Comparison of Neighborhood-Scale, Residential Property Flood-Loss Assessment Methodologies

Mostafiz, Rubayet Bin; Friedland, Carol J.; Rahman, Md Asif; Rohli, Robert V.; Tate, Eric; Bushra, Nazla; Taghinezhad, Arash

doi:10.3389/fenvs.2021.734294

ORIGINAL RESEARCH article

Front. Environ. Sci., 15 October 2021

Sec. Environmental Economics and Management

Volume 9 - 2021 | https://doi.org/10.3389/fenvs.2021.734294

This article is part of the Research Topic Flood Management: Multi-Disciplinary Approaches for Data Observation, Analysis, Forecasting, and Management View all 4 articles

Comparison of Neighborhood-Scale, Residential Property Flood-Loss Assessment Methodologies

Rubayet Bin Mostafiz¹*

Robert V. Rohli^1,4

Arash Taghinezhad²

¹Department of Oceanography and Coastal Sciences, College of the Coast and Environment, Louisiana State University, Baton Rouge, LA, United States
²Bert S. Turner Department of Construction Management, College of Engineering, Louisiana State University, Baton Rouge, LA, United States
³Department of Geographical and Sustainability Sciences, College of Liberal Arts and Sciences, University of Iowa, Iowa City, IA, United States
⁴Department Coastal Studies Institute, Louisiana State University, Baton Rouge, LA, United States

Leading flood loss estimation models include Federal Emergency Management Agency’s (FEMA’s) Hazus, FEMA’s Flood Assessment Structure Tool (FAST), and (U.S.) Hydrologic Engineering Center’s Flood Impact Analysis (HEC-FIA), with each requiring different data input. No research to date has compared the resulting outcomes from such models at a neighborhood scale. This research examines the building and content loss estimates by Hazus Level 2, FAST, and HEC-FIA, over a levee-protected census block in Metairie, in Jefferson Parish, Louisiana. Building attribute data in National Structure Inventory (NSI) 2.0 are compared against “best available data” (BAD) collected at the individual building scale from Google Street View, Jefferson Parish building inventory, and 2019 National Building Cost Manual, to assess the sensitivity of input building inventory selection. Results suggest that use of BAD likely enhances flood loss estimation accuracy over existing reliance on default data in the software or from a national data set that generalizes over a broad scale. Although the three models give similar mean (median) building and content loss, Hazus Level 2 results diverge from those produced by FAST and HEC-FIA at the individual building level. A statistically significant difference in mean (median) building loss exists, but no significant difference is found in mean (median) content loss, between building inventory input (i.e., NSI 2.0 vs BAD), but both the building and content loss vary at the individual building scale due to difference in building-inventory-reported foundation height, foundation type, number of stories, replacement cost, and content cost. Moreover, building loss estimation also differs significantly by depth-damage function (DDF), for flood depths corresponding with the longest return periods, with content loss differing significantly by DDF at all return periods tested, from 10 to 500 years. Knowledge of the extent of estimated differences aids in understanding the degree of uncertainty in flood loss estimation. Much like the real estate industry uses comparable home values to appraise a home, flood loss planners should use multiple models to estimate flood-related losses. Moreover, results from this study can be used as a baseline for assessing losses from other hazards, thereby enhancing protection of human life and property.

Introduction and Background

Flood is one of the most damaging natural hazards, causing massive destruction to property, floodplain structures, and agricultural lands. National Severe Storms Laboratory (2021) recognizes five main types of flooding: 1) riverine, which occurs when excessive precipitation in the higher reaches of a watershed causes a stream to overflow its banks; 2) coastal, which is caused by high tides and/or onshore winds often exacerbated by heavy rainfall; 3) storm surge, associated with landfalling marine storms, often accompanied by strong winds and rain; 4) inland, in which persistent precipitation accumulates locally or a stream overflows due to an ice or debris jam or the failure of a dam or levee; and 5) flash, in which water rushes downslope during an intense local downpour, levee or dam failure, or sudden unclogging of an ice jam or debris blockage. Multiple types of floods may be occurring simultaneously in a given location, creating a compound flood, such as when high tides of coastal floods and storm surges are met by riverine floods from inland precipitation.

