(a) India-fixed GPS north velocities summarized by Vernant et al. (2014) and Kreemer et al. (2014) are coloured coded to indicate their southwards velocity relative to India. The blue line indicates the 8.5 mm a⁻¹ velocity contour which approximately follows the locking line at 18 km depth below the 3.5 km contour. The dashed green line is small-circle fit to this centred on a pole at 42.1° N and 90.72° E with a radius of 1642 km (Vernant et al. 2014) used in subsequent plots. (b) Deviation of GPS vectors (from arc normal) within the 1 radian quadrant dashed. The weighted least-squares fit to these (red line) is used to calculate the deviation from arc normal velocities depicted as arrows in (a). (c) Arc-normal velocities within the small-circle quadrant compared with synthetic velocities for a 6°–9° dipping dislocation locked at 18 km depth, and with convergence velocities of 16 and 18 mm a⁻¹ (red lines), and synthetic vertical displacements (green lines).

(a) Seismicity of Himalaya and Tibet. (b) Time–distance graph showing the rupture lengths of historical earthquakes since 1800 (violet), inferred ruptures (dashed violet) discussed in the text, and the increase in recorded earthquakes since 1900. Red are deeper than 40 km: blue are shallower. (c) Polar plot centred at 42.1° N, 90.72° E straightens the Himalaya and illustrates that the locking line closely follows a small circle. Recent earthquakes are dated and colour coded according to depth (scale as in a). Historical earthquakes and unruptured segments of the décollement are colour coded according to the legend below the figure. Incomplete ruptures leave a substantial fraction of the décollement unruptured (see Fig. 4b). (d) The same polar plot showing the radial width of the décollement as a function of distance along the arc. The lower edge of the shaded region represents the Main Frontal Thrust; the upper edge the locking line. The red lines indicate along-arc spatial averages with their numerical values.

Geometry of the slip in Himalayan earthquakes (section from Mencin et al. 2016). Updip blind thrusts from oil exploration seismic lines (Bashyal 1998). The square wave indicates locked, the bold line indicates rupture and the sine wave indicates aseismic slip. (a) The region of interseismic strain accumulation is the locus of microseismicity and occasional moderate earthquakes occurring as high-level thrusts. (b) Incomplete rupture in the 2015 Gorkha M_w 7.8 earthquake, with subsequent induced post-seismic creep (4 min–4 years). No slip occurred on the Main Frontal Thrust (MFT) but triggered surface slip was recorded on the Main Dun Thrust. InSAR (interferometric synthetic aperture radar) imagery shows that this occurred in the form of decaying creep for 26 km along strike with no slip below c. 5 km. The slip increased over a few weeks to ≥5 cm and then ceased (Elliott et al. 2016). (c) Complete rupture as inferred to occur in great earthquakes (e.g. in 1505 and 1950 M_w ≥ 8.6). Occasionally these may activate blind thrusts south of the MFT.

Location of the Qtub Minar and the Buddhist Kesariya Stupa relative to the nearby Himalayan earthquakes. The omission of the multifluted ornamentation of the lower three tiers in higher levels of the minar has been interpreted as a result of post-seismic 14th century reconstruction (Rajendran et al. 2018b). The Qtub is alleged to have lost its summit cupula in the 1803 earthquake, and the uppermost 7 m of the Stupa collapsed in the 1934 earthquake.

The location of sixteenth century earthquakes, and reported damage from the 6 June 1505 and 1555 earthquakes. White circles clustered near inferred rupture zones (shaded black) indicate conformed reports; open squares are accounts of dubious validity. Dated events with no rupture areas are the location of accounts from one location only with no indication of severity. The inset polar plots are shown for westwards (nucleation at 28.3° N, 83.5° E) or eastwards (nucleation at 30.5° N, 79.5° E) propagation for the 6 June 1505 8.7 < M_w < 8.9 earthquake, assuming it was a single event (see the text). The radial direction from each hypothesized nucleation point is represented by the log of the distance in kilometres from the hypothetical epicentre at the centre of the polar plot, and the azimuth indicates the proximity and direction of major cities and the (in this case, sparse) location of intensity observations. The grey area within the dashed red line represents the rupture area (from Table 2) using the same projection and permits a first-order evaluation of the possible influence of rupture propagation directivity, relative to reported observations of shaking intensity. That the earthquake was not reported in Kathmandu, where sediment amplification would have been pronounced, provides weak evidence that the earthquake ruptured westwards (see the text).

