Compound disasters arising from interacting hazards increasingly challenge urban resilience, yet spatially explicit tools for diagnosing where hazards co-vary remain limited. To address this gap, this study develops a spatial explicit inferential diagnostic framework that applies moving-window cross-correlation to harmonized raster risk layers, providing a map-based assessment of local inter-hazard dependence. The framework implements three complementary statistics within a moving-window design—Dutilleul's modified t -test (spatial-autocorrelation-adjusted inference), mutual information (nonlinear dependence detection), and Kendall's τ (rank-based robustness)—and benchmarks their performance in terms of diagnostic consistency, computational efficiency, reliability, and scale sensitivity across multiple spatial scales. The Guangdong–Hong Kong–Macao Greater Bay Area, China, serves as the case study, focusing on water-related compound risk. Results show that high-to-very-high dependence zones concentrate in coastal–estuarine corridors and major urban drainage basins, highlighting interaction patterns not recoverable from single-hazard overlays alone. Under an uncertainty-aware validation framework, Dutilleul's modified t -test achieved the strongest fine-scale hotspot delineation, with 88.0% of river flood–urban inundation event and 65.7% of urban inundation–storm surge events intersecting mapped high-dependence zones. Kendall's τ provides balanced robustness and stable performance across scales, supporting regional screening and cross-validation of spatial patterns. Mutual information generates smoother dependence fields with low scale sensitivity, indicating value for reconnaissance-style mapping but reduced fidelity for localized hotspot targeting. Runtime comparisons further demonstrate that methodological performance is scale-conditional, with Dutilleul's test fastest at fine resolutions but increasingly computationally intensive at larger window sizes. Rather than replacing indicator-based risk assessment, the proposed framework functions as a complementary diagnostic tool that quantifies where hazards reinforce one another under spatial autocorrelation. By explicitly mapping localized dependence regimes, the study advances multi-hazard interpretation and provides a transferable workflow for compound-risk screening in data-rich urban regions. Future research should expand compound-event validation datasets, improve computational scalability, and extend the hazard-agnostic workflow to additional hazard combinations and scenario-based planning contexts. • Spatially explicit moving-window diagnostics reveal localized inter-hazard dependence beyond single-hazard overlays. • Dutilleul's modified t -test yields the clearest fine-scale hotspot, capturing 88.0% of dominant compound events. • Kendall's τ offers robust, rank-based screening with stable spatial behavior across multiple scales. • Mutual information emphasizes smooth regional dependence gradients and nonlinear association patterns. • The framework functions as a complementary diagnostic layer to support multi-hazard planning and compound-risk screening.
Lu et al. (Fri,) studied this question.