Michele Prestifilippo UPNV: Unità di Progetto Nubi Vulcaniche Mauro Coltelli, Michele Prestifilippo, Simona Scollo, Gaetano Spata
20 July 22 July 23 July 2001 Etna Eruption 24 July Scollo et all. 2007
58 days of continuous fallout Merged Pulsating 2002 Etna Eruption 58 days of continuous fallout Sustained Diluite Andronico et all. in press JGR
2002 Etna Eruption – Fallout problems Catania Via Etnea Catania Vincenzo Bellini International Airport
Main typologies of explosive eruptions at Etna volcano Short-lasting eruptions Duration from minutes to a few hours Magnitude from violent strombolian to subplinian (from 4 to 15 km high eruptive plumes) One plinian eruption in historical time (26 km high plume) Occurrence more then 150 in the last 25 years Frequency up to several tens in a few months Long-lasting eruptions Duration from days to a few months Magnitude violent strombolian (low-troposphere plumes) Occurrence only two in the last century but at least 8 occurred in the last four century
Bonadonna et Philips 2003
Tephra Dispersal Models Three dimensional Models Lagrangian Models PUFF (Searcy et al. 1998) HYSPLIT (Draxler and Taylor 1982) VOLCALPUFF (Barsotti et al. 2008) Three dimensional Models Numerically airborne particles are hypothetical : Advected Dispersed Gravitationally settled through the atmosphere.
Tephra Dispersal Models Gaussian Models HAZMAP (Macedonio et al. 2005) TEPHRA (Bonadonna et al. 2005) Advection-Diffusion Models Mass conservation equation Constant Diffusion Coefficients Vertical wind and diffusion negligible Isotropic atmospheric diffusion in the plane (x,y) Total mass ejected at t=0
Tephra Dispersal Models Three dimensional Models Eulerian Models FALL3D (Costa et al. 2006) Three dimensional Models Buoyant plume theory Wind data from local models Time dependent
Volcano Input Parameters Meteorological Input Parameters Wind distribution Temperature Pression Humidity Volcano Input Parameters Erupted Mass Topography Column Height Eruption Column Model Total Grain-size Distribution Particle Shape Factor
Meteorological Input Parameters Meteorological data of ARPA/CINECA EUMETSAT MSG
Volcano Input Parameters Andronico et all. in press JGR
There are the following problems Acquire meteorological data. (and validate it!) Define in the most precision way the volcano input parameters values. (and validate it!) Make a prediction of volcanic ash distribution on ground and on air. (and validate it!)
UPNV activities
Why is the parallelization required? Ash forecasting is made everyday using 4 different tephra dispersal models and simulating 2 possible eruptive scenarios: Eruptive scenarious Eruption Time Mass flow[kg/s] Column height [km] Grain-size [F] Sigma Test case Weak Plumes continua 5.E+04 3.5 0.5 1.5 2002-03 Eruption Strong Plumes 5 min 1.E+06 8 -0.5 22 July 1998 Eruption The output produced is relative to 2 days with a time resolution of 3 hours and spatial resolution of 171x171x10 km. In this way we obtain the result after the days of interest. All the simulation codes are parallelized and scheduled hierarchically by a decentralized-supervisor scheduler allocated in the “Server UPNV”. The produced outputs are elaborated by the central server and are published in the web and sent to the Civil Protection Department (DPC). Everyday the INGV receives the meteorological data at 06:00 GMT and the processing is complete at 10:30 GMT (4.5h << 64h). The DPC receives the elaborated outputs at 07:30 GMT.
The scheduler Supervised cluster
Volcanic ash forecasting
Volcanic Ash forecasting
Volcanic Ash forecasting
Volcanic Ash forecasting
The PUFF model The output of the lagrangian PUFF modelis a collection of particles so the output is only qualitative but one simulation require few seconds. We have worked to make a quantitative PUFF. The new simulations require about 4 millions of particles and the new elaboration time is about 12 minutes so we have also parallelized the PUFF model.
Results Reliable warning of 24/11/2006
Results Reliable warning of 4/09/2007
Why is the validation/calibration required? The real eruption can be notably different from the scenarios assumed. The mismatch between field and simulated data can be due: uncertainty of input parameters (calibration) choice of the tephra dispersal model (validation) Calibration and validation of models require data measurement and hundred or thousand simulations to verify the matching between field and simulated data, so the parallelization code is necessary to obtain a result within an acceptable time. Computed data (kg/m2) Field data (kg/m2)
Details about the hardware structure of the system will be explained in the following presentation. Thanks