a. panizza, & g. de natale

From: TROISE, C., DE NATALE, G. & KILBURN, C. R. J. (eds) 2006. Mechanisms of Activity and Unrest at LargeCalderas. Geological Society, London, Special Publications, 269, 159–171.0305-8719/06/$15.00 © The Geological Society of London.

Volcanic hazard assessment at the Campi Flegrei caldera

G. MASTROLORENZO, L. PAPPALARDO, C. TROISE, S. ROSSANO,A. PANIZZA, & G. DE NATALE

Istituto Nazionale di Geofisica e Vulcanologia, Sezione di Napoli Osservatorio Vesuviano,Via Diocleziano, 328, 80024 Naples, Italy (e-mail: [email protected])

Abstract: Previous and new results from probabilistic approaches based on availablevolcanological data from real eruptions of Campi Flegrei, are assembled in a comprehensiveassessment of volcanic hazards at the Campi Flegrei caldera, in order to compare the volcanichazards related to the different types of events. Hazard maps based on a very wide set ofnumerical simulations, produced using field and laboratory data as input parameters relativeto the whole range of fallout and pyroclastic-flow events and their relative occurrence,are presented. The results allow us to quantitatively evaluate and compare the hazardrelated to pyroclastic fallout and density currents (PDCs) in the Campi Flegrei area and itssurroundings, including the city of Naples.

Due to the dominant wind directions, the hazard from fallout mostly affects the area east ofthe caldera, and the caldera itself, with the level of probability and expected thickness decreas-ing with distance from the caldera and outside the eastern sectors. The hazard from PDCsdecrease roughly radially with distance from the caldera centre and is strongly controlled bythe topographic relief, which produces an effective barrier to propagation of PDCs to the eastand northeast, areas which include metropolitan Naples. The main result is that the metro-politan area of Naples would be directly exposed to both fallout and PDCs. Moreover, thelevel of probability for critical tephra accumulation by fallout is relatively high, even formoderate-scale events, while, due to the presence of topographic barriers, the hazard fromPDCs is only moderate and mostly associated with the largest events.

Hazard evaluation for explosive volcanoes indensely populated areas is a major goal ofresearch in volcanology. Tephra fallout andpyroclastic density currents coexist in mostexplosive eruptions, and their relevance and theway that they extend the area affected by volca-nic hazards will depend on the eruption style andmagnitude as well as environmental conditions(topography, wind pattern). Therefore, hazardevaluation at an active volcano requires theacquisition of as much data as possible relativeto past eruptions and their related effects. Never-theless, due to the high variability of the eruptivephenomena, the volcanological records are notexhaustive enough to be able to define the wholerange of possible events for probabilistic evalua-tion. A more accurate approach requires theexploration of a wider repertoire of possibleevents which can be reproduced by numericalsimulations, in order to produce probabilistichazard maps.

Campi Flegrei caldera, which has been activesince at least c. 39 ka ago, produced a total of c.60 highly explosive eruptions, with VEI rangingbetween 0 and 6. In most cases tephra falloutfrom sustained column and pyroclastic densitycurrents has been produced. The mechanism oftheir formation, their intensity and the extension

of the affected area, mainly depend on the erup-tion magnitude. However, volcanological studies(e.g. Mastrolorenzo 1994; Mastrolorenzo et al.2001) and modelling (Rossano et al. 1998) haveshown that tephra fallout distribution is stronglyinfluenced by seasonal wind changes, while thespreading of pyroclastic density currents ismostly controlled by the topography of thecaldera and its surroundings.

Due to the widespread urbanization of thearea around the caldera, and particularly of itssurroundings, a quantitative evaluation of thehazards related to such phenomena is necessaryto better define of the emergency planning.

In this paper, previous and new results ofintegrated probabilistic and volcanologicalapproaches to numerical simulations of falloutand PDCs events are presented in order tocompare the hazards for different types of events.

Explosive eruptions at Campi Flegri caldera

The Campi Flegrei (CF) is a volcanic fielddominated by a Quaternary 12-km-wide calderadepression formed during two high-magnitudeeruptions: the 39 ka BP Campanian Ignimbrite(De Vivo et al. 2001) and the 14 ka BP NeapolitanYellow Tuff (Deino et al. 2004) (Fig. 1). After

160 G. MASTROLORENZO ET AL.

these two main events, c. 60 eruptions occurred inthe last 14 000 years, from scattered vents locatedwithin the caldera.

Explosive eruptions in volcanic history ofCampi Flegrei have VEI values of 0 to 5, witherupted masses ranging between less than 108

and 1013 kg; column height from less than 1 to45 km, magma eruption rate from hundreds ofkg s−1 to 108 kg s−1; and wide variability in totalgrain-size distribution, fragmentation and dis-persion (Table 1). However, extremely largeevents are rare, since VEI is generally lower thanfour. A few very large-scale surge-and-flowdeposits are the prevalent eruptive mechanismsin large explosive eruptions, while the tens ofintermediate and small-scale deposits occurred assubordinate episodes of more complex eruptivesequences. Effusive activity, with associated lavaflows and domes, contributes only a few per centto the total volume of the volcanic products.

