Observing Community Resilience from Space: Using
Nighttime Lights to Model Economic Disturbance and
Recovery Pattern in Natural Disaster
YiQianga, Qingxu Huangb, Jinwen Xua
aDepartment of Geography and Environment, University of Hawaii – Manoa, USA
bCenter for Human-Environment System Sustainability (CHESS), State Key Laboratory of Earth
Surface Processes and Resource Ecology (ESPRE), Beijing Normal University, China
Published in Sustainable Cities and Society in February 2020.
Due to climate change and rapid population growth, human society is faced with increasing threats
from natural disaster that can cause significant socio-economic consequences. Coastal
communities around the world are particularly vulnerable to natural disasters including both large-
scale rapid-moving disturbances such as hurricane and storm surges (Tebaldi et al., 2012), and the
slow-moving processes such as coastal erosion, sea level rise (Nicholls et al., 1999) and reduction
of ecosystem services (Spalding et al., 2014). According to the data from U.S. Census Bureau
(2011), 39% of the total population in the United States are living in counties directly on the
shoreline and the population density in coastal counties is more than four times the average density
of the whole United States. Since 2005 when Hurricane Katrina and Rita caused catastrophic
damage in Central Gulf Coast, much attention has been paid the resilience and long-term
sustainability of coastal communities. Empirical observations suggest that, under the same strength
of disaster, different communities endured different levels of disturbance and presented different
recovery patterns in socio-economic (Finch et al., 2010; Fussell et al., 2010), health (Burton, 2006;
Sastry and VanLandingham, 2009), and psychological conditions (Adeola, 2009). These observed
disparities can be attribute to various resilience of the communities.
Resilience describes the ability of an individual or a system to adapt to and recover from external
shocks or stresses (Adger, 2000). Although substantial knowledge has been gained on ecological
resilience (Perz et al., 2013) and engineering resilience (Yodo and Wang, 2016), there is yet a
consensus on how to measure resilience of human communities due to their complexity. In general,
quantitative assessment of community resilience is challenged by two issues. First, the definition
of community resilience various in different domains, which will be discussed in Section 2.1.
Moreover, resilience is often used interchangeably with other relevant concepts such as
vulnerability and adaptive capacity. The various definitions and conceptual frameworks of
community resilience influence how researchers select indicators and models to measure resilience
(Cutter et al., 2008; Lam et al., 2016; Sherrieb et al., 2010). The definition disagreement hampers
the development of standard metrics to measure resilience. Second, there is lack of empirical data
and approaches to quantify community resilience. Most of the existing assessments are based on
an index approach, which integrates a set of presumed indicators into a composite score to measure
resilience (Cutter et al., 2010; Hung et al., 2016; Sempier et al., 2010; Sherrieb et al., 2010). The
model specification, indicator selection and weighting are based on prior knowledge or expert
opinions. Although these indices provide general guidance for predicting community resilience,
their accuracies have not been validated against empirical observations in disasters (Bakkensen et
al., 2017; Beccari, 2016).
Empirical data about human activities and states are difficult to obtain in a disaster condition when
many social systems fail to function. Traditional data sources for resilience assessment (e.g.
surveys and census data) have limitations in various aspects, which will be elaborated in the next
section. Recently, remote sensing imageries become popular instruments to monitor human
dynamics on the earth surface such as urban growth (Shahtahmassebi et al., 2016), land cover
change (Joshi et al., 2016) and socio-economic conditions (Kuffer et al., 2016). Among the various
remote sensing products, night-time light (NTL) remote sensing has unique ability to capture
fluctuations of human activities, which can provide empirical data for resilience assessment. This
study introduces a quantitative framework to assess community resilience using the DMSP-OLS
NTL annual composite images as the data source. Specifically, stable lights in the time series of
DMSP-OLS annual images are used as a proxy to model recovery patterns of economic activity in
Hurricane Katrina in 2005. Spatial and statistical analyses are conducted to explore the
geographical disparities of the recovery patterns and their relationships with selected resilience
indicators. Specific questions answered in the case study are: 1) which communities appeared to
be more or less resilient in the disaster; 2) how the observed resilience levels are associated with
the environmental and socio-economic conditions? The introduced framework aims to fill the
critical gap of empirical data and assessment methods for community resilience. The analysis
results from the case study increase our understanding about community resilience and provide
actionable information to predict and prompt community resilience.
The rest of the article is organized as follows. Section 2 briefly reviews the related work about the
definitions, conceptual frameworks and assessment methods of community resilience. Section 3
introduces the data sources, assessment framework of community resilience based on NTL data
and statistical analyses. Section 4 presents the analysis results in the case study of Hurricane
Katrina, followed by the discussions in Section 5 and conclusions in Section 6.
2 Related Work
2.1 Definition and Conceptual Framework
The concept of resilience was first introduced by Holling (1973), who views resilience as the
ability of an ecological system to absorb change in the face of extreme perturbation and yet
continue to persist. Later, Timmerman (1981) applied the concept of resilience to social systems
and defined resilience as the measure of a system’s capacity to absorb and recover from disastrous
events. The resilience of a social system is also known as community resilience. Extending
Timmerman’s definition, Cutter et al., (2008) further elaborated that community resilience
includes both the inherent conditions of a system to absorb impacts and cope with an event and
post-event, adaptive processes that facilitate the ability of the social system to re-organize, change,
and learn in response to a threat. Norris et al., (2008) considered resilience as a process linking
communities’ capacities in response to the disturbance. In the field of engineering and
infrastructure systems, resilience describes the ability of resisting and absorbing disturbances and
the ability of adapting to disruptions, and returning to normal functionalities (Faturechi and Miller-
Hooks, 2015). Extensive reviews about the definitions of resilience and the related terms can be
found in (Cutter et al., 2008; Lam et al., 2016; Liao, 2012; Peacock, 2010). Despite the various
definitions in the literature, community resilience is often associated with two abilities: 1) the
ability to absorb/resist/withstand disturbance, and 2) the ability to respond/recover/restore the
acceptable level of functioning and structure.
