Case studies with SAMOS comparison with observed avalanches

Transcription

1 Case studies with SAMOS comparison with observed avalanches Rudolf Sailer Department of Avalanche and Torrent Research Federal Office and Research Centre for Forests (BFW) Hofburg, Rennweg 1, A-6020 Innsbruck Abstract SAMOS (Snow Avalanche MOdelling and Simulation) is an advanced tool for hazard mapping and is already in operational use at the Forest Technical Service for Avalanche and Torrent Control in Austria. The SAMOS-software includes a Pre-/Postprocessor, which is an adaption of the FIRE V.7 Pre-/Postprocessor, and a CFD-simulation-program based on the FIRE V.7 solver. Snow avalanches typically comprise two-layers: a powder snow layer on top of a dense-flow-layer. The powder snow layer is computed in 3D with the FIRE solver, the dense-flow-layer is computed in 2D with an embedded shallow-water-model, specifically developed for SAMOS. Case studies with observed and well documented (artificially or naturally released) avalanches are used to check the reliability of the model. Furthermore enhancements of the model are based on these studies. Simulation results of avalanches in the Alps (Austria, Wolfgruben avalanche, St. Anton a/a, Tyrol; Swiss, Creta Besse avalanche, Vallee de la Sionne, Canton du Valais) and Norway (Ryggfonn) are presented in this paper. The results are compared with observed (measured, mapped) parameters such as velocity, deposit distribution, deposit depth or impact pressure. Introduction First written information on avalanches (in Caucasian mountains) are available from ancient Greeks and Romans (about 60 B.C.). In the European Alps avalanches threatened mainly settlements and travellers. In the 14 th century for example a hospice was founded to harbour travellers in St. Christoph / Arlberg (Tyrol, Austria) after an avalanche killed seven people. In Austria the first illustration of an avalanche were done by Schäufelein in the year 1517 (Figure 1). Figure 1: The first avalanche illustration in Austria (Knight Theurdank is threatened by an avalanche close to Innsbruck, Tyrol, Austria; xylograph by Schäufelein, 1517; Archiv Ferdinadeum). At the end of the 18 th century (1773) the Jesuit Josef Walcher already divided avalanches in Grund- und Staublähnen. These are the German expressions for dense flow avalanches and powder avalanches. During the First World War ( ) thousands of soldiers were victims of avalanches. In 1916 after a snow fall period of one week approximately 6000 soldiers died within 48 hours due to avalanches. In more recent years two catastrophic winter occurred in the European Alps (1950/1951 and 1953/1954). Several hundred people were killed by avalanches in Switzerland and Austria. The last avalanche winter occurred in 1999 in these two Alpine countries and in France. The worst accident happened in Galtür (Tyrol, Austria) with 31 victims (Figure 2). A more detailed list of the history of avalanches can be found in the Austrian Avalanche Handbook (Land-Tirol, 2000)

2 Figure 2: Cleanup efforts in Galtür after the avalanche disaster occurred on 23 rd of February 1999 (Photo: Martin Klimmer) Figure 3: By a powder snow avalanche destroyed chair lift (1999, Photo: Martin Klimmer) Generally, the increasing exploitation of alpine regions by tourism (Figure 3), traffic and power industry exposes human lives, buildings and roads to snow avalanches in the European Alps. In view of this evolution big efforts have been made to reduce avalanche danger during the last decades. Between 1949 and 1989 an amount of 350 million EUR was spent for defence structures (Figure 4), supporting constructions or protective dams in Austria. These efficient but cost and time intensive permanent measures (snow-stabilizing works, deflecting dams, catching dams, tunnels, hazard zone mapping, etc.) are increasingly complemented by temporary measures (avalanche warning service, avalanche commissions, road blocks, risk and crisis management, artificial release of avalanches, etc.; Sailer, 2001a; Sailer, 2001b) and supported by numerical simulations (e.g. SAMOS, Snow Avalanche MOdelling and Simulation). A B Figure 4: Defence structures (Wolfgruben avalanche overview A, detail B with a divided catching dam) beyond St.Anton am Arlberg, Tyrol, Austria (Photos: Lambert Rammer, BFW) Due to the climatic conditions dry avalanches occur quite frequently in the Alps and the catastrophic avalanches mentioned above are of this type, so simulation activities in Austria have been focused thereon. Currently widely used models as statistical run out models that are restricted to avalanche tracks that fit into different categories, or one-dimensional centre of mass models are very simplified and can not describe the deformation of the avalanche body. Additionally, the avalanche path has to be prescribed by the user. Therefore, two- and three-dimensional models based on fundamental fluid mechanics gained in importance throughout the recent years. Two- or (higher-) dimensional models (two dimension tangential to the terrain surface) are able to predict the avalanche track as well as lateral spreading and run out distances and thus provide additional criteria for their verification

