systematic avalanche warning in Switzwhereas the regional level has been intrwithin Avalanche Warning Switzerland 2000national level is represented by SLF. The nbulletin, the swisswide collection of input davalanche warning and severa! auxiliary pare the main focus of the top level. At thelevel local avalanche warnings are subm"order to prevent avalanche accidents. In 0

bridge the gap between local and nauthorities the regional level has been introd

The flow of information especiallysnow cover stability and local estimationsavalanche hazard (see also Brabec et aI.,from the local level to SLF are the kimproving the precision of avalanche wamiorder to motivate and include local forecastthe construction process of avalanche fornew product has been launched for the r:level in 1997: regional avalanche forecastsfigure 1 for an example). In 1996/97 first tregional forecasts have been done inSince then regional forecasts haveintroduced in Grison (1998), Central SwitZ(1999) and Wallis (2000). For winter 200more regions are planned: SOdbunden and

60

Avalanche Warning has been a key taskof the Swiss Federal Institute for Snow andAvalanche Research in Davos (SLF) since it hasbeen set up over half a century ago. Since 1996the way avalanche warning is done has gonethrough a major change: The project AvalancheWarning Switzerland 2000 (Russi et. aI., 1998)has introduced "computer-aided avalancheforecasting" in order to cope with the increasingdemand for more and more precise avalancheforecasts. The state of avalanche warning inSwitzerland before this effort has beendocumented by Meister (1994). Stucki et al. (1998)have presented a 3-level concept for avalanchewarning in Switzerland that is in use since then:The local and national level of avalanche warninghave been in use since the beginning of

• Corresponding author address:Bernhard Brabec, Swiss Federal Institute for Snowand Avalanche Research, FlUelastr. 11, CH-7260Davos-Dort, Switzerland; tel: ++41-81-417 01 56;fax: ++41-81-417 0110; email: brabec@slf.ch

1. INTRODUCTION

ABSTRACT: Besides the text-based and Swisswide national avalanche forecasts regional avaforecasts have been introduced in Switzerland in 1996. Each of the regional forecasts covers an aabout 3.000 km2 and shows the avalanche hazard in graphical form. The regional bulletins are pUb!"at 8 a.m. .RAIFoS is a computer based system that contains• a statistical model (NXD-REG) that forecasts the avalanche hazard for Switzerland. The m

based on the nearest neighbor approach which is applied for 60 manual weather stations. The rof each station are then compiled into a map of the current and future avalanche hazard. The mresults are an additional information for constructing avalanche hazard estimations and areintegrated with BULLED.

• a GIS-based editor (BULLED) for constructing an estimation of the past or current avalanchefor Switzerland. The maps can be edited, copied, printed, exported in all important graphics fand saved iri a relational database.

• a text-based editor (REGBUL) for writing the forecasts for each individual region. Theautomatic stations and the official weather forecast of Switzerland are included automa'Furthermore a set of structured sentences allows the avalanche warner to quickly compilesentences about the current situation. The structure of the sentences allows us to translateautomatically into French. The hazard estimation created with BULLED is used for the region

These three complementary approaches are the basis of Computer Aided Regional Avalanche ForThe data of automatic stations and model calculations of SNOWPACK are further important prereqbut are not presented here.

Bernhard Brabec·, Roland Meister, Norbert Raderschall, Urs Stockli, Andreas Stoffel and ThomasSwiss Federal Institute for Snow and Avalanche Research, Davos Dort, Switzerland

RAIFoS: Regional Avalanche Information and Forecasting System

KEYWORDS: avalanche warning, regional avalanche forecast, statistical models, nearest neighbou

___omdW (V7:00h):

statistical regionalavalanche

NXD·REG ---+ forecast

~.........,

Swisswide avalanche

BULLED f------+forecast (map of current

hazard levels)

-~

..........

regional avalanche

REGBUL f----. forecast

- ..........

-

2. ARCHITECTURE

(1992) focused on computer applications foravalanche warning. Durand et. al. (1999) presentthe state of the French avalanche forecastingchain SAFRAN-CROCUS-MEPRA. McClung et al.(1993) introduce conventional and numericalavalanche forecasting. Gabl et. al. (1994) presentan introduction to avalanche warning in Austria.The current state of avalanche warning in differentcountries can be derived from the annual winterreports of each avalanche warning service.

Russi et al. (1998) have presented thearchitecture of Avalanche Warning Switzerland2000. Here we focus on the Regional ForecastingModule and present details of the implementation.Figure 2 shows the general architecture ofRAIFoS, the Regional Avalanche Information and

Figure 2: General architecture of RAIFoS(Regional Forecasting Module of AvalancheWarning Switzerland 2000).