Maps of the Kashmir Valley. (a) Section of Montgomerie (1858) map showing villages of Nila and Dampur mentioned in one of several accounts of the 1555 earthquake (location a in d). The Nila location is also mentioned as venting methane in an account of the 1885 Kashmir earthquake. (b) Segment from 1858 map with a dashed box showing the location of the 2014 Google Image (c). The villages Hussainpur (Hu), Hossainpur (Ha) and Dampur (Wampur, Wa) near Arwin allegedly switched sides across the Vesha River in 1555. The Vesha floodplain is approximated by the dark forest cover. With the exception of a more pronounced meander near Tulkhan, the path of the Vesha River has changed little since 1858. The present village of Wanpora may correspond to the village of Dampur (Wampur) mentioned in Ferishta's account of the legend. (d) Location map. The village of Khadinyar near Baramulla lies close to the earthquake-induced landslide that in 844 CE dammed the Jhelum and flooded the valley up to Bijbehara (Bilham & Bali 2013). Bilham et al. (2013) described a minor normal fault near the summit of the Sintham Pass that slipped near this time (c. 700 CE ¹⁴C detrital date). The Sughandesa Temple near Patan was damaged in an earthquake in 1123 CE and again in 1885 (Bilham & Bali 2013).

Seventeenth and eighteenth century earthquakes. With one exception, the earthquakes are known from single accounts and thus permit no estimate of magnitude. The exception is the 1714 Bhutan M_w = 8.0 ± 0.5 earthquake (Hetényi et al. 2016) felt north and east of the epicentre, and associated with dated slip in two trenches (white circles).

Nineteenth century earthquakes. Approximate location of the 1803 rupture and its felt reports (white circles), and the location of all nineteenth century felt reports (1800–99, black dots) from Martin & Szeliga 2010. Approximate EMS intensity contours for the 1803 earthquake are shown.

The inferred epicentral region of the 1803 earthquake (dashed) is loosely drawn to encapsulate the EMS intensity 7 observations (red numbers) and named temples (violet italics) repaired after 1803 (Rajendran et al. 2013). The repair or ruin of temples to the SE of the depicted rupture is not specifically associated with the 1803 earthquake. The rupture areas of the Uttarkashi 1991 and Chamoli 1991 earthquakes are shaded blue, and their intensity ≥7 felt locations are shaded by orange and green squares, respectively. The conjectural west end of the 1505 rupture is outlined with light grey shading. The arrow depicts the local convergence rate and direction.

Polar log–distance plots showing the conjectural rupture zone and observed and residual EMS assessments for the 1803 Almora/Srinagar earthquake. The residual is the difference between the observed EMS value and that predicted by the Himalayan attenuation curve developed by Szeliga et al. (2010). SE rupture propagation is evident from the residual plot (right), and is consistent with their calculated epicentre near Gangotri and Uttarkashi (Fig. 10).

(a) A cartoon of sequential slip in the 1505, 1803, 1991 and 1999 earthquakes depicting the static slip residual resulting from contiguous or overlapping elliptical slip distributions. Assuming the 1991 and 199 earthquakes reduced the wedge-shaped grey slip residuals between the larger earthquakes, the hatched region shows considerable slip potential remains to which must be added the growing 1.5 m/century slip potential from the convergence rate. (b) The time sequence of the growing slip potential assuming zero strain after the 1505 earthquake at these longitudes. The region between the 1991 and 1999 earthquakes includes a slip deficit that will grow by c. 2100 to 4.5 m. The current slip deficit is sufficient to drive a 7 < M_w < 7.3 earthquake in the Almora/Srinagar/Tehri region.

Polar plot comparing the intensity distributions of the 1833 and 2015 earthquakes (from Mencin et al. 2016). The left-hand plot shows only those locations for which an intensity was evaluated both in 1833 and in 2015; whereas the right-hand plot numbers all the 1833 observations in black, and the 2015 observations in colour.

The crescent-shaped uncertainty in the location of the 1866 M_w ≈ 7 aftershock (Szeliga et al. 2010) surrounding Kathmandu results from the weak geometrical distribution of felt EMS intensities (numbered) mostly along the Ganges. The square wave depicts the locking line. If the 2015 Gorkha rupture area is adopted as a proxy for the 1833 rupture; the 12 May 2015 aftershock would be consistent with a recurrence of the 1866 earthquake.

