National Risk Index Primer

National Risk Index

Primer

December 2020

This page intentionally left blank

National Risk Index Primer

National Risk Index Primer i

Table of Contents

1. Introduction ............................................................................................................................. 1

2. Background ............................................................................................................................. 1

2.1. Natural Hazard Selection ................................................................................................ 2

2.2. Working Groups ............................................................................................................... 3

2.3. Literature Review ............................................................................................................ 4

2.4. Subject Matter Expert Review ........................................................................................ 4

2.5. Data and Methodologies ................................................................................................. 5

3. Risk Analysis Overview ............................................................................................................ 5

3.1. Risk Calculation ............................................................................................................... 6

3.2. Scores and Ratings ......................................................................................................... 6

3.3. Assumptions and Limitations ......................................................................................... 9

4. Risk Components Overview ..................................................................................................... 9

4.1. Social Vulnerability ....................................................................................................... 10

4.1.1. Social Vulnerability Source Data ................................................................................. 10

4.1.2. Processing Social Vulnerability Source Data for the NRI ........................................... 11

4.2. Community Resilience ................................................................................................. 12

4.2.1. Community Resilience Source Data ........................................................................... 12

4.2.2. Processing Community Resilience Source Data for the NRI ..................................... 12

4.3. Expected Annual Loss .................................................................................................. 13

4.3.1. Calculating Expected Annual Loss .............................................................................. 13

4.3.2. Analytical Techniques .................................................................................................. 15

NRI Processing Database ...........................................................................................................15

Geographic/Administrative Layers ............................................................................................15

Determining County-Level Possibility of Hazard Occurrence ....................................................16

Base Calculation and Aggregation .............................................................................................16

Representation of Hazards as Spatial Polygons .......................................................................17

Intersection ..................................................................................................................................17

Tabulation ....................................................................................................................................19

5. Natural Hazard Expected Annual Loss Components ............................................................ 21

National Risk Index Primer

National Risk Index Primer ii

5.1. Natural Hazards............................................................................................................ 21

5.2. Natural Hazard Annualized Frequency ....................................................................... 22

5.2.1. Selecting Source Data ................................................................................................. 22

5.2.2. Annualized Frequency Methodology ........................................................................... 22

5.2.3. Data Aggregation ......................................................................................................... 26

5.3. Natural Hazard Exposure ............................................................................................. 28

5.3.1. Selecting Source Data ................................................................................................. 28

5.3.2. Consequence Types ..................................................................................................... 28

Buildings ......................................................................................................................................28

Population ....................................................................................................................................28

Agriculture ...................................................................................................................................29

5.3.3. Exposure Methodology ................................................................................................ 29

Approach 1. Developed Area Density Concentrated Exposure ................................................31

Approach 2. Widespread Hazard Event Exposure .....................................................................33

Approach 3. Hazard-Specific Representative Exposure ...........................................................33

5.3.4. Data Aggregation ......................................................................................................... 33

5.4. Natural Hazard Historic Loss Ratio ............................................................................. 34

5.4.1. Selecting Source Data: SHELDUS ............................................................................... 34

5.4.2. Selecting Source Data: NWS Storm Events Database ............................................... 36

5.4.3. Consequence Types ..................................................................................................... 37

Property .......................................................................................................................................37

Population ....................................................................................................................................37

Agriculture ...................................................................................................................................38

5.4.4. Historic Loss Ratio Methodology ................................................................................. 38

Loss Record Expansion to per Basis Records ...........................................................................38

Loss Ratio per Basis Calculation ................................................................................................39

Non-Loss Causing Hazard Occurrence ......................................................................................41

Bayesian Credibility .....................................................................................................................42

HLR Inheritance ..........................................................................................................................48

5.4.5. Limitations and Assumptions in Historic Loss Ratio Methodology ........................... 48

5.5. Validating Expected Annual Loss Estimates to Historical Losses ............................. 49

6. Using the National Risk Index .............................................................................................. 50

6.1. The NRI Website ........................................................................................................... 50

6.2. Downloadable and Online Datasets ............................................................................ 50

National Risk Index Primer

National Risk Index Primer 1

1. Introduction

The National Risk Index (NRI) is a dataset and an application that help identify communities most at-

risk for natural hazards. The NRI leverages available source data for 18 natural hazards, social

vulnerability, and community resilience to develop a baseline relative risk measurement for each

United States county and Census tract. The NRI is intended to help users better understand the

natural hazard risk of their communities or assigned areas. Intended users include planners and

emergency managers at the local, regional, state, and federal levels, as well as other decision

makers and interested members of the general public. Specifically, it can support decision-making

to:

▪ Update emergency operations plans

▪ Enhance hazard mitigation plans

▪ Prioritize and allocate resources

▪ Identify the need for more refined risk assessments

▪ Encourage community-level risk communication and engagement

▪ Educate homeowners and renters

▪ Support enhanced codes and standards

▪ Inform long-term community recovery

This report provides a detailed overview of the National Risk Index, including its background, data

sources, and processing methodologies. It describes the high-level concepts used to develop the NRI

and calculate its components. The methodologies for computing each hazard’s Expected Annual

Loss (EAL) are also explained in depth in the NRI Technical Documentation.

2. Background

All communities in the United States experience natural hazards, and there is a wide range of

environmental, social, and economic factors that influence each community’s risk to natural

hazards. The likelihood that a community may experience a natural hazard can vary drastically, as

can the associated consequences. Additionally, a community’s risk is influenced by many social,

economic, and ecological factors. FEMA, along with numerous federal, state, and local governments,

academic institutions, nonprofit groups, and private industry (see Figure 1), collaborated to develop

the National Risk Index as a baseline risk assessment application.

Beginning in 2016, FEMA’s Natural Hazards Risk Assessment Program (NHRAP) started work on the

NRI by adopting an established vision for a multi-hazard view of risk that combines the likelihood and

consequence of natural hazards with social factors and resilience capabilities. The goal was to take

a broad, holistic view and create a nationwide baseline of natural hazard risk. Through various

partnerships and working groups, FEMA developed a methodology and procedure to create the NRI

dataset, and then researched, designed, and built the NRI website and application.

National Risk Index Primer

National Risk Index Primer 2

Figure 1: Timeline of the Development of the National Risk Index

The NRI Team conducted multiple workshops and sessions to discuss and determine the

methodologies for translating raw source data into natural hazard risk factors for input into the NRI.

The key objective of these exercises was to ensure that a vetted risk model or equation was

leveraged throughout all methodological development and that certain factors were not being

interpreted inconsistently across the 18 natural hazards.

2.1. Natural Hazard Selection

Natural hazard exposure across the country varies from location to location. The 18 natural hazards

evaluated by the NRI were chosen after reviewing FEMA-approved State Hazard Mitigation Plans for

all 50 states. Tribal hazard mitigation plans were not available at the time of the analysis, and island

territories were excluded from the hazard selection process since data for most NRI hazards are not

available. Note that Washington, DC, was initially excluded from the hazard selection analysis

process; however, it was added to the project scope in 2017 after the hazard selection.

Natural hazards that were included in at least half of the FEMA-approved state plans, or those that

were deemed to be of regional significance, were selected to the NRI (see Figure 2). A regionally

significant hazard is defined as having the capacity to cause widespread, catastrophic damage, such

as Hurricanes, Tsunami, and Volcanic Activity, but otherwise affected fewer than 25 states. It should

be noted that one natural hazard, Subsidence, fit these criteria, but could not be evaluated by the

NRI as there was no reliable, nationwide dataset cataloging this type of hazard event.

National Risk Index Primer

National Risk Index Primer 3

Figure 2: Determination of Hazard Inclusion Based on State Hazard Mitigation Plans

The State Hazard Mitigation Plan review revealed that both Dam Failure and Levee Failure hazards

are profiled by many states, but the datasets needed to develop the EAL component of NRI are not

nationally or publicly available. Levees may be incorporated into the riverine or coastal flood

components if these manmade features are not included on floodplain maps or reflected in NOAA

storm surge and coastal flood analysis. These hazards should not be discussed from traditional risk

assessment. The State Hazard Mitigation Plan hazard analysis was completed in early 2016 and was

limited to the FEMA-approved State Hazard Mitigation Plans. No territorial or tribal plans were

reviewed due to their limited availability.

2.2. Working Groups

After a detailed literature review and hazard analysis, the NRI Team convened three working groups

made of intended users, subject matter experts (SMEs), and interested stakeholders from all levels

of government, private industry, nonprofits, and academia. Each working group was responsible for

an aspect of the NRI’s development and methodology. Experts in each group helped guide the NRI

data and application development.

The Natural Hazards Working Group assessed and recommended datasets associated with the

identified 18 natural hazards selected (as well as Subsidence prior to its recommended removal)

and determined the best ways to incorporate associated data into the NRI.

National Risk Index Primer

National Risk Index Primer 4

The Social Vulnerability and Community Resilience Working Group reviewed and evaluated existing

efforts to measure social vulnerability and community resilience to understand which components

were most important (vulnerability or resilience, or both) and which indices should be used in the

NRI. As a result, both Social Vulnerability and Community Resilience are components of the NRI.

The Data Analytics Working Group oversaw the spatial processing, normalization, and aggregation of

data to arrive at a risk indexing methodology and calculation procedure that integrated the datasets

identified by the other two working groups.

Together, the groups discussed and developed the National Risk Index, including the datasets and

indices to incorporate, definitions of index components, data management strategies and metadata

requirements, data processing and index creation methodologies, and the data visualization and

interactive web mapping application requirements.

2.3. Literature Review

The NRI’s project team reviewed literature in the fields of hazard mitigation, emergency

management, hazard risk science, and other related fields. Centering around a search for natural

hazard and exposure variables, the literature review identified multiple datasets, risk indices,

research reports, methodologies, indicator lists, and existing risk assessment at national and global

scales.

The team identified important risk indicator categories and specific indicators during the review (see

Table 1).

Table 1: Literature Review Risk Indicators and Categories

Risk Indicator Categories

Individual Risk Indicators

▪ Social

▪ Economic

▪ Environmental

▪ Infrastructure

▪ Income

▪ Age

▪ Illnesses

▪ Hospitals

▪ Road Systems

▪ Economic Productivity

▪ Housing

▪ Community Revenue

After review, the team concluded the NRI would involve three components: natural hazard risk

(likelihoods and consequences), social vulnerability, and community resilience.

2.4. Subject Matter Expert Review

Extensive development of the NRI began in 2017 and proceeded through the end of 2019. Over this

period, the NRI team continually iterated on their data processing and risk calculation

methodologies, and engaged with SMEs throughout. A full list of organizations whose members

contributed to the SME reviews is available in the NRI Technical Documentation.

National Risk Index Primer

National Risk Index Primer 5

At major milestones, the team paused development to engage in broader, more comprehensive SME

review periods. The first major milestone arrived in January 2019 where teams of SMEs were tasked

to evaluate two competing draft methodologies: “Methodology 1,” which relied on unitless

standardization of EAL, and “Methodology 2,” which standardized EAL to a dollar value

measurement. Over the course of two weeks and many meetings, dozens of SMEs provided

feedback to the NRI team, resulting in a clear consensus that, although both methodologies were

valid, Methodology 2 created a more robust measurement of risk and a more valuable dataset for

the hazard planning and mitigation communities.

With clear direction on the methodology, the NRI team continued iterating through improvements to

data sourcing and processing. From July through September 2019, they conducted a final

comprehensive SME review period to focus on the new methodology’s results. More than 40 SMEs

participated in over 20 review sessions and helped the team reach concurrence on the validity and

value of the dataset. From these sessions, the NRI team was equipped to begin final iterations of the

methodology and source data processing.

2.5. Data and Methodologies

Over the course of several years, with the help of hundreds of collaborators and contributors, and

through unknown iterations of planning, design and development, the NRI working groups concluded

their work by reviewing and providing feedback on an iterative version of the National Risk Index

dataset (December 2019).

Briefly stated, the NRI is a first-of-its-kind, nationwide, holistic assessment of baseline risk to natural

hazards. Although it is based on extensive research and best practices in the risk assessment fields,

the NRI’s methodology is unique and carefully constructed the specific needs of natural hazard risk

assessment at both small and large geographic scales. A detailed overview of the risk calculation is

available in the Risk Analysis Overview section.

The NRI’s most important and central component, Expected Annual Loss (EAL), is a robust

measurement that quantifies the anticipated economic damage resulting from natural hazards each

year. Details of its equation and analytical techniques are available in the Expected Annual Loss

section. EAL consists of the best-available datasets for 18 natural hazards of national and regional

significance, with source data being processed to match the unique nature of each natural hazard.

Full processing details for each hazard are available in the NRI Technical Documentation. Per the

direction established at initiation, the dataset also includes measurements of social vulnerability and

community resilience to quantify overall risk. These key components are detailed fully in the Social

Vulnerability and Community Resilience sections.

3. Risk Analysis Overview

Risk, in the most general terms, is often defined as the likelihood (or probability) of a natural hazard

event happening multiplied by the expected consequence if a natural hazard event occurs. The

generalized form of a risk equation is given in Equation 1.

National Risk Index Primer

National Risk Index Primer 6

Equation 1: Generalized Risk Equation

3.1. Risk Calculation

In the National Risk Index, risk is defined as the potential for negative impacts as a result of a

natural hazard. The risk equation behind the NRI includes three components: a natural hazards

component, a consequence enhancing component, and a consequence reduction component. EAL is

the natural hazards risk component, measuring the expected loss of building value, population,

and/or agricultural value each year due to natural hazards. Social vulnerability is the consequence

enhancing component and analyzes demographic characteristics to measure a community’s

susceptibility of social groups to the adverse impacts of natural hazards. Community resilience is the

consequence reduction component and uses demographic characteristics to measure a

community’s ability to prepare for, adapt to, withstand, and recover from the effects of natural

hazards. These three risk components are combined into one risk value using Equation 2.

Equation 2: NRI Risk Equation

An overall composite Risk Index score and individual hazard Risk Index scores are calculated for

each county and Census tract included in the NRI. A composite Risk Index score measures the

relative risk of a location considering all 18 natural hazards included in the index. An individual

hazard Risk Index score measures the relative natural hazard risk of a location for a single natural

hazard. All scores are relative as each Census tract or county’s score is evaluated in comparison with

all other Census tracts or counties.

3.2. Scores and Ratings

In this NRI Risk Equation, each component is represented by a unitless index value, representing a

community’s score relative to all other communities. From the three indices, the Risk Index score is

calculated to measure a community’s risk to all 18 natural hazards. The Risk Index is also a unitless

index and represents a community’s risk relative to all other communities. The Risk Index and EAL

are provided as both composite scores from the summation of all 18 natural hazards, as well as

individual-hazard scores where each hazard is considered separately.

All calculations are performed separately at two levels-of-detail—county and Census tract—so scores

are relative only within their level-of-detail. It must be stressed that scores are relative, representing

a community’s relative position among all other communities for a given component and level-of-

detail. Scores are not absolute measurements and should be expected to change over time either by

their own changing measurements or changes in other communities.

National Risk Index Primer

National Risk Index Primer 7

All scores are constrained to a range of 0 (lowest possible value) to 100 (highest possible value). To

achieve this range, the values of each component are rescaled using min-max normalization, which

preserves their distribution while making them easier to understand. EAL values are heavily skewed

by an extreme range of population and property value densities between urban and rural

communities. To account for this, a cube root transformation is applied before min-max

normalization. By applying cube root transformation, the NRI controls for this characteristic and

provides scores with greater differentiation and usefulness. If the minimum value of the EAL is a

nonzero number before normalization, an artificial minimum is set to 99% of that value, so that

entities expected to experience loss do not receive a 0 EAL score.

For every score there is also a qualitative rating that describes the nature of a community’s score in

comparison to all other communities, ranging from “Very Low” to “Very High.” Because all ratings are

relative, there are no specific numeric values that determine the rating. For example, a community’s

Risk Index score could be 8.9 with a rating of “Relatively Low,” but its Social Vulnerability score may

be 11.3 with a rating of “Very Low.” The rating is intended to classify a community for a specific

component in relation to all other communities.

To determine ratings, a methodology known as k-means clustering or natural breaks is applied to

each score. This approach divides all communities into groups such that the communities within

each group are as similar as possible (minimized variance) while the groups are as different as

possible (maximized variance).

In the NRI application’s maps and data visualizations, standard color schemes have been applied to

the qualitative ratings. Risk Index ratings are represented using a diverging blue (Very Low) to red

(Very High) color scheme. Ratings for EAL, Social Vulnerability and Community Resilience are

represented using sequential color schemes (e.g., single color at various intensities). According to

the NRI, higher EAL, higher Social Vulnerability, and/or lower Community Resilience increase your

overall risk. In general, darker shading in the map layers represents a higher contribution to overall

risk. When source data is not available or a score cannot be calculated, then additional ratings are

used and shown in white or shades of gray. The NRI’s standard color schemes are shown in Figure 3.

Figure 3: National Risk Index Qualitative Rating Legend

National Risk Index Primer

National Risk Index Primer 8

Scores of 0 (zero) or missing values (“nulls”) in the EAL components receive ratings that reflect the

logic behind the score. A county or tract whose EAL is zero either has no building value, population,

or crop value exposed to the hazard, or has a calculated hazard frequency of zero, except for hazards

that apply a minimum annual frequency. These areas are displayed in the NRI application as having

“No Expected Annual Loss” for the designated hazard.

In collaboration with SMEs most familiar with individual hazards and the source data used in the

NRI, a priori definitions of hazard applicability have also been applied to help distinguish between

where no hazard risk exists and where the hazard is deemed to be not possible. For example,

Coastal Flooding EAL is not computed for inland areas. These areas are displayed in the NRI

application as “Not Applicable” for EAL computation for the designated hazard.

Finally, if a component used to calculate the EAL of a Census tract or county for a hazard has a null

value, the community is rated as “Insufficient Data.” For example, certain hazards, such as Wildfire,

Lightning, and Landslide, only have source data used to determine frequency or exposure for the

conterminous United States, meaning that both Alaska and Hawaii are rated as “Insufficient Data” to

compute the EAL for those hazards. When a hazard is not applicable or there is insufficient data for a

community, EAL for that hazard is simply not included in the community’s final summation and

scoring. A summary of non-numerical ratings is provided Table 2.

Table 2: Definitions of Ratings without Numerical Scores

Rating

Risk Index

Expected Annual

Loss

Social

Vulnerability

Community

Resilience

No Rating

EAL is zero. SoVI

and/or HVRI BRIC

are not available.

n/a

No Expected

Annual Loss

n/a

Hazard exposure

or frequency is

zero.

n/a

Not Applicable

Location is not

considered at-

risk for hazard

occurrence.

Location is not

considered at-

risk for hazard

occurrence.

n/a

Insufficient Data

Hazard source

data is not

available.

Hazard source

data is not

available.

n/a

Data Unavailable

n/a

SoVI is not

available.

HVRI BRIC is not

available.

National Risk Index Primer

National Risk Index Primer 9

3.3. Assumptions and Limitations

The National Risk Index dataset and application are meant for planning purposes only and are

intended for use as a tool for broad, nationwide comparisons. Nationwide datasets used as inputs

for the NRI are in many cases not as accurate as locally available data. Users with access to local

data for each NRI risk factor should consider substituting those data to calculate a more accurate

EAL value at the local level.

The NRI does not consider the intricate economic and physical interdependencies that exist across

geographic regions. The user should be mindful that hazard impacts in surrounding counties or

Census tracts can cause indirect losses in a location regardless of the location’s risk profile.

The NRI’s most recent source datasets only include a period of record up to 2017. It should be noted

that the EAL values represent an extrapolation based on a snapshot in time. Extending source data

collection beyond that time may result in varying Census tract or county EAL values due to changes

in recorded hazard intensity and frequency, as well as fluctuations in local economic value and/or

population density.

Most of the hazards evaluated by the NRI use a frequency model to determine EAL. This makes it

difficult to accurately estimate EAL for high consequence, low frequency events. Certain rare hazards

(such as Earthquake, Hurricane, Tsunami, and Volcanic Activity) benefit from using a probabilistic

model that estimates the likelihood of a hazard event occurring over an extended period of time,

which can then be annualized. Of these, only Earthquake has probabilistic source data that is

sufficient for accurately estimating EAL.

Best available nationwide data for some risk factors are rudimentary. More sophisticated risk

analysis methodologies are available but require more temporally and spatially granular data for

hazard exposure, frequency, and historic loss measurements.

The NRI methodology makes various efforts to control for possible discrepancies in source data, but

cannot correct for all accuracy problems present in that data. The NRI processing database is a

complex system and localized inaccuracies in source data have the potential to propagate.

Therefore, the NRI and its components should be considered a baseline measurement and a

guideline for determining hazard risk but should not be used as an absolute measurement of risk.

4. Risk Components Overview

The risk score in the NRI is based on three components: Social Vulnerability, Community Resilience,

and EAL, with EAL based on Exposure, Annualized Frequency, and Historic Loss components, for a

total of five risk factors. Each risk factor contributes to either the likelihood or consequence aspect

of risk and can be classified as one of two risk types: either risk based on geographic location or risk

Federal Emergency Management Administration (FEMA). (2017). Hazus Estimated Annualized Earthquake Losses for the

United States: FEMA Publication 366. Retrieved from https://www.fema.gov/sites/default/files/2020-

07/fema_earthquakes_hazus-estimated-annualized-earthquake-losses-for-the-united-states_20170401.pdf

National Risk Index Primer

National Risk Index Primer 10

based on the nature and historical occurrences of natural hazards. The five risk factors are

summarized in Table 3 and further described in this section.

Table 3: National Risk Index Score Risk Factors

NRI Risk

Component

NRI Risk Factors

Risk Factor

Description

Risk Equation

Bin

Risk Type

Assignment

Social

Vulnerability

Social

Vulnerability

Consequence

Enhancer

Consequence

Geographic Risk

Community

Resilience

Community

Resilience

Consequence

Reducer

Consequence

Geographic Risk

Expected Annual

Loss

Exposure

Expected

Consequence

Natural Hazard

Risk

Expected Annual

Loss

Annualized

Frequency

Probability of

Occurrence

Likelihood

Natural Hazard

Risk

Expected Annual

Loss

Historic Loss

Expected

Consequence

Natural Hazard

Risk

4.1. Social Vulnerability

Social vulnerability is broadly defined as the susceptibility of social groups to the adverse impacts of

natural hazards, including disproportionate death, injury, loss, or disruption of livelihood. Social

vulnerability considers the social, economic, demographic, and housing characteristics of a

community that influence its ability to prepare for, respond to, cope with, recover from, and adapt to

environmental hazards.

As a consequence-enhancing risk factor, the Social Vulnerability score represents the relative level of

social vulnerability for a given county or Census tract. A higher social vulnerability score results in a

higher risk score. Because social vulnerability is unique to a geographic location—specifically, a

county or Census tract—it is a geographic risk factor.