Forecasting System, containing all modules andinformation paths.

WInd.m...--__

=- ~ -soc ~"::__ HE ~ f-

\

000 __...--=--

Figure 1: Example of a regional forecast for UpperWallis: Nr 115 of winter 1999/2000.

Oberland. After that nearly all Swiss alpine regionswill be covered by regional forecasts.

61

__"' guCeT........-......

IloIT_.mV_ d."••-...--Anotiogdw..--__In-.__und.ll'COn.indio_T ....

_.sind~ .. don MiIIIIgaI1nlon mOgIc:h.

The regional avalanche forecasts arepublished during the main winter season everyday at 8:00 a.m. The main target group of theregional forecasts are out-of-bounds skiers andsnowboarders who usually do not have a lot ofavalanche education. The data of automatic snowand weather stations, precise weather forecastsand the feedback of local avalanche forecastersare compiled to the new bulletin. Additionalinformation comes from SNOWPACK (Lehning etaI., 1998) and NXD-REG (Brabec et aI., 2000).This paper focuses on a system for decisionsupport and the construction of regional avalancheforecasts. Three complementary modules (NXD­REG, BULLED and REGBUL) are the parts of thissystem.

Similar software systems are in use at allmodern avalanche warning services around the~Orld. The main differences are in the amount ofI~tegration that is provided and the number andkind of models that are in use. Williams (1998) haspre~ented an overview of tools and systems thatare In use in North America. Before that Tremper

2.1 NXD-REG

The abbreviation NXD-REG stands for Next-X­Days-Regional. It is a nearest neighbors basedmodel that has been generalized to the task ofregional avalanche forecasting (see Brabec et aI.,2000). The results of 60 nearest neighborcalculations based on the Swiss manual observernetwork are compiled to a map of averagedhazard levels called hazard values and a secondmap of hazard levels. Both maps are presented tothe avalanche warners as a suggestion for a newavalanche forecast. Besides the data of 10 yearsof manual observation stations BULLED has beenused to construct a database of past avalancheforecasts. Both databases are integrated into theofficial SLF snow and avalanche database. Theevaluation of NXD-REG has shown that thesystem is in agreement with the conventionallyestimated hazard levels in about 52% of thecases. It has to be noted that the system does notyet include data about the current snow cover

stability. By downscaling weather forecast mto the location of the weather stations a trueforecast for the next day can be derived.

The results of NXD-REG are directlywritten to the database of BULLED which allothe avalanche forecaster to directly access thOf course NXD-REG is only an extension of thinformation that the Swiss avalanche warningservice uses. The conventional method presby Meister (1994) still plays a predominant rolewill be supported by computer aided avalancheforecasting more and more.

The software implementation of NXD­is based on a combination of database (OracleGIS (ArcView) and statistical software (S-Plus).data, results and parameters of the model arestored within the database. Different versionsthe model and all calculations (even testwiseare also kept there. This allows us to keep traabout the historic development of the system.statistical part is used for doing all statisticalestimations, implementation of the NXD-appr

Figure 3: User interface of bulletin-graph-editor (BULLED).

62

for each station and comparison to other statistical .approaches. The GIS software produces all maps.

2.2 BULLED

The bulletin-graph-editor (BULLED) is atool to construct past, current and futureavalanche hazard maps. Figure 3 shows the userinterface of BULLED. The software has beenimplemented using ArcView and a dat~base forstoring all maps. For each map date, time,authors, origin and type are stored. For the originnational forecasts, regional forecasts, models andquestionnaires are distinguished. For the typeforecast for the next day, now-cast for the currentday, check of the forecast and verification can bechosen. Check of the forecast and verification aredistinguished because one is interested in.checking the forecast on a daily basis whereas averification is usually done several days later whenall available information is available (see Fohn etal.,1995).

Maps can be constructed in two ways:Either a region is selected via a menu or a mapand the hazard estimation for that area is doneafterwards or vice versa. To make the map-basedbulletins compatible to the text-based nationalavalanche forecast a new division of Switzerlandinto subareas of about 50-1000 km2 has beenderived. Figure 3 bottom right shows these 100subareas that can be selected individually byclicking on the map Teilgebiete or via a dynamicmenu (see part Gebietseinteilung of userinterface). Besides the names of the subareas 70additional geographic terms have been introducedand organized into five levels, e.g. CH-Regionencovers the whole alpine area of Switzerland.

