Quantifying Bias in Cloud Analysis–Based Fragility Under Ground Motion Selection Uncertainty

Abstract The cloud analysis (CA) method is widely used in the seismic fragility analysis of structures due to its computational efficiency and its ability to avoid biases resulting from ground motion scaling. However, its reliability remains strongly dependent on the selection of ground motion records, and the extent to which improved input selection can mitigate model bias is still unclear. This study proposes a practical strategy to examine the limitations and potential of the CA method with a focus on the selection of ground motion records. To this end, a rigorous selection scheme based on Latin hypercube sampling and sufficiency and efficiency checks for intensity measures was first implemented. Then, the performance of the CA method using this optimized ground motion input was evaluated against a high-fidelity benchmark derived from multiple stripe analysis employing hazard-consistent conditional spectrum–based ground motion selection. The comparative results indicate that rigorous ground motion selection significantly stabilizes fragility estimates for slight and moderate damage states; however, it cannot overcome the inherent theoretical limitations of the CA method in predicting structural collapse. These findings demonstrate that, regardless of input selection quality, the log-linear assumption between structural demand and ground motion intensity adopted in the probabilistic seismic demand model tends to underestimate seismic demand at high-intensity levels. This study provides insight into the valid range of applicability of the CA method. The validity boundary of the CA method is identified, beyond which refined ground motion selection alone is insufficient. This study provides a risk-informed perspective on the applicability of the CA method and highlights the necessity of advanced nonlinear probabilistic models for reliable collapse assessment.