Quantifying the potential exposure of property to damages associated with storm surges, extreme weather and hurricanes is fundamental to developing frameworks that can be used to conceive and implement mitigation plans as well as support urban development that accounts for such events. In this study, we aim at quantifying the total value and area of properties exposed to the flooding associated with Hurricane Florence that occurred in September 2018. To this aim, we implement an approach for the identification of affected areas by generating a map of the maximum flood
extent obtained from a combination of the flood extent produced by the
Federal Emergency Management Agency's (FEMA's) water marks with those obtained from spaceborne radar remote-sensing data. The use of radar in the creation of the flood extent allows for those properties commonly missed by FEMA's interpolation methods, especially from pluvial or non-fluvial sources, and can be used in more accurately estimating the exposure and market value of properties to event-specific flooding. Lastly, we study and quantify how the urban development over the past decades in the regions flooded by Hurricane Florence might have impacted the exposure of properties to present-day storms and floods. This approach is conceptually similar to what experts are addressing as the “expanding bull's eye effect”, in which “targets” of geophysical hazards, such as people and their built environments, enlarge as populations grow and spread. Our results indicate that the total value of property exposed to flooding during Hurricane Florence was USD 52 billion (in 2018 USD), with this value increasing from
USD
The projected rise in sea level, increased floods and storm surge, and
associated consequences over the 21st century has the potential to do
immense economic harm. The economic impact is particularly worrisome in the US because the most valuable real estate, densest communities and most
productive economic engines are situated disproportionately in coastal regions (Fu et al., 2016; NOAA, 2013; Kildow et al., 2014). Recent research has highlighted an ongoing economic signal associated with high-probability flooding events and real estate transactions in coastal communities that can be observed with historical data (see McAlpine and Porter, 2018; Keenan et al., 2018; and Bernstein et al., 2019), suggesting that sea level rise (SLR) already produces negative economic consequences on coastal communities. Furthermore, there is abundant evidence indicating that we are only seeing the first signs of a much more problematic issue both in terms of the flooding scale and the magnitude of associated economic losses (see Fu et al., 2016; Hallegatte et al., 2011; Bin et al., 2008, 2011; Parsons and Powell, 2001; Michael, 2007). In this regard, a SLR of
Recent events such as hurricanes Katrina, Irma and Florence have
highlighted the issues related to floods and extreme events even more. In
particular, Florence was one of the most devastating hurricanes in history,
as it combined storm surge, strong winds and extreme precipitation. It began
as a tropical storm on 1 September 2018 over the islands of Cabo Verde off the
coast of western Africa and peaked as a Category 4 hurricane with winds up to
225 km h
The human cost of Hurricane Florence was a reminder of the power of such storms, and these storms are likely becoming more impactful, as their surge reaches further inland due to changing tracks, increased strength and rising seas. The increasing exposure of the public and properties to events similar to Hurricane Florence has unintended consequences of raising the awareness and concern to all types of climate-related events (Borenstien and Fingerhut, 2019). Such is likely the case in much of the recent research on real estate market responses to higher-probability flooding associated with nuisance tidal flooding events (McAlpine and Porter, 2018). In their study, MacAlpine and Porter (2018) found that properties in Miami-Dade County at risk of frequent tidal flooding had lost over USD 430 million in property value relative to homes that were not a risk of repeated tidal flooding events. Likewise, and also centered in the Miami-Dade region, Keenan et al. (2018) found that homes at lower elevations were being penalized on the market relative to homes at higher elevations. Moreover, in another analysis, Bernstein et al. (2019) found a similar penalty for homes at risk of flooding from an increase in the SLR but found that this penalty was primarily driven by investors and uneven access to information associated with risk. All of these studies identify an increase in awareness of SLR-related flooding events, and all document the fact that this trend is relatively new (since about the middle of the last decade). Of particular importance to the recent market response is the fact that increased probability is an important driving force. In the work undertaken by Bernstein et al. (2019), for example, the price penalty for homes at risk of flooding is explicitly driven by the sophistication of investors and their access to risk tools aimed at helping them to make decisions about property value as well as long-term appreciation over time. McAlpine and Porter (2018) also found, in this regard, that risk associated with being impacted by a Category 1 hurricane is correlated with potential loss property value but not the probability of being impacted by a higher-category storm. In each of these cases, the research suggests that the real estate market is becoming more sensitive to the probability of damage associated with inundation from flooding events due to rising seas, storm surges, nuisance flooding and consequences of a changing climate. On the other hand, research out of the University of Pennsylvania's Wharton Risk Center by Kunthreuther et al. (2018) found that the elasticity concerning the housing market tends to show quick recoveries in areas where the experience of climate catastrophes is characterized as a market shock. Market shocks are generally thought of as one-time (or contiguous time period) events that negatively impact the housing market. Due to the nature of market shocks being lower probability and harder to predict, the housing market tends to see them as unlikely and related to collective internalizations associated with myopia, amnesia, optimism, inertia, simplification and herding (Kunthreuther et al., 2018). However, market stressors are ubiquitous, high-probability events that are generally predictable and have historical certainty. In the context in which we are working, increased and unmanageable tidal flooding could be considered a market stressor, while the impact of a single hurricane event could constitute a market shock. Historically, market shocks (such as hurricanes) are much more expensive, in terms of actual economic impacts, and consume more media attention, in terms of the coverage of the events.
