Quantifying Exposure Information Uncertainty in Regional Risk Assessment

TL;DR

This paper introduces a methodology to quantify and decompose bias and uncertainty in regional risk assessments caused by incomplete exposure data, using analytical and simulation approaches.

Contribution

It presents a novel framework combining analytical and simulation methods to assess how missing exposure information impacts regional risk estimates.

Findings

Decomposes total uncertainty into exposure, hazard, and damage contributions.

Develops a high-resolution bridge exposure inventory using data augmentation, machine learning, and imputation.

Demonstrates the methodology on bridge-specific and regional risk assessments.

Abstract

Exposure characterization in regional risk assessment aims to assign physical properties to the assets of interest so they can be associated with damage and loss functions. While this process has benefited from the growing availability of public infrastructure inventories, these datasets often lack the detailed attributes required for high-resolution risk assessment. Missing attributes are commonly inferred using predictive models or engineering-based rulesets. However, these imputations are inherently imperfect and can introduce bias and additional uncertainty in regional risk estimates. This study proposes a methodology to quantify the bias and uncertainty in regional risk assessment that arises from probabilistic exposure characterization. By integrating analytical and simulation-based approaches, the methodology decomposes the total uncertainty into contributions from incomplete exposure information as well as other sources, including hazard and damage characterization. This decomposition clarifies how bias and uncertainty associated with missing exposure information are generated and propagated through the risk assessment pipeline. The methodology is applied to both bridge-specific and regional risk assessments. A high-resolution bridge exposure inventory is developed using a data augmentation framework that combines publicly available information with machine learning and engineering-based imputation methods.