3 SAMOS short model description Dry snow avalanches are assumed to consist of small, non-cohesive ice particles. Every avalanche starts as a dense flow avalanche (Figure 5) with flow densities ranging from 100 to 400 kg m -3. The momentum is transferred by particle contacts and collisions. On steep terrain after a few seconds a powder part layer with relative small particles of ice suspended in air is formed (Figure 6). The time delay is controlled mainly by the quality of snow (density, liquid water content) and the topography (inclination of slope, velocity). In the powder layer with generally low densities (< 15 kg m -3 ) the momentum is transferred by the viscosity and turbulence of the air. Due to the different formation and behaviour of the dense flow part and the powder snow layer two different models are used in SAMOS. The turbulent flow of the powder-snow is computed by the standard FIRE V.7 solver, using the κ-ε -turbulence model. The variable snow-concentration in the powder cloud is mapped to a FIRE-passive-scalar. This way, an additional mass-balance for the snow is solved for. The variable density of the powder-snow - air-mixture is computed from this scalar by a user-density-function. The dense-flow-layer is typically very shallow, with flow depths up to a few meters and length of 100 meters or more. A shallow-water-model with a Mohr- Coulombian rheology-model for the dense-flow-snow was specifically developed for SAMOS and coupled with the 3D-flow. The avalanche starts as pure dense-flow, with snow concentrations in the 3D powder-part being zero. Depending on the shear force between the 2D-dense-flow and the 3D-flow above, snow mass and momentum is transferred from the dense-flow to the powder-flow (Figure 7), resulting in mass and momentum sources for the 3D-flow. This means that SAMOS computes the amount of powder-snow generated by itself. A detailed description of SAMOS can be found in (Sampl and Zwinger, submitted; Sampl et al., 2000). Figure 5: Slab release by explosives. The avalanche starts as a dense flow avalanche. (Photo: Tobias Hafele, Arlberger Bergbahnen) Figure 6: After few seconds the dense flow avalanche forms a powder snow layer. Photo: Tobias Hafele, Arlberger Bergbahnen) Figure 7: Dense flow part and powder part are coupled by by a simple transition model (Sampl et al., 2000) SAMOS was developed for the Ministry of Agriculture and Forestry of Austria (BMLFUW) by AVL-List GmbH (Graz, Styria), the Department of Fluid Dynamics and Heat Transfer (Vienna University of Technology), the Service for Torrent and Avalanche Control (Austria) and Austrian Department for Avalanche and Torrent Research (BFW, Innsbruck, Tyrol) and is under further development. The latest model version allows the definition of entrainment areas with the specification of snow depth and density. Therein the entrainment process considered is not erosion - 3 -

