Professional Documents
Culture Documents
To cite this article: Joanne R. Stevenson , Christopher T. Emrich , Jerry T. Mitchell & Susan L. Cutter (2010) Using Building Permits
to Monitor Disaster Recovery: A Spatio-Temporal Case Study of Coastal Mississippi Following Hurricane Katrina, Cartography and
Geographic Information Science, 37:1, 57-68, DOI: 10.1559/152304010790588052
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the
publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or
warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of the authors, and are not the views of or
endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently
verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising
directly or indirectly in connection with, in relation to or arising out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is
expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-
and-conditions
Using Building Permits to Monitor Disaster
Recovery: A Spatio-Temporal Case Study
of Coastal Mississippi Following Hurricane Katrina
Joanne R. Stevenson, Christopher T. Emrich,
Jerry T. Mitchell, and Susan L. Cutter
ABSTRACT: The pace of disaster recovery varies considerably from one place to another. Even
when places suffer impacts from the same event, recovery studies often lack the spatial and tempo-
ral resolution to fully understand such local variability in the recovery process and patterns. This
paper discusses the novel use of building permits and a spatial scan statistic to identify the spatial
and temporal dimensions of recovery in coastal Mississippi following Hurricane Katrina. Our work
identifies significant space-time clusters of recovery activity and indicates that the amount of damage
experienced and the amount of pre-event housing strongly influence the timing and location of
building permit clusters. This analytical method and the use of publicly available data are valuable
for a better understanding of long-term recovery processes.
Downloaded by [RMIT University] at 03:11 10 July 2013
KEYWORDS: Building permits, SaTScan, Hurricane Katrina, disaster recovery, rebuilding, Gulf
Coast
R
following Hurricane Katrina have revealed initial
ecovery from a natural disaster is under- recovery disparities driven by class and gender
stood as a dynamic and multifaceted pro- (Cutter et al. 2006; Elliott and Pais 2006), few
cess, yet we know little about its spatial methods or metrics are capable of capturing trends
and temporal variability. The inability of most of recovery throughout the entire impacted area
methods to provide information about the pace and over longer time frames (Pais and Elliott 2008;
and progression of disaster recovery leads to the Zottarelli 2008). A disaster of this magnitude affords
problematic conclusion that recovery is spatially an opportunity to study how long-term recovery
uniform and consistent from one time period to is manifested within an affected landscape and
another (Cutter et al. 2006; Zottarelli 2008). to uncover the drivers of recovery as they shift
Hurricane Katrina stands as the most damag- through space and time.
ing disaster in U.S. history. One model suggests In this paper we use a spatial scan statistic,
that long-term recovery of the Gulf Coast could SaTScan, to examine the space–time trends of
take 11 years (Kates et al. 2006). Within that time, built environment recovery following Hurricane
several billion dollars of aid and countless hours Katrina. Scan statistics are a common tool used to
will be spent rebuilding the damaged structures determine if points (in this case, rebuilding activi-
ties) are randomly distributed in space and time
Joanne R. Stevenson, Hazards and Vulnerability Research or if they are clustered (Kulldorff 1997; Kulldorff
Institute, Department of Geography, University of South
2005). This research specifically investigates the
Carolina. Columbia, South Carolina 29208, E-mail: <rinaldij@
spatial and temporal patterns of recovery using
mailbox.sc.edu>. Christopher T. Emrich, Hazards and
Vulnerability Research Institute, Department of Geography, building permits issued in three municipalities
University of South Carolina. Columbia, South Carolina 29208, E- on Mississippi’s Gulf Coast. The spatio-temporal
mail: <emrich@mailbox.sc.edu>. Jerry T. Mitchell, Hazards relationships between permits issued, damage level,
and Vulnerability Research Institute, Department of Geography, and the pre-event number of housing units in the
University of South Carolina. Columbia, South Carolina 29208, affected area form the basis for this inquiry. We
E-mail: <mitchell@sc.edu>. Susan L. Cutter, Hazards and question whether spatial and temporal clusters of
Vulnerability Research Institute, Department of Geography, building permits, if they exist, are related to the
University of South Carolina. Columbia, South Carolina 29208, level of damage caused by the storm or the density
E-mail: <scutter@sc.edu>. of pre-event housing. The techniques employed
Cartography and Geographic Information Science, Vol. 37, No. 1, 2010, pp. 57-68
here improve our understanding of recovery with need to integrate betterment processes throughout
data and methods which highlight recovery as a disaster recovery, assess vulnerability issues which
process rather than as an outcome. may have exacerbated the effects of the disaster,
and use the recovery period following a disaster
What is Recovery? as an opportunity to address other pre-existing
social and environmental issues (Cutter et al. 2006;
Defining what recovery is and what it means for Kates et al. 2006; Olshansky 2006; Rubin 2009).