In addition to the qualitative descriptions, a number of theoretical frameworks have been
developed to quantify community resilience. For instance, Cutter et al. (2010, 2014) define that
resilience consists of six components including social, economic, infrastructural, institutional,
community, and environmental, which is used as a guidance to select indicators for resilience
indices. Lam et al. (2016) measure resilience from the relationships among exposure, damage and
recovery. Additionally, resilience can be conceptualized as a dynamic process. The recovery
trajectory (also known as recovery curve) describes the continuous change of a functional capacity
of a system affected by a disturbance (e.g. natural disaster). The functional capacity could be the
social and economic capacity of a human community (White et al., 2015), biomass or population
of an ecological community (Qiang and Xu, 2019; Vercelloni et al., 2019), or the functionality or
serviceability of an infrastructure system (Koliou et al., 2018). The Curve A, B and C in Figure 1
illustrate common scenarios of resilience from high to low, where the functional capacity suddenly
declines after a disturbance, gradually recovers afterwards, and finally restores to the pre-disaster
condition (e.g. Trajectory B) or a new equilibrium (Trajectory A and C). Given the same strength
of disaster, the variation of the recovery trajectory is indicative of community resilience. The
maximum deviation from the pre-disaster condition (i.e. maximum disturbance) reflects the ability
of a system to absorb/resist/withstand disturbance from the disaster. The recovery speed or time
indicates the ability to respond/recover/restore the functional capacity.
Figure 1: Recovery trajectory of a system due to an external disturbance. Curve A, B and C represent three
possible scenarios of resilience from high to low. (Modified from White et al., 2015)
2.2 Resilience Assessment – Index Approaches
Previous work of resilience assessment is mostly based on an index approach, which aggregates a
number of socio-economic and environmental indicators into an overall resilience index. The
Baseline Resilience Index for Communities (BRIC) developed by Cutter et al., (2010) is one of
the first and most cited resilience index. Analogous to the previous work of the Social Vulnerability
Index (SoVI) (Cutter et al., 2003), BRIC is aggregated from 36 indicators of the baseline
characteristics of community resilience. The indicators are rescaled into [0,1] and then aggregated
in five categories including social, economic, institutional, infrastructure, and community. The
overall BRIC index is the summation of the aggregated indices in the five categories. Analogously,
the Community Disaster Resilience Index (CDRI) by (Peacock, 2010) categorize resilience
indicators into a 4×4 matrix with capital domains (i.e. the social, economic, physical, and human
capital) and disaster phases (i.e. the mitigation, preparedness, response and recovery phase). The
selected indicators are aggregated in the 16 categories in the matrix, which are then aggregated
into the CDRI index. Other work on resilience indices include (Foster, 2012; Hung et al., 2016;
Sherrieb et al., 2010). Although these indices provide general predictions of resilience by
integrating prior knowledge and expert opinions, they do not inform specific disaster outcomes.
For instance, it is unclear whether a high resilience index implies low economic loss, low casualty
and injury, or fast economic recovery. The unspecified outcomes diminish the value of the indices
for specific decision-making. Moreover, most of the resilience indices have not been calibrated or
validated against empirical observations. Bakkensen et al., (2017) validated the BRIC and CDRI
with observed losses, fatalities, and disaster declarations, and found low or even contradictory
correlations between the indices and disaster outcomes.
2.3 Resilience Assessment – Empirical Approaches
In addition to the index approach, efforts have been made to assess community resilience using
empirical observations in disasters. For instance, Lam et al., (2009) and LeSage et al., (2011)
conducted a series of telephone and street surveys to continuously monitor business reopening in
New Orleans after Hurricane Katrina. The time series of open businesses resemble the recovery
trajectory illustrated in Figure 1, which declines sharply after Katrina and gradually recovers. The
various recovery patterns in different communities were associated with environmental and socio-
economic variables to explain why some communities restored businesses more quickly than
others. However, the surveys were costly, labor-intensive and time consuming, which is not widely
applicable. Later, Lam et al., (2015, 2016) developed the Resilience Inference Model (RIM) which
uses the number of disasters, damage and population growth as proxies to measure community
resilience. Based on aggregated data in a 10-year period, the RIM model assumes that a resilient
community can resist damage and maintain high population growth while endured a high number
of disasters. Recently, with the advent of the Big Data era, data crowdsourcing and social media
platforms provide new opportunities to observe individuals’ activities and narratives at finer spatial
and temporal resolutions. For instance, Zou et al., (2017) used the frequency and sentiment of geo-
tagged Twitter messages (tweets) to monitor dynamic conditions of communities in Hurricane
Sandy. The recovery trajectories of communities can be reflected from time series of ratios and
average sentiment of tweets during the disaster. Timeliness, low cost and scalability are the main
advantages of social media data. However, the Big Data approaches are criticized for the biased
user profile (Zou et al., 2018) and low data quality (much noise and misinformation) (Li et al.,
2.4 Remote Sensing for Disaster Management
Remotely sensed imageries have been widely applied in disaster risk mapping (Bates, 2004; Hong
et al., 2007) and damage assessment (Cooner et al., 2016; Dong and Shan, 2013; Vetrivel et al.,
2018). However, most remote sensing products are not very useful for resilience assessment due
to their insensitivity to decline of human activity. For instance, it is difficult to detect a ‘ghost town’
or a damaged city from Landsat images. As an alternative, nighttime light (NTL) remote sensing
images have been proved an effective means to observe the dynamics (both increase and decline)
of population (Zhuo et al., 2009), urbanization (Xie and Weng, 2017) and economic activities (Li
et al., 2013). A comprehensive review of the applications of NTL data can be found in (Huang et
al., 2014). For disaster management, the NTL data have been applied to identify damage
(GILLESPIE et al., 2014; Kohiyama et al., 2004), power outage (Hultquist et al., 2015; Zhao et
al., 2018), and analyze the change of human activities (Li et al., 2018) and urbanization (Huang et
al., 2019) affected by disasters. Despite these applications, the utility of NTL data in modeling
community resilience has not been fully exploited in a theoretical framework. Extending the
conceptual framework of recovery trajectory, this study introduces a quantitative approach to
model resilience using DMSP/OLS NTL images as the data source.
3.1 Inter-Calibration of NTL Images
The Stable Lights images collected by the Defense Meteorological Satellite Program Operational
Line Scanner (DMSP/OLS) were used for this study. The images are cloud-free composites created
using all the available archived DMSP-OLS smooth resolution data from the year 1992 to 2013.