The Social Vulnerability and Community Resilience Working Group reviewed multiple top-down and

bottom-up indices and chose to recommend the University of South Carolina’s Hazards and

Vulnerability Research Institute (HVRI) Social Vulnerability Index (SoVI).

4.1.1. SOCIAL VULNERABILITY SOURCE DATA

Social Vulnerability source data provider: University of South Carolina's Hazards and Vulnerability

Research Institute (HVRI) Social Vulnerability Index (SoVI)

National Risk Index Primer

National Risk Index Primer 11

SoVI is a location-specific assessment of social vulnerability that utilizes 29 socioeconomic variables

(listed below) deemed to contribute to a community’s reduced ability to prepare for, respond to, and

recover from hazards.

▪ Median gross rent for renter-occupied

housing units

▪ Median age

▪ Median dollar value of owner-occupied

housing units

▪ Per capita income

▪ Average number of people per household

▪ % population under 5 years or age 65 and

over

▪ % civilian labor force unemployed

▪ % population over 25 with <12 years of

education

▪ % children living in married couple families

▪ % female

▪ % female participation in the labor force

▪ % households receiving Social Security

benefits

▪ % unoccupied housing units

▪ % families with female-headed households

with no spouse present

▪ % population speaking English as second

language (with limited English proficiency)

▪ % Asian population

▪ % African American (Black) population

▪ % Hispanic population

▪ % population living in mobile homes

▪ % Native American population

▪ % housing units with no car available

▪ % population living in nursing facilities

▪ % persons living in poverty

▪ % renter-occupied housing units

▪ % families earning more than $200,000

income per year

▪ % employment in service occupations

▪ % employment in extractive industries (e.g.,

farming)

▪ % population without health insurance

(County SoVI only)

▪ Community hospitals per capita (County

SoVI only)

Data was acquired from HVRI’s SoVI website and users looking for more information should consult

HVRI.

4.1.2. PROCESSING SOCIAL VULNERABILITY SOURCE DATA FOR THE NRI

For the NRI, the SoVI dataset was incorporated using min-max transformation (0.01-100.00 scale).

County-level and Census tract-level Social Vulnerability scores were classified into five qualitative

categories, from “Very Low” to “Very High,” using k-means clustering. Social Vulnerability scores are

available for all counties, but they are absent for 292 Census tracts that have no population. Risk

cannot be calculated for tracts without Social Vulnerability scores, so those Census tracts are rated

“Insufficient Data.”

Cutter, S.L., Boruff, B.J. & Shirley, W.L. (2003). Social vulnerability to environmental hazards. Social Science

Quarterly, 84(2): 242-261. Retrieved from https://doi.org/10.1111/1540-6237.8402002

National Risk Index Primer

National Risk Index Primer 12

4.2. Community Resilience

Community Resilience is defined by FEMA as the ability of a community to prepare for anticipated

natural hazards, adapt to changing conditions, and withstand and recover rapidly from disruptions.

There are multiple, well-established ways to define community resilience at the local level, and key

drivers of resilience vary between locations. Because there are no nationally available, bottom-up

community resilience indices available, the Social Vulnerability and Community Resilience Working

Group chose to utilize a top-down approach. The NRI relies on using broad factors to define

resilience at a national level and create a comparative metric to use as a risk factor. The Social

Vulnerability and Community Resilience Working Group reviewed multiple top-down indices and

chose to recommend the University of South Carolina’s Hazards and Vulnerability Research Institute

(HVRI) Baseline Resilience Indicators for Communities (HVRI BRIC) index.

The Community Resilience score is a consequence reduction risk factor of the NRI and represents

the relative level of community resilience for a given location. A higher Community Resilience score

results in a lower Risk score. Because Community Resilience is unique to a geographic location—

specifically, a county—it is a geographic risk factor.

4.2.1. COMMUNITY RESILIENCE SOURCE DATA

Community Resilience source data provider: University of South Carolina’s Hazards and Vulnerability

Research Institute (HVRI) Baseline Resilience Indicators for Communities (BRIC)

Community Resilience data for the NRI is supported by the HVRI BRIC. HVRI BRIC provides a sound

methodology for quantifying community resilience by identifying the ability of a community to prepare

and plan for, absorb, recover from, and more successfully adapt to the impacts of natural hazards.

The HVRI BRIC dataset includes a set of 49 indicators that represent six types of resilience: social,

economic, community capital, institutional capacity, housing/infrastructure, and environmental. It

uses a local scale within a nationwide scope, and the national dataset serves as a baseline for

measuring relative resilience. This data can be used to compare one place to another and determine

specific drivers of resilience, and a higher HVRI BRIC score indicates a stronger and more resilient

community.

4.2.2. PROCESSING COMMUNITY RESILIENCE SOURCE DATA FOR THE NRI

For the NRI, the HVRI BRIC dataset was in incorporated using min-max transformation (0.01-100.00

scale). Because HVRI BRIC has a potential range of 0.0 to 6.0, but the full range does not exist in the

dataset, the normalized score for Community Resilience ranges from 41.2 to 64.7. HVRI BRIC is only

available at the county-level, so Community Resilience scores were inferred from counties to Census

tracts by assigning each Census tract the value of its parent county. Community Resilience scores

National Institute of Standards and Technology (NIST). (2020). Community Resilience. Retrieved from:

https://www.nist.gov/topics/community-resilience

National Risk Index Primer 13

National Risk Index Primer

were classified into five qualitative categories, from “Very Low” to “Very High,” using k-means

clustering.

For more information on the creation of the HVRI BRIC, please refer to HVRI’s BRIC website or the

geographies of community disaster resilience paper published by Cutter, Ash, and Emrich (2014).

4,5

4.3. Expected Annual Loss (EAL)

The EAL for each Census tract or county is the average economic loss in dollars resulting from

natural hazards each year. EAL is computed for each hazard type and only quantifies loss for

relevant consequence types (i.e., buildings, people, or agriculture). For example, most natural

hazards only significantly impact buildings and population, so the loss to agriculture is not included

in the computation. However, the EAL for Drought only quantifies the damage to crops and livestock

(agriculture) in its computation. A consequence type is only included in the EAL computation for a

hazard if at least 10% of the total reported economic loss due to the hazard (see the Natural Hazard

Historic Loss Ratio section) is of that consequence type.

All loss is quantified as a dollar amount. While building and agriculture loss are quantified in dollars

in the source data, population loss is quantified as the number of fatalities and injuries and must be

converted to ensure all EAL values use a common unit of measurement. Population loss is

monetized using the value of statistical life approach in which each fatality or ten injuries is treated

as $7.4 million of economic loss, an inflation-adjusted Value of Statistical Life (VSL) used by FEMA.

To adjust for inflation, all historic losses are converted to 2016 dollars.

4.3.1. CALCULATING EXPECTED ANNUAL LOSS

EAL is calculated using a multiplicative equation that considers the consequence risk factors of

natural hazard exposure and historic loss, and the likelihood risk factor of natural hazard frequency

for 18 natural hazards. The EAL value for each consequence type is calculated by multiplying the

total exposure value of an area by the estimated annual frequency of a natural hazard event and by

the historic loss ratio (see Equation 3). See the Natural Hazard Expected Annual Loss Components

section for further explanation of these EAL components and how they are computed. EAL values are

computed at the Census block level (or for some hazards, the Census tract level) for each relevant

consequence type and summed to produce a composite EAL for each hazard (see Equation 4). A

cubic root transformation is applied to each hazard-specific EAL value to address skew. The resulting

transformed values are then min-max normalized (0.00 – 100.00 scale) to produce an EAL score for

Cutter, S.L., Ash, K.D., & Emrich, C.T. (2014). The geographies of community disaster resilience. Global Environmental

Change, 29, 65-77. https://doi.org/10.1016/j.gloenvcha.2014.08.005

See also Mitigation Framework Leadership Group (MitFLG), Federal Emergency Management Agency (FEMA). (2016).

Draft Interagency Concept for Community Resilience Indicators and National-Level Measures. Washington, DC: Department

of Homeland Security (DHS). Retrieved from https://www.fema.gov/media-library-data/1466085676217-

a14e229a461adfa574a5d03041a6297c/FEMA-CRI-Draft-Concept-Paper-508_Jun_2016.pdf

Federal Emergency Management Agency (FEMA). (2016). Benefit-cost sustainment and enhancements: baseline

standard economic value methodology report. Retrieved from

https://www.caloes.ca.gov/RecoverySite/Documents/Benefit%20Cost%20Sustainment.pdf

National Risk Index Primer

National Risk Index Primer 14

each hazard. A total EAL is also summed from all hazard EALs for the area and a total EAL score is

calculated using the same cubic root transformation and min-max normalization process.

Hazard-specific Risk Index scores are calculated using individual hazard EAL scores. Overall Risk

Index scores are calculated using the composite EAL score.

Equation 3: Hazard-Specific Expected Annual Loss by Consequence Type

Equation 4: Composite Hazard-Specific Expected Annual Loss

While each hazard uses the same components to calculate EAL, these computations require

different approaches due to the varying nature of the hazards and the differences in source data

format. A set of common analytical techniques (see the Expected Annual Loss section) are leveraged

to achieve the best possible normalization between all hazards for accurate NRI calculation. The

process for computing the EAL and its components for each individual hazard are described in the

hazard-specific sections of the NRI Technical Documentation.

See Table 4 for a simplified example of a county-level EAL calculation for the hazard Hail. All three

consequence types are included in the calculation of Hail EAL. By multiplying the county’s

consequence exposure, hazard frequency, and consequence-specific historic loss ratio, an EAL value

for that consequence type is determined. The values for each consequence are summed to produce

the composite EAL for the county. This composite EAL is used to derive the hazard’s EAL score for

that county. This computation includes a min-max normalization using the hazard-specific composite

EAL values of all counties in the nation. The composite EAL for Hail is summed with the composite

EAL values for the 17 other hazards to calculate the total EAL, which is scored in the same way.

Table 4: Example of a County-Level EAL Calculation for Hail

EAL Component

Building Value

Population

Agriculture Value

Exposure

$23.14 M

182,265 people or

$1.35 T

$120,000

Frequency

9.7 events/year

Historic Loss Ratio

1.6e-8

3.2e-8

1.4e-7

National Risk Index Primer

National Risk Index Primer 15

EAL Component

Building Value

Population

Agriculture Value

Expected Annual Loss

$3,478

0.054 people or

$399,954

$156

4.3.2. ANALYTICAL TECHNIQUES

Arriving at a dollar value representing the EAL due to each of the 18 hazards for every county and

Census tract in the United States requires multiple analytical techniques utilized across all hazards

to ensure the most accurate representation of loss.

NRI Processing Database

To support the processing of the NRI, a dedicated SQL Server database environment was

established. Using a relational database to store and analyze each dataset used to compute the NRI

provides a variety of benefits. The database allows for computational efficiencies when calculating

the components of the EAL for more than 11 million Census blocks in the United States. Grouping

and aggregation functions can be used easily to roll these values into the Census tract and county

level values displayed in the NRI application. Implementation of NRI methodologies in stored

procedures allows for application and adaptation of complex business logic and spatial analysis. The

NRI processing database also makes quality control easier by allowing complex calculations to be

processed in steps with output for each step accessible in its own table. Records for each Census

block can be checked to identify outliers and any possible problems with the methodology or

algorithms. Additionally, repeatable processes can be modified and run in smaller portions, cutting

down on processing time as methodology is adapted. For example, a change in source data for a

hazard only requires the replacement of hazard-specific source data tables and for the re-processing

of a single hazard to be executed. The NRI processing database also supports version control and

allows backups of each version to be stored securely.

