Mapping social vulnerability to natural hazards in Italy: A suitable tool for risk mitigation strategies
Ivan Frigerio*, Mattia De Amicis
EarthandEnvironmentalScienceDepartment,UniversityofMilano-Bicocca,PiazzadellaScienza1,20126Milano,Italy
ARTICLE INFO
Articlehistory:
Received17February2016
Receivedinrevisedform17May2016 Accepted3June2016
Availableonlinexxx
Keywords:
Socialvulnerability Indicators Naturalhazards Italy
Mappingvulnerability Riskmitigation
ABSTRACT
ItalyisoneoftheEuropeancountriesthataremostheavilyexposedtoawiderangeofnaturalhazards, whichmightcauselargeeconomiclosses.Inthiscontext,theassessmentofsocialvulnerabilityhasan importantroleforevaluatingthecapacityofacommunitytopreparefor,respondtoandrecoverfrom disasters.Howevertherearecurrentlynopublishedstudiesanalysingsocialvulnerabilityanditsspatial distributioninItaly.Withinthisframework,thispaperaimstoapplyaprovenmethodforassessingsocial vulnerability at the national scale, while considering the contribution of the socioeconomic and demographicfactorsthataffecttheItalianpopulation.Theproposed methodologyis basedonthe Hazard-of-Place Model approach, and uses free and open source software applications (FOSS).
Specifically,weselectedsignificantcomponentsthroughPrincipalComponentAnalysisandderivedtheir spatial distribution.Usingcomponentscores,wederivedasocial vulnerabilityindex,evaluatedits geographicdistribution,andperformedaclusteranalysisonitsspatialvariation.Theanalysisidentified differentspatialpatternsacrossItaly,providingusefulinformationforidentifyingthecommunitiesmost likelytoexperience negativenaturaldisasterimpactsduetotheirsocioeconomicanddemographic characteristics.Thisresearchrepresentsanimportantcontributiontoimprovethepotentialityofrisk mitigationstrategiesandindesigningriskcustom-madepolicies.
ã2016ElsevierLtd.Allrightsreserved.
1.Introduction
The concept of vulnerability in the framework of classical natural hazard, multi-hazard and risk assessment traditionally referstothefractionofthetotalvalueatriskofbeinglostaftera specificadverseevent(Marzocchietal.,2009).However,overthe lastfewyears,theterm‘vulnerability’hasbeenfrequentlycitedin thescientificliteratureinrelationtothesocio-economicaspects thatinfluencesocietalconditions,includingexposure,sensitivity, coping, adaptive capacity and social capital. Therefore, the definition ofvulnerabilitydependsontheresearchperspectives employed.Someperspectivesfocusonasinglefactor,suchasthe physical side of vulnerability (biophysical vulnerability), while othersconsidersocialaspects,suchasgender,employment,age,or acombinationthereof(Cutteretal.,2003;Wisneratal.,2004).In thecontext of climate change,the IPCC (Pachauri et al., 2014) definesvulnerabilityasthepropensityof humanandecological systemstosufferharm,and theirability torespondtostresses imposed as a result of climate change effects. Many other
definitionscanalsobefoundintheliterature.Forexample,Adger (2006) defines vulnerability as the exposure of a group or an individualtostressduetosocialaswellasenvironmentalchanges thatdisruptslivelihoods.Todate,asharedtheoreticalconstructof vulnerability does not exist (Cutter, 1996; Fekete et al., 2014).
Within this backdrop, vulnerability assessment is frequently appliedindifferentstudies,suchasthosethatmeasuretheeffect ofclimatechangeonsocieties(e.g.,drought,heatwavesand/or intensehurricanes),studylivelihoodandpoverty(Prowse,2003), andconceptualizedisasterriskreduction(Cardona,2005;Cutter, 2008;Feketeetal.,2014).
In thecontextofdisaster riskreduction, threeelementscan revealthevulnerabilityofasocietyandplace:risk(R),hazard(H) andvulnerability(V).Theseelementsarerelatedtoeachotherand can be schematize using the following pseudo-equation (1) ( Wisneretal.,2004):
R¼HV ð1Þ
Considering this equation, risk is the result of interaction betweenhazard,adangerouseventthatmayaffectdifferentplaces at different times (with a certain intensity and frequency of occurrence) and vulnerability, the characteristics and
* Correspondingauthor.
E-mailaddress:[email protected](I.Frigerio).
http://dx.doi.org/10.1016/j.envsci.2016.06.001 1462-9011/ã2016ElsevierLtd.Allrightsreserved.
ContentslistsavailableatScienceDirect
Environmental Science & Policy
j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a te / en v s c i
circumstances of a community, system or asset that make it susceptible to the damaging effects of a hazard. According to Wisner(2004)«theriskofdisasterisacompoundfunctionofthe naturalhazardandthenumberofpeople,characterizedbytheir varying degrees of vulnerability to that specific hazard, who occupy the space and time of exposure to the hazard event».
Indeed,thedisasterentityvariesdrasticallydependingonthelocal context.
There is an important relationship between the natural elementsand thecultural, social andeconomic organization of theaffected society.As mentioned by Cutter et al. (2003),the probabilitythatanaturaldisasterhasmoredevastatingeffectson oneplacethananotherdependsonthelocalvulnerabilityofeach place.Forthisreason,thelackofsocialstudiesonhazardandrisk assessmenthighlightstheneedtobetterintegrateinformationon socialvulnerability into emergencyand risk managementdeci- sion-making.
Over the last years, several approaches to measure social vulnerabilityatdifferentscaleswereproposed.Thesestudieswere developed for different scopes as evaluating climate-change impacts, defining what are the societal responses to natural hazardsormeasuringprogress towarddisasterrisk reduction ( DolanandMessen,2012;Tateetal.,2010).Somecountries,likethe UnitedStatesof America (Wu et al., 2002;Cutter et al., 2003;
Chakrabortyetal.,2005),Australia(Dwyeretal.,2004),theUnited Kingdom(Tapselletal.,2002;Kuhlickeetal.,2011),12countriesin LatinAmericaandtheCaribbean(Cardona,2005),Spain(Weich- selgartner,2002),Israel(FelsensteinandLichter,2014),Germany (Fekete,2009)andIndonesia(Siagianetal.,2014)developedsocial vulnerabilityindicesatthenationalorsubnationalscales.
