9
Forecasting articially-triggered avalanches in storm snow at a large ski area Edward H. Bair US Army Corps of Engineers Cold Regions Research and Engineering Laboratory, Hanover, NH, United States abstract article info Article history: Received 7 February 2012 Accepted 4 October 2012 Keywords: Snow Avalanche Ski area At ski areas, a majority of avalanches fail in storm snow. Using thousands of observations from avalanche control work at Mammoth Mountain, CA, USA, a large coastal ski area, I analyzed important predictors of avalanche ac- tivity. New (24 h) precipitation increased avalanche activity, while changing temperatures and different wind patterns had no effect. If slopes remained undisturbed for one day after snowfall, the number and size of ava- lanches as well as the explosive yield (avalanches per shot) were all signicantly reduced. I also examined a smaller dataset of Extended Column Test (ECT) results and their relation to avalanche activity. ECT propagation was a powerful predictor; days with ECTs that propagated had signicantly more avalanches and larger sizes. Days with propagating ECTs also had signicantly greater new snow amounts, with a threshold value of 0.29 m of new snow, very close to the 0.31 m threshold from Atwater's 10 factors. That new precipitation above a threshold causes greater avalanche activity is not a new nding; the new nding is that ECT propagation (versus non-propagation) also has a similar new snow threshold. Thus, I suggest that ECT propagation is an important tool to predict explosively-triggered avalanches in storm snow. © 2012 Elsevier B.V. All rights reserved. 1. Introduction At ski areas in North America that record failure layers, a majority of avalanches are estimated to fail in storm snow (Bair, 2011; Stethem and Perla, 1980; Williams and Armstrong, 1998). Storm snow crystals are called nonpersistent because they metamorphose into rounded or faceted forms within a few days (Jamieson, 1995). Older faceted crys- tals are called persistent because, once buried, they can remain in the snowpack for weeks or months. Because they are often deeper and more destructive, avalanches that fail on persistent crystals have re- ceived signicant study, while those on nonpersistent crystals have received little study. Despite the lack of research, avalanches that involve only storm snow are a signicant hazard. For instance, the worst avalanche accident at a North American ski area was caused by an avalanche that only involved storm snow. The accident oc- curred on 31 Mar 1982 and killed seven people at Alpine Meadows, CA (Heywood, 1992). The storm snow accumulation at the time of the accident was 2.2 m on top of a well bonded melt-freeze crust. Crown heights were 23m(Penniman, 1986). Strong winds, averag- ing up to 39 m s 1 , easily account for the added wind load. 1.1. Avalanche forecasting Conventional avalanche forecasting has been performed by experi- enced avalanche professionals using a variety of measurements and information sources to guide decisions, which are largely based on em- piricism and intuition (LaChapelle, 1980). Various attempts at statistical (Buser, 1983; Buser et al., 1985; Davis et al., 1999; Gassner and Brabec, 2002; Heierli et al., 2004; McCollister, 2004; Purves et al., 2003; Roch, 1966) and physically based (Bartelt and Lehning, 2002; Casson, 2009; Conway and Wilbour, 1999; Föhn, 1987; Gauthier et al., 2010; Hayes et al., 2004; Jamieson, 1995; Jamieson and Johnston, 1993, 1998; Zeidler, 2004) stability models have been made, but have not gained widespread acceptance in the avalanche community. With increased computing power, spatially explicit (Durand et al., 1999; Hirashima et al., 2008; Pozdnoukhov et al., 2008; Rousselout et al., 2010; Schirmer et al., 2009, 2010) models have emerged, but have not proven capable of providing better guidance than conventional forecasting methods. Further, verica- tion is difcult since avalanche activity is often not observed or reported and depends on triggering factors such as skier trafc. Avalanche hazard forecasts (Brown and Jamieson, 2008; Elder and Armstrong, 1987; Schweizer et al., 2003) and models (Durand et al., 1999; Rousselout et al., 2010; Schirmer et al., 2009, 2010) have been veried with a posteriori hazard estimates, snow pit data and stability tests, or a limited number of avalanche occurrences. Avalanche hazard forecasting largely depends on scale and type of area. For example, for the same absolute hazard level, a backcountry area will have far fewer avalanche occurrences than a ski area with extensive avalanche control measures. Many stability models were developed for backcountry forecast areas, so their accuracy cannot be directly compared to models specically developed for ski areas or other areas with extensive avalanche control measures. At many ski areas, one will nd snow safety professionals still implicitly using Atwater's 10 contributory factors (Atwater, 1954; Atwater and Koziol, 1953): 1) old snow depth, 2) old snow surface, Cold Regions Science and Technology 85 (2013) 261269 Earth Research Institute, University of California, Santa Barbara, United States. Tel.: +1 11 805 893 4885. E-mail address: [email protected]. 0165-232X/$ see front matter © 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.coldregions.2012.10.003 Contents lists available at SciVerse ScienceDirect Cold Regions Science and Technology journal homepage: www.elsevier.com/locate/coldregions

Forecasting artificially-triggered avalanches in storm snow at a large

Embed Size (px)

Citation preview

Page 1: Forecasting artificially-triggered avalanches in storm snow at a large

Cold Regions Science and Technology 85 (2013) 261–269

Contents lists available at SciVerse ScienceDirect

Cold Regions Science and Technology

j ourna l homepage: www.e lsev ie r .com/ locate /co ldreg ions

Forecasting artificially-triggered avalanches in storm snow at a large ski area

Edward H. Bair ⁎US Army Corps of Engineers Cold Regions Research and Engineering Laboratory, Hanover, NH, United States

⁎ Earth Research Institute, University of California, San+1 11 805 893 4885.

E-mail address: [email protected].

0165-232X/$ – see front matter © 2012 Elsevier B.V. Alhttp://dx.doi.org/10.1016/j.coldregions.2012.10.003

a b s t r a c t

a r t i c l e i n f o

Article history:Received 7 February 2012Accepted 4 October 2012

Keywords:SnowAvalancheSki area

At ski areas, a majority of avalanches fail in storm snow. Using thousands of observations from avalanche controlwork at Mammoth Mountain, CA, USA, a large coastal ski area, I analyzed important predictors of avalanche ac-tivity. New (24 h) precipitation increased avalanche activity, while changing temperatures and different windpatterns had no effect. If slopes remained undisturbed for one day after snowfall, the number and size of ava-lanches as well as the explosive yield (avalanches per shot) were all significantly reduced. I also examined asmaller dataset of Extended Column Test (ECT) results and their relation to avalanche activity. ECT propagationwas a powerful predictor; days with ECTs that propagated had significantly more avalanches and larger sizes.Days with propagating ECTs also had significantly greater new snow amounts, with a threshold value of0.29 m of new snow, very close to the 0.31 m threshold from Atwater's 10 factors. That new precipitationabove a threshold causes greater avalanche activity is not a new finding; the new finding is that ECT propagation(versus non-propagation) also has a similar new snow threshold. Thus, I suggest that ECT propagation is animportant tool to predict explosively-triggered avalanches in storm snow.