affected communities is fundamental to finding The lack of clarity in the extant literature on
appropriate ways to measure it. Recovery varies recovery is easily defined by two simple questions:
depending on the context of the disaster, the Recovery for whom? And recovery to what? The
level of impact and the extent of the damage, answers are social choices embedded in the politics
and the pre-event conditions (Bates and Peacock of the local communities affected by the disaster.
1989; Quarantelli 1999). In addition to physical These choices, whether part of a “return to nor-
destruction and disruption, disasters interrupt malcy” or as sustainable redevelopment, result in
the highly connected social fabric of communi- measurable changes to the landscape.
ties (Bolin 1976). While some of the literature
addresses recovery as a multi-dimensional con-
Measuring Recovery
cept from a theoretical perspective, most case
studies of recovery distill a single aspect of the The process of recovery includes a set of activi-
recovery process for in-depth analysis, focusing ties which ameliorate the negative impacts of
Downloaded by [RMIT University] at 03:11 10 July 2013
on specific recovery topics such psycho-social disasters and restore individuals, the built envi-
(Gault et al. 2005), institutional (Rubin and ronment, and the natural environment to pre-
Barbee 1985), economic and business (Chang disaster functioning. Recovery also includes
2000; Webb et al. 2002), built environment (Liu outcomes, or the extent to which the recovery
and Plyer 2009; McCarthy and Hanson 2008), activities are judged as successful or complete,
or the natural environment (Orr and Ogden using subjective (qualitative) or more objec-
1992). tive (quantitative) measures (National Research
Oftentimes, the term recovery has been used Council 2006). Unfortunately, there is a paucity
interchangeably with rebuilding, restoration, and of empirical studies on this point.
redevelopment (Mileti 1999). These phases are The most basic quantitative analyses and those
instrumental to recovery. For example, rebuilding most often reported by government agencies and
of residential, commercial, and public structures aid organizations are simple numerical compari-
not only requires the largest amount of resources, sons of pre- and post-event conditions. Examples
but is also an important precursor to repopulation, include measuring household recovery by identi-
and the reestablishment of commerce and social fying when a home value returns to its pre-event
networks (Rubin et al1985; Kamel and Loukaitou- level or comparing (at the county or city level) the
Sideris 2004). It must be acknowledged, however, number of housing units which have been rebuilt to
that these other indicators are inadequate when what was in place before the event (McCarthy and
trying to generalize about community recovery Hanson 2008). These simple numeric approaches
as a process, one that strives to restore, rebuild, can provide useful measures of demographic trends
and reshape the physical infrastructure, natural and physical recovery post disaster, but do not
environment, and socio-economic systems through provide information on the differential rates of
pre-event planning and post-event actions (Smith recovery within the affected area.