Most spatial functions, such as buffering and intersection, are performed within the NRI processing

database. However, some processes, such as land use tabulation necessitate the use of ArcGIS tools

and functions. The output of these externally performed processes is transferred and stored within

the NRI processing database where it is used to compute the components of the EAL.

Geographic/Administrative Layers

EAL components may be calculated at three different administrative layers: Census block, Census

tract, and county. The most granular level is the Census block and, when possible, values are

calculated at this level and then aggregated. The source of the boundaries for these layers is the US

Census Bureau’s 2017 TIGER/Line shapefiles.

The shapefiles include US territories and some large

bodies of water. These are either manually removed or clipped based on a County boundary

US Census Bureau. (2017). Cartographic Boundary Shapefiles [cartographic dataset]. Retrieved from

https://www.Census.gov/geographies/mapping-files/time-series/geo/carto-boundary-file.2017.html

National Risk Index Primer

National Risk Index Primer 16

shapefile provided by Esri.

All spatial layers use the North America Albers Equal Area Conic

projection. Figure 4 provides examples of block, tract, and county boundaries.

Figure 4: Example of County, Census Tract, Census Block Shapes

Determining County-Level Possibility of Hazard Occurrence

Not all hazards are able to occur in all areas. For example, Coastal Flooding cannot occur in Kansas

and Avalanches cannot occur on flat terrain. The NRI logically differentiates areas where a given

hazard is unlikely or has never occurred from areas where that hazard is impossible using a control

table in the database that designates where each hazard can occur. This table is based on counties

that intersect past hazard event polygons generated through spatial processing or which have some

possibility of occurrence as identified by probabilistic or susceptibility source data or which have

recorded loss due to hazard occurrence.

Base Calculation and Aggregation

One of the NRI’s strengths is that it determines the EAL for an area at the lowest geographical level

deemed appropriate, predominantly the Census block level. EAL is determined by assessing the

combination of a specific location’s frequency of occurrence and associated consequence if it were

to occur (for example, how often Riverine Flooding occurs in the area and what buildings, population

Esri, TomTom North America, Inc., & US Census Bureau. (2012). USA County Boundaries [cartographic dataset]. Retrieved

from https://www.arcgis.com/home/item.html?id=f16090f6d3da48ec8f144a0771c8fec4

National Risk Index Primer 17

National Risk Index Primer

and crops are potentially affected). For many hazard types, frequency and exposure can be highly

localized. Modeling the event frequency in coordination with its exposure provides the best

assessment of its expected impact.

The Census block is currently the lowest administrative level at which population and building value

data are nationally, consistently, and publicly available. By performing the EAL calculation at the

Census block level, the NRI is more accurately assessing EAL by looking at specific frequency and

exposure combinations at the lowest possible resolution. The NRI provides the most relevant

aggregations to its users, namely EAL values at the Census tract and county levels. For most

hazards, Census tract and county level exposure and frequency are calculated by “rolling up” or

aggregating values from the Census block level.

Representation of Hazards as Spatial Polygons

EAL components for each hazard are derived from one or more sources of spatial hazard

information. This can include identified hazard-susceptible areas, spatiotemporal records of past

hazard occurrences, and countywide records of economic loss due to a hazard event. The format of

spatial source data varies by hazard. Frequency and exposure calculations typically require

spatiotemporal records of past hazards or probabilistic modelling. To achieve a uniform level of

accuracy, any spatial hazard source data were converted to vector polygon format and intersected

with the Census blocks or tracts.

Necessary conversions are performed either with tools available in Esri’s ArcGIS software or with SQL

Server’s spatial operations. Common methods of hazard conversion used for NRI calculation are the

buffering of points and lines to form polygons, and raster-to-polygon conversion.

Point and line representations of hazard events or hazard-susceptible areas are buffered by different

distances depending on the hazard. Point buffers allow for better representation of event coverage

or area of possible impact. Path representations, such as those for Tornado and Hurricane, are

included in the source data as a series of points with a common identifier (e.g., StormID). These are

connected by a line or multi-segmented line. The line is then buffered by a distance depending on

the intensity of the Tornado (Enhanced-Fujita scale) or Hurricane (Saffir-Simpson scale) event. See

the spatial processing discussion in the hazard-specific sections of the NRI Technical Documentation

for more detail on buffering techniques.

Conversion from raster to polygon vector format is performed by using ArcGIS’s Create Fishnet tool to

form a grid of rectangular cells that match the extent and dimensions of the original raster and then

using the Extract Values to Table tool to insert the cell values of the raster into the corresponding

fishnet polygon’s attribute table. In vector format, attributes from the source raster data can be used

to filter or select the data needed for hazard specific methodology calculations.

Intersection

Determining areas of spatial intersection between hazard events or susceptible areas and the

various levels of reference layers is an essential function used in calculating EAL. The results of

National Risk Index Primer

National Risk Index Primer 18

these intersections are stored in the NRI processing database and used for multiple purposes. For

many hazards, the quantification of a hazard’s exposure is done at the Census block level. This

requires the computation of intersecting areas of exposure. Figure 5 provides an example of a

hazard event shape intersecting a Census block.

Frequency computations also typically involve counting the number of hazard event polygons that

intersect the Census block. Widespread hazards, like Hurricanes, often require a larger

administrative layer to more accurately represent the frequency of Hazard events. For these types of

hazards, the intersection is performed with a 49-by-49 km fishnet grid and the count of the fishnet

grid cell is inherited by the Census blocks it encompasses, using an area-weighted value when a

Census block intersects more than one cell.

Figure 5: Example of Intersection Between Hazard Event and Census Block

The 49-by-49 km grid cell size was used because of analysis conducted early in the project which

roughly estimated the average Census tract size to be 4,900 m

(or 70-by-70 m) and the average

county size to be 2,500 km

(or 50-by-50 km), which was reduced slightly to 49-by-49 km to ensure

the county size was a multiple of the tract size. Though the use of a grid at the average Census tract

resolution was discarded, the use of the 49-by-49 km fishnet grid was maintained for the

calculation of frequency for widespread hazards.

National Risk Index Primer 19

National Risk Index Primer

Tabulation

Tabulation refers to the process of calculating the composition of a vector shape by overlaying it on

a raster layer inside a GIS. The GIS computes the area of raster cells completely contained within

the vector shape by raster value.

The land use tabulation process is performed by using the Tabulate Area tool in Esri’s ArcGIS

software. All spatial layers use the North America Albers Equal Area Conic projection. A layer

containing county boundaries is tabulated against the 2017 CropScape raster file

, which describes

the land use of the conterminous United States in 30-by-30-m cells using 132 distinct raster values.

The output layer contains a record for each county (by county FIPS code) with fields for each class

(crop types, developed areas, etc.) displaying the area (in square meters) of each type of land use

within the county. There are five classes of developed area (Developed, Developed Open Space, and

Developed Low, Medium, and High Intensity) which can be summed to get the total developed area

of the county. The area values of all crop classes can be summed to give a total agricultural area.

This same tabulation is performed at the Census tract and Census block level to support the

computation of developed area densities at these levels. The EAL calculations for most hazards

utilize the developed area density values at the Census block level (see the Approach 1. Developed

Area Density Concentrated Exposure).

The CropScape layer only contains information for the conterminous United States. For Alaska and

Hawaii, a similar tabulation process is carried out substituting the 2016 National Land Cover

Database (NLCD) raster files

for both states. NLCD uses the same classification types for

developed land as CropScape. It has two classifications for agricultural land: Pasture/Hay and

Cultivated Crops.

Primary tabulation involves summing the total area of interest (e.g., developed land use) and

dividing by the total area of raster cells contained. The shape area (e.g., Census block, Census tract,

or county) is multiplied by this developed area percent to calculate the developed area (in square

kilometers). To speed up calculations, the intersected shapes are classified as whether they

completely contain the Census block, tract, or county (for which developed area and crop/pasture

area had already been calculated). For such shapes, the values were transferred over without

tabulation. Tabulated areas are approximations based on the cell size of the source raster and can

exceed the area of the shape being tabulated. In these cases, the total area of the shape is set as

the ceiling of the tabulation area results.

Very small intersections of hazard event shapes with Census blocks can be too small to tabulate

against 900-m

raster cells. If not, all shapes are tabulated using the primary method, secondary

methods are pursued. Secondary methods are hazard-specific. For example, secondary tabulation of

US Department of Agriculture (USDA), National Agricultural Statistics Service (NASS). (2017). Published crop-specific data

layer [online dataset]. Retrieved from https://nassgeodata.gmu.edu/CropScape/

Multi-Resolution Land Characteristics Consortium. (2016). National Land Cover Database (NLCD) [online dataset].

Retrieved from https://www.mrlc.gov/data

National Risk Index Primer 20

National Risk Index Primer

Drought-Census tract shapes involves extracting the raster value at the centroid of the shape. The

entire area of the shape is classified as the raster value extracted at the centroid. On the other

hand, Riverine Flooding shapes, as many administrative boundaries are drawn using rivers, are

winding and narrow (see the shape on the right in Figure 6). A centroid-based approach is not the

most accurate. For this reason, raster cell centroids representing developed areas were exported.

SQL Spatial routines then calculated whether a developed land-use was within 42 meters (the

hypotenuse distance of a 30-by-30 m raster cell). If so, the entire shape was deemed developed.

If not, the shape was considered to have zero developed area.

Figure 6: Land Use Raster Tabulation

(This section intentionally left blank.)

National Risk Index Primer 21

National Risk Index Primer

(This section intentionally left blank.)

5. Natural Hazard Expected Annual Loss Components

The NRI represents natural hazard in terms of EAL, which incorporates data for natural hazard

exposure, annualized frequency, and historic loss. A single “mental model” was leveraged

throughout all methodological processes in calculating these EAL components, so that certain risk

factors were not being interpreted inconsistently across the 18 natural hazards.

5.1. Natural Hazards

Natural hazards are defined as environmental phenomena that have the potential to impact

societies and the human environment. These should not be confused with other types of hazards,

such as manmade hazards. For example, a flood resulting from changes in river flows is a natural

hazard, whereas flooding due to a dam failure is a considered manmade hazard by the NRI.

Natural hazard events can induce secondary natural hazard events. For example, Landslides can be

caused by an Earthquake. Natural hazards are distinct from natural disasters. A natural hazard is the

threat of an event that will likely have a negative impact. A natural disaster is the negative impact

following an actual occurrence of the natural hazard in the event that it significantly harms a

National Risk Index Primer

National Risk Index Primer 22

community. The NRI only considers primary natural hazard events and not their results or after-

effects.

The NRI considers 18 natural hazards, including: Avalanche, Coastal Flooding, Cold Wave, Drought,

Earthquake, Hail, Heat Wave, Hurricane, Ice Storm, Landslide, Lightning, Riverine Flooding, Strong

Wind, Tornado, Tsunami, Volcanic Activity, Wildfire, and Winter Weather. These hazards are listed

below and described in more detail in the NRI Technical Documentation.

5.2. Natural Hazard Annualized Frequency

The annualized natural hazard frequency is defined as the expected frequency or probability of an

event happening per year. Frequency is derived either from the number of recorded events each year

over a given period or the modeled probability of an event occurring each year. The NRI considers

that natural hazards can occur in places where they may have not yet been recorded to-date and

that hazards may have occurred in locations without being recorded. Therefore, the NRI has built-in

minimum representative frequency values for certain geographical areas and hazards, such as

Hurricane, Ice Storm, Landslide, Tornado, and Tsunami.

5.2.1. SELECTING SOURCE DATA

Annualized frequency data are derived from multiple sources and depend on the natural hazard.