© 2012 Elsevier B.V. All rights reserved.

1. Introduction

At ski areas in North America that record failure layers, a majorityof avalanches are estimated to fail in storm snow (Bair, 2011; Stethemand Perla, 1980; Williams and Armstrong, 1998). Storm snow crystalsare called nonpersistent because they metamorphose into rounded orfaceted forms within a few days (Jamieson, 1995). Older faceted crys-tals are called persistent because, once buried, they can remain in thesnowpack for weeks or months. Because they are often deeper andmore destructive, avalanches that fail on persistent crystals have re-ceived significant study, while those on nonpersistent crystals havereceived little study. Despite the lack of research, avalanches thatinvolve only storm snow are a significant hazard. For instance, theworst avalanche accident at a North American ski area was causedby an avalanche that only involved storm snow. The accident oc-curred on 31 Mar 1982 and killed seven people at Alpine Meadows,CA (Heywood, 1992). The storm snow accumulation at the time ofthe accident was 2.2 m on top of a well bonded melt-freeze crust.Crown heights were 2–3 m (Penniman, 1986). Strong winds, averag-ing up to 39 m s−1, easily account for the added wind load.

1.1. Avalanche forecasting

Conventional avalanche forecasting has been performed by experi-enced avalanche professionals using a variety of measurements and

ta Barbara, United States. Tel.:

l rights reserved.

information sources to guide decisions, which are largely based on em-piricism and intuition (LaChapelle, 1980). Various attempts at statistical(Buser, 1983; Buser et al., 1985; Davis et al., 1999; Gassner and Brabec,2002; Heierli et al., 2004; McCollister, 2004; Purves et al., 2003; Roch,1966) and physically based (Bartelt and Lehning, 2002; Casson, 2009;Conway and Wilbour, 1999; Föhn, 1987; Gauthier et al., 2010; Hayes etal., 2004; Jamieson, 1995; Jamieson and Johnston, 1993, 1998; Zeidler,2004) stability models have beenmade, but have not gainedwidespreadacceptance in the avalanche community. With increased computingpower, spatially explicit (Durand et al., 1999; Hirashima et al., 2008;Pozdnoukhov et al., 2008; Rousselout et al., 2010; Schirmer et al., 2009,2010) models have emerged, but have not proven capable of providingbetter guidance than conventional forecastingmethods. Further, verifica-tion is difficult since avalanche activity is often not observed or reportedand depends on triggering factors such as skier traffic. Avalanche hazardforecasts (Brown and Jamieson, 2008; Elder and Armstrong, 1987;Schweizer et al., 2003) and models (Durand et al., 1999; Rousselout etal., 2010; Schirmer et al., 2009, 2010) have been verifiedwith a posteriorihazard estimates, snowpit data and stability tests, or a limited number ofavalanche occurrences. Avalanche hazard forecasting largely depends onscale and type of area. For example, for the same absolute hazard level, abackcountry area will have far fewer avalanche occurrences than a skiarea with extensive avalanche control measures. Many stability modelswere developed for backcountry forecast areas, so their accuracy cannotbe directly compared to models specifically developed for ski areas orother areas with extensive avalanche control measures.

At many ski areas, one will find snow safety professionals stillimplicitly using Atwater's 10 contributory factors (Atwater, 1954;Atwater and Koziol, 1953): 1) old snow depth, 2) old snow surface,

Page 2: Forecasting artificially-triggered avalanches in storm snow at a large

262 E.H. Bair / Cold Regions Science and Technology 85 (2013) 261–269

3) new snow depth, 4) new snow type, 5) new snow density,6) snowfall intensity, 7) precipitation intensity, 8) wind action,9) air temperature, and 10) snow settlement. Perla (1970) examinedthe impact of the 10 contributory factors on avalanche hazard over107 storms at Alta, UT. He found that all measures of precipitation(e.g. Atwater's factors 3, 6, and 7) show clear positive relationshipswith avalanche hazard. Other factors, such as wind speed and changesin air temperature, did not.

With a few exceptions (i.e. Atwater and Koziol, 1953; Davis et al.,1999; Föhn et al., 1977; McCollister et al., 2003; Perla, 1970;Rosenthal and Elder, 2003; Stethem and Perla, 1980), much of the av-alanche forecasting literature focuses on backcountry areas. Sincethere are few studies that focus on ski areas or on avalanches thatfail in storm snow, the aim of this study is to describe a few simplevariables that have proven successful for predicting avalanches at alarge ski area.

2. Location, data, and methods

2.1. Mammoth Mountain

MammothMountain is a silica dome cluster (Hildreth, 2004) with abase elevation of 2424 m and a summit at 3369 m. LaChapelle's (1966)avalanche climate classificationswould place it in the Coastal TransitionZone. Like other Pacific Coast areas,MammothMountain receives heavywinter precipitation, accumulating an average of 890 mm of snowwater equivalent (SWE) and 719 cm of snow depth from Decemberthrough March. Mammoth's Main Lodge elevation, 2712 m, is higherthan most Pacific Coast ski areas and is similar to Intermountain areas,which have an average base elevation of 2605 m (Armstrong andArmstrong, 1987).Mammoth's higher elevation leads to colder temper-atures and infrequent mid-winter rain, uncommon characteristicsfor Coast areas. The average Main Lodge December to March daily

Fig. 1. Map of Mammoth Mountain ski area. Study Plot is where precipitation is measu

temperature is −2.4 °C, slightly lower than the average Coast baselodge temperature, –2.0 °C, but higher than the average Intermountainbase lodge air temperature, –6.0 °C, and much higher than the averageRocky Mountain base lodge air temperature, –8.7 °C (Mock andBirkeland, 2000). In one study (Mock and Birkeland, 2000), themixtureof Coast and Intermountain avalanche climate characteristics causesMammoth to be classified as a Coast area in half the years and anIntermountain area in the other half.

2.2. Mammoth Mountain ski patrol (MMSP) daily weather observations

At Mammoth, trained observers have taken daily morning weath-er observations and measurements on over 6000 days since 1982.Total depth, new snow, new snow density, new SWE, temperatures,relative humidity, visibility, and several other measurements aretaken every day between 5:00 and 8:00 am during the winter season(November–April) at the patrol snow study site (Study Plot, Fig. 1).“New” refers to 24-h accumulations (Fierz et al., 2009). Often, newsnow is manually weighed to determine SWE.

