Snow avalanche speed determination using seismic methods

Abstract We present a new method to determine the average propagation speed of avalanches using seismic techniques. Avalanche propagation speeds can reach 70 m/s and more, depending on a wide range of factors, such as the characteristics of the avalanche track (e.g. topography) and the snowpack properties (e.g. density). Since the damage produced by the avalanche depends primarily on the size and on the speed of the avalanche, the knowledge of the latter is therefore crucial for estimating avalanche induced hazard in inhabited mountain areas. However, our knowledge of this basic physical parameter is limited by the difficulty of conducting various measurements in the harsh winter weather conditions that often accompany this natural phenomenon. The method of avalanche speed determination presented in this paper is based on cross-correlation and time-frequency analysis techniques. The data used in this study come from the Ryggfonn (Norway) avalanche experimental site operated by the Norwegian Geotechnical Institute (NGI), and recorded by an array of 6 geophones buried along the main avalanche path during the 2003-2004 and 2004-2005 winter seasons. Specifically, we examine the speeds of 11 different events, characterized by size and snow type. The results obtained are compared with independent speed estimates from CW-radar and pressure plate measurements. As a result of these comparisons our method was validated and has proved to be successful and robust in all cases. We detected a systematic behaviour in the speed evolution among different types of avalanches. Specifically, we found that whereas dry/mixed type flow events display a complex type of speed evolution in the study area with a gradual acceleration and an abrupt deceleration, the speed of the wet snow avalanches decreases with distance in an approximately linear fashion. This generalization holds for different size events. In terms of time duration and maximum speed of the studied events, dry/mixed type avalanches lasted between 8 to 18 s and reached speeds up to 50 m/s, whereas the duration of wet avalanches ranged between 50 and 80 s and their maximum speeds were 10 m/s.


U N C O R R E C T E D P R O O F
41 experiments are compared with those obtained by the 42 models in order to calibrate them. 43 Avalanche speeds depend on a wide range of factors, 44 including the characteristics of the track (e.g. topogra-45 phy) and of snowpack properties (e.g. grain/clod size 46 and density). Avalanches can reach speeds up to 70 m/s 47 and more. Since the damage produced by the avalanche 48 depends on its propagation speed, the knowledge of this 49 parameter is crucial for estimating avalanche induced 50 hazards in inhabited mountain areas. However, our 51 knowledge of this basic physical parameter is limited by 52 the difficulty of conducting measurements, which is 53 compounded by the harsh weather conditions that often 54 accompany snow avalanches and by the complexity of 55 the physical phenomenon itself. 56 Earlier studies of speed determination of snow 57 avalanches have been based on the processing of 58 video images (Granada et al., 1995;McElwaine, 59 2004), the determination of internal clod speeds using 60 LED-photocells (Dent et al., 1998), on the interpretation 61 of data from pressure load cells (Norem et al., 1985), 62 and Doppler-radar (Gubler, 1987) techniques among 63 others. The few studies of avalanche speed determina-64 tion based on seismic methods include works by 65 Schaerer and Salway (1980), Nishimura et al. (1993), 66 and Nishimura and Izumi (1997), who used basic 67 picking techniques to obtain the arrival time of the 68 avalanche body over geophones installed along the 69 avalanche path. However, none of these studies describe 70 in detail the criteria of the selection techniques used. 71 In this paper we present a new method for determining 72 the average propagation speed of snow avalanches from Plates (LP) measurements .

83
The methods for determining the avalanche speed 84 presented in this paper are based on cross-correlation

125
A shelter containing the control and recording 126 instruments is located in the valley 500 m east of the 127 dam, and is provided with power, telephone line and 128 ISDN connection. The recording system is triggered by 129 geophone GF1 when the avalanche front hits the steel 130 tower. All the measurements from the different instru-131 ments are recorded at a sampling rate of 150 sps with a 132 local common base of time. The total length of the 133 records is 150 s including 25 s of pre-triggered data.