Data sources were identified through public knowledge, guidance by SMEs, and research. Examples

of selected data sources include the National Weather Service (NWS), the National Oceanic and

Atmospheric Administration (NOAA), the US Geological Survey (USGS), the US Army Corps of

Engineers (USACE), the Smithsonian databases, and the US Department of Agriculture (USDA). See

the hazard-specific sections in the NRI Technical Documentation for more information on spatial

data sources.

5.2.2. ANNUALIZED FREQUENCY METHODOLOGY

The natural hazard annualized frequency is the expected frequency for a given hazard event and

measures the actual or expected number of events or event days each year. Not all events are

considered relevant for frequency calculation. SMEs established that some hazards meet certain

criteria to be included as a hazard event capable of causing damage e.g., Hail size of diameter

greater than 0.75 in. (see the hazard-specific sections for more information on these criteria).

Annualized frequency can be defined as the number of historical occurrences of a natural hazard

within a known period of record per geographic area, as seen below in Equation 5:

National Risk Index Primer

National Risk Index Primer 23

Equation 5: Annualized Frequency Equation

In some cases, as with Wildfire and Earthquake, the best available source data consists of

probabilistic statistics contained in raster files which are used to compute an annualized frequency.

In these cases, the frequency value represents the probability of a hazard event occurring in a given

year.

For hazards that track actual hazard occurrences, the historical event count quantifies either the

number of distinct hazard events that have occurred (e.g., Hurricanes to hit the area) or the count of

days on which a hazard has occurred (e.g., on how many days a Hail event was reported). The

determination of whether hazard occurrence was defined by event-days or discrete events was

based on SME review of the source data. This determination depended on how hazard occurrence

was recorded as well as how economic loss was reported. Table 5 gives the frequency basis (event

or event-day) for each hazard.

Table 5: Geographic Level of Event Count Determination and Hazard Occurrence Basis

Natural Hazard

Geographic Level of Historic Event Count

Determination

Hazard Occurrence Basis

Avalanche

County

Distinct events

Coastal Flooding

No event count

Cold Wave

Census Block

Event days

Drought

Census Tract

Event days

Earthquake

No event count

Hail

49-km Fishnet

Event days

Heat Wave

Census Block

Event days

Hurricane

49-km Fishnet

Distinct events

Ice Storm

49-km Fishnet

Event days

Landslide

Census Tract

Distinct events

Lightning

4-km Fishnet (Source raster cell)

Distinct events

Riverine Flooding

County

Distinct events

Strong Wind

49-km Fishnet

Event days

Tornado

49-km Fishnet

Distinct events

Tsunami

Census Tract

Distinct events

National Risk Index Primer

National Risk Index Primer 24

Natural Hazard

Geographic Level of Historic Event Count

Determination

Hazard Occurrence Basis

Volcanic Activity

Census Block

Distinct events

Wildfire

No event count

Winter Weather

Census Block

Event days

While the NRI application reports information at the Census tract and county level, often the data

used to determine this information is captured at either a lower or higher level. Predominantly, EAL

components are assessed at the Census block level, so the number of hazard events (or event-days)

which have historically occurred is determined for each Census block.

Depending on the nature of the hazard and its source data, the event count used to calculate

frequency can be initially captured at the Census block, Census tract, county, or 49-by-49 km fishnet

grid cell level (see each hazard’s frequency section in the NRI Technical Documentation for specific

hazard event count methodology). Table 5 provides the geographic level at which event count

information is determined for use in frequency calculations for each hazard.

For large geographic areas and areas with a statistically significant number of events recorded, the

logic supporting Equation 5 is sound and is used as one approach for calculating annualized

frequency in the NRI for some natural hazards. However, for hazards with few events historically

recorded, due to urban bias and varying demographics across the country, this equation is not

always accurate or representative. Additionally, as geographic boundaries are partitioned into much

smaller regions (counties, Census tracts, and Census blocks), further challenges are uncovered

resulting from the fact that geographic areas that have not been historically impacted by a hazard

and/or recorded hazard events are being calculated as having no risk from that hazard (since the

EAL and NRI risk equation is multiplicative, and therefore any individual factor of zero results in a

total NRI score of 0).

Consider an example (Figure 7) where four Tornadoes hit a single Census tract (e.g. “Tract A”) near

its geographic border. Using Equation 5, the annualized frequency for “Tract A” would be calculated

using a 4 in the numerator. However, given the Tornado event locations (specifically, their proximity

to the neighboring tracts), these four events could easily have occurred within, say, “Tract B”.

Therefore, “Tract B” should not be represented as having no (zero) risk, and, yet, it would be zero if

annualized frequency was deemed to be zero based on the fact that no Tornado has historically

occurred in “Tract B”. Natural hazard events cannot be expected to respect arbitrarily drawn political

boundaries, so, in evaluating risk, hazard occurrence definition should account for events in nearby

Census blocks or tracts that easily could have impacted a given area.

National Risk Index Primer

National Risk Index Primer 25

Figure 7: Example of the Issues with a Simplistic Annualized Frequency Methodology

Three main solutions were incorporated to spread the area of hazard influence used to calculate

frequency and/or exposure. Hazard-specific frequency methodologies may use some or all of these

approaches:

1. Hazard Event Counting Using a 49-by-49 km Fishnet Grid: This approach involves creating a

fishnet grid covering the United States and counting the number of events (or event-days) of

hazard occurrence within each cell. Areas within the cell inherit the event count (or receive

an area-weighted event count when intersecting multiple cells; see the Data Aggregation

section) and frequency is then calculated according to Equation 5. Hazards using this

approach include Hail, Hurricane, Ice Storm, Strong Wind, and Tornado.

2. Minimum Annual Frequency: A minimum annual frequency (MAF) is assigned to areas which

have not experienced a hazard occurrence recorded by the source data, but are determined

to be at some risk due to their location (see the Determining County-Level Possibility of

Hazard Occurrence section). Appropriate MAF values were identified by natural hazard SMEs.

The estimated values were typically low, given the fact that historic events had never been

recorded over the period of record, which sometimes dated back multiple centuries.

Minimum values were typically defined in the format of “once in the period of record,” or

similar. Hazards using this approach include Avalanche, Hurricane, Ice Storm, Landslide,

Riverine Flooding, Tornado, and Tsunami.

3. Hazard Event Shape Buffering: Hazards with widespread and/or unpredictable event

locations are buffered using SME-determined distances to create more representative areas

with potential exposure to natural hazards. Buffering also allows events with relatively small

surface areas to be smoothed together into general representative shapes to eliminate gaps

that may exist between historically recorded hazard events (see Figure 8). Hazards using this

approach include Hail, Hurricane, Strong Wind, Tornado, Tsunami, and Volcanic Activity.

National Risk Index Primer

National Risk Index Primer 26

Figure 8: Example of Buffering Hazard Events to Determine Areas Applicable to Minimum

Frequency Values

Some hazards do not require any of these solutions due to the nature of the source data or the

widespread prevalence of the hazard. For example, the spatial data for Cold Wave, Heat Wave, and

Winter Weather events cover areas the size and shape of NWS forecast zones and counties. These

events can occur across the entire United States, so it is not necessary to spread the hazards’ area

of influence any further.

5.2.3. DATA AGGREGATION

In most instances, annualized frequency is calculated first at the Census block level. In cases where

the event count is evaluated at the fishnet level (see Table 5), the Census block inherits the event

count from the fishnet cell that encompasses it, performing an area-weighted count if a Census

block intersects multiple fishnet cells, as computed in Equation 6. Applying this equation to the

example in Figure 9 results in a Census block event count of about 22. This fishnet-aggregated count

is used to calculate the Census block frequency.

Equation 6: Census Block Area-Weighted Fishnet Event Count Calculation

National Risk Index Primer

National Risk Index Primer 27

Figure 9: Aggregation from Fishnet Cell to Census Block Example

The NRI rolls up data from the Census block to the Census tract and county level, usually by

leveraging area-weighted aggregation as computed in Equation 7. These Census tract and county

level frequency values may not exactly match that of dividing the Census tract and county level

number of historical hazard events by the period of record, as they are based on an area-weighted

aggregation.

Equation 7: Census Tract and County Frequency Aggregations

For a few natural hazards (typically those that are widespread, such as Tsunami or Drought),

annualized frequency is calculated at the Census tract level, after which the Census block simply

inherits the value of its parent tract (see Table 5). Only annualized frequency of the Avalanche and

Riverine Flooding natural hazards are calculated at the county level directly, where the Census tracts

and blocks inherit the value of their parent county.

National Risk Index Primer

5.3. Natural Hazard Exposure

Natural hazard exposure is defined as the representative value of buildings, population, or

agriculture potentially exposed to a natural hazard event. Data sources with the best available

national-level data for each hazard were selected to perform a spatial analysis and compute areas of

exposure.

5.3.1. SELECTING SOURCE DATA

The initial spatial processing of the source data for each hazard is used to identify areas of natural

hazard exposure. Data sources were selected for their accuracy, long period of record, and spatial

component, based on the best available, national-level data per natural hazard. Sources were

identified through public knowledge, subject matter expert recommendations, and research.

Providers of natural hazard exposure data include:

▪ National Oceanic & Atmospheric Administration (NOAA)

▪ USC Hazards & Vulnerability Research Institute (HVRI)

▪ Spatial Hazard Events & Losses Database for the United States (SHELDUS)

▪ United States Army Corps of Engineers (USACE)

▪ United States Geological Survey (USGS)

▪ United States Department of Agriculture (USDA)

▪ National Weather Service (NWS)

▪ Federal Emergency Management Agency (FEMA)

5.3.2. CONSEQUENCE TYPES

A natural hazard consequence is defined in the NRI as economic loss or bodily harm to individuals

that is directly caused by a natural hazard event. Consequences of natural hazard events are

categorized into three different types: buildings, population, and agriculture.

Buildings

Building exposure is defined as the dollar value of the buildings determined by the source data to be

exposed to a hazard according to a hazard-specific methodology. The maximum possible building

exposure of an area (Census block, Census tract, or county) is its building value as recorded in

Hazus 4.2, Service Pack 01 (SP1),

which provides 2018 valuations of the 2010 Census.

Population

Population exposure is defined as the estimated number of people determined by the source data to

be exposed to a hazard according to a hazard-specific methodology. The maximum possible

population exposure of an area (Census block, Census tract, or county) is its population as recorded

Federal Emergency Management Agency (FEMA). (2018). Hazus 4.2, Service Pack 01 Release. Retrieved from

https://msc.fema.gov/portal/resources/hazus

US Census Bureau. (2010). 2010 Census. Retrieved from http://www.Census.gov/2010Census/data/

National Risk Index Primer

National Risk Index Primer 29

National Risk Index Primer

in Hazus 4.2 SP1. The Value of Statistical Life (VSL) was used to express population exposure in

terms of dollars.

Agriculture

Agriculture exposure is defined as the estimated dollar value of the crops and livestock determined

by the source data to be exposed to a hazard according to a hazard-specific methodology. This is

derived from the USDA 2017 Census of Agriculture

county-level value of crop and pastureland.

5.3.3. EXPOSURE METHODOLOGY

Natural hazard exposure is typically calculated at the Census block level and then aggregated to the

tract and county level by summing the block exposure values within the parent tract or parent

county. See each hazard’s exposure section for more information.

Some hazard exposure areas are represented as polygons in the source data, while others are

represented as points, lines, or raster cells. Hazard exposure is based on either historic event

locations or areas of identifiable risk, e.g., Tsunami inundation zones. Eventually, every relevant

record in the source data is processed into a polygon via a hazard-specific methodology. This polygon

represents an area of exposure to the hazard.