The DMSP/OLS images include 34 annual composites at a 30 arc second resolution collected by
six different satellites (F10, F12, F14, F15, F16, and F18). Due to the absence of inter-satellite
calibration and onboard calibration, the digital numbers (DN) in the DMSP-OLS images cannot
be converted to exact radiance. To analyze continuous recovery trajectories over time, the DMSP-
OLS images need to be calibrated to make the images in different years and satellites comparable.
A widely applied inter-calibration procedure was proposed by (Elvidge et al., 2009), which uses a
quadratic polynomial regression to adjust the DNs against a reference image (see Equation 1).
The inter-calibration process follows the same procedure as introduced in (Elvidge et al., 2009).
By reviewing the data, it was found that the image F121999 has the highest average DN in the
United States. Due to the saturation of the DMSP/OLS in bright areas (e.g. city centers), the
F121999 was used as the reference image and all other images were calibrated to match the DNs
in F121999. Los Angeles was chosen as the reference site, as it has been a mature metropolis where
the light change is negligible (Hsu et al., 2015). Due to the uneven distribution of DNs in the
images, a random sampling will lead to overfit near the two extremes of DNs (i.e. 1 and 63) where
pixels are concentrated. To ensure the regression equations equally fit the entire value range, a
stratified sample of lit pixels (200 pixels in each DN value) were extracted in the reference site for
the calibration. The 2nd order regression equation was calibrated for each image with the reference
to F121999. The coefficients of the regression equations are in Table 1. After the calibration,
images in the same year were averaged into one image, leading to a time series of annual images
from 1992 to 2013.
Table 1: Coefficients of the quadratic polynomial regression for inter-calibration.
3.2 Estimation of Gross Domestic Product
The inter-calibrated DMSP-OLS images were clipped in the affected area in Hurricane Katrina,
which include 179 counties declared as disaster areas by the Federal Emergency Management
Agency (FEMA). Hurricane Katrina made the first landfall in Florida on August 25th, 2005 and
the second in Louisiana on August 29th, 2005. The disaster areas include the entire Louisiana (64
parishes) and Mississippi (82 counties), 22 counties in Alabama, and 11 counties in Florida (Figure
2). Zonal operation was applied to aggregate the DNs within individual counties. After logarithm
transformation, the sum of DNs and number of lit pixels can well predict (= 0.92) the gross
domestic product (GDP) in the 179 counties in a linear model (Equation 2). The goodness of fit
(i.e. ) is highest for GDP compared to other economic indicators (e.g. personal income, number
of employees and business establishments). As county-level GPD data in the U.S. is only available
in 2012-2015 (BEA, 2018), the GDP data and DMSP-OLS images in 2012 and 2013 (the
overlapping years of the two data sources) were used to derive the regression equation. All the
GDP data are adjusted to the value of current U.S. dollar. Finally, the GDP of the counties from
1992 to 2013 was estimated using Equation 3.
) = 3.375 log(
Figure 2: The track of Hurricane Katrina and counties that are presidentially declared as disaster area.
3.3 Measurement Framework
The estimated GDP was analyzed in the framework in Figure 3 to assess the resilience of the
counties. The GDP of each county is rescaled into [0, 1] to be comparable. Two regression lines
were derived for the normalized GDP from 1992 to 2004 and from 2005 to 2013 to represent the
pre- and post-Katrina economic trajectory respectively. Although various models are available for
GDP projection, the linear model is still considered a valid benchmark to project GDP growth in
the U.S. (Ferrara et al., 2015; Marcellino, 2008), especially in the recent decades when the
volatility of US GDP growth tend to decline (Burren and Neusser, 2010). Despite the slight decline
(-1.8%) in 2009 due to the financial crisis, the annual GDP in the U.S. from 1990 to 2018 generally
follows a linear trend (World Bank, 2019). The extension of the pre-Katrina trajectory (thin dashed
line in Figure 3) represents the business-as-usual condition as if the pre-Katrina economic growth
Three metrics were calculated for each county. First, the difference between the projected GDP in
the business-as-usual trajectory ( (2005)) and the actual 2005 GDP in the post-Katrina
trajectory ( (2005)) was calculated to represent instant economic disturbance caused by
Katrina (Equation 4). This metric (denoted as D for simplicity) measures the ability of a system to
absorb/resist/withstand disturbance in a disaster. Second, the difference between the slopes of the
post-disaster trajectory () and the business-as-usual trajectory () was calculated to indicate the
recovering rate () of GDP after Katrina (Equation 5). A high (positive) would implies a
strong ability to respond/recover/restore the GDP growth to catch up with the business-as-usual
trajectory. Third, the accumulated difference between the post-disaster GDP and the business-as-
usual GDP from 2005 to 2013 was calculated using Equation 6. This metrics measures the
accumulated economic loss (L) due to the deviation of GDP growth from the business-as-usual
trajectory, which is illustrated as the grey area in Figure 3. L represents the combined effect of D
and RR. A high L can be interpreted as low resilience, which typically consists of a high instant
disturbance (D) and slow recovery (Rr). A low L means the opposite. L differentiates the
intermediate situations such as high D and high RR or low D and low RR. L is calculated from 2005-
2013 after Katrina when the DMSP-OLS images are available.
Figure 3: Framework of GDP trajectory in Hurricane Katrina.
3.4 Regression Analysis
Regression analyses were applied to examine the associations of the three resilience metrics (D,
RR and L) with environmental and socio-economic variables. The selected variables fall into four
categories, including impact intensity, environmental, socio-economic, and industrial structure.
Max wind gust speed and accumulated rainfall represent the destructive power of a hurricane
(NOAA, 2006). Elevation, proximity to coast and ratio of urban in flood zones indicate the
exposure to Hurricane-induced storm surge and flooding (Dasgupta et al., 2011; Jonkman et al.,
2009). The socio-economic category includes 10 commonly used indicators in community
resilience measurements (Cai et al., 2018; Cutter et al., 2014). Ratios of establishments in different
scales and industrial sectors represent the industry structure, which potentially influences
economic recovery (Martin and Sunley, 2015). Most of the variables were acquired from datasets
released before 2005 to represent pre-disaster conditions. Variables that are not reported in
counties were preprocessed and averaged into counties (Table 2). Instead of composing a
comprehensive index for resilience, the objective of the analysis is examining the relationships
between the GDP recovery trajectories and the hypothetical resilience indicators.
Two types of regression analysis were carried out in this study. First, univariate regression was
applied to examine the relationships between D, RR and L and each individual variable. Second,
multivariate regression was utilized to relate the three metrics with all the variables and variables
in the four categories. R2 of the multivariate regression indicates the proportion of variance
explained by the different categories of variables. Adjusted R2 were compared to evaluate the
prediction power of the different categories for the recovery metrics. The regression analyses aim
to explore the underlying factors that influence resilience and proportion of variance of resilience
can be explained by the variables. All variables were re-scaled into z-scores for the regression
analyses. Thus the b coefficient of the regression represents the direction and standard deviations
of the relationship. The regression analyses were conducted in the linear model (lm) function in R.
Table 2: Description of variables used in the regression analyses.
Recorded maximum gust speed from
Aug. 23 to 30, 2005
Knobb et al. (2005)
Total rainfall from Aug. 23 to 30,
Distance to Coast
Mean distance to coast line
Euclidean distance, Zonal
% of urban in
Percent of developed land in flood
FEMA flood map,
National Land Cover
Method in (Qiang, 2019)
Percent of population in one race:
U.S. decennial census
Percent of population in one race:
Black or African American
U.S. decennial census
Percent of population in one race:
U.S. decennial census
% Hispanic &
Percent of Hispanic or Latino
U.S. decennial census
Percent of population under 18 year s
U.S. decennial census
% elderly adult
Percent of population above 65
U.S. decennial census
Percent of owner occupied housing
U.S. decennial census
Percent of population (> 25 years
old) with 4 or more years of college
or bachelor's degree or higher
U.S. decennial census
Percent of population whose income
is below poverty level
U.S. decennial census
Per cap. income
Per capita income
U.S. decennial census
Percent of establishments with < 20
Percent of establishments with >500
Percent of establishments in
agriculture, forestry, fishing and
Percent of establishments in mining,
quarrying, and oil and gas extraction
Percent of establishments in
4.1 Overall Economic Impact
Figure 4 demonstrates the change of NTL brightness (DN value) from 2004 to 2005 near the
landfall location of Katrina, where a decline of NTL brightness can be observed in New Orleans
and the surrounding coastal cities. The annual GDP estimates from the NTL data (see Figure 5)
reveal that Hurricane Katrina has fundamentally altered the economic growth in the declared
disaster area. In contrast to the steady growth in the Southeast Region (despite the slight drop in
2009 due to the financial crisis), the GDP in the disaster area declined sharply in 2005 and grew
at a slower rate in the following years. The Southeast Region is delineated by Bureau of Economic
Analysis (BEA) (2015) for releasing economic statistics (see Figure 2). According to the
estimation, the GDP in 2005 in the entire disaster area declined 24.2% compared with the business-
as-usual condition if the pre-Katrina growth trend persists. The accumulated loss of GDP from
2005 to 2013 (gray area in in Figure 5) is about 2.2 billion current dollar and the GDP is unlikely
to restore to the business-as-usual trajectory in the following years. Note, this loss is estimated
only from the declined GDP in the FEMA declared disaster area, which does not include other
indirect losses in a broader area. Note, due to the instability of the DMSP-OLS images, the
estimated GDP fluctuates from year to year. The estimation of GDP in a specific year may not be
accurate. However, the general trend indicates the strong economic impact of the disaster.
Figure 4: Changes of NTL brightness (DN value) in the 2004 and 2005 DMSP-OLS annual composite
Figure 5: Annual estimated GDP in current dollars in the affected area (refer to the left axis) and the
Southeast Region (refer to the right axis) (Data source: Bureau of Economic Analysis).
4.2 Spatial Variation of Resilience
The GDP recovery trajectory varies in counties. As illustrated in Figure 6, the estimated GDP in
Orleans Parish (blue) and St. Bernard Parish (red) had a substantive decline in 2005 when Katrina
stroke. Both parishes are in the metropolitan area of New Orleans near the landfall location of
Katrina. The post-Katrina GDP in St. Bernard Parish shows a faster recovery rate than Orleans
Parish after Katrina. In contrast, St. Tammany Parish (green) in the north shore of Lake
Pontchartrain had little economic impact and maintained a steady GDP growth after Katrina,
despite its proximity to the hurricane track.
Figure 6: Estimated GDP time series of Orleans Parish (blue), St. Bernard Parish (red) and St. Tammany
As shown in Figure 7 (a), Hurricane Katrina caused large instant disturbance () in counties near
the two landfall locations in Florida and Louisiana. Note, the high in southwest Louisiana
(around Lake Charles Parish) is possibly due to Hurricane Rita, a Category 3 hurricane landed near
the Louisiana-Texas border a month after Katrina. Figure 7 (b) shows that Louisiana counties have
a higher recovery rate (Rr) of GDP after Katrina, except Jefferson and Orleans Parish near the
center of New Orleans which were struggling to recover after Katrina. High accumulated GDP
loss (L) are distributed in coastal cities, including Gulfport (MS), Mobile (AL), Panama City and
Miami (FL). The inland counties generally have lower accumulated loss.
Figure 7: Spatial variation of (a) instant disturbance (), (b) recovering rate (Rr) and (c) accumulated GDP
loss (). Red indicates high instant disturbance (D), slower recovery (Rr) and high accumulated loss (L).
4.3 Regression Analysis
Various relationships were found between the three metrics (D, Rr, L) and selected variables (Table
Instant disturbance (D) has the strongest relationship with the percent of Asian people (highest R2)
among the 20 selected variables. The positive coefficient b indicates that counties with a higher
ratio of Asian people have endured greater GDP disturbance in Katrina. This result confirms the
findings in Vu et al. (2009) that 78% of Vietnamese (the largest Asian group in Louisiana and
Mississippi) left their home during the hurricane and gradually returns afterwards. D is also highly
correlated with the physical impacts and environmental conditions. Specifically, counties with
high gust speed, high accumulated rainfall, low elevation, high ratio of urban in flood zone, or
close to the coast line tend to have a high D. This reflects the fact that New Orleans, which has
low elevation and high ratio of urban flood exposure, was serious damaged by flood inundation
due to levee breech. Additionally, counties with a higher percentage of agricultural industry have
a lower D.
High recovery rates (Rr) are associated with high percentages of energy industry (e.g. mining,
quarrying, and oil and gas extraction), implying that the energy industry bounced back more
quickly after the disturbance. Moreover, low elevation, proximity to coast, high ratios of children
and elderly people tend to impede the GDP recovery.
