Skip to main content

There has been a lack of tools and platforms for real-time prediction of wildfire movement and risk. Commonly used models do not address the dynamic nature of an area’s current meteorological conditions such as the wind, humidity, and precipitation when determining the direction and speed of fire propagation. Near-real time predictions would allow users to estimate localized wildfire risk and is crucial for focused and efficient firefighting. In this project, openly available datasets from USGS, LANDFIRE, NASA, InciWeb, and NOAA along with others are used for modeling and visualization purposes. The computation of fire risk utilizes Apache Spark, an open source big data computation platform, so that platform users could instantly view modeling results. Two-tier modeling is used to speed up the computation process. Initially a large spatial scale computation model will be run on a regional area of interest identifying and prioritizing areas that need to be further processed for local risk assessment. Prioritized discrete areas would then be scrutinized in much finer resolution to improve model accuracy. Historical fire behavior data will be used to train the model using machine-learning algorithms. Testing and calibration both on accuracy and computing performance of the model is performed through benchmarking against the most popular fire simulation models and libraries used in wildland fire assessment. Modular design approach is implemented to ensure a scalable and dynamic model for future additions and expansions, such as a web interface where real time wind and fire spread animations can be viewed. This is part of the Missoula Fire Sciences Laboratory 2018 Seminar Series.

Media Record Details

May 17, 2018
Thomas A. Minckley

Cataloging Information

Topic(s):
Fire Behavior
Simulation Modeling

NRFSN number: 17777
Record updated: