Triggering Mechanisms of Gayari Avalanche, Pakistan


 A massive snow avalanche occurred on April, 2012 at Gayari, located in NE part of Pakistan, close to India and China Border. The catastrophic avalanche killed nearly 148 people, majority of which were Pakistan army personnel destroying army base camp. To mitigate its future hazard, different triggering mechanisms have been investigated in this study. We contemplate that the avalanche was triggered due to snow pack existence on favorable slope in combination with different meteorological conditions and anomalous ground vibration. The avalanche occurrence clock was advanced by two earthquakes: M4.1 at a distance ∼ 125 km that occurred about 21 hours before and another comparatively larger (M5.6) earthquake that occurred comparatively at larger distance (∼ 370 km) and longer time (∼ 25 days) before which have significantly changed the loading conditions. The latter event (M 5.6) has imparted maximum peak dynamic stress and cumulative seismic moment a month before the avalanche. Interestingly the avalanche occurred within the seismic coda of M2.8 earthquake from Hindu Kush region, located at 560 km distance. Although the size and its expected impact on avalanche might be minor but its role in instantaneous triggering cannot be ruled out. Even smaller events at larger distance have been reported to cause snow avalanches in same environments. The presence of cracks within the avalanche, were further weaken by persistence of extremely low temperature (lowest in the past decade), causing high precipitation rate along with altering the mechanical properties of the weak layer within the snow pack. Robust wind pressure pattern highest and lowest in March and April, 2012 respectively might be responsible for abrupt changes in loading conditions.

some observations showed delay in time between earthquake shaking and avalanche triggering. This is due to indirect 44 triggering, earthquake activated process such as change pore pressure, stress weakening or development of cracks. Thus, 45 evaluating the snow slope failure requires a complex system of variables, compose a number of different factors, with 46 scales variation in time and space (Laternser and Schneebeli, 2003;Schweizer, 2008). 47 Snow catalog is normally based on visual observations or news reports, after occurrence of the avalanche. These 48 catalogs are incomplete and have poor resolution in time and space (Laternser and Schneebeli, 2002 and Inverse Slope. Both methods were used to find the representative slope for a specified number of data points. The 99 slopes are then used in combination with their associated correlation values to determine an infrasonic signal. Processing 100 infrasound array data using frequency slowness analysis gives us three outputs per processing step: correlation, azimuth 101 and trace velocity (or slowness).

102
DEM data is considered as the primary input for glacier avalanche studies. High-resolution Global-DEM 30-m 103 horizontal resolution were used to generated from the ASTER satellite covering the whole gayari glacier (NASA, 2020 104 (accessed October 11, 2020; DAAC, 2015), as show in Figure 1. We analyze different metreological conditions such as; 105 temperature, precipitation and wind speed data of the region to find out any abnormal behavior, we analyze climate data 106 from different sources. The historical spatial distributed data sets of Climate research unit (CRU) also provided in by the 107 world bank organization were used to find out variations in temperature and rainfall at different latitudes and longitudes 108 within the region. This data is monthly averaged for the period 1991-2016 (Harris and Jones, 2017; CRU, 2016 (accessed 109 November 10, 2020). The other data we analyze is the gridded data sets of temperature, precipitation and wind speed 110 downloaded from National data center, he other data we analyze is the gridded data sets of temperature, precipitation 111 and wind speed downloaded from National data center, Physical sciences labratory data bank where different types of 112 gridded data sets of climate variables are avaliable in netCDF format The temperature and the precipitation data is of 113 0.5 • and 0.5 • grid of CPC global temperature data sets and GPCC global precipitation data set (GPCC, 2016 (accessed 114 November 10, 2020; CPC, 2020 (accessed November 05, 2020) and wind data sets is of 2.5 • and 2.5 • grid from NCEP 115 reanalysis of 40 years of climate variables (Kalnay et al, 1996), this data covers daily, monthly mean and annual variation 116 in the climates variables within the grid. We take several points in the region to find out variations of climate variables.

118
The location of Gayari snow avalanche is in high terrain and difficult to access, along with close to boundary of India, 119 China in extreme NE of Pakistan as show in Figure 1. The rescue operation was difficult due to harsh weather conditions.