The post-NYT intracalderic activity,prevalently hydromagmatic, mainly formedpyroclastic cones, consisting of interbedded

pyroclastic surge and fallout beds. The puremagmatic strombolian, sub-plinian and plinianevents, only subordinate, consist of scoria andlapilli beds, with dispersion ranging betweena few km2 and some thousands of km2. Small-scale strombolian deposits are mostly dispersedwithin the caldera rim, while the larger depositsspread outside the caldera in a large area, mostlytowards the East.

Probabilistic approach

Probabilistic hazard maps have been developedfor Campi Flegrei for pyroclastic falls and cur-rents, using a generalization of the approachesproposed by Rossano et al. (2004) and De Nataleet al. (2006).

Pyroclastic fall

The transport of tephra in the atmosphere, dueto the processes of convection, diffusion and

Fig. 1. Digital topographic map of the Campi Flegrei Volcanic Field, including monogenetic volcanoes andvolcanic formations: AS, Astroni tuff ring; AV, Averno tuff ring; BA, Baia tuff ring; BM, Breccia Museopyroclastic-flow formation; FB, Fondi di Baia cinder cone; MI, Miseno tuff ring; MN, Monte Nuovo cindercone; MP, Montagna Spaccata strombolian formation; MS, Monte Spina small-scale pyroclastic flow formation;NI, Nisida tuff cone; SF, Solfatara tuff ring; SG, Senga spatter cone; MO, Minopoli violent strombolianformation; PP, Pomici Principali plinian formation; TGN, Neapolitan Yellow Tuff.

161VOLCANIC HAZARD ASSESSMENT

settling, may be described by the followingwell-known convection–diffusion–migrationequation (Friedlander 2000):

∂∂

∂∂

∂∂

∂∂

∂∂

∂∂

∂∂

∂∂

∂∂

χ χ χ χ

χ χ

tux

vy

wz

xK

x yK

y zx y

+ + +

= + + KKz

Vx

z

∂∂

∂∂

χ

χ- ts

(1)

where x is the concentration of tephra (kg m−3);vw(u,v,w) is the wind-velocity vector, Kx, Ky, Kz

are diffusion coefficients; and Vts is the terminalsettling velocity. In the hypothesis of low-volumeconcentration of tephra, v, Ki and Vts depend onlyon (x,y,z,t) but not on x, so the equation is linear.Given an expression for such coefficients andsuitable boundary and initial conditions, eq. (1)may be solved numerically to give x=x (x,y,z,t).The mass distribution of tephra after all particleshave settled may then be calculated by the inte-gration of vertical fluxes at the surface, at a timemuch greater than the settling time of the small-est particles in which we are interested. However,such an approach is computationally expensive;in order to calculate the 2D distribution of tephraon the ground for a time larger than the maxi-mum settling time (tF) tF > ts , we have to computea 3D field for all times t < tF. According to Suzuki(1983), we can neglect vertical diffusion and con-vection in the atmosphere with respect to hori-zontal diffusion and convection, because theformer has a much smaller effect than the latterabove the atmospheric boundary layer. We have:

∂∂

∂∂

∂∂

∂∂

∂∂

∂∂

∂∂

χ χ χ χ χt

ux

vy x

Kx y

Kyx y+ + = + (2)

x measures of mass of accumulated tephra perunit surface (kg m−2). The settling term is implic-itly taken into account by considering the solu-tion of eq. (2) at the time corresponding to the falltime of the particles.

Following Bonadonna et al. (2005), we usetwo functional forms for the diffusion coeffi-cient, depending on the fall times (and hence,indirectly, from the sizes) of the particlesinvolved in the diffusion process. The horizontaldiffusion of large particles, which is characterizedby short fall times, is better described by lineardiffusion, in which K=k is constant in time.Diffusion of small particles, on the contrary, ischaracterized by long fall times, and so is betterdescribed by power-law diffusion: K=Ct3/2, withC=0.04 m2 s−5/2. The parameter K is not a

turbulent diffusion coefficient but a partlyempirical parameter, which takes into accountall effects which tend to spread the plume, suchas eddy diffusion, gravity flow, and so on. Weassume the following initial conditions (IC) andboundary conditions (BC) for equation (2):

IC(3)

BC

x dx y Q x x y y

r t r x yr

, , ,

,lim

0

0

0 0

2 2

( ) ( )( )