Accumulated economic loss (L) is negatively correlated (highest R2) with ratio of owner-occupied
housing units, indicating communities with more home-owner residents managed to prevent the
overall economic loss. Communities with a higher ratio of Asian, Hispanic and Latino, children
and elderly adults have suffered more accumulated loss. High L is also found in counties with a
high urban exposure to flood zone. Additionally, counties reliant on agricultural and mining
industries tend to have a lower long term loss.
Table 3: Univariate regression analysis of instant disturbance (D), recovery (Rr) and economic loss (L) with
environmental and socio-economic variables. Bold font indicates statistical significance (p<0.01). Detailed
statistics of the regression analysis (e.g. confidence intervals, standard errors of residuals, and residual
distributions) can be found in the supplementary material.
Instant disturbance (D)
Recovery rate (Rr)
Economic loss (L)
Distance to Coast
% of urban in flood
% Hispanic & Latino
% elderly adult
% owner occupied
% bachelor degree
Per cap. income
% small businesses
% large businesses
R2 of the multivariate regression indicates that the selected variables can only explain 39.9%, 29.3%
and 27.4% variance of D, Rr, and L respectively (Table 4), which implies that other variables
should be considered to fully explain the variance of recovery pattern. Due to the different numbers
of variables in the four categories, adjusted R2 was used to compare the variance explained by the
different categories. In general, socio-economic variables can best predict (highest adjusted R2)
instant disturbance (D), followed by environmental condition and industrial structure. Industrial
structure is the best indicator of recovery rate (Rr), followed by environmental and socio-economic
variables. Socio-economic variables can best predict the accumulated economic loss (L). It is
worth-noting that the disaster impacts explain the least variance in all the three metrics, which
implies that the intrinsic community capacities (e.g. socio-economic conditions) are more decisive
to economic recovery than the physical disaster impacts.
Table 4: Adjusted R2 of the multivariate regressions between the resilience metrics (D, Rr, and L) and
variables in different categories.
Instant disturbance (D)
Recovery rate (Rr)
Economic loss (L)
Despite the extensive discussions in the literature, quantitative assessment of community resilience
is still a challenge due to the lack of empirical data continuously collected in disasters. In the U.S,
census data are released decennially and county-level GDP data is only available in limited years,
not to mention the developing world. Other data collection methods such as field surveys and
interviews are costly and time consuming. As an alternative, NTL remote sensing is an efficient
means to observe human activities (such as population, GDP, and energy consumption) from space.
Most importantly, the NTL images have the unique ability to detect declines of human activities,
which is not easy for other types of remote sensing imageries. The continuous scan of NTL images
can capture the disturbance and recovery pattern of human activity during natural disasters at low
cost and in a timely manner. The introduced framework can increase our understanding about
community resilience and help to improve resilience prediction models in terms of variable
selection and weighting.
In addition to Hurricane Katrina, the introduced assessment framework based on NTL data is
applicable to other natural disasters (e.g. tsunami and earthquake) that cause economic disturbance.
The framework is particularly useful in situations where official socio-economic data are not
available at the desired spatio-temporal resolution. Not limited to the DMSP/OLS images, the
introduced framework can use other data sources as input. For instance, the Visible Infrared
Imaging Radiometer Suite (VIIRS) images, the new generation NTL remote sensing products since
2011, provide many new features (e.g. higher spatial and radiometric resolution) that are useful
for measuring community resilience in disasters with a smaller impact area and shorter period.
As demonstrated in this study, the GDP time series estimated from the NTL data can capture the
various recovery trajectories at the county level. The analysis results uncover the strong economic
impact of Katrina in the affected region, where the GDP has given up its rapid growth and shifted
to a different trajectory since 2005. This finding confirms the tremendous and long-lasting impact
of Katrina on population migration (Fussell et al., 2014, 2010), declined urban growth (Qiang and
Lam, 2016) and economic activity (Baade et al., 2007; Petterson et al., 2006). According to the
recent estimates of the U.S. Census (2019), until 2017 the population and housing price in New
Orleans has not yet recovered to the pre-Katrina level. The various GDP recovery patterns reveal
geographical disparities of community resilience related to the local environmental and socio-
The recovery pattern of a social system is dependent on both intensity of disaster impacts and
resilience of the system. The analysis results (see Table 4) indicates the physical impact variables
(including max gust speed and accumulated rainfall) can only explain limited variance (R2 = 0.065)
in the recovery pattern. Instead, the inherent conditions of communities (including environmental,
socio-economic and industrial conditions) play a central role in shaping the recovery pattern. The
univariate regression with individual variables provides empirical evidence about the underlying
factors that influence the recovery. The results may also inform specific plan to prompt resilience
at different phases of a disaster. For instance, the strong correlation between instant disturbance
(D) and the environmental conditions (e.g. elevation, proximity to coast and % of urban in flood
zone) suggests that reducing exposure is the most effective way to reduce the direct impact caused
by disasters. In the recovery phase, communities with some demographic and socio-economic
characteristics may have difficulty to bounce back, pinpointing areas where special assistance or
policy levers s be applied. Compared with the traditional resilience indices without specific outputs,
these analysis results can better provide more actionable information to support decision-making
at different phases of a disaster.
Despite the merits demonstrated in this study, the introduced approach can be improved in the
following aspects. First, despite the good model fit at the county level, the DMSP-OLS images
cannot perfectly predict economic activity due to their inherent limitations (e.g. low spatial
resolution, lack of onboard and inter-satellite calibration, and limited dynamic range). In future
studies, the assessment results needs to be validated against additional datasets such as other NTL
products (e.g. VIIRS images) or socio-economic data continuously collected during the disaster.
Second, other factors that influence economic recovery should be taken into account in future
studies. For instance, economic cycles (e.g. great recessions) can slow down business recovery
after a disaster. In the introduced approach, linear models were used to generalize the economic
growth in the pre- and post-Katrina periods, where the yearly fluctuations (e.g. the financial crisis
in 2009) are averaged in the trend lines. Ideally, the “business-as-usual” trajectories should be
projected with more sophisticated models to eliminate the effect of economic cycles. Third, the 20
selected variables can only explain ~40% variation of the measured metrics, which reveal the
complexity of community resilience. The prediction power of the model can be improved by
including more variables and using more sophisticated model specifications (e.g. non-linear
models). Robust model for community resilience prediction can be developed by accumulating
empirical evidence in more disaster events.
This study introduces a quantitative framework for resilience assessment using DMSP-OLS
Nighttime Lights images. The framework was applied to model the recovery patterns of economic
activity in the affected area in Hurricane Katrina 2005. The analyses show the great economic
disturbance caused by Katrina and the slow recovery in the entire affected area. The county-level
analyses indicate strong spatial variation of the recovery pattern. Statistical analyses were carried
out to explore the underlying factors that influence the recovery patterns. This study demonstrates
the utility of NTL images in monitoring human dynamics in natural disasters, which filled the
critical gap of empirical data and assessment methods for resilience research. Based on the
framework of recovery trajectory, resilience is modelled as a dynamic process. Compared with the
traditional resilience indices, the modeling results can provide more specific and actionable
measures resilience promotion in diverse communities and in different phases of a disaster. This
study re-visits Hurricane Katrina using DMSP-OLS NTL images. However, the assessment
approach is applicable for other disaster events using other NTL products (e.g. VIIRS DNB
images). The results increase our understanding about the complexity of community resilience and
provide support for decision-makers to develop resilient and sustainable communities.
Adeola, F.O., 2009. Mental health & psychosocial distress sequelae of Katrina: An empirical study
of survivors. Hum. Ecol. Rev. 16, 195–210.
Adger, W.N., 2000. Social and ecological resilience: are they related? Prog. Hum. Geogr. 24, 347–
Baade, R.A., Baumann, R., Matheson, V., 2007. Estimating the Economic Impact of Natural and
Social Disasters, with an Application to Hurricane Katrina. Urban Stud. 44, 2061–2076.
Bakkensen, L.A., Fox‐Lent, C., Read, L.K., Linkov, I., 2017. Validating Resilience and
Vulnerability Indices in the Context of Natural Disasters. Risk Anal. 37, 982–1004.
Bates, P.D., 2004. Remote sensing and flood inundation modelling. Hydrol. Process. 18, 2593–
Beccari, B., 2016. A Comparative Analysis of Disaster Risk, Vulnerability and Resilience
Composite Indicators. PLOS Curr. Disasters.
Bureau of Economic Analysis, U.S. Department of Commerce, 2018. Prototype Gross Domestic
Product by County, 2012-2015.
Bureau of Economic Analysis, U.S. Department of Commerce, 2015. Gross Domestic Product by
State, Advance 2014 and Revised 1997-2013.
Burren, D., Neusser, K., 2010. The decline in volatility of US GDP growth. Appl. Econ. Lett. 17,
Burton, A., 2006. Health Disparities: Crisis Not Over for Hurricane Victims. Environ. Health
Perspect. 114, A462.
Cai, H., Lam, N.S.N., Zou, L., Qiang, Y., 2018. Modeling the Dynamics of Community Resilience
to Coastal Hazards Using a Bayesian Network. Ann. Am. Assoc. Geogr. 108, 1260–1279.
Cooner, A.J., Shao, Y., Campbell, J.B., 2016. Detection of Urban Damage Using Remote Sensing
and Machine Learning Algorithms: Revisiting the 2010 Haiti Earthquake. Remote Sens. 8,
Cutter, S.L., Ash, K.D., Emrich, C.T., 2014. The geographies of community disaster resilience.
Glob. Environ. Change 29, 65–77. https://doi.org/10.1016/j.gloenvcha.2014.08.005
Cutter, S.L., Barnes, L., Berry, M., Burton, C., Evans, E., Tate, E., Webb, J., 2008. A place-based
model for understanding community resilience to natural disasters. Glob. Environ. Change
18, 598–606. http://dx.doi.org/10.1016/j.gloenvcha.2008.07.013
Cutter, S.L., Boruff, B.J., Shirley, W.L., 2003. Social vulnerability to environmental hazards. Soc.
Sci. Q. 84, 242–261. https://doi.org/10.1111/1540-6237.8402002
Cutter, S.L., Burton, C.G., Emrich, C.T., 2010. Disaster Resilience Indicators for Benchmarking
Baseline Conditions. J. Homel. Secur. Emerg. Manag. 7, 14. https://doi.org/10.2202/1547-
Dasgupta, S., Laplante, B., Murray, S., Wheeler, D., 2011. Exposure of developing countries to
sea-level rise and storm surges. Clim. Change 106, 567–579.
Dong, L., Shan, J., 2013. A comprehensive review of earthquake-induced building damage
detection with remote sensing techniques. ISPRS J. Photogramm. Remote Sens. 84, 85–99.
Elvidge, C.D., Ziskin, D., Baugh, K.E., Tuttle, B.T., Ghosh, T., Pack, D.W., Erwin, E.H., Zhizhin,
M., 2009. A Fifteen Year Record of Global Natural Gas Flaring Derived from Satellite
Data. Energies 2, 595–622. https://doi.org/10.3390/en20300595
Faturechi, R., Miller-Hooks, E., 2015. Measuring the Performance of Transportation Infrastructure
Systems in Disasters: A Comprehensive Review. J. Infrastruct. Syst. 21, 04014025.
Ferrara, L., Marcellino, M., Mogliani, M., 2015. Macroeconomic forecasting during the Great
Recession: The return of non-linearity? Int. J. Forecast. 31, 664–679.
Finch, C., Emrich, C.T., Cutter, S.L., 2010. Disaster disparities and differential recovery in New
Orleans. Popul. Environ. 31, 179–202. https://doi.org/10.1007/s11111-009-0099-8
Foster, K., 2012. In search of regional resilience, in: Weir, M., Wial, H., Wolman, H. (Eds.),
Building Regional Resilience: Urban and Regional Policy and Its Effects. Brookings
Institute Press, Washington, DC.
Fussell, E., Curtis, K.J., DeWaard, J., 2014. Recovery migration to the City of New Orleans after
Hurricane Katrina: a migration systems approach. Popul. Environ. 35, 305–322.
Fussell, E., Sastry, N., VanLandingham, M., 2010. Race, socioeconomic status, and return
migration to New Orleans after Hurricane Katrina. Popul. Environ. 31, 20–42.
GILLESPIE, T.W., FRANKENBERG, E., CHUM, K.F., THOMAS, D., 2014. Nighttime lights
time series of tsunami damage, recovery, and economic metrics in Sumatra, Indonesia.