The 1885 Kashmir earthquake and felt locations (solid circles) showing approximate isoseismal contours for EMS intensities 5, 6 and 7 surrounding Baramulla.

Uplift contours from one of several synthetic models consistent with horizontal triangulation changes before and after the 1897 Shillong rupture (England & Bilham 2015). Green boxes show (independent) observed vertical uplift relative to trigonometrical points on the southern margin of the plateau consistent with the synthetic model. Light blue to the SE and south of the rupture indicates transient lakes or the subsidence reported by Oldham (1899). Below the map is Oldham's (1899) photograph of the impounded (5 m deep) transient lake near Jhira, reproduced from Marr (1900). Red lines indicate faulting reported by Oldham (1899) with the sense of motion. To avoid running aground on shoals, a pilot was usually hired by captains steering shipping along the Brahmaputra between Dhubri and Gauhati. For many months after the earthquake this stretch of the river (light blue) was free from shoals and pilotage was not required.

Twentieth century felt reports (dots), major earthquakes and two great earthquakes.

Bounds of rupture for the Kangra 1905 earthquake (Table 6) shown by violet shading. Mean GPS convergence vectors (c. 15 mm a⁻¹) are indicated by arrows (from Fig. 2). The mainshock and principal aftershock are separated by c. 150 km (cf. Gorkha 2015 earthquake). Slip partitioning between the Kangra Valley Fault and the frontal thrust faults varies from 50 to 66% depending on the strike of the fold belts currently active (shaded range in vector summation is shown in the lower right).

EMS intensities (Martin & Szeliga 2010; Szeliga & Bilham 2017) for the 1905 Kangra earthquake and log polar plots (on the right with the rupture zone in grey outlined by a red dashed line) centred on Dharmsala. Residuals (lower right) from removing synthetic amplitudes predicted by Himalayan attenuation parameters (Szeliga et al. 2010) suggest amplification towards the SE and attenuation to the NW, consistent with directivity caused by nucleation near Dharmsala and propagation 125–150 km to the SE. High intensities near Dehra Dun have hitherto been interpreted to have resulted from a triggered deep earthquake near the base of the Indian crust (Hough et al. 2005b, Hough & Bilham 2008).

(a) Intensities observed in the 1934 Mw 8.4 rupture (from Ambraseys & Douglas 2004; Martin & Szeliga 2010) with nearby 1505, 1714 and 2015 rupture zones, dashed where inferred. The size of epicentral stars correspond to approximate mainshock location uncertainties. Various epicentral locations for the 1934 earthquake are indicated by letters: RD=Dunn et al. (1939), GR=Gutenberg and Richter (1949) which they rounded to the nearest half degree, IG=ISC/GEM v.5, CM=Chen & Molnar, 1977. White stars depict M > 5.7 aftershocks (ISC/GEM v.5) in the four years following the earthquake. Three 1934 scenario ruptures areas (a, b & c) are depicted (see the text). Large arrows show inferred rupture propagation directions. Dunn et al. (1939) propose that the mainshock was a major subevent, west of the nucleation phase, that occurred 11 s after the first arrival. Assuming a 2.6 km/s rupture propagation velocity its location would be c. 30 km west of the CM epicenter. Polar plots (below map) reveal apparent 1934 directivity to the SW related to rupture propagation in scenario a. Amplification in the Ganges sediments is partly responsible for positive residuals to the SW, but the absence of amplification in the SE quadrant at all distances is presumed to arise from directivity to the SW, consistent with scenario a and the location of the dominant subevent west of the nucleation event. (b) Scenes from the ‘slump belt’ of the Bihar province of India after the 1934 Bihar/Nepal earthquake. Sitamarhi (left at 26.58° N, 85.48° E), was long considered close to the epicentre of the 1934 earthquake. A typical fissure was 80 m long, 2.5 m wide and filled with sand to within 1m of the top (Dunn et al. 1939, p. 149, and p. 210). Extensive lateral spreading and sand venting occurred during liquefaction of sediments in the Ganges Plain south of the rupture.

