ArticlePDF Available

Abstract and Figures

Since 1979, Piton de la Fournaise (La Réunion) has erupted on average two times per year, with 95 % of these eruptions occurring within an uninhabited caldera. However, lava flows have occasionally impacted populated regions on the island, as in 1977 and 1986. Since 2014, an integrated satellite data–driven multinational response to effusive crises has been developed to rapidly assess lava inundation area and flow runout distance. In 2018, this protocol was implemented as a standalone software to provide a lava flow hazard map showing the probability of flow coverage and runouts as a function of discharge rate. Since 2019, the produced short-term hazard map is shared with local civil protection in the first few hours following the start of an eruption to aid in mitigation actions. Multiple exchanges between scientists, the observatory, and civil protection has improved the delivered hazard map, ensuring a common understanding, a product which is of use and usable, and helping to build effective mitigation strategies at Piton de la Fournaise. In this work we illustrate this effective near real-time protocol with case studies and document how the produced short-term hazard map has been tailored to meet the needs of civil protection.
Content may be subject to copyright.
Editor: L. Gallant | Typesetter: J. Farquharson
Volcanic crisis management supported by near real-time lava flow hazard
assessment at Piton de la Fournaise, La Réunion
Magdalena Oryaëlle Chevrelα, Andrew J. L. Harrisα, Aline Peltierβ,γ, Nicolas Villeneuveβ,γ,δ,
Diego Coppolaε, Mathieu Gouhierα, and Stéphane Drenneζ
αUniversité Clermont Auvergne, CNRS, IRD, OPGC, Laboratoire Magmas et Volcans, 63000 Clermont-Ferrand, France.
βUniversité Paris Cité, Institut de Physique du Globe de Paris, CNRS, 75005 Paris, France.
γObservatoire Volcanologique du Piton de la Fournaise, Institut de Physique du Globe de Paris, 97418 La Plaine des Cafres, France.
δUniversité de La Réunion, Laboratoire GéoSciences Réunion, 97744 Saint Denis, France.
εDipartimento di Scienze della Terra, Universita degli Studi di Torino, Torino, Italy.
ζEMPZCOI, Préfecture de la Réunion, 97744 Saint Denis, France.
Since 1979, Piton de la Fournaise (La Réunion) has erupted on average two times per year, with 95 % of these eruptions occurring
within an uninhabited caldera. However, lava flows have occasionally impacted populated regions on the island, as in 1977 and
1986. Since 2014, an integrated satellite data–driven multinational response to effusive crises has been developed to rapidly
assess lava inundation area and flow runout distance. In 2018, this protocol was implemented as a standalone software to
provide a lava flow hazard map showing the probability of flow coverage and runouts as a function of discharge rate. Since
2019, the produced short-term hazard map is shared with local civil protection in the first few hours following the start of an
eruption to aid in mitigation actions. Multiple exchanges between scientists, the observatory, and civil protection has improved
the delivered hazard map, ensuring a common understanding, a product which is of use and usable, and helping to build effective
mitigation strategies at Piton de la Fournaise. In this work we illustrate this effective near real-time protocol with case studies
and document how the produced short-term hazard map has been tailored to meet the needs of civil protection.
Depuis 1979, le Piton de la Fournaise est en éruption en moyenne deux fois par an et 95 % des éruptions a lieu dans la caldera
principale qui est inhabitée. Des coulées de lave peuvent néanmoins être émises en dehors de cet espace et envahir des villages
comme ce fut le cas en 1977 et 1986. Depuis 2014, un protocole a été mis en place afin de modéliser et d’anticiper rapidement la
trajectoire des coulées et la zone potentiellement affectée par la lave lorsqu’une éruption se déclenche. En 2018, ce protocole
a été amélioré afin de produire une carte d’aléas indiquant la probabilité de couverture de la coulée et les distances qu’elle
peut atteindre en fonction du débit de lave émit. Depuis 2019 cette carte est communiquée à la préfecture dans les premières
heures du début de l’éruption afin d’aider à la prise de décision et d’atténuer les risques potentiels. De multiples échanges
entre les scientifiques, l’observatoire volcanologique et l’Etat-Major de Zone a permis d’améliorer cette carte afin d’assurer une
compréhension commune du risque et d’apporter une aide efficace pour élaborer des stratégies d’atténuation des risques au
Piton de la Fournaise. Dans ce présent article, nous illustrons ce protocole avec des études de cas et décrivons comment les
cartes produites ont été adaptées pour répondre aux besoins de la protection civile.
K: Volcanoes; Lava flow; Hazard map; Civil protection; Modeling.
Lava flows are one of the most frequent volcanic hazards in
the world, yet they are considered to rarely pose a risk to peo-
ple [Witham 2005;Harris 2015]. Indeed, at most sites, lava
flows are slow enough to allow populations to be evacuated
[e.g. Weisel and Stapleton 1992;Duncan et al. 1996;Chester
et al. 1999]. However, examples where lava flows have caused
fatalities do exist [Blong 1984], such as during the flank erup-
tions of Nyiragongo in the Democratic Republic of Congo in
1977, 2002, and 2021. Lavas associated with these Nyiragongo
eruptions were extremely fluid [Giordano et al. 2007;Morri-
son et al. 2020] and thus advanced extremely rapidly (at tens
of km h1[Tazieff 1977]). As a result, lava flows reached the
city of Goma within a few hours, causing dozens of fatalities
either directly (burns, asphyxia) or indirectly (e.g. explosion of
a gas station or car accidents during evacuation) [Harris 2015].
However, by far the greatest problem facing a community
in the path of a lava flow is the destruction of structures by
burning, flooding, and/or burial [Blong 1984]. Any unmov-
able structure, and its contents, will be destroyed including
buildings, businesses (factories, garages, shops), essential com-
munity facilities (schools, churches, sports centers, hospitals),
infrastructure, utilities, and agriculture [e.g. Finch and Mac-
donald 1950;Luhr and Simkin (eds) 1993;Duncan et al. 1996].
Destruction of all of these essential elements of a community
have lasting repercussions on social fabric, mental health, and
local economies [Harris 2015]. For example, after the 2002
Nyiragongo eruption, about 120,000 people were left home-
less, and 15 % of the city and a third of the airport runway
were completely buried by lava [Tedesco et al. 2007]. Fol-
lowing the 2021 Nyiragongo eruption, 3600 houses were de-
Lava flow hazard assessment at Piton de la Fournaise, La Réunion Chevrel et al. 2022
stroyed and more than 400,000 people were affected either by
water supply difficulties or displacement (UNOCHA report of
5 June, 2021). In June 2018, the effusive eruption of K¯ılauea
in the Puna region of Hawaii destroyed more than 700 struc-
tures and caused economic losses estimated at more than 800
million US dollars[Meredith et al. 2022]. Finally, the erup-
tion of La Palma (Canaries), which began in September 2021
and ended in January 2022, buried ca. 12km2of land includ-
ing agricultural fields and villages, destroying more than 1600
houses. As a result, more than 7000 people were displaced§.
To better assess lava flow hazard, run real-time appraisals
of potential lava flow inundation areas and, thus, respond to an
effusive crisis in an effective manner, scientists and civil pro-
tection agents must work together closely to test and validate
operational tools before an event occurs. With an average of
two, mainly effusive, eruptions per year over the last 40 years,
Piton de la Fournaise volcano (La Réunion, France) is a perfect
laboratory to develop and test effusive crisis response proto-
cols [Peltier et al. 2020;2022]. Piton de la Fournaise is an active
basaltic hotspot volcano on La Réunion Island, a French over-
seas department located in the Western Indian Ocean (Fig-
ure 1A). Since 2014, scientists have been working on an in-
tegrated satellite data–driven response for effusive crises at
Piton de la Fournaise, with the aim of assessing, in near real-
time, potential lava inundation areas and flow runout distances
[Harris et al. 2017;2019]. This protocol has now been imple-
mented as a standalone software package that can be tied to
any Geographic Information System (GIS) allowing produc-
tion of a short-term hazard maps within the first minutes to
hours after the start of an eruption [Peltier et al. 2021]. Maps
are a common way of communicating comprehensive hazard
information allowing the spatial extent of the hazard, as well
as information as to key infrastructure and land type at risk,
to be quickly assessed by emergency managers [Pareschi et al.
2000;Thompson et al. 2015].
In the present study, we detail this near real-time protocol,
package, and product. We first provide background on miti-
gating the risks associated with lava flows at effusive centers
in general, and then focus on crisis management at Piton de
la Fournaise. Within this context, we describe the implemen-
tation of the system, including the lava flow numerical mod-
eling, the use of satellite sensor to derive the time-averaged
discharge rate (TADR) and flow outlines, and the production
and communication of lava hazard maps giving the probabil-
ity of lava flow coverage during an on-going crisis. An initial
map is shared within the first hours of any eruption and, when
needed, the map is updated during the eruption. As we ex-
plain, these commonly understandable maps have been devel-
oped through open and iterative discussion between scientists,
the observatory responsible for monitoring and reporting du-
ılauea Eruption Recovery. How much damage was caused by the
2018 eruption?