4 at the sliding surface of the avalanche but a ploughing-in of the resting snow in front of the dense flow avalanche layer (for more details concerning the entrainment module cf. Sailer et al., 2002). Recalculation of observed avalanches The validation and verification of numerical models is very important to create an interface between model developers and practitioners. At the Department of Avalanche and Torrent Research different studies have been carried out to estimate the reliability and accuracy of SAMOS. The practical knowledge of these studies leads to intense discussions with practitioners (e.g. Forest Technical Service for Avalanche and Torrent Control) and contains useful information for the further development and improvement of the model. Below three examples are presented compendiously. In Sampl et al., 1999 and Sampl et al., 2000 the results of calculations with particular objective targets are shown. In the last two years sensitivity analyses have been carried out. Therein the purpose was focused on i) the main driving input parameters (snow mass, particle diameter and Figure 8: Dense flow pressure (red denotes high pressure) of the Wolfgruben avalanche: red line mapped dense flow impact, yellow line mapped avalanche impact (powder layer), arrows direction of flow part. Figure 9: Powder peak pressure (orange denotes high pressure) of the Wolfgruben avalanche: red line mapped dense flow impact, yellow line mapped avalanche impact (powder layer), arrow direction of powder layer; Figure 1 Tschol, Figure 2 Zangerl. snow density of the dense flow part) and ii) the influence of different digital elevation models on simulation results (Klebinder and Sailer, submitted; Schmidt et al., 2003). Both studies have been based on four avalanches that were used to calibrate and to verify the simulation model at the early stages of development. These catastrophic (victims and damages) avalanches occurred in the years 1984 and 1988 in the western part of Tyrol. Release areas, flow paths and depositions were mapped or reconstructed precisely. Furthermore, the forces that caused the damages on buildings were recalculated. Figure 10: Snow stabilizing works at the release area of the Wolfgruben avalanche (St. Anton a/a, Tyrol, Photo: Rudi Sailer). For instance Figure 8 and Figure 9 show the result of the SAMOS recalculation of the Wolfgruben avalanche (St. Anton a/a, Tyrol) that occurred on March 13 th 1988 early in the morning. Approximately 52 kilo tons were released and reached the eastern part of St. Anton a/a as a catastrophic avalanche that killed seven persons and destroyed 31 buildings (two of them totally). The calculated deposits and impacts are good in agreement with the observed ones. At two buildings impact pressures were calculated (Tschom, 1988). At Haus Zangerl the pressure is between 13.6 kpa and 14.4 kpa

5 SAMOS gives a pressure of 15 kpa at that point and is good in agreement with the back calculated values. Based on that promising results SAMOS was used to find the best arrangement of snow stabilizing works at the release area of the Wolfgruben avalanche (cf. Figure 4 and Figure 10). Figure 11: Front velocities from radar measurement and SAMOS calculation along the avalanche track On 10 th February 1999 a joint experiment was carried out by the Swiss Federal Institute for Snow and Avalanche Research (SFISAR) and the Austrian Department of Avalanche and Torrent Research to measure forces and velocities at the full scale experimental site Creta Besse in Vallee de la Sionne, Canton du Valais, Switzerland. A huge avalanche could be released artificially. The results of the velocity measurements from the pulsed Doppler avalanche radar and the recalculation with SAMOS are explained in (Sailer et al., 2002) in detail. A description of the experimental site as well as an introduction to the frequency pulsed Doppler avalanche radar (Rammer, 2000; Rammer et al., 1998) can be found therein too (cf. also Issler, 1999). The comparison of the measured velocity values of the pulsed Doppler radar are good in agreement with the recalculated values given by SAMOS. The measurements derived from the radar differ less then 10 % from calculated values using a global entrainment depth of 1.0 m. Figure 12: Velocities calculated with SAMOS against observed velocities (figures refer to date of avalanche and version) Reducing the entrainment depth for example to 0.5 m increases the deviation to measured values up to approximately 20 % (Figure 11). Neglecting entrainment leads to large differences between the results derived from Doppler radar and SAMOS values. The velocities are generally by far to low (> 10 m s -1 ) in this context