(a) Estimated rupture zones and inferred propagation directions for the 1947 M_w = 7.7 and 1950 M_w = 8.7 earthquakes (the size of epicentral stars approximate the location uncertainties). Yellow and black stars are 1947 and 1950 mainshock locations from ISC/GEM v.5, and Chen and Molnar (1977) respectively. Red pentagons are the first 100 days of 1950 aftershocks from ISCGEM. Numerical EMS intensities and smoothed isoseismal contours are shown for the 1950 rupture. Felt intensities in the polar plots are consistent with westwards propagation in 1950, but high intensities are also influenced by sediment amplification in the Brahmaputra Valley. (b) Photographs of rail damage near Dibrugarh in the Brahmaputra Valley after the 1950 Assam earthquake photographed by M.C. Poddar (1950). (ii) is a close-up of the engine shed visible in the background of (i). Liquefaction sanding is visible in the foreground of (i) and is responsible for the subsidence beneath the tilted engine. A lateral spreading crack has severed the house in (iii).

View looking north at the 2005 earthquake fault scarp east of Muzafferabad. In the background can be seen some of the many thousands of landslides triggered by the earthquake. Shaking intensity was assessed as EMS 8–9 within 10 km of the surface rupture.

Kashmir M_w = 7.6, 8 October 2005: subsurface propagation (open arrows) and surface slip (solid arrows, largest surface slip = 8 m) from Avouac et al. (2006), with intensity distribution from Martin & Szeliga (2010). EMS intensities 7 and 8 are contoured with individual observations between EMS6 (green) and EMS 9 (red) indicated. The poorly constrained bounds of the 1555 earthquake (dashed) embrace possible slip of a rectangular steeply dipping patch SE of the 2005 rupture, or rupture beneath the Zanskar.

Polar plots for the Kashmir M_w = 7.6, 8 October 2015. Circles are logarithmically increasing distances from the hypocentre in kilometres. The grey area bordered by a dashed red line indicates the inferred subsurface rupture which ruptured bilaterally and updip. The solid red line indicates the surface rupture. Far-field intensities were amplified by 1–1.5 intensity units in the region of thick sedimentary cover in the Punjab and Ganges basins. Amplification was additionally enhanced by directivity effects during nucleation. Approximate isoseismal contours (EMS 4–7) are shown as dashed lines in map view.

2015 Gorkha rupture area and preceding contiguous earthquakes (from ISC/GEM v. 5). EMS-98 intensities are from Martin et al. 2015. Intensities above the rupture area average EMS 6.6. Intensities in the Terai south of the rupture average EMS 5, compared to intensities of 6–8 at similar distances from the 1934 rupture (Fig. 18). The polar plot shows anomalous low residuals above the rupture (shaded blue) predicted from the Himalayan attenuation parameters of Szeliga et al. (2010). Microseismicity during 1995–99 are shown as red dots. Palaeoseismic trenches are indicated by triangles (from Wesnousky et al. 2018).

Bedrock motions from the southern edge of the Gorkha rupture displayed as hodograms (map views of horizontal motions). (a) GPS positions sampled every 0.2 s at point KKN4 on the northern edge of the basin reveal the wholesale 1.5 m southwards displacement of the surface rocks of Nepal over the Indian Plate. Simultaneously, the basin rose 1 m more in the north than the south (Galetzka et al. 2015). The horizontal displacement is known as ‘fling’ because of its dynamic non-recoverable offset (in this case more than 1.5 m). (b) The GPS velocity at KKN4 is the time derivative of (a). (c) KKN4 acceleration, the time derivative of (b). (d) KTP acceleration measured by a strong motion accelerometer (Takai et al. 2016). (e) view of ridge amplification in northern Nepal. The distant village at lower elevation is shaken by lower shaking intensity.

The first 20 days of aftershocks (log of time since the mainshock) including the M_w = 7.3 aftershock show a general eastwards migration of seismic activity. The xx events symbolize the numerous radiating phases that occurred during eastward propagation along the northern edge of the locked décollement in the first two minutes of the rupture. Five aftershocks ≥M_w = 4 occur in the first 14 min (0.01 days). More than 553 M_w > 4 earthquakes occurred within 45 days of the mainshock.