To calculate the natural hazard’s representative size for a given area, the NRI leverages a few

techniques, such as using subject matter expertise to define a single representative hazard size,

calculating historical average event occurrence sizes, or defining the size of probabilistic/susceptible

zones for hazards within the area of interest using existing source data (Figure 10).

US Department of Agriculture. (2017). 2017 Census of Agriculture. Retrieved from

https://www.nass.usda.gov/Publications/AgCensus/2017/index.php

National Risk Index Primer

National Risk Index Primer 30

Figure 10: Examples of Representative Hazard Size

To estimate exposure, the hazard event or susceptible area polygons are intersected with the

appropriate administrative layer polygons and the resulting intersect shape defines the area of

hazard exposure. Once the area of exposure is defined, one of three generalized approaches are

executed within the NRI processing database to estimate the exposure value within the

administrative area. The approach used for a natural hazard was determined by the hazard’s

recorded historic events, hazard susceptibility maps, and subject matter expertise. The type of

exposure method used for each of the 18 hazards is described further in the NRI Technical

Documentation. The general approaches to modeling exposure include:

1. Developed Area/Agricultural Area Density Concentrated Exposure. The NRI determined area

of hazard exposure intersected with the administrative area is multiplied by the density of

either the population or building value within the developed land of the area to calculate the

worst-case concentration of hazard consequence. To estimate agriculture exposure, this

method uses the density of crop and livestock value within the agricultural land of the area.

2. Widespread Hazard Event Exposure. The entire Census block is considered to be exposed.

This approach is leveraged for hazards whose extent likely spans the entire area of interest

and whose boundaries are indefinable.

3. Pre-Defined Representative Exposure. Subject matter experts defined a default,

representative exposure value for areas of interest deemed at risk of natural hazard events.

National Risk Index Primer

National Risk Index Primer 31

Approach 1. Developed Area Density Concentrated Exposure

Exposure is calculated for most of the natural hazards using the developed area density approach.

This approach uses the area of the hazard event exposure shape (intersection of hazard shape with

the administrative area) multiplied by the developed area density of the administrative area to

generate the worst-case representative property damage or population that could result from a

future natural hazard event within the area.

The Hazus 4.2 SP1 data provides building value and population estimates at each administrative

reference layer (Census block, Census tract, and county). For certain hazards, a density estimate was

needed for the hazard’s exposure calculation. Rather than only calculating an average density value

for each administrative layer (i.e., by dividing the population of a Census block by the area of the

Census block), effort was made to refine the density estimate by first estimating where people and

buildings might exist within an area. Using the USDA CropScape 2017 raster, which categorizes land

types and use (see Figure 11), a spatial tabulation process was used to derive an estimate of the

developed area within each administrative reference layer. This same tabulation process was used

to estimate the crop and pasture area as well (see the Tabulation section).

With an estimate of the developed area and crop and pasture area for each record of the

administrative reference layers, densities were then calculated. Using the Hazus data’s Building

Stock Value and Population estimates for each administrative layer, the ratio of developed area

within an administrative reference over its whole area was used to calculate the building value and

population densities. These densities represent an assumption that population and the presence of

buildings are concentrated in developed areas rather than being equally distributed across an

administrative area.

Note that, in cases where the Hazus data reports population and or building value and the tabulation

process did not identify any developed land area, the record was assigned an average density value

calculated as the building value (or population) divided by the total area of the record. For cases

where the tabulation process identified developed area but the Hazus data did not report any

population or building values, the densities where set to 0. This ensures that the tabulation process,

which can be spatially imprecise due to the resolution of the source rasters, does not count adjacent

developed area as developed area within the administrative area when Hazus data does not

consider it populated or developed.

To compute the building and population value densities, the building and population values of the

administrative layer (Census block, Census tract, or county) are divided by the total developed area

(determined for the tabulation process) of the administrative layer, as in Equation 8.

Equation 8: Census Block Building and Population Value Density

National Risk Index Primer

National Risk Index Primer 32

where:

is the building value density calculated at the Census block level (in dollars

per square kilometer)

is the total building value of the Census block, as recorded in Hazus 4.2 (in

dollars)

is the total developed area of the Census block, tabulated from CropScape or

NLCD raster files (in square kilometers)

is the population density calculated at the Census block level (in people per

square kilometer)

is the total population of the Census block, as recorded in Hazus 4.2

For agriculture, the USDA 2017 Census of Agriculture provides an estimated dollar value of crop and

livestock within each county. The county value is divided by the total agricultural area of the county

to find its crop value density (see Equation 9). The county level agricultural value density is inherited

by any Census tracts or Census blocks that contain crop or pastureland.

Equation 9: County Crop Value Density

where:

is the agricultural value density calculated at the county level (in dollars per

square kilometer)

is the total crop and livestock value of the county, as reported in the 2017

Census of Agriculture (in dollars)

is the total agricultural area of the county (in square kilometers

National Risk Index Primer

National Risk Index Primer 33

Figure 11: CropScape Developed Land Layer

Approach 2. Widespread Hazard Event Exposure

For certain natural hazards whose extent is widespread with indefinable boundaries, the entire area

of interest is considered exposed. For these natural hazards, exposure values are defined to be the

entire area of interest’s building value, crop and livestock value, or population as recorded by Hazus

4.2 SP1 or the 2017 Census of Agriculture.

Approach 3. Hazard-Specific Representative Exposure

Avalanche and Tornado each have a unique method of calculating exposure. For Avalanche, a single

exposure value, defined by SMEs, is pre-determined and assigned to all areas deemed at risk of

Avalanche events. A review of the source data found that 98% of historical Tornado events impact an

area of 50 km

or less, with the average damage area being 2.07 km

, so a 2 km

area was used to

estimate an average area impacted by a Tornado. This representative footprint area is multiplied by

the average building or population density of the Census tract to find exposure.

5.3.4. DATA AGGREGATION

Natural hazard exposure is calculated at the Census block level and then is aggregated to the tract

and county level by summing the block exposure values within the parent tract or parent county (with

the exception of Avalanche, Drought, Earthquake, and Tornado which are initially calculated at the

National Risk Index Primer

National Risk Index Primer 34

tract level). Detailed methodologies per hazard are explained in the hazard-specific sections of the

NRI Technical Documentation.

5.4. Natural Hazard Historic Loss Ratio

The Historic Loss Ratio (HLR) is an area-specific estimate of the percentage of the exposed

consequence type (building value, population, or agriculture value) expected to be lost due to a

single hazard occurrence. In concept, it is the average of the loss ratios associated with past hazard

events and is used to estimate the potential impact of a future hazard events. To begin the

determination of this value, a Loss Ratio per Basis (event or event-day) (LRB) is calculated for each

historical loss-causing hazard occurrence (for each relevant consequence type) as the value of the

loss divided by the exposed consequence value.

A Bayesian credibility analysis is then performed with the individual LRBs at multiple geographic

levels (county, surrounding area, regional, and/or national) to better balance historic loss accuracy

with geographic precision and characteristics. The resulting HLR (by consequence type) is a

Bayesian-adjusted ratio that is the summed weighted average of various geospatial groupings of the

consequence LRBs at the relevant geographic levels for the hazard. This Bayesian-adjusted resulting

HLR value, computed for each County-Hazard-Consequence type combination, serves as a prediction

of the ratio of loss to exposed consequence value that can be expected from a single hazard

occurrence. Computation of the HLR also considers hazard events which resulted in no loss prior to

performing the Bayesian credibility spatial modelling analysis. This ensures that HLR can be

multiplied by frequency within the risk equation without over-inflating the EAL value.

5.4.1. SELECTING SOURCE DATA: SHELDUS

Historic Losses source data provider: Arizona State University, Spatial Hazard Events and Losses

Database of the United States (SHELDUS)

Arizona State University’s Spatial Hazard Events and Losses Database of the United States

(SHELDUS) loss data were used for most natural hazards. SHELDUS provides county-level data that

correspond to nearly all of the natural hazards represented by the NRI. It offers a further degree of

description by identifying events by peril as well as hazard. SHELDUS aggregates property damage,

crop losses, injuries, and fatalities due to a peril by month, year, and county since 1960. Most of this

data, at the event level, were collected by NOAA and published in the monthly Storm Data and

Unusual Weather Phenomena report, though information for some hazards is extracted from

additional resources.

SHELDUS represents the best available data on economic, population, and agricultural losses due to

natural hazards. However, there are many cases where the geographic precision of the recorded loss

is imperfect. In these cases, the exact location of injuries and fatalities may be unknown due to

regional reporting from the source data interpreted by SHELDUS, often based on a forecast zone that

Center for Emergency Management and Homeland Security, Arizona State University. (2017). Spatial Hazard Events and

Losses Database for the United States, Version 16.0. [online database]. Retrieved from https://cemhs.asu.edu/sheldus

National Risk Index Primer

National Risk Index Primer 35

covers multiple counties. For example, in Table 6, an Ice Storm injury is recorded as 0.5 for two

neighboring counties and both have the same level of property damage. This signifies that the

precise location of the damage associated with this event could not be determined between the two

counties, so the damage is split evenly between them. The NRI utilizes SHELDUS data as it is

compiled and does nothing to alter the source information.

Table 6: Sample SHELDUS Data, Aggregated by Peril, County, and Year-Month

County

FIPS

Year

Month

Peril

Number

Records

Duration

Days

Crop

Damage

(2016 $)

Property

Damage

(2016 $)

Injuries

Fatali-

ties

01001

1996

Hail

3,115.02

18,690.11

32003

1996

WindVortex

7,787.55

05007

2009

Ice

17,643,421.41

0.5

05143

2009

Ice

17,643,421.41

0.5

Data were downloaded at the peril level, aggregated to a county-month level, and mapped via a

control table in the NRI processing database to the appropriate NRI-defined natural hazards. Peril

data were downloaded because natural hazard types as defined in SHELDUS do not directly map into

the natural hazard definitions utilized in the NRI. For example, SHELDUS classifies all flooding perils

under the hazard Flood while the NRI explores two flooding hazards (Coastal and Riverine) and

classifies the different flooding perils accordingly (see Table 7).

Table 7: NRI Hazard to SHELDUS Peril Mapping

NRI Hazard

SHELDUS Perils

Avalanche

Avalanche, AvalancheDebris, AvalancheSnow, SnowSlide

Coastal

Flooding

Coastal, CoastalStorm, FloodCoastal, FloodTidal

Drought

Earthquake

Earthquake, Fire-following Earthquake, LandslideFollowingEQ, Liquefaction

Hail

Heat Wave

Heat, HeatWave

National Risk Index Primer

National Risk Index Primer 36

NRI Hazard

SHELDUS Perils

Hurricane

CycloneExtratropical, CycloneSubtropical, CycloneUnspecified,

HurricaneTropicalStorm, NorEaster, StormSurge, TropicalDepression,

TropicalStorm

Ice

Ice Storm

Landslide

Landslide, LandslideSlump, MudFlow, Mudslide, RockSlide

Lightning

FireStElmos, Lightning

Riverine

Flooding

FloodFlash, FloodIceJam, Flooding, FloodLakeshore, FloodLowland,

FloodRiverine, FloodSmallStream, FloodSnowmelt

Strong Wind

Derecho, Wind, WindStraightLine

Tornado

FireTornado, Tornado, Waterspout, WindTornadic, WindVortex,

Tsunami

Tsunami, TsunamiSeiche

Volcanic

Activity

Ashfall, Lahar, LavaFlow, PyroclasticFlow, Vog, Volcano

Wildfire

FireBrush, FireBush, FireForest, FireGrass, Wildfire

Winter

Weather

Blizzard, StormWinter, WinterWeather

5.4.2. SELECTING SOURCE DATA: NWS STORM EVENTS DATABASE

National Weather Service, Storm Events Database

Unlike the other natural hazards included in the NRI, the loss information for Cold Wave is derived

from the NWS’s Storm Events Database. Loss data for property damage and crop damage is

recorded in the same manner as the SHELDUS data, much of which originates from the Storm

Events Database. Unlike SHELDUS, the Storm Events Database includes natural hazard events with

no reported loss.

Dollar amounts in the Storm Events Database are not inflation-adjusted, so these were converted

to 2016 dollars using the Bureau of Labor Statistics Consumer Price Index

to correspond with the

SHELDUS inflation-adjusted dollar amounts, using Equation 10.

National Weather Service. (2017). Storm Events Database, Version 3.0.

[online database]. Retrieved from

https://www.ncdc.noaa.gov/stormevents/

Bureau of Labor Statistics. (2019). Consumer Price Index for all urban consumers [online dataset]. Retrieved from

https://www.bls.gov/data/

National Risk Index Primer

National Risk Index Primer 37

Equation 10: Conversion to 2016 Dollars

where:

is the dollar value in 2016 dollars

is the original dollar value (assumed dollar value at the time of the loss event)

is the Consumer Price Index for the month of the loss event in 2016

is the Consumer Price Index for the month/year of the loss event

Some loss records in the Storm Events Database are designated with a forecast zone rather than a

county, so each must be joined to a county via a county-zone correlation table with data that is also

provided by the NWS. Cold Wave events also have beginning and end dates recorded, so the number

of event-days can be computed. Cold Wave events extracted from the Storm Events Database use

the same date range as most of the data utilized from SHELDUS, 1/1/1995 to 12/31/2016. The

resulting extracted records mimic the structure of the SHELDUS data in that all records are

aggregated by county, peril, year, and month.

5.4.3. CONSEQUENCE TYPES

The consequence types in the loss data sources are treated as direct corollaries to consequence

types measured for NRI Hazard exposure.

Property

Property loss is defined as the SHELDUS -- or NWS -- reported damage to property caused by the

hazard event in 2016 dollars. In the calculation of HLR, property loss is treated as the equivalent of

building value recorded in Hazus 4.2 SP1. However, SHELDUS property damage can include other

types of property, like vehicles or infrastructure, which would not be reported in the Census data

used by Hazus to estimate building value. This is a caveat to consider when working with this data.

SHELDUS and Hazus data remain the best available estimates of loss and value that could be

utilized for the NRI.

Population

Population loss is defined as the SHELDUS -- or NWS -- reported number of fatalities and injuries

caused by the hazard event. To combine fatalities and injuries for the computation of population loss

value, an injury is counted as one-tenth (1/10) of a fatality.

National Risk Index Primer

National Risk Index Primer 38

The NWS Storm Events Database classifies injuries and fatalities as direct or indirect. For the

purposes of the NRI, both direct and indirect injuries and fatalities are counted in the population loss

value.

Agriculture

Agriculture loss is defined as the SHELDUS – or NWS – reported damage to crops and livestock

caused by the hazard event in dollars. SHELDUS also tracks crop indemnity payments for USDA-

insured crop loss, however the total crop/livestock damage value was considered to be more

inclusive and the crop indemnity data is not used.

5.4.4. HISTORIC LOSS RATIO METHODOLOGY

Conceptually, the Historic Loss Ratio (HLR) is the representative percentage of a location’s hazard

exposure area that experiences loss due to a Hazard, or the average rate of loss associated with the

occurrence of a hazard.

This could be computed as the average of the individual occurrence loss rates (referred to here as

Loss Ratios per Basis). However, HLR cannot be calculated in these simple terms and be considered

accurate. Many counties which have not experienced a loss-causing event during the time period

captured from SHELDUS may be in close proximity to counties which share similar characteristics

that have experienced loss to the hazard. For example, it may be inaccurate to say that a county’s

likely loss ratio to Hurricane is zero just because it has not experienced a loss-causing Hurricane

event during the 22-year window, especially if it borders counties which have experienced loss to

Hurricanes. A better approximation of the HLR is achieved by applying a Bayesian spatial weighting

matrix to smooth the loss ratio data spatially and ensure that historic loss is represented in a rational

way without allowing anomalous Hazard events to distort the data. To implement Bayesian credibility

weighting, loss ratio averages and variances need to be computed for spatial groupings of national,

surrounding, county and, for some hazards, regional levels. The nature of the source data requires

some pre-processing within the database to ensure that all historical hazard events are included in

the loss ratio calculations, including per-basis record expansion of the native SHELDUS records and

the insertion of records representing hazard occurrences which did not result in economic loss. See

the Limitations and Assumptions in Historic Loss Ratio Methodology section for more information.

Loss Record Expansion to per Basis Records

Native SHELDUS and NWS records represent loss aggregated on a county, year, month, and peril

basis. Each row includes the number of reported loss-causing peril events for the month in the

county and the total duration days of the events. For example, the January 2009 Ice Storm event in

Table 8 lasted three days. The basis of Ice Storm occurrences is the event-day as this definition

better captures the variability in duration for Ice Storm events. Without the resolution of knowing

which event-day the damage occurred on, the loss is divided among the days so that each event-day

record has an equal portion of the total loss (see Table 9). In this example, the three event-day

records replace the native SHELDUS record. Similarly, a single native SHELDUS peril month record

for an event-based hazard like Hurricane could describe two separate events. This native record

National Risk Index Primer

National Risk Index Primer 39

would be replaced with two records, each representing a single event with half the loss of the native

aggregated record. Because SHELDUS does not specify the amount of loss associated with each of

the events, each SHELDUS record is expanded based on the occurrence basis (Number of Records

for event basis and Duration Days for event-day basis) if the basis count is greater than one (see

Table 5). This record count expansion process is performed because loss ratios will ultimately be

computed for each event (or event-day) record. Having a record for each hazard occurrence per basis

unit better supports the process of determining loss ratio averages and variance.

Table 8: Native SHELDUS Loss Records

County

FIPS

Year

Month

Peril

Event

Records

Duration

(Days)

Crop

Damage

(2016 $)

Property

Damage

(2016 $)

Injuries

Fatalities

5007

2009

Ice

17,643,421.41

0.5

1097

1998

Hurricane

Tropical

Storm

681,464.65

23,749,724.65

Table 9: Expanded SHELDUS Loss Records

County

FIPS

Year

Month

Peril

Native

Loss

Record

Expanded

per Basis

Crop

Damage

(2016 $)

Property

Damage

(2016 $)

Injuries

Fatalities

5007

2009

Ice

EventDay

5,881,140.47

0.1666

5007

2009

Ice

EventDay

5,881,140.47

0.1666

5007

2009

Ice

EventDay

5,881,140.47

0.1666

1097

1998

Hurricane,

Tropical

Storm

Event

340,732.33

11,874,862.33

0.5

1097

1998

Hurricane,

Tropical

Storm

Event

340,732.33

11,874,862.33

0.5

Loss Ratio per Basis Calculation

After this expansion of records to convert the loss data to loss per single event or event-day is

performed, the Loss Ratio per Basis (LRB) is calculated for each event or event-day occurrence for

each consequence type (building, population, or agriculture) according to Equation 11.

National Risk Index Primer

National Risk Index Primer 40

Equation 11: Loss Ratio per Basis Calculation

where:

is the Loss Ratio per Basis (event or event-day) representing the ratio of loss

to exposure experienced by a specific county due do the occurrence of a

specific Hazard event, performed for each relevant consequence type

(building, population, and agriculture)

is the Loss (by consequence type) experienced from the Hazard event (or

event day) documented to have occurred in the county (in dollars)

is the total value (by consequence type) estimated to have been exposed to

the event or event-day Hazard occurrence (in dollars)

The definition of the HLR exposure variable in the LRB formula does not always match the definition

of the exposure component utilized in the EAL formula. For hazards which can occur almost

anywhere or affect large geographic areas, the HLR exposure is the entire county’s building,

population, or agriculture value. Hazards which only occur in certain susceptible areas, such as

floodplains and tsunami inundation zones, use the HLR exposure value associated with those areas.

Tornado HLR exposure is defined by the area footprint of specific historical Tornado paths.

Avalanche is a unique case which requires the use of default exposure values. The HLR exposure

types utilized for each hazard can be seen in the table below. Specific methods of determining HLR

exposure in the LRB calculation can be found in the HLR section for each hazard. Table 10 lists the

exposure types used in each hazard’s LRB calculation.

Table 10: HLR Exposure Types Used in Loss Ratio per Basis Calculation

Natural Hazard

HLR Exposure Type Used in Loss Ratio per Basis Calculation

Avalanche

Default Value

Coastal Flooding

Value Defined by Hazard Intersect

Cold Wave

Total County Value

Drought

Total County Value

Earthquake

Total County Value

Hail

Total County Value

National Risk Index Primer

National Risk Index Primer 41

Natural Hazard

HLR Exposure Type Used in Loss Ratio per Basis Calculation

Heat Wave

Total County Value

Hurricane

Total County Value

Ice Storm

Total County Value

Landslide

Value Defined by Hazard Intersect

Lightning

Total County Value

Riverine Flooding

Value Defined by Hazard Intersect

Strong Wind

Total County Value

Tornado

Historical Footprint Matched to Specific SHELDUS Loss

Tsunami

Value Defined by Hazard Intersect

Volcanic Activity

Value Defined by Hazard Intersect

Wildfire

Value Defined by Hazard Intersect

Winter Weather

Total County Value

Non-Loss Causing Hazard Occurrence

Hazards may occur without resulting in recorded loss to buildings, population, or agriculture. For

example, Lightning may strike with a high frequency, but have few loss-causing events. SHELDUS

does not record events in which no loss was reported. In an effort to capture events that do not

cause loss, a count of historic year-month events is produced from hazard source data and

compared to a count of loss-producing events from SHELDUS. When the hazard historic event source

records more events than SHELDUS, a number of zero-loss records are inserted into the set of Loss

Ratios per Basis to make up the difference between historic events and loss-causing events from

SHELDUS so that the event counts for both metrics are equal.

Computing loss ratio averages and variances without including the zero-loss records produces very

different results than when they are included. For example, a county with 100 historical Lightning

strikes may only have two loss-causing events, one causing $40,000 in damage to buildings and the

other causing $60,000. If the building exposure value is $10M, the loss ratios for each loss-causing

event would be 0.004 and 0.006, respectively. If only the LRBs for two loss-causing events were

considered, the average would be 0.005. Including the 98 Lightning strikes that did not result in loss

lowers the average to 0.0001, a more accurate approximation of the average Lightning strike’s

impact on the county as not every Lightning strike is a loss-causing event.

The output of the Loss Ratio per Basis calculation (see Equation 11) and all corrective record

insertion is stored in the LRB table within the NRI processing database, and are then used to

National Risk Index Primer

National Risk Index Primer 42

compute Bayesian metrics and calculate the weighting factors that are applied to find the Bayesian-

adjusted HLR for each consequence type for the county. Table 11 illustrates the content of the LRB

database table after the corrective record insertions. Notice the loss ratios for three Ice Storm event-

days in one county in January of 2009. These have been expanded from a single SHELDUS record

based on duration days and consequence types. Also, one zero-loss record for each relevant

consequence type has been inserted to recognize an Ice Storm event-day which occurred within the

county (based on the historical event source data) but resulted in no economic loss. These records

can then be used to calculate loss ratio averages and variance.

Table 11: Sample Data from the Loss Ratio per Basis Table

Bayesian Credibility

To apply Bayesian credibility weighting factors and balance Historic Loss accuracy with geographic

precision in areas where small sample sizes result in volatile Historic Loss estimates, LRB averages

and variance may be calculated at the level of: county, surrounding 196-by-196-km fishnet grid

Hazard

Peril

Basis

Year

Month

Conseq.

