Spatial decoupling of fire hazard and vulnerability in dryland Australia: a compound risk index integrating 24 years of remote sensing and climate reanalysis data

Operational wildfire management in Australia prioritises areas of high fire danger, implicitly assuming that fire is most consequential where it is most likely. We developed a Compound Fire Risk Index (CFRI) for the Brigalow Belt Bioregion, Queensland, informed by the IPCC AR6 risk framework, in which a hazard index is multiplied by a vulnerability component combining exposure, sensitivity and adaptive capacity. The index integrates remote sensing, climate, census and infrastructure data at 1 km resolution, with weights assigned a priori through expert-based allocation. All time-varying predictors were built from 2001–2019 data, and the index was validated against withheld 2020–2024 burn scars. Hazard and vulnerability were spatially decoupled (r = −0.01). The decoupling persisted against observed fire history, after controlling for population density and for each vulnerability component, and the top-decile hazard and vulnerability areas overlapped less than expected by chance (5.9% versus 10%). In regional landscapes, vulnerability was concentrated where hazard was lowest (r = −0.50 to −0.25). The CFRI discriminated 2020–2024 burned areas modestly but robustly better than hazard alone (AUC = 0.554 versus 0.501; ΔAUC = +0.053, spatial block-bootstrap 95% CI 0.037–0.069). Under the wet La Niña conditions of the validation period, the vulnerability component alone discriminated burned areas best (AUC = 0.626), consistent with ignitions concentrated near people and infrastructure. The CFRI identified areas largely distinct from Queensland's Bushfire Prone Area mapping (ρ = 0.11), which maps where fire can propagate and discriminated burned areas more strongly (AUC = 0.714); the two products are complementary. Rankings were stable under ±10% and ±20% perturbation of the three top-level vulnerability weights (Spearman ρ ≥ 0.993); lower-level weights were not tested. Prioritising by hazard alone would shift 35% of the highest-priority area and largely miss the most vulnerable areas. The workflow is transferable to comparable dryland regions.