Multi-Modal Monitoring and Modelling of Extreme Hydro-Geomorphological Events: Bridging the Gap Between Local Dynamics and Catchment-Scale Predictions

The central challenge in understanding extreme hydro-geomorphologic events is the persistent lack of integrated, quantitative observations capable of developing and constraining predictive models. While flash floods and associated sediment transport represent an escalating hazard under climate change, their underlying dynamics remain poorly understood across the spatio-temporal scales required for effective risk mitigation. Existing monitoring is often fragmented, with upcoming novel approaches only partly resolving key unknowns when used in isolation. For instance, optical methods such as UAV-based photogrammetry and camera gauges provide high resolution surface process data but cannot resolve subsurface bedload dynamics, whereas environmental seismic methods capture particle-riverbed interactions and signatures of turbulence but produce indirect, composite signals that are difficult to isolate and quantify.

To bridge this gap, we envision a multi-modal approach that moves beyond those single-technique or single-sensor proxies. To reliably and robustly observe temporarily evolving interlinked key parameters, i.e., water level, flow velocity, and hydraulic geometry, major steps involve using stereo-vision for precise scaling and channel cross-section updates, alongside AI-based optical flow for complex velocity fields. By integrating low-cost, event-triggered sensors (e.g., thermal & multispectral cameras, seismometers, and LiDAR), we can automate the retrieval of discharge as well as additional parameters such as turbidity and granulometry. Using photogrammetric change detection and AI-driven image processing we can further bridge terrestrial and aerial perspectives (e.g., from UAV), moving toward a physically consistent characterization of extreme events. By integrating high-resolution 3D imaging and seismic data inversion, it becomes possible to capture water and sediment dynamics simultaneously, resulting in unique complementary information on the same event.

In this framework, laboratory experiments provide the necessary controlled conditions to infer the capabilities, caveats and calibration measures for this sensor integration. Highly resolved computational fluid dynamics multiphase flow modelling will generate synthetic reference datasets to disentangle environmental signals and sensor noise. These heterogeneous data streams are integrated via AI-based fusion and uncertainty modelling to resolve non-linear relationships governing coupled water–sediment dynamics. Ultimately, hydrological and hydraulic modelling serves as a testbed for upscaling, in which models are informed by improved process knowledge-based observation data and its uncertainty to evaluate how small-scale insights alter catchment-scale predictions.

From this framework, significant gaps emerge that define the current research frontier. A critical unresolved challenge is the systematic separation of source terms from the superimposed signals generated by the actively evolving sediment-carrying river during flood events. Furthermore, the transition from "data-rich" local observations to “data-poor” but "process-informed" regional models is still hindered by the lack of scalable frameworks that can maintain physical consistency across different scales, i.e., climatic and geomorphological regimes. Addressing these gaps requires a coordinated shift from observing isolated parameters to an integrative, physics-based monitoring loop that can provide truly scalable, model-ready information for extreme events.