Approximately 99 percent of U.S. counties were affected by flooding of one or more of these types during the 1996–2019 period (FEMA 2021a). According to National Oceanic and Atmospheric Administration (National Oceanic and Atmospheric Administration, 2020), the 33 flood events in the U.S.A. from 1980 through 2020 that each caused more than $1 billion in damage (consumer price index- (CPI-) adjusted to 2020$) were responsible for a total of more than (CPI-adjusted) $151.2 billion in U.S. damage. Recent estimates suggest that perhaps three times more Americans (41 million) live within the 100-yr floodplain than suggested via Federal Emergency Management Agency (FEMA) flood maps (Wing et al., 2018).

Property loss (Mostafiz et al. 2020a; Mostafiz et al. 2020b)estimation from floods is important for assessing risk and therefore improving planning efforts to mitigate this hazard. However, such loss estimation from flooding is extremely complicated for several reasons. First, many components of loss must be considered, including direct losses such as to infrastructure, agriculture, and transportation, in addition to indirect, non-property losses such as missed work and productivity, and items of sentimental, non-quantifiable value. Second, many losses, particularly the uninsured losses, are uncatalogued. Third, because the different types of flooding (i.e., flash flooding, riverine flooding, storm surge flooding) include order-of-magnitude differences in the various types of losses, flood loss evaluation does not fall into a “one size fits all” methodology. As a result of these difficulties, despite the importance of having accurate and precise flood loss information, existing data are often inadequate and impossible to validate due to the lack of accurate empirical data.

Two major flood loss models – FEMA’s Hazus (Schneider and Schauer 2006) and U.S. Army Corps of Engineers (USACE) Hydrologic Engineering Center’s Flood Impact Analysis (HEC-FIA; Dunn 2000) – are widely used in the U.S.A. Another promising model – FEMA’s Flood Assessment Structure Tool (FAST; FEMA 2021b) – introduced in November 2019, is likely to become commonly used. Among the two leading flood models, FEMA’s Hazus, introduced in 1998 and now incorporating three “levels” of analysis based on the sophistication of input and output data (Figure 1), incorporates estimates of flood damage to structures, but also to agricultural and utility infrastructure, transportation networks, and other flood-related impacts. Hazus has been shown to be a valuable tool for flood adaptation planning (Banks et al., 2014) and has been used widely in recent years. Such applications include examining the advantages of levee setbacks (Dierauer et al., 2012), assessing feasibility of flood relocation (Cummings et al., 2012), cost-benefit analysis of floodplain conservation (Kousky and Walls 2014), evaluating potential for bridge scouring (Banks et al., 2016), comparing urban vs rural flood vulnerability (Remo et al., 2016), validating other flood estimation methods (Gutenson et al., 2018), estimating modern hurricane-induced flood damage in the context of historical hurricanes (Paul and Sharif 2018), determining impacts of sea-level rise (Ghanbari et al., 2020), and assessing community flood resilience (Allen et al., 2020). Sensitivity analysis on the various input functions in the flood component of Hazus has suggested that the choice of digital elevation data is important (Tate et al., 2015) and that the DDF, flood level, and restoration duration all affect model sensitivity (McGrath et al., 2015).

FIGURE 1

FIGURE 1. Schematic of information involved in each level of Hazus (modified from FEMA 2018).

HEC-FIA has also been used widely in the last two decades. Nafari (2013) elaborated on advantages and disadvantages of HEC-FIA for flood damage assessment. Prominent among the advantages of HEC-FIA software is the ability to incorporate quantitatively an estimation of consequences with uncertainty, thereby estimating and comparing the benefits of existing and future flood risk management measures (Lehman and Light 2016). HEC-FIA also has the advantage of providing separate estimates of direct economic, indirect economic, agricultural, and life loss consequences for flood hazards, by incorporating structure inventory, geospatially and externally derived flood depth grids, and depth-damage functions (DDFs; Lehman et al., 2014). An additional feature is HEC-FIA’s ability to incorporate information from other USACE software, such as the River Analysis System (HEC-RAS) for hydrograph-based or geographic information systems-based estimation of consequences for life and property (Dunn et al., 2016).