and Wenger 2006). An extension of these quantitative comparisons is
Another point of contention in the disaster recov- the development of recovery indices. Indices more
ery literature is whether recovery means returning formally compare recovery to a data baseline in order
to a stable state following a disaster, returning the to track the progress of recovery. The New Orleans
affected area to pre-event conditions, or if recov- Index reports extensively on several recovery indica-
ery necessitates a betterment process (Bates and tors, including population recovery, the amount and
Peacock 1989; Mileti 1999; Quarantelli 1999; Kates location of new construction and repairs, housing and
et al. 2006; Anderson 2008; Alesch et al. 2008). employment vacancy rates, school enrollment, retail
For example, Rubin and Popkin (1990) describe sales, and the availability of schools, libraries, and
a model of recovery which reconciles the view of childcare (Liu and Plyer 2009). Despite their ability
recovery as both a return to normalcy and a bet- to reflect several types of recovery and document
terment process. Recent research emphasizes the them empirically, indices do not always answer the
Median Household
Total Population Total Housing Units % White/Black/Other
Income ($)
Bay St. Louis 8,209 3,817 34,106 80.2/ 16.6/ 3.2
Pass Christian 6,579 3,351 40,743 65.9/ 28.2/ 5.9
Long Beach 17,320 7,203 43,289 87.5/ 7.4/ 5.1
Source: American FactFinder (2000).
Table 1. Study area demographic comparisons.
important question of why or where certain recovery Mississippi from wind, rain-induced flooding,
trends are occurring. breaches in flood containment structures, and
Physical and spatial indicators of recovery—changes storm surge. The eastern eye wall of the storm
to the built-environment—have been assessed through passed over Waveland and Bay St. Louis on the
geographic information systems and other analysis western Mississippi coast. Approximately 60 per-
and visualization techniques. These include aerial cent of the housing stock in Mississippi’s three
and satellite photography and spatial video acquisi- coastal counties suffered some level of damage
tion systems (Jarmin and Miranda 2006; Curtis et al. (Jaycox et al. 2006). Three municipalities—Bay
2007; Mills 2008; Liu and Plyer 2009). Unfortunately, St. Louis, Pass Christian, and Long Beach—had
many of these data sources only provide a snapshot some of the greatest and most direct impacts.
of recovery at certain points in space and time and These communities constitute our study area
do not take into account the underlying spatial and (Figure 1). Each community differs from the
temporal interactions between the drivers, activities, other in overall population size, racial composi-
and outcomes of the recovery process itself. tion, and median household income (Table 1).
To address some of these gaps in the literature Bay St. Louis (county seat of Hancock County)
and provide an improvement in measuring recovery, is a middle-income residential community with a
this paper empirically demonstrates temporal and largely white population. As one of the three oldest
spatial changes in recovery activities and outcomes. cities along the Gulf of Mexico, Bay St. Louis pre-
We explicitly focus on the built environment as our dates the founding of New Orleans. Bay St. Louis
recovery measure. became a summer home destination for wealthy
New Orleanians after the completion of the rail-
road between Mobile and New Orleans in 1870.
Study Area Today, the community is focused on services such
Hurricane Katrina struck the U.S. Gulf Coast as tourism and gaming. Across St. Louis Bay is Pass
on August 29, 2005, causing substantial damage Christian (Harrison County), a city that began as
and loss of life in Alabama, Louisiana, and a summer resort community for the urban elite of
exterior damage…some mobile homes and of the freely available SaTScan software from its
light structures are destroyed, and many are original purpose—spatial analytics for epidemi-
damaged or displaced. ology—into hazards applications. For example,
4. Extensive Damage: Some solid structures are Vadrevu (2008) used SaTScan to analyze the
destroyed, most sustain exterior damage…most significance of wildfire occurrence clusters in
mobile homes and light structures are destroyed. India, Witham and Oppenheimer (2005) evalu-
5. Catastrophic Damage: Most solid and all light ated historic mortality clusters in England fol-
or mobile structures are destroyed. lowing the 1783-1784 Laki Craters eruption,
6. Flood: Area under water or ground saturation. and Kulldorff et al. (2005) detected disease out-
The damage polygon dataset may not account breaks. However, SaTScan technology has not
for all damages in each enumeration area or to been utilized to track the progress of recovery
every structure; however, these data are the most from major disaster events. This is likely due to
comprehensive geospatially enabled damage the historical lack of data with the level of spatial
information available for the entire study area. or temporal resolution needed to evaluate “clus-
As discussed by Richardson and Renner (2007) ters” of recovery.
