Analysis of landslide inventories for accurate prediction of debris-flow source areas
Jan Blahut a ,b ,⁎,Cees J.van Westen c ,Simone Sterlacchini b
a Department of Environmental and Territorial Sciences,University of Milano-Bicocca,Piazza della,Scienza 1,20126Milan,Italy
b Institute for the Dynami
c of Environmental Process,National Rearch Council,(CNR-IDPA),Piazza della Scienza 1,20126Milan,Italy
c
International Institute for Geo-Information Science and Earth Obrvation,ITC,Hengelostraat 99,P.O.Box 6,7500AA,Enschede,The Netherlands
a b s t r a c t
a r t i c l e i n f o Article history:
Received 9June 2009
Received in revid form 18February 2010Accepted 23February 2010Available online 1March 2010Keywords:
Landslide inventory Debris flow
Susceptibility analysis GIS
Italian Alps
For the generation of susceptibility maps on medium scales (1:25,000to 1:50,000)using statistical techniques,a reliable landslide inventory is needed,together with factor maps ud as inputs.This paper compares landslide susceptibility maps obtained with the same methodology but using different landslide inventories:the of ficial Italian landslide inventory GeoIFFI for the Lombardy Region and a recently mapped inventory (DF2001).The analysis included four main steps:(i)preparation of debris flow inventories using both random and spatial partitions and factor maps as explanatory variables;(ii)calculation of accountability and reliability indices for a preliminary susceptibility analysis and lection of an appropriate combination of the factor maps for detailed an
alysis;(iii)evaluation and validation of the obtained susceptibility maps;and (iv)comparison of the results and lection of the final map.The study area is located in the Valtellina Valley in the Central Italian Alps.The analysis identi fied highly susceptible areas of shallow landslides that may generate debris flows.It was demonstrated that more precily delimited source areas for landslide-induced debris flows produce better susceptibility maps.However,the improvement of the maps was relatively limited when the inventories were randomly subdivided.Higher improvements were obrved after the subdivision of the inventories into three geographical parts with different geomorphological characteristics.Although the modelling showed very similar results if evaluation is made using standard techniques,the spatial pattern of the susceptibility maps was highly variable and dependent on the combination of the factor maps ud.
©2010Elvier B.V.All rights rerved.
1.Introduction
Landslides are among the most signi ficant natural damaging events in mountain environments.They are one of the primary caus of property damage,loss of life and injuries of persons.To better predict future occurrences of landslides and improve protection,hazard or susceptibility analys are
performed.Landslide susceptibility analysis (or so called spatial probability of landslide occurrence)using statistical techniques is bad on the asssment of terrain conditions in an area subjected to previous landslides (Carrara et al.,1995).The conditions that caud the landslides are assumed to be the same for future landslides.Such a landslide susceptibility analysis on a medium scale (1:25,000to 1:50,000)has been ud as one of the first steps in landslide hazard asssment (Remondo et al.,2005;Fell et al.,2008).The performance of the models can be effectively evaluated and the prediction power of the models could be validated using techniques such as ROC plots,or success and prediction rate curves with areas under curves (Chung and Fabbri,1999,2003;Beguería 2006).
A main problem in landslide hazard asssment is the de finition of magnitude and frequency of prospective events.Although there are many methods for landslide susceptibility asssment,only a few techniques convert the result into landslide hazard maps bad on temporal probability asssment.One of them is the u of event bad inventory maps.The return periods of landslide triggering events are ud for asssing temporal probability which is then combined with size and spatial probabilities generated from event-bad inventories (Guzzetti et al.,2006a ).However,only a few complete landslide inventories are available.Italy is one of the countries where such inventory databas have been made in a consistent manner.
This study focus on the mapping of source areas of landslide-induced debris flows in the Valtellina Valley.According to Crosta et al.(1990),the majority of debris flows in the study area originate from soil-slips or shallow slides.They usually leave broad sheet-like scars which are easily recognizable on aerial photographs.