→ ∞

= - -

= = +x

corresponding to the instantaneous release att=0 of a mass Q from a vent at coordinates(x0, y0), in a doubly infinite domain, with thephysical constraint that the concentration oftephra decreases at sufficiently large distancesfrom the source. Eq. (2) then yields the analyticalsolution:

xp

=

- - + - -

58

5

8

5 2

0

2

0

2

5 2

QCt

x ut x y vt y

Ct

/

/exp( ) ( ){ }

for small particles (4)

xp

=- - - -Q

Kt

x ut x y vt y

Kt4 4

0

2

0

2

exp( ) + ( ){ }

for large particles (linear diffusivity)

To obtain the distribution of tephra on theground, we set t equal to the fall time of the par-ticle. The variation of wind speed and directionwith altitude, z, can be accounted for by dividingthe height of the atmosphere into layers of heightDzi, each of which is characterized by a constantwind velocity and direction. Particles in the erup-tion column have different diameters, and arriveat different heights into the column. Particlesof grain size wj arriving at height zk in the column(zkfH, the total column height) fall throughlayers Dzk=zk-zk-1, . . . Dz1=z1-z0, (z0=0).Each layer is crossed in a time Dti=Dzi/Vts(zi,wj),where the terminal settling velocity Vts dependsboth on height and on grain size, but not ontime, because we can assume that atmosphericdensity and viscosity at heights of kilometresdo not vary with time over the time-scales oftephra deposition. During this time interval, thecentre of the Gaussian distribution is displacedby Dxi=u(zi)Dti and Dyi=v(zi)Dti. Hence, whenthe particles reach the ground, the distributionis:


x

p D

D

=

-

- -

=

5

8

5

1

5 2

01

2

Q

C t

x x x

ii

K

ii

K

∑

∑

=

/

exp

∑

∑=

8

5

1

5 2

01

2

C t

y y y

ii

K

ii

K

D

D

=

-

- -

/

∑81

5 2

C tii

K

D=

/

(5)x

p

D

D

=

-- -

=

=

=

Q

K t

x x x

K t

ii

K

ii

K

ii

K

4

4

1

01

2

1

∆∑

∑

∑

exp

∑

∑-

- -=

=

y y y

K t

ii

K

ii

K

01

2

1

4

D

D

for large particles

(linear diffusivity).

The tephra mass is distributed along theheight of the column, and has a continuousgrain-size distribution. Summing on all heightsfrom 0 to H (column height), and from minimumto maximum diameter, we get the completegrain-size distribution:

x w www

w

=

+

f z fQ

Cz

V zt

z

H

z,

,

ex

min/

max

( ) ( )

( )

∫∫∫

0

0

5 2

5

8πφ

d

tss

pp,

-

- -

=

5

8

00

2

1

x xu z z

V z

C t

z

ii

K

( )( )

∫

∑

d

ts φ

∆

( )( )

∫

5 2

00

2

5

8

/

,-

- -y yv z z

V z

C t

z d

ts w

∆ iii

Kz

=1

5 2

∑

/ d dw

for small particles, and (6)

x w w

pw

w

w

w

=

+

-

-

f z fQ

Kz

V zt

x

z

H

z,

,

exp

min

max

( ) ( )

( )

∫∫

∫0

0

4d

tss

xxu z z

V z

K t

y yv

z

ii

K

00

2

1

0

4

-

-

- -

=

( )( )

∫

∑

d

ts ,w

D

zz z

V z

K t

z

z

ii

K

( )( )

∫

∑

d

d dts ,w

D

w0

2

1

4=

for large particles (linear diffusivity) (where fz

and fw are the distribution functions with respectto height and to diameter, respectively). To thediffusion time, we added the value ts correspond-ing to the diffusion of particles inside the erup-tion column, due to the fact that the eruptioncolumn is not really a vertical line, but has anon-zero cross-section.

To calculate ts we use the semi-empiricalexpression of Suzuki (1983) for small particles:

tz

Cs=5

288

2 2 5

/

and, for large particles, the expression fromBonadonna et al. (2005):

tz

Ks=0 0032 2.

We considered release from an instantaneouspoint source at time t=0 and position (x0, y0),while in reality an eruption has a finite durationand spatial extent. To investigate the implicationof finite dimensions, we describe a continuouseruption as the integral with respect to time ofinstantaneous emissions of intensity I=dQ/dt.This is possible because of Duhamel’s principlefor parabolic linear PDEs (such as eq. (1) andeq. (2)). If fz and fw are independent of time, thenall quantities in eq. (6) will be independent oftime, with the only possible exception of I.Accordingly, we take all parameters except I outof the time integral, and, because the integral ofI from 0 to t is equal to Q, we have precisely thesame result as in the case of an instantaneouseruption. Physically, the reason is that we havealready assumed that transport phenomena in

for small particles


the atmosphere do not depend on initial and finaltime, but only on the difference between them,and they are also independent of particle concen-tration. As a result, identical particles released atdifferent times from the same height will remainin the atmosphere for the same time, and fall tothe same place.