Several studies have recently focused on assessing damages from Hurricane
Florence. Roberson et al. (2019) use overhead imagery, including synthetic-aperture radar (SAR) and optical data, to study the impact of Florence on
livestock wastewater and crop health. Srikanto et al. (2019) study the spatial distribution of fatalities and associated demographics, indicating
that 93 % of the affected buildings were residential structures. The proper quantification of the impact of Hurricane Florence (or more in general of extreme events) is helpful not only for addressing the recovery of the communities impacted by the event but also for providing tools to policymakers, urban planners and city managers that will ultimately guide them through the decision process of reducing the impacts of future events. If it is true, indeed, that climate change is and will be influencing the frequency and strength of storms and floods, it is also true that the exposure of anthropogenic structures and lives is a function of urbanization factors such as, for example, the building of new properties in proximity to the coast and to water bodies. In this context, it becomes crucial to understand and quantify how urban development has impacted the exposure of properties and population to present-day storms and floods. For example, one of the most devastating hurricanes over the same region before Florence was Hurricane Hugo, reaching the Carolinas on 10 September 1989, with winds up to 260 km h
We aim at understanding the usefulness of remotely sensed satellite data as a method for the identification of impacted areas and for delineating the maximum flood extent. Specifically, we report results concerning the mapping of the flood extent associated with Hurricane Florence estimated from SAR data and compare such extent with the maximum flood extent provided by the Federal Emergency Management Agency (FEMA). From that exposure, we are able to quantify the property value and total area exposed to Hurricane Florence by combining the flood extent coverage with a database containing publicly available property value attributes. Despite recent studies that have started to focus on the spatio-temporal variability in property values and human settlements in hurricane-prone areas (e.g., Huang et al., 2019) and on the market responses to increases in observed flooding events (e.g., McAlpine and Porter, 2019; Keenan et al., 2018), no study, to our knowledge, has focused on the impact of urban growth on the property exposed to Hurricane Florence. Addressing this point is crucial to account for those impacts related to the choices that our society makes to continue the expansion of urban areas and that have been addressed by experts as the expanding bull's eye effect (Ashley and Strader, 2016), in which targets of geophysical hazards, such as people and their built environments, enlarge as populations grow and spread. The term bull's eye is used here to define the eye or center of a storm. Our approach is complementary to the approaches focusing on the bull's eye effect and to those calculating the impact of floods under future climate scenarios (e.g., sea level rise or storm surge is changing but the property distribution remains the same). Specifically, we assess the potential exposure of properties using a dataset containing, among other information, when each property was built and use that information to estimate the potential exposure of buildings to Hurricane Florence that would have occurred in the past decades.