2.3. MMSP avalanche database

The avalanche control records are stored in a database withover 15,000 avalanches and over 40,000 total records (includesnon-avalanches, avalanches, and unseen results due to visibility)from 1982–2012. Only 1% of avalanches were naturally triggered;99% were artificially triggered, in decreasing order by: explosives, ar-tillery, and ski cuts. Records are estimates of avalanche propertiesthat can be easily observed, such as: relative class size, crown height,slab width, and total length (Greene et al., 2010). Bed surface has onlybeen recorded in the database since 2006, so there are only about3700 avalanches with a bed surface estimate.

Study PlotSesame West

Lost Lake

red. Sesame West and Lost Lake are nearby sites where stability tests were made.

Page 3: Forecasting artificially-triggered avalanches in storm snow at a large

1 2 3 4 50

500

1000

1500

2000

R-size

Ava

lanc

hes

groundoldinterfacestorm

4 50

10

20

30

40

50

60

Fig. 2. Class size and bed surface estimates in MMSP occurrence records. Vertical axis isavalanche count and the horizontal axis is class size, relative to the path (Greene et al.,2010). Colors are bed surface estimates. Sample size N=3743.

263E.H. Bair / Cold Regions Science and Technology 85 (2013) 261–269

2.4. Avalanche activity

In addition to counting the number of avalanches, I also used theunweighted sum of relative class sizes (R-size, avalanche size relativeto a path's maximum). Other studies have used weighted size sums toaccount for nonlinearity in size classification (McClung, 2009; Mockand Birkeland, 2000; Salway, 1979; Schweizer et al., 1998; Zeidler,2004). An absolute size classification (D-size, the avalanche size interms of destructive force; see Greene et al., 2010) is preferable tocompare results across different avalanche paths, but D-size hasonly recently been recorded at Mammoth Mountain. Since thisstudy involves mostly small avalanches, unweighted sums of R-sizeswere preferred because this metric emphasizes activity rather thansize (Davis et al., 1999; Gauthier et al., 2010). Most small avalanchesare classified as size R1, which is similar to D1 for most avalanchepaths. Ski areas, at least in the US, devote considerable resources topreventing small avalanches from occurring while slopes are open,as they are a legal liability. For this reason, I included R1 avalanchesin the following analysis rather than excluding them as in some stud-ies (Birkeland and Landry, 2002; Davis et al., 1999).

I only examined records from 92 selected avalanche paths. Onlyavalanches triggered by explosives were counted. Naturally-triggeredavalanches and those triggered by ski-cutting or artillery were exclud-ed. Avalanches triggered by ski-cutting were excluded because theyare consistently not recorded. Paths shot by artillery have been exclud-ed since those paths are often shot blind and results are often not seen.

2.5. Explosive yield

Explosive yield was computed as the count of avalanches trig-gered by explosives divided by the sum of shots for the selected ava-lanche paths. As with R-size, avalanches triggered by artillery, by skicutting, or triggered naturally were excluded.

2.6. Avalanche activity predictors

To determine which variables were important avalanche activitypredictors, I analyzed the effect of eight weather variables, based onAtwater's ten factors, over 737 storms using hazard probability Hdefined by Perla (1970):

H ¼ number of storms in interval with S≥ttotal number of storms in interval

ð1Þ

where S is the sum of R-sizes for each storm and t is an arbitrary thresh-old, corresponding to 25th (low) and 75th (high) percentiles of R-sizesfor all storms. Here, storms are defined as continuous days with newsnow and include the following day after snowfall ends. Storms wereplaced into five evenly-spaced bins based on each weather variable. Thevariables examined were: maximum snow density change, maximumtemperature change, mean wind speed, mean wind direction, totalSWE, maximum SWE intensity, mean SWE intensity, and total snow. Tolimit storms to dry snow avalanches, only storms from December toApril were included.

2.7. Extended Column Test

The Extended Column Test (ECT), developed by Simenhois andBirkeland (2006, 2009), is similar to the compression test developedby Canadian Park wardens in the 1970s (Jamieson, 1999), exceptthat it employs a wider column, so that crack propagation is also ob-served. For details on how an ECT is performed, see Simenhois andBirkeland (2009). At least three ECTs were performed each day toensure valid results. Although there were differences in the numberof taps needed for propagation, all of the tests were consistent with

regard to propagation; on any given day, all tests propagated on thesame failure layer or none propagated.

Like other studies that evaluate the predictive accuracy of the ECT(Schweizer and Jamieson, 2010; Simenhois and Birkeland, 2009),I only examined two cases: propagation and no propagation. Whenusing the propagation criterion alone, the ECT has been shown to bea more accurate stability test than Compression Tests, PropagationSaw Tests, or Rutschblock Tests (Schweizer and Jamieson, 2010;Simenhois and Birkeland, 2009).

2.8. Hypothesis test

Probabilities in the results are based on the Kruskal-Wallis (1952)test, which assumes that populations have the same, not necessarilynormal, distribution and that all observations are independent. Thenull hypothesis is that the two groups come from the same population.I used a threshold p value for significance at p≤0.05 below which thenull hypothesis is rejected and the two groups are significantlydifferent.

2.9. Classification trees

To determine threshold values for snow and SWE associated withECT propagation, I constructed classification trees. Classification treesare a method of supervised learning where values are placed intoclasses that minimize a chosen measure of impurity. In this study,there are two classes: “propagation” (N=10) and “no propagation”(N=24). The measure of impurity minimized is Gini's diversityindex (Breiman et al., 1984).

3. Results

3.1. MMSP avalanche database

Most of the avalanches at Mammoth Mountain are small (Fig. 2).This agrees with previous studies of avalanches at Mammoth Moun-tain (Bair et al., 2008; Rosenthal and Elder, 2003) and ski areas(Birkeland and Landry, 2002; Faillettaz et al., 2004; Louchet et al.,2002) that show there are many small avalanches and few largeones. About 56% of avalanches were estimated to have failed in astorm layer. Another 31% were estimated to have failed on the inter-face at the top of the old snow layer.

Page 4: Forecasting artificially-triggered avalanches in storm snow at a large

(a) (b)

(c) (d)

(e) (f)

(g) (h)

Fig. 3. Avalanche hazard probabilities for selected paths and weather variables. Thevertical axes are avalanche hazard probabilities, and the horizontal axes are bin centersfor different weather measurements. The colored lines represent thresholds for R-sizesums, based on percentiles for all storms; low hazard (gray) exceeds the 25th percen-tile and high hazard (orange) exceeds the 75th percentile. These graphs can be com-pared with those in Perla (1970). The number of storms analyzed N=737.