In order to model a continuous eruption as aninstantaneous release, therefore we must assumethat fz and fw do not depend on time, i.e. that theeruption column is a quasi-steady state. Thisassumption is valid for plinian eruptions after theinitial explosion (Woods 1988).

In order to complete the mathematicaldescription of our model, we must specify:

(1) How to compute the terminal settlingvelocity;

(2) The functional form for grain-size distribu-tion and for vertical mass distribution in thecolumn;

(3) Vertical distribution of wind velocity anddirections, with their relative probabilities.

We assume that particles always fall at theirterminal settling velocity, i.e. thus neglecting thetransient phase of acceleration, which is reason-able, because of the thickness of the verticallayers into which we divide the atmosphere. Theexpression for the terminal settling velocity of aparticle is a critical point; when the velocity ofthe particle is equal to Vts, drag forces due to theair and to the weight of the particle must bebalanced:

m g AC Vp a D ts=12

2r (7)

where mp is particle mass, g is gravitational accel-eration, ra is air density, A is particle cross-section and CD is a non-dimensional parametercalled the drag coefficient, defined as the ratiobetween the aerodynamic drag force on theparticle and the product of dynamic pressure1/2raV for the cross-section A. CD depends onthe Reynolds number Re, defined as:

Re=raVd /m (8)

where V is the particle velocity relative to the air;d is the particle diameter (mean of the three prin-cipal axes for aspherical particles); and m is airviscosity. If Re<<1, we are in Stokes’ regime,where, for spherical particles, CD=24/Re, so thatwe get the following analytical expression for Vts:

Vd g

tsp=r

m

2

18(9)

Outside Stokes’ regime (Re<<1), no analyti-cal solution for Vts has been found up to now,even for simple shapes like spheres. This is due tothe non-linearity of the Navier–Stokes equa-tions: partial differential equations governing theviscous flow of Newtonian fluids like air. Experi-mental expressions are thus needed, but differentsets of experimental results for volcanic particlesare often in disagreement. Suzuki (1983) foundan expression which is a good fit to the variousexperimental data:

Vgd

F

gd

F gd F

da b

tsp

p

a p

=

++ -

=+ +

-

-

r

m

r

m rr

2

0 32

2

2 0 64 3

9

81 1 5 1 07

.

. . .

→

←

cc

Fb c

a

3

2=+

(10)

where rp is particle density; ra is air density; a, b,c are the three principal axes with a being thelongest one; m is air viscosity; F is called the shapefactor and is equal to 1 for spherical particles;and d is the mean diameter.

Such an expression works well for low tomoderate Reynolds numbers: its limit for Rem0and F=1 is eq. (9), correctly. However, it fails forhigher values, because it predicts a decreasing CD,while experimental evidence for volcanic par-ticles shows that the drag coefficient approachesthe value of 1 for high Re. We thus chose to com-pute the Reynolds number ReL which, accordingto Suzuki’s expression, corresponds to a dragcoefficient of 1, and use Suzuki’s expressionfor RefReL, while if Re>ReL, we set CD=1 andobtain Vts from equation (7). We write Vts fora spherical particle, in order to show that Vts

increases with the diameter and the density of theparticles:

Vdg

tsp

a3=

4r

r (11)

Air density and viscosity in the atmospheredepend on height; thus, particles will fall quicklythrough the higher atmospheric layers, but willslow down at lower heights. The phenomenonis less important for the smallest particleswhich move in the Stokes’ regime; their terminalsettling velocity does not depend on density.We will model the dependence of atmospheric


thermodynamic properties according to the USStandard Atmosphere 1976 (Anon. 1976).

We assume that the grain-size distributionis log-normal, e.g. the distribution in w units isnormal (w=-log2(d) where d is the particlediameter in millimetres):

fd

d

d

w wps

w m

s( ) = −

−( )

1

2 2

2

2exp (12)

where md and sd are, respectively, mean andstandard deviation.

The vertical mass distribution in the column istaken from Connor et al. (2001):

f z

WY z Y z

V H Y Yz ( )( ) ( ){ }

( ) ( ) ( ){ } ( )=

-

- + -

b

w

0

0 1 1 0 0

0

exp

, expts

( ) ( )( )

( )

0 H

otherwise

ts

f f

=

= -

z

Y zW z

V

W z WzH

b

w

l

0

10

,

,

where W(z) is gas velocity at height z and W0 isgas velocity at the vent. In this work it is alwaysassumed that l=1 (Connor et al. 2001), since,according to Carey & Sparks (1986), linear varia-tion of W with z is a satisfactory approximationfor the major part of the column, except nearthe top and the bottom. From eq. (13) it maybe easily proved that the integral of fz over [0,H]is 1, and that fzg0 for any z, so it is indeed adistribution.