Copernicus Emergency Service mapping: https://emergency. components/EMSR546
§Cabildo de la Palma: https://riesgovolcanico-lapalma.hub.arcgis.
ties (Observatoire Volcanologique du Piton de la Fournaise
Institut Physique du Globe de Paris; OVPF-IPGP), and local
civil protection (État-Major de Zone et de Protection Civile
de l’Océan Indien; EMZPCOI). We provide details on items
that was requested by the EMZPCOI to improve the delivered
map. We discuss the limitation of our protocol by providing
details on the uncertainties of the model itself, the influence of
the topography that needs to be regularly updated, and the un-
certainties related to the estimation of the discharge rates from
satellite sensors. Finally, we examine the challenges of imple-
menting such a multi-agency protocol, including the timeline
of production and advances for future crises, and consider the
potential for possible export of our protocol to other effusive
1.1 Background on mitigating the risks associated with lava
Risk mitigation associated with lava flow consists of prepared-
ness, evacuation, and sometimes attempts to divert or delay
the advance of the flow [for review see Harris 2015]. As high-
lighted by Harris [2015]:
preparation and planning for losses remains the
best way to reduce the severity, or moderate the
impact (of lava flow ingress into a populated area)
while also alleviating the loss incurred”.
During effusive eruptions, monitoring lava flow emplacement
is a common action carried out by volcano observatories to
track and assess land and infrastructure lost or at risk. Since
the 1990–2000s, the development of event onset alerting tools
using high-spatial resolution satellite sensors operating in the
infrared have been providing support to observatories [Har-
ris 2013;Harris et al. 2016]. More recently, numerical lava
flow models (stochastic or deterministic) have also been devel-
oped to include satellite-derived source terms (mostly TADR
in m3s1) and to allow real-time lava flow emplacement sim-
ulations and projections. This has permitted construction of
operational tools, such as generation of hazard maps to assess
potential lava flow inundation area and to project lava flow
paths [Wright et al. 2008;Ganci et al. 2012;Harris et al. 2019].
These tools have been provided as support for stakeholders
involved in preparing and planning for loss (e.g. volcano ob-
servatory, civil protection, and/or national, regional, and local
government agencies).
Since 2007, the Hawaiian Volcano Observatory (HVO) has
been evaluating the short-term threat to communities posed
by lava flows through frequent airborne and satellite map-
ping of lava flow fields and assessing steepest descent paths
[Kauahikaua 2007]. These techniques have regularly been used
during effusive activity at K¯ılauea to build hazard maps in-
cluding likely lava flow paths, and routes for communication
with emergency managers and the public have been well-
developed [Kauahikaua et al. 2017;Brantley et al. 2019]. In
2014, during the lava flow crisis in the vicinity of the town of
ahoa, scientists from the HVO measured lava flow advance
rates and provided a range of potential arrival times that were
used to build emergency plans [Poland 2016]. During the 2018
Presses universitaires de rasbourg
Page 314
5(2): –.
eruption in the Puna district, HVO rapidly produced prelim-
inary lava flow path forecasts using the DOWNFLOW model of
Favalli [2005], a commonly used stochastic model that com-
putes the line of steepest descent (LoSD) and probable lava
flow inundation area. Throughout the eruption, lava flow path
simulations were run on regularly updated topography, and
from active flow fronts, new fissures, and channel overflow lo-
cations, to yield maps of likely future flow directions and inun-
dation areas [Neal et al. 2019]. These maps facilitated commu-
nication of up-to-date hazard information to emergency man-
agers and provided situation awareness for eruption response
field crews.
On Mount Etna (Italy), the Istituto Nazionale di Geofisica
e Vulcanologia (INGV) at the Etna Observatory (EO) mod-
els and tracks lava flow emplacement using a combination of
satellite-derived discharge rate estimates and numerical mod-
els [Vicari et al. 2011]. INGV-EO employs the HOTSAT satellite
monitoring system to convert infrared data from both MODIS
and SEVIRI sensors into TADR that is then input in the nu-
merical model GPUFLOW.GPUFLOW, an evolution of MAGFLOW,
is based on a cellular automaton, physics-based model which
can reproduce, spatially and temporally, the likely lava flow
propagation [Vicari et al. 2011;Ganci et al. 2012;Del Negro et
al. 2013;Bilotta et al. 2016]. The combination of HOTSAT and
GPUFLOW is integrated into a web-GIS system, LAV@HAZARD,
that produces real-time hazard assessment by providing time
of propagation of lava flow fronts, maximum runout distance,
and area of inundation [Ganci et al. 2016]. This operational
tool was described and validated using a retrospective analy-
sis of Etna’s 2008–2009 flank eruption by Ganci et al. [2012]. It
has since been operational by providing automatic reports in-
cluding the radiant heat flux, TADRs, and a short-term hazard
map. These products are communicated to INGV and local
civil protection through periodic bulletins. Although the mon-
itoring strategy implemented in LAV@HAZARD was designed for
Etna, it has been adapted and employed during volcanic crises
at other sites, such as Fogo at Capo Verde [Cappello et al. 2016]
and for Nabro in Eritrea [Ganci et al. 2020].
At Piton de la Fournaise, a well-established protocol be-
tween the OVPF-IPGP, local civil protection (EMZPCOI), and
scientists from a multi-national array of institutes has allowed
effective tracking of effusive crises and hazard management
[Peltier et al. 2020;2022]. Forecasting and modeling lava flow
inundation areas through a satellite data–driven protocol to
mitigate their impact started to be developed in 2014 [Harris
et al. 2017]. The DOWNFLOW model was used to provide the
LoSD and probable lava flow inundation area [Favalli 2005]
and FLOWGO was used to calculate the maximum runout of
lava for a given effusion rate [Harris and Rowland 2001;Harris
2015]. This protocol and near real-time application, as tested
and validated during the April 2018 eruption of Piton de la
Fournaise, was presented in Harris et al. [2019] and its im-
provement and final, operational implementation is provided
and discussed herein.
1.2 Eruptive activity at Piton de la Fournaise
Since the first documented eruption on La Reunion in 1708,
95 % of the eruptions of Piton de la Fournaise have oc-
curred inside the Enclos Fouqué caldera (Figure 1B) [Vil-
leneuve and Bachèlery 2006;Chevrel et al. 2021]. The En-
clos Fouqué caldera is open towards the ocean to the east
and its formation has been accompanied by successive land-
slides [Bachèlery 1981], forming a horseshoe-shaped depres-
sion, hereafter called the Enclos. Vents generally open within
the Enclos, with lava flows being contained within the depres-
sion and flowing downhill to the east and towards the ocean
(Figure 1B, C). Eruptions are fed by magma dikes or sills that
propagate along preferential paths within, tangentially, or ra-
dially around the summit crater (i.e. the Dolomieu crater) of
the central terminal shield inside the caldera and along three
main rift zones orientated southeast, northeast, and 120°N
[Bachèlery 1981;Dumont et al. 2022]. Since at least the last
century, the majority of eruptions have consisted of fissure-
fed lava fountaining and lava flows [Peltier et al. 2012].
Eruptive activity is controlled by cycles related to the evolu-
tion of the superficial edifice stress field that usually last from
one to eleven years and control the location of eruptive fis-
sures [Peltier et al. 2009;Got et al. 2013;Derrien 2019] and to
magma source recharge [Vlastélic et al. 2018]. These cycles
usually start with summit or summit-proximal eruptions and
end with a distal or eccentric eruption that occurs more than
4 km from the summit and can be located inside or outside
of the Enclos. In contrast, deep source-related cycles are one
to three decades long and control eruption output rates and
erupted volumes. These cycles end with summit crater col-
lapse and voluminous eruptions (a “hotspot surge”) followed
by few years of quiescence [Vlastélic et al. 2018]. This was the
case, for example, in April 2007 when the most voluminous
low-elevation flank eruption so far observed on La Réunion
[Staudacher et al. 2009] marked the end of the 1998–2007 cy-
cle [Vlastélic et al. 2018]. The April 2007 vent was located at
590 m a.s.l., and lasted one month (30 March–1 May, 2007)
and produced 140 to 240 ×106m3of lava [Staudacher et al.
2009;Roult et al. 2012]. Lava flows were fed by a high effusion
rate (around 200m3s1) and reached the island belt road, lo-
cated 2.2 km from the vent, in less than 5 hours [Harris and
Villeneuve 2018]. By the end of the eruption, the belt road that
links the south and north of the island had been buried over a
length of 1.4 km by a 50 m thickness of lava [Staudacher et al.
2009]. The April 2007 event was followed by three years of
low intensity activity inside or close to the summit at a mean
output rate of 1.8 ×106m3yr1. This compared with 24 ×
106m3yr1towards the end of the 1998–2007 cycle [Roult
et al. 2012] and was followed by four years of inactivity from
2010–2014 [Peltier et al. 2018].
In June 2014, Piton de La Fournaise entered a new period
of high eruptive frequency. The single day eruption on 20
June, 2014 was followed by four eruptions in 2015, four dur-
ing 2016–2017, and then 14 during 2018–2021 (Figure 1C).
The 2014–2021 eruptive period was characterized by both
short and long duration eruptions (0.7 to 55 days) and mean
output rates of 1.5–18m3s1(mean of 6.8m3s1; annual av-
erage of 14 ×106m3yr1). The majority of eruptive fissures
were located within 2.5 km of the summit above 1800 m a.s.l.
Average based on database extracted from several reports that can be
found here:
Presses universitaires de rasbourg
Page 315
Lava flow hazard assessment at Piton de la Fournaise, La Réunion Chevrel et al. 2022
and inside the Enclos (Figure 1C). Exceptions were the two
last eruptions of 2019 (starting on 11 August and 25 Octo-
ber, respectively) during which fissures opened in the Grandes
Pentes (steep slopes) area at distances of 4.5 and 6 km from the
summit at 1450 and 1000 m a.s.l., respectively. For both erup-
tions, lava flows reached the area of the Grand Brûlé (deeply
burnt) and stopped 2 km and 220 m from the belt road dur-
ing the two eruptions, respectively (Figure 1C). Additionally,
lava flows associated with the September 2016, July 2018, and
February–March 2019 eruptions buried several hundred me-
ters of the hiking trail network [ Peltier et al. 2022].
