Understanding how urban environments interact with hydrometeorological extremes is becoming increasingly important as cities face growing risks from extreme precipitation, flooding, drought, and compound events. The characteristics of these extremes are shaped by interactions among local atmospheric and hydrological processes, land-surface properties, and urban form. Yet these interactions remain difficult to generalise across cities, climates, and spatial and temporal scales. This complexity reflects the high heterogeneity of urban environments and the multiple physical pathways through which urbanisation can influence hydrometeorological extremes. Addressing this challenge requires approaches that can disentangle nonlinear and scale-dependent relationships while retaining physical interpretability.
Machine learning provides opportunities to identify dominant drivers, characterise nonlinear relationships, and reveal spatially and climatically varying responses that are difficult to isolate based on conventional approaches alone. In particular, explainable machine learning (XAI), physics-informed machine learning, causal modelling and hybrid modelling can further support the transition from predictive performance to robust interpretation, hypothesis testing, and process understanding.
This session welcomes studies using machine learning and related data-driven approaches to advance understanding of hydrometeorological extremes in complex urban environments. We particularly welcome studies that use machine learning for physical interpretation, hypothesis testing, and process understanding. Key topics of discussion include:
• Identifying the drivers and nonlinear interactions shaping extreme precipitation, flooding, drought, and compound events.
• Investigating how urban form, land-surface properties, and infrastructure interact with atmospheric and hydrological processes across spatial and temporal scales.
• Applying explainable AI (XAI), physics-informed machine learning, causal modelling, and hybrid approaches to support physical interpretation and process understanding.
• Integrating remote sensing and other multi-source observations to characterise heterogeneous patterns and their scale dependence.