H is a normalization factor which does notchange the shape of the function, so there areapparently three parameters b,W0,Vts(0, f)=V0,in eq. (13). However, fz depends only on the

productYW

VWV

Y00

0

0

00( ) ( )

= = =b b , so it is a one-

parameter distribution. In practice, W0 and V0

are constrained by the physics of the problem,so we study the dependence of fz on b only. Omit-ting the simple but tedious calculation, we saythat fz has a maximum in [0,H] if and only ifb3=V0/W0. The terminal settling velocity atsea-level of big particles is of the order oftens of metres per second, while the eruptionvelocity is of the order of hundreds of metresper second, so this condition is usually satisfiedfor all particles when b3=0.1. The maximum is

attained for z HVWM= -( )1 0

0band has the value

f zW V

HWV

WV

z M( )

=- + -

b

e b b

0 0

0

0

0

0

1

1 1( exp

.

So we note that as bm∞, zMmH, fz(zM)m∞:since the integral of fz must be 1, this means thatlarger values of b correspond to more mass beingaccumulated near the top of the column, and lessat lower heights.

In order to handle the wind-velocity term ineq. (1), we adopted the 1953–2000 database ofwind velocity and directions measured between0 and 34 km of elevation from the ground atBrindisi (southern Italy) meteorological station.Average values of wind velocities and frequencyof occurrence of wind direction have been com-puted, from the database, in sectors of 10° andfor 10 selected heights. Thus, probabilities can beassigned, at each elevation, to each direction insteps of 10°, and the corresponding wind velocitycan be assigned. As previously evidenced byCornell et al. (1983), the average wind directionsin the troposphere and stratosphere are almostindependent of the height, above 5 km with anearly westerly prevailing direction in all seasons;in summer, however, there is a higher variabilitywith, above 20 km, a nearly easterly prevailingdirection. Figure 2 shows histograms of falloutdeposits’ dispersal axes relative to several tensof past eruptions from Neapolitan volcanoes(Vesuvius, Campi Flegrei and Ischia). It is evi-dent that there is a good correlation of observeddispersal axes in past eruptions with the relative

(13)

Fig. 2. Comparison between the frequency of winddirections collected at Brindisi (Southern Italy)meteorological station in the period 1953–2000 (red)and the observed tephra dispersions of Campanianeruptions in the last 100 ka (white).


frequencies of wind directions as inferred frommeteorological data.

Pyroclastic density currents

Simulation of pyroclastic density currents(PDCs) was carried out by Rossano et al. (2004),by direct and inverse numerical simulation ofpyroclastic flows, based on a simple model of agravity-driven pyroclastic current which stopsfor en masse freezing (Rossano et al. 1996, 1998,2004). The flow is assumed to be incompressible,with a Newtonian or Bingham rheology. Emis-sion from the vent is considered as being steadystate from the initiation of eruption.

The physical model of PDCs is described bythe equations of uniform flow for a Binghamor Newtonian fluid (McEwen & Malin 1989;Rossano et al. 1996). The steady, uniform veloc-ity profile for a Bingham flow in infinitely widechannels is given by:

v zg D z

k D z( )sin

,=-

- -1

2

2 2

g

r h( ) ( )

(14)

where zgDc is the height (measured from thebottom of the channel), k is yield strength, r isthe density of flow material, g is accelerationdue to of gravity, h is the slope of ground, g is theviscosity, D the total flow depth, and Dc is theplug thickness,

DkD v kD v k D

kc =+( ) − +( ) −2 2 2 4

2

2 2 2g g. (15)

The acceleration of the plug in a wide channel is(McEwen & Malin 1989):

d

dp

c

p

c

v

ta

kD D

v

D D= -

+-

+2 2

2 2r

g

r( ) ( ) , (16)

where vp is plug velocity and |a| is the modulusof the gravity contribution to the flow motion.Resistance terms in the equation describing aBingham flow unit depend on several factors.For a Bingham flow, the transition from laminarto turbulent flow depends upon two dimension-less numbers, the Reynolds number Re=rv D/g,and the Bingham number Bi=kD/gv. Fromempirical relations, Middleton & Southard(1978) observed that, when Bi exceeds about 1.0,the onset of turbulence occurs for Re/Big1000.According to McEwen & Malin (1989), the fron-tal air drag is assumed to be negligible, so that thedeceleration due to air drag on the upper surface(Perla 1980) of the flow is:

dd a

avt

cvD

=-r

r

2

2(17)

where the constant ca ranges ranges between0.1 and 1 (Perla 1980). The matrix of the inputparameters for the generation of the PDCs modelset includes: vent coordinates; initial velocityand direction of the flow; flow density, thicknessand rheological parameters (viscosity and yieldstrength).

Flow trajectories were computed for a digitaltopographic model of Campi Flegrei, approxi-mating PDCs to idealized, one-dimensional,mass-independent flows. The kinematics of thematerial point are governed by the equationsgiven above, with the known values of the localcomponent of acceleration, and the assumedphysical properties of the moving flow (initialvelocity, density and rheology) and air friction.A large family of flow lines simulates the travelpaths of a linear front departing from the vent.