6 Twelve well documented dry snow avalanches from the Ryggfonn avalanche path in western Norway were selected for back calculations with several dynamic avalanche models (Issler et al., in prep.). The study is part of the EU project SATSIE (Avalanche Studies and Model Validation in Europe). Nearly all analysed avalanches show a net entrainment between release and deposit up to 220 %. The first SAMOS simulations with standard parameters (bed friction angle δ = 16 degree, internal friction angle φ = 35 degree, dynamic friction coefficient c = 0.022, flow density 200 kg m -3, particle size 1 mm) showed discrepancies compared with the observed deposit distributions. Although the velocities showed a better agreement, the results were not satisfying at all. Therefore, calculations with different bed friction angles and including entrainment have been carried out. The aim was to find a parameter combination (particularly bed friction angle, net entrainment) that leads to more realistic results concerning velocity and deposit distribution (cf. Table 1 and Figure 11). The avarage difference from the calculatet to the observed run out lengths is -10 m (Std. deviation 35 m). The calculated front velocities differ on an average -3.0 m s -1 from the measured values (Std. deviation 4.1 ms -1 ). The largest deviation of the run out length (-100 m) can be found at the last line of Table 1 referring to the avalanche occurred on February 17 th The observed run out distance of that avalanche indicates the reach of the saltation layer and it is not surprising that the models used in the mentioned study generally and SAMOS particularly can not reproduce such a run out with reasonable parameter values (Issler et al., 2003). The calculated velocity of the event of January 10 th 1983 is interestingly 10 m s -1 to low whereas the simulation with the standard parameter set and without entrainment lead to a correct velocity but to a to long run out length. An important feature of the Ryggfonn site is a catching dam (16 m high and 100 m wide). The dam is overflowed quite often but retains a considerable fraction of avalanche mass. The deposit distributions (percentage of mass stopped above the dam) is quite good in agreement with the observed values. But, SAMOS does not, as well as the other dynamic simulation models used in the study, reproduce all avalanches with appropriate accuracy. Table 1: Ryggfonn SAMOS calculation results Summary Although high quality field data are rare some back calculations could be done to check the reliability of SAMOS during the last years. The analyses of catastrophic avalanches (e.g. Wolfgruben avalanche in this paper) support important information on pressures needed to destroy objects, distribution of deposits or impact of avalanche dynamics generally. The results of the - 6 -

7 Wolfgruben avalanche are good in agreement with the observed values. Both, the calculated distribution of the deposits and the effect caused by the powder snow avalanche are close to the reality. SAMOS pressure values are also good in agreement with the analysed forces that were necessary to damage or destroy buildings. The design of instrumented full scale test sites gains in importance. In this paper the result of two different test site studies are shortly described. Both studies with different main focuses yield to reliable results. The recalculation of the Creta Besse avalanche was emphasised on the potential influence of entrainment to the simulation results. It is shown that the disregard of entrainment does not lead to satisfying simulation results. During the recalculation of the Ryggfonn avalanche the main focus was to find the correct bed friction angle to reproduce the observed run out lengths and velocities. The bed friction angle is the driving internal parameter that influences the run out length as well as the velocity and has been retained unchanged at 16 degree (corresponds to a dry friction coefficient of 0.29) until now. The calculation of the twelve Ryggfonn avalanches shows only in one case (February 17 th 2000) the best result with a bed friction angle of 16 degree. Only the simulation of the avalanche occurred on March 27 th 1993 lead to a reliable result using a small bed friction angle (15 degree). Generally the best results were reached with relatively high dry friction values (> 19 degree). The average bed friction angle is 20 degree with a Standard deviation of 5 degree. These figures are good in agreement with the results obtained by the other simulations models used in the Ryggfonn study (Issler et al., 2003). At the Ryggfonn test site also the net entrainment is estimated generally. The net entrainment between release mass and deposit vary from -15 % to 220 %. The corresponding entrainment was included in the SAMOS calculations by using global entrainment with specified depths and densities, whereas the densities are predetermined values from observations. The entrainment depth was estimated to approximate the given net entrainment mass. The consideration of entrainment improved the results particularly the deposit distribution and velocities. A detailed analysis of the influence of the entrainment is not finished yet. But, it seems to be obvious that both bed friction angle and entrainment affect and improves the result of avalanche calculation significantly. Conclusions Although the mentioned SAMOS simulation results are auspicious on the whole, congruence with nature is not possible. There are a lot of reasons. For example the model inherent simplifications allow only a best fit. But, the determining factors such as digital elevation models (summer terrain vs. winter terrain, modified terrain due to previous avalanches, etc.), release areas, release depth or release densities are significant elements of uncertainty. The difficulties of the determination of release areas, release depths and densities are well known by snow and avalanche experts. In that sense the handling of sophisticated avalanche simulation models such as SAMOS require special accurateness and is generally on the responsibility of the experts (Service for Torrent and Avalanche Control, civil engineers, etc.). Particularly the pre calculation (calculation of a less known avalanche in order to create hazard zoning, etc.) of avalanches increases these difficulties. Directives and guidelines are helpful but sometimes not up to date and aware of the ability of modern simulation technologies. Back calculations in the sense of validation and verification of avalanche simulation models of observed (catastrophic) avalanches and scientific studies on full scale test sites are important tools to link model developers and practitioners. The results of such tests show the limits and the application area of a specific model. Furthermore, it is possible to support practitioners with advices of reasonable parameter combinations based on observations. Validation studies are also an excellent tool to figure out the difficulties that are results of model architecture and those that might be attributed to determining factors. To introduce new model components, as this was the case with the entrainment module of SAMOS, a careful validity check has to be passed prior to an operational use of the targeted module and its interaction with the entire model. An increasing acceptance of avalanche simulation models is addicted to a careful validation and verification - 7 -