Prominent applications for HEC-FIA have been in dam safety (McClelland and Bowles 2002; Needham 2010) and other risk assessments. For instance, Mokhtari et al. (2017) used HEC-FIA to identify vulnerable buildings and croplands, and estimate property and crop losses, in the Ghamsar watershed of Iran for different flood return periods. HEC-FIA has also been used in recent years to predict storm surge impacts (Brackins and Kalyanapu 2016), compare the variation of loss using different sources of digital elevation models (DEM; Bhuyian and Kalyanapu 2018), and validate a new flood loss model (Gutenson et al., 2018).

While the major existing flood loss models are helpful, their utility is limited by three major factors which produce research gaps in the literature. First, at the macro-scale or in study areas with unavailable local input data, their accuracy relies on data from the national-level building inventory. For example, the National Structure Inventory (NSI; https://github.com/HydrologicEngineeringCenter/NSI) provides general information about individual buildings, such as number of stories, foundation type, first-floor height, building type, occupancy, building replacement cost, and content cost. However, data from NSI are incomplete and are often only estimated based on features observable from remote sensing. First-floor elevation is particularly important for assessing potential flood losses, but these building data sets fall short because they rely on Light Detection and Ranging (LIDAR) data that may be older and less accurate than more recently collected data at the micro-scale, particularly in areas that are subject to subsidence, such as Louisiana. Thus, improvements in estimation accuracy of flood loss requires more comprehensive input data at various spatial scales to enhance current understanding. The result is a research gap at the micro-scale. Data compiled from the Jefferson Parish (Louisiana) building inventory (JPBI; Jefferson Parish Geoportal 2021) is an example of a source that offers hope for improved loss estimates for individual buildings, if such collected data and other available data would be used in the flood loss models.

Several potential improvements to the national-level building inventories have gone underexplored, to date. One is the use of ground-truth data, such as Google Street View^® (GSV, https://earth.google.com/) and Zillow^® (https://www.zillow.com/) for input into flood loss models. GSV can be used to determine the number of stories and first-floor elevation perhaps more accurately than existing national-level building data sets. Likewise, Zillow, which provides information on number of stories, building square footage, property value, and other information, may allow for more accurate flood loss estimates. Both GSV and Zillow require visual interpretation of ground-based imagery to determine first-floor elevation and manual recording of building data. Finally, the 2019 National Building Cost Manual (NBCM; Moselle 2018) provides localized construction costs, which may assist in calculating building replacement cost of structures due to flood. Assembly of a “best available data” (BAD) database that includes information from these sources can enhance flood loss estimates and address this gap.

Another important component of flood loss modeling is accuracy and proper selection of the DDFs. Typically, a flood model includes a set of default (i.e., non-site-specific) DDFs, which are percentages of building (and, separately, content) damage as a function of flood depth in the building, with one default function for each combination of building inventory attributes (e.g., one-story residential building with no basement). All too often, users may run the flood loss estimation model with the default DDF without considering the implications of this “one size fits all” assumption, which can contribute to inaccuracies based on the quality of building inventory data upon which such a default DDF rests. Each DDF requires accuracy in the unique set of building attributes, so the appropriateness of the DDF selection depends on the type and accuracy of the data in the building inventory that must be input into the DDF calculation. The generalized set of default DDFs in the two leading models (i.e., Hazus (which uses the same set of default DDFs as FAST) and HEC-FIA) differ substantially from each other, even for the same building attributes (e.g., one-story, single-family, residential home with basement). A thorough examination of the sensitivity of estimated flood losses to the DDFs between the two leading models is lacking in the literature. While it is possible to overwrite the default DDFs when running either Hazus or HEC-FIA, it is difficult to know which DDFs to use in micro-scale analysis.

