Empirical Fragility-Based Adjustment of Embodied Carbon Using Machine-Learning-Derived Post-Hurricane Fragility Data

Climate mitigation and hazard resilience are often treated as separate goals, even though storm-driven repairs can add substantial embodied greenhouse gas emissions over a building's lifetime. This paper presents a data-driven workflow that links post-disaster field observations (StEER), site-specific wind fields (ARA), and machine-learning fragility models to support hazard-adjusted embodied carbon assessment. A pooled dataset of 979 residential buildings inspected after Hurricanes Laura (2020) and Michael (2018) was harmonized to a common schema and assigned peak 3-second gust wind speed using nearest-neighbor geospatial matching. Gradient boosting classifiers estimate the probability of roof and wall replacement from gust intensity and building attributes. The models achieve AUC = 0.75 (roof) and AUC = 0.92 (wall). Gust speed is the main driver, with roof age and wall cladding group acting as key modifiers. These fragility outputs can be combined with regional hazard frequency and component-specific repair carbon factors to estimate expected repair-related emissions over a service life. The framework treats resilience as an emissions-reduction strategy by translating replacement risk into lifecycle CO2e consequences.