Statistically bad susceptibility asssment for the source areas of landslide-induced debris flow was performed using different land-slide inventories in order to evaluate the effect of the accuracy of the input data on the prediction capabilities of the resulting susceptibility maps.The same input data and analytical methods were ud for both inventories.This study also evaluates the improvement of the
篮球队员位置名称Geomorphology 119(2010)36–51
女性的声音⁎Corresponding author.Department of Environmental and Territorial Sciences,University of Milano-Bicocca,Piazza della,Scienza 1,20126Milan,Italy.Tel.:+390264482854;fax:+390264482895.
E-mail address:jan.blahut@unimib.it (J.
Blahut).0169-555X/$–e front matter ©2010Elvier B.V.All rights rerved.doi:
10.ph.2010.02.017
Contents lists available at ScienceDirect
Geomorphology
j o u r n a l h om e p a g e :w w w.e l s e v i e r.c o m /l oc a t e /g e o m o r p h
predictions when the area is divided into geomorphologically homogeneous zones.2.Study area
The study area,the Valtellina valley (Fig.1),is a typical alpine valley located in the Lombardy Region in northern Italy.The valley has U-shaped transversal pro files derived from Quaternary glacial activity.The axis of the valley corresponds to the Adda River,flowing through the towns of Bormio,Tirano and Sondrio to the Como Lake.The valley has prevalently an E –W orientation from Dubino to Teglio,where it enters the study area and takes an NE turn for a few kilometres,and then turns almost to N after Grosio.The orientation of the valley is related to the location of a regional fault that parates the proper Alps (the Austroalpine,Penninic and Helvetic nappes)from the Variscan bament of the Southern Alps.This Periadriatic Fault (or so called Insubric Line or Tonale Fault)runs on the northern slopes of the valley,some 500m above the Adda river floodplain.The bedrock in the valley is mainly compod of metamorphic rocks (gneiss,mica schist,phyllite and quartzite)and intrusive rock units,with subordinate dimentary rocks.Due to the proximity of the fault,cataclastic and mylonitic zones are prent.The alluvial plain of the Adda River is up to 3km wide,and alluvial fans at the outlet of tributary valleys can reach a considerable size,with a longitudinal length up to 3km.
The study area lies in Consortium of Mountain Municipalities of Valtellina di Tirano,an area of about
450km 2.Its territory is subdivided in 12municipalities and has about 29,000inhabitants,mainly on the valley bottom.The northern part of the study area is compod mainly of gneiss,while in the south micaschists and dimentary rocks dominate.Both flanks of the valley are covered by morainic diments and colluvial deposits of variable thickness.The bottom of the valley is covered by fluvial diments.The lowest altitude in the study area is about 350m a.ar San Giacomo di Teglio where the Adda River flows out from the study area.The highest elevation is reached in the northern part of the study area on Cima Viola:3370m a.s.l.
The Valtellina Valley has a long history of inten and extensive landsliding.A large percentage of landslides are reprented by rainfall-induced small slides and soil slips which are the sources for debris flows (up to 1.5m thick),with volumes ranging from a few to thousands m 3(Crosta et al.,1990,2003).The phenomena
affect
Fig.1.Simpli fied geomorphological map of the study area of the Consortium of Mountain Municipalities of Valtellina di Tirano.The location of the study area in the Lombardy Region and in Italy is shown on the right.
皮肚37
J.Blahut et al./Geomorphology 119(2010)36–51
cultivated areas,cau the interruption of transportation corridors and disrupt inhabited areas,sometimes leading to temporary evacuation of people.The study area suffered from inten rainfall and conquent landslides veral times in the past.The major events occurred in 1983,1987and 2000.
In 1983a vere precipitation event triggered more than 200shallow landslides and debris flows between Tirano and Sondrio,with a density of 60landslides per km 2and causing 17casualties (Cancelli and Nova,1985).Two major storms were obrved on May 14th to 16th and May 21st to 23rd,1983,in the Valtellina valley.In Aprica a cumulated precipitation of 453mm was measured which
corresponds to 34%of the total annual precipitation.The average cumulative rainfall for the whole event was in the order of 260mm (Guzzetti et al.,1992).The most affected part of the study area was in Trenda (a part of Teglio Municipality)where a debris flow caud 14casualties.
