Citation

Himoto K, Hayashi Y. Fire Safety J. 2022; 134: e103679.

Copyright

(Copyright © 2022, Elsevier Publishing)

DOI

10.1016/j.firesaf.2022.103679

PMID

unavailable

Abstract

Spot ignition by wind-dispersed firebrands is an important factor for large outdoor fires. However, the large probabilistic variability in the generation, transport, and ignition processes makes it difficult to quantify the risk. Data collection under a wide range of conditions is essential for quantifying uncertainty. Generally, this can be addressed only with relatively small-scale experiments conducted repeatedly under the desired conditions, simultaneously involving ambiguity in the consistency of the full-scale behavior. However, data collection in full-scale experiments is complex, particularly for behaviors such as large outdoor fires, which makes it difficult to acquire sufficient data for statistical analysis. To fill this gap, we employed a hierarchical Bayesian approach to alleviate the estimation variance of the probabilistically varying characteristics of the spot-ignition processes involved in full-scale behavior. In the present analysis, the results of two previous experiments were used: a full-scale experiment that burnt a three-story wooden building, and a wind tunnel experiment that burnt wood cribs in an open-top combustion apparatus. Pooled models, using mixed data from all experiments without considering the relationship between data sources, were less successful in developing probabilistic distributions in some instances. By contrast, hierarchical Bayesian models (HBMs) are relatively robust, compensating for the scarcity of full-scale data with small-scale data. The obtained HBMs were applied to estimate the probabilistic distributions for the size and transport distance of firebrands under the hypothesized fire conditions. A comparison between the HBMs and existing probabilistic models demonstrated a significantly short dispersal range of firebrands by the latter, which may lead to an underestimation of hazards in practical applications. This study presents a methodology for constructing a reasonable probabilistic model for the processes of spot ignition behavior, which often encounter the issue of data limitations for statistical analysis.

Language: en

Keywords

Firebrand generation; Firebrand transport; Hierarchical bayesian analysis; Large outdoor fires; Statistical analysis