DespiteItalybeingoneofthefiveEuropeancountrieswiththe highestprobabilityofadisasterandrelatedeconomiclosses(Beck etal.,2012),therearenostudiesassessingitssocialvulnerability.
Understanding the differential impact of hazardous events is criticaltoreducingthenegativeimpactofnaturaldisasters.The overallaimofthispaperistodefineasocialvulnerabilityindex (SVI)forItalybyapplyinganinductivemethod,wheremeasure- mentsofsocialvulnerabilityarebasedontheunderlyingfactors thatinfluenceacommunity’sabilitytopreparefor,dealwithand recoverfromanimpact(Cardona,2005;Tate,2012).ThisSVIaims toincreasetheunderstandingofthedynamicsandsocio-economic characteristicsoftheItalianpopulation,aswellastoidentifyareas withdifferentabilitytoreacttocatastrophicnaturalevents.
Specifically, based on the Hazards-of-Place model approach (Cutter,1996), this paper first outlines the main variables that indicateaspectsofvulnerabilityofItalytonaturalhazards.Itthen appliesa PrincipalComponentAnalysis (PCA)toidentifyingthe underlyingcomponentsthatmakeaterritorysociallyvulnerableto naturalhazards.Finally,thearticleexploresthespatialpatternsof SVIacrossItaly,usingclusteranalysis.
This revealed to be a useful tool to identify potentially vulnerable areas, proposing finally the possibility to integrate thesocialvulnerabilityindexintoemergencymanagementplans anddisasterriskreductionstrategies.
2.Materialsandmethods 2.1.Toolsanddata
TheproposedmethodologyisbasedonFOSS(FreeandOpen SourceSoftware)applications.Bystartingfromtheimplementa- tionofhazardandvulnerabilitydatabase,itisabletostatistically analyse different socioeconomic variables in order to produce mapsofSVI.FOSSapplicationswereselectedbecausetheypermit themethodologytobeeasilyreplicablewithoutanyroyaltycost.In particular, the following tools were used: DBMS-system
PostgreSQL/PostGIS(http://www.postgresql.org)fordatastorage, R software (http://www.r-project.org) for statistical analysis, GeoDasoftware(http://geodacenter.asu.edu)forspatialstatistics analysisandQGISsoftware(http://www.qgis.org)tovisualizeand producemaps.
Secondarydataaboutdemographyinthe2011Italianpopula- tionandhousingcensusweretakenfromtheofficialrepositoryof theItalianNationalInstituteofStatistics(ISTAT)(ISTAT,2015).The information used included population, social variables and geographicallyreferencedinformationatdifferentadministrative units(censustract,municipalityandregional).Censussocioeco- nomicvariablesweredownloadedascsvfileforall8092Italian municipalities and, in order to improve the management and processingdata,thevariables werestoredasa singletableinto PostgreSQL/PostGISdatabase,usingISTATmunicipalitycodeasthe primarykey.Thisprocedurehasallowedtoobtainanaggregate informationofsocioeconomicdataatmunicipalitylevel.Therange of residents population counts for the Italian municipalities is betweenmorethan2.6millionpeopleforRomemunicipalityand 30 people for Pedesina municipality, a small town in Sondrio province(northItaly).Then,ametadatatablewithdescriptionof thevariableswasprepared,storedandlinkedtothesocioeconomic censusdata.Additionally,ageographiclayerofItalianmunicipali- tiesborderswasaddedand joinedtothecensussocioeconomic variablesfor thevisualizationoftheirspatialdistributionalong Italianterritory.
2.2.Constructionofsocialvulnerabilityindicators
Social vulnerability is a theoretical concept that is hard to quantify.Amethodmustthenbedevelopedtomakethisconcept countable(Hinkel,2011).Throughtheuseofdifferentindicatorsit ispossibletoassesssocialvulnerability;indeed,anindicatorisa quantitativeorqualitativevariablethatisderivedfromaseriesof measurements,hasdifferentmeaningsandcanbeobservedacross timeandspace.Forexample,indicatorsofsocialvulnerability,like ageorsocioeconomicstatus,canbeassessedfortheirchangeover timeorforhowtheycompareacrossgeographicorsocialentities (Hinkel,2011).
Withinthisframework,wefirstselectedsuitableindicatorsin ordertocreatetheirproxyvariables.Weexaminedcensusdataand thencreated18variablesrepresentingtheItaliansocio-economic conditions,usingappropriateSQLqueries(Table1).
Abriefdescriptionofeachindicatoralongwithitsinfluenceon socialvulnerabilityfollows.
1.Familystructure:largefamiliesoftenhavelimitedfinancesto outsource carefor dependents, and thus must balancework responsibilitiesandcareforfamilymembers.Largefamilysize may also reduceevacuation ability and resilience tonatural hazards(Cutteretal.,2003).
2.Education: it reveals the ability to understand information aboutemergency plansorwarninginformation and toavoid dangeroussituations(Elstad,1996;Morrow,1999;Cutteretal., 2003).
3.Socioeconomicstatus:thisindicatorrepresentsameasureofthe ability to absorblosses and enhance resilience tohazards ( CollinsandBolin,2009;Adgeretal.,2009;Lee,2014).Among variables that compose this indicator we identified (i) a containment index,a measure of the number of peoplewho workwithintheirresidencemunicipality,and(ii)anattraction index,whichevaluatesthecapabilityofamunicipalitytoappeal commuters.
4.Employment: it is often related to the potential loss of job activities after a hazardous event, increasing therefore the
numberofunemployedworkersinacommunity(Cutteretal., 2003).