in their evaluation of damage data and GIS in SaTScan uses a scan statistic to analyze either
disaster response and recovery there are serious spatial, temporal, or space-time point data (Abe
difficulties of trying to combine damage assessments et al. 2006). The software is useful for this study
which are not confined to political boundaries as it provides outputs in a format compatible with
to enumeration units such as census blocks. Our multiple commercial grade GIS software, including
solution was to use a GIS to overlay the damage ESRI’s ArcGIS. SaTScan makes no assumptions
category shapefile to the census blocks shapefile, about if and where clusters exist. Therefore, unlike
thus creating smaller “sub-block” units containing other clustering and spatial techniques, SaTScan
a single damage category. We then aggregated our does not require the user to define a number of
point permit and pre-storm housing unit data to clusters desired or any other parameters. Once
these “sub-blocks.” A final concern is that the broad clusters are identified by SaTScan, they are tested
and somewhat subjective damage categories do for statistical significance using a Monte Carlo test.
not provide detailed information about the type The Monte Carlo statistic tests the significance (p
of structures damaged, insurance status, or any value) of each cluster by analyzing its maximum
indication of the number of structures damaged likelihood, or the likelihood that a cluster could
in an area (Richardson and Renner 2007). This have occurred randomly in the data set. For this
study’s analysis is restricted to damages at the study, only those clusters with the highest sig-
aggregate level and not on damage to individual nificance (0.001 chance of occurring randomly)
homes. are discussed. The analysis is conditioned by the
The pre-event housing units were derived from total number of observed points to calculate an
the Mississippi Housing Recovery Data Project expected value. The number of points in each scan
window is then compared against the expected 0.001) were found, each with varying spatial and
value to identify areas with higher-than-expected temporal distributions. Each cluster identifies
concentrations of permits (Kulldorff 1997; Abe an area of higher-than-expected concentrations
et al. 2006). The scan window is composed of of construction permits based on location and
thousands of overlapping cylinders, with the base date.
of the cylinder scanning the spatial component The first cluster (a .23 mi radius) is located in Bay
of the data and the height of the cylinder scan- St. Louis. The cluster consists of two large apart-
ning the temporal component. For each window, ment complexes permitted in May 2008 (Figure
the expected number of cases is compared to the 2). The first complex is the Bay Side Apartments
observed number of cases in order to identify where at 701 Union Street. This complex has eleven
clusters occur. The SaTScan output includes a list buildings with 50 apartment units, a main service
of all clusters, a list of the unique permit identifi- building, and a maintenance shop. The residential
cation numbers associated with each cluster, the construction permits for each unit were issued
numbers of observed and expected cases, and the over a three-day period from May 6-8, 2008. The
p value for each cluster. second complex in Cluster 1, Sheffield Park, is
The space–time probability model was chosen located nearby, at 635 Carroll Avenue. Here there
as the statistical test for this dataset as knowledge are seven buildings with 131 separate apartments.
of both where and when permit clusters occurred These were permitted between May 21, 2008 and
is important in understanding the progression of May 29, 2008. Neither complex is beachside nor
recovery (as measured by building reconstruction). can they be considered luxury condominiums. The
Although permits are recorded daily, time was Sheffield Park apartments, for example, rent for
aggregated by calendar month for this analysis. $474 per month for a one-bedroom apartment
(Gulf Coast Apartment Guide 2009). All permits
associated with Cluster 1 are located completely
Results within the moderate damage category.