Another ries of major event occurring in July 1987claimed 12lives and triggered veral hundreds of soil slips and debris flows (Crosta,1990;Crosta et al.,2003).The main rainfall event occurred on July 17th to 19th and was marked by increasing rainfall intensity (Guzzetti et al.,1992).Unfortunately a complete landslide inventory was not compiled,mainly becau of the constraints of time and resources (Guzzetti et al.,1992).Aerial photographs after the 1987event are limited to a narrow strip along the Adda River and do not allow us to map the source areas of the landslides.
Landslides affecting the Valtellina Valley on November 14th to 17th,2000,were mostly concentrated on terraced slopes ud as vineyards.A prolonged inten rainfall event triggered 260shallow landslides on an area of 270km 2.The highest landslide density was obrved around Bianzone,with 49landslides per km 2,and near Tirano with 26.8landslides km −2(Crosta et al.,2003).
In order to capture different characteristics of debris flow source areas in the different places (Fig.2)of the studied area,the landslide inventories were divided in three subts:Northern,Central and
Southern parts (Fig.3).The Northern subt lies in the Val Grosina Valleys,which are two tributary valleys belonging to the Rhaetic Alps.They reprent the highest altitudes in the area with typical alpine relief.Glaciers played a major role in the development of the morphology and they are still prent in a limited area in the highest altitudes.The majority of this area is underlain by gneiss bedrock.Rock glaciers and landslide deposits are typical in this part of the territory.The central subt lies on both flanks of the Valtellina Valley.Slopes are covered mainly by moraine deposits.On both flanks Pleistocene glacial terraces are prent.The Southern subt covers five parallel valleys among which Val Belviso is the largest one.The southern territory is mainly compod of micaschists but dimentary rocks are also prent in the southernmost part.Geomorphologically it is part of the Orobic Alps.
电脑桌面图标怎么随意摆放
3.Materials and methods
In order to compare susceptibility maps created from different inventories,the methodology prented in Fig.4was applied.First,the existing landslide inventory was compared with a newly generated inventory.Bad on an initial t of factor maps,the accountability and reliability indices were estimated to choo different combinations of factor maps as inputs for Weights-of-Evidence (WofE)modelling.The two landslide inventories were randomly and spatially subdivided and WofE m
odelling was applied to create susceptibility models.The model performance and prediction power of the susceptibility maps was assd using success and prediction rate curves with corresponding areas under curves (Chung and Fabbri,1999,2003).Afterwards,highly performing models were compared and their spatial variability was assd.Finally,the best model was chon to create the final debris flow susceptibility
map.
Fig.2.Photographs of debris flow scarps in the study area.A:debris flow scarp in the Val Grosina Valley in highly fractured gneiss;B:Scarp in moraine deposits in the central part of the study area;C:debris flow in unconsolidated colluvial diments in the Val Belviso Valley.Locations of the photos are shown in Fig.3.
38J.Blahut et al./Geomorphology 119(2010)36–51
3.1.Landslide inventories
There are three of ficial landslide inventory databas available for the study area:
•The AVI databa:A bibliographical and archive inventory of landslides and floods in Italy (Guzzetti et al.,1994),which is updated regularly.The AVI Databa was originally designed to inventorize all places in Italy which were affected by landslides or floods.No spatial scale was de fined for this databa,and the information was visualized as points with coordinates.
•The regional databa of landslides of the Lombardy Region (Lombardy Region,2002),mapped at 1:10,000scale.This databa has been compiled since 1998and is systematically updated.