5.Age:itisarelevantdimensiontoassesssocialvulnerabilityand it includes variables likethe percentageof children under 5 yearsandofelderlypeopleover65,themostvulnerablegroups in a society (Bolinand Stanford,1991;Morrow,1999; Cutter etal.,2006).
6.Raceandethnicity:thisindicatorisusedforthesocietiesthat includeseveralethnicgroupswithdifferentlanguages,cultures andeducationallevelsthatcoulddetermineculturalbarriersin acommunity(BolinandStanford,1991;Cutteretal.,2003).
7.Populationgrowth:it indicatesthedegreeofurbanization; a rapidpopulationgrowthisoftenunlikelytobeabsorbedbythe countrybymakinginefficientservicestothepopulation(Cutter etal.,2003).
2.3.Multivariatestatisticalanalysis
TheinductiveapproachtoconstructingaSVIrequiresalargeset of variables in order to identify the meaningful components.
Multivariate statistical analysis is the most commonly used statistical methods for reducing the number of variables in a dataset.Principalcomponentsanalysis(PCA)isespeciallyuseful whenthevariableswithinthedatasetarehighlycorrelatedwith oneanother,andwhenthereisahighratioofexplanatoryvariables tothenumberofobservations(EverittandHothorn,2011).
For this reason, R data analysisprogramming languagewas usedtoperformPCAinordertoreduceacomplexmultivariatedata settoalowerdimension.
Extendeddatasetliketheonesusedhere(e.g.ISTATcensus data,municipalitiesandcensusblockgeoreferencedpolygons)can leadtoanalyticalissues whenthestaticmemorymodelofRis used.To overcomethis, weusedthe“RPostgreSQL” and “sqldf” packages,asiscommonpracticeinthefieldofgeospatialanalysis( Bivand and Neteler, 2000). This way, R was linked directly to PostgreSQL, thus querying the database using standard SQL commands.Also,inordertoeasestatisticalanalysis,adataframe inRwascreatedsothatuserscouldreadandwriteittoandfrom PostgreSQLdatabase,speedingupstatisticalanalysis.
2.3.1.Exploratorydataanalysisandprincipalcomponentanalysis ProcessingdataandPCAwerecomputedusing“Vegan”,aCRAN packagedeveloped for theanalysisof ecologicalcommunities ( Borcard et al., 2011). This package has several useful tools for multivariatestatisticalanalysis(e.g.NMDS,pCCA,pRDAandPCA).
Furthermore others R libraries such as “outliers”, “gclus” and
“fbasics”wereused.
Before computing PCA, we carried out the following data processingsteps:
A boxplotwas generated for each proxy variablein order to visualize its statistical distributionand toindicate variability outsidetheupperandlowerquartiles.Thisisusefultodetect possibleoutliers;
TheoutliersconfirmedbyDixsonstatisticaltest(Dixon,1950;
Kimetal.,2012)werereplacedbythemeanofthedatathrough
“Outliers”package(Nardoetal.,2005;Krishnanetal.,2010);
Z-scorestatisticalmethodwasusedtoconvertall17variablesto acommonscalewithmeanof0andstandarddeviationof1by thefollowingequation(2):
Zij¼x
m
js
j ð2ÞWhere
m
jis themeanands
jis thestandarddeviation ofthepopulation. The absolute value of Z represents the distance betweentherawscoreandthepopulationmeaninunitsofthe standard deviation.Standardization allows comparing variables thatareexpressedindifferentmeasurementunits.
VIF value (variance inflation factor) was calculated to test multicollinearity of the variables. Values greater than 10 may indicatemulticollinearityamongvariables(Gaitheretal.,2011).
At the end of this step, the variables family structure, containment and attraction indices were excluded from the analysisandaPCAwasperformed.
2.3.2.Extractingandinterpretingthecomponents
The number of components to be considered after a PCA dependsonthechosenmethod.Differentselectionsatthisstage, i.e.theselectionoftoofewortoomanycomponents,mayleadto differentconclusionsintheanalysis(LedesmaandValero-Mora, 2007).Forthisreason,threeselectionmethodsweretested:
1. Kaiser–Guttman criterion: this method is the most used in social vulnerability literature and consistsin computing the meanofalleigenvaluesandininterpretingonlytheaxeswhose eigenvaluesarelargerthanthatmean(Kaiser,1960);
2. Brokenstickmodel:thismethodrandomlydividesastickofunit lengthintothesamenumberofpiecesastherearePCAaxes.It repeatsthisproceduremanytimesandaveragestheresultsof allpermutations.Thepiecesarethenputinorderofdecreasing Table1
Indicators,proxyvariablesandtheirimpactonsocialvulnerability.
Variables Mean Minimum Maximum Std.Dev. Indicators Impactonsocialvulnerability
Familywith1component 0.33 0.11 0.85 0.08 Familystructure Increase
Familywithmorethan6components 0.01 0 0.11 0.009
Higheducationindex 0.35 0 0.63 0.06 Education Decrease
Loweducationindex 0.59 0 0.89 0.15 Increase
Containmentindex 0.49 0 1 0.16 Socioeconomicstatus Decrease
Attractionindex 0.15 1 1 0.34
Commutingrate 0.27 0 0.69 0.10 Increase
Femalelabourforceemployed 0.39 0.17 0.57 0.04 Employment Decrease
Labourforceemployed 0.89 0.57 1 0.06
Unemploymentrate 0.10 0 0.42 0.06 Increase
Rateofchildren<14years 0.13 0 0.23 0.02 Age Increase
Rateofold>65years 0.22 0.05 0.61 0.05
Agingindex 1.95 0 28.5 1.41
Dependencyratio 0.56 0.30 1.87 0.10
Populationdensity 296.94 1 12224.0 631.9 Populationgrowth Increase
Urbanizedindexforresidentialuse 0.11 0 3.90 0.22
Crowdingindex 2.41 1.20 3.43 0.26
Foreignresidents 0.58 0 3.67 0.04 Race/Ethnicity Increase
lengthandcomparedtotheeigenvalues.Oneinterpretsonlythe axes whose eigenvalues are larger than the length of the correspondingpieceofthestick(Borcardetal.,2011);
3.Parallelanalysis:thismethod(Horn,1965)consistsincompar- ingeigenvaluescalculatedonobserveddatawiththeeigenval- ues calculated ona random matrice through a Monte Carlo simulation.Acomponentisconsideredsignificantifithasan eigenvaluegreaterthan1andatthesametimegreaterthanthe correspondingeigenvalueproducedfromrandomdata.Itisthus basicallythecomponentmust provetobestrongerthanthe randomone.