The second cluster is located primarily along
Time–Space Clusters of Recovery Pass Christian’s gulf side. With a radius of 1.57
A total of 6661 residential construction permits mi, the cluster includes much of the western half
were used in this analysis, covering the period of Pass Christian. This cluster contains permits
from September 2005 through to December issued between October 2006 and April 2008. The
2008. For building permit analysis, only permit expected number of permits for this area, 272, was
data, including the issue date of the permit and far lower than the 537 cases identified by SaTScan.
its location (latitude, longitude), were used in The majority of the permits were for single family
SaTScan. Four significant clusters (p value of residences; only nine percent of the permits were
block (H= high number per block based on natural breaks defined in GIS relative to entire dataset, M= medium, L=low).
for multi-unit residences. Accordingly, this clus- in the same location) were more easily and quickly
ter of rebuilding reflects individual homeowners made where damages were less.
applying for building permits. Approximately 67 The identified spatio-temporal clusters of recovery
percent of the permits associated with this cluster are activity correspond with the level of damage and
located within the catastrophic damaged areas. The the amount of preexisting housing in the area. For
remaining permits are distributed throughout the example, the larger Long Beach cluster not only
moderate, limited, and no damage categories. appears earlier, but is also the cluster located farthest
The third and smallest cluster is a complex of 15 from the storm’s eye path and its more damaging
buildings in a Long Beach resort condo/apartment effects. This temporal variation is consistent in
development. Known as the “Beach Club,” this cluster the clusters from East to West, with the earliest
at 2012 West 2nd Street represents permits issued in clusters appearing farther east than the western
September 2007. This development is owned by an clusters which emerged later in the study period.
out-of-state developer, and rental prices are upwards The longer permitting activity period identified
of $750 per month for a one-bedroom unit (Gulf in Pass Christian suggests that greater damage
Coast Apartment Guide 2009). All permits within levels slowed the rebuilding process. Similarly, we
this cluster are located within a catastrophic damage would expect more early activity in areas with
area close to the water front. limited damage where homeowners would simply
The final cluster is the largest (2.05 mi radius) rebuild on the same property. We explore these
and earliest in terms of permits issued (between relationships further in the next section.
October–November 2005). Also located in Long
Beach, the cluster represents 464 different resi- Recovery Activity
dential units compared to an expected 236 cases and Pre-event Housing
predicted by the likelihood calculation in SaTScan.
This cluster is primarily characterized by single A reasonable expectation is that areas with
family homes. While occurring only 2-3 months higher pre-storm housing concentrations would
following the storm, this cluster contains nearly also see a higher number of construction per-
30 percent of all the building permits issued in mits, given the resident’s investment in the land.
Long Beach throughout the study period. Seventy What normally changes is the nature of the con-
percent of the permits in this cluster fall within struction (e.g., conversion from lower density
limited and no damage areas, an indication that single family homes to condos, or from smaller
reconstruction decisions (i.e., the decision to stay single family homes to larger ones).
Correlation Between
# Pre-storm Houses Percentage of Permits in Each Damage Category
and Permits
Entire Study Area Bay St. Louis Pass Christian Long Beach
Catastrophic .322** 28.04 11.35 47.52 35.67
Extensive .859** 4.02 8.12 0.98 0.18
Moderate .679** 42.02 78.39 21.35 0.59
Limited .731** 15.99 0 14.84 45.87
None .743** 9.92 2.14 15.31 17.69
N= 679 Blocks 6,661 Permits 3,031 Permits 1,938 Permits 1,692 Permits
**Correlation significant at 0.01 level (2-tailed).
Table 4. Relationship between pre-storm housing and construction permits by damage category.
pre-storm housing and building permits remains building permits. In the no damage category, the
strong and positive for all damage categories except positive association between pre-storm housing
the catastrophically damaged area. In catastrophi- and permits could reflect renovations to existing
cally damaged areas there is a modest (R=.322) structures. A second possibility is the establishment
but significant correlation between pre-event of new subdivisions that reflect new post-storm
housing and building permits. This relationship settlement patterns.