For each subarea the hazard estimation isgiven by the hazard level, according to theEuropean Avalanche Hazard Scale, type ofdanger, aspect, height and type of endangeredterrain part. The result of the map construction isshown in the left bottom window. Black and whiteas well as color are supported.

The system is in operational use sinceWinter 1998/99 at SLF. During two years about500 maps of the avalanche hazard for Switzerlandhave been constructed. As a database for NXD­~EG the bulletins of 10 winters have been broughtI~to t~e new formalized form. There were only fewsituations that were difficult to transform especiallywhen the hazard estimation was rather vague anddepended on weather factors.

2.3REGBUL

63

The Regional Bulletineditor REGBUL isused to efficiently construct regional avalancheforecasts for Switzerland. It can easily beextended for new regions that can also overlap.Consistency for the hazard estimations ofsubareas that appear in several regional forecastsis guaranteed. The hazard estimation which is themain part of the product (see Figure 1) is doneusing BULLED. REGBUL has an interface to theBULLED database that allows to fullfill theconstistency requirement.

Figure 4 shows the user interface ofREGBUL. The structure is similar to the documentthat is published (see Figure 1). At the top one canfind the input fields for general information. Next isthe Flash (german and french) and a 3-linetextfield that can be generated using a sentencegenerator. This generator allows the forecaster tocreate the most important sentences for REGBULin german. These sentences are automaticallytranslated into french (for regional bulletinUnterwallis). The sentence generator can easilybe extended to further languages. The weatherforecast from MeteoSwiss is included. The data ofautomatic stations is also automatic input that onlyhas to be checked by the forecasters. Theresulting document is created using ArcView.ArcView automatically reads the input of thehazard map from BULLED where a hazard maphas to be constructed before starting REGBUl.

The dissemination of the bulletin is highlyintegrated. From one matrix of targets (printers,internet, fax and fax on demand) several optionscan be selected.

3. EVALUATION

In this paper we focus on the evaluation of thesystem from a consumer perspective. Theevaluation of NXD-REG can be found in Brabec etaI., 2000. To quantify the interest in products ofSLF two complementary approaches have beenapplied: on the one hand we have submitted aquestionnaire to a group of security personnel andmountain guides to ask them which informationand which products are most useful for decisionsconcerning avalanche questions. This inquiry hasshown that for this group the regional avalancheforecasts have become as important as theswisswide national forecasts within a few years.On the other hand detailed information about theamount of information and products that havebeen requested from SLF have been collected. Inwinter 1996/97 the national avalanche forecastshave been requested 212'386 times from SLF.This number has increased up to 501 '004 in winter

are highly integrated but also offer openinterfaces. The system is in operational usewinter 1999/2000 and has proven to be usefstable. Internet hits and user inquiries indicatthe new products are well accepted and areneed.

For the next years it is planned to acover all areas of Switzerland with regionalbulletins. Furthermore several analysis toolsbe developed to compare different hazardassessments within BULLED. NXD-REG willextended by input data from automatic weastations, snow cover stability and output fromNXD2000 (Gassner et aI., 2000). Furthermoseveral other approaches for statistical modbe applied to the task of regional avalancheforecasting.

F: h _ _ , _. 'w'_._ _ __ .

lurU......br [-Cl II1nd Unital I1ft ,__ IM.) ....-..- S...... .... 1_'_1

105:31 I!85:40,;15:.

2d.t dw __

Figure 4: User interface of the regional bulletin editor (REGBUL).

~ ScMees~~------:----Sc:hneehiihe-[CII) I: Zeit.......... lIS dis (2'tIt) , pI'I ~::=-~- '~:!:I ·-:i+----~li

6a1Aiharrt 2430R. !4)5:31 238 I ·1 j o.!1__ Stolli._ 29100 IOG:31 154; -3'~::

1-..uea 21100 \06:31 19 9' "ii'• Cicrfterret 2t5Oa 05:. 159 .z i 0 ,If

~el Touren und Valianlenabfa!trlen muss der wAtmdledfngte tageszelUIChE! Anstieg der lawinengefa:h l Ibeachtet werden. In stellen, besonnten Rinnen und Mulden, in die fl1scher Trtebschnee elngeblasen j

lWUrde. sind Lawinenausl6sungen ab den Mittagsstunden Ildglich. . j_ .... .•__._.__ ._. .__. .__. .__ . ..__.. ._.._ ..__.•... , ,,_, . ... _.. ..•..... _ .. _ ..._._ .....__.~_...J

