Advancing characterization of rainstorm variabilities for urban flood hazard assessment and mitigation

Abstract

Intensified rainfall extremes coupled with ongoing urbanization are anticipated to increasingly expose populations to urban flood hazards. However, the pace at which flood hazard assessment theories and practices are updated is not keeping up with this trend. Traditional flood hazard assessments and infrastructure design are typically based on regular-shaped and uniform rainfall patterns with stationary frequencies; the mapping of flood hazards largely depends on riverine flood peaks, and the severity is assumed to align with the return level of the designed rainstorms. Yet, the nonstationary frequency of rainfall extremes is prevalent in a warming climate; complex spatiotemporal variabilities (STV) lead to significant discrepancies in the flooding processes compared to idealized designs. Moreover, the intensely local nature of urban floods necessitates dynamic and distributed estimation of inundation hazards. Consequently, high spatiotemporal-resolution rainfall inputs that accurately represent space-time structures are essential. This thesis seeks to enhance flood hazard assessment and mitigation by advancing the characterization of rainstorm variabilities, encompassing both long-term frequency and event-level STV. Using a multi-source merged gridded dataset, this study first evaluated the non-stationarity (NS) of short-duration rainfall extremes in the rapidly developing Greater Bay Area of China. The findings indicate that NS exhibits significant land cover and duration dependencies, with urban areas experiencing a more pronounced intensification of events over short durations and short return periods compared to rural areas. Besides, urban areas display a higher degree of variabilities in NS across time scales and return periods with higher peak scaling rates. Then, this study tracked rainstorms across various watersheds in Hong Kong using radar data and explored the interplay among rainstorm kinematics, heterogeneity, scaling, and event severity through a vine copula-based model. The prevailing STV challenges the equal-probability assumption in the frequency analysis regarding event severity across scales. The study also demonstrated the integration of rainstorm STV into generating a stochastic rainstorm catalog that reflects the realistic STV distribution over a rural-urban watershed. 3000 scenarios were drawn from the catalog to conduct numerical experiments using a full 2D shallow water equations-based inundation model. The intricate joint probability structures of rainstorm severity and STV variables tend to obscure the mechanistic impact of individual factors on flood response, resulting in the inconsistency between the exceedance probabilities of the rainstorm and flood variables, particularly between rainstorm intensity and flood hazard index (with Kendall's tau being 0.22). Notably, significant underestimation of inundation hazards may occur when street-level inundation hazards are represented by watershed-level hazards for the same return period, again underscoring the fallacious outcomes of the equal-probability assumption across scales. Additionally, this thesis assessed the performance of a typical nature-based solution, the flood retention lake (FRL), within the same watershed. The results reveal a pronounced nonlinearity with rainstorm severity, characterized by an L-shaped band of satisfactory effectiveness on the duration-return period diagram. Furthermore, the performance of multiple FRLs in various geographic configurations shows contrasting results from hydrologic and hydrodynamic perspectives, emphasizing the importance of clear objectives in the strategic blueprints prior to optimization.