Bayesian networks for knowledge synthesis and translation: Examining post-wildfire impacts on aquatic ecosystems
On this page:
Wildfire is a natural component of many freshwater riverine ecosystems; however, uncharacteristically frequent and severe wildfires threaten stream habitat and biota crucial to long-term ecosystem stability. The existing body of literature explores post-fire abiotic (environmental) and biotic (macroinvertebrate, fish, primary producer) variable responses, but limitations in knowledge synthesis may be contributing to an incomplete picture of ecosystem responses and recovery. We demonstrate a Bayesian network analysis in an ecological risk assessment (ERA) format to systematically synthesize and translate knowledge, aiming to 1) improve understanding of the relative importance/impact of different post-fire environmental changes on post-fire biotic responses and 2) elucidate best practices for wildfire observation study design. We applied naïve and tree-augmented naïve structure learning to data from a prior systematic review on aquatic ecological impacts from wildfires (Erdozain et al, 2024, “Fire impacts on the biology of stream ecosystems: A synthesis of current knowledge to guide future research and integrated fire management.” Global Change Biology, 30, 317389, https://doi.org/10.1111/gcb.17389). The dataset was filtered to exclude variables with a missing data threshold of >10% and niche cases that would bias the model fitting. Labeling convention for the selected variables (nodes) and their associated data values (states) was standardized. The cleaned and filtered data were split into training and testing subsets with an 80:20 ratio and then supplied as test cases to our Bayesian network model in the Norsys Netica software. Preliminary model results found observed endpoint responses were most sensitive to the occurrence of post-fire hydrologic events (floods, debris flows, high inputs of sediment into channels) and to the amount of time elapsed since the fire.