Assessing cascading flood hazards in a warming climate

Pluvial, fluvial, and coastal floods seriously threaten lives and properties in low-lying and densely populated coastal communities. Since these types of floods may occur simultaneously or successively in coastal watersheds, the compound and cascading impact could be more devastating than either alone. Many studies have examined the compound effects resulting from disparate driving forces (including upstream riverine flow, local rainfall, storm tides, and sea level rise) during a single tropical cyclone landfall or the compound effects of back-to-back tropical cyclones occurring within relatively short time intervals (typically 15 days).

However, other meteorological systems, such as atmospheric rivers, monsoons, and strong convection, can produce extreme rainfall in addition to tropical cyclones. This rainfall can be linked to the preceding or subsequent floods caused by tropical cyclones through the cascading hydrologic effects of catchment soil moisture storage capacity, affecting the onset, intensity, and duration of flood outbreaks over a longer timescale. A recent example of such cascaded flooding was Bangladesh's monsoon floods in July 2020. The monsoon rainfall was an ``ordinary" event. Still, the premature soil saturation caused by cyclone Amphan in May 2020 generated more effective runoff during the monsoon season, leading to historically early and prolonged monsoon flooding. Quantifying cascading flood hazards in a warming climate and providing a hyper-resolved integrated inundation map is critical for decision-makers to take adaptation measures. However, the absence of fast, high-performance downscaling models that simultaneously handle tropical cyclones and seasonal weather extremes remains a significant challenge. Furthermore, the cascading effect of soil water storage capacity on floods at varying time scales has been underestimated, resulting in an inaccurate flood hazard (and risk) assessment.

To bridge this gap, here, for the first time, we propose an integrated downscaling-hydrodynamic inundation framework that seamlessly drives a fast inundation model using a new approach to generate temporally-cascaded synthetic stochastic timelines of downscaled high-resolved seasonal rainfall extremes and cyclone-induced compound events across multiple global warming scenarios. Our framework couples simplified physics with ensemble-approximated Gaussian Processes in an adversarial learning approach with bias correction to rapidly downscale rainfall, uses a fast statistical-hydrodynamical approach for flood simulation, and actively learns surrogate models to identify flood extremes with high precision. That enables improved probabilistic inundation maps to support flood hazard assessment in a warming climate. We selected Florida in the United States and Bangladesh in South Asia as illustrative examples to validate our modeling framework. It is important to note that our modeling toolkit applies to any place in the world to assess flood risk at the regional scale for the benefit of local decision-makers in taking adaptation measures under climate change.