Theresultsobtainedfromtheapplicationofthethreemethods arepresentedinFig.1.BothKaiser–GuttmanandParallelanalysis suggesttheextractionofthefirstfourcomponents,explaining76%
ofthetotalvariance.Thebrokenstickmodelsuggeststheselection of three components explaining 67% of the total variance.
Consideringtheconcordancein theresultsof thetwomethods andthehighervarianceexplained,wechosetoselectthefirstfour components. Varimax rotation was subsequently applied to simplify the results of PCA and to better understand the significanceofthecomponents.
Thecomponents,withtheirpercentageofvarianceexplained, wereinterpretedasthefollow:age(30.6%),employment(23.4%), populationgrowth(13.4%)andeducation(8.4%).Thedirectionofthe components was determined with respect to their known influences on social vulnerability, identified from the existing literature.Weassignedpositive (+)directionalitytofactorsthat increasevulnerability(age,population growth),and negative() directionalitytofactorsthatdecreaseit(employment,education)( Cutteretal.,2003).
2.4.Constructingandmappingsocialvulnerabilityindex
Acompositeindicatorisformedwhentwoormoreindicators are aggregated into a single index that should theoretically measuremultidimensional concepts,which cannotbecaptured bya singleindicator(Nardoet al.,2005).OurSVIcombinedall derivedcomponentscoresintoasinglemeasure.Theaggregation methodtogeneratecompositeindicesisthemainmethodusedby
most researchers interested in social vulnerability studies (Ge etal.,2013).However,giventhattheindividualcomponentsdon't all havethe same influenceon social vulnerability,a weighted index is usually preferred (Nardo et al., 2005). Therefore, we calculatedSVIscoresbyweightingeachcomponentscorebyits percentage of variance, and dividing by the total variance explainedbyallselectedcomponents.Inthis way,acomponent withhighervariancecontributesmoretothefinalSVIvalues.After that,theSVIwas generatedsummingeach componentscorein accordingtotheirinfluenceonsocialvulnerability;thistechnique represents the most widelyapplied aggregation method (Tate, 2013).Finally,theSVIwasthenclassifiedinfivecategories:very high social vulnerability (>1.5 standard deviation); high social vulnerability(0.5–1.5standarddeviation);mediumsocialvulner- ability (-0.5–0.5 standard deviation); low social vulnerability (from 1.5 to 0.5 standard deviation) and very low social vulnerability(<1.5standarddeviation).
Maps were then produced toshow the geographic areasof higher potential vulnerability of a community in response to natural hazards.SVIwas mappedusing QGIS desktopsoftware (Fig.3).Thedescriptionof localcontexts withinavulnerability assessment is possible through a spatial representation of geographicallyheterogeneousdeterminants ofvulnerabilityand their interactions (Preston et al., 2011). Vulnerability, in its different fields of application, is complex to explain,especially tothe publicand decision makers. Maps allow non-experts to easilyunderstandthepotentialextentandmagnitudeofaspecific dangerouseventanditsinteractionwithhumansettlements,thus supportingspatial planning(Clark etal.,1998).Additionally, as several authorsindicated(Kellyand Adger2000;Prestonet al., 2011), communicatinghazardsand theirinteractionwithsocial vulnerability is a complex task for the social institutions that addressdisasterriskreduction.
2.5.Spatialclustering
Thisstepfocusesonthespatialaspectofthesocialvulnerability index. Following the approach developed by Anselin (1988), Moran’s I was applied to measure spatial autocorrelation, identifying spatial clusters along Italian country. The Local
Fig.1.Threemethodsusedtoevaluatethechoiceofthecomponents.OntheleftKaiser–GuttmancriterionandBrokenstickmodel(PC1-PC14arethenumberoftotal componentsderivedfromPCAandthefirstfour,relevantforthestudy,arerespectively:age,employment,populationgrowth,andeducation).Ontheright,theresultofParallel analysis(1–14arethesamenumberoftotalcomponentsderivedfromPCA).
IndicatorofSpatialAutocorrelation(LISA)wasusedtomeasurethe linearassociationbetweenSVIandthespatiallyweightedvalueof thesameindex.Theanalysisteststhestrengthofglobalspatial autocorrelation of social vulnerability index, and requires a weightedmatrix to define a local neighbourhood around each municipality. The SVIvalues related toeach municipality were
compared with the weighted average of the values of its neighbours.Tothisend,arookcontiguityweightsfile(W)from municipality boundaries was created using GeoDa software (Anselin, 2004). Rook contiguity was used because is a more stringentdefinitionofpolygoncontiguitythanqueenordistance bandstodefineneighbours;inthiscase,contiguousmunicipalities Fig.2.SpatialdistributionofthefourcomponentsacrosstheItaliancountry:a)age,b)employment,c)populationgrowthandd)education.
aretreatedasneighbourswhilenon-contiguousmunicipalitiesare not(i.e.twomunicipalitiesareneighboringiftheirdistrictsshare commonborders)(AnnoniandKozlovska,2010;Imranetal.,2014).