137
In this study we analysed the data from 11 avalanches 138 that occurred during the winter seasons 2003-2004 and 139 2004-2005 and one from 2000 (Table 1). The ava-140 lanches differed in size and type of snow. All the 141 avalanches with the exception of events 6 and 7 were 142 triggered naturally. Unfortunately, the availability of 143 complementary information including field observa-144 tions and video images is limited. Seismic and load plate 145 data analyses indicated that in avalanche 2 the dense 146 body did not reach the dam although the aerosol passed 147 over the dam. Avalanche 4 did not reach the dam (field 180 the spectra is performed as presented in Fig. 2b. We 181 attribute this time interval to the arrival of the avalanche 182 front over the sensor. The increase in the amplitude in 183 the RS with time is produced by the reduction of the 184 distance source-receiver as the avalanche front 185 approaches the sensor (Suriñach et al., 2005). This is 186 physically supported by the anelastic attenuation of 187 seismic waves with distance (Aki and Richards, 1980),  (Biescas, 2003;Biescas et al., 2002).

197
The selection of the window is performed by means

U N C O R R E C T E D P R O O F
204 and frequency resolutions (Flandrin, 2002); and 2) 205 selection of the first time window showing maximum 206 amplitudes at the highest frequencies in the RS (Fig. 2b).

207
The arrival time in the seismic time series is 208 performed by picking a discontinuity in the amplitude 209 or/and frequency of the time series in the selected 210 window using the standard seismological technique 211 (PK) (Fig. 2c). However, there are no characteristic 212 features to identify the arrival of the front.

213
As a result of this procedure the arrival time of the 214 avalanche front at each geophone is obtained. Subse-215 quently, using the distance between the geophones the 216 average propagation speed of the avalanche between the 217 pairs of geophones is obtained. When the PK procedure 218 is not easy due to the lack of clarity of the wave arrival 219 as in the case of small avalanches which generate weak 220 vibrations or in the case of signals produced by slow 221 dense avalanches recorded in the sensors located in the 222 dam (GF4 and GF5) (further discussion is presented 223 below) an alternative, cross-correlation procedure (XC) 224 is applied.

225
The XC procedure consists of determining the lag 226 time between the previously selected seismic time series 227 windows corresponding to two separate geophones. The 228 lag time, which indicates the time shift between the two 229 time series, yields the difference in arrival time of the 230 avalanche front needed to obtain the average propagation 231 speed of the avalanche between the two points (Fig. 3).

232
The XC method assumes that signals generated by the 233 same source must be comparable. Earlier studies 234 demonstrate that the main sources of the seismic signals 235 are snow erosion, changes in the slope and impacts with 236 obstacles (Suriñach et al., 2001). In our case, the signals 237 to be identified are mostly produced by snow erosion, 238 which is assumed to occur mainly near the front of the 239 avalanche (Gauer and Issler, 2004).

240
The cross-correlation was performed on a windowed The XC method involves pairs of seismic time series.

263
As a consequence, we were able to obtain the differences Fig. 4. Comparison of the speeds of 11 avalanches released at Ryggfonn obtained from the data of various instruments (CW-radar, geophones, load plates). Continuous lines shows the path profile and dots indicate the position of the measuring instruments. The speeds are presented at the midpoint between adjacent instrument locations.

U N C O R R E C T E D P R O O F
264 between the arrival times for four pairs of sensors (GF1-265 GF2, GF2-GF3, GF3-GF4 and GF4-GF5). This is not the 266 case with the PK method with which we obtain an 267 estimate of the avalanche arrival time at each geophone. 268 Avalanche average speed estimates are obtained using the 269 distances involved. Data from GF-6, located on top of the 270 dam (Fig. 1) were of no use because of the spatial 271 distribution of the geophones and the ground character-272 istics of the dam (formed by loose gravel with poor 273 vibration transmission). Using the methods described in the previous section 276 (PK and XC) we estimated the speeds for the 11 ava-277 lanches at Ryggfonn listed in Table 1. The estimates of the 278 average avalanche propagation speeds using both methods 279 are shown in Fig. 4. To assess the reliability of our me-280 thods, we compared our results with the speed measure-281 ments from a CW-Doppler radar (Sigurðsson et al., 2004) 282 and with estimates from the arrival time at the load plates 283 . In Fig. 4 the average avalanche 284 propagation speeds are represented as a function of the 285 horizontal distance, the steel tower being the origin of the 286 distances. The average avalanche propagation speed 287 between two adjacent geophones is depicted midway 288 between the two sensors to show the evolution of the 289 avalanche along the path. The positions of the sensors are 290 indicated by dots on the path profile in Fig. 4. Averaged 291 speeds obtained from the load plates (steel tower, concrete 292 structure and dam) are plotted midway between the load 293 plates, and speeds from the radar (obtained by averaging 294 the corresponding values from Sigurðsson et al. (2004)) 295 are also indicated in Fig. 4.