264 E.H. Bair / Cold Regions Science and Technology 85 (2013) 261–269

Seven crown face profiles of avalanches that failed on storm/oldinterfaces at Mammoth show that none of them contained persistentcrystals; all contained nonpersistent crystals in the slab and the sub-strate (Bair, 2011). Also, a majority (20/23) of crown face profilescontained failure layers of storm snow. Thus, we assume that inter-face failures at Mammoth usually involve only storm snow. Together,interface failures and failures in storm layers comprise 87% of ava-lanches at Mammoth. The proportion of storm snow avalanches de-creases with R-size, but there are still three (27%) interface failuresand five (45%) storm failures out of eight total R-size 5 avalanchesin the database.

3.2. Weather variables important to avalanche hazard

Of the eight weather variables analyzed (Fig. 3), only the precipi-tation variables (Fig. 3e–h) had positive relationships with avalancheactivity or “first order” (Perla, 1970) effects. Changes in snow densityor air temperature (Fig. 3a,b) and wind speed/direction (Fig. 3c,d) os-cillate near a mean value and thus do not affect hazard probability.The “upside down storm” (increasing temperature, Atwater, 1954)did not affect avalanche hazard at Mammoth Mountain.

3.3. ECTs

I performed ECTs on 37 days over two winters, 2010–11 and2011–12. It was snowing or there was new snow on 34 of thosedays (92%). Tests on days without snowfall were all performed withinone day of snowfall, at the end of a storm. For the selected avalanchepaths, there were only six days (out of 967 or b1%) with avalanchesrecorded more than one day after new snowfall, which suggestednoneed to perform ECTsmore than one day after snowfall to predict av-alanches inside the ski area. All ECTs took place at Lost Lake or SesameWest (Fig. 1). Propagation occurred on 13 out of 37 days (35%).

All failures occurred on nonpersistent crystals, except for threesampling dates, 16 Feb, 17 Feb, and 16Mar 2011, which were excludedfrom the analysis. Failures on nonpersistent crystals propagated on 10of 34 days (29%). Whether or not an ECT propagated correlated withthe amount of new SWE and snow (Fig. 4). Results are given for bothbecause new snow is more reliably measured than SWE. For newSWE, themedian is 46 mm for propagation and 18 mm for no propaga-tion (p=0.03). For new snow, themedian for propagation is 54 cm and17 cm for no propagation (pb0.01). Threshold values from classifica-tion trees between “propagation” and “no propagation” are 25 mmSWE (26% misclassification) and 29 cm snow (15% misclassification).

Days with propagation also have higher median snow storm totals(65 cm) than days with no propagation (26 cm), but the difference isnot significant (p=0.06), suggesting that new snow and SWE have agreater impact on propagation. Still, the p-value is close to 0.05, sothere is some statistical evidence that days with propagation havegreater storm totals than days without propagation.

Days with propagation also have significantly greater R-size sums(pb0.01). The median R-size sum for propagation days is 49 vs. 0 fordays without propagation (Fig. 5). Days without propagation do nothave significantly different explosive yields than days with propaga-tion (p=0.27).

3.4. Explosive yield

The total explosive yield for the 92 selected paths at Mammoth is24%. Explosive yield shows little relationship (R2b0.06 or less) to pre-cipitation variables (i.e. new snow/SWE, storm snow, and new snowdensity), aggregated over single days or multi-day storm periods.Although yield shows year-to-year variation, from 0.15 to 0.35, linearregressions of the number of shots, avalanches, and yield per wateryear (Fig. 6a–c) have slopes with 95% confidence intervals thatcross zero, which means there is no significant statistical trend; the

three variables are stationary. Explosive yield varies on a path-by-path basis, from 0.03 to 0.61, with slight correlation to maximumstart zone angle (R2=0.15, Fig. 7).

3.5. Effect of performing avalanche control work after one day

To see the effect of performing control work more than 24h aftersnowfall, rather than while snowing or immediately after snowfall, Iexamined avalanche activity on the selected avalanche paths andgrouped them based on how many days it had been since snowfall.There were 904 avalanche control days with snowfall in the past24 h, 28 avalanche control days without snowfall in the past 24 h,7 days without snowfall in the past 48 h, 3 days without snowfall inthe past 72 h, and 2 days without snowfall in the past 96 h. Becausethere were so few avalanche control days after 24h with no

Page 5: Forecasting artificially-triggered avalanches in storm snow at a large

(a) (b)

Fig. 4. New SWE and snow grouped by ECT propagation, storm snow failures only. Left (a) shows new SWE for days with and without ECT propagation. Right (b) shows the same fornew snow. N=10 days with propagation and 24 days without propagation. Line at center is the median, boxes are 25th and 75th percentiles, whiskers are non-outlier ranges, andcrosses are outliers. Outliers are points larger than p75+1.5(p75−p25) or smaller than p25−1.5(p75−p25), where p75 and p25 are 75th and 25th percentiles. Non-overlappingnotches indicate statistical significance at p=0.05, based on the Kruskal–Wallis test.

265E.H. Bair / Cold Regions Science and Technology 85 (2013) 261–269

snowfall, I focused only on the first two groups (new snow and nonew snow in the past 24 h). Avalanche control days with new snowhad significantly more avalanches, greater R-size sums, and greaterexplosive yields (all p-valuesb0.01, Fig. 8a–c). It is striking that thedays without new snow had median values of zero for avalanches,R-size sums, and explosive yield, meaning that usually, explosivescaused zero avalanches more than 24h since snowfall ended.

4. Discussion

4.1. Avalanche size and bed surface

Most avalanches at Mammoth were size R1 and failed in the stormlayer. Most of the others failed at the storm/old snow interface. Thus,87% of avalanches at Mammoth involved only storm snow. Althoughthe proportion of avalanches that only involve storm snow decreaseswith R-size, even the majority (72%) of R5 avalanches involved onlystorm snow. Avalanche control and ski compaction are the moststraightforward explanations. Constant release of unstable stormsnow by explosives and increased strength via compaction from theweight of thousands of skiers are effective measures to reduce ava-lanches on persistent weak layers.

