Near-real-time forecasting of lava flow hazards during the 12–13 January 2011 Etna eruption



[1] Forecasting the lava flow invasion hazard in near-real time is a primary challenge for volcano monitoring systems. The paroxysmal episode at Mount Etna on 12–13 January 2011 produced in ∼4 hours lava fountains and fast-moving lava flows 4.3 km long. We produced timely predictions of the areas likely to be inundated by lava flows while the eruption was still ongoing. We employed infrared satellite data (MODIS, AVHRR, SEVIRI) to estimate in near-real-time lava eruption rates (peak value of 60 m3 s−1). These time-varying discharge rates were then used to drive MAGFLOW simulations to chart the spread of lava as a function of time. Based on a classification on durations and lava volumes of ∼130 paroxysms at Etna in the past 13 years, and on lava flow path simulations of expected eruptions, we constructed a lava flow invasion hazard map for summit eruptions, providing a rapid response to the impending hazard. This allowed key at-risk areas to be rapidly and appropriately identified.

1. Introduction

[2] In the night between 12 and 13 January 2011 the pit crater on the low eastern flank of the South-East Crater (SEC) cone, in the summit area of the Mount Etna (Sicily, South Italy), was the site of a paroxysmal episode that produced high lava fountains, a lava overflow toward the Valle del Bove (VdB) depression, and a tephra-rich eruption column several kilometers high (Figure 1). This was a typical paroxysmal episode, similar to dozens of events from the summit craters of Etna over the past decades [Behncke et al., 2006].

Figure 1.

(a) Sketch map of eruptive products of the 12–13 January 2011 paroxysm at (b) Mt Etna. Contours are in meters. (c) Photograph shows initial phase of the paroxysm; courtesy of Miryam Grasso.

[3] The growing number of these abrupt and violent events, which may produce lava flows capable of flowing over long distances enough to invade vulnerable areas on the flanks of Etna [Behncke et al., 2005], requires effective tools for timely predictions of lava flow hazards [Wright et al., 2008; Crisci et al., 2010] to help local authorities in making the necessary decisions during a volcanic eruption. This challenge has inspired the INGV-CT to develop an integrated system that uses near-real-time infrared satellite data to drive numerical simulations of lava flow paths [Vicari et al., 2009]. This system represents the central part of an extensive methodology that allows us to draw the probable evolution of lava flow-fields and hazard mapping of different eruptive scenarios.

[4] We applied our methodology to the forecasting of lava flow hazards during the powerful paroxysm at the SEC on 12–13 January 2011. This event was exceptionally well documented by a variety of monitoring instruments of the INGV-CT multi-parameter network, field observations, and spaceborne thermal infrared measurements. Satellite data acquired by MODIS, SEVIRI and AVHRR sensors were promptly elaborated to locate thermal anomalies related to volcanic features and to estimate lava eruption rates using the measurements of the total thermal flux of active surface flows [Vicari et al., 2009]. These data are needed to enable advanced flow simulation models to be used in an operational use, making predictions in time for response action to be taken. Therefore, we used satellite-derived discharge rate estimates, calculated in near-real-time, to simulate the temporal evolution of lava emplacement using the MAGFLOW model [Del Negro et al., 2008]. The capability of the MAGFLOW code to reproduce the 12–13 January 2011 lava flow was tested by comparison with the actual, mapped lava flow-field. In this way, our MAGFLOW model provided a reasonable starting point from which to analyze flow emplacement processes, as well as assess the hazard posed by hypothetical and ongoing eruptions. The MAGFLOW code was then used, together with knowledge of the eruptive history of the SEC, for the production of a lava flow invasion hazard map of the SEC area for summit eruptions, providing a rapid response to the impending hazard.

