FireFlux Experiments

FireFlux Experiments

The performance of our prototype data-driven wildland fire spread model is being evaluated by comparisons with data from field-scale controlled fire experiments. We first consider the Fireflux experiment that took place on February 23, 2006, at Houston Coastal Center in Texas [1]. The Fireflux experiment corresponds to a 0.63 km2 plot size, flat terrain, approximately homogeneous tall grass vegetation (fuel parameters: fuel loading = 1.08 kg/m2; average depth = 1.5 m; density = 400 kg/m2; surface-to-volume ratio = 5000 m-1; moisture content = 9%; heat of combustion = 15.4 MJ/kg), and relatively low wind conditions (speed = 2 m/s; direction = 10 degrees into south-west direction). The rate-of-spread (ROS) of the fire in the Fireflux experiment is approximately 1 m/s (in the wind direction) and the duration of the experiment is 15 minutes.

The Fireflux experiment was originally proposed as a validation experiment for coupled fire-atmosphere models and it has been used by a number of research groups as a benchmark test for CFD-based wildland fire dynamics. For instance, Filippi et al. performed detailed numerical simulations of the Fireflux experiment using a fire rate-of-spread model (ForeFire) coupled with an atmospheric boundary layer model (Meso-NH) and obtained results that compared well with experimental data [2]. In this project, we use the high-resolution fireline data from Filippi’s simulations as observation data for basic testing of our data assimilation algorithms.

fireflux

Fireflux experiment: picture showing grass fire moving from left to right towards a 43-m-tall meteorological instrument tower. Credit: Craig Clements [1].

Data- Driven Modeling

Our current prototype data-driven wildland fire spread model uses an in-house fire rate-of-spread model (FireFly) for forward modeling and an Ensemble Kalman Filter (EnKF) for inverse modeling. FireFLY is based on Rothermel’s classical model: Rothermel’s model was originally developed to describe fire spread in the wind direction and/or in the direction of the slope of the terrain, i.e. to describe spread of head fires; the model, however, has limited accuracy when applied to flank fires or rear fires. In this project, we use data assimilation to correct the modeling limitations of FireFly and predict the fire behavior observed in the Fireflux experiment.

The figure below shows a typical result from our testing campaign. The figure shows a comparison between the observed fireline position (taken from Filippi’s detailed numerical simulations), the predicted fireline position based on a FireFly simulation and without data assimilation (free run), and the predicted fireline position based on an ensemble of FireFly simulations and with data assimilation (data-driven run). The data-driven prediction assumes that the wind magnitude and direction is unknown and may change both temporally and spatially (the algorithms use between 12 and 48 control variables, and 200 ensemble members). The data-driven run shown in the figure corresponds to a forecast of the fireline position at time t = 400 s informed by a data assimilation update previously performed at time t = 300 s. The good level of agreement between the data-driven run and the observations (much improved compared to the poor level of agreement between the free run and the observations) confirms the ability of data assimilation to provide an optimized forecast of the fire spread dynamics.

firefluxsim

Fireflux experiment: comparison between the observed fireline position (observations) and the predicted fireline positions, without (free run) and with (data-driven run) data assimilation.

Operational Modeling

We consider here again the Fireflux experiment. This experiment is now simulated using FARSITE [3]. FARSITE uses Rothermel’s classical model to describe the spread of head fires and an assumed local ellipsoidal fireline shape combined with empirical correlations to describe the spread of flank fires or rear fires. The simulator has been enhanced over the years with a database of vegetation fuels. FARSITE is the official wildland fire spread simulator of the US Forest Service.

We here evaluate the ability of FARSITE to predict the fire behavior observed in the Fireflux experiment. The figure below shows a typical result. The figure shows a comparison between the observed fireline positions (taken from Filippi’s detailed numerical simulations [2]) and the predicted fireline positions based on a FARSITE simulation (and without data assimilation, also called a free run). The FARSITE prediction assumes flat terrain, homogeneous tall grass vegetation (fuel parameters: fuel loading = 1.08 kg/m2; average depth = 1.5 m; dead fuel moisture content = 9%) and relatively low wind conditions (speed = 2.1 m/s; direction = 10 degrees into south-west direction).

Fireflux experiment: comparison between the observed fireline positions (observations) and the fireline positions predicted by FARSITE. The fireline positions are plotted at different times after ignition: t = 120, 240, 360, 480, 600 and 720 seconds.

Fireflux experiment: comparison between the observed fireline positions (observations) and the fireline positions predicted by FARSITE. The fireline positions are plotted at different times after ignition: t = 120, 240, 360, 480, 600 and 720 seconds.

References:

[1] Clements, C.B., S. Zhong, S. Goodrick, J. Li, X. Bian, B.E. Potter, W.E. Heilman, J.J. Charney, R. Perna, M. Jang, D. Lee, M. Patel, S. Street, G. Aumann (2007) “Observing the dynamics of wildland grass fires: FireFlux – A field validation experiment,” Bulletin of the American Meteorological Society, 88(9),1369-1382.

[2] Filippi, J.B., X. Pialat, C.B. Clements (2013) “Assessment of FOREFIRE/MESONH for wildland fire/atmosphere coupled simulation of the FireFlux experiment.” Proceedings of the Combustion Institute. 34, 2633-2640.

[3] M.A. Finney, M. Alexander, P. Andrews, J. Beck, B. Keane, J. Scott, “FARSITE : Fire Area Simulator — Model Development and Evaluation,” USDA Forest Service, Rocky Mountain Research Station, Research Paper RMRS-RP-4 (2004).