Type

Conseq.

Exposure

Conseq. Loss

per Basis

Conseq.

Ratio per

Basis Unit

Record

Type

Ice

Storm

Ice

Event

Day

2009

People

221339

0.01666667

7.53E-08

Peril Basis

Expansion

Ice

Storm

Ice

Event

Day

2009

People

221339

0.01666667

7.53E-08

Peril Basis

Expansion

Ice

Storm

Ice

Event

Day

2009

People

221339

0.01666667

7.53E-08

Peril Basis

Expansion

Ice

Storm

Ice

Event

Day

2009

Property

2.3138E+10

5881140.47

0.00025

Peril Basis

Expansion

Ice

Storm

Ice

Event

Day

2009

Property

2.3138E+10

5881140.47

0.00025

Peril Basis

Expansion

Ice

Storm

Ice

Event

Day

2001

People

221339

SHELDUS

Native

Record

Ice

Storm

Ice

Event

Day

2001

Property

2.3138E+10

310468.525

0.0000134

SHELDUS

Native

Record

Ice

Storm

Inserted

Zero-

Loss

Record

Event

Day

People

221339

Inserted

Zero-Loss

Record

Ice

Storm

Inserted

Zero-

Loss

Record

Event

Day

Property

2.3138E+10

Inserted

Zero-Loss

Record

National Risk Index Primer 43

National Risk Index Primer

cell,

region, and national. These geographic levels define which spatial grouping (or set) of LRBs

are used to calculate the average and variance values. The county level grouping includes all LRBs

for the county, the surrounding grouping includes LRBs for all counties that intersect the same 196-

by-196-km fishnet cell, and national includes all LRBs. The formulas in Equation 12 illustrate the

computation of the loss ratio average and variance.

Equation 12: Geographic Level Consequence Ratio Average and Variance Computations

where:

is the Average value of all Loss Ratio per Basis (event or event-day) (LRB)

records of the consequence type for the geographic level due to the

Hazard

is the Loss Ratio per Basis (event or event-day) of the consequence type

within the geographic area due to the Hazard occurrence basis

is the total number of records of Hazard events or event-days occurring in

the geographic area (includes any non-loss causing events/event-days

identified)

is the consequence LRB variance of the geographic level due to the

Hazard

Credibility increases as a function of sample size and decreased LRB variance. In other words, the

higher the credibility at a given geographic level, the higher the contribution to the location’s

calculated Historic Loss value. Figure 12 illustrates possible loss ratio variance in neighboring

The 196-by-196 km fishnet grid cell is roughly the area of four average counties. See the Intersection section for more

information on the use of the 49-by-49 km fishnet resolution to represent average county area.

National Risk Index Primer

National Risk Index Primer 44

counties. Weighting factors in the Bayesian credibility calculation are what determines the

contribution of each geographic level to the final HLR value.

Figure 12: Example of Variance in County Loss Ratio Values

Weighting factors are derived from the variance values (calculated using Equation 12) at each

geographic level according to Equation 13. For the surrounding fishnet level, if the county intersects

more than one fishnet grid cell, the cell with the lowest LRB variance value is used as this provides

the data with the best fit.

National Risk Index Primer

National Risk Index Primer 45

Equation 13: HLR Bayesian Weighting Factor Calculation

where:

is the sum of the inverted variances calculated at each geographic level,

used as a denominator for the level weighting factors

is the weighting factor to be applied to the average consequence LRB for the

Hazard at X level (national, regional, surrounding, county)

is the consequence LRB variance for the Hazard at X level (national, regional,

surrounding, county)

For several hazards, regional Bayesian HLR weighting supplies a more accurate estimation of historic

loss for areas which have not experienced economic loss due to hazard events during the hazard’s

period of record. This is especially true for areas whose hazard frequency and severity are dependent

on their geographic location and climate. For example, Ice Storm, Winter Weather, and Cold Wave

will have a very different degree of impact on the Northeast than on the Southwest. For this reason,

the Bayesian spatial weighting incorporates regional weighting rather than national for these

hazards.

Most hazard-specific HLR region definitions are derived from the FEMA administrative region

definitions, the only difference being that FEMA Regions I, II, and III are merged to form a region

whose size is closer to that of the other regions (see Figure 13). The definition of regions for

Hurricane utilizes the FEMA administrative region definitions, but further divides them into coastal

regions (for the East and Gulf coasts) and inland regions along a county-level boundary which

National Risk Index Primer 46

National Risk Index Primer

approximates the hurricane prone regions identified in the American Society of Civil Engineers

(ASCE) 7-05, Minimum Design Loads for Buildings and Other Structures (see Figure 14).

Figure 13: Historic Loss Ratio Region Definitions

American Society of Civil Engineers. (2005). Minimum design loads for buildings and other structures (ASCE/SEI 7-05).

Reston, VA: American Society of Civil Engineers.

National Risk Index Primer

National Risk Index Primer 47

Figure 14: Hurricane Historic Loss Ratio Region Definitions

The Historic Loss Ratio for each relevant consequence type is calculated as the sum of its weighted

average county, surrounding fishnet, regional, and national average LRB (see Equation 14). Levels

not used for a specific hazard are removed from the computation.

Equation 14: County Bayesian-Adjusted HLR Calculation

where:

is the Historic Loss Ratio for the Hazard at the county level, by consequence

type

National Risk Index Primer

National Risk Index Primer 48

is the average LRB by consequence type for the Hazard at X level (national,

regional, surrounding, county)

is the weighting factor applied to the LRB by consequence type for the Hazard

at X level (national, regional, surrounding, county)

HLR Inheritance

The Bayesian-adjusted county Historic Loss Ratio is inherited by the Census blocks and Census

tracts within the county when used in the NRI EAL calculations, as in Equation 15.

Equation 15: Census Tract and Census Block HLR Inheritance

where:

is the Bayesian-adjusted Historic Loss Ratio, a hazard-county-consequence

type specific value

is the Inherited Historic Loss Ratio for the Hazard at the Census tract level

is the Inherited Historic Loss Ratio for the Hazard at the Census block level

5.4.5. LIMITATIONS AND ASSUMPTIONS IN HISTORIC LOSS RATIO METHODOLOGY

Several factors are not entirely accounted for in the calculation of HLR. Certain processes, such as

Bayesian credibility adjustments, attempt to correct some of these limitations. This section

addresses some of the assumptions that are intrinsic within the current methodology and how these

can limit the accuracy of the calculation.

Evaluating historic economic loss from SHELDUS over a relatively brief period of time and comparing

it to a static HLR exposure value does not account for changes in development patterns over these

years. For example, a hazard event in 1995 may have a low HLR when its loss is compared to its

2010 Hazus-derived exposure value, though because of increased development and population

influx over the years, its HLR would be much higher if the same loss were compared to the actual

1995 exposure value. There is an inherent assumption in the methodology that all buildings,

population, and agriculture exposed to the hazard are static in economic value and quantity over the

data period. Additionally, the SHELDUS loss values are inflation-adjusted to 2016 dollars while

National Risk Index Primer

National Risk Index Primer 49

Hazus-derived exposure values are in 2018 dollars based on 2010 valuations and there is an

assumption that these dollar values are comparable.

Since the HLR calculation is based on historical events, it does not project reductions due to

enhanced mitigation efforts and improved building standards that have changed over time (i.e., a

seawall being built after a destructive flooding event may reduce the damage caused by subsequent

flooding events).

Characterizing agriculture losses from events is highly complex and can vary based on a number of

factors, including supply and demand, substitution effects, crop rotation, and seasonality. The

simplified HLR calculations use crop and livestock distribution and values based on agriculture data

from CropScape and the Census of Agriculture.

There are many cases where the geographic precision of the recorded loss is imperfectly captured in

SHELDUS. The regional reporting data used to compile SHELDUS may mention multiple counties for

a loss-causing event. In these cases, the loss is spread equally over the counties where the hazard

occurred, though the loss may have only occurred in one county. Also, loss may only occur in a

portion of the county, yet the HLR will apply to the entire county due to loss not being recorded with

any granularity below the county level.

5.5. Validating Expected Annual Loss Estimates to Historical Losses

The diversity of the hazards and source data included in the calculation of the NRI presents a

significant challenge to provide accurate and meaningful results for the variety of potential lenses

through which the results may be viewed, such as:

▪ Hazard EAL rankings within a county;

▪ County EAL rankings within a hazard;

▪ County EAL rankings across all hazards;

▪ Hazard EAL rankings all counties.

As an attempt to validate the EAL, historic loss from SHELDUS for the period from 1995 to 2016 was

aggregated for the entire nation for each hazard and divided by the period of record (22 years) to

give a rough nationwide hazard annualized loss estimate.

This value was compared to the

aggregated EAL estimate calculated for the NRI for its corresponding hazard. All but two (Earthquake

and Volcanic Activity) of the natural hazard EALs are within the same order of magnitude as the

experienced historic losses and half of the hazards are within a factor of 2.

When evaluating the historical record, losses for some hazards are driven by relatively few events.

For example, from 1960 to 2016, 50% of all hurricane consequences were caused by only 8 storms.

Similarly, from 1995 to 2016, 50% of all riverine flooding consequences were caused by only 48

events. The same pattern applies to Earthquakes and Volcanic Activity. These events are statistical

For Cold Wave, the historic loss data was aggregated from the NWS Storm Events Database for 1996 to 2016 and

divided by the 21-year period of record.

National Risk Index Primer

National Risk Index Primer 50

outliers where high-value urban areas have been impacted by severe hazard events. For Wildfire and

Earthquake, annualized frequency uses probabilistic statistics to compute an annualized frequency.

Use of probabilistic data to calculate EALs for these hazards account for the probability that the

outlier event may occur. Reliance on historical data alone for the other hazards will generally

underestimate the EALs for hazards where losses are driven by the rare catastrophic events. For this

reason, Hurricane EALs are significantly lower (~75%) than their historical losses. This is because, for

every severe hurricane that directly strikes a major city, there may be dozens of glancing blows from

minor hurricanes or tropical storms that cause minimal damage. The HLR approach calculates an

average value; so, HLRs are weighted toward the more common, lower loss events rather than the

rare catastrophic events.

Despite these outliers, a relatively high level of agreement between the NRI-calculated EAL and the

historical loss records serves as an indication that the NRI estimated annual hazard loss is fairly

aligned with actual recorded historic loss.

6. Using the National Risk Index

The NRI is available to the public through https://www.fema.gov/nri. FEMA provides access to the

NRI data and information through multiple venues, including a website, an interactive map and data

exploration tool, tabular and spatial dataset files, and GIS-based REST services.

6.1. The NRI Website

The NRI website is the hub of access to the vast array of information in the National Risk Index. The

website provides an overview of the NRI and links to documentation with important details about the

source data and source data providers, descriptions of the methodology, and guidance on

interpreting the results. With the interactive mapping tool, users can visually explore components of

the NRI dataset and then delve into any location and examine its risk factors.

6.2. Downloadable and Online Datasets

File-based versions of the NRI dataset can be retrieved from the Data Download feature of the NRI

website. Tabular and spatial formats are provided for both Census tract-level and county-level

datasets. Tabular data, provided as CSV files, can be used in a wide variety of applications, and

shapefiles are available for spatial applications.

The NRI dataset can also be used from web services that are hosted in the FEMA Hazards

Geoplatform and accessible through ArcGIS Online. These services are a convenient way to explore

the data with online tools other than the NRI website, and developers can leverage the REST

services to integrate NRI data into their own applications.