2. The Eruptions of the South-East Crater

[5] The South-East Crater (SEC) formed in the spring of 1971 at the southeastern base of the central cone of Etna, and has since then gone through numerous eruptive periods often characterized by sequences of brief but high-eruption-rate paroxysmal episodes that generate tall lava fountains and eruption columns, and high-rate outpouring of lava [Calvari et al., 1994; Behncke et al., 2006]. During the past 15 years, activity of the SEC has predominantly consisted of sequences of paroxysmal episodes, amounting to 105 events between 1998 and 2001 [Behncke et al., 2006] and several more during the intermittent activity of 2006–2008 [Neri et al., 2006; Andronico et al., 2008; Langer et al., 2011]. Mass eruption rates during the lava fountaining episodes in the period 1998–2001 never exceeded 170 m3 s−1; lava flows advanced a maximum of 3 km, and most of them lasted less than 1 hour in their culminating phases. All activity since mid-2007 has occurred from a new vent (called pit crater) that opened on the lower eastern flank of the SEC cone in late-May 2007. The shift of the main vent location brought about a distinct change in the character of the eruptive activity. Between September 2007 and February 2011, the new vent produced five paroxysms with tall lava fountains and voluminous lava flows; durations varied from 3.5 to 14 hours, and lava flow lengths ranged from 3.3 to 6.2 km. Mass eruption rates during the culmination of the 10 May 2008 paroxysm probably exceeded 300 m3 s−1, making this one of the highest rates recorded during an Etnean eruption.

[6] In order to characterize the eruptive paroxysms of the SEC, we used the main quantitative volcanological parameters (i.e., duration and lava volume) of ∼130 episodes in the past 13 years [Behncke et al., 2006; Andronico et al., 2008]. We established short, medium and long eruption durations, that is, ≤4, 4–8 and >8 hours, respectively. Then we defined three different ranges of emitted lava volume: ≤1, 1–2 and >2 million cubic meters. Combining these values, we obtained 9 possible eruptive classes representing the lava volume in relation to the eruption duration. However, only six classes are populated (Table 1). These parameters are used to simulate the expansion of the lava flows and the construction of a hazard map of the SEC area for summit eruptions.

Table 1. Classification on Durations and Lava Volumes of Paroxysmal Eruptive Episodes at SEC Since 1998a
Lava Volume (×106 m3)Duration (hours)
  • a

    Mean Output Rate (MOR) is calculated as the final volume of erupted lava divided by total eruption duration.

[0–1]Class IClass IIClass III
65% (MOR ≈ 68 m3 s−1)5% (MOR ≈ 35 m3 s−1)23% (MOR ≈ 18 m3 s−1)
[1–2]Class IV0%Class V
3% (MOR ≈ 157 m3 s−1) 3% (MOR ≈ 35 m3 s−1)
[>2]0%Class VI0%
 1% (MOR ≈ 166 m3 s−1) 

3. The 12–13 January 2011 Paroxysm

[7] After the 10 May 2008 paroxysm, the SEC reactivated in late-December 2010 with small Strombolian bursts and continued at fluctuating levels into early January 2011. A gradual increase in the intensity of Strombolian explosions started on 11 January, and on the 12th the activity became progressively stronger. Lava began to flow over the rim of the active vent around 20:10 (GMT = local time − 1), initially feeding a series of short, sluggish flows before the effusion rate increased and a new flow advanced about 1 km toward the western rim of the VdB.

[8] At about 21:50, the Strombolian activity rapidly passed into sustained lava fountaining, accompanied by the rise of a tephra-rich eruption column and a manifold increase in the lava effusion rate. A voluminous surge of lava exited the vent at 21:51 descending, in several branches, the western slope of the VdB. Lava fountaining to heights of 300–500 m above the vent continued until about 23:00, after which the fountain height decreased. At 23:25, all activity was reduced to a single, narrow jet, and at 23:53 lava fountaining stopped, followed by a number of isolated explosions and minor ash emissions until the early afternoon of the next day.

[9] The main lava flow field reached a total length of 4.3 km (Figure 1). Although sluggish flow continued through the forenoon of 13 January, feeding of the lava had ceased some minutes before the end of lava fountaining, so that the main part of the lava had been emitted in about 3.5–4.0 hours. Field mapping revealed that the lava had covered an area of little more than 1 km2 with a thickness of 1.3 to 2.3 m, giving a total bulk volume of 1.3–2.3 × 106 m3 [Behncke et al., 2011], corresponding to 1.1–1.8 × 106 m3 Dense Rock Equivalent (DRE). On the basis of erupted volume and duration of the event, the paroxysm of 12–13 January 2011 belongs to class IV (Table 1).

