Avalanche danger variability in level 2 moderate and 3 considerable of the European danger scale following the EAWS bavarian matrix: experimental use of icons representing different weight within one degree and scenarios frequency in the last few winter seasons Mauro Valt 1,3, * and Flavio Berbenni 2,3 1 ARPA Veneto- Centro Valanghe di Arabba, Via Pradat, 5 32020 Arabba (BL),Italy 2 ARPA Lombardia, Centro Nivometeorolog)ico, Via- Monte Confinale, 23032 - Bormio (SO),Italy 3 AINEVA, Vicolo dell Adige, 18 38100 Trento, Italy ABSTRACT: The avalanche forecaster should describe and represent, unambiguously and clearly to users, the avalanche danger through the bulletin. The EAWS bavarian matrix, adopted in 2005, is a valuable support to the forecaster to determine avalanche hazard scenarios and to define, objectively, the level of danger. In recent years, its systematic use allowed us to verify the existence of intermediate hazard scenarios and their frequency for level 2 moderate and 3 considerable and to represent them in the bulletins, on a trial basis, by means of an icon. The collection of data relating to the frequency of intermediate situations shows that it is important to explicitly provide users with the information relating to the weight of the avalanche danger within each single scale degree. KEYWORDS: Bavarian Matrix, Avalanche forecast, Avalanche danger scale 1 INTRODUCTION Avalanche danger prediction is a complex decision making based upon the knowledge of snow physical processes, the availability of data concerning snowpack stability and the avalanche forecaster's experience (LaChapelle, 1980). The avalanche danger is described by a danger scale, unified at European level, which intersects the snowpack stability and the possibility of the occurrence of the avalanche phenomena generated by natural or artificial overload (skiers, explosives, etc.). In an organized structure, where the avalanche forecasters are alternated during the winter season, it is essential to have procedures which allow a high standard of forecasting objectivity and determining the level of avalanche danger. The expression of a danger level often has some repercussions on economic activity at high altitude (opening of ski slopes, road closures, etc..). For this purpose, the Bavarian Avalanche Service, developed, in the early 2000, a support matrix to the avalanches forecasting (known as the Bavarian Matrix BM) based on the European danger scale. The matrix is based on the probability of avalanche release and the aerial distribution of hazards sites. The present work attests the use, in recent years, of the BM by Italian regional avalanche forecasting services, highlighting the most Corresponding author address: Flavio Berbenni ARPA Lombardia - Centro Nivometeorologico - Bormio (SO),Italy email: f.berbenni@arpalombardia.it common avalanche scenarios used inside the matrix. It also describes danger situations which are often located in between two danger levels and proposes a system of weights, for each level, following the scenario identified in the matrix. 2 AVALANCHE DANGER IN ITALY The avalanches danger forecasting onto the Italian Alps is carried out by AINEVA s regional or provincial avalanche forecasting offices. Each region or province is divided into small homogeneous climatic areas regarding snow cover characteristics. Actually 45 homogeneous areas are present and for each daily avalanche danger level is forecasted (www.aineva.it). For each winter season, is therefore possible to determine the danger level most frequently used (Fig.1). Figure 1. Use of the avalanche danger level during the most recent winter seasons on the Italian Alps. The analysis of danger level percentage of use, for each winter season, allows also an estimation of the season type regarding the ava- 203