Another research gap regarding studies using the leading flood loss models is that they tend to ignore local idiosyncrasies because their purpose is usually to plan for and protect an entire county, state, or country. Yet variation within the spatial scale of analysis can be very important at the neighborhood or individual building scale. While there is a trend toward more spatially-integrated flood risk management in many countries (Bubeck et al., 2017), only a few comprehensive studies have evaluated the flood risk methodologies at the micro-scale, especially for individual buildings in a census block.

This study is motivated by the need to identify the effects of building features (i.e., building inventory), model selection, and the DDFs used in estimating building and content losses due to the flood hazard. The results of this study explore the extent to which the results of neighborhood-level Hazus, which has been widely used in flood studies, corresponds with the results from building-level FAST and HEC-FIA models. More specifically, the contribution and objectives of this research are to show the extent to which, at the neighborhood-scale:

1) a localized building inventory assembled from GSV, Zillow, a local parish (i.e., county)-level building inventory, and localized information from the 2019 NBCM differs from the national-level building inventory (i.e., NSI 2.0); the null hypothesis is no difference of means.

2) flood loss estimations differ among Hazus Level 2, FAST, and HEC-FIA, when all are using the same building inventory (i.e., NSI 2.0 or BAD) and DDF; the null hypothesis is no difference of means among the three.

3) improved, customized local building inventory information acquired in (1) above improves the within-model performance of Hazus Level 2, FAST, and HEC-FIA (separately) while using their respective default DDFs, over NSI 2.0; the null hypothesis is no difference of means.

4) the default DDFs in Hazus Level 2, FAST, and HEC-FIA produce different results from each other when they are switched among these three models with all keeping the same building inventory; the null hypothesis is no difference of means.

In general, the results call attention to model sensitivity to the input parameters and DDFs, which will aid in assessing vulnerability to losses in local flood planning and decision-making.

Materials and Methods

Study Area

Census block 220510220012004, in levee-protected, suburban Metairie, Louisiana, United States, which consists of 29 single-family residential dwellings (Figure 2), is selected for this analysis, for several reasons. First, Metairie is a densely-populated area that lies in a severe hazard area for flooding due to local rainfall that is expected to become more hazardous over time. Second, the availability of recently-collected flood depth grids at multiple return periods (i.e., 10-, 50-, 100-, 500-yr) for Metairie facilitates probabilistic analysis of the flood loss-return period relationship. Several flood and hurricane events in recent decades in this community have led to increasing amounts of flood mitigation spending and flood payouts each year (Coastal Protection and Restoration Authority (Coastal Protection and Restoration Authority (CPRA)) 2012; Taghi Nezhad Bilandi 2018), including many properties that have sustained repetitive flood losses (Ergen 2006; Mattei et al., 2009; Kick et al., 2011) on subsiding land (Zou et al., 2016). This census block is selected over other nearby census blocks because some part of it is expected to flood at every return period analyzed. Additionally, this census block is advantageous for study because it is bounded by easily identified roads, to ensure accuracy in analysis. Finally, the uniformity of building types (i.e., 100 percent residential, the subject of this study) offers ease of analysis because the same DDFs can be applied for the entire analysis.

FIGURE 2

FIGURE 2. Study area census block in Jefferson Parish, Louisiana, with building footprint polygons and a gray rectangular dot in the index map.

Data

de Moel and Aerts (2011) noted that flood loss assessment requires input from four types of data: (1) magnitude of the flood hazard, which is provided by a flood depth grid; (2) land use type (e.g., residential, commercial, agricultural); (3) economic value of the object(s) of interest; and (4) DDFs, which, although not required, can improve data analysis over space. In practice, for property loss analysis, (2) and (3) above are often combined in a building inventory. Hazus, FAST, and HEC-FIA require these three components – flood depth grid, building inventory, and DDFs – to calculate the damage for an individual flood event (Table 1).