4. Effusion Rate Estimations by Infrared Satellite Data

[10] To monitor Etna's thermal activity, we used the multiplatform system HOTSAT [Ganci et al., 2011] that is capable of managing multispectral data from different sensors aboard meteorological satellites as NOAA-AVHRR, EOS-MODIS and MSG-SEVIRI. HOTSAT takes advantage from integrating information acquired by sensors with different spatial, spectral and temporal resolution: AVHRR and MODIS sensors are aboard polar satellites and provide data at least 4 times a day with a spatial resolution of about 1 km, while SEVIRI is on geostationary satellite providing images every 15 minutes with a spatial resolution of about 3 km. We implemented several algorithms for hot spot detection (thermal anomalies that possibly relate to dynamic volcanic processes) and heat flux computation at active volcanoes [Ganci et al., 2009]. We converted the total thermal flux measured from thermal infrared satellite images to time average discharge rate (TADR) following Harris et al. [1998] and Wright et al. [2001]. The conversion from heat flux to volume flux depends on lava parameters such as density, specific heat capacity, eruption temperature, solidus temperature, latent heat of crystallization, and volume percent of crystals that form during cooling of the lava [Harris et al., 2007]. To date, laboratory measurements have not been carried out on the 12–13 January lavas. Therefore, we decided to infer the lava parameters by comparison with rocks sampled in recent Etna lava flows. Given that we cannot fix a single value to characterize each lava parameter, the most reasonable solution is to use a range of possible values. In particular, we defined a range of variability for each parameter adopting the extreme values (Figure 2) found by Harris et al. [2000, 2007], which proved to be reasonable in calibrating satellite thermal data technique for Etna lavas.

Figure 2.

(top) TADR ranges and cumulative lava volumes estimated from AVHRR, MODIS and SEVIRI data. Lava parameter values used to convert satellite thermal data to TADR are given by Harris et al. [2000, 2007]: dense rock density (ρ = 2600 kg m−3); specific heat capacity (Cp = 1150 J kg−1 K−1); vesicularity (ϖ = 10 ÷ 34%); lava cooling between eruption temperature and temperature at which flow is no longer possible (ΔT = 100 ÷ 200 K); crystallization in cooling through ΔT (ΔΦ = 0.3 ÷ 0.54 fraction); and latent heat of crystallization (CL = 3.5 × 105 J kg−1). (bottom) SEVIRI scenes of Sicily island, South Italy, between 21:30 and 23:15 on 12 January 2011. Inset images show the Etna volcano. Dark pixels are ash clouds and bright pixels are hot spots associated with the eruptive paroxysm at SEC.

[11] HOTSAT was able to monitor in near-real-time the thermal behavior of Etna during the 12–13 January 2011 episode. The first hotspot in the SEC area was revealed at 18:56 on 11 January by AVHRR, followed, five hours later, by MODIS at 00:40 on 12 January, while SEVIRI detected the first thermal anomaly at 17:30 on 12 January. After this, an almost continuous thermal activity was recorded from 20:15 on 12 January, with a peak of about 18 GW at 23:45. The formation of the ash cloud from 21:45 to 23:15 led to a sharp decrease of the radiative power (see SEVIRI images in Figure 2). A continuous decrease in thermal activity was observed at 01:15 on 13 January, with the last hotspot detected by SEVIRI at 06:30 on 14 January. Lower intensity thermal anomalies lasted until 17:55 (detected by AVHRR) and 20:55 (detected by MODIS) on the same day.

[12] Lava discharge rate estimates from satellite-measured heat flux were calculated up to four times per hour (Figure 2). We obtained minimum and maximum estimates for TADR by taking into account in the variability range of each lava parameter the largest and smallest values, respectively. Peak estimates for TADR were reached around midnight on 12 January, ranging between 20 and 60 m3 s−1. By integrating separately minimum and maximum estimates of TADR, we computed two cumulative curves of erupted lava (Figure 2). Over the entire period of thermal emission, lava volumes are estimated to be between 0.4 and 1.2 × 106 m3. However these computations are affected by uncertainties and assumptions such as the ash cloud attenuation and/or the inability to distinguish between lava draining channels and cooling phenomena. Assuming that the lava stopped flowing around 6:00 on 13 January (at the end of the 10-hour period of more intense thermal activity), the satellite-derived final volume is estimated in the range 0.3–0.9 × 106 m3.