Less clear is why most avalanches at Mammoth fail within stormsnow rather than at the storm/old snow interface. This finding suggests

Fig. 5. Sum of R-sizes grouped by ECT propagation, storm snow avalanches on selectedpaths. N=10 days with propagation and 24 days with no propagation.

that snow directly above an interface is usually better bonded to oldersnow below than to newer snow above. One reason may be that mostSierra Nevada storms start with periods of low precipitation that in-crease gradually (Fig. 9). Periods of low precipitation intensity allowstorm snow to strengthen via compaction and sintering before addi-tional weight causes enough stress to trigger an avalanche. Changingstorm conditions, such as decreasing temperature and wind may alsocause weaker snow to be deposited in the middle of the storm, ratherthan at the beginning. Onemight think that another reason for the prev-alence of failures within storm snow is that skier traffic causes a raggedsurface at the interface that inhibits fracture propagation. The distribu-tion of 196 bed surfaces from avalanches on four avalanche paths thatare permanently closed to skiing and only controlled with explosivesshows a similar distribution to the ski area as a whole: ground (1%),old (10%), interface (31%), and storm (58%). Similar distributions be-tween these paths and thewhole ski area suggest that skier compactionand surface roughing do not affect bed surface distribution.

4.2. Factors that affect avalanche hazard

As other studies focusing on ski areas have shown (Atwater, 1954;Davis et al., 1999; Föhn et al., 1977; McCollister et al., 2003; Mock andBirkeland, 2000), new precipitation is an important predictor of ava-lanche activity. The 29 cm of new snow threshold from the classifica-tion tree is very close to the 31 cm threshold originally given byAtwater (1954) in his 10 contributory factors. The upside downstorm, commonly perceived by avalanche workers to increase ava-lanche activity, did not affect avalanche hazard. Similarly, Perla(1970) showed that the probability of low avalanche hazard wasnot affected by changing storm temperatures, although Perla foundthat the probability of high avalanche hazard increased with stormsthat increased in temperature. One explanation may be that stormswith increasing temperature may be correlated with higher SWEamounts because of so called warm “Pineapple Express” storms thatdraw moisture from the southern Pacific. Thus, it may be high SWEamounts that increase hazard, not warming during a storm.

It's unclear why wind speed and direction do not have more of apronounced effect on avalanche hazard. For instance, an earlierstudy using classification trees on the Mammoth Mountain avalanchedatabase (Davis et al., 1999) showed that a derived wind load param-eter (24 h SWE×24 h wind speed4) was the most accurate factor forpredicting R-size sums. It is likely that wind speed and direction affectavalanche hazard at the individual path scale over short time frames.Yet, it may be that wind speed and direction do not have an effect

Page 6: Forecasting artificially-triggered avalanches in storm snow at a large

1985 1990 1995 2000 2005 20100

500

1000

1500

2000

2500

3000

Water year

Sho

ts

(a)

1985 1990 1995 2000 2005 20100

100

200

300

400

500

600

700

Water year

Ava

lanc

hes

(b)

1985 1990 1995 2000 2005 20100

0.05

0.1

0.15

0.2

0.25

0.3

0.35

Water year

Yie

ld

(c)

Fig. 6. Number of shots, avalanches, and explosives yield by water year. For the selected avalanche paths, the total number of shots (a), avalanches (b), and explosive yield (c) areplotted by water year.

266 E.H. Bair / Cold Regions Science and Technology 85 (2013) 261–269

over larger spatial and longer temporal scales. The narrowdistribution of speeds (3–7 m s−1) and directions (south to west)suggest that most storms have similar wind patterns at Mammoth.

30 40 50 60 70

0.1

0.2

0.3

0.4

0.5

0.6

Max. starting zone angle,

Exp

losi

ve y

ield

Fig. 7. Explosive yield vs. maximum start zone slope. Maximum starting zone slope forselected avalanche paths, derived using a high resolution (0.3 m2) digital elevationmodel. Gray line is a least-squares linear regression. The slope is 0.005 (95%confidence interval is 0.002 to 0.008), R2=0.15.

This lack of variability suggests that wind is a poor predictor of ava-lanche hazard, which exhibits high variability between storms.

4.3. ECTs

Most days with new snow did not have ECT propagation. ECTpropagation is an important predictor of avalanche activity. If anECT does not propagate, there is almost no avalanche activity, butwith propagation, there is high avalanche activity. The threshold forECT propagation is very close to Atwater's 0.31 m threshold for condi-tions favorable to avalanches. This threshold seems to be when newsnow can sustain crack propagation. Low-density new snow is theweakest of all types (Roch, 1966), so it fractures easily, but those frac-tures often do not propagate far. At the slope scale, small failures arecalled sloughs. Sloughs are relatively harmless and are similar to ECTsthat crumble in the new snow (ECTN or ECTX). When slabs aredeeper or denser and therefore tougher, they can sustain a propagat-ing collapse wave (Heierli et al., 2008) without slab fracture; that iswithout tensile fractures arresting fracture at the bed surface. Thesefailures are similar to propagating ECTs (ECTP), usually with cleanand planar fracture surfaces. Thus, the ECT can distinguish betweennew snow that is likely to slough and new snow that is likely toform a slab avalanche.

Although more explosives were thrown on days with propagation,explosives were still used on days without propagation and often did

Page 7: Forecasting artificially-triggered avalanches in storm snow at a large

new snow no new snow new snow no new snow

0

20

40

60

80

100

120

R−

size

sum

Ava

lana

ches

(b)(a)

new snow no new snow

0

0.2

0.4

0.6

0.8

1

Exp

losi

ve y

ield

(c)

50

40

30

20

10

0

Fig. 8. Effect of new snow on avalanche activity and explosive yield for selected avalanche paths. Box plots are: (a) number of avalanches, (b) sum of R-sizes, and (c) explosive yieldgrouped by snow in the past day. The “no new snow” group had no new snow in the last 24 h, but snow prior. The “new snow group” had snow within the last 24 h. N=132 days for “no new snow” and N=904 days for “new snow”.

267E.H. Bair / Cold Regions Science and Technology 85 (2013) 261–269

not produce any avalanches. Median explosive yields for days with-out propagation were lower (0.19 vs. 0.37), but not significantlydifferent (p=0.27), probably because of the small sample size. I

0 5 100

0.5

1

1.5

2

2.5

3

3.5

Elapsed time, hr

Hou

rly w

ater

equ

ival

ent,

mm

Fig. 9. Study Plot hourly water equivalent vs. elapsed time since precipitation began.Vertical axis is median hourly water equivalent from a precipitation can at the StudyPlot (Fig. 1). Horizontal axis is elapsed time since precipitation began. N=110 storms.

suggest certain new snow stratigraphy conducive to avalanchingcaused high avalanche activity on propagation days, not increaseduse of explosives.