8 throughout their life time. Therefore, back calculations as shown in this paper exemplarily are essential instruments to improve and extend (e.g. with a module that allows to define secondary release areas) more dimensional dynamic avalanche models. Acknowledgements Peter Sampl (Advanced Simulation Technologies, AVL-List, Graz) wrote the helpful comments on the FIRE V.7 solver. The Ryggfonn study is partly founded by the European Commission through the project SATSIE, contract EVG and the Norwegian Geotechnical Institute. References Issler, D., European Avalanche Test Sites, Overview and Analysis in View of Coordinated Experiments. Mitteilungen 59, Eidgenössisches Institut für Schnee- und Lawinenforschung, Davos. Issler, D. et al., in preparation. Simulations of observed dry-snow avalanches in the full-scale test site Ryggfonn, Norway., in preparation. Klebinder, K. and Sailer, R., Sensitivitätsanalyse des Lawinensimulationsmodells SAMOS, Interpraevent 2004, Riva sul Garda. Land-Tirol, Lawinenhandbuch. Tyrolia, Innsbruck, 260 pp. Rammer, L., Velocity measurements of avalanches by a pulsed Doppler Radar, IUFRO, Natural Disasters, St. Christoph a/a, Tyrol, Austria. Rammer, L., Kristensen, K., Lied, K., Schreiber, H. and Randeu, W., Radar measurements of snow avalanche full scale experiment in Ryggfonn. In: E. Hestnes (Editor), 25 Years of Snow Avalanche Research. Norwegian Geotechnical Institute, Oslo, Voss May 1998, pp Sailer, R., 2001a. Risk Assessment and Crisis Management for a winter tourist resort (St.Anton a/a, Tyrol, Austria) - a case study (Part I). Proceedings of 21st annual ESRI International User Conference. San Diego, USA., Sailer, R., 2001b. Risk Assessment and Crisis Management for a winter tourist resort (St.Anton a/a, Tyrol, Austria) - a case study (Part II). Proceedings of the First Annual IIASA-DPRI Meeting on Integrated Disaster Risk Management: Reducing Socio-Economic Vulnerability. IIASA, Laxenburg, Austria., Sailer, R., Rammer, L. and Sampl, P., Recalculation of an artificially released avalanche with SAMOS and validation with measurements from a pulsed Doppler radar. Natural Hazards and Earth System Sciences, 2: Sampl, P. and Zwinger, T., submitted. Avalanche Simulation with SAMOS. Annals of Glaciology. Sampl, P., Zwinger, T. and Kluwick, A., SAMOS - Simulation von Trockenschneelawinen. Wildbach- und Lawinenverbau, 63(138): Sampl, P., Zwinger, T. and Schaffhauser, H., Evaluation of Avalanche Defense Structures with the simulation Model SAMOS. Rock and Soil Engeneering, 1/2000: Schmidt, R., Heller, A. and Sailer, R., Die Eignung verschiedener digitaler Geländemodelle für die dynamische Lawinensimulation mit SAMOS. In: J. Strobl, T. Blaschke and G. Griesebner (Editors), Angewandte Geographische Informationsverarbeitung XV. Beiträge zum AGIT-Symposium Salzburg 2003, Salzburg, pp Tschom, H., Lawinenabgang Wolfsgrubenlawine im Gemeindegebiet St. Anton - Nasserein am : Berechnung der statischen Lawinenlast., Gebietsbauleitung Oberes Inntal des Forsttechnischen Dienstes für Wildbach- und Lawinenverbauung., Imst