296
In general, there is a reasonable agreement between 297 the different speed estimates. However, the different 298 sensitivity of the different types of equipments used to 299 detect the avalanche, which is reflected in the results, 300 should be pointed out. As regards the LP sensors, the 301 values are obtained from the arrival time of the 302 avalanche at the load cells. This determination is usually 303 easy and unambiguous because of a sudden increase in 304 pressure observed in the record of the LP. Nevertheless, 305 although the strong peaks which are probably produced 306 by the saltation layer or by the dense part of the 307 avalanche are easy to detect, it is much harder to detect 308 the signals generated by the small powder part, which in 309 some cases precedes the dense core. The impacts 310 produced by the saltation layer on the LP are of slightly 311 lower amplitude and higher frequency than that 312 produced by the dense part. The impacts of the powder 313 part are only detected when the LP is not covered by 314 snow. As for the radar, it is also difficult to detect the 315 powder part of small avalanches because of the poor 316 reflectivity of the powder cloud of a low density.

317
Moreover, the estimated speeds obtained by radar are 318 not directly comparable to the front speeds determined 319 by the other methods given that the radar signal is 320 composed of the reflections from many parts of the 321 avalanche body. This is a consequence of the wide angle 322 range from which the radar data are obtained (Gubler, 323 1987).

324
As regards the seismic method, previous studies 325 allow us to determine the origin (source) of the seismic 326 signals associated with snow avalanches. It has been 327 confirmed that the powder part is detected despite the 328 low signal amplitude (Suriñach et al., 2001). One of the 329 problems of using seismic sensors to study avalanches is 330 that the origin of the signal, unlike earthquakes is not

366
As regards the load plates, the speed values are in 367 general consistent with those obtained from the seismic 368 data. The original measurements obtained by radar yield 369 the avalanche speed as a function of time in intervals 370 ranging from 1.5 to 4.5 s. In order to assign a horizontal 371 distance to the radar speed values between two given 372 sensors, these speeds were averaged between the arrival 373 times of the avalanche front over the sensors. These 374 arrival times were extracted from the RS calculations 375 and LP data analysis. The speed values from the radar 376 show differences in relation to the LP and geophones 377 measurements. This was expected given that the 378 geophones and the load plates measure the front 379 speed, whereas the radar records a composite signal 380 proceeding from all the moving parts of the avalanche 381 within the range of the radar. We believed that the 382 averaged speeds correspond to two different parts of the 383 avalanche. Whereas the high speed values from the 384 radar at approx. 100 m in event 1 (Fig. 4) correspond to 385 the aerosol front of the avalanche (simultaneous 386 pressure measurements show low values (Sigurðsson 387 et al., 2004)), the lower values of the speed obtained 388 from the load plates and seismic sensors correspond to 389 the dense body. In cases where the dense and aerosol 390 parts coexist, geophones detect the former preferentially 391 because of the higher vibrations produced; the same 392 behaviour is observed for the load cells. This is the case 393 for event 2, which is a dry/mixed avalanche with a large 394 aerosol part and a reduced dense body.

395
Two different distance evolution profiles of the speed 396 along the path are observed in Fig. 4, which corresponds 397 to the two distinct types of avalanche (dry/mixed and 398 wet). Interestingly, the character of the profile is 399 relatively independent of the size of the avalanche. 400 Dry/mixed avalanches have higher speeds than wet 401 avalanches. If we consider the slope profile in Fig. 4 as a 402 reference curve, all speed values for dry/mixed 403 avalanches are above this curve, whereas the speeds of 404 the wet avalanches are below the curve. Dry/mixed 405 avalanches even seem to accelerate along this part of the 406 path in some cases: The average seismic speed values 407 for the GF2-GF3 interval appear slightly higher than for 408 the GF1-GF2 interval for events 1 and 4, although this is 409 not the case for the radar derived speed values for these 410 events nor for the LP derived speeds for event 1. Nor is 411 this the case for the wet avalanches where an 412 approximately linear decrease in velocity with distance 413 is observed. The speed values deduced from load plates 414 are also consistent with the observations. Note that the 415 velocity profile of event 7, which corresponds to the 416 largest dry/mixed avalanche, has a shape similar to that 417 of the other dry/mixed avalanches, but with higher