Sunday, 28 January 2024
Hall E (The Baltimore Convention Center)
Sita Baaba Nyame, University of Connecticut, Storrs, CT; and W. O. Taylor, D. Cerrai, A. Spaulding, M. Denton, M. Koukoula, and F. Yang
Handout
(598.1 kB)
In western United States, the peak of wildfire season occurs between the end of summer and the beginning of the fall. The effects of climate change and the resulting alterations of environmental conditions, such as an increase in aridity, pose a big threat in increasing the time frame of this fire season. In anticipation of wildfires, many utility companies cut off power to neighborhoods out of fear of contributing to the ignition and spread of wildfires. Currently, there are many models for understanding the spread of wildfires but not much on predicting ignition sites. These models allow for creating mitigation plans for when these fires occur, however, they require wildfire ignition location to be usable. Creating a Wildfire Ignition Model (WIM) will allow federal government, states, utilities, and stakeholders to be proactive in identifying the location of possible fires and preventing their occurrence.
The WIM uses machine learning techniques to predict the location of possible wildfires on a daily basis. The model is trained on 23 years of historical fire, environmental, terrain, leaf conditions and vegetation data on the State of California. This model focuses on predicting wildfire occurrence locations during the wildfire season in California, which starts in the month of July and ends in October. Model predictions are cross validated to increase its generalizability.
In this presentation, I will be explaining the trends in fire ignitions seen over the years, the influence of different input data on the model’s predictions, and model performances in identifying fire ignition sites. Model skills will be presented after extensive evaluation across different climate zones, seasons and outlier wildfire events. Future works on this project include adapting it to other states and countries facing wildfire issues; narrowing down key variables that contribute to making wildfire ignition sites; and creating a refined model to look at ignition sites in proximity to power lines.

- Indicates paper has been withdrawn from meeting

- Indicates an Award Winner