Theweightsfile(W)wasusedtogenerateMoran'sIscatterplot andtoperformLISAstatisticinordertoproducesignificanceand clustermaps(Fig.4).Thefollowingequation(3),basedonAnselin (1995),providesthecomputationofunivariateLISA:
Ii¼Zi
X
j
wijzj ð3Þ
whereZiandZjarestandardizedscoresofattributevaluesforuniti andj,andjisamongtheidentifiedneighboursofiaccordingtothe weights matrix Wij in standardized form. Moreover, sensitive analysiswas performed for significant local clustersin model, applyinga randomizationapproachof999permutationswitha thresholdvalue of p=0.05. (Anselin et al., 2006).This analysis allows to identifysignificant (p-value<0.05) concentrations of highvalues(high–high),concentrationsoflowvalues(low–low) andspatialoutliers(low–high,andhigh–low).
3.Resultsanddiscussion
The PCAextractedfour components,explaining 76%of total variance.Consideringtheextractedcomponents,theserevealedto besignificant, since theyrepresent clearlysocial and economic conditionsofItalianpopulation. Inordertovisualizethespatial distributionofage,employment,populationgrowthandeducation, wegeneratedfourmapsusing thescoresforthefourprincipal components(Fig.2).
Tocontextualizetheresultsofthisstudy,abriefdescriptionof Italian natural hazards spatial distribution also need to be considered.Fig.3bcouldhelpreadersthatarenotfamiliarwith the Italian geography. Because of its particular location, geo- dynamics context and high population density, Italy is highly affected by natural hazards: hydro-geological and earthquakes eventsarecertainlythemostrelevantnaturalphenomenafortheir highdiffusion,butalsovolcanichazardisfarfrombenegligible,
duetothepresenceofactivevolcanoesclosetodenselypopulated areas(e.g.VesuviusareainCampaniaregion,andSicily).
Consideringthehydrogeologicalhazard,overthelast50years (1964–2013)landslideandfloodeventsaffected2034municipali- ties(25%ofthetotal),causing2007dead,87missingandatleast 2.578injuredin(Guzzettietal.,2005;Guzzettietal.,2013;Salvati etal.,2014).AccordingtotheItalianLandslideInventoryIFFI,an important landslides database used as tool for Civil Protection Plans,528.903arethelandslidessurveyedinItaly(7.3%ofthetotal nationalarea)(Ispra,2014).Inparticular,theProvinceofTrento (locatedinTrentinoaltoAdigeregion),Valled'Aosta,Campania, Liguria and Tuscany are the areas more affected by landslides (Ispra,2014).Furthermore,someregionsarecharacterizedbyhigh values of flood hazard,as Emilia Romagna,Tuscany, Lombardy, PiedmontandVeneto.
Referringtoseismichazard,thestudycarriedoutbyNational Instituteof Geophysics and Volcanology (INGV) highlights that 37.6%ofItalianmunicipalitiesfallintothetwohigherclassesof earthquake hazard (Zones 1, the most dangerousareas, where major earthquakes may occur and Zone 2, areas that may be affectedbyratherstrongearthquakes).Themostseismicallyactive zones are concentrated along the central part of Apennine mountain range, in Calabria and Sicily regions and in some northern areas, such as Friuli Venezia Giulia, Veneto and the westernpartofLiguria.OnlySardiniaisnotparticularlyaffectedby seismicevents(INGV,2003).
The most significant aspect of the first component is representedbytheagingindex.Thisvariabledescribestheratio ofthepopulationaged65andovertothepopulationagedlessthan 15 years. Spatial distribution of the aging phenomenon is considerableand widespreadin Italy (Fig.2a).It isobservedin Friuli-VeneziaGiulia,inthewesternpartoftheAlps(Piedmontand LiguriaRegions)anddiffuselyalongtheApenninemountainrange, showingaquitestrongspatialrelationwiththetopographicand social characteristicsof Italy.In fact, overthe last decades,the structureanddynamicsoftheItalianpopulationhasundergone profoundchangesduetoa depopulationofmountainareasand Fig.3. a)Spatialdistributionofsocialvulnerabilityindexclassifiedinto5classesusingstandarddeviationmethod;b)StudyareawithregionslabelledwithDTM(Digital TerrainModel)superimposed;thegreenshadesrepresentsthemountainareas(AlpineandApenninemountainranges).(Forinterpretationofthereferencestocolourinthis figurelegend,thereaderisreferredtothewebversionofthisarticle.)
landabandonment,resultinginthemigrationoftheyounglabour forcetothevalleys,thankstothepresenceofindustrialandservice sectors.TheagingphenomenoninItalyisfurthershownbytwo other variables: the ratio of older people (65 and older) and childrenunder6yearsofage.Thesetwodemographicvariablesare importanttoperceivetheimbalancenotonlybetweenyoungand adults,butalsobetweenyoungandoldageclasses.TheItalianage structure evolution is a result of a trend, started in the late nineteenthcentury,towardtheincreaseinlifeexpectancyandthe decreaseoftheaveragebirthrate(Castagnaroetal.,2013).These trendsbroughtItalytobeoneoftheoldestcountriesintheworld (Italia,2012).
The second component represents employment. From a geographicalpointofview,Fig.2bclearlyillustratesthatItalyis divided in two macro-regions based on economic status: a northernpartwithhighereconomicdevelopmentandbetterjob opportunities thanks to the high concentration of industrial activitiesand services,anda southernpartwithsmallbusiness companyandareascharacterizedbyanestablishedconditionof disadvantagerelatedto:health,lackofservicesandthesignificant socialinequalities(Italia,2012).
The geographic distribution of population growth(map 2c) draws attention to the Italian major metropolitan aggregation areas.Thiscomponent isan indicatorof communityinstability, which often contributes to an increase in social vulnerability (Myers, 2007). Regions experiencing rapid population growth, mostlyinurbanareas,maynothavehadenoughtimetoadjustto
therecentpopulationincrease(Cutteretal.,2003).Thevariables
‘populationdensity’and‘urbanizedindexforresidentialuse’are theonesmosthighlycorrelated withthis component.Themap revealsanexpectedincreaseinsocialvulnerabilityinthemajor Italian metropolitan areas, particularly those of Milan, Rome, Naples,andPalermo.Thiscomponentisalsoausefulindicatorfor evaluatingtheimpactofhumanpressureontheenvironment( Bradshawetal.,2010).