As discussed by Rossano et al. (1998), even ifwe neglect lateral expansion; interaction betweendistinctive streamlines; and the effects of entrain-ment and sedimentation, the high density of1D flow-lines approximates a 2D model. Thisis particularly true for the high ratio betweenaverage PDC thickness and small-scale averagelandforms of the area.

A large number of individual 2D pyroclasticdensity currents (PDCs) were simulated spanninga wide range of properties and vent conditions.Each 2D flow, characterized by specific physicalproperties, is defined by a family of 1D flow-linesthat move over a digital elevation model as agravity-driven current. The whole set of flowmodels is obtained by sampling a multidimen-sional matrix of vent position, initial velocityand rheological parameters which control thepyroclastic flow movement.

Equations (14) to (17) have been used to simu-late the flows. These show that the parameterswhich characterize individual pyroclastic cur-rents are: initial velocity, viscosity, density, thick-ness and yield strength.

Typical values for each parameter have beenestimated using the eruption catalogue and fielddata. The results are summarized in Table 2,which shows the probability of a particular valueoccurring. All the combinations of the givenparameter values are considered in the simula-tion, each one taken with the cumulative prob-ability given by the product of probabilities ofthe single parameters. With these choices, hazardmaps are computed by extracting, from thewhole set of 2800 simulated eruptions, the prob-abilities of the flow crossing each given area of1 km2 on the topographic map. The crossing


probabilities are computed as the sum of prob-abilities from each given eruption for which atleast one flow crosses the area.

Hazard maps

Hazard map computation was achieved usinga probabilistic approach to simulate a large,mostly complete set of eruptions, whose relativedistribution in energy classes (VEI) matches theone observed in the intrinsically incomplete setof past eruptions. The incompleteness of the his-torical record derives from two main sources:the first is the scarce sampling of the eruptionspectrum, which, however, does give us the gen-eral statistical distribution of eruptions in eachclass of VEI; the second is the scarce spatialsampling of eruptive vents, from which we canstill infer some information about its statisticaldistribution.

Both fallout and PDCs have been simulatedby discrete sampling of a multidimensionalmatrix of input parameters (Tables 1 & 2), basedon field evidence and inferences as well as datareported in the literature and corresponding tothe whole range of volcanic explosivity index(VEI) that occurred in the history of CampiFlegrei.

For fallout hazard assessment, three differentscenarios have been considered, correspondingto distinctive sampling of the whole matrix of

input parameters: (1) an upper-limit scenariorelative to the worst-case event (VEI 5); thisis useful to determine the upper-limit value oftephra-fall accumulation; (2) an eruption rangescenario (VEI 1–4) which excludes the upper-limit high-risk but very-rare events, as well asthe lowest VEI events that are frequent but havevery low risk; this represents the most probablescenario expected for medium–short term; (3) awhole-range eruption scenario (VEI 0–5), whichincludes all the possible VEI events (long-lastingactivity scenario). For the three scenarios, hazardmaps relative to the critical load for the roofcollapse of 200 kg m−2 have been drawn (Fig. 3).

In addition, the absolute probability (con-ditional probability) that the critical load willbe exceeded in the case of eruption, has beenalso computed for the whole eruption scenario(Fig. 4).

For PDCs hazard map, due to the mass-independent nature of the flow model, a singlewhole-range scenario was considered byRossano et al. (2004), (Fig. 5). It includes allpossible combinations of flow parameters which

Table 1. Matrix of input parameters and their relative probability for fallout simulations

VEI 0 VEI 1 VEI 2 VEI 3 VEI 4 VEI 5

Probability 0.01 0.09 0.2 0.4 0.2 0.09Column height (km) 0.01–0.1 0.1–1 1–5 3–15 10–25 25–35Erupted mass (kg) 106–107 107–109 109–1010 1010–1011 1011–1012 1012–1013

Initial velocity (m s−1) 10–100 10–200 100–200–400 200–400–600 200–400–600 200–400–600Mdw (probability) –4,–3,–1 –4,–3,–1 –3,–2,–1 –2,–1,0 –2,–1,0 –2,–1,0

(0.25,0.5,0.25) (0.25,0.5,0.25) (0.25,0.5,0.25) (0.2,0.4,0.4) (0.2,0.4,0.4) (0.2,0.4,0.4)Mdw min 7 7 7 7 7 7Mdw max –7 –7 –7 –7 –7 –7sw 1, 0.5 1, 0.5 1, 0.5 1, 0.5 1, 0.5 1, 0.5b 0.008–0.2 0.008–0.2 0.008–0.2 0.008–0.2 0.008–0.2 0.1–0.6k 2000–3000 2000–3000 2000–3000 2000–3000 2000–3000 2000–3000