Remote Sens. Lett. Print 5, 286–294. https://doi.org/10.1080/2150704X.2014.900205
Holling, C.S., 1973. Resilience and Stability of Ecological Systems. Annu. Rev. Ecol. Syst. 4, 1–
Hong, Y., Adler, R., Huffman, G., 2007. Use of satellite remote sensing data in the mapping of
global landslide susceptibility. Nat. Hazards 43, 245–256. https://doi.org/10.1007/s11069-
Hsu, F.-C., Baugh, K.E., Ghosh, T., Zhizhin, M., Elvidge, C.D., 2015. DMSP-OLS Radiance
Calibrated Nighttime Lights Time Series with Intercalibration. Remote Sens. 7, 1855–1876.
Huang, Q., Yang, X., Gao, B., Yang, Y., Zhao, Y., 2014. Application of DMSP/OLS Nighttime
Light Images: A Meta-Analysis and a Systematic Literature Review. Remote Sens. 6,
Huang, X., Wang, C., Lu, J., 2019. Understanding Spatiotemporal Development of Human
Settlement in Hurricane-prone Areas on U.S. Atlantic and Gulf Coasts using Nighttime
Remote Sensing. Nat. Hazards Earth Syst. Sci. Discuss. 1–22.
Hultquist, C., Simpson, M., Cervone, G., Huang, Q., 2015. Using Nightlight Remote Sensing
Imagery and Twitter Data to Study Power Outages, in: Proceedings of the 1st ACM
SIGSPATIAL International Workshop on the Use of GIS in Emergency Management, EM-
GIS ’15. ACM, New York, NY, USA, pp. 6:1–6:6.
Hung, H.-C., Yang, C.-Y., Chien, C.-Y., Liu, Y.-C., 2016. Building resilience: Mainstreaming
community participation into integrated assessment of resilience to climatic hazards in
metropolitan land use management. Land Use Policy 50, 48–58.
Jonkman, S.N., Maaskant, B., Boyd, E., Levitan, M.L., 2009. Loss of Life Caused by the Flooding
of New Orleans After Hurricane Katrina: Analysis of the Relationship Between Flood
Characteristics and Mortality. Risk Anal. 29, 676–698. https://doi.org/10.1111/j.1539-
Joshi, N., Baumann, M., Ehammer, A., Fensholt, R., Grogan, K., Hostert, P., Jepsen, M.R.,
Kuemmerle, T., Meyfroidt, P., Mitchard, E.T.A., Reiche, J., Ryan, C.M., Waske, B., 2016.
A Review of the Application of Optical and Radar Remote Sensing Data Fusion to Land
Use Mapping and Monitoring. Remote Sens. 8, 70. https://doi.org/10.3390/rs8010070
Kohiyama, M., Hayashi, H., Maki, N., Higashida, M., Kroehl, H.W., Elvidge, C.D., Hobson, V.R.,
2004. Early damaged area estimation system using DMSP-OLS night-time imagery. Int. J.
Remote Sens. 25, 2015–2036. https://doi.org/10.1080/01431160310001595033
Koliou, M., Lindt, J.W. van de, McAllister, T.P., Ellingwood, B.R., Dillard, M., Cutler, H., 2018.
State of the research in community resilience: progress and challenges. Sustain. Resilient
Infrastruct. 0, 1–21. https://doi.org/10.1080/23789689.2017.1418547
Kuffer, M., Pfeffer, K., Sliuzas, R., 2016. Slums from Space—15 Years of Slum Mapping Using
Remote Sensing. Remote Sens. 8, 455. https://doi.org/10.3390/rs8060455
Lam, N.S.N., Pace, K., Campanella, R., LeSage, J., Arenas, H., 2009. Business Return in New
Orleans: Decision Making Amid Post-Katrina Uncertainty. PLoS ONE 4, e6765.
Lam, N.S.-N., Qiang, Y., Arenas, H., Brito, P., Liu, K.-B., 2015. Mapping and assessing coastal
resilience in the Caribbean region. Cartogr. Geogr. Inf. Sci. 0406, 1–8.
Lam, N.S.N., Reams, M., Li, K., Li, C., Mata, L.P., 2016. Measuring community resilience to
coastal hazards along the Northern Gulf of Mexico. Nat. Hazards Rev. 17, e193–e193.
LeSage, J.P., Pace, R.K., Lam, N., Campanella, R., Liu, X., 2011. New Orleans business recovery
in the aftermath of Hurricane Katrina. J. R. Stat. Soc. Ser. A Stat. Soc. 174, 1007–1027.
Li, S., Dragicevic, S., Castro, F.A., Sester, M., Winter, S., Coltekin, A., Pettit, C., Jiang, B.,
Haworth, J., Stein, A., Cheng, T., 2016. Geospatial big data handling theory and methods:
A review and research challenges. ISPRS J. Photogramm. Remote Sens., Theme issue
“State-of-the-art in photogrammetry, remote sensing and spatial information science” 115,
Li, X., Ge, L., Chen, X., 2013. Detecting Zimbabwe’s Decadal Economic Decline Using Nighttime
Light Imagery. Remote Sens. 5, 4551–4570. https://doi.org/10.3390/rs5094551
Li, X., Zhan, C., Tao, J., Li, L., 2018. Long-Term Monitoring of the Impacts of Disaster on Human
Activity Using DMSP/OLS Nighttime Light Data: A Case Study of the 2008 Wenchuan,
China Earthquake. Remote Sens. 10, 588. https://doi.org/10.3390/rs10040588
Liao, K.-H., 2012. A Theory on Urban Resilience to Floods—A Basis for Alternative Planning
Practices. Ecol. Soc. 17. https://doi.org/10.5751/ES-05231-170448
Marcellino, M., 2008. A linear benchmark for forecasting GDP growth and inflation? J. Forecast.
27, 305–340. https://doi.org/10.1002/for.1059
Martin, R., Sunley, P., 2015. On the notion of regional economic resilience: conceptualization and
explanation. J. Econ. Geogr. 15, 1–42. https://doi.org/10.1093/jeg/lbu015
Nicholls, R.J., Hoozemans, F.M.J., Marchand, M., 1999. Increasing flood risk and wetland losses
due to global sea-level rise: regional and global analyses. Glob. Environ. Change 9, S69–
NOAA, 2006. Service Assessment: Hurricane Katrina August 23-31, 2005.
