You are here:
Using Bayesian networks with Fisher information to assess and manage wildfire risks
Citation:
Carriger, J. AND H. Cabezas. Using Bayesian networks with Fisher information to assess and manage wildfire risks. SETAC North America 41st Annual Meeting, NA, N/A, November 15 - 19, 2020.
Impact/Purpose:
Slides for a platform presentation at the 2020 SETAC North America meeting on combining Bayesian networks with Fisher information for wildfire assessments.
Description:
Wildfires are important for managing and regulating ecosystems for ecological and human use. Humans have long played a role in prescribing wildfires for usages such as hunting and safety, and many ecological systems are fire-adapted. However, severe wildfires may cause injuries to the environment ,property and human health. Recent events have demonstrated the dangers from severe wildfires and the severity and number of wildfires is only expected to increase in the future. Causal factors for severe wildfires in a region of concern include prior fire history, soil moisture, vegetation characteristics, and atmospheric conditions. An approach to incorporate these causal factors and examine spatial regions for severe wildfire potential is proposed. The approach relies on Bayesian networks and Fisher information. Bayesian networks are used to develop causal or semi-causal probabilistic models that link causal factors with severe fire predictions. Machine learning may be used to help derive an optimal network structure from data. Fisher information is used on virtual data generated from scenarios in the Bayesian networks for examining uncertainties in variables used to predict severe wildfires. For analyzing scenarios, the uncertainty of individual variables can be tracked but multiple variables can be combined for emergent properties with joint Fisher information calculations. A hypothetical demonstration will be given using spatial data from the 2013 Rim Fire (Sierra Nevada) that spanned private and public lands including the Yosemite National Park and Stanislaus National Forest. Combining Bayesian networks and Fisher information might provide an approach for interpreting the uncertainty in the knowledge of severe wildfire risks.