Table 2. Matrix of input parameters and their relative probability for PDC simulations (Rossano et al. 2004)

Initial velocity (m/s) 2 (0.1) 35 (0.3) 65 (0.3) 100 (0.2) 150 (0.1)Flow thickness (m) 0.5 (0.1) 3 (0.3) 6 (0.3) 10 (0.2) 20 (0.1)Viscosity (Pa s) 1 (0.2) 10 (0.3) 100 (0.3) 500 (0.2)Yield Strength 0 (0.5) 10 (0.2) 100 (0.2) 1000 (0.1)Density (kg m−3) 10 (0.5) 100 (0.4) 1000 (0.1)

Fig. 3. Yearly probability hazard maps computed forthe minimum load for the roof collapse of 200 kg m−2

run for (a) upper limit scenario (VEI 5); (b) eruption-range scenario (VEI 1–4); (c) whole-range eruptionscenario (VEI 0–5).


Fig. 5. Yearly probability hazard map for PDCs computed for whole-range scenario (VEI 0–5).

Fig. 4. Conditional probability hazard maps relative to the minimum load for the roof collapse of 200 kg m−2 runwhole-range eruption scenario (VEI 0–5).


can characterize both small- and large-scaleevents. In fact, according to the volcanic historyof Campi Flegrei, even eruptions of a moderatemagnitude could generate high-mobility PDCs.Dynamic overpressure maps computed byRossano et al. (2004) are also presented inFigure 6.

Discussion and conclusions

Probabilistic volcanological approaches providethe first complete, statistically accurate descrip-tion of the expected tephra fallout and PDCshazard from the whole variety of possible explo-sive events at Campi Flegrei, in the Neapolitanarea and its surroundings.

For fallout events, in the case of small andintermediate eruptions (VEI not exceeding 3)with a relatively low eruptive column, highlyvariable low tropospheric winds prevail, causingthe tephra to be potentially dispersed in alldirections with near-homogeneous probability.

However, in these cases, critical tephra load andthe associated hazards are confined to withinfew km of the vents inside the caldera boundary.

Outside the caldera, the hazard level ofdepends strictly on the scenario being consid-ered. In the case of the eruption range scenario,which is the most useful for civil-defence, signi-ficant (0.00025 events/year) hazards for criticaltephra load characterize a wide sector northeastto southeast within a range of about 10 km fromthe caldera centre, which includes the westernpart of Naples. The area north of Campi Flegreiis exposed to a moderate hazard level. The sec-tors west and south of caldera, exposed to a lowhazard level, are also less important in terms ofrisk, being mostly occupied by the sea. In thewhole range and upper-limit scenarios, an areaup to 60 km from the caldera is potentiallyexposed to a critical load. In the whole rangescenario, the yearly hazard for the area northeastto southeast of the caldera, between c. 10 and60 km, ranges respectively between about

Fig. 6. Dynamic overpressure for PDCs.


0.0005 events/year to 0.0001 events/year poten-tially affecting the districts of Naples, Caserta,Avellino and Salerno. For the upper-limit sce-nario, the yearly probability for critical-loadinghazard maps is ³0.0015 events/a within 30 km ofthe vent, and a decrease up to ³0.001 events/a ata distance of 60 km. In addition, the absoluteprobability (conditional probability) that thecritical load will be exceeded in the case of aneruption has been also computed for the wholeeruption scenario. Figure 6 shows that a signifi-cant probability level of 10% affects the areawithin 30 km of the caldera.

The numerical simulation of millions of pyro-clastic flow-lines, for a wide range of reasonablerheological properties over a topographic modelof Campi Flegrei, reveals that:(1) The maximum range of the most mobile

flows is c. 20 km. Such flows are mostlyNewtonian, and are very fast (v0>100 m s−1),with a>0 m moving front, which spreadsin a circular pattern. They pass over up to400-m-high topographic barriers with onlymoderate effects. However, these flows arerare in the recent history of Campi Flegrei.

(2) Most of the flows with intermediate mobil-ity suffer the effects of the main topographicbarriers, and in particular of the westernsteep slope of Posillipo Hill (c. 200 m a.s.l.).They pass through the valleys contouringthe crater rims and generate an irregularareal pattern of distribution. Most of themoderate- and small-scale pyroclastic flowsdo not affect the city of Naples.

(3) Very viscous, slow-moving flows arestrongly controlled by the rugged topogra-phy of the volcanic field. They are mostlyconfined within a few km of the vent insmall valleys and even within older craterrelicts.

(4) Very high hazard levels and high dynamicoverpressure mostly affect the inner calderaarea; however, the eastern part of the plainnorth of Campi Flegrei, and even Naples,are exposed to the effects of PDCs, but witha probability one order of magnitude lessthan that calculated for Campi Flegrei.

