Evaluating the scale-dependent performance of cetacean hotspot–based spatial indicators in a dynamic estuarine ecosystem

Hotspot maps derived from mobile marine predators are widely used as spatial indicators for conservation and marine spatial planning, yet their decision utility is rarely validated in dynamic estuarine systems. We evaluated whether Indo-Pacific humpback dolphin ( Sousa chinensis ) hotspot intensity can identify management-relevant ecological conditions in the Jiangmen waters of the western Pearl River Estuary. Using dolphin photo-identification sightings (2013–2016), trawl-based fish surveys, environmental covariates, and vessel-tracking fishing effort, we quantified dolphin space-use intensity with kernel density estimation (KDE) at two temporal scales (multi-year vs within-season). We conducted area-based comparisons (KDE50, KDE95, Outside) for fish community structure, taxonomic and functional diversity, functional traits, CPUE, and anthropogenic pressure, and used db-RDA variation partitioning to assess the relative contributions of environmental gradients versus dolphin space-use areas. To validate indicator performance directly, we extracted normalized KDE intensity at fish sites and quantified discrimination effects with ROC-AUC for three operational targets: CPUE hotspots, species richness (SR) hotspots, and low fishing-pressure areas. Multi-year KDE50 showed a pronounced mismatch with reserve boundaries and clearer ecological differentiation among areas than within-season analyses. However, environmental gradients explained more variance in fish metrics than dolphin areas alone. ROC-AUC revealed indicator performance patterns that were both metric- and scale-dependent, including inverse discrimination for SR and low fishing-pressure targets, and only weak discrimination for CPUE hotspots. Overall, dolphin KDE hotspots showed limited reliability as fine-scale spatial indicators without explicit, scale-aware validation. • Cetacean hotspot intensity was evaluated as a spatial indicator across temporal scales. • Zone-based ecological differences did not ensure reliable indicator discrimination. • ROC–AUC revealed weak and scale-dependent performance of KDE-based indicators. • Environmental gradients explained more variation than predator space use. • Hotspot-based indicators require explicit, scale-aware validation before application.