From an observational point of view, spaceborne and airborne remote sensing
(e.g., Schumann et al., 2011), as well as UAV-based approaches (e.g.,
Gebrehiwot et al., 2019), offer powerful tools to monitor flood extent (e.g.,
Domeneghetti et al., 2019; Schumann et al., 2018; Schumann, 2018; Giordan et al., 2018). Optical data can map the presence of surface water at relatively high spatial resolution and accuracy, but it is limited by the presence of clouds (Schumann et al., 2018). On the other hand, datasets collected in the microwave region, such as those collected by synthetic-aperture radar (SAR), are not limited by the presence of clouds (Schumann et al., 2018; Manavalan, 2017; Huang et al., 2018). The recent launch of Sentinel-1 ESA sensors in September 2014 (Sentinel-1A) and April 2016 (Sentinel-1B;
We supplement the radar-derived flood extent with FEMA's high-water-mark-based depth grids and inundation polygons from observed and collected
Hurricane Florence data. High-water marks (HWMs) are point data collected
using high-resolution real-time kinematic (RTK) GPS systems or other methods. HWM points represent the highest extent of riverine flood or coastal storm surge inundation. The raw data are available at the FEMA Natural Hazard Risk Assessment Program (NHRAP) site and were downloaded for all basins available per FEMA's collection efforts following the hurricane event (
The FEMA maximum water extent is distributed as a GIS raster file created to
represent the extent of riverine or coastal storm inundation following larger flooding events. The file is created as a derived product following the creation of the maximum depth grid raster file, which is obtained using FEMA HWM data and FEMA's Digital Flood Insurance Rate Map (DRIRM) Base Flood Elevations (lidar-based elevation data). Using those datasets, a grid is
obtained to estimate the height of water at any given point between HWMs based on base elevation. From this, we extracted a secondary file measuring only the extent of inundation from the storm surge. The FEMA dataset is distributed as an ArcGIS® geodatabase (.gdb format), and we rasterized it at a spatial resolution of 10 m to match the spatial resolution of the SAR data. More information on the FEMA approach for
estimating maximum flood extent can be found at
Property value data are compiled from each individual property's county assessor in the form of the property tax assessed value. The data were obtained from a third-party provider, ATTOM™ Data Solutions, which provides high-quality parcel-level information on all properties in the
United States and in a value-added format (
Inundated areas (including permanent water bodies) obtained from Sentinel-1 data and FEMA are reported, respectively, as blue (radar) and red (FEMA) regions in Fig. 1. We used a total of 12 Sentinel-1 images collected between 14 and 19 September 2018 and whose footprints are shown in the inset in the top left corner of Fig. 1. Specific names and acquisition times of the radar images are reported in the Supplement. We used the 12 images in order to maximize the covered area and to account for the temporal evolution of surface water after the landfall of Hurricane Florence associated with heavy, persistent rainfall.
Map of inundated areas estimated by FEMA (red) and by the Sentinel-1 radar images (blue). The inset in the top left corner shows the footprint of the several radar images to create the composite water extent map. Acquisition times and other details concerning the radar images are available in the Supplement.
The comparison between the maximum water extent estimated by FEMA and the
water extent mask obtained from Sentinel-1 indicates a matching score (defined here as the percentage of flooded pixels identified by Sentinel-1
with respect to the total number of flooded pixels identified by FEMA) of
11.3 % and a commission error (defined as the relative percentage of pixels when Sentinel-1 detects flooded areas but FEMA does not with respect to the total number of FEMA flooded pixels) of 9.2 %. We note here that the FEMA map is based on a combination of modeled and measured quantities and might miss flooding associated with heavy rains, as in the case of Hurricane Florence. Consequently, it is possible that some areas that were flooded according to the radar images were not included in the FEMA maps. As an example, Fig. 2 shows the maximum water extent from FEMA (red) together with the one derived from Sentinel-1 data (blue) nearby the town of Bennettsville, South Carolina (34.6174
Map of inundated areas estimated by FEMA (red) and by Sentinel-1 (blue) near the town of Bennettsville, South Carolina (34.6174
Time series of
Another factor complicating the comparison between Sentinel-1 and FEMA
inundated regions regards the acquisition time of the radar images, which are collected before or after the time of the maximum water extent. Figure 3a shows the time series of the water height (mean sea level in meters) for the ocean tide gauge located in Wrightsville Beach, North Carolina (ID no. 8658163), where Hurricane Florence made landfall. Maximum water height was reached on the same day at around 15:00 UTC. The image also shows the acquisition time of
the Sentinel-1B (14 September 2018, 11:15:05 UTC) and Sentinel-1A (14 September 2018, 23:05:48 UTC) as vertical, dashed lines, indicating that
such images were, indeed, acquired before and after the time when the water
reached the maximum extent. River gauge data also show that, because of the
heavy precipitation, the maximum water discharge and gauge heights inland
occurred a few days after Hurricane Florence made landfall. In this regard,
Fig. 3b and c show, respectively, the daily discharge (in m
Flooded areas detected by
Given these considerations, for this study we merge the FEMA and Sentinel-1 flood extent maps to generate a maximum composite flood extent map that will be used to assess the property exposure to Hurricane Florence flooding. We will refer to this dataset simply as the “maximum flood extent” in the remaining sections of the paper.