•The GeoIFFI landslide inventory databa for the Lombardy Region (GeoIFFI,2006),which is part of the IFFI National Databa.Unfortunately,there is only limited information about debris flow source areas in the AVI and Lombardy Regional databas.In the AVI databa there are a total of 80events within the study area,but only 12of them are classi fied as debris flows and only three events have a preci date (day,month and year).In the Lombardy databa there are 501events within the study area,of which 46are classi fied as debris flows,but only ven have information about the exact date of occurrence.Another problem with the databas is related to the spatial location of the events.Both databas have coordinates showing event locations but only in some cas the points are located in the scarp areas,and mostly they are in the transport or deposition areas.
The GeoIFFI databa was made by incorporating the two previously mentioned inventories.The databa consists of different types of landslides such as debris flows,earth flows,shallow landslides,and deep ated gravitational slope deformations mapped by points,lines,and polygons.Unfortunately,there is no information on the time of occurrence of the debris flows;thus,it is impossible to divide the inventory into temporal subts.For this study,only debris flow scarp areas mapped as points were considered becau there are only a few debris flows mapped as polygons in the databa.Moreover,in the ca of polygons and lines,the scarp areas may not be clearly distinguishable from the rest of the flow.
The GeoIFFI inventory contained 1478landslide scarps.Becau the inventory included mistakes with the positions of the scarps,we decided to make a new inventory (abbreviated as DF2001).We prepared this inventory by the interpretation of aerial photographs taken in 2001.A total of 573landslide scarp polygons (with a
侃侃滴答
total
Fig.3.Three subdivisions of the debris flow inventories.1–3:Locations of the scarps in Fig.2.Rectangle shows the extent of Fig.5.
39
J.Blahut et al./Geomorphology 119(2010)36–51
area of 4.4km 2)were mapped.The ILWIS software (ITC,2009)was ud for the preci delimitation of the scarp polygons using the aerial photographs and a DEM.The DF2001inventory has veral advan-tages compared to the GeoIFFI databa such as the u of polygons as mapping units instead of points of GeoIFFI databa (Fig.5)and the exclusion of the debris flow scarps that were initiated on anthropo-genic terraced terrain due to the collap of man-made dry stone walls supporting the terraces.Becau of the scale of this study and a lack of data about the prent state of the dry stone walls,only natural landslide-induced debris flow scarps were taken into account.
周厉王
Both the GeoIFFI and the DF2001inventories were randomly subdivided into two subts with the same size.As already mentioned,both inventories were also spatially subdivided into the three sub
ts.The training subts were ud for the construction of the model,and the validation subt was ud for independent validation of the predictive power of the resulting models.3.2.Factor maps
After evaluating the literature (Carrara et al.,1991;Soeters and van Westen,1996;Guzzetti et al.,1999;Castellanos Abella,2008),10causal factor maps were prepared (Fig.6).The maps can be divided in two groups:DEM derived factors and other geo-factors.
For the preparation of the maps,a DEM of the study area with a 10m resolution was ud.The DEM was provided by the Cartographical Of fice of the Mountain Consortium of Municipalities of Valtellina di Tirano,from contour lines with an interval of 1m in urbanized areas and 10m in the rest of the territory,and additional points with spot heights obtained by photogrammetry from the 2001air photos.The following factors were derived from the DEM using ArcGIS tools:altitude,internal relief,planar curvature,pro file curvature,slope,slope aspects,and flow accumulation.The values of each factor were classi fied into 10class using quantiles,except for the aspect map,which had nine class (eight for the main compass directions and one for flat areas).
Usage of quantile classi fication may cau important conquences if data distribution is extremely skewed.The usage may make it possible to better explore the behaviour of the factors with respect to the landslide occurrence,becau the rank-ordered variables are proportionally distributed.
For the preparation of the geo-factor maps veral sources were ud:
•A land u map,derived from the 1:10,000scale map of DUSAF Project (2003),made by the Lombardy Region using orthophotos taken in 2001.The map contains 23class of which the largest ones are coniferous forests and scarce
vegetation.
Fig.4.Flowchart of the applied methodology.抽筋是什么原因引起的
解酒方40J.Blahut et al./Geomorphology 119(2010)36–51
