interpreted to act as persistent barriers to rupture on strike-slip faults (Wesnousky, 2006), but for thrust faults, the role of subsurface geometry is less well known (King and Nabelek, 1985). For these faults, characteristics such as changes in fault zone properties or footwall or hanging-wall structure are thought to control the distribution and size of ruptures. Our results suggest that geometric changes are equally important in collision megathrusts. Detailed studies of the geometries of these convergent systems should therefore help better assess seismic hazard by identifying constraints on the sizes and locations of future earthquakes.
We thank the editor and the reviewers (Jean-Philippe Avouac, Richard Briggs, and two anonymous reviewers) for their thoughtful comments and suggestions, which helped us to improve the manuscript. This research was supported by the National Research Foundation (NRF) Singapore under its NRF Fellowship scheme (award no. NRF-NRFF2013–06) and by the EOS and the NRF Singapore and the Singapore Ministry of Education under the Research Centres of Excellence initiative. This work comprises Earth Observatory of Singapore (EOS) contribution 106.