Continental-Scale Dynamics of Flood Extremes: A Unified Spatio-Temporal Modelling Approach

Flood risks are escalating globally as climate change intensifies land–atmosphere interactions, resulting in large-scale flooding. Yet, existing assessments remain constrained by site-based and stationary assumptions that obscure how extreme floods propagate across space and evolve over time. Addressing this gap, we present the first continent-scale analysis of changing spatio-temporal dependence in Australian flood extremes. Towards this, we develop a comprehensive suite of 3,937 spatial and spatio-temporal max-stable process (MSP) models, integrated with large-scale climate modes and physiographic controls. Leveraging annual maximum floods from 325 Hydrologic Reference Stations and a 30-year moving-window framework (1973–2002), we quantify evolving spatio-temporal dependencies in floods and benchmark MSP performance against non-stationary GEV models to assess the added value of jointly modelling spatial dependence and temporal non-stationarity. Results reveal fundamentally different spatial and temporal behaviours of frequent and rare floods. Frequent floods (with a 2-year return period) weaken across much of southern Australia but intensify in the north, reflecting the dominance of local hydrological and topographic controls. Rare floods (with return periods of 25–100 years) exhibit strong spatial heterogeneity and widespread increases along the east coast, southeast, and tropical north, driven primarily by ENSO, IOD, and SAM, which emerge as the strongest modulators of temporal variability. Physiographic gradients, particularly catchment area, elevation, and slope, govern spatial dependence across the continent. A striking north–south divergence in the evolution of flood coherence is uncovered: southern Australia exhibits increasing synchronisation of floods, whereas northern regions show growing spatial fragmentation. Critically, spatio-temporal MSPs capture these dynamic shifts in flood clustering—features that NSGEV models cannot detect—resulting in substantial reductions in uncertainty in rare-flood quantiles, particularly in data-sparse regions. By integrating local catchment attributes and large-scale climate variability into spatial extremes theory, we provide a unified modelling framework that uncovers how Australia’s flood hazard landscape is being structurally reorganised under climate change, offering a new foundation for continent-scale risk assessment, infrastructure planning, and climate-resilient adaptation.