Urban Flood Modeling: Uncertainty Quantification and Physics‐Informed Gaussian Processes Regression Forecasting

Abstract

Multiphysics urban flood models are commonly used for urban infrastructure planning, risk management, and forecasting (Hemmati et al., 2020;Qi et al., 2021;Rosenzweig et al., 2021). Because these models have many uncertain parameters and rely on assumptions (e.g., the effect of the groundwater flow on urban flooding), it is important to quantify uncertainty in the model-based predictions and forecasts.In recent years, several continent-scale and regional flood modeling approaches were introduced that are based on one-and two-dimensional surface hydraulics equations (Merwade et al., 2008). Flood modeling in urban areas is more challenging because urban systems are generally complex, and their models require a detailed representation of multiple components. Recent advances in computational capabilities have enabled hyper-resolution urban flood predictions using integrated flood models (Saksena et al., 2019). Some examples of these models are the Interconnected Channel and Pond Routing (ICPR), ParFlow, and GSSGA models (Downer et al., 2006;Kollet & Maxwell, 2006; Streamline-Technologies, 2018). The integrated models have the capability to simultaneously calculate urban watershed processes, including channel flow, overland flow, unsaturated flow, groundwater flow, and stormwater flow. Each of these individual processes is simulated using an appropriate set of equations (and the associated parameters) and their numerical approximations. The unknown parameters are a source of parametric uncertainty. The approximations and omitted physical processes result in model uncertainty (Saksena et al., 2020(Saksena et al., , 2021.The effect of the model and parametric uncertainties must be understood, characterized, and quantified to make actionable risk-based decisions in response to intense rainfall events in urban areas. In this work, we investigate the impact of both parametric and model uncertainties on the urban flood predictions of the ICPR model.

Quantifying cascading uncertainty in compound flood modeling with linked process-based and machine learning models

“…(i) initial condition, (ii) forcing (or boundary) conditions, (iii) model parameters, and (iv) model structure (Beven et al, 2005;Moradkhani et al, 2018;Vrugt, 2016). Initial and forcing conditions are essentially model inputs to any process-based models, however, their isolated effects on WL dynamics are often analyzed separately as reported in diverse hydrological (Abbaszadeh et al, 2019;Jafarzadegan et al, 2021a;Kohanpur et al, 2023) and coastal studies (Bakhtyar et al, 2020;Marsooli and Wang, 2020;Muñoz et al, 2022a). The first source of uncertainty involves inaccuracies in the geometry of the system, which is spatially represented with light detection and ranging (LiDAR) elevation data.…”