Statistical models for disaster waste estimation: Incorporating uncertainty using Bayesian regression

Accurately estimating the amount of disaster waste is crucial to achieve sustainable disaster waste management. In light of the insufficient consideration of uncertainty in previous studies, this study aimed to develop Bayesian regression-based estimation models that provide predictions incorporating uncertainty through a simple structure. To this end, 21 candidate models with different likelihoods and linear models were compared using cross-validation and metrics of probabilistic prediction accuracy, point prediction accuracy, and interval prediction accuracy. The results showed that a hierarchical model using a log-normal or Weibull distribution, with the number of damaged houses as predictor and disaster type and degree of urbanization as grouping variables, achieved the best prediction performance. Applying the model to regional datasets enables the construction of area-specific estimation models, allowing for reasonable estimation of disaster waste towards better decision making. • Bayesian regression model developed to estimate disaster waste with uncertainty. • Hierarchical log-normal or Weibull models showed superior performance. • Key factors were number of damaged houses, disaster type, and urbanization level. • The model enables region-specific estimation with region-specific data.