The last component, education, not only influences risk perception, skills and knowledge but also reduces poverty, improveshealth,jobopportunitiesandincome.Educatedindivid- uals,householdsandsocietiesareassumedtobetterrespondto, preparefor,andrecoveryfromdisasters(Elstad,1996;Morrow, 1999;Cutteretal.,2003).Map2dshowsthathighereducationare found in northern and central Italy and in major metropolitan areas.Thisisrelatedtohigheremploymentandbetteraccessto education in the northern part of the country as well as in populatedcities.
Fig.3representsthespatialdistributionofsocialvulnerability index: scores greater than 1.5 SDs indicate very high social vulnerability.TheSVImapallowsvisualizingthespatialdistribu- tion ofsocial vulnerabilityatthenational scale,thus supplying decisionmakerswithausefultoolforemergencymanagementand territorialplanning.
SVI values greatly variedacross Italy and depended on the interplayofthefourcomponentsthataffectsocial vulnerability Fig.4.Localindicatorsofspatialautocorrelationclustermap(4a)andsignificancemap(4b).Thescatterplot(4c)showsthespatiallagofthevariableontheverticalaxisand theoriginalvariableonthehorizontalaxis.Bothvariablesarestandardizedandthegraphisdividedintofourquadrants:high–high(upperright)andlow–low(lowerleft) indicatingpositivespatialautocorrelation;andhigh-low(lowerright)andlow-high(upperleft)indicatingnegativespatialautocorrelation.Theslopeoftheregressionlineis Moran’sI.
(Table 2). In fact, this is the result of the combination of the vulnerabilityindicatorsmentionedbefore.
Cluster analysis,performedbythecomputationofunivariate LISA,confirmsthemeaningfulnessofthespatialdistributionofthe socialvulnerabilityindex.ImportanttohighlightisthattheLISA clustermapcanidentifytheoverallspatialpatterns,butdoesnot explainwhythesepatternsoccur.Theclustermap(Fig.4a)shows, inredcolour,apositivespatialauto-correlationforthe18.6%of municipalities that reveals high values of social vulnerability.
Differently, the blue clusters (23.4%) denote a low social vulnerability.Incontrast,theroseandlightbluecoloursindicate spatialoutliers,respectivelyhighsurroundedbylow(0.56%),and lowsurroundedbyhigh(0.86%).Thismaprepresentsthelocations withsignificantLocalMoranstatisticsandclassifythoselocations bytypeofassociation.Morebroadly,theoverallpatterndepicts clustersofhighsocialvulnerabilityonthenorth-westernpartof Italy,alongthesouthernpartofApenninemountainrangeandfor thetwomajorislands(SicilyandSardinia).
The sensitive analysis (Fig. 4b), performed by applying a randomizationmethod,providedanadhocapproachtoassessthe sensitivityoftheresultsdue tomultiplecomparisons(i.e.,how stableistheindicationofclustersoroutlierswhenthesignificance barrierislowered).Clusterswithhighandlowresidualvaluesat themunicipalitieslevel mayindicatespatialvariationin model performance.
AnalysingmapsinFig.2andcomparingthemwithSVIindex and clustermaps (Fig.3 and 4),it is possibletounderline the contributeof eachcomponentin thedefinitionof overallsocial vulnerability.Forexample,inthenorthernpartofItaly,thespatial distributionofsocial vulnerabilityisprimarilylinkedtotheage component, while in the southern the employment condition characterizetheHigh–Highcluster.Thisspatialpatternhighlights thestrongterritorialdifferenceinwhichtheemploymentrateof northernregionsishigherthanthesouthern,withanegativetrend that persists from 2002 (ISTAT, 2013). This difference is also reinforcedbytheeducationcomponent. Indeed, asobservedby Istatforthe2011year,thepercentageofadultpopulationwithno higher education is very high for the two big islands Sardinia (53.5%) and Sicily(53.2%)and for Campania region(52.9%). The componentpopulationgrowthislessclearinexplainingclusters;
probablybecauseitisabsorbedbythestrongpatternofthefirst twocomponents.Despitethis,itisinterestingtonotethatsome urbanareas(e.g.Milan,Bologna,Venezia)areclassifiedasHigh- Lowclusterswherepopulationgrowthandurbanizationarevery highthantheirsurroundings.
3.1.Riskmitigationmeasuresandpolicyimplications
The results of this study underline that the vulnerability characteristicsofpeoplehavearobustspatialvariation.Forthis reason,accordingtoKoksetal.(2015),theexclusionofthespatial variabilityofsocialvulnerabilityinthecontextofriskreduction strategiesmayleadtonon-efficiencydecisions.
In line with EU countries, Italian policy to reduce risks is gradually changed from a structural-security based approach towardanintegratedriskmanagementapproach.Thischangeis
particularlyevidentconsideringhydrogeologicalrisk.Forexample, Scolobigetal. (2014)haveanalyseda timelineoflandsliderisk policy in Italy evidencing how the legislation is changed from structural defense measures and building restrictions (Royal DecreeDR523/1904),to a newframeworkfor risk assessment and mapping (L. 225/1992, one of the main pillar of Italian legislative in matterof Civil Protection), finally reachinga risk governancewiththeimplementationofEuropeandirective(2007/
60/EU).Furthermore,theupdateofthe225/1992lawexplicitsthe conceptofriskmitigation,identifyingtheactivitiestominimize thepotentialdamagecausedbydangerousevents;theseactivities, called“non-structural”,are:earlywarning,emergencyplanning, trainingactivities,disseminationofCivilProtectionknowledgeand informationcampaigns(L.100/2012).
Theperspectivechangeaboutriskmitigationactivitiesisalso strengthened by the Decree of the Prime Minister issued on February 18th, 2008 creating the Italian National Platform for DisasterRiskReduction,coordinatedbytheItalianNationalCivil Protection Department (DPC) that aims at ensuring the full implementation of the Hyogo Declaration and of the Hyogo FrameworkforActioninItaly.Amongthetasksmentionedinthe Art.1,thePlatformshouldpromotetheemploymentofthemost suitableoperationaltoolstoimprovecommunityresilience.