Figure 5 shows the spatial distribution of the properties within our database overlaid with an image of the eye of Hurricane Florence when it made landfall. Our analysis indicates that the total area of properties affected by the maximum flood extent water was 70 964 700 m
Distribution of properties within our database used to estimate the exposed-property damage from Hurricane Florence. An image of the Hurricane Florence making landfall is also reported as a reference. ©Hurricane image courtesy: Cyclocane.
The total exposed-property value estimated using the maximum flood extent is
USD 52 079 520 584 (2018 USD, corresponding to
In Fig. 6 we report the distribution of the number of properties exposed to flooding as a function of the corresponding property value. A power law function (as reported in Eq. 1),
Distribution of the number of properties exposed to flooding as a function of property value. Dashed line represents the power law curve fitting the distribution. The parameters of the fitting power law function are reported in the top right section of the figure.
Number of properties as a function of distance from water bodies. Dashed line represents the power law curve fitting the distribution. The parameters of the fitting power law function are also reported in the top right section of the figure.
Spatial distribution of the properties within our database that
were built during the
Distance from water bodies, especially coastal and riverine bodies, is also a useful indicator of properties vulnerability and potential exposure in hurricane-prone areas. Consequently, we expanded our analysis to consider
the distance of the properties that were flooded during Florence within our
database from permanent water bodies (Fig. 7). Values along the
Addressing this point is crucial to account for those impacts related to urban growth and the expansion of urban areas as addressed by experts when considering the so-called expanding bull's eye effect (Ashley and Strader, 2016), in which targets of geophysical hazards, such as people and their built environments, enlarge as populations grow and spread. In this case, the bull's eye expansion does not refer to the increased storm size but rather to the area where the impact of the geophysical hazard occurs, expanding because of the urbanization process over the past decades. This concept is well synthesized in what has been named the expanding bull's eye effect (Ashley and Strader, 2016), arguing that targets – people and their built environments – of geophysical hazards enlarge as populations grow and spread. We point out that, to demonstrate their bull's eye effect, Ashley and Strader (2016) work with a semi-empirical spatio-temporal model of housing stock in tornado zones over time. A major difference between this study and that of Ashley and Strader (2016) is that in our case we have a snapshot of stock rather than a continuous record through time (i.e., annual records of all properties). This means that our dataset might be skewed toward newer properties as old buildings get replaced and that, despite the bull's eye expansion effect being evident from space, our property dataset might only capture it indirectly. With this in mind, we calculated what would have been the property area and values exposed to the Florence flood that would have occurred 10, 50 or 100 years ago by using the information contained within our database on the years when properties were built. For the purpose of this analysis, we clarify that we are assuming the same sea levels and topography of today.
Figure 8 shows the spatial distribution of the properties within our database that were built during the (a) 1800–1900, (b) 1900–1950, (c) 1950–2000 and (d) 2000–2018 periods. We considered the first period to be a 100-year one (Fig. 8a) because of the relatively small number of properties that were built then. Most of the urban growth between 1900 and 1950 (Fig. 8b) occurred inland and along the coast north of Wilmington, with a relatively small number of new properties built close to water bodies (either rivers or ocean). An explosion in new properties occurred between 1950 and 2000 (Fig. 8c), likely as a consequence of the economic stimulus following World War II. The period 2000–2018 shows a relatively smaller number of new properties with respect to the previous periods (Fig. 8d). In this regard, our analysis performed on the 10-year period of number of properties built within our database (Fig. 9) shows that before 2010 the number of houses built had been increasing exponentially (
Number of properties (in thousands) built within our data record during different decades (red bars; left axis) and relative change between two consecutive periods (blue line; right axis). Note that the number of properties built between 1800 and 1900 are aggregated as a single value because of the small number of properties built during that period.