It should be noted that the stability tests in this study suffer fromlimitations of study plots (Jamieson et al., 2007). Mainly, the newsnow and stratigraphy at the study plot may not represent those inthe avalanche starting zones. Spatial variation issues are inherentwhen using study plots to predict stability across a large area. To ad-dress issues with spatial variation in stability, ski areas use manyexplosives, even during times of moderate stability, to cover their av-alanche terrain.

4.4. Explosive yield

The number of shots, avalanches, and yield are stationary over thepast three decades. This result suggests that the avalanche threat didnot change, nor did the effectiveness of avalanche control work atMammoth Mountain. This result emphasizes the quality of the ava-lanche database, since long-term stationary databases are desirablefor analysis.

The variation in explosive yield of an order of magnitude betweenpaths is interesting and explains some of the path names. For in-stance, “Old Faithful” (Fig. 10a) implies frequent avalanching andhas an explosive yield of 45% (73 avalanches over 164 shots), while“Climax” (Fig. 10b) implies accumulation and infrequent avalanching

Page 8: Forecasting artificially-triggered avalanches in storm snow at a large

27

60

27

70

2 7 4 0

27

202

73

0

2 7 9 02 8 0 0

2 8 1 02 8 2 0

2 8 3 0

2 8 4 02 8 5 02 8 6 0

27

10

28

70

2 8 8 0

27

00

28

90

0610 80 Meters

3260

3240

3220

3180

3200

3300

3160

3320

3340

31403120

3360

3180

3200

0430 170 Meters

2(a) Old Faithful (b) Climax

27

08

Fig. 10. Satellite images of two avalanche paths. Old Faithful (a) is on the left, Climax (b) is on the right. The labeled contours are the elevations in m.

268 E.H. Bair / Cold Regions Science and Technology 85 (2013) 261–269

and has an explosive yield of only 7% (70 avalanches over 1067shots). Aside from Climax being a much larger path, the two havesimilar steep starting zones and aspect; 60° maximum slope and NEaspect for Climax; 65° maximum slope and E aspect for Old Faithful.One reason for the vastly different yields might be wind exposure.Old Faithful is considerably more wind-protected by trees thanClimax, which is entirely above the treeline. As a result of its wind ex-posure, Climax accumulates vast amounts of wind-loaded snow,which form permanent snowfields. Windslabs can be very strongand can thus accumulate to dangerous depths before failing. SinceOld Faithful accumulates less wind-loaded snow, its new snow loadis probably weaker and subject to more frequent avalanching thanClimax.

4.5. Effect of waiting one day to perform avalanche control work

Avalanche control days with no new snow recorded in the past24h had significantly fewer avalanches, lower R-size sums, and lowerexplosive yields. This is probably because new snow strengthens signif-icantly in the first 24 h via viscous compaction and sintering. Waitingone day to perform control workwas the only significant factor that de-creased explosive yield. Thus, waiting is a potential cost-savingmeasurein terms of labor, explosives, and risk of post-control avalanches. Yet,waiting a day or longer is often not a realistic option. Ski area managersmust weigh costs with the financial benefits of opening terrain within24 h of snowfall, since new snow is popular with resort guests. Thetradeoff is that processes that increase strength (e.g. viscous compac-tion and sintering) decrease the quality of powder skiing. Also, explo-sives are used to test slopes, as well as to trigger avalanches, so lowexplosive yields after waiting a day may not be undesirable, but ratheran affirmation of a forecast for low avalanche activity.

4.6. Applying results to other ski areas

Explosive yields of 20–30% for Swiss ski areas reported by Gubler(1977) suggest little difference between yields in this study and thosein Switzerland. Similarly, Perla and Everts (1983) report a yield of 24%over 12 years of heli-bombing missions for the Sunshine Ski AreaRoad, Alberta, Canada; although they excluded size D1 events.

It would be interesting to analyze a similar database from a conti-nental ski area. I suspect that onewouldfind similar results, but a higher

prevalence of buried persistent weak layers and lower air temperaturesmight increase the waiting time needed to reduce avalanche activity.Likewise, the threshold new snow and SWE values would probably bedifferent. Because these results are only based on measurements fromone ski area, the specific thresholds are probably not broadly applicable.

5. Conclusion

This study examined explosively-triggered avalanches atMammothMountain, where the great majority of avalanches only involve stormsnow. Two datasets were examined. One dataset contains thousandsof avalanches on selected paths over three decades. This dataset wasused to find factors that affected avalanche activity. The findings werethat avalanche activity increased with all measures of precipitation,while changes in temperature and snow density did not affect ava-lanche activity. There was also not a specific wind speed or directionthat increased avalanche activity. Explosive yield was about one ava-lanche per four shots, similar to yields reported in other studies.Waiting one day after snowfall ended to perform control work resultedin significantly fewer avalanches and lower R-size sums, and it is theonly significant factor that decreased explosive yield.

The second dataset is smaller and contains two seasons of ECT re-sults at Mammoth Mountain. These ECT results were compared withresults from avalanche control work using explosives and with newprecipitation amounts. The ECT was a powerful predictor of avalancheactivity; days with ECTs that propagated had significantly greaternumbers of avalanches and R-size sums than days without ECT prop-agation. Days with ECT propagation also had significantly greatersnowfall, with the threshold between propagation (ECTP) and nopropagation (ECTN/X) being 0.29 m of new snow, very close toAtwater's threshold value of 0.31 m. That new precipitation above athreshold causes increased avalanche activity is not a new finding;the new finding is that ECT propagation also has a similar threshold.Thus, I suggest that ECT propagation is an important tool to predictexplosively-triggered avalanches in storm snow.

Acknowledgments

I thank Jeff Dozier, Karl Birkeland, and two anonymous reviewersfor their helpful comments.

Page 9: Forecasting artificially-triggered avalanches in storm snow at a large

269E.H. Bair / Cold Regions Science and Technology 85 (2013) 261–269

References

Armstrong, R.L., Armstrong, B., 1987. Snow and avalanche climates of the westernUnited States: a comparison of maritime, intermountain and continental condi-tions. In: Salm, B., Gubler, H. (Eds.), Avalanche Formation, Movement and Effects:IAHS Publ. No. 162, pp. 281–294 (Wallingford, UK).

Atwater, M.M., 1954. Snow avalanches. Scientific American 190 (1), 26–31.Atwater, M.M., Koziol, F.C., 1953. Avalanche Handbook, (149 pp.).Bair, E.H., 2011. Fracture mechanical and statistical properties of nonpersistent snow

avalanches. Ph.D. Thesis, University of California, Santa Barbara, CA, 183 pp.Bair, E.H., Dozier, J., Birkeland, K.W., 2008. Avalanche crown-depth distributions.