1.3 Volcanic crisis management at Piton de la Fournaise
Although most eruptions take place inside the uninhabited
Enclos, this area remains highly frequented, with at least a
hundred thousand hikers access the summit every year [Der-
rien et al. 2018]. Furthermore, the lower eastern flank of the
Enclos is crossed by the island belt road (RN2), which repre-
sents the only access between the southern to northern parts
of the island along the east coast. Since 1950 the road has
been cut 10 times by lava flows [Peltier et al. 2022]. On 12 oc-
casions since the beginning of the 18th century eruptions have
fed lava flows outside of the Enclos (called Hors Enclos erup-
tions) where populations are at risk; this was the case recently,
in 1977, 1986, and 1998 [Chevrel et al. 2021]. While the 1998
event produced short lava flows at a high elevation, the 1977
and 1986 eruptions fed lava flows that entered the populated
coastal area. The April 1977 eruption required evacuation
of Piton Sainte-Rose, burnt down or damaged ~30 buildings
(houses, the church, the police station and a gas station), and
buried a portion of the village [Kieffer et al. 1977].
Following the 1977 eruption, the French Ministère des Uni-
versités, the département de La Réunion, the Institut Na-
tional d’Astronomie et de Géophysique (now called CNRS)
and the Institut de Physique du Globe de Paris (IPGP) estab-
lished the Observatoire Volcanologique du Piton de la Four-
naise (OVPF) with a remit to monitor the volcano. Addition-
ally, a national response organization plan was established in
the case of eruptions at Piton de la Fournaise by the French
government for civil protection and associated agencies. This
plan, still valid today, is named the “ORSEC-DSO (Organi-
sation de la Réponse de SEcurité Civile- Dispositions Spéci-
fiques Opérationnel), Volcan du Piton de la Fournaise”, and
defines the warning levels to be set, as well as the actions
to be taken and restrictions to be implemented at each level,
the protocol to change the level, and the communication strat-
egy for alert diffusion [Peltier et al. 2018;2020;2022]. As part
of the plan, in the case of a change in activity (i.e. onset of
unrest, new fissure opening, end of eruption), OVPF must in-
form the civil protection headquarters of the Indian Ocean
zone (EMZPCOI) and propose a change in the alert level. The
decision to officially change the alert level is the responsibility
of the Préfecture, (i.e. local government headquarters). The
Préfecture then communicates with other actors (town coun-
cils, gendarmerie (police), central authorities, institutions, and
the media).
The four levels of alerts are:
Vigilance: possible eruption in the medium term (a few
days or weeks) or presence of risk (rockfalls, increased gas
emissions, inactive but still hot and cooling lava flows ...);
Alert 1: probable or imminent eruption;
Alert 2: ongoing eruption;
Alert 2-1: ongoing eruption inside the Enclos without
threat to the safety of people, property or the environment;
Alert 2-2: ongoing eruption inside the Enclos with threat
to the safety of people, property or the environment;
Alert 2-3: ongoing eruption outside the Enclos with
threat to the safety of people, property or the environment;
Sauvegarde: end of eruption.
As soon as Alert 1 is triggered, an inter-services meeting
is organized at the Préfecture to (i) establish a situation re-
port to inform the response services, and (ii) prepare the crisis
management team and response system. The inter-service
consortium of responders includes:
the police force (gendarmerie),
the town councils,
the fire department (SDIS: Service départemental
d’incendie et de secours),
the department of environment, planning and housing
(DEAL: Direction de l’environnement, de l’aménagement et
du logement from Préfecture),
the national forest office (ONF: Office National des
Forêts), and
if required, the departmental and regional council (con-
seil départemental de la Réunion,conseil Régional), the
army (FAZSOI: Forces armées dans la zone sud de l’Océan
Indien), the emergency health services (SAMU: Service d’Aide
Médicale Urgente, and ARS: Agence Régionale Santé) are also
included in the consortium.
From alert level 1 onwards, access to the Enclos Fouqué is
strictly limited to authorized personnel.
During an eruptive crisis OVPF communicates daily with
EMZPCOI via daily bulletins and phone calls. In the case of
a “special situation”, for example when there is an imminent
threat to population or infrastructure, as is the case for erup-
tions located outside the Enclos or when there is a risk of the
belt road being cut, the Centre des Operation de la Préfec-
ture (COP) is activated. Activation of the COP means that the
system to organize and implement mitigation actions for the
population and infrastructure at threat, or road closure, is trig-
gered. If the situation requires, one or more operational com-
mand posts (PCO: Poste de Commandement Opérationnel)
and advanced medical stations (PMA: Poste Médical avancé),
may also be installed as close as possible to the impacted area.
This is a decision made by the Préfet (head of the Préfecture).
To date, the COP has been activated only once, in April 2007.
Presses universitaires de rasbourg
Page 316
5(2): –.
Figure 1: Location of Piton de la Fournaise and its lava flows since 2014. [A] Location of La Réunion island (credits Wikimedia
Commons, the free media repository). [B] Map of Piton de la Fournaise showing the municipalities outline (pink dashed lines)
and towns (buildings are in black), roads (yellow lines), trails (orange lines), the vegetated areas (in green) and the limits of the
Enclos caldera (blue line). The background is a hill-shaded DEM released in 2010 (coordinates are given in meters within system
WGS84-UTM 40S) by the Institut national de l’information geographique et forestiere (IGN). Buildings, roads, and trails are from
2019 BD TOPO®IGN. Yellow dots are the volcano observatory monitoring stations. [C] Zoom of the Enclos showing the fissures
(black lines) and associated lava flows (color coded) since 2014 and up to early of 2022 (legend indicates the eruption starting
date: YYYYMMDD). The dashed black lines divide the Enclos into three main domains: Enclos Fouqué Caldera, Grandes Pentes,
and Grand Brûlé. Black arrow shows the summit Dolomieu Crater.
At the alert level Sauvegarde, the eruption is over but re-
strictions on access to the volcano remain in place until activity
indicators (seismicity, gas flux, and ground deformation) are
considered to have returned to “normal”. Reconnaissance is
also carried out to check that the area is safe for reopening.
The alert level then reverts to vigilance”.
     OVPF
A number of numerical models exist to support lava flow mod-
eling (see review by Dietterich et al. [2017]). Here, we com-
bine the stochastic model DOWNFLOW [Favalli 2005] with the de-
terministic model FLOWGO [Harris and Rowland 2001;Harris
2015] to support hazard assessment at Piton de la Fournaise.
DOWNFLOW provides the most likely lava flow paths, includ-
ing the LoSD and area of coverage. For responding to current
crises at Piton de la Fournaise, this model was calibrated by
fitting the output flow coverage to the actual areas of all flow
fields since 2016 [Chevrel et al. 2021].
FLOWGO calculates the runout distance of a lava flow along a
slope line for a given effusion rate [Harris and Rowland 2001].
It is 1-D model adapted for cooling-limited lava flow in a chan-
nel of uniform depth that, once calibrated with a suitable rhe-
ological model, only needs the slope from the vent along the
steepest descent line and a discharge rate as its source terms
(see Chevrel et al. [2018] for more details). Once initialized,
to assess flow runout, FLOWGO takes a control volume of lava
and tracks its cooling and crystallization down-channel until it
becomes too viscous for further forward motion to occur, i.e.
velocity approaches zero [Harris and Rowland 2001]. FLOWGO
was calibrated to be run on the LoSD obtained via DOWNFLOW
and using a retrospective analysis of the April 2018 eruptions
[Harris et al. 2019]. If needed, new calibration can be car-
ried out at any time if source terms are known to change or
if the model fit with the actual flow begins to break down. It
must be stressed that FLOWGO output is only appropriate for
channel-fed, cooling-limited flow of basaltic composition.
I    - 
 
3.1 Original chain of tasks
The rapid response lava flow simulation protocol began with
the eruption of June 2014 [Harris et al. 2017]. Between 2014
and 2018, the chain of tasks comprising this protocol involved
multinational partners and each step was operated by a dif-
ferent institute in a different country and was collated on a
centralized reporting page [Harris et al. 2019]. As part of this
protocol, five sequential actions were executed:
1) Action 1: At the start of an eruption, OVPF communi-
cated the vent or fissure coordinates (obtained either via visible
observation or location of the tremor that is available online
on the WebObs platform [Beauducel et al. 2020]) to INGV in
Pisa (Italy) where a lava flow inundation map was produced
using DOWNFLOW on the most recent Digital Elevation Model
2) Action 2: The DOWNFLOW-derived LoSD was sent to
the Laboratoire Magmas et Volcans (LMV) at the Université
Clermont-Auvergne (UCA, France) where the FLOWGO model
was run to estimate maximum the lava flow runout distances.
Presses universitaires de rasbourg
Page 317
Lava flow hazard assessment at Piton de la Fournaise, La Réunion Chevrel et al. 2022
Note that uncertainties for runout distances are of the order
of 30 % [Harris et al. 2019].
3) Action 3: The key source term for FLOWGO is the dis-
charge rate. This value was derived from satellite-based ther-
mal infrared images with the MIROVA platform at the Uni-
versita di Torino in Italyand with the HOTVOLC platform at
the Observatoire de physique du globe de Clermont-Ferrand
(OPGC) at UCAtwo volcanic hotspot detection systems. Un-
certainties on the satellite-derived TADRs are of the order
of 50 % for HOTVOLC and 35 % for MIROVA (see discussion).
The cumulative volume can also be calculated by integrating
TADR values over the period of time in between two satellite
images using the trapezium rule.
4) Action 4: The resulting lava flow area coverage obtained
via Action 1, and the lava flow runout maximum distance ob-
tained for the given discharge rate (Actions 2 and 3) repre-
sented as a graph of lava velocity versus distance from the vent
to the flow front were then communicated to OVPF [Harris et
al. 2017].
