1. Introduction
Tropical cyclones (TCs), associated with extreme wind, rainfall and storm surge, are responsible for significant property damage and loss of life in coastal areas (Smith and Katz, 2013). Often, TC-induced storm surge together with astronomical high tide have been suggested as the primary causes of episodic flooding in low-lying coastal regions. Recent flood catastrophes illustrate that coastal cities, as the agglomeration of population and assets, are particularly vulnerable to storm tide flooding. For example, the unprecedented flooding of New Orleans due to Hurricane Katrina (2005) caused hundreds of fatalities and more than $40 billion in economic losses, making it the worst natural disaster in U.S. history (Kates et al., 2006). Hurricane Sandy (2012) resulted in the deaths of tens of people due to drowning and at least $20 billion in flood-related damages in New York City (Blake et al., 2013). Coastal flooding is expected to be more frequent and devastating, owing to rapid urbanization, sea level rise (SLR) and TC intensification in a warming climate (Emanuel, 2013; Woodruff et al., 2013; Hallegatte et al., 2013; Lin and Shullman 2017; Vitousek et al., 2017; Marsooli et al. 2019; Knutson et al., 2020). Hence, it is essential to quantify TC-induced flood hazards at city or regional scales.
To ensure an accurate assessment of TC-induced flood hazards for coastal cities, a large number of TC events are required particularly for addressing extreme flood hazards induced by low-probability storms. Due to the lack of sufficient measurements over a long period, statistical methods have been commonly used to generate large samples of synthetic TCs for a specific site (e.g., a city) of interest (Batts et al., 1980; Georgiou et al., 1983). The basic idea of these models is that probability distributions of key TC parameters (e.g., central pressure and radius of maximum wind) are fitted to the historical record, and then a Monte Carlo approach is applied to sample from these distributions. To overcome the limitation of local data, Vickery et al. (2000, 2009) applied the basin-wide data to develop a probabilistic track model with TC intensity along the track estimated based on storm persistent and environmental variables such as the sea surface temperature. Furthermore, Emanuel et al. (2008), Lee et al. (2018), and Jing and Lin (2020) developed more advanced probabilistic track models by incorporating more environmental and oceanic variables that are essential to storm development. The advent of synthetic TC simulation has enabled probabilistic TC hazard assessment for coastal cities.
Recently, synthetic TC simulation has been increasingly used to drive storm tide simulation and assess flood hazards at the city scale. Lin et al. (2010, 2012, 2016) applied the statistical-deterministic model developed by Emanuel et al. (2008) to produce large numbers of synthetic storms for New York City (NYC), and then they used two hydrodynamic models: the Sea, Lake and Overland Surges from Hurricanes (SLOSH) model (Jelesnianski et al., 1992) and the Advanced Circulation (ADCIRC) model (Luettich et al., 1992; Westerink et al., 1994) with a high resolution mesh (up to 10 m around NYC) to simulate storm surges and storm tides induced by synthetic TCs. They estimated the probabilistic distribution of storm tides along NYC coast and found a heavy tail for very low occurrence probability events. Furthermore, the simulated surge heights along the coast from Lin et al. (2012) were interpolated and applied to terrain elevation for determining the flood inundation (extent and depth) and associated probabilities for NYC (Aerts et al., 2013). However, the simple nearest neighbor extrapolation method used in this analysis does not consider the dynamic nature of flood routing and may cause biases in the final inundation maps (Barnard et al., 2019). More recently, Yin et al. (2016) described a new method for coastal flood simulation by coupling ADCIRC and a simplified 2D flood inundation model (FloodMap) and applied the analysis to NYC for the case of Hurricane Sandy. The coupled model demonstrated efficiency and improved prediction over ADCIRC-only simulations for estimating storm-induced flooding in coastal areas.
In this study, an integrated statistical-hydrodynamic approach is proposed by running the statistical-deterministic hurricane model, ADCIRC, and FloodMap in sequence, to derive probabilistic flood hazard maps and to investigate possible worst-case flood scenarios for coastal cities under the current climate conditions as well as projected future SLR scenarios. To our knowledge, this study is the first that couples probabilistic hurricane model, storm surge model, inundation model, coastal protection data, and SLR projections to quantify the inundation hazards (probabilities). The proposed approach is applied to analyze the coastal flood hazard for the city of Shanghai, which is within the Northwest Pacific Basin and is highly exposed to typhoon-induced coastal flooding. For instance, Typhoon Winnie (1997) caused storm tides up to nearly 7 meters above Wusong Datum, the highest recorded in Shanghai’s history. Meanwhile, Shanghai is protected by high-standard flood defense, which needs to be taken into account when estimating the inundation risk and effects of SLR. The remainder of this paper is organized as follows. Section 2 describes materials and methods, including synthetic TC dataset, storm tide simulation, flood probability analysis, and coastal inundation modeling. Section 3 presents the results and discusses the key findings. Section 4 provides the conclusions and offers suggestions for further research.