TABLE 1

TABLE 1. Years analyzed and data sources, for property losses, by hazard in Louisiana.

Flood Depth Grid

Flood depth grids for four available return periods – 10-, 50-, 100-, and 500-yr – for Jefferson Parish, Louisiana, were developed at a scale of 3.048 m × 3.048 m, by FEMA through its Risk Mapping, Assessment and Planning (Risk MAP) program (FEMA 2021c). These data are input into all three models examined here (i.e., Hazus Level 2, FAST, and HEC-FIA). While the complexities associated with Risk MAP products have been identified (Johnson and Maloney 2011), these model-output data represent the best available input for understanding the flood risk.

Building Inventory

Building inventory consists of information about the fundamental features about each building, including, but not limited to, foundation height (first-floor height), foundation type, number of stories, square footage, building replacement cost, and content cost. The unavailability or inaccuracy of data regarding the building inventory variables listed above is among the most important barriers in flood loss studies (Taghinezhad et al., 2020a). Among the leading flood loss models, Hazus incorporates the General Building Stock (GBS) building inventory data set (Shultz 2017a) as the default input. FAST and HEC-FIA require input of building inventory data by the user.

Here, these data are considered from six sources (Table 2). The two national-level databases – GBS and NSI 2.0 – offer advantages of completeness but are disadvantageous because their accuracy is compromised where local-scale attributes might improve estimates of building damage. For example, use of GBS data has been observed to provide severe overestimates of replacement cost (Shultz 2017a), and the need for adjustment has been emphasized (Shultz 2017b). Furthermore, only five buildings in the study area are included in GBS, leading to further concerns that GBS may not represent the study area well.

TABLE 2

TABLE 2. Sources of attribute/data used in this study.

NSI is a structure-specific building dataset for the continuous United States. NSI 2.0 was developed from many sources, including Hazus, USACE, U.S. Geological Survey, National Center for Education Statistics, U.S. Census Bureau, Microsoft, Esri, and Homeland Infrastructure Foundation-Level Data (Barnett 2010). Availability of NSI 2.0 is typically limited to approved federal government applications and academic work such as this research. NSI 2.0 may prove useful, but its very recent release and its restricted availability leaves its utility largely untested.

Because of the uncertainty of NSI 2.0, this research collects and incorporates building attribute data not only for running FAST and HEC-FIA, but also for Hazus Level 2, for comparison between model runs using NSI 2.0 data vs that assembled from four local-level databases – Zillow, JPBI, visually-interpreted GSV imagery, and 2019 NBCM. These local-level sources may provide enhanced local-level accuracy but fail to include all of the variables needed for Hazus Level 2, FAST, and HEC-FIA. Therefore, an original data set named BAD is assembled here using the attribute data available from a compilation of each of these four data sets; the BAD is then input into the three models for comparison against runs using NSI 2.0 input data.

One of the four additional databases considered is Zillow, a popular web-based platform, typically used to identify property for sale and evaluate residential real estate attributes. The underlying data are accessible through the Zillow Transaction and Assessment dataset (ZTRAX), but in this research, data are collected manually through the web platform. Because the Zillow-reported property value combines the land and building value, and because some attribute data available from Zillow are incomplete and could be acquired elsewhere (Table 2) to assemble the BAD, ultimately it was decided that Zillow would not be used as a primary source in this study, but only to confirm data from other sources.

The other three data local-level databases are ultimately chosen for use in compiling BAD. JPBI is provided by the Jefferson Parish (Louisiana) Department of Floodplain Management & Hazard Mitigation. These data are useful because they are collected at the individual building level rather than being estimated en masse over a swath of properties. However, as shown in Table 2, the types of building attribute data available through JPBI are limited. GSV is another locally collected data source. Direct visual interpretation of GSV imagery is used to estimate occupancy type, number of stories, foundation type, and foundation height. Finally, the NCBM (Moselle 2018, p. 12) is used to collect construction cost per square foot. Thus, BAD is prepared by collating information from JPBI, GSV, and NBCM, with the data sources for each building attribute shown in Table 3. Zillow and GBS are also consulted to confirm data or resolve uncertainties.