5) Action 5: In parallel, the evolution of lava flow field
(vent location, channel dimension, flow outline) was obtained
from airborne photogrammetry acquired by OVPF. In addi-
tion, interferograms and coherence maps derived from InSAR
via the OI2platform developed at the OPGC, UCAand satel-
lite infrared and visible imagery acquired by the ASTER Ur-
gent Request Protocol [Ramsey et al. 2016] were used to assess
zones covered by lava. These products were also used to close
the loop by up-dating and cross-checking the performance of
DOWNFLOW and FLOWGO [Harris et al. 2019].
This protocol was triggered in real time at the start of any
given eruption through release of an email from the director of
OVPF with vent coordinates and was designed so that it could
be repeated and updated as the eruption conditions evolved
(e.g. new vents opening, extension of tubes, change in TADR).
The response protocol was reviewed and updated throughout
each of the ten eruptions from June 2014 to April 2018, at
which point the protocol became fully operational [Harris et
al. 2019].
3.2 Improved chain of tasks
The system became centralized in 2019 to improve efficiency.
The execution of DOWNFLOW and FLOWGO was combined in
DOWNFLOWGO as part of the upgrade [Harris et al. 2019;Peltier
et al. 2021], thereby merging Actions 1 and 2. This was possi-
ble through PyFLOWGO, an integrated python code that allows
the actions to be executed by a single operator at a single lo-
cation on any operating system [Chevrel et al. 2018]. Action
3 was also streamlined because the original process required
a new simulation to generate the flow runout projection with
each change in the TADR, which was time-consuming and
inefficient. Simulations are now run for a range of likely effu-
sion rates (typically between 5 and 50m3s1at increments of; [Coppola et al. 2016].; [Gouhier et al. 2016].; [Hrysiewicz 2019;Richter
and Froger 2020].
5m3s1). In essence, once the vent coordinates are entered in
the code line, a raster of the most probable area to be covered,
the path of steepest descent, and the runout locations (coor-
dinates and elevation) along this path are computed. This is
carried out on a 10 m resolution DEM (here we use the DEM
acquired via LiDAR in 2010) that is 130 ×106m2and where
10,000 DOWNFLOW runs are calculated, as calibrated by Chevrel
et al. [2021]. Then 𝑥runs of FLOWGO for 𝑥effusion rates, at
10 m steps down the LoSD are made, this being the ideal it-
eration step size to ensure numerical convergence [Chevrel et
al. 2018] to give the runout points. The complete computing
time is less than two minutes. The results are then imported
into Q-GIS, a free and open-source Geographic Information
System, that already contains a pre-prepared template having
all the needed layers, including locations of the OVPF mon-
itoring network and all features required by civil protection
(see Section 5 for details). This improved protocol is given in
Figure 2 and only requires the vent coordinates of the starting
eruption to prepare and deliver the map, with the map being
prepared in a few minutes by a single operator.
The delivered short-term hazard map (e.g. Figures 3,4,5,
6and 7) provides:
1) The path of steepest descent from the main vent location
(in case of multiple vents along a fissure, several simulations
may be run);
2) The probable area of lava flow coverage (i.e. the proba-
bility that a pixel will be hit by lava); and
3) The runout distances (i.e. where the lava should stop)
for the pre-defined effusion rate range along the LoSD.
The map is then used as a guide to assess probable runout
given current TADR obtained from the satellite detection sys-
tems. Actions 3 and 5 are still implemented through email
exchange and/or consultation the dedicated websites. This
includes update of TADR from MIROVA or HOTVOLC to allow
the look-up-based assessment to be revised, as well as provi-
sion of ASTER images and InSAR coherence maps to allow
on-the-fly validation of the hazard map and possible update.
The map is first reviewed and approved by the OVPF di-
rector, who is then responsible for transmitting the map to the
local civil protection. Due to difficulties of representation, un-
certainties are not represented on the maps but known by all
actors. For this, regular explanation of the modeling is done
during meetings with civil protection (see Sections 5and 6).
This is essential to ensure that all duty staff know that these
maps are a first-order estimation of the area to be covered
and that the lava can always take another direction or go fur-
ther than predicted. By common agreement, the maps are not
open to the general public during an eruption. This is to avoid
misinterpretation from untrained operators and the spread of
false information. However, following any given eruption, the
maps are published in the OVPF monthly bulletin§.
fevrier-2022; [OVPF 2022]
Presses universitaires de rasbourg
Page 318
5(2): –.
Figure 2: Protocol for the near real-time hazard assessment response at Piton de la Fournaise since 2018. From top to bottom,
the first level (light grey boxes) are the input data collected during the eruption. The second level (white boxes) are the output
parameters collected in level 1 and used to implement the software in level 3. The third level (dark grey) are the two employed
software DOWNFLOWGO and any Geographic Information System. DOWNFLOWGO is composed of (i) DOWNFLOW that needs the topog-
raphy (DEM) and the calibrated simulation parameters (Δhand N) and computes the line of steepest descent (LoSD) and the
probable lava flow area, and (ii) PyFLOWGO that uses the lava thermo-rheological properties to compute the runout distances
along the LoSD for a given effusion rate. The GIS needs to contain all required layers (geology and infrastructure) and a map
template so to produce the consistent final product that is the short-term lava flow hazard map to be delivered to the authorities.
Acquired lava flow outline by field and airborne survey or satellite data acquired during the eruption can be added as a layer as
and when available.
S : S 
To illustrate our operational near real-time protocol, we
present the response to three eruptive events as case studies,
these being the eruptions of 25–27 October, 2019 (an eruption
that threatened the road belt), 10–16 February, 2020 (an erup-
tion that started only 20 minutes after the start of the seismic
crisis), and 7–8 December, 2020 (when the response protocol
was triggered in a record time after the eruption onset). Addi-
tionally, we provide a hypothetical case study of an eruption
outside of the main caldera in an inhabited area, based on
the scenario proposed in Tadini et al. [2022]. For each case
study we describe the eruption timeline (given in the local
time zone, i.e. RET, UTC+4) and include the maps that were
communicated to the authorities (note that original maps have
legends in French but they are here translated into English), a
map showing the final lava flow outline, TADRs as function of
time as obtained from HOTVOLC and MIROVA, and the respec-
tive cumulative volume. The end of the eruption, as declared
by OVPF when tremor ceases, is also given.
4.1 25–27 October, 2019
The 25–27 October, 2019 eruption was characterized by the
location of the vent that was the lowest in elevation since the
2007 eruption, and the island belt road was threaten by the
lava flow. A seismic crisis started at 04:15 RET (UTC+4) on 25
October, 2019 after ~14 days of edifice inflation. At Piton de la
Fournaise, such seismic crises have been interpreted as a sign
that the shallow magma reservoir has pressurized to the ex-
tent that a diking event occurs, triggering magma propagation
out of the chamber and towards the surface [Peltier et al. 2009;
2018]. After less than 1 hour of seismic crisis, the location of
the earthquakes and of the origin of ground deformation indi-
cated magma propagation towards the eastern flank. A total of
827 shallow volcano-tectonic earthquakes below the summit
and ground deformation of ~12 cm were recorded over the
following ten hours [OVPF 2019]. Volcanic tremor, which at
Piton de la Fournaise is associated with magma reaching the
surface with opening of a fissure, and thus interpreted as the
onset of eruption, appeared on the seismic records at 14:40.
From the location of the tremor source, the eruptive site was
located at a low elevation in the Southeast sector of the Enclos
in the Grandes Pentes area (Figure 1). We produced the first
hazard map at 16:00 local time using this approximate loca-
tion (Figure 3A). This initial product was communicated with
OVPF only because the exact location of the eruptive fissure
was not yet confirmed through visual observation, and was
used to assess what parts of the monitoring network could be
at risk and what likely scenarios to prepare operations for.
An aerial survey via helicopter took place 2.5 hours after the
eruption onset and revealed two eruptive fissures in the south-
southeastern sector of the Enclos, at 1060 m and 990 m a.s.l.,
respectively. At the time of the survey, a third, smaller upper
eruptive fissure was no longer active, and activity had focused
on the lower fissure. At 16:45, a first TADR of 6m3s1was
obtained from HOTVOLC. This value was likely underestimated
due to extremely cloudy conditions and was thus treated with
caution. From the improved vent location obtained from ob-
servations on the overflight, the lava flow simulation was re-
run and a new map produced (Figure 3B), which was sent
to EMPZCOI at around 18:00 (3.3 hours after the start of the
eruption). From this map it became clear that the belt road
could be cut given lava flow fed at any discharge rate greater
than 20m3s1.
At 22:30, lava discharge rate from HOTVOLC exceeded
25m3s1but a first MODIS image was acquired 3 hours later
at 01:30 (on 26/10, 11 hours after the start of the eruption) from
Presses universitaires de rasbourg
Page 319
Lava flow hazard assessment at Piton de la Fournaise, La Réunion Chevrel et al. 2022
Figure 3: [A] First map produced on 25 October, 2019 at 16:00 RET with the location of the vent estimated from the tremor map.
Yellow stars represent the distance at which lava could extend along the LoSD for a given TADR (numbers are in m3s1). [B]
Update of the map produced on 25 October, 2019 at 18:17 with the location of the vent estimated during a helicopter overflight.