Historical earthquakes (red) and dated palaeoseismic ruptures of Himalayan Frontal Thrust faults (shown by error bars). Least-squares fits to the east and west Himalaya are indicated with weighted deviations proportional to the range of reported dates at each site. Blue lines are weighted least-squares fit representing a hypothetical west-propagating rupture front with a velocity of 7.5 ± 5 km a⁻¹ starting in Assam at c. 1100 and arriving at Kashmir >1400 (dates that ignore systematic detrital ¹⁴C errors). Grey dashed bars indicate synchronous slip in single earthquakes with surface ruptures exceeding 200 km proposed variously by Mugnier et al. (2013), Rajendran et al. (2018b) and Wesnousky et al. (2019). A, Arora & Malik (2017), Malik & Nakata (2003); Malik et al. (2003, 2008, 2010a, b); B, Kumahara & Jayangondaperumal (2013); C, Kumar et al. (2006); D, Malik et al. (2017); E, Rajendran et al. (2015, 2018b); F, Yule et al. (2006); G, Hossler et al. (2016); H, Mugnier et al. (1998, 2013); J, Murphy et al. (2013); K, Wesnousky et al. (2017b); L, Lave et al. (2005); M, Sapkota et al. (2013); N, Upreti et al. (2000, 2007); P, Kumar et al. (2010); Q, Le Roux-Mallouf et al. (2016); R, Jayangondaperumal et al. (2011); S, Priyanka et al. (2017); T, Wesnousky et al. (2018).

Schematic descent of the Indian Plate beneath Tibet showing the volume of Himalaya subject to 10 µstrain compression (200 years at 17 mm a⁻¹) for seismic-decoupling transition widths of 3–60 km. The tip of the 3 km-wide decoupling zone reaches 100 µstrain in 200 years (small black area). A 5 km wide shaded area can store elastic strain approaching failure in a much larger region (shaded). A second shaded area extending 60 km down-dip can store significantly more strain before failure (dashed contours show cumulative strain ampliitude after 200 years). Wider decoupling zones take longer to reach 100 µstrain failure levels, but can drive correspondingly larger earthquakes due to their longer downdip widths. The steep reverse fault represents the preferred path for faulting in the presence of a fully locked décollement (c.f. Fig. 4a).

Seismic slip deficit assuming a mean convergence rate of 15 mm a⁻¹ for the Himalaya. The dashed line indicates the theoretical moment release; the solid incremental lines represent the observed rate taking the high and low estimate for historical magnitudes. The divergence of the synthetic and dashed lines imply that one or two M_w ≥ 8.6 earthquakes, or an equivalent moment release contributed by several M_w > 8 earthquakes, are missing and must occur in the future.

Seasonal occurrences of historical and recent Himalayan earthquakes. (a) The declustered time history of Himalayan earthquakes 1904–2018 from the ComCat USGS catalog, and is plotted east (https://earthquake.usgs.gov/data/comcat/). Earthquakes east of 78°E are shaded red, and earthquakes to the west are shaded green. (b) map view with same shading. Blue stars depict the approximate locations of historical earthquakes (Table 1). (b) map view indicating locations of Himalayan earthquakes coloured in (a) and (c). (c) Monthly counts of significant pre-1900 earthquakes and post 1900 M_w ≥ 5.5 earthquakes. Due to the westwards progression of the monsoon, earthquakes in the Western Himalaya should respond later than the Eastern Himalaya; however, this effect and the direct influence of the monsoon on the timing of M_w ≥ 5.5 earthquakes appears to be insignificant contrary to claims by Panda et al. (2018). In each panel in (c), bars indicate cumulative earthquakes per month; east (red) and west (green), and the sum of these two – total counts (blue historical; black 1904–2018).

Cumulative fatalities from Himalayan earthquakes 1800–2018. None of these recent earthquakes have occurred at night, although the Kangra earthquake occurred at 6 am. It is likely that should a nocturnal earthquake occur, especially close to a large population such as Dehra Dun or Kashmir or Kathmandu, the death toll may exceed 100 000. Wyss (2005, 2017) estimated maximum death tolls of twice this for some scenario earthquakes.

(a) Temple at Saharn, Sutlej Valley with interleaved stone and timber (photographed by Carl Griesbach in 1883). (b) A Koti Banal dwelling in Manikarn, Kulu that survived the 1905 M_w = 7.8 Kangra and earlier earthquakes (photographed by Charles Middlemiss in 1905). Some Koti dwellings succumbed to a 1906 M_w 6.5 aftershock (Szeliga & Bilham 2017). Rautela & Joshi (2008) speculated that this style of architecture was introduced more than 800 years ago as a response to a widespread damaging Himalayan earthquake, and that some of the original structures still remain.