120
The avalanche signal on infrasound array located (375 km away from avalanche) in Islamabad was clearly recorded 121 (Figure 2, 3). The quality of signal was very good and clear as at midnight there was no air due to convection process and 122 no microbaroms. Only that part of signal with co-relation co-efficients greater than 80% on array components have been  The velocities difference (∼ 4.0 km/s) between event and avalanche corresponded to the expected avalanche occurrence 137 time ∼ 25 sec late as that of first arrival from the event. Avalanche is a moving and extended source but unfortunately 138 our seismic station are away from source and was not able to capture its moving velocity. The recorded velocity of an 139 avalanche might represent body wave velocity that generated after the interaction between avalanche and earth. Compare  Comparing CE14 and CE11 stations respectively to close (351 km) and away (462 km) from avalanche and vice versa 152 in case of Hindu Kush 2.8 Ml earthquake, as shown in Figure 5. In case of CE14 station, avalanche within the coda of 153 earthquake, whereas at that of CE11 both are separated ∼ 115 sec from each other. 154 We located all events within 24 hours (12 hours before and after) of the avalanche time, with clear P−, S− arrivals, 155 good signal to noise ratio and recorded on minimum three stations. Because of some small events were missing in our 156 catalog that were further added. Majority of these events were located in Hindu Kush region (∼ 550 km away from 6 Bilal et al avalanche location), whereas couple of events located within 200 km from avalanche ( Figure 6). Average event occur-158 rence rate before (0.75 events/hour) were higher than that of after (0.3 events/hour) i.e. 22.5%. A significant accelerated 159 rate that is, events rate during the occurrence of avalanche time is respectively 62.5% and 85% higher than that of av-160 erage event rate and after the avalanche occurrence. This increase in earthquake rate start nearly three hours before the 161 avalanche time (0.7 and 2 events/hour earthquake rate exist 6 hours and 3 before the avalanche time). We extend the 162 seismicity pattern from time period of 01 day to 01 year before and after the avalanche time. First the threshold magni-163 tude or completeness magnitude for the catalog was calculated as 3.5 M L (local magnitude) using maximum curvature 164 (Wiemer and Katsumata, 1999) and Ogata technique (Ogata, 1983). The slope of frequency magnitude, which is b-value 165 of Gutenberg Richter law (Gutenberg and Richter, 1954) were close to 1, depict same ratio of large to small events pro- significantly (µ -1σ , with µ is mean, σ is standard deviation) lower than that of previous one and winter 2012 was 208 coldest in the Gayari region. Similarly precipitation in first six months were higher (µ+1σ ) than that of average one. 209 We analyze wind speed data monthly and daily basis, and then convert that wind speed into pressure using Bernoulli's 210 equation (Kalnay et al, 1996). In March 2012 the wind pressure was significantly high than the average wind speed. But 211 in April 2012 it was opposite, i.e. wind pressure was very low than the average value (Figure 15, 16). 212 We also analyze the daily hourly wind speed data of uwind and vwind components of wind speed for year 2012.

265
The three month avalanche delay can be acquired from small k value (i.e., k=0.005) with t r and l e have their maximum 266 values.

267
Lower the stress ratio (k = τ g −τ r τ p0 −τ g , τ g = ρghsinα) value will slow the shear fracture development, thus yield longer 268 time delay for avalanche triggering and vice versa. For accelerating fracture growth the gravitational shear stress (τ g ) 269 should be larger than residual shear stress (τ r ) but smaller than the peak strength (τ p0 ). For mild slope or lower α value 270 the τ g is lower and longer time required for avalanche to be triggered. Avalanche triggering will be alter by density and       Fig. 7 Cumulative number of events, 12 hour and after avalanche. These events were manually picked and located, using minimum three stations. All these events were from Hindu Kush region. Three hours before the avalanche seismicity rate was significantly high (2 events/hour) as compare to average seismicity rate (0.7 events/hour). After Avalanche hourly rate 0.3 events/hour, which is lower than that of average of rate.    Fig. 12 Similar as that of Figure 11 but daily rate calculated on basis of events occurrence before and after avalanche. Maximum earthquake rate with clear aftershock decay observed for event of magnitude 5.6, occured 25 days before avalanche. Horizontal dotted line show average or reference seismicity of the region. Inset figure at top show decay of aftershock sequence, avalanche was occurred within the average or reference seismicity rather than relaxtion phase of aftershock sequence.