Modeling wildland fire behaviour using a multi-physics system on hpc platforms

• Autores: Àngel Farguell Caus
• Directores de la Tesis: Ana Cortés Fité (dir. tes.), Jan Mandel (codir. tes.), Josep Ramon Miró Cubells (codir. tes.)
• Lectura: En la Universitat Autònoma de Barcelona ( España ) en 2018
• Idioma: español
• Tribunal Calificador de la Tesis: Mélanie C. Rouchoux (presid.), Eduardo Cesar Galobardes (secret.), Arnau Folch Duran (voc.)
• Programa de doctorado: Programa de Doctorado en Informática por la Universidad Autónoma de Barcelona
• Materias:
  • Lógica
    • Metodología
  • Matemáticas
    • Ciencia de los ordenadores
      • Informática
      • Simulación
• Enlaces
  • Tesis en acceso abierto en:  TESEO  TDX 
• Resumen

  • Damages resulting from wildfires have arisen as a major threat worldwide. Properly accounting for the interaction between the fire and the atmosphere surrounding the hazard could aid firefighters and civil protection staff in making more informed, better decisions during an ongoing event. In that sense, WRF-SFIRE is a wildfire simulator which couples the meteorological model WRF-ARW and the fire spread model SFIRE solving Rothermel's equation through the level set method. This model solves the complex interaction between the atmosphere and the fire through a Computational Fluid Dynamics (CFD) approach. However, it has some limitations which provide the motivation for this investigation.

    The aforementioned coupled system needs to run fast enough to assure real-time execution. A deep analysis of the parallelism programmed in WRF-SFIRE is an important matter to get operational results. The best way to run WRF-SFIRE fast is using a distributed memory parallelism with MPI, but it has some limitations because of the dimension of the division of the domain.

    Another important element of WRF-SFIRE, which evolves the fire being modeled and keeps it updated, is the level set method. The level set method with strong and heterogeneous rates of spread suffers from instabilities, resulting in spurious fires. This is solved by enforcing the constraint that the level set function at a point may not decrease below the minimum value at neighbors.

    Finally, a new method of fitting the fire arrival time to observed perimeter data is proposed. This new method can be used to generate an artificial fire history, which can be used to spin up the atmospheric model for the purpose of starting a simulation from the observed fire perimeter. The main idea is to minimize a non-linear objective function, which is zero when the fire arrival time satisfies the eikonal equation. This new method, unlike position or additive time corrections, respects the dependence of the fire rate of spread on topography, diurnal changes of fuel moisture, winds, as well as spatial fuel heterogeneity. This interpolation method could be used to assimilate fire perimeters and satellite fire detections into the coupled atmosphere-fire model.