TABLE 3

TABLE 3. Sources of best available data (BAD).

Depth-Damage Functions (DDFs)

Water depth relative to the first-floor elevation is the main variable in flood loss analysis (Taghinezhad et al., 2020b). Flood loss functions are used commonly to evaluate the percentage of loss to the building itself and (in a separate analysis) contents based on the flood depth variable (USACE 2003; Gulf Engineers & Consultants (Gulf Engineers and Consultants (GEC)) 2006). While the default DDFs for a given set of building inventory attributes are the same in Hazus and FAST, the default DDFs in HEC-FIA for the corresponding set of building inventory attributes differ substantially from those of the other models, as shown in the example of a one-story building with no basement in Figure 3. Moreover, Hazus (but not FAST) offers alternative DDFs that the user can select to match the suggested DDF for that set of building attributes. In addition, in any of the three models, the user can input any DDF, even if it is not among the suggested alternatives, for use in flood loss estimation.

FIGURE 3

FIGURE 3. Comparison of default depth-damage curves for building (left panel) and contents (right panel), used in Hazus/FAST and HEC-FIA, for a one-story building with no basement.

Methodology

To address the four objectives, several combinations are tested against each other, with the following naming convention used. Building inventory ( $I$ ) is specified as $I 1$ for NSI 2.0 and $I 2$ for BAD. The model/software ( $S$ ) is represented as $S 1$ for Hazus Level 2, $S 2$ for FAST, and $S 3$ for HEC-FIA. The input $D D F$ is denoted by $D D F 1$ for that provided by Hazus Level 2 and FAST, and by $D D F 2$ for that provided by HEC-FIA. The damage/loss ( $L$ ) is designated as $L 1$ for building and $L 2$ for content. Finally, flood depth ( $D$ ) is shown as $D 1$ for 10-yr, $D 2$ for 50-yr, $D 3$ for 100-yr, and $D$ 4 for 500-yr flood return period.

Building Inventory Comparison

Within each of the six features in the building inventory comparison (i.e., foundation height, foundation type, number of stories, square footage, building replacement cost, and content cost), all combinations of options are examined comparatively, in this all-residential census block. Regarding the first feature – foundation height, the general egress requirements of the International Building Code (Ching & Winkel 2018) stipulate that stair risers in residential uses are to be built to a height between 4 and 7.75 inches. Because construction using maximum rise is more cost- and space-efficient, here a stair rise near that of the maximum allowed – 7.5 inches – is assumed for calculating FFE. Foundation height is calculated from GSV by multiplying 7.5 inches by the number of stairs. The difference between foundation heights from NSI 2.0 ( $I 1$ ) and BAD ( $I 2$ ) is examined at the neighborhood (census block) scale using pairwise t-test.

The second feature of the building inventory is foundation type. Because foundation type is a nominal-level variable (e.g., slab, crawl-space, pier-and-beam), non-parametric chi-square tests of independence are used to test the null hypotheses of no relationship between foundation types in $I 1$ vs $I 2$ . The third feature – number of stories – is available from JPBI, but GSV is used to compile the BAD data set because it is more accurate and visible, making the data verifiable. This variable also requires the non-parametric chi-square test because it is of ordinal level. The null hypothesis is of no relationship between number of stories in $I 1$ vs $I 2$ . Building area (square footage) is the fourth feature to be considered. Although square footage can be calculated from the JPBI shapefiles, it is not possible to calculate areas for multiple-story buildings. Therefore, square footage data from JPBI is used here to prepare BAD. The difference between square footage from $I 1$ and $I 2$ is examined at the neighborhood scale using pairwise t-test.