[C] Update of the map produced on 26 October, 2019 at 18:00 with the location of the vent and the lava flow front obtained during
field survey. [D] Final map showing the probability of lava flow invasion, red arrows represent the runout distances as a function
of discharge rate. The fissure locations and final lava flow outline obtained by photogrammetry are also given in green and black,
respectively. This map was produced after the end of the eruption and published in the OVPF monthly report of October 2019
[OVPF 2019].
which a lower discharge rate of between 7 and 14 m3s1was
obtained. By the next morning, HOTVOLC indicated a decline in
discharge rate with values <10m3s1(Figure 3D), thus sug-
gesting also a decrease of runout potential. On the same day
(26 October), improved weather conditions also allowed a field
survey to obtain a more accurate vent location. The hazard
map was subsequently updated and sent to EMPZCOI at the
end of the day. This map showed the exact location of the
main vent and the position of the lava flow front at that time,
which matched the simulated lava flow path and runout for a
TADR of 20m3s1(Figure 3C).
The eruption stopped the following day, on 27 October
at 16:30 RET after 1 hour of gas pistoning activity [OVPF
2019]. Overall, lava discharge rates estimated from the satel-
lite platforms were up to 27m3s1for the first 20 hours of the
eruption, which then decreased to <10m3s1thereafter (Fig-
ure 4A), although these values were likely underestimated due
to extensive cloud cover. Final cumulative volume obtained
from TADRs was 1.0±0.35 ×106m3and 0.8±0.4×106m3for
MIROVA and HOTVOLC, respectively (Figure 4B), which repre-
sents 50 to 62 % less than the total lava volume of 1.6 ×106m3
obtained from photogrammetry (for details on the method, see
Derrien [2019]), which confirms an underestimation in TADR.
The final lava flow length was 3.6 km and the flow front
stopped 230 m short of the belt road. The flow front location
corresponded to the flow front runout modeled for a TADR of
20m3s1(Figure 3D). The final lava flow field outline from
post-eruption airborne photogrammetry revealed a good fit
with the simulated path (Figure 3C), further showing that our
calibrated model performed well. Note that uncertainties on
runout distances are of the order of 30 % [Harris et al. 2019],
which in this case is ±1 km. Given this degree of uncertainty
it was thus not possible to disregard the threat to the road.
4.2 10–16 February, 2020
This eruption was marked by a very short time between the
onset of precursory activity and the start of the eruption. In-
deed, the eruption began only 20 minutes after the onset of
Presses universitaires de rasbourg
Page 320
5(2): –.
Figure 4: [A] Time-average discharge rate as a function of time since the start of the eruption, and [B] lava cumulative volume
obtained by HOTVOLC and MIROVA. Uncertainties are 35 % for MIROVA and 50 % for HOTVOLC.
the seismic crises which gave little reaction time to produce
the short-term hazard map.
On 10 February, 2020, a seismic crisis started at 10:27 RET
(06:27 UTC) with more than 200 shallow volcano-tectonic
earthquakes in 30 minutes. This was accompanied by rapid
surface inflation of up to 27 cm [OVPF 2020]. Volcanic tremor
was detected at 10:50, located on the southeastern flank of the
summit shield. Due to fog, no information on the precise fis-
sure location was known until 11:15, when a local resident
called the OVPF while observing a gas plume sourced on the
east of the volcano, in the vicinity of an existing scoria cone
(Marco Crater). The first simulation was therefore run ar-
bitrarily from a few hundreds of meters above of this cone
and a map was delivered to EMPZCOI at 12:00 (Figure 5A).
Given that the fissure extended across the boundary of two
municipalities, EMPZCOI requested that these boundaries be
added to the map. The first lava discharge rate estimation
with MIROVA was obtained at 13:20 (2.5 hours after the onset of
the eruption) and indicated a range of 3 to 5m3s1.HOTVOLC
provided an initial, isolated (single-point) peak at 36 m3s1, af-
ter which values were more consistent with those of MIROVA.
Due to the cloudy conditions these values were likely under-
estimated, but these initial estimates allowed the first assess-
ments of likely flow runout (Figure 5B).
In the early afternoon, an aerial survey revealed that
the main eruptive fissure was oriented parallel to the slope
from west to east, extending downslope between 2450 m to
1990 m a.s.l. It was not possible to place a single point for
the source of emission because the fissure was parallel to the
slope. We therefore performed a series of simulations along
the fissure to consider the full range of source elevations. To
avoid overcrowding the map, we decided to only represent
the paths from the upper and lower ends of the fissure, these
representing the en-member bounds on flow runout. An up-
dated map was then delivered to EMPZCOI at 15:00 on which
MIROVA-derived TADRs were highlighted (Figure 5B). For this
new map, EMPZCOI asked for the addition of vegetated areas
so that they could quickly inform the relevant municipality
as where their respective fire departments should intervene
in case of wildfire. Our simulations suggested that the lava
flow would reach the Grande Pentes and the Grand Brulé at
discharge rates of about 20 and 30m3s1, respectively, and
that the municipality of Sainte Rose, to the north, was the
most likely to be affected by wildfires (Figure 5B). Two other
smaller flows located to the southwest of the main fissure were
also observed but these were not active by the time of the
On 11 February, a coherence map was acquired via InSAR
and the lava flow field outline was extracted by the OI2plat-
form. This outline showed a lava flow front near the expected
runout distances for discharge rates of 10–12m3s1, meaning
that the flow would have already reached its cooling-limited
length if the discharge rate was as reported by HOTVOLC and
MIROVA (Figure 5C). For the next two days, discharge rate esti-
mation was not possible due to cloud conditions. Images were
acquired on February 13 and 14 (63 to 109 h after the start of
the eruption), indicating TADRs of 8–13m3s1, as estimated
by MIROVA and around 5m3s1as estimated by HOTVOLC (Fig-
ure 5D).
Seismic tremor decreased abruptly around 14:00 on
15 February, and the eruption stopped 24 hours later at 14:12
on 16 February. Once the final lava flow outline had been
obtained from photogrammetry after the eruption, a new sim-
ulation was performed from the exact location of the main lava
source to compare performance of the modeling (Figure 5C).
This revealed that the flow had indeed reached its cooling-
limited length within the first two days, and did not extend
further; with the eruption continuing for three more days at
approximately the same discharge rate. During this period,
instead of extending downslope, the flow extended laterally,
forming branches to the north as revealed by the difference
between the outline obtained on 11 February by InSAR and
final flow outline (Figure 5C). Thickening of the flow field by
superposition of lava units could have also taken place. In
total, the erupted volume was estimated from the aerial pho-
togrammetry at 3.8 ×106m3, and this was consistent with the
value of 3.5±1.2 ×106m3obtained with MIROVA (Figure 5D).
The HOTVOLC system provided a lower volume of 2.2 ±1.1 ×
106m3, and we note that this underestimation was likely due
to cloudy conditions.
4.3 7–8 December, 2020
For the 7–8 December, 2020 eruption the response protocol
was triggered in a record time, with the map being provided
to EMPZCOI just a few minutes after the initial aerial survey
of the eruption site, which itself was just a couple of hours
after the eruption began.
Presses universitaires de rasbourg
Page 321
Lava flow hazard assessment at Piton de la Fournaise, La Réunion Chevrel et al. 2022
Figure 5: [A] First map produced on 10 February, 2020 at 12:00 RET. Note that the color code for the lava flow inundation
probability is now from light yellow to red (see Section 6 for the reason of the change). [B] Updated map produced on 10
February, 2020 at 16:00 with the precise location of the vent. This map shows two simulations, one from the highest elevation of
the fissure and one from the lowest. Green arrows show the location of the lava runout along the LoSD for given discharge rates
(numbers are in m3s1) and red arrows locate the TADR obtained from MIROVA (5 m3s1) and from HOTVOLC (35 m3s1). Dashed
black line represent the lava flow outline obtained from MODIS image with the front location as observed at 13:30. [C] Final map
showing the probability of lava flow invasion from the main vent (as identified after the end of the eruption), the syn-eruptive
lava flow outline obtained on 11 February, 2020 using InSAR data and the final outline obtained by photogrammetry. This final
map was produced after the end of the eruption and is an updated version of the one published in the OVPF monthly report of
February 2020 [OVPF 2020]. [D] Lava discharge rate and cumulative volume since the start of the eruption obtained by HOTVOLC
On 4 December, 2020, after a slight ramp-up in seismic-
ity starting on 3 December and inflation in the first few
days of December, a dramatic increase in seismicity began
at 05:10 RET and was accompanied by small levels of ground
deformation (<1 cm) below the Dolomieu crater. Although
this crisis did not lead to an eruption right away, the seismic-
ity remained high for three days until a second crisis began
on 7 December at 02:28. This new crisis was accompanied
by larger ground displacement (reaching about 30 cm) and lo-
calized to the south-southwestern flank of the terminal shield.
Two hours later, at around 04:40, the volcanic tremor was
recorded and localized on the southwestern sector of the En-
clos [OVPF 2021].
A first overflight of the eruptive site was possible at 07:30,
which allowed OVPF to precisely locate the eruptive fissures
just after the sunrise. On return to the observatory and in less
than ten minutes, the coordinates of the eruptive site were
entered and the first lava flow hazard map was produced, ap-
proved, and transmitted to EMZPCOI, by 08:00 (Figure 6A).
The production and delivery of the map was extremely rapid
because all operators were onsite at OVPF and because the
map template had been pre-prepared by setting up all the
GIS layers over the previous days knowing an eruption was
imminent. The eruption was located near the heavily vege-
tated south wall of the Enclos, which also necessitated rapid
notification to EMZPCOI due to the possibility of wildfires.
Discharge rate estimates given by HOTVOLC during the first
hours of the eruption ranged between 10 and 42m3s1, and,
given this large range, it was not possible to discount lava
reaching the vegetated rampart. The first map communi-
cated to EMZPCOI therefore did not contain the lava flow
runout estimates, but only the lava flow path and location of
Presses universitaires de rasbourg
Page 322
5(2): –.
the front position at the time of the flight survey (Figure 6A).