indicates that either 1) rebuilding is occurring
in these areas at a slower rate than in the other
damage categories, or 2) there are other factors
(e.g., the damage) that explain the variance in
Discussion and Conclusion
residential permits (Table 4). Sharply contrasting The analysis of building permit clustering, both
this relationship is the strong positive correlation spatial and temporal, provides strong evidence of
between pre-event housing and building permits in differential rebuilding across coastal Mississippi
areas categorized with extensive damage (R=.859). and is a novel approach to understanding long-
For the remaining damage categories, including term rebuilding from disaster events. Additional
no damage, there are equally strong and signifi- information from building permits may provide
cant positive relationships (R=.679 for moderate a more thorough understanding of the pro-
damage; R=.731 for limited damage, and R=.743 cesses at work along the Mississippi Gulf Coast
for no damage) between pre-storm housing and following Hurricane Katrina. Our analysis spe-
the recovery is largely due to rebuilding of existing Although there has been an increased interest in
structures in heavily damaged areas. What we see understanding the dynamics of long-term recovery
are three different types of recovery: the build- following Hurricane Katrina, it is still under-studied
ing/rebuilding of multiple family unit structures; and the mechanisms driving recovery as a process
the rebuilding of single family structures in heav- are not well understood. While these results are
ily damaged areas (Pass Christian and parts of relatively intuitive, a major contribution of this
Long Beach); and the building of new single family research is the application of a technique, a spa-
housing in areas with no damage (Long Beach). tial scan statistic, and the utilization of building
This new construction may help explain why the permit data to empirically assess locally based
correlation between permits and pre-storm hous- trends in disaster recovery. The application of this
ing in the catastrophic damage zone is so low. We technique demonstrates that space–time clusters
speculate that we might be seeing a retreat from of rebuilding during the period of recovery fol-
the coastline, with new residential construction lowing Hurricane Katrina (or any disaster) can
taking place further inland, rather than simply be identified and analyzed using freely available
replacing the waterfront housing. To document data and software. This research also fills a sig-
this explicitly is beyond the scope of this present nificant gap in the current literature by provid-
paper but is worth future study. ing a much higher temporal resolution for the
Our results indicate that rebuilding following a recovery process.
disaster does not occur uniformly. The process is In this work, the pace and distribution of recov-
concentrated at various points in space and time ery were examined utilizing both SaTScan and
and is under the influence of damage levels and ArcGIS software. This method focuses on long-term
the pre-event housing concentration. Specifically, recovery to understand where and when recovery
we find that the amount of damage experienced by has occurred. Researchers such as Laben (2002)
an area in conjunction with the pre-event number and Curtis et al. (2007) have developed tools to
of houses influences the timeliness of rebuilding track various aspects of recovery, make decisions
and how that rebuilding is spatially distributed. about aid distribution, and identify the drivers of
Areas experiencing higher levels of damage also recovery. Taken together, such work can identify
have new building code requirements related the major influences on rebuilding and “best prac-
to base flood elevations which often reduce the tices” that can be utilized for areas which have not
number of people who are capable of rebuilding experienced the desired level of recovery.
within a confined space and time. The lack of This research can act as a springboard for future
clusters in any specific geographic area does not investigations into the relationships between the
indicate that no rebuilding was occurring, simply level of damage, pre-event housing densities, and
that recovery activities were spatially and tempo- other variables with spatial and temporal clus-
rally diffuse (equal to or less than the expected ters of rebuilding. In addition to increasing our
statistical amount of recovery). understanding of how long-term rebuilding and
and recovery. Cityscape (Washington, DC) 9: 189-215. Hurricane Andrew. Environmental Hazards 4: 45-58.
Rubin, C.B. 2009. Long term recovery from Witham, C.S., and C. Oppenheimer. 2005. Mortality
disasters—The neglected component of emergency in England during the 1783-4 Laki Craters eruption.
management. Journal of Homeland Security and Bulletin of Volcanology 67: 15-26.
Emergency Management. 6(1): Article 46. [http://www. Zottarelli, L.K. 2008. Post-Hurricane Katrina
bepress.com/jhsem/vol6/iss1/46]. employment recovery: The interaction of race and
Rubin, C.B., and D.G. Barbee. 1985. Disaster place. Social Science Quarterly 89(3): 592-607.
recovery and hazard mitigation: Bridging the
intergovernmental gap. Public Administration Review
45(Special Issue): 57-63.