Wildland Fire Monitoring and Modeling

Research Team: Anthony Vodacek, Ambrose Ononye, Robert Kremens (LIAS), Ying Li
(Ph.D. student), Zhen Wang (Ph.D. student)

Sponsor(s): NASA and NSF.

Project Scope: The goal of this project is to develop a dynamic data driven system for modeling the propagation of wild land fires. The end product will give wild land fire managers a tool for predicting the propagation of wild land fire with steering of the modeling results by data from ground sensors and remotely sensed data. RIT is developing the interface to use thermal imagers such as WASP or WASP-lite as airborne data sources of fire location. Dr. Kremens is also continuing to develop ground sensor measurement capabilities and experimental data on the optical and thermal radiation emitted by fires. RIT is creating synthetic images from the model output to facilitate comparison to the ground sensor and airborne remote sensing data to provide steering of the model results. We are also using DIRSIG as a visualization tool for generating synthetic 3D fire scenes including the 3D structure of flames.

Project Status: This multi-year project with multiple sources of funding began in 2000 and will continue to at least August 2009. Ph.D. student Ying Li finished her dissertation in 2006. Her work consisted of algorithm development for fire detection and mapping. Her work on contextual fire detection methods produced a statistically based method for detecting fire in multispectral images. Her approach is flexible in that it is not sensor specific and can be applied to a wide variety of satellite and airborne sensors with different spectral bands and spatial resolutions. An example of wildland fire detection in a Moderate Resolution Imaging Spectrometer (MODIS) image is shown in Figure 1.

Research Scientist Dr. Ambrose Ononye worked specifically with high resolution multispectral images of fire and developed an algorithm for extracting fire line parameters that are useful for fire management. His approach used a variety of image processing tools combined with spectral processing tools to outline the fire perimeter, isolate the actively burning areas of the fire, and determine the direction of the fire propagation automatically. An example of his approach is shown in Figure 2.

Ph.D. student Zhen Wang finished her dissertation in 2007. Her work was to use the output of fire propagation models combined with DIRSIG's atmospheric plume capabilities to generate 3D synthetic scenes of wildland fire. See Figure 3. This capability is a key component of the feedback process in which the fire propagation model output is steered according to the available remotely sensed images of a fire. In this case the fire propagation model output is used to generate a synthetic image that can then be directly compared to a real image.

Figure 1: An image of the 2003 Southern California fires obtained with the NASA MODIS sensor. Pixels with active fire found using the statistical fire detection algorithm are marked in red	Figure 2: An image of a fire obtained with the NASA AVIRIS sensor with overlays of the fire perimeter and the fire direction automatically drawn by Dr. Ononye's algorithm
Figure 3: A simple infrared DIRSIG scene containing a grass fire. The output of a fire propagation model was used as input to describe the 3D shape of the flames