At 10:05 on the same day (7 December), MIROVA revealed
a TADR of 18±6m3s1. These agreed with the HOTVOLC
data for the same time (Figure 6B). A new map was thus pro-
duced and communicated to EMZPCOI showing that, with
this lower value, it was unlikely that the lava flow would reach
the caldera wall (Figure 6C). However, on the map it was clear
where the point of contact would occur if the lava discharge
rate increased.
By the end of 7 December, discharge rate had decreased
to <2m3s1and was accompanied by a gradual decrease in
tremor. The eruption stopped the following day (8 Decem-
ber) at around 07:15 after 26 hours and 35 minutes of effusive
activity. The post-eruption lava flow field outline from aerial
photogrammetry showed that flows extended to 2 km, with
the flow front stopping 500 m short of the caldera wall. A
Sentinel-2 image (acquired at 10:35 RET on 7 December and
processed the next day) also provided the lava flow field area.
The fit between the flow field outline and the model was good
in the proximal zone but diverged in the distal part (Figure 5D).
This discrepancy was due to the presence of a previous lava
flow (of April 2018) that was not present on the DEM. As dis-
cussed later, the DEM needs to be regularly updated at such an
active site, although unfortunately an update could not have
been done in time for this eruption. The total erupted lava
volume was estimated at 0.8 ×106m3from photogramme-
try, which was consistent with the cumulative volumes ob-
tained from discharge rate time series (0.8±0.3 ×106m3and
0.9±0.45 ×106m3, with MIROVA and HOTVOLC respectively).
4.4 Hypothetical eruption outside of the caldera
This last study case focuses on a hypothetical eruption that
would occur outside of the main caldera (i.e. Hors Enclos
eruption) in an inhabited area. This scenario is based on the
expert elicitation exercise proposed by Tadini et al. [2022]. We
chose this case to test our protocol for an eruption that could
potentially impact population in inhabited areas.
In this hypothetical scenario, tremor started at 12:05 RET
on Day 0 on the north-western flank of the volcano. Erup-
tive fissures opened about 15 km north-west of the caldera in
the town of La Plaine des Palmistes. Initial thermal anoma-
lies were detected at the same time, with an estimated TADR
of 120m3s1. Such a high value is consistent with the ac-
tual highest discharge rate to date of 129m3s1recorded by
HOTVOLC on 24 August, 2015, at 15:45 on Piton de la Fournaise.
A vent opening in a populated zone requires particularly
rapid response, and our previous case study in Section 4.3 (7–8
December, 2020 eruption) shows that, if all actors are together
on site, such map can be produced and sent to EMZPCOI at
best within 10 minutes of the communication of the vent lo-
cation. In the case of the expert elicitation exercise proposed
by [Tadini et al. 2022] only the path of steepest descent was
provided to the experts and this was about 6 hours after the
start of the eruption because it included reports from initial
field reconnaissance by OVPF-IPGP staff, confirming vent lo-
cations and flow front locations. Here we show the first hazard
map that we could have been produced following our proto-
col (Figure 7A). From this map it is possible to rapidly identify
the potentially affected infrastructure. Note that simulations
do not capture interaction with buildings that could possibly
slow down or locally divert the lava flow [e.g. Dietterich et al.
2015]. Proximally, we see the flow impacting the north side of
La Plaine des Palmistes (Figure 7B) and distally the predicted
range of potential lava flow paths is confined to a deep ravine,
except where the flow enters the ocean. Such ravines are
characteristic of the geomorphology of La Réunion, and in the
case of an Hors Enclos eruption, could be advantageous be-
cause these features may confine the lava. Note, however, that
in some areas of the Piton de la Fournaise the Digital Terrain
Model (DTM) is not reliable due to dense tropical vegetation.
The use of a DEM calculated from the DTM may therefore
lack vertical precision because the canopy tends to smooth
out the details of the terrain [Stevens et al. 1999;Bigdeli et al.
2018]. The importance of the gullies in channelizing the lava
flow should therefore be treated with care when lava flow
simulations are performed in such dense tropical forest areas.
Such a situation happened in Hawaii during the ahoa lava
flow of 2014, where the lava flowing through a dense forest
entered a preexisting series of ground cracks [Poland 2016], so
the progress of the lava flow could only be tracked following
the gas plume [Patrick et al. 2017].
Because the eruptive fissure is situated at high elevation
and about 15 km from the ocean, the modeled runout dis-
tances indicate that at the discharge rate given in this scenario
(120m3s1) the lava would not reach the ocean but would
require even higher discharge rate (>250m3s1) in order to
enter the ocean (Figure 7A). In such a case, the flow would
cut all towns and roads between La Plaine des Palmistes and
the populated coastal zone around Saint-Benoit. Now, if we
compare with an analogous large Piton de la Fournaise erup-
tion of such high discharge rate, such as that of 2007, lava
tubes could form [Rhéty et al. 2017]. Formation of a tube will
greatly reduce cooling rates by enhancing flow insulation, al-
lowing the flow system to extend much further at any given
discharge rate [Keszthelyi and Self 1998]. Considering that a
tube system would develop, we can simulate a tube condition
by assigning the crust cover fraction in FLOWGO to be 100 %
and a low temperature [Rowland et al. 2005]. Applying this,
we find that the lava could travel from the vent to the ocean
(at a distance of 15 km) at discharge rates as low as 35m3s1
(Figure 7A).
I    
Between 2014 and 2018, the products shared with OVPF com-
prised a map with the DOWNFLOW results (showing the probabil-
ity distribution of lava flow paths) and the results of FLOWGO
(presented as a graph showing the lava velocity versus dis-
tance from the vent to the flow front for the range of MIROVA-
given TADR) [Harris et al. 2017]. The OVPF director could
then inform EMZPCOI of the possible runout of the lava flow,
but no physical product was shared. The primary format of
communication was a telephone exchange between the OVPF
and EMZPCOI, supported by information from visual obser-
vations made during flight surveys and/or field crews. Be-
cause the communication lacked product support, officers in
charge of civil protection actions at the Préfecture took much
Presses universitaires de rasbourg
Page 323
Lava flow hazard assessment at Piton de la Fournaise, La Réunion Chevrel et al. 2022
Figure 6: [A] First map produced on 7 December, 2020 at 08:00 RET. Note that the background is different than for previous
maps and is the BD TOPO IGN®from IGN (see Section 5). This map was published in the OVPF monthly report of December
2020 (after the end of the eruption) [OVPF 2021]. [B] Lava discharge rate and cumulative volume since the start of the eruption
obtained by HOTVOLC and MIROVA. [C] Update of the map produced on 7 December, 2020 at 10:30. [D] Final map showing the
probability of lava flow invasion, runout distances as function of discharge rate and the final lava flow field outline obtained by
photogrammetry. The lava flow outline drawn from the Sentinel-2 image acquired on 7 ecember at 10:30 is also shown. This
map was produced after the end of the eruption.
Figure 7: [A] First map that would have been produced for the hypothetical scenario of a fissure opening in La Plaine des
Palmistes. This map shows the runout distance for a given discharge rate in the case of an open channel system (as usu-
ally first assumed, in blue) and in the case formation of a tube system (in green). [B] Zoom of the map to identify proximal
buildings and roads that will be the first to be at risk.
Presses universitaires de rasbourg
Page 324
5(2): –.
caution by, for example, over-estimating the zone at risk in
their assessments areas to be closed. The belt road to the east
of the Enclos would also be closed when lava flows arrived
in the Grandes Pentes area at a distance of 3 km from the
road. The other difficulty with this communication format
was, without a hazard map, assessment of the situation con-
sisted of simply pointing to the potentially affected area on a
topographic map without precision of the probability distri-
bution of lava flow paths. A first short-term hazard map was
built for the April 2018 eruption [Harris et al. 2019] but the
first time a short-term hazard map was shared with EMPZ-
COI was for the August 2019 eruption. Since then, maps have
been generated and shared for all seven eruptions between
2019 and 2021, with maps being sent to EMZPCOI within the
first minutes to a few hours of eruption onset. These dedicated
hazard maps have allowed EMZPCOI to have a better sense of
the likely flow inundation area, front position and advance of
the lava flow with discharge rate. Most importantly, the maps
have allowed EMZPCOI to plan their response actions with
higher degrees of confidence (e.g. to decide when and where
to close the road, where to deploy wildfire countermeasures,
or even when and where to evacuate populations, including
hikers, and/or to close tourist overlooks that are identified as
potential fire traps).
Throughout the years the maps have been improved after
each eruption upon specific requests from the EMPZCOI. The
first short-term hazard maps only included lava flow path and
runout distances tailored to the vent location and discharge
rate of the specific eruption in hand and this was presented
on a greyscale hill-shaded DEM background (Figure 8A). Now,
and as detailed successively, maps include a more complete
background where remarkable items and vulnerable struc-
tures are visible (Figure 8B). To ensure that the maps are
as useful as possible in aiding EMZPCOI in improved crises
management and that communication within EMZPCOI as
well as with OVPF was fully informed, several exchanges and
meetings were organized between EMZPCOI and OVPF, as
well as LMV-OPGC. These exchanges served as a means of
training for the emergency managers, so that EMZPCOI fully
understood the scientific approach behind the modeling, in-
cluding uncertainties in models. In return, it allowed the sci-
entific group to understand the needs of the emergency man-
agers. This work focused on improving the maps in terms
of graphical representation following constructive comment
from EMZPCOI. As a result, there is a great deal of difference
between the first map produced for the April 2018 eruption
and one of the most recent maps generated to date for the
April 2021 eruption (Figure 8).
