Empirical Stage-Damage Functions for SME Content Losses in Australian Flood Vulnerability Assessment

Small and medium-sized enterprises (SME) dominate Australian business activity, yet empirically calibrated functions for nonresidential flood-related contents losses remain limited. Postevent surveys of 185 SMEs across 12 business types and 14 towns affected by the February–April 2022 Northern Rivers floods were used to derive stage–damage functions for stock, equipment, and fit-out. Contents-specific damage indices combined peak internal flood depth (0–6 m, relative to finished-floor level) with self-reported losses and preevent replacement values. Pooled curves were fitted using LOESS and a beta-family generalized additive model (Beta-GAM); a hierarchical Beta-GAM (HBGAM) generated business-type functions under partial pooling, while random forest and XGBoost provided depth-only predictive benchmarks. Across the pooled models, performance on the corresponding within-event 25% holdout sets was moderate ( R 2 = 0.57 – 0.69 ; RMSE = 0.23 – 0.27 ), and neither machine-learning benchmark consistently improved performance. The Beta-GAM was retained because it provided smooth, bounded, interpretable relationships, model-based and bootstrap uncertainty intervals, and extractable threshold depths. The fitted mean damage index reached 0.50 at 1.03 m for stock, 1.09 m for equipment, and 1.27 m for fit-out, with uncertainty generally increasing above 3 m as data density declined. HBGAM-derived D50 values ranged from 0.66 to 1.18 m for stock, 0.81 to 1.18 m for equipment, and 1.09 to 1.48 m for fit-out, indicating model-derived sector variation. These event-calibrated functions support screening-level relative SME contents-loss assessment and evaluation of loss-reduction options in comparable settings; application elsewhere requires external validation or local recalibration.