Geophysical Research Letters 35, L23502. http://dx.doi.org/10.1029/2008GL035788.Bartelt, P., Lehning, M., 2002. A physical SNOWPACK model for the Swiss avalanche

warning: Part I: numerical model. Cold Regions Science and Technology 35 (3),123–145. http://dx.doi.org/10.1016/s0165-232x(02)00074-5.

Birkeland, K.W., Landry, C.C., 2002. Power-laws and snow avalanches. GeophysicalResearch Letters 29 (11), 1554. http://dx.doi.org/10.1029/2001GL014623.

Breiman, L., Friedman, J., Olshen, R., Stone, C., 1984. Classification and Regression Trees.CRC Press, Boca Raton, FL . (368 pp.).

Brown, C., Jamieson, B., 2008. Shear strength and snowpack stability trends in non-persistent weak layers. 2008 International Snow Science Workshop, Whistler, BC,Canada, pp. 939–947.

Buser, O., 1983. Avalanche forecast with the method of nearest neighbours: an interac-tive approach. Cold Regions Science and Technology 8 (2), 155–163. http://dx.doi.org/10.1016/0165-232X(83)90006-X.

Buser, O., Föhn, P.M.B., Good, W., Gubler, H., Salm, B., 1985. Different methods for theassessment of avalanche danger. Cold Regions Science and Technology 10 (3),199–218. http://dx.doi.org/10.1016/0165-232x(85)90033-3.

Casson, J., 2009. Evaluating the importance of crystal type on new snow instability.M.Sc. Thesis, University of Washington, Seattle, 73 pp.

Conway, H., Wilbour, C., 1999. Evolution of slope stability during storms. Cold RegionsScience and Technology 30 (1–3), 67–77. http://dx.doi.org/10.1016/S0165-232X(99)00009-9.

Davis, R.E., Elder, K., Howlett, D., Bouzaglou, E., 1999. Relating storm and weatherfactors to dry slab avalanche activity at Alta, Utah, and Mammoth Mountain,California, using classification and regression trees. Cold Regions Science and Tech-nology 30 (1–3), 79–89. http://dx.doi.org/10.1016/S0165-232X(99)00032-4.

Durand, Y., Giraud, G., Brun, E., Mérindol, L., Martin, E., 1999. A computer-based systemsimulating snowpack structures as a tool for regional avalanche forecasting. Journal ofGlaciology 45 (151), 469–484.

Elder, K., Armstrong, B., 1987. A quantitative approach for verifying avalanche hazardratings. In: Salm, B., Gubler, H. (Eds.), Avalanche Formation, Movement and Effects:IAHS Pub. no. 162, pp. 593–603 (Wallingford, UK).

Faillettaz, J., Louchet, F., Grasso, J.-R., 2004. Two-threshold model for scaling laws ofnoninteracting snow avalanches. Physical Review Letters 93 (20), 208001. http://dx.doi.org/10.1103/PhysRevLett.93.208001.

Fierz, C., Armstrong, R.L., Durand, Y., Etchevers, P., Greene, E., McClung, D.M.,Nishimura, K., Satyawali, P.K., Sokratov, S., 2009. The International Classificationfor Seasonal Snow on the Ground. IHP-VII Technical Documents in HydrologyN°83(IACS Contribution N°1), (90 pp.).

Föhn, P.M.B., 1987. The stability index and various triggering mechanisms. In: Salm, B.,Gubler, H. (Eds.), Avalanche Formation, Movement, and Effects: IAHS Pub. no. 162,pp. 195–214 (Wallingford, UK).

Föhn, P., Bois, P., Obled, C., 1977. Evaluation and comparison of statistical and conventionalmethods of forecasting avalanche hazard. Journal of Glaciology 19 (81), 375–387.

Gassner, M., Brabec, B., 2002. Nearest neighbour models for local and regionalavalanche forecasting. Natural Hazards and Earth System Sciences 2, 247–253.http://dx.doi.org/10.5194/nhess-2-247-2002.

Gauthier, D., Brown, C., Jamieson, J.B., 2010. Modeling strength and stability in stormsnow for avalanche forecasting. Cold Regions Science and Technology 62,107–118. http://dx.doi.org/10.1016/j.coldregions.2010.04.004.

Greene, E., Atkins, D., Birkeland, K.W., Elder, K., Landry, C., Lazar, B., McCammon, I.,Moore, M., Sharaf, D., Sterbenz, C., Tremper, B., Williams, K., 2010. Snow, Weather,and Avalanches: Observational Guidelines for Avalanche Programs in the UnitedStates. American Avalanche Association, Pagosa Springs, CO . 136 pp.

Gubler, H., 1977. Artificial release of avalanches by explosives. Journal of Glaciology 19(91), 419–428.

Hayes, P., Wilbour, C., Gibson, R., Marshall, H.P., Conway, H., 2004. A simple model ofsnow slope stability during storms. 2004 International Snow Science Workshop,Jackson, WY, pp. 165–171.

Heierli, J., Purves, R.S., Felber, A., Kowalski, J., 2004. Verification of nearest-neighboursinterpretations in avalanche forecasting. Annals of Glaciology 38, 84–88. http://dx.doi.org/10.3189/172756404781815095.

Heierli, J., Gumbsch, P., Zaiser, M., 2008. Anticrack nucleation as triggering mechanismfor snow slab avalanches. Science 321 (5886), 240–243. http://dx.doi.org/10.1126/science.1153948.

Heywood, L., 1992. Alpine Meadows: March 31, 1982 the story. The Avalanche Review10 (5).

Hildreth, W., 2004. Volcanological perspectives on Long Valley, Mammoth Mountain,and Mono Craters: several contiguous but discrete systems. Journal of Volcanologyand Geothermal Research 136 (3–4), 169–198. http://dx.doi.org/10.1016/j.jvolgeores.2004.05.019.

Hirashima, H., Nishimura, K., Yamaguchi, S., Sato, A., Lehning, M., 2008. Avalancheforecasting in a heavy snowfall area using the SNOWPACK model. Cold RegionsScience and Technology 51 (2–3), 191–203. http://dx.doi.org/10.1016/j.coldregions.2007.05.013.

Jamieson, J.B., 1995. Avalanche prediction for persistent snow slabs. Ph.D. Thesis,University of Calgary, Calgary, AB, Canada, 258 pp.

Jamieson, J.B., 1999. The compression test – after 25 years. The Avalanche Review 18(1), 10–12.