Map changes as a result of user needs include:
Color of lava inundation: One of the first exchanges was
related to the colors. The blue shades, as presented in the
first map (Figure 8A) were too confusing as, for EMZCOI,
the colors related to water flooding rather than lava inunda-
tion. Thus, as recommended by Thompson et al. [2015], colors
“commonly associated with volcanoes and hazard, and [which]
reduce the potential for confusion with hydrological hazard
map types” were used (see also Harris et al. [2012]). This in-
volved application of a yellow-to-red sequential color scheme,
with increasing the probability of inundation (Figure 8B).
Boundaries of municipalities: Drawing the boundary of
the two municipalities that shares the Enclos (Sainte-Rose to
the north and Saint-Philippe to the south) was requested so
that the right municipality could be quickly inform if needed.
This is fundamental in cases where the lava flow reaches
the road, as Sainte-Rose is responsible for closure, diversion
and replacement/repair to the north, and Saint-Philippe to the
south. It also helps the two municipalities to prepare for losses.
To fulfil the request, the municipality boundaries where sim-
ply added to the background (Figure 5A).
Land cover: Including the vegetated areas was requested
to visualize where potential wildfires could start. This allows
civil protection to alert the local fire department(s) so they
could intervene in time to stop fires igniting or spreading. This
was also of use to the French National Forestry Office in case
any endemic flora and fauna at risk could be saved. To com-
plies with the request, we therefore added the vegetated areas
as a GIS layer to the greyscale hill-shaded DEM (Figure 5B).
Well-known features: To improve easy geographical lo-
cation of the potential lava flow impact, the map needed to in-
clude well-known features, such as trails, past lava flows, and
well-known scoria cones. The anonymous shaded-relief back-
ground which is suitable for scientific purposes and model
testing, turn out to not be appropriated for outreach and com-
munication purposes. We therefore finally came into the
conclusion that the most meaningful and usable background
for EMZPCOI was the commonly used map of the sector
from the Institut national de l’information geographique et
forestiere (IGN) (Figure 8B). This provides real added value
to the product for response purposes and allows for rapid
decision-making by the authorities.
Specific items on the maps: EMZPCOI wanted the possi-
bility to see specific items on the maps as organized into opera-
tional, strategic, and human resources. Operational resources
include vital centers (rescue centers, hospitals and clinics, mil-
itary camps, town halls, etc.) and strategic sites (airport, wa-
ter reserves, gas stations, etc.). Strategic resources are roads,
some specific sites and establishments such as schools and
retirement homes, as well as electricity, phone, and internet
networks. Human resources comprise homes, schools and
colleges, as well as commercial or industrial areas. EMZP-
COI therefore provided these layers as GIS-compatible format
so that they could be implemented within our map template.
Upon request they are activated to be visible or not.
The maps will certainly continue to be improved in the fu-
ture. Indeed, some requests have been made but have not
yet been fulfilled. For example, a critical tool for civil pro-
tection issues would be to automatically assess the number
and location of people and buildings likely to be affected by
the eruption, and therefore allow improved evacuation plan-
ning and loss preparation. This could potentially be done by
adding the information of the map or providing a separate
map where expected losses could be highlighted. Because
Presses universitaires de rasbourg
Page 325
Lava flow hazard assessment at Piton de la Fournaise, La Réunion Chevrel et al. 2022
Figure 8: [A] The first map produced during the April 2018 eruption showing the lava flow inundation area (simulation with Δh= 0.8
in light blue and Δh= 2.5 in dark blue) and run out distances (yellow stars) for a given TADRs (numbers are in m3s1) along the
LoSD descent (red line). The background is the hill-shaded 2010 LiDAR DEM produced by Institut national de l’information géo-
graphique et forestière (IGN). [B] Map produced during the April 2021 eruption showing the lava flow inundation area (probability
increases from yellow to red; Δh= 2 and N= 10,000) and runout distances (blue arrows) for given discharge rate (numbers are
in m3s1) along the LoSD (red line). The background is the BD TOPO IGN®from IGN.
(since the production of the maps) inhabited areas have not
been threatened, this has not yet been implemented. Another
request from the EMZPCOI is to include an assessment of the
time before the lava flow front would reach critical morpho-
logical features (such as the caldera wall, the Grandes Pentes,
or the coast) or infrastructure (such as the road and inhabited
areas). This relates to the improvement of the the numerical
model itself, rather than just the map; accordingly, this has not
yet been provided.
6 D
6.1 Uncertainties of the modeling and future improvements
Between 2019 and 2021, ten eruption responses at Piton de
la Fournaise used our lava flow response protocol. Because
prior work [Harris et al. 2017;Rhéty et al. 2017;Harris et al.
2019;Chevrel et al. 2021] had gone into initialization and vali-
dation of the base model (DOWNFLOWGO), our confidence in the
numerical model was sufficient to ensure a fast response and
evolution of the map product. However, to ensure a complete
understanding of the delivered map, it is crucial to communi-
cate the uncertainties and the limitations of the model to the
end-user. Here we first present the uncertainties related to the
calibration of the model then we discuss the uncertainties due
to of the model itself and finally the potential model mismatch
due to the topography.
For DOWNFLOW, a key uncertainty is that the accuracy of the
projected lava flow area depends on the calibration of the nu-
merical code that computes the steepest descent paths, while
randomly applying a given vertical perturbation (Δ) at every
pixel of the DEM during each of 𝑁iterations [Favalli 2005].
The code computes a spectrum of possible paths of the flow
from the vent to the edge of the DEM by iterating over a large
number of runs (𝑁), producing a probabilistic estimation of
the lava flow inundation area. Calibrating DOWNFLOW consists
of finding the parameters Δand 𝑁that are able to best fit
the model output to the actual lava flow area. For Piton de
la Fournaise, Chevrel et al. [2021] found that these parameters
may change with the volcano eruptive cycle and the available
DEM resolution. For lava flows emitted between 1998 and
2007, simulations on a DEM produced in 1997 that has a 25 m
resolution are the best for 𝑁> 6,000 and Δ> 4 m. While
for lava flows emplaced between 2010 and 2019 on a DEM
from 2010 with a 10 m resolution, the best fit parameters are
𝑁=10,000 and Δ= 2 m. The best fit parameters changed
because of the increasing DEM resolution and because after
2007, the lava flows are thinner and shorter [Chevrel et al.
2021]. This last calibration ensures that more than 50 % of
the real lava flow area is well-considered by the simulation
[Chevrel et al. 2021]. These best fit parameters are therefore
used in the protocol, but must verified at each eruption to
avoid over- or under-estimation of the lava flow coverage.
Changes in the initialization parameters for FLOWGO corre-
spond to changes in lava composition, temperature, gas con-
tent, and crystallinity, all of which affect the rheology of the
lava and thus also the flow velocity, thickness, and length.
Harris et al. [2019] showed that variation in crystal and bubble
content could affect the lava flow length by 28 %, while uncer-
tainty on surface temperature could affect the lava flow length
by 23 %. Moreover, Thompson and Ramsey [2021] showed
that variable emissivity may have a measurable effect on heat
flux (>30 % decrease), which translates to potentially of longer
flows (~7 % increase). In addition, [Ramsey et al. 2019] ar-
gued that a two-component emissivity model produces flows
that are a maximum of 34 % longer than flows modeled us-
ing a single constant emissivity. Here, we use the same in-
put parameters for FLOWGO as defined by best-fitting to lava
channel dimensional, crystal, and temperature properties for
well-constrained channel-fed flow at Piton de la Fournaise by
Harris [2015], Harris et al. [2017], Rhéty et al. [2017], and Har-
Presses universitaires de rasbourg
Page 326
5(2): –.
ris et al. [2019]. Based on post-event cross-checks of the model
performance and the fact that lava composition has not been
highly variable over the considered period, the model input
parameters have been proven to still be valid (Figure 3D, 5C,
6D). However, should the validity begin to break down we are
able easily to change these parameters; moreover, the best fit
is always appraised as part of model debrief at the end of each
eruption. The modeled lava flow runout length uncertainty is
30 %. This is not yet represented on the map but this is reg-
ularly communicated to the end-user. Future improvement
could include this uncertainty directly on the map.
Another source of modelling inaccuracy originates from the
model itself. DOWNFLOWGO combines a range of probable lava
flow paths and provides the cooling-limited distance the lava
could reach down the LoSD. The model is unable to simu-
late levées, breakouts, overflows, lobes, tubes, bifurcation, etc.
[Tarquini and Favalli 2011]. Also, FLOWGO is a 1-D model and
assumes that the lava flow is channelized and cooling-limited
[Harris and Rowland 2001;Rowland et al. 2005]. Collectively,
this means that complex lava flow fields, volume-limited lava
flows, and long duration flows cannot be modeled accurately
because syn-eruptive topography changes that would influ-
ence the flow are not captured on the DEM. Nonetheless, at
Piton de la Fournaise, effusive events are of relatively short
duration. Additionally, this numerical model is computation-
ally inexpensive, rapid, and can be operated on any operat-
ing system. Therefore, DOWNFLOWGO is, so far, the best option
to rapidly build short-term hazard maps that show the most
probable inundation area and the runout length for the given
discharge rate at the beginning of an eruption when flow is in
a well-defined channel. In the event of a long-duration flow,
update of the DOWNFLOWGO modeling can be operated at the
site of breakouts [Harris et al. 2019] and insulation condition
changes from open channel to tubes can also be modeled (see
Section 4.4). Future options to better incorporate syn-eruptive
lava flow morphology and flow front propagation could in-
clude use of a physics-based model, similar to MAGFLOW [Del
Negro et al. 2013], which can spatially and temporally repro-
duce lava flow propagation [Cappello et al. 2022]. At Piton de
la Fournaise, the Rheolef model has been used [Bernabeu et
al. 2016], and VOLCFLOW [Kelfoun and Vargas 2016] has been
tested on previous eruptions [Lemaire 2022]. Further work is
needed to validate the use of such models during an eruption
crisis and to ensure that they can be implemented in a timely
Errors in the modeled lava flow path may also come from
error on the terrain topography [Tarquini and Favalli 2010]. In
the Enclos, the high frequency of eruptive activity results in
frequent change of the topography and these changes cannot
always be added to the available DEM and therefore will not
be considered by the model. For example, in this study we
showed that the simulation of the lava flow emitted during the
eruption of February 2020 did not compare well with the distal
part of the flow. Here, the lava flow divided into two branches,
while the modeled steepest descent line ran between the two
branches (Figure 6). This discrepancy is due to the April 2018
lava flow that was not present in the DEM that we used. This
shows how important it is to update the DEM after every erup-
tion. At Piton de la Fournaise, the frequent bad weather con-
ditions on the volcano flanks and difficult terrain prevent the
use of Unmanned Aerial Vehicles (UAV) for rapidly producing
DEMs. Although this has been done in some cases and has
been being completed by stereophotogrammetry from aircraft
surveys [Derrien et al. 2015;2018;Derrien 2019;Derrien et al.
