Extreme water levels (EWLs) resulting from cyclones pose significant flood hazards and risks to coastal communities and interconnected ecosystems. To date, physically-based models have enabled accurate prediction of EWLs despite their inherent high computational cost. However, the applicability of these models is limited to data-rich sites with diverse characteristics. The dependence on high quality spatiotemporal data, which is often computationally expensive, hinders the applicability of these models to regions of either limited or data-scarce conditions. To address this challenge, we present a Long Short-Term Memory (LSTM) network framework to predict the evolution of EWLs beyond site-specific training stations. The framework, named LSTM-Station Approximated Models (LSTM-SAM), consists of a collection of bidirectional LSTM models enhanced with a custom attention mechanism layer embedded in the architecture. LSTM-SAM incorporates a transfer learning approach applicable to target (tide-gage) stations along the U.S. Atlantic Coast. Importantly, LSTM-SAM helps analyze: (i) the underlying limitations associated with transfer learning, (ii) evaluate EWL predictions beyond training domains, and (iii) capture the evolution of EWL caused by tropical and extratropical cyclones. The framework demonstrates satisfactory performance with “transferable” models achieving Kling-Gupta Efficiency (KGE), Nash-Sutcliffe Efficiency (NSE), and Root-Mean Square Error (RMSE) ranging from 0.78 to 0.92, 0.90 to 0.97, and 0.09 m to 0.18 m at the target stations, respectively. We show that LSTM-SAM can accurately predict not only EWLs but also their evolution over time, i.e., onset, peak, and dissipation, which could assist in operational flood forecasting in regions with limited resources to set up physically-based models.

Several studies have focused on the importance of river bathymetry (channel geometry) in hydrodynamic routing along individual reaches. However, its effect on other watershed processes such as infiltration and surface water (SW) – groundwater (GW) interactions has not been explored across large river networks. Surface and subsurface processes are interdependent, therefore, errors due to inaccurate representation of one watershed process can cascade across other hydraulic or hydrologic processes. This study hypothesizes that accurate bathymetric representation is not only essential for simulating channel hydrodynamics but also affects subsurface processes by impacting SW-GW interactions. Moreover, quantifying the effect of bathymetry on surface and subsurface hydrological processes across a river network can facilitate an improved understanding of how bathymetric characteristics affect these processes across large spatial domains. The study tests this hypothesis by developing physically-based distributed models capable of bidirectional coupling (SW-GW) with four configurations with progressively reduced levels of bathymetric representation. A comparison of hydrologic and hydrodynamic outputs shows that changes in channel geometry across the four configurations has a considerable effect on infiltration, lateral seepage, and location of water table across the entire river network. For example, when using bathymetry with inaccurate channel conveyance capacity but accurate channel depth, peak lateral seepage rate exhibited 58% error. The results from this study provide insights into the level of bathymetric detail required for accurately simulating flooding-related physical processes while also highlighting potential issues with ignoring bathymetry across lower order streams such as spurious backwater flow, inaccurate water table elevations, and incorrect inundation extents.

Multiphysics urban flood models are commonly used for urban infrastructure development planning and evaluating risk due to climate change and sea level rise. However, these integrated flood models rely on several parameters that are hard to measure directly, and the resulting uncertainty in model prediction needs to be quantified, often without observable data. As a part of the Urban Flooding Open Knowledge Network (UFOKN) project, in this study we quantify parametric uncertainty in urban flood models. UFOKN incorporates flood model predictions in combination with machine learning, data and computer science, epidemiology, socioeconomics, and transportation and electrical engineering to minimize economic and human losses from future urban flooding in the United States. As a case study, we choose the Interconnected Channel and Pond Routing (ICPR) numerical model to simulate flooding in the city of Minneapolis in response to the design storms (e.g., 100-year rainfall). Through a sensitivity study, we reduce the number of uncertain model parameters to the Manning’s roughness coefficient and vertical hydraulic conductivity of soil, and construct the distributions of these parameters using open databases. We employ the multilevel Monte Carlo (MLMC) method that combines a small number of high-resolution ICPR simulations with a larger number of low-resolution simulations to reduce the computational cost of computing the key statistics of the quantities of interest describing the urban flooding. Our results show that the uncertainty in the flood predictions (as described by the coefficient of variation of the flood water depth) is distributed highly non-uniformly in the urban area with the coefficient of variation exceeding 0.5 limited to a relatively few computational elements in the ICPR model. Our results demonstrate that urban flood models such as ICPR can provide reliable flood predictions and can be used for a targeted data acquisition to further reduce the parametric uncertainty.