To this regard, one of the current Italian natural hazard mitigation activities is the Italian application of the Floods EuropeanDirective2007/60/EU(lawD.Lgs. 2010n49).Thelaw defines how to prepare flood risk maps as a tool for risk management, combining flood hazards, physical vulnerability and populationexposure. Regarding social dimension, the only indicatorsuggestedbythelawisthenumberofresidentialpeople potentiallyaffectedbyfloodevents.Consequently,thevulnerabili- tybecomes homogeneousfor theentirepopulationandneither resiliencenorsocialvulnerabilityaredefinedorimplementedin thisdirectiveonfloodriskmanagement.
Consideringthedescribedbackdrop,theresultsofthis study emphasizehowsocioeconomicdifferencesinapopulationshould be included in risk mitigation strategies. In fact, the social vulnerability indexindicatesanhighspatial patternalong Italy withsignificantclustersthatevidencethedifferentvulnerability characteristicsofacommunity.Theconceptofsocialvulnerability is proactive as it can reduce the probability of loss before it becomes a real threat, and can minimize the magnitude of damages.Forthisreason,thepre-conditionoflocalcommunities toadaptandrespondtohazardouseventsisextremelyimportant for the evaluation of their impacts and for the successful implementationinpolicyofriskreductionmeasures.Thecapacity toadaptandrespondis largelyfunction ofalocalcommunity's socio-demographicstatus,thatisrelatedtotheirsocialvulnera- bility(Cutteretal.,2003;SmitandWandel,2006).Assuch,the socialcharacteristicsoflocalcommunitieslivinginareasexposed to natural hazards can be considered an important factor in determiningthefeasibilityofriskmitigationpolicies(Koksetal., 2014).
For policymakers,it iscrucial toknowwhich arethe more socially vulnerable zones against hazards in order to identify appropriatecost-effective riskreduction strategies tobeimple- mentedatnationalandatthelocallevel.Consideringthepotential implicationofthisstudyinriskmitigationmeasures,theSVImap canbeimplementedinTerritorialGovernmentPlansatregional scaleandataRiverBasinscale;atanationalscale,mapsallowto obtainawidespreadvisionofsocialvulnerabilityspatialvariation, identifying the most vulnerable areas that require a more deepened analysis at a local scale. This is feasible due to the multiscalepropertyofthesocialvulnerabilityindexasitispossible tobuilditfromcensustracts(inItalycensustractrepresentsthe lowest levels of aggregation). In that way, mapping the social Table2
PercentageandnumberofmunicipalitiesbylevelsofsocialvulnerabilityinItaly.
Vulnerabilityclasses Numberofmunicipalities Percentage(%)
Veryhigh(>1.5StdDev) 619 7.65
High(1.5<StdDev<0.5) 1688 20.8 Medium(0.5<StdDev<0.5) 3014 37.2
Low(0.5<StdDev<1.5) 2473 30.5
Verylow(>1.5StdDev) 298 3.6
vulnerabilityindexatadetailscalerevealstobeaveryefficienttool inthecontextofCivilProtectionPlansforemergencymanagement.
Forexample,asmentionedbyKoksetal.(2014)regardingfloods riskmitigation,«elderlyarelessabletorapidlyplacesandbagsin frontoftheirdoortoprotectthemfromthewater,andlow-income householdsmaynotbeabletoaffordflood-proofingtheirhouse»;
thus,theidentificationofareaswiththesevulnerabilitiesbecomes a priority, in order to develop place-specific risk mitigation strategies.Similarly,differentlanguages,cultureandeducational levelsinsocieties thatinclude several ethnicgroupscancreate barriersduringpreandpost-disasteractivitiessuchascommuni- cationofemergencyprocedure(Cutteretal.,2003;Morrow,1999;
Bolin,2006;Peguaro,2006).Inthis respect,knowledgeofareas withmanymigrantscanbeanimportanttooltoorganizespecific communication activities or to install panels that describe evacuation routes and recovery areas in different languages.
Therefore,usingaSVIandanalysingthecomponentscontributing tosocialvulnerability,riskinformationcampaignscanbetailored todifferentvulnerablegroups(Thiekenetal.,2007).Moreover,the combination of theSVI map withinstitutional hazard and risk mapscangiveasignificantcontributiontoidentifyhot-spotareas toaddressmitigationriskefforts.
InItaly,thelackofstudiesaboutsocialvulnerabilityhighlights animportantquestion:noattentionispaidtotheunderlyingsocial conditions of the communities living in hazard prone areas.
AccordingtoTapselletal.(2010)itisthereforepossibletodeduce that«thenotionofsocialvulnerabilityisnotamainstreamconcept indealingwithnaturalhazardsinItaly.Theterm‘vulnerability’on thewebsitesofthenationalandlocalunitsofCivilProtectionis defined by ‘identifying natural risks’, instead of being seen as somethingthatalreadyexistsinthesocialstructureandcouldbe mitigatedinordertoreducethemagnitudeoftheimpact».
Finally,anintegrationofsocialvulnerabilitycomponentinrisk assessmentstudiesisrecommended.Thisisacriticalelementasit fits in all phases of the risk management cycle (emergency preparedness,immediateresponse,mitigationplanningandlong- termrecovery).InreferencetotheItalianregulatoryframework, this work suggests a new methodology for the assessment of vulnerabilitythatcanbeintegratedintheproposalsforthefuture legislationinmatterofriskmitigationpolicies.
4.Conclusions
Inconclusion,thisstudyrepresentsthefirstapproachanalysing themajordimensionsandspatialvariationofsocialvulnerability tonaturalhazardsinItaly.Thisresearchrevealsthesignificanceof social vulnerability assessment, which could have important applications in forecasting and prevention measures for risk mitigation. In fact, the SVI map could be important tool for territorialplanningandemergencymanagementinprovidingan overall vision on the areas where the social structure is determinantinrespondingto,preparingfor,andrecoveringfrom disasters.
