Position details

Data Assimilation for wildfire modeling
Training › Climate Modelling And Global Change - Data Assimilation

Required Education / Niveau requis

Master 2

From / Date de début

1st February 2013

Context / Contexte

Predicting the speed and the direction of wildfire propagation has obvious

applications, but remains a challenge because many parameters are unknown, in

particular what is burning, at which speed, and what is the local humidity.

A promising method to predict fire evolution is to couple Computational

Fluid Dynamics (CFD) and data assimilation, so as to incorporate, into the

CFD solver, ground or airborne information obtained in real-time on the fire

position and evolution. This technique, initially developed in meteorology,

is useful to compare experiment to simulation as well as to reduce

uncertainties in the inputs and the outputs of a model.

This project is well-suited for students with an interest in fire physics

(combustion, heat transfer, fluid mechanics), CFD, data assimilation and

programming (Fortran). A PhD project may follow this initial internship. The

project is a collaboration between CERFACS and University of Maryland.

CERFACS has a long experience in CFD and data assimilation, and University

of Maryland provides the expertise in the field of fire modeling.

This subject corresponds to an innovative application of data

assimilation to the

area of combustion and fire science. The internship could possibly be

performed at CERFACS and at UMD.

Description / Description

This work builds on the expertise of three previous internships in 2010, 2011 and 2012 that proved the feasibility of data assimilation for fire spread modeling

using both synthetical and real observations. The concept is that a data-driven fire model can lead to more accurate predictions of fire spread when using observations

similar to airborne pictures.

Via the data assimilation algorithm, these observations are compared to a simplified model of fire propagation, based on a level-set equation and in which the main physical quantity is the speed of the fire front, also called the fire rate of spread (ROS). The ROS depends on the local fuel characteristics and environmental conditions (typically, local wind and slope). In this context, an ensemble based data assimilation algorithm (Ensemble Kalman Filter) calibrates the model input parameters, which are significant sources of uncertainties in the estimation of the ROS. This ensemble technique implies that the uncertainties are statistically described and reduced as observations are sequentially assimilated.

To continue this work, efforts should be made on the extension of the control vector in order to correct the front position in addition to the input model parameters. This will allow for a spatialized correction accounting for all sources of errors at once and lead to the improvement of forecast simulations. The development will be made using the OPALM coupling software that offer a parallel environment particularly adapted for stochastic algorithms. The trainee will learn data assimilation methods and the basic

physics of fire modeling, and contribute in the further development of a data assimilation prototype for fire propagation.

Contacts / Contacts

Name: RICCI Sophie
Phone: +33(0)5 61 19 31 28
Fax: +33(0)5 61 19 30 00
Email: sophie.ricci@cerfacs.fr

Name: Cuenot Bénédicte
Phone: 05 61 19 30 44
Fax: 05 61 19 30 00
Email: cuenot@cerfacs.fr

Salary / Rémunération

580 euros/month