~ ~~~.._. ~--Ill.:'''=t=ei='.:'Dob=·.-o=t:..1_....~iA>o~lp=b=e_======~I-Bull_.t!a_IIr_ ..::l§J:::.._::::j_IIot_......:~:::)lon:::...='~_~~. j Zeit~ oUlUc lois~

Iw. s....1 .: ~ __.._. .._---,"_._-_ _----_J F: li..,..__. ._. . :.. .1

jlllle so.- I 0: IT .._ . . .... ..__ ... < • __ .._ __ . ._ j F: IL......__......__... ___.__. .._ _.__.. .. .._.........._... ,IIllle Situ I 0: k.. . . __ ..._.... .._..... __ .._ _. F: 11...__..._. . _......._ __

IlIlle·Si 10: lC. _ _ _____ _.. __ _,..___1 F: li __ __, , _ , ..__ .....J

i l'lllle so.-I .: [:::==-..:=-...::.....: -=::·-==::·-,~==:=~:--~~~~:~=-:.::-===:1-·-------:- .._·-.JI

4. CONCLUSIONS AND OUTLOOK

In this paper three complementarysoftware packages for regional avalancheforecasting are described: NXD-REG is a nearestneighbor based statistical model that allows theSLF to calculate a forecast for the complete alpinearea of Switzerland. BULLED includes the resultsof NXD-REG and is used for constructing maps ofthe current avalanche hazard. REGBUL is aneditor for editing Swiss regional avalancheforecasts for several regions. All three packages

1999/2000. At the same time the requests forregional forecasts increased from 0 to 277'557. AllSLF products have been requested 1'453'630times by internet, fax, InfoBox (see Russi et aI.,1998) or telephone during winter 1999/2000. 80%of the information have been disseminated byinternet.

64

5. ACKNOWLEDGEMENTS

We would like to acknowledge the support of theSwiss National Science Foundation, the Swissgovernment and the Board of the Swiss FederalInstitutes of Technology. Furthermore, a numberof people at SLF and ETH are involved insupporting the development.of the models a~ddelivering input data. In particular we would like tothank Walter Ammann, Othmar Buser, Paul Fohn,Martin Gassner, Martin Machler, Tom Russi,Martin Schneebeli, JOrg Schweizer, WernerStahel, Thomas Wiesinger and Martin Zimmerli fortheir contributions and discussions.

6. REFERENCES

Brabec, B., and Stucki, T. 1998. Verification ofavalanche bulletins by questionnaires.Proceedings 25 Years of Snow AvalancheResearch, Voss, 12·16 May 1998, 79-83.

Brabec, B., and Meister, R. 2000. A NearestNeighbor Model for Regional AvalancheForecasting. Annals of Glaciology, Vol 32,Innsbruck, accepted.

Durand, Y., Giraud, G., Brun, E., Merindol, L. andMartin, E.,. 1999. A computer-basedsystem simulating snowpack structures asa tool for regional avalanche forecasting.Journal of Glaciology, Vol 45, No 151,469·484.

Fohn, P. and Schweizer, J.,. 1995. Verification ofavalanche danger with respect toavalanche forecasting. Int. SymposiumSciences and mountain - The contributionof scientific research to snow, ice andavalanch~ safety, Chamonix.

Gabl, K., and Lackinger, B., editors. 1994.Lawinenhandbuch. Tyrolia Verlag.

Gassner, M., Birkeland, K. and Leonard, T. 2000.NXD2000: An improved avalancheforecasting program based on the nearestneighbor method, this issue.

Lehning, M., Bartelt, P., Brown, B., RuSSi, T.,Stockli, U. and Zimmerli, M. 1999.SNOWPACK model calculations foravalanche warning based upon a newnetwork of weather and snow stations.cold regions science and technology, 30,pages 145-157.

65

McClung, D. and Schaerer, P. 1993. Theavalanche handbook. The Mountaineers.

Meister, R. 1994. Country-wide AvalancheWarning in Switzerland. ISSW'94Proceedings, 58-71.

Russi, T., Ammann, W., Brabec, B., Lehning, M.and Meister, R. 1998. Avalanche WarningSwitzerland 2000. ISSW'98 Proceedings,146-153.

Stucki, T., Ammann, W., Meister, R. and Brabec,B. 1998. A new concept for avalanchewarning in Switzerland. Proceedings 25Years of Snow Avalanche Research Voss,12-16 May 1998, 278-282.

Tremper, B. 1992. Computer applications foravalanche forecasting. ISSW'92Proceedings, 105-115.

Williams, K. 1998. An overview of avalancheforecasting in North America. ISSW'98Proceedings, 161-169.