INGV, BSC, IMO
Tephra accumulation at the ground from volcanic eruptions; airborne ash concentration from volcanic eruptions
FALL3D
Services:
- PVHA_WF_lt for the long-term hazard assessment
- PVHA_WF_st for the short-term hazard assessment
| Main objective / mission | The objective of this PD is to provide innovative hazard maps with uncertainty, and overcoming the current limits of PVHA imposed so far by the high computational cost required to adequately simulate complex volcanic phenomena (such as tephra dispersal) while fully exploring the natural variability associated to such volcanic phenomena, on a country-size domain (~thousands of km) at a high resolution (one to few km). We focus on the case of tephra hazard assessment, both in terms of tephra accumulated at the ground and airborne tephra concentration: to achieve unbiased PVHA it is mandatory to account for the natural variability linked to the eruption scale or type, the consequent Eruptive Source Parameters (ESP), the position of eruptive vents, and the wind field. This translates into a large number of numerical simulation for tephra transport and deposition, on a large-scale domain, that we perform with the flagship code Fall3D optimized in ChEESE. In this PD, we are proposing two types of Probabilistic Volcanic Hazard Assessment (PVHA), that may become services: short- (days to weeks) and long-term (years to decades) PVHA. Further, we focus on two target volcanic systems: Campi Flegrei caldera, in Southern Italy, located in one of the most populated areas in Europe and characterized by a long-lasting unrest (since 2004), and Jan Mayen In North Atlantic (Norway), a little island located on the highest-traffic air routes between Europe and North America. The short-term PVHA is proposed as an illustrative example at the Campi Flegrei volcano, on a particular set of dates (5 to 7 December 2019), when an earthquake of magnitude 3.1 was located in the caldera. The long-term hazard assessment is proposed both for Campi Flegrei (for ground tephra accumulation and airborne ash concentration), and Jan Mayen island (for the airborne ash concentration only). |
| Workflow description | PVHA_WF_st fetches the monitoring data (seismic and deformation) and, together with the configuration file of the volcano, calculates the eruptive forecasting (probability curves and vent opening positions) and uses the output file from alphabeta_MPI.py to create the volcanic hazard probabilities and maps. PVHA_WF_lt uses the configuration file of the volcano to calculate the eruptive forecasting and, together with the output file from alphabeta_MPI.py, creates the volcanic hazard probabilities and maps. Meteo data download process is fully automated. PVHA_WF_st and PVHA_WF_lt connect to the Climate Data Store (Copernicus data server) and download the meteorological data associated with a specified analysis grid. These data will later be used to obtain the results of tephra deposition by FALL3D. |
| Tested architectures | MareNostrum-4 at BSC Joliot-Curie (Irene-rome and Irene-skylake partitions) at CEA/TGCC. Ceneri in INGV-Osservatorio Vesuviano. |
| Target TRL | 6 or 7 View TRL chart |
| Relevant stakeholders | Civil protection agencies |
| Achievements up to M18 |
|
HPC benefits in both capacity and capability. On the one hand, since we are analyzing a high-resolution grid, we need a bigger capability. The simulations need to be parallelized in three-dimensional subdomains. Each of these subdomains describes a subregion of the grid (latitude, longitude, and altitude) and is assigned to one core optimizing the time and computation resources. On the other hand, since we are simulating a few thousand eruptive scenarios to obtain hazard and probability maps and reduce the uncertainty, we need a bigger capacity. HPC allows to launch hundreds of simulations at the same time, reducing the time to achieve results by several orders of magnitude. The global results are similarly influenced by both capability and capacity. In this sense, the Exascale target is widely justified.
By increasing the computational resources we obtain faster results, so we are able to explore the epistemic variability associated to unknow eruption size in the short-term case (whereas, before ChEESE, due to limited computational resources only 3 simulations per day could be run in useful time for short-term hazard forecast).
Also, we can explore larger domains with high resolution. This allows analysing, for long-term hazard assessment, the impact of low-probability but distal-consequence events.
Scalability is being improved by the POP project. On the other hand, we are faced with the problem of handling too large data matrices which we should manage changing the information we save.
| Number of cores / GPUs: | Memory (GB): | Storage (GB) both temporal and permanent: | #files written both temporal and permanent | I/O data traffic per hour during job | |
| Minimum: |
768 cores 16 nodes 48 cores/node |
4 GB/core 256 GB/node Total 4096 GB |
150 TB |
22680 CF_short term 63000 FC_long term 4200 JM |
|
| Average: | |||||
| Maximum: |
1024 cores 22 nodes 48 cores/node |
256 GB/node Total 5632 GB |
150 TB |
22680 CF_short term 63000 CF_long term 42000 JM |