lanche danger. During the 2008-2009 season the danger level 4 high has been used more often than during other seasons, that season was, as a matter of fact, the third snowiest winter since 1930 (Valt and Cianfarra, 2010). During the 2009-2010 season the danger level 3 considerable has been widely used: due to its frequent snowfalls and strong winds, that was in Italy one of the season with the highest number of avalanche victims (more than 40 for the period 1984-2013the average is 20). During the two seasons 2006-2007 and 2011-2012, characterized by scarce snow falls, the danger level 1 low has been the more frequent one. Each AINEVA s regional avalanche forecasting service of the Italian Alps has its own organization, but the method of danger forecasting is common (using as an aid the BM) and data and observations are shared. 3 THE BAVARIAN MATRIX (BM) The Bavarian Matrix (BM) is based on the probability of avalanche release versus the aerial distribution of hazards sites. The matrix was presented, for the first time, at the 2003 general meeting of the European Avalanche Warning Services (EWAS) in Munich Bavaria Germany (Zencke, 2003). During such meeting and in Davos (2005) the BM (Fig. 2) was discussed and approved and, subsequently integrated into the majority of European avalanche services forecasting workflows (Frigo et al., 2012 Valt, 2012). The BM is divided into two sections, one relating to the triggering of avalanches with additional load (left side) and one relating to the spontaneous release (right side). Crossing the probability of release versus the aerial distribution of hazards sites, we can obtain the numerical value of the avalanche danger level for each box of the two sections. Usually, the forecaster which writes the avalanche bulletin, chooses the highest danger level issuing from the comparison between the two sections. The BM is characterized by colored and uncolored boxes (see Fig. 2) with a danger level and by empty ones (without a level). The colored boxes represent the most common avalanche scenario for each danger level, discussed and approved by EAWS, whereas the boxes with a danger level but uncolored are less common or poorly studied scenario, finally, the empty boxes are unreal situations. The possible combinations in the section triggering of avalanches with additional load are: 1 box with danger level 1 - low 7 boxes with danger level 2 - moderate 4 boxes with danger level 3 -considerable 3 boxes with danger level 4 - high 1 box with danger level 5 very high The possible combinations in the section spontaneous release of avalanches without additional load are: 1 box with danger level 1 - low 3 boxes with danger level 2 - moderate 4 boxes with danger level 3 -considerable 5 boxes with danger level 4 - high 3 boxes with danger level 5 very high The number of boxes with the same danger level are due to different combinations of the probability of avalanche release with the aerial distribution of hazards sites which allows to better differentiate the danger scenario. During the EAWS general meeting in Grenoble, Berbenni and Valt (2011) presented a new classification of the danger level boxes into the BM. They proposed to add a symbol "+" or "-" following the deviation of each BM avalanche scenario from the original definition adopted for the European Avalanche Danger Scale (Fig. 3). 4 METHOD Figure 2. The Bavarian Matrix (see EAWS www.avalanches.org). The avalanche bulletin describes the actual situation and the forecast for the next 24-72 hours. Usually, avalanche forecasters, using all available information and analyzing the snowpack characteristics (stable versus unstable snowpack) for the day or simulating its evolution in the near future, select for each section (triggered versus spontaneous) of the BM the two boxes which better describe the avalanche danger scenario and level related to the probability of release versus the aerial distribution of hazards sites. 204
Figure 3. The Bavarian Matrix modified by Berbenni and Valt (2011). Note the added symbol "+" or "-" for each box. Often avalanche forecasters identifies multiple avalanche scenarios possible for the same section. Our work was to collect all the BM handwritten (Veneto and Lombardy Regions) or stored in local database (Friuli Venezia Giulia, Piedmont Regions) into a single data set numbering the individual boxes in rows and columns. The first analysis was conducted clustering the boxes as proposed by Berbenni and Valt (2011) within the same danger degree: Danger level Berbenni and Valt (2011) 1 - low 1 2 - moderate 2-, 2, 2+ 3 - considerable 3-, 3, 3+ 4 high 4-, 4, 4+ 5 very high 5 It was possible to discover, level by level, how many days each single level floated between the -" and +" scenario. The second analysis was conducted comparing, for the same day and homogenous climatic area, the danger level assigned for each section of the BM and with the level issued in the avalanche bulletin. 