2020], update of the DEM has not been systematic. DEMs are
also produced by space agency services from optical sensors
such as ASTER GDEM or radar sensors such as SRTM and
tanDEM-X [Zink et al. 2014]. Additionally, at Piton de la Four-
naise, with the help of the Centre National d’Etudes Spatiales
(Kalidéos program), tri-stereo Pleiades have been acquired ev-
ery year since 2013. These images are acquired in triplets and
can be used to generate DEMs of resolutions better than 1 me-
ter [Bagnardi et al. 2016]. However, these DEMs need to be
carefully “cleaned” and post-processed to avoid artifacts that
can greatly affect the modeling. Updating the DEMs is there-
fore done as often as possible, but it cannot be systematic.
6.2 Satellite sensor–derived TADR
The assessment of runout within our protocol depends on the
reliability of the satellite-derived estimates of TADR, which
are affected by uncertainties in the conversion between heat
flux and volume flux as caused by thermal, rheological, and
terrain factors [Harris and Baloga 2009]. Other uncertainties
result from errors in the measurement of the thermal flux,
mainly due to the presence of clouds that can attenuate the
thermal signal detected by the satellite [Coppola et al. 2013],
as well as pixel overlap and the point spread function at high
scan angles [Harris et al. 1997;Harris 2013]. At Piton de la
Fournaise, weather conditions are often cloudy, which there-
fore reduces the usability of satellite images to extract accurate
TADR [Peltier et al. 2020;Thivet et al. 2020].
The method of discharge rate estimation by MIROVA differs
from the approach of HOTVOLC.MIROVA system uses Low-Earth
Orbiting platforms (Terra/Aqua) with pixels that are nomi-
nally 1km2and requires a single image to provide a lava dis-
charge rate. The error associated with cloud-free satellite-
derived time-averaged lava discharge rates with MIROVA is
about ±50 % which can be reduced to ±35 % after calibra-
tion on previous eruptions [Harris et al. 2019]. These TADRs
represents an estimate of the volume of lava emitted in the
hours preceding image acquisition, typically 6–18 hours in the
case of data acquired in the mid-infrared [Coppola et al. 2010].
Consequently, the discharge rate obtained by MIROVA does not
represent an instantaneous effusion rate value, but is a lava
flux time-averaged over several hours [cf. Harris et al. 2007].
This temporal resolution (one image every 6 hours) does not
allow for detailed evaluation of the lava discharge rate vari-
ation for short (a couple of days) eruptions as can happen at
Piton de la Fournaise.
With HOTVOLC, discharge rate is estimated using lava vol-
umes acquired every 15 minutes, as calculated from infrared
data provided by the MSG-SEVIRI (Spinning Enhanced Visi-
ble and Infrared Imager) sensor [Labazuy et al. 2012;Gouhier
et al. 2016;Gouhier 2022]. Using this geostationary sensor
allows detailed time-series to be generated over short periods
but the main limitation is the low spatial resolution, where pix-
Presses universitaires de rasbourg
Page 327
Lava flow hazard assessment at Piton de la Fournaise, La Réunion Chevrel et al. 2022
els are ~24.5 km2at the image location of La Réunion and this
prevents the detection of low amplitude thermal anomalies.
HOTVOLC uses the pixel-integrated thermal anomaly measured
at a given time that is the balance of contributions related to
hot lava material newly emplaced and the cooling lava mate-
rial previously emplaced [Gouhier et al. 2016;Dumont et al.
2021]. This method has uncertainties of the order of ±50 %
and topographic shadowing [Dehn et al. 2002], as well as prob-
lems associated with radiance smearing due to the detector’s
point-spread function [Mannstein and Gesell 1991] and pixel
overlap [Cahoon et al. 1992], are particularly troublesome at
high scan angles like at Piton de la Fournaise [Harris 2013].
In our protocol, we use the two systems together for cross-
checking and on-the-fly validation. This strategy is also
adopted by LAV@HAZARD [Vicari et al. 2011;Ganci et al. 2012].
Future improvements include the development of ad-hoc
satellite missions with a volcanological focus that would con-
stitute a turning point in monitoring these types of events, al-
lowing us to get closer to instantaneous values [Ramsey et al.
2021]. In the meantime it is possible to increase the temporal
frequency and spatial resolution of thermal images through
multi-platform data analysis [Harris et al. 1998]. For exam-
ple, the implementation of VIIRS (Visible Infrared Imaging
Radiometer Suite) data within the MIROVA system will dou-
ble the number of daily images useful for an accurate esti-
mate of TADR [Campus et al. 2022]. Similarly, the analysis of
high-spatial-resolution images, such as those from Sentinel 2
and Landsat 8 and 9 will increase the probability of obtaining
maps useful for locating the vents and active lava fronts.
6.3 Timeline to map production
Running DOWNFLOWGO is a very rapid process. Once the opera-
tor of the model has the vent coordinates, the map production
and delivery can thus be executed in about 10 minutes, as it
was the case for the December 2020 eruption. In practice this
time is often longer, because of readjustment of some param-
eters, the need for scaling, or addition of particular features
if needed or requested by the director of OVPF or EMZP-
COI. With the map template ready, the key factor to ensure
an efficient hazard assessment is therefore the availability and
accuracy on the vent location.
The time constraint for delivery of the map within the cur-
rent protocol therefore mostly depends on the time between
the start of the eruption and the confirmation of vent loca-
tion. This may take from several minutes to a few hours, and
depends on:
1) The location of the main vent or fissures, which may be
difficult to confirm if located far from any viewpoints, roads
or trails, or out of view of the monitoring camera network;
2) The weather conditions, where if it is too cloudy or
foggy, the eruptive fissures may not be visible from the various
viewpoints and a flight survey cannot be conducted;
3) The availability of the OVPF staff for carrying out field
surveillance. The staff involves a small team [Harris et al. 2017;
Peltier et al. 2022], and staffing problems can and do arise if
an eruption occurs in the middle of the night, at a weekend,
during vacations or periods of illness or lockdown [Peltier et
al. 2020]; and
4) The helicopter used for airborne surveillance operations
is operated by the Gendarmerie and also serves for mountain
rescue services around the island. Thus, the aircraft may be
unavailable in a timely fashion if already out on, or called for,
a rescue mission which is prioritized over surveillance duties.
These time constraints may be significantly reduced if the
eruptive site is, instead, located using the tremor maps that
are produced every 15 minutes by the OVPF and which are
available online [Beauducel et al. 2020]. Inside the Enclos,
the tremor map provides a relatively accurate vent location
because this is where the monitoring network is the densest
(there are 20 seismic stations within the Enclos within an area
of 100 km2,Figure 1). The center of the tremor location is
obtained by a triangulation method, which locates the source
of the tremor, interpreted as the main fissure, with an error
of hundreds of meters [Battaglia 2003]. Eruptions outside of
the Enclos would be more difficult to precisely locate with the
tremor map because of the scattered and less dense station
distribution. Stations should be placed on the western flank
of the volcano to improve triangulation in this area. Vent loca-
tion may also be improved with the increase of satellite sensor
spatial resolution that would enhance our ability to locate the
vent and hence to quickly respond to an emergency crisis.
6.4 Crisis management for future eruptions at Piton de la
Since the protocol has been in place and operational, the erup-
tions of Piton de la Fournaise have been characterized as “typi-
cal” with relatively low effusion rates that reach a maximum of
30 to 40m3s1. All eruptions have occurred inside the Enclos
where there is no population or infrastructure, aside from the
belt road, hiking trials, and the OVPF monitoring network. For
these reasons, the present protocol has been employed with-
out need for adjustment for eruption location and source terms
(see Section 6.2). However, as shown by the hypothetical sce-
nario considered in Section 4.4, if a fissure opens outside of the
Enclos to feed a high-effusion-rate flow (>100m3s1) into an
inhabited area, the response protocol will become extremely
important as a response-support tool and will need careful
tracking to assess its performance. The time between the on-
set of the eruption and delivery of product will be also crucial
in such an event, but we run into the circular problem as to
how far we can trust runs based on “typical” source terms for
an “atypical” scenario. However, we stress that the most im-
portant parameter is the location of the eruptive fissure and its
timely provision because it quickly provides an initial approx-
imation on the hazard. This first-approximation need may be
pre-empted, to an extent, by precursor analyses such as lo-
cations of pre-eruptive seismic activity, deformation, and soil
degassing anomalies (CO2). Further improvements may also
rely on simulations of, and expert elicitations for, hypothetical
scenarios [e.g. Tadini et al. 2022].
Presses universitaires de rasbourg
Page 328