Comparison of the pyroclastic fall and flow-hazard maps reveals the following evidence.Within the caldera rim, the hazard from falloutand PDCs is nearly equivalent (higher than0.0002 events/year), while outside the calderawithin a range of c. 20 km, which includes theprovinces of Naples and Caserta, the PDC haz-ard and dynamic overpressure drop to valuesmore than one order of magnitude less than thecorresponding hazard for critical tephra load.

The area of Naples outside the caldera rim isexposed to a relatively low level of PDChazard (2E-6 events/year), but lies within thezone of significant fallout hazard (0.0005events/year).

References

ANON. 1976. US Standard Atmosphere 1976. USGovernment Printing Office, Washington, DC.

BONADONNA, C., CONNOR, C. B., HOUGHTON, B. F.,CONNOR, L., BYRNE, M., LAING, A. & HINCKS,T. K. 2005. Probabilistic modeling tephra dispersal:hazard assessment of a multiphase rhyoliticeruption at Tarawera, New Zealand. Journal ofGeophyical Research, 110(B03203), doi:10.1029/2003JB002896.

CAREY, S. & SPARKS, R. S. J. 1986. Quantitative modelsof the fallout and dispersal of tephra from volcaniceruption columns. Bulletin of Volcanology, 48,109–125.

CORNELL, W., CAREY, S. & SIGURDSSON, H. 1983.Computer simulation of transport and deposition ofthe Campanian Y-5 ash. Journal of Volcanology andGeothermal and Research, 17, 89–109.

CONNOR, B. C., HILL, E. B., WINFREY, B., FRANKLIN,M. N. & LA FEMINA, C. P. 2001. Estimation of vol-canic hazards from tephra fallout. Natural HazardsReview, February, 33–42.

DEINO, A. L., ORSI, G., DE VITA, S. & PIOCHI, M. 2004.The age of the Neapolitan Yellow Tuff caldera-forming eruption (Campi Flegrei caldera – Italy)assessed by 40Ar/39Ar dating method. Journalof Volcanology and Geothermal Research, 133,157–170.

DE NATALE, G., TROISE, C., PINGUE, F.,MASTROLORENZO, G. & PAPPALARDO, L. 2006.The Somma – Vesuvius volcano (Southern Italy)Structure, dynamics and hazard evolution. Earth-Science Reviews, 74, 73–111.

DE VIVO, B., ROLANDI, G. et al. 2001. New constraintson the pyroclastic eruptive history of theCampanian Volcanic Plain (Italy). Mineralogy andPetrology, 73, 47–65

FRIELANDER, S. K. 2000. Smoke, Dust and Haze:Fundamentals of Aerosol Behaviour. John Wiley,New York.

MCEWEN, A. S. & MALIN, M. C. 1989. Dynamics ofMount St. Helens’ 1980 pyroclastic flows, rockslide-avalanche, lahars, and blast. Journal Volcanologyand Geothermal Research, 37, 205–231

MASTROLORENZO, G. 1994. Averno tuff ring in CampiFlegrei (south Italy). Bulletin of Volcanology, 56,561–572.

MASTROLORENZO, G., BRACHI, L. & CANZANELLA, A.2001. Vesicularity of various types of pyroclasticdeposits of Campi Flegrei volcanic field: evidenceof analogies in magma rise and vesiculationmachanisms. Journal of Volcanology and Geother-mal Research, 109, 41–53.

MIDDLETON, G. V. & SOUTHARD, J. B. 1978. Mecha-nism of Sediment Movement. SEPM, EasternSection, Short Course Lecture Notes.


PERLA, R. I. 1980. Avalanche release, motion, andimpact. In: COLBECK, S. C. (ed.) Dynamics of Snowand Ice Avalanches. Academic Press, New York,NY, 397–462.

ROSSANO, S., MASTROLORENZO, G., DE NATALE, G. &PINGUE, F. 1996. Computer simulation of pyroclas-tic flow movement: an inverse approach. Geophysi-cal Research Letters, 23, 3779–3782.

ROSSANO, S., MASTROLORENZO, G. & DE NATALE, G.1998. Computer simulations of pyroclastic flows onSomma–Vesuvius volcano, Journal of Volcanologyand Geothermal Research, 82, 113–137.

ROSSANO, S., MASTROLORENZO, G. & DE NATALE, G.2004. Numerical simulation of pyroclastic densitycurrents on Campi Flegrei topography: a tool forstatistical hazard estimation. Journal of Volcanol-ogy and Geothermal Research, 132, 1–14.

SUZUKI, T. 1983. A theoretical model for dispersionof tephra. In: SHIMOZURU, D. & YOKOYAMA, I.(eds) Arc Volcanism: Physics and Tectonics, TerraScientific Publishing Company (Terrapub), Tokyo,95–113.

WOODS, A. W. 1988. The fluid dynamics and ther-modynamics of eruption columns. Bulletin ofVolcanology, 50, 169–193.

a. panizza, & g. de natale

Documents