5 DATA SOURCES Two different data sets of the BM were analyzed: a) 395 BM compiled by avalanche forecasters of the ARPA Veneto - Arabba Avalanche Centre (AINEVA) during the period 2008-2012; b) 689 BM compiled by AINEVA avalanche forecasters of several regional offices during the 2013 winter season: Friuli Venezia Giulia Region (Julian and Carnic Alps): from 27th December 2012 to 9th May 2013; Veneto Region Region (Dolomites): from 1st December 2012 to 2nd June 2013; Lombardy Region: from 3rd December 2012 to 13th May 2013; Piedmont Region (Ligurian and northern Lepontine Alps): erratically from 23rd January 2013 to 13th May 2013. Totally 1084 BM were analyzed. 6 RESULTS First data set: 5 winter seasons from 1 Avalanche Forecasting Service: 205
The danger level was the same in both BM sections for 296 of the 395 analyzed days (75%); for 26 days the danger level of the spontaneous avalanches section was greater than 1 level compared to the triggered section (7 episodes for the level 4 high, 9 days for the level 3 considerable and 10 for the level 1 - low). For the others 73 days, the danger level of the spontaneous avalanches section was greater than 1 level compared to the triggered section was greater than 1 degree in 70 episodes and of 2 levels in 3 cases. The danger level more issued onto the avalanche bulletin was level 2 moderate (51% - 200 days) followed by level 3 considerable (31% - 123 days), level 1 low (13% - 51 days) and level 4 high (5% - 21 days). There was no day with level 5 very high of avalanche danger. Detailed analysis of the data clustering, using the approach of Berbenni and Valt (2011), show interesting results (see Figure 4). Figure 4. Avalanche danger scenario of the BM for the 1st data set (5 winter seasons, 1 Avalanche Forecasting Service). Considering the danger level 1 low there is only one box and therefore the result is unique. The danger level 2 moderate show the highest percentage (46%) onto the mid value, a smaller percentage (19%) on the level "2 -" and quite a good value (35% - 69 days) on the level "2 +". The danger level 3 considerable show the highest percentage (50%) onto the level 3-, as well as the danger level 4 high on the level 4-. Therefore, it is clear that the avalanche forecaster observe quite a large number of days when the avalanche scenario is interposed between the danger level 2 moderate and level 3 considerable (34% - equal to 132 days during 5 winter seasons). The scenario clustering in the following groups: 2-, 2 ; 2+,3- and 3, 3+ show that the largest group fall under the category "2 +, 3 (see Figure 5). Figure 5. Intermediate avalanche danger scenario for the 1st data set (5 winter seasons, 1 Avalanche Forecasting Service). Second data set: 1 winter season from 4 Avalanche Forecasting Services: The danger level was the same in both BM sections for 413 of the 689 analyzed days (60%); for 222 days (32%) the danger level of the triggered avalanches section was greater than 1 level compared to the spontaneous section. Although the winter season 2012-2013 was quite snowy the more severe avalanche danger scenarios were those related to the triggered avalanches. The danger level more issued onto the avalanche bulletin was level 3 considerable (48%) followed by level 2 moderate (41%), level 4 high (6) and level 1 low (5%). There was no day with level 5 very high of avalanche danger. As performed on the first data set, detailed analysis of the data clustering, using the approach of Berbenni and Valt (2011), show interesting results (see Figure 6). Figure 6. Avalanche danger scenario of the BM for the 2nd data set (1 winter season, 4 Avalanche Forecasting Services). Also for this data set, quite a large number of days (27%) showed an avalanche scenario interposed between the level 2 and 3 (2+ and 3-) 206