Jamieson, J.B., Johnston, C.D., 1993. Shear frame stability parameters for large scaleavalanche forecasting. Annals of Glaciology 18, 268–273.

Jamieson, J.B., Johnston, C.D., 1998. Refinements to the stability index for skier-triggered slab avalanches. Annals of Glaciology 26, 296–302.

Jamieson, J.B., Zeidler, A., Brown, C., 2007. Explanation and limitations of study plotstability indices for forecasting dry snow slab avalanches in surrounding terrain.Cold Regions Science and Technology 50, 23–34. http://dx.doi.org/10.1016/j.coldregions.2007.02.010.

Kruskal, W.H., Wallis, W.A., 1952. Use of ranks in one-criterion variance analysis.Journal of the American Statistical Association 47 (260), 583–621.

LaChapelle, E., 1966. Avalanche forecasting – a modern synthesis. Scientific Aspects ofSnow and Ice Avalanches, pp. 350–356.

LaChapelle, E., 1980. The fundamental processes in conventional avalanche forecasting.Journal of Glaciology 26 (94), 75–84.

Louchet, F., Faillettaz, J., Daudon, D., Bédouin, N., Collet, E., Lhuissier, J., Portal, A.-M.,2002. Possible deviations from Griffith's criterion in shallow slabs, and conse-quences on slab avalanche release. Natural Hazards and Earth System Sciences 2(3/4), 157–161. http://dx.doi.org/10.5194/nhess-2-157-2002.

McClung, D.M., 2009. Dimensions of dry snow slab avalanches from field measure-ments. Journal of Geophysical Research 114, F01006. http://dx.doi.org/10.1029/2007JF000941.

McCollister, C.M., 2004. Geographic knowledge discovery techniques for exporinghistorical weather and avalanche data. M.Sc. Thesis, Montana State University,Bozeman, MT, 117 pp.

McCollister, C., Birkeland, K., Hansen, K., Aspinall, R., Comey, R., 2003. Exploring multi-scale spatial patterns in historical avalanche data, Jackson Hole Mountain Resort,Wyoming. Cold Regions Science and Technology 37 (3), 299–313.

Mock, C.J., Birkeland, K.W., 2000. Snow avalanche climatology of the western United Statesmountain ranges. Bulletin of the AmericanMeteorological Society 81 (10), 2367–2390.http://dx.doi.org/10.1175/1520-0477(2000)081b2367:SACOTW>2.3.CO;2.

Penniman, D., 1986. The Alpine Meadows avalanche trial: conflicting viewpoints of theexpert witnesses. 1986 International Snow Science Workshop, pp. 189–194.

Perla, R.I., 1970. On contributory factors in avalanche hazard evaluation. CanadianGeotechnical Journal 7 (4), 414–419. http://dx.doi.org/10.1139/t70-053.

Perla, R.I., Everts, K., 1983. On the placement and mass of avalanche explosives: expe-rience with helicopter bombing and preplanted charges. Annals of Glaciology 4,222–227.

Pozdnoukhov, A., Purves, R.S., Kanevski, M., 2008. Applying machine learning methodsto avalanche forecasting. Annals of Glaciology 49, 107–113. http://dx.doi.org/10.3189/172756408787814870.

Purves, R.S., Morrison, K.W., Moss, G., Wright, D.S.B., 2003. Nearest neighbours for av-alanche forecasting in Scotland – development, verification and optimisation of amodel. Cold Regions Science and Technology 37 (3), 343–355. http://dx.doi.org/10.1016/S0165-232X(03)00075-2.

Roch, A., 1966. Les variations de la resistance de la neige. Scientific Aspects of Snow andIce Avalanches: IAHS Publ. No. 69, pp. 182–195 (Gentbrugge, Belgium).

Rosenthal, W., Elder, K., 2003. Evidence of chaos in slab avalanching. Cold RegionsScience and Technology 37 (3), 243–253. http://dx.doi.org/10.1016/S0165-232X(03)00068-5.

Rousselout, M., Durand, Y., Giraud, G., Merindol, L., Daniel, L., 2010. Analysis andforecast of extreme new-snow avalanches: a numerical study of the avalanchecycles of February 1999 in France. Journal of Glaciology 56 (199), 758–770.http://dx.doi.org/10.3189/002214310794457308.

Salway, A.A., 1979. Time-series modelling of avalanche activity from meteorologicaldata. Journal of Glaciology 22 (88), 513–528.

Schirmer, M., Lehning, M., Schweizer, J., 2009. Statistical forecasting of regionalavalanche danger using simulated snow-cover data. Journal of Glaciology 55,761–768. http://dx.doi.org/10.3189/002214309790152429.

Schirmer, M., Schweizer, J., Lehning, M., 2010. Statistical evaluation of local to regionalsnowpack stability using simulated snow-cover data. Cold Regions Science andTechnology 64 (2), 110–118. http://dx.doi.org/10.1016/j.coldregions.2010.04.012.

Schweizer, J., Jamieson, B., 2010. Snowpack tests for assessing snow-slope instability.Annals of Glaciology 51 (54), 187–194.

Schweizer, J., Jamieson, B., Skjonsberg, D., 1998. Avalanche forecasting for transporta-tion corridor and backcountry in Glacier National Park (BC, Canada). Proceedingsof the Anniversary Conference: 25 Years of Snow Avalanche Research. NorwegianGeotechnical Institute, Voss, Norway, pp. 238–244.

Schweizer, J., Kronholm, K., Wiesinger, T., 2003. Verification of regional snowpackstability and avalanche danger. Cold Regions Science and Technology 37 (3), 277–288.http://dx.doi.org/10.1016/s0165-232x(03)00070-3.

Simenhois, R., Birkeland, K., 2006. The extended column test: a field test for fractureinitiation and propagation. 2006 International Snow Science Workshop, Telluride,CO, pp. 79–85.

Simenhois, R., Birkeland, K., 2009. The extended column test: test effectiveness, spatialvariability, and comparison with the propagation saw test. Cold Regions Science andTechnology 59 (2–3), 210–216. http://dx.doi.org/10.1016/j.coldregions.2009.04.001.

Stethem, C., Perla, R., 1980. Snow-slab studies at Whistler Mountain, British Columbia,Canada. Journal of Glaciology 26 (94), 85–91.

Williams, K., Armstrong, B., 1998. Westwide Avalanche Network Electronic Database:Avalanche Observations.

Zeidler, A., 2004. Forecasting skier-triggered avalanches in the Columbia Mountains ofCanada. Ph.D. Thesis, University of Calgary, Calgary, AB, Canada, 297 pp.