Figure 10 shows the time series of total value of exposed property (in 2018
billion USD). The inset reports the relative change of the exposed area and value between two consecutive time steps (10 years). Consistent with the results discussed above, a relatively small increase in the exposed-property value occurs before the 1940s (from USD
Time series of total value of exposed buildings (in 2018 USD) to the maximum flooded extent region between 1800 and 2018. The inset shows the relative change of the exposed area and value between two consecutive time steps (10 years).
As mentioned, distance from permanent water bodies can play a critical role
in terms of exposure, with flooding due to Hurricane Florence reaching
properties that were up to
Distribution of properties (red areas) built
Total value within our database of properties exposed to flooding as a function of distance from water for the different periods reported in the inset.
Increased flooding associated with sea level rise, storm surges and other extreme events has the potential to economically disrupt many areas around the world, with the most valuable real estate, densest communities and most productive economic engines situated in coastal regions. The specific goal of our study was to quantify the exposure of properties to the flooding associated with Hurricane Florence that hit the Carolinas in September 2018 and to study how the spatio-temporal evolution of new built properties along the most recent decades has impacted the property exposure. It is important to note that much of the vulnerability associated with building development in these areas should be considered independent of climate change to this point. However, moving forward, these types of storms are expected to increase in intensity and the link between climate change and potential exposure is likely to be tied more closely together. In fact, we are already seeing these trends as they relate to tidal flooding events, and one might expect that the low-probability, larger storms are likely to become more linked to our changing climate as well.
In order to properly quantify the exposure of properties to Florence flooding, we developed a maximum flood extent map from the combination of the FEMA maximum extent map (generated through the merging of high-water marks) and the flooded areas detected by means of spaceborne radar data acquired by the ESA Sentinel-1 sensors. We found that the total value of property exposed to flooding was USD
Our work not only provide new insights for policymakers and city planners but also provides a tool to better estimate how the property market will respond to future disasters. Recent work (e.g., McAlpine and Porter, 2018; Keenan et al., 2018) has found that homes at lower elevations were being penalized on the market relative to homes at higher elevations and that houses exposed to sea level rise sell for approximately 7 % less than observably equivalent unexposed properties equidistant from the beach (Bernstein et al., 2019). For our future work, we plan to expand our analysis to other modern-day (e.g., Irma, Michael, Katrina and Sandy) and historical (e.g., Hugo in 1989) hurricanes to address similar questions to those addressed in this study. Moreover, we plan to improve the detection of maximum flood extent through the implementation of machine-learning techniques combining radar maps with tide gauge interpolated data and other ancillary information. Lastly, the combination of the knowledge on how property distribution changed over the years in conjunction with outputs of physical or probabilistic models that can separate the different contributions associated to flood due to sea level rise, storm surge and rain will allow for properly quantification of the impact of the different components of the climate–economic system on the total exposure and, eventually, damage. This will provide a crucial tool for policymakers, governments, citizens and those who are, rightly, interested in quantifying the impact of climate change on the economic and housing markets.
We used a combination of publicly available software and
codes developed ad hoc for the purpose of this study. Specifically, the ESA
SNAP software used to preprocess the Sentinel datasets is available at
Sentinel-1 data are freely available at
The supplement related to this article is available online at:
MT conceived the study and wrote the first draft of the paper. JRP and SM provided and analyzed the property data, permanent water bodies and LULC data. MT developed and implemented the code for the radar dataset. All authors contributed to the final analysis and final version of the paper.
The authors declare that they have no conflict of interest.
This article is part of the special issue “Hydroclimatic extremes and impacts at catchment to regional scales”. It is not associated with a conference.
Marco Tedesco acknowledges financial support from Columbia University through the RISE program and from the FirstStreet Foundation.
This paper was edited by Fernando Domínguez-Castro and reviewed by two anonymous referees.