The building replacement cost using BAD is calculated by multiplying the per-square-foot construction cost from 2019 NBCM by the building area from JPBI. Because the average cost of constructing a single-family residence of average standard quality class in the New Orleans area is two percent higher than the national average (Moselle 2018, p. 7), construction costs used in the calculations here are increased by 2.0 percent over the values suggested in NBCM. Zillow was considered but ultimately is not used in this analysis, as it reports land value as part of the home purchase price. The difference between the building replacement cost from $I 1$ and $I 2$ is examined at the neighborhood scale using pairwise t-test.

Finally, content cost must be considered. In the NSI 2.0 data set, content cost is assumed to be equal to the building replacement cost. However, NSI 2.0 includes not only residential structures, but also commercial and industrial facilities that are likely to contain very expensive and abundant equipment and merchandise. Because this research includes only residential buildings in a relatively modest neighborhood where expensive jewelry and other possessions are likely to be minimal and vehicles are situated outside of the structure, it is assumed here that content cost is half of the building replacement cost. This assumption is supported by Hazus manual guidelines for similar situations (FEMA 2013, pp. 6–9). The difference between content cost from $I 1$ and $I 2$ is examined at the neighborhood scale using pairwise t-test.

Model Selection Sensitivity

To test the hypothesis resulting from Objective 2, the null hypothesis of equality of population mean (µ; used when the data are distributed normally, as indicated by a Shapiro-Wilk normality test) from among the three software/models,

µ_{I = I|S = S 1 |L = L|D = D|DDF = DDF} = µ_{I = I|S = S 2 |L = L|D = D|DDF = DDF} = µ_{I = I|S = S3 |L = L|D = D|DDF = DDF} (1)

is examined, again at the neighborhood scale (Eq. 1). To test this hypothesis, the values of all variables input (i.e., flood depth, DDF, and building inventory) are the same across model runs, if that variable is mandatory or optional for the given software. However, as is shown in Table 4, some variables are not eligible to be included in some of the software. Thus, any differences in building and/or content loss resulting from each model output would be attributable to differences in the structure information requirements and/or algorithms used in the three software modules. It should be noted that even though some additional variables are optional for input into one or more of the models (e.g., demographic data, vehicle type, fatalities), such variables are ignored in this analysis because they are deemed to be only indirectly related to building/content loss. The sections that follow provide an overview of how each of the three models is run.

TABLE 4

TABLE 4. Structure information used in the testing of software, and their categorization as mandatory or optional.

Hazus Level 2 analysis conducted in this study relies on user-input depth grids and user-defined building inventory rather than the system-generated flood conditions (see again Table 1). The first step is the generation of the study area (i.e., census block level). Then, the 10-yr flood depth grid is uploaded as user data. Next, the building inventory (NSI 2.0 or BAD) is imported under “User Defined Facilities (UDF).” Then, a new hazard scenario is created with the flood depth grid specified, and the riverine floodplain is delineated for a single return period using the default cell size (3.048 m × 3.048 m). This process is repeated for the 50-, 100-, and 500-yr flood depth grids to obtain the results. The DDF can be modified from the “damage functions” on the “analysis” tab.

FAST, which requires Anaconda for Python and the Hazus Python package (HazPy), offers improvements over Hazus Level 1, 2, and 3 in ease of use and processing efficiency. For example, only flood depth grid(s) (in .tiff format) and the structure attributes/table (.csv) are needed as input. Moreover, Hazus DDFs/curves are invoked automatically to calculate the flood loss. A complete explanation of using FAST for flood hazard assessment analysis is given at https://github.com/nhrap-hazus/FAST, but a brief overview to allow for replicability of our method is provided below.

After downloading the “FAST-master” folder from Github™ (https://github.com/nhrap-dev/FAST), the structure table (.csv), consisting of the mandatory fields of occupancy type, building replacement cost, building content value, number of stories, foundation type, first-floor height, total area of building, latitude, and longitude is built. Then, with the structure table moved into the “UDF” folder and the flood depth grids (.tiff) of interest (in the World Geodetic System 1984 (WGS84) projection) into the “rasters” folder, the “FAST.py” icon is executed. “Browse to Inventory Input (.csv)” is selected to input the structure inventory (.csv). Finally, the “Coastal Flooding attribute” is set to “riverine,” with the appropriate return period in the “depth grid” box. After “executing,” the output then appear in the “UDF” folder in .csv format. The DDF of building and contents can be modified from the “lookuptables” folder.

