1. INTRODUCTION
Flooding, as a natural hazard, has been increasingly threatening human lives and economies (Gourley et al., 2017; Hirabayashi et al., 2013; Li et al., 2021a). In the United States, most billion-dollar natural hazards are tied to either local or regional flooding, making it the major cost to human society. Unfortunately, under a warming climate with anthropogenic pressure, flood crises are likely to continue expanding, as the flood frequency accelerates and flood magnitude rises (Bates et al., 2021; Hirabayashi et al., 2013; Tabari, 2020; Triet et al., 2020; Swain et al., 2020; Viero et al., 2019;). To combat flood risks, researchers have been developing hydrologic/hydraulic models to deliver accurate and timely flood information for local communities and decision-makers (Gourley et al., 2017). In the United States, two pronounced flood forecasting systems – the National Water Model (NWM) and FLASH – are capable of both simulating real-time floods and forecasting floods in a short range (Cohen, Praskievicz, & Maidment, 2018; Gourley et al., 2017; Viterbo et al., 2020; Yussouf et al., 2020).
These large-scale flood monitoring systems, although claimed to offer inundation maps and predictions, weaken their hydrodynamic simulation due to computational requirements. For instance, the NWM adopts the Height Above Nearest Drainage (HAND) method to produce flood inundation maps along the river channels by mapping discharge to stage via rating curves (Johnson, Munasinghe, Eyelade, & Cohen, 2019). This conceptual method, however, overlooks the physics of floodwater propagation because of no flow dynamics being represented (Wing et al., 2017). Moreover, it cannot simulate the pluvial flood, which is a local effect caused by intense rainfall rates and does not normally occur along river channels (Bates et al., 2021). More recently, some emerging hydrodynamic models have been successfully deployed and evaluated at continental or global scales (Bates et al., 2021; Grimaldi et al., 2019; Sampson et al., 2015; Wing et al., 2017; Yamazaki, Kanae, Kim, & Oki, 2011). These models simplify the full Shallow Water Equation (SWE) to speed up the flood simulation. Nevertheless, they normally do not represent the hydrologic process well, especially for the infiltration process, which is proven to be critical in flood simulations (Li et al., 2021b; Ni et al., 2020). As such, a coupled physically-based hydrologic & hydraulic (H&H) model appears to be a better choice, which takes the complementary advantages for accurate flood modeling (Dullo et al., 2021; Felder, Zischg, & Weingartner, 2017; Kim et al., 2012; Nguyen et al., 2016; Pontes et al., 2017; Sebastian et al., 2021). Readers are referred to Teng et al. (2017) and Grimaldi et al. (2019) for a detailed review of coupled models. Most of such models, however, adopt one-way and weak coupling, meaning that there is no interplay between the hydrologic component and hydraulic component (Bravo, Allasia, Paz, & Collischonn, 2013). They normally produce surface runoff outputs first to drive the hydraulic model. The recent development of the Coupled Routing and Excess STorage inundation MApping and Prediction (CREST-iMAP) version 1.0 also falls into this category, although it is online coupled (Chen et al., 2021; Li et al., 2021b).
Two-way coupling for the H&H models has not hitherto been well-recognized. The accumulated surface water (hydraulic feature), along with excess surface runoff during a flood event in principle would alter infiltration rates, whereby both the flood magnitudes and timings could differ. Therefore, we activate the surface water infiltration along its way to downslope, which is called run-on infiltration or re-infiltration in short (Smith & Hebbert, 1979; Nahar, Govindaraju, Corradini, & Morbidelli, 2004; Zhang, Lin, Gao, & Fang, 2020). Nahar et al. (2004) defined this re-infiltration as the infiltration of surface water that, as it moves downslope, encounters areas where moisture deficit has not yet been satisfied. It is often ignored in rainfall-runoff studies, while it can be significant when the random nature of infiltration properties is taken into account (Corradini, Morbidelli, & Melone, 1998; Nahar et al., 2004). Smith and Hebbert (1979) simulated the run-on process with varying saturated hydraulic conductivity and rainfall rates, and they reported that the effect of the run-on process is to decrease the ponding time dramatically. Corradini et al. (2002) compared models with and without re-infiltration, and they suggested that re-infiltration greatly reduces surface flow and alters both rising and recession limbs of the hydrograph. Nahar et al. (2004) emphasized the influence of re-infiltration in hillslope hydrograph using the Green-Ampt model with a 1D kinematic wave surface routing. A recent study by Zhang et al. (2020) takes it one step further, in which they applied the community model WRF-Hydro (Weather Research Forecasting model-Hydrological modeling system) to explore the influence of rainfall rates, topography, soil types on the re-infiltration process. However, none of these studies have considered the implication of re-infiltration to hydrodynamic studies, where the overland flow is driven by the 2D Shallow Water Equation (SWE) instead of 1D routing. To do so, we can obtain a more realistic view of how re-infiltration plays a role in flood simulations.
During extreme flood events, the infiltration process is often disregarded because the infiltration rates are relatively low compared to excess rainfall rates. Yet, some studies claim that the infiltration process is critical to determine flood wave propagation, such as arrival and dissipation, especially in flat plain or regions with highly permeable soil media (Corradini et al., 2002; Mahapartra, et al., 2020; Nahar et al., 2004; Li et al., 2021b; Woolhiser, Smith, & Giraldez, 1996). A hydrodynamic model without infiltration is likely to overestimate flood depth (Kim et al., 2012; Li et al., 2021b; Ni et al., 2020). Nevertheless, it still remains unclear, or at least not as clear as infiltration, whether re-infiltration is essential for H&H models in extreme flood events. In other words, whether it is worth encapsulating such a scheme in modern flood simulation frameworks. To our knowledge, few studies have attempted to answer this question under the context of extreme flood events. Moreover, further questions can arise as to what the determining factors are during such a process and how it interacts with both flood magnitude and dynamics. In light of these questions, the objectives of this study are to explore 1) the effectiveness and importance of the re-infiltration scheme to an H&H model, 2) the contributing factors to the differences between with and without re-infiltration, and 3) whether and to what extent the re-infiltration process can help improve flood inundation mapping and prediction of extreme events. We first test its effectiveness and importance on a 100-year design extreme rainfall event during a sensitivity test and then apply it to a real case study – Hurricane Harvey – to validate the efficacy. It is anticipated to provide insightful information for model developers and researchers to understand the importance of the re-infiltration process to flood modeling. In this study, we also release our latest development of CREST-iMAP V1.1, which features a two-way coupling and re-infiltration scheme on top of the previous version (Chen et al., 2021; Li et al., 2021b).
The rest of this paper is structured as follows. Section 2 introduces the study area and necessary datasets for the model setup, followed by experimental designs. Section 3 presents the results from the sensitivity test and the Hurricane Harvey event. Section 4 discusses limitations of this study as well as recommendations for input data and future model development. At last, Section 4 concludes the main findings of this study.