Norris, F.H., Stevens, S.P., Pfefferbaum, B., Wyche, K.F., Pfefferbaum, R.L., 2008. Community
Resilience as a Metaphor, Theory, Set of Capacities, and Strategy for Disaster Readiness.
Am. J. Community Psychol. 41, 127–150. https://doi.org/10.1007/s10464-007-9156-6
Peacock, W.G., 2010. Final Report Advancing the Resilience of Coastal Localities 10-02R.
Perz, S.G., Muñoz-Carpena, R., Kiker, G., Holt, R.D., 2013. Evaluating ecological resilience with
global sensitivity and uncertainty analysis. Ecol. Model. 263, 174–186.
Petterson, J.S., Stanley, L.D., Glazier, E., Philipp, J., 2006. A Preliminary Assessment of Social
and Economic Impacts Associated with Hurricane Katrina. Am. Anthropol. 108, 643–670.
Qiang, Y., 2019. Disparities of population exposed to flood hazards in the United States. J. Environ.
Manage. 232, 295–304. https://doi.org/10.1016/j.jenvman.2018.11.039
Qiang, Y., Lam, N.S.-N., 2016. The impact of Hurricane Katrina on urban growth in Louisiana:
an analysis using data mining and simulation approaches. Int. J. Geogr. Inf. Sci. 30, 1832–
Qiang, Y., Xu, J., 2019. Empirical assessment of road network resilience in natural hazards using
crowdsourced traffic data. Int. J. Geogr. Inf. Sci. 0, 1–17.
Sastry, N., VanLandingham, M., 2009. One Year Later: Mental Illness Prevalence and Disparities
Among New Orleans Residents Displaced by Hurricane Katrina. Am. J. Public Health 99,
Sempier, T.T., Swann, D.L., Emmer, R., Sempier, S.H., Schneider, M., 2010. Coastal Community
Resilience Index: A Community Self-Assessment (No. MASGP-08-014).
Shahtahmassebi, A.R., Song, J., Zheng, Q., Blackburn, G.A., Wang, K., Huang, L.Y., Pan, Y.,
Moore, N., Shahtahmassebi, G., Sadrabadi Haghighi, R., Deng, J.S., 2016. Remote sensing
of impervious surface growth: A framework for quantifying urban expansion and re-
densification mechanisms. Int. J. Appl. Earth Obs. Geoinformation 46, 94–112.
Sherrieb, K., Norris, F.H., Galea, S., 2010. Measuring Capacities for Community Resilience. Soc.
Indic. Res. 99, 227–247. https://doi.org/10.1007/s11205-010-9576-9
Spalding, M.D., McIvor, A.L., Beck, M.W., Koch, E.W., Möller, I., Reed, D.J., Rubinoff, P.,
Spencer, T., Tolhurst, T.J., Wamsley, T.V., van Wesenbeeck, B.K., Wolanski, E.,
Woodroffe, C.D., 2014. Coastal Ecosystems: A Critical Element of Risk Reduction.
Conserv. Lett. 7, 293–301. https://doi.org/10.1111/conl.12074
Tebaldi, C., Strauss, B.H., Zervas, C.E., 2012. Modelling sea level rise impacts on storm surges
along US coasts. Environ. Res. Lett. 7, 014032. https://doi.org/10.1088/1748-
Timmerman, P., 1981. Vulnerability, Resilience and the Collapse of Society - a Review of Models
and Possible Climatic Applications., University of Toronto Institute For Environmental
Studies Environmental Monograph. Institute for Environmental Studies, University of
U.S. Census, 2019. American Community Survey.
U.S. Census Bureau, 2011. Census 2010. U.S. Census Bureau.
Vercelloni, J., Kayal, M., Chancerelle, Y., Planes, S., 2019. Exposure, vulnerability, and resiliency
of French Polynesian coral reefs to environmental disturbances. Sci. Rep. 9, 1027.
Vetrivel, A., Gerke, M., Kerle, N., Nex, F., Vosselman, G., 2018. Disaster damage detection
through synergistic use of deep learning and 3D point cloud features derived from very
high resolution oblique aerial images, and multiple-kernel-learning. ISPRS J. Photogramm.
Remote Sens., Geospatial Computer Vision 140, 45–59.
Vu, L., VanLandingham, M.J., Do, M., Bankston, C.L., 2009. Evacuation and Return of
Vietnamese New Orleanians Affected by Hurricane Katrina. Organ. Environ. 22, 422–436.
White, R.K., Edwards, W.C., Farrar, A., Plodinec, M.J., 2015. A Practical Approach to Building
Resilience in America’s Communities. Am. Behav. Sci. 59, 200–219.
World Bank, 2019. World Bank national accounts data.
Xie, Y., Weng, Q., 2017. Spatiotemporally enhancing time-series DMSP/OLS nighttime light
imagery for assessing large-scale urban dynamics. ISPRS J. Photogramm. Remote Sens.
128, 1–15. https://doi.org/10.1016/j.isprsjprs.2017.03.003
Yodo, N., Wang, P., 2016. Engineering Resilience Quantification and System Design Implications:
A Literature Survey. J. Mech. Des. 138. https://doi.org/10.1115/1.4034223
Zhao, X., Yu, B., Liu, Y., Yao, S., Lian, T., Chen, L., Yang, C., Chen, Z., Wu, J., 2018. NPP-
VIIRS DNB Daily Data in Natural Disaster Assessment: Evidence from Selected Case
Studies. Remote Sens. 10, 1526. https://doi.org/10.3390/rs10101526
Zhuo, L., Ichinose, T., Zheng, J., Chen, J., Shi, P.J., Li, X., 2009. Modelling the population density
of China at the pixel level based on DMSP/OLS non‐radiance‐calibrated night‐time light
images. Int. J. Remote Sens. 30, 1003–1018. https://doi.org/10.1080/01431160802430693
Zou, L., Lam, N., Cai, H., Qiang, Y., 2017. Mining Social Media Data for Improved understanding
of Disaster Resilience, in: The 2017 Annal Conference of the Association of American
Zou, L., Lam, N.S.N., Shams, S., Cai, H., Meyer, M.A., Yang, S., Lee, K., Park, S.-J., Reams,
M.A., 2018. Social and geographical disparities in Twitter use during Hurricane Harvey.
Int. J. Digit. Earth 0, 1–19. https://doi.org/10.1080/17538947.2018.1545878