[13] As a matter of fact, the uncertainty in satellite derived effusion rate estimates is quite large, up to about 50%, but it is comparable to the error in field-based effusion rate measurements [Calvari et al., 2003; Harris and Neri, 2002; Sutton et al., 2003; Harris et al., 2007]. The main uncertainties arise from the lack of constraint on the lava parameters used to convert thermal flux in effusion rate. Moreover, the presence of ash strongly affected the lava discharge rate estimates between 21:45 and 23:15 GMT, when the maximum intensity of eruptive activity and peak rates of lava emission occurred. This necessarily led to an underestimation of the satellite-derived final volume, and to a difference in the timing of simulated lava flow emplacement.

5. Scenario Forecasting Using MAGFLOW Model

[14] To simulate lava flow emplacement in space and time, we employed the MAGFLOW model [Vicari et al., 2007; Del Negro et al., 2008], previously used to reproduce the lava flow paths during the 2004, 2006 and 2008 Etna eruptions [Del Negro et al., 2008; Herault et al., 2009; Bonaccorso et al., 2011]. More recently, it was applied to more sophisticated issues including the impact evaluation of protective barriers placement on lava flow diversion [Scifoni et al., 2010] and the production of a hazard map for lava flow invasion on Etna [Cappello et al., 2010].

[15] MAGFLOW is a Cellular Automaton (CA) model based on physical modeling of lava flows, as its evolution function is derived from a steady-state solution of the Navier-Stokes equation for Bingham fluids, coupled with a simplified heat transfer model [Vicari et al., 2007]. The implementation of physical equations in MAGFLOW allows the definition of features essential for hazard purposes such as the time of propagation of lava flows and the maximum run-out distance. To produce a dynamic picture of probable lava flow paths, MAGFLOW requires constraint of many parameters. However, for a given composition, the instantaneous lava flow output by the vent is the principal parameter controlling final flow dimensions, and in order to obtain more reliable simulations for the same total lava volume it is better to have continuous monitoring of the effusion rates, even if with moderate errors, rather than sparse but accurate measurements (G. Bilotta et al., Sensitivity analysis of the MAGFLOW Cellular Automaton model, submitted to Environmental Modelling and Software, 2011). As such, simulations that take into account the way in which effusion rate changes during an eruption, and how this influences the spread of lava as a function of time, are of special interest, particularly as effusion rates can be highly variable [Harris and Rowland, 2009].

[16] To this end, we used satellite-derived time-varying discharge rates to drive the MAGFLOW simulations of lava flow paths, for predicting the area most likely to be inundated with lava during the 12–13 January 2011 paroxysm. Such forecasting was calculated on the afternoon of 13 January [Cappello et al., 2011], using the maximum estimates of TADR (Figure 2). A 5 m—resolution Digital Elevation Model (DEM) of the Etna and updated to 2005 was the basis for the numerical simulations. The eruptive vent was centered on the pit crater at the base of the SEC. The simulated lava flow path after 12 hours is shown in Figure 3b, where the actual flow is also reported. The lava flow field has extended downslope to an elevation of 1,700 m with a maximum thickness of ∼7 m. Even if a time of simulation of 12 hours has been adopted, the maximum length of simulated lava flow is already reached after ∼6 hours. In the next 6 hours, only an increase in lava thickness is observed. Comparing this timing with that observed in the field, a mismatching is found, since the real lava flow was mostly emplaced in about 4 hours. Moreover, even if the simulated flow field well approximates the observed flow field, some small branches of the flow are not reproduced. These space and time differences were probably due to the DEM that is updated to 2005, and excludes the lavas emitted since 2005, which all modified the morphology of the VdB. Also the missing estimations of TADR due to ash cloud, during the most intense phase of lava fountains and highest lava output rates, could have introduced a discrepancy (a) in the maximum length of the simulated lava flow paths, (b) in the timing of the lava flow emplacement (especially considering the time in which the maximum length is reached), and (c) in the lateral expansion of the upper part of simulated flow. However, the zones of highest probability of lava coverage in the simulation fit well with observed lava flow field.

Figure 3.

(a) Lava flow invasion hazard map associated with the 12–13 January 2011 paroxysm. The black contour represents the actual lava flow emitted in this event. A comparison between the lava flow simulated by MAGFLOW (vent UTM coordinates: 500285 E, 4177748 N) and (b) the real lava flow. (c) Inset diagram shows the bell-shaped effusion rates retrieved for the six different typologies of summit eruptions (see Table 1).