To use HEC-FIA, the layer files (i.e., study area boundary of the selected census block, NSI 2.0 data for the census block(s) of interest) are first converted into shapefiles. Likewise, all raster files (i.e., DEM and flood depth grids) are converted into Tag Image File Format (TIFF or TIF), and all shapefiles and raster files are assigned the same geographic projection system (WGS 1984 UTM Zone 15N).

For risk assessment, after creating and naming a “new study,” “add map layer” is selected from the “map layers” folder, and shapefiles are added (i.e., the study area and NSI 2.0 data). Then, “add terrain model” is selected under the “terrain grids” folder, and the DEM data are “imported.” Next, a new “watershed configurations” is created using the terrain grid (i.e., DEM) that was added. Then, the study area is imported in the “boundaries” and “impacted areas” folders separately, under “geographic data,” choosing from “shapefile name.” Under the “inundation data” folder, “new” is selected and the 10-, 50-, 100-, and 500-yr flood inundation configuration is created separately choosing “grid only” under the “hydraulic data type” and “inundation grid” under “grid only” with the watershed configuration created earlier. Then, “define event” is selected for each flood inundation configuration. The DDFs are imported using the “import from default” option under the “structure occ. types” folder. Next, “structure inventories” is selected, along with “generate from point shapefile” (i.e., NSI 2.0), and “use terrain+foundation height” is chosen under the “first floor elevation source” header. Then, the appropriate “shapefile fields” that were added from NSI 2.0 (i.e., “fd_id,” “st_damcat,” “occtype,” and “val_struct”) are uploaded into the “structure ID,” “damage category,” “occupancy type,” and “replacement value” inventory fields, respectively. Under “optional data” the following inventory fields are selected: “content value,” “foundation height,” “foundation type,” and number of stories,” and the following corresponding shapefile fields are selected: “val_cont,” “found_ht,” “found_type,” and “num_story.” Then, the “new” option is selected within the “alternatives” folder. The study area is clicked in the “impact area set” line, and the appropriate flood depth of interest is chosen for the “inundation configuration,” and NSI 2.0 is selected for the “structure inventory.” In the “simulation” folder, “new” is chosen, and the appropriate flood return period is selected in the “alternative” and “event” fields, with the output sent to the filename specified in the “name” line. Finally, the “compute” option is selected by clicking on the name given to the simulation in the previous step. The results show the aggregate and individual building damage reports, including both structure and content separately, for the selected census block. The DDF can be modified from the “structure occ. types” in the “structure inventories” under the “inventory” folder.

Building Inventory Sensitivity

To address Objective 3, the sensitivity of building inventory (NSI 2.0 vs BAD) is examined at the neighborhood scale, changing only the building inventory while leaving the model (i.e., Hazus Level 2, FAST, or HEC-FIA), loss type (i.e., building or content), and flood depth by return period (i.e., 10-, 50-, 100-, or 500-yr depth) the same, with default model DDF in each comparison, as described by the Eq. 2:

µ_{S = S|L = L|D = D|I = I 1} = µ_{S = S|L = L|D = D|I = I 2} (2)

DDF Sensitivity

To address Objective 4, the sensitivity of DDFs is examined, changing only the depth damage function while leaving building inventory (i.e., NSI 2.0 or BAD), the model (i.e., Hazus Level 2, FAST, or HEC-FIA), loss type (i.e., building or content), and flood depth by return period (i.e., 10-, 50-, 100-, or 500-yr depth) the same in each comparison, as described by the Eq. 3:

µ_{I = I|S = S|L = L|D = D|DDF = DDF 1} = µ_{I = I|S = S|L = L|D = D|DDF = DDF 2} (3)