and were higher than the single percentage of level 2 moderate (10%) and 3 considerable (18%). Compared to the first data set, the avalanche scenario interposed between level 3 and 4 (3+ and 4-) shows lower percentages (8%). The scenario clustering in the following groups: 2-, 2 ; 2+,3- and 3, 3+ show, in this data set, that the largest group fall again under the category "2 +, 3 (see Figure 7). 8 CONCLUSION The MB is a good tool for analysis and identification of scenarios for triggered or spontaneous avalanches. Collection and storage of danger level scenarios identified for each bulletin, allows to compare in an objective way the analysis about avalanches danger. Danger levels most frequently used in Italy are level 2 moderate and level 3 - considerable. Data analysis and clustering showed that there are several avalanche scenario interposed between level 2 moderate and level 3 - considerable and also between level 3 considerable and level 4 high. Such outcomes are supported also by Italian avalanche accidents statistic which shows that the highest number of victims occurred when danger was level 2 or 3 (See Figure 8). Figure 7. Intermediate avalanche danger scenario for the 2nd data set (1 winter season, 4 Avalanche Forecasting Services). 7 DISCUSSION Analyzing the two data sets, the first one including a 5 winter seasons data of one avalanche forecasting service and the other one concerning one winter season but from different avalanche forecasting services, it appears that in both cases is frequent the percentage of intermediate situations identified by avalanche forecasters. In 79% of danger level scenarios identified as 2+, forecasters showed a danger level 2 moderate, and in remaining cases they marked a danger level 3 considerable. In 82% of danger level scenarios identified as 3-, avalanche forecasters selected a danger level 3 considerable, and in the remaining events they marked a danger level 2 - moderate. From these data we can deduce that almost half of the cases (42%) in which the scenario of danger level is between moderate and considerable, it is indicated a less severe avalanche scenario than the avalanche forecaster identified. Regarding the scenarios of "3+" in almost all situations (96%) it was indicated the level 3 considerable while in scenarios "4-" the level 3 considerable was shown only in 22% of the situations and in the remaining 78% it was indicated level 4 -high. Figure 8. Avalanche accidents versus avalanche danger level in Italy (period 2007-2013 source AINEVA). The European avalanche danger scale is set upon 5 levels; in the BM there are 16 possible avalanche scenarios both for triggered and spontaneous avalanches (numbered 32 boxes with and without color) which correspond to 11 intermediate avalanche danger levels (1, 2-, 2,2+, 3-, 3, 3+, 4-, 4,4+, 5). Therefore, we propose two solutions in order to improve the avalanche forecasting, enhancing objectivity of the process, as well as the communication to the public by making it more explicit and detailed: a) Only for avalanche scenarios "2+" and "3+" we propose an additional small exclamation point in the top right hand corner of EAWS icon to emphasize the importance of such scenario (Fig. 6). For avalanche scenarios "2 -", 3- and 4- we propose to maintain the standard EAWS icon for the danger level as a precautionary measure. 207
10 REFERENCES Figure 6. Danger level 2 - moderate. The exclamation point in the top right hand corner is used for avalanche danger intermediate scenario 2+ of the BM. b) Complete the EAWS icon with an additional bar able to indicate which intermediate avalanche scenario is described (2-, 2, 2+; 3-,3,3+; 4-, 4, 4+)-(Figure 7a,b). Berbenni F., Valt M., 2011, Avalanche danger scale - new ideas to provide a better information to users: 16 th EWAS Meeting, Grenoble, France, 15 th 17 th September 2011. Cagnati A., 1993, La nuova scala unificata per la classificazione del pericolo da valanghe: Neve e Valanghe, v. 19, pag. 26-31. Frigo B., Prola M.C., and Faletto M., 2012, Valutazione della stabilità del manto nevoso - linee guida per la raccolta e l interpretazione dei dati: Regione Autonoma Valle d Aosta, pp. 103. Valt M., and Cianfarra P., 2010, Recent snow cover variations and avalanche activities in the Southern Alps: Cold Regions Science and Technology, v. 64 (2),pag. 146-151. Valt M., 2012, 5 anni di esperienze con la matrice bavarese: Neve e Valanghe, v. 76 (4), pag. 4-13. Zencke B., 2003, Die Lawinengefahrenstufen - Wie kommt der Lawinenwarner zur aktuellen Gefahrenstufe: http://www.lawinenwarndienstbaern.de/download/infothek/vortrag_wklzg_2003 _print.pdf Figure 7a. Danger level 2 - moderate. The arrow and the bar are used to identify the avalanche danger intermediate scenario 2+ of the BM. Figure 7b. Danger level 2 - moderate. The arrow and the bar are used to identify the avalanche danger intermediate scenario 2- of the BM. 9 THANKS We thank Bernard Zenke and AINEVA avalanche forecasting offices for their cooperation. 208