Thisstudyprovidesalsoanexampleofanalysisandinterpreta- tionof theindicators spatial distribution.In addition,a careful comparisonoftheobtainedmaps(indicatorsandSVI)canleadto identifyandunderstand,foragivenarea,thesocialdynamicsthat arerelevantinthedeterminationofhighsocialvulnerability,thus introducingappropriatemeasurestoreduce it.Thisis a crucial aspectbecausethemostvulnerableareascouldnotbepredictedby singleindicatorsmapbutwiththecombinationofdifferentsocio- economicaspectsthathaveanoverallpatternmoreextensively than on the single patterns for the individual indicators.
Furthermore,resultsshowaclearspatialclusteringbetweenhigh and low social vulnerable groups, indicating some degree of segregationinItaly.
Finally, mappingsocial vulnerabilityis recommendedin risk assessmentstudies,whichallowforacomprehensivestudyofthe feasibilityofriskreductionstrategies.Asmentionedbefore,this research can provide policymakers (especially Public Adminis- trationsandCivilProtection)newusefultoolstosocialvulnerabil- ityassessmentandtohighlightlocaldifferencesincapacitiesto implementstrategiesandneedsforriskmitigation.
Futuredevelopments,someofwhicharecurrentlyunderstudy, consist in assessing the spatial relationship between social vulnerabilityanddifferenthazards, includingearthquakes,land- slidesandwildfires.Bycombiningthemapsofsocialvulnerability indices with those of hazardous events through a GIS-based approachitwillbepossibletoidentifyareasthataremoreaffected byhazardous events,which wouldfurtherhighlighttheimpor- tanceofintegratingsocialanalysisincountriespronetonatural hazards.
Acknowledgments
WearegratefultoLuisaMaggioniandAndreaBonisoliAlquati fortheirsuggestions.Wearealsogratefultothetwoanonymous reviewers fortheircommentsthat greatlyimprovedthemanu- script.
References
Adger,W.N.,etal.,2009.Aretheresociallimitstoadaptationtoclimatechange.
Clim.Change93(3–4),335–354.
Adger,W.N.,2006.Vulnerability.GlobalEnviron.Change16(3),268–281.
Annoni,P.,Kozlovska,K.,2010.RCI2010:somein-depthanalysis.European CommissionJointResearchCentre.
Anselin,L.,Syabri,I.,Kho,Y.,2006.GeoDa:anintroductiontospatialdataanalysis.
Geographicalanalysis38(1),5–22.
Anselin,L.,1988.SpatialEconometrics:MethodsandModels,vol.4.Springer Science&BusinessMedia.
Anselin,L.,1995.Localindicatorsofspatialassociation-LISA.GeographicalAnalysis 27(2),93–115.
Anselin,L.,2004.ExploringspatialdatawithGeoDaTM:aworkbook.Urbana51, 61801.
Beck,M.,etal.,2012.Worldriskreport2012.AllianceDevelopmentWorksin CollaborationwithUNU/EHS,TheNatureConservancy.AllianceDevelopment Works,Bonn.
Bivand,R.,Neteler,M.,2000.Opensourcegeocomputation:usingtheRdataanalysis languageintegratedwithGRASSGISandPostgreSQLdatabasesystems.
Proceedingsofthe5thInternationalConferenceonGeoComputation. Bolin,R.,Stanford,L.,1991.Shelter,housingandrecovery:acomparisonofUS
disasters.Disasters15(1),24–34.
Borcard,D.,Gillet,F.,Legendre,P.,2011.NumericalEcologywithR.SpringerScience
&BusinessMedia.
Bradshaw,C.J.,Giam,X.,Sodhi,N.S.,2010.Evaluatingtherelativeenvironmental impactofcountries.PLoSOne5(5),e10440.
Cardona,O.D.,2005.IndicatorsofDisasterRiskandRiskManagement:Programfor LatinAmericaandtheCaribbean:SummaryReport.TechnicalReport.Inter- AmericanDevelopmentBank.
Castagnaro,C.,deAzevedo,R.C.,etal.,2013.AgeingandcounterageingIntheitalian censuses1RIEDS-Rivistaitalianadieconomia,demografiaestatistica-italian reviewofeconomics.DemographyandStatistics67(2),21–30.
Chakraborty,J.,Tobin,G.A.,Montz,B.E.,2005.Populationevacuation:assessing spatialvariabilityingeophysicalriskandsocialvulnerabilitytonaturalhazards.
Nat.Hazard.Rev..
Clark,G.E.,etal.,1998.Assessingthevulnerabilityofcoastalcommunitiesto extremestorms:thecaseofrevere,ma.,usa.Mitigationandadaptation strategiesforglobalchange.MitigationandAdaptationStrategiesforGlobal Change3(1),59–82.
Collins,T.,Bolin,B.,2009.Situatinghazardvulnerability:people’snegotiationswith wildfireenvironmentsintheU.S.Southwest.Environ.Manage.44(3),441–455.
Cutter,S.,Boruff,B.,Shirley,W.,2003.Socialvulnerabilitytoenvironmentalhazards.
SocialScienceQuarterly84(2),242–261.
Cutter,S.,Boruff,B.J.,Shirley,W.L.,2006.Socialvulnerabilitytoenvironmental hazards.Hazards,Vulnerability,andEnvironmentalJustice115–132.
Cutter,S.L.,1996.Societalresponsestoenvironmentalhazards.InternationalSocial ScienceJournal48(150),525–536.
Cutter,S.L.,2008.VulnerabilityAnalysisforEnvironmentalHazards.Encyclopedia ofQuantitativeRiskAnalysisandAssessment.
Dixon,W.J.,1950.Analysisofextremevalues.AnnalsofMathematicalStatistics 488–506.
Dolan,G.,Messen,D.,2012.Socialvulnerability:anemergencymanagersplanning tool.JournalofEmergencyManagement10(3),161.