6. Lava Flow Hazard Assessment

[17] To evaluate the likely lava flow paths from various eruptive scenarios, whose real-time prediction may lead to civil defense decision and impact mitigation action, we produced a lava flow invasion hazard map for the SEC area (Figure 3a). We here adopt the probabilistic lava flow path modeling approach [Cappello et al., 2010], which has been previously used to investigate the lava flow hazard at Etna. Based on a definition of the vent opening probability, on the characterization of the past flank eruptions, and on lava flow path simulations, we constructed a probabilistic mapping of the potential for lava flow inundation on the flanks of entire Etna volcano.

[18] To assess the hazard associated with the 12–13 January 2011 paroxysm, the probabilistic approach, originally designed for flank eruptions, was adapted to summit eruptions. We defined a 400 × 400 m grid centered on the active pit crater with 100 m spacing, where the 25 grid points represent possible eruptive vents. We assigned the same activation probability (Pa) to all 25 hypothetical vents. The typologies of expected eruptions to be simulated were derived by the six populated classes shown in Table 1. For each eruptive class we estimated an event probability (Pe), based on the total number of observed events, that represents the probability of occurrence of the considered class for every grid point. Moreover, we retrieved the effusion rate trends associated with each class, considering a bell-shaped signal [Vicari et al., 2007], where the maximum peak occurs at 1/4 of the entire eruption duration (see Figure 3c). For every vent of the grid, we executed six simulations by MAGFLOW, each characterized by its own effusion rate and duration. Finally, the lava flow invasion probability has been computed for each point (x,y) of the area covered by flow simulations taking into account the information on lava flows overlapping, activation probability of vents vi, and event probability of eruptive classes cj, as follows:

equation image

[19] Results are shown in Figure 3a, where the probability of lava flow invasion is shown for the areas potentially affected by lava flows emitted from vents in the SEC zone. The inundated area measures about 4.3 km2, and the maximum distance reached by a lava flow path is almost 9.5 km. The threatened zone is almost completely contained within the huge uninhabited depression of the VdB. It is worth to note that the observed lava flow paths of the 12–13 January 2011 eruption mostly (about 66%) match the areas in which the highest values of invasion probability are reached.

7. Conclusions

[20] Near-real-time forecasting of lava flow hazards on Etna is difficult due to uncertainties concerning the duration of an incipient eruption, the estimate of the effusion rate, and the location of future vents. The methodology presented in this paper to evaluate the hazard of lava flows outpoured during the 12–13 January paroxysm of Etna tackles these aspects from different directions. By using satellite thermal data with low spatial and high temporal resolution, we obtained a system of early warning combined with a preliminary estimation of the lava discharge rates. These satellite-derived discharge rate estimates were used in the MAGFLOW model, allowing us to effectively simulate the time of advancing and the maximum length of the lava flow. In this way, an eruptive scenario was provided promptly enough for a response to be effective. By simulating the inundation areas for diverse typologies of possible future eruptions at the SEC, we produced a hazard map that is able to consider possible abrupt changes of eruptive conditions, furnishing the probable paths of lava flows and the associated inundation probability. Field observations permit to validate our methodology in an operational context. The time of intervention is the key point of the system presented, even if large uncertainties are present in the preliminary data analyzed.

[21] In spite of these limits, the simulated lava flow (Figure 3b), and the most likely invaded areas predicted in the hazard assessment (Figure 3a) show a good match with the lava flow actually emplaced on 12–13 January 2011, confirming that our approach can be applied with confidence. We also demonstrate how the MAGFLOW model can be driven by satellite-derived lava effusion rates. As these data can be obtained in near-real-time, such an approach allows flow simulations to be updated in response to changing eruption conditions. Moreover, it was found that the construction of a lava flow invasion hazard map allows to rapidly assess the threat posed by an eruption from a given area and thus represents a support tool for decision makers. In this case, the results obtained from the hazard map predict that summit eruptions like at the SEC pose no threat to the local population, with the added value that all the procedures developed showed a very short time of intervention (from few minutes to hours), representing a critical point during an emergency.


[22] We are grateful to EUMETSAT for SEVIRI data, to NASA for MODIS data, and to NOAA for AVHRR data. The authors thank one anonymous reviewer and V. Acocella for their helpful and constructive comments. This study was performed with the financial support from the V3-LAVA project (INGV-DPC 2007-2009 contract).

[23] The Editor thanks Valerio Acocella and an anonymous reviewer for their assistance in evaluating this paper.