Developments in the Built Environment 4 (2020) 100034

Contents lists available at ScienceDirect

Developments in the Built Environment

journal homepage: www.editorialmanager.com/dibe/default.aspx

A spatio-temporal cluster analysis of structurally deficient bridges in the

contiguous USA
R.W. Amin a, S. Kocak b, *, H.E. Sevil c, G.P. Peterson d, J.T. Hamilton e, B. Rivera e, T. Barraco e
a
Mathematics & Statistics at University of West Florida, 11000 University Parkway, Building 4 Room 336, Pensacola, FL, 32514, USA
b
Construction Management at University of West Florida, 11000 University Parkway, Building 70 Room 125, Pensacola, FL, 32514, USA
c
Intelligent Systems & Robotics at University of West Florida, 11000 University Parkway, Building 4 Room 139, Pensacola, FL, 32514, USA
d
Eisman & Russo, Inc., 220W. Garden St. Suite 200, Pensacola, FL, 32502, USA
e
Mathematics & Statistics at University of West Florida, 11000 University Parkway, Building 4 Room 223, Pensacola, FL, 32514, USA

A R T I C L E I N F O A B S T R A C T

Keywords: The disease surveillance software SaTScan™ is used to identify spatial and space-time clusters of counties with
Structurally deficient bridges in U.S. counties unusually high counts or rates of SD bridges. Initially, a descriptive data analysis of over 600,000 bridges, on
Disease surveillance software SaTScan which data were available for 2017, identified the kind of material and design of all bridges. This was followed by
Spatio-temporal cluster analysis
analyzing data on SD bridges for the 3108 counties. The clusters were tested for significance with Monte Carlo
Epidemiologic approach to structurally
deficient bridges
study to designate significant SD clusters. While the purely spatial analysis was based on data for 2017, the space-
Heat maps time analysis used data for the years 2006–2017. A Negative Binomial regression model was used in addition to a
cluster analysis. Regression analysis was performed to adjust SD counts for several covariates or risk factors. This
study identified counties with high rates of SD bridges as rural counties with old bridges where there is cold
weather and low daily traffic.

1. Introduction which require less maintenance and which provide longer durability
(Cheng and Karbhari, 2006). The literature on the structural integrity of
It is known that outdated roads, bridges, and other vital infrastructure bridges previously introduced methods, such as fiber reinforced polymer
in America need maintenance or repair (Layne, 2019). In 2017, the (FRP) grid system (Yost and Schmeckpeper, 2001), steel-free FRP con-
American Society of Civil Engineers (ASCE) gave a Dþ grade to US crete deck (Cheng et al., 2005), pultrusion deck (Lopez-Anido et al.,
infrastructure (Layne, 2019). In the report from ASCE, there are 16 cat- 1997), and rectangular hollow-based Strongwell System (Hayes et al.,
egories, including roads, parks, bridges, and 12 categories were listed as 2000; Dumlao et al., 1996; Salim and Davalos, 1999), to achieve such an
being “poor” or “at-risk” (Gillies, 2019). Although conditions of high- objective.
ways and roads are equally important, as bad conditions can contribute In today’s economy driven world, one other major point related with
to accidents, the most crucial infrastructure condition to monitor is the any type of construction is the sustainability of bridge constructions
structure of bridges, as their failure can lead to more catastrophic results. (Penades-Pla et al., 2016). Three pillars of sustainability, have been a
One aspect of this aging infrastructure is maintaining and investi- focus of the decision-making process in bridge design in recent years, are
gating bridges for repair. It is known that there are infrastructures in economic, social, and environmental (Penades-Pla et al., 2016). More-
America without having received proper maintenance for more than five over, the areas for decision-making can be grouped into four categories;
decades (Thompson and Matousek, 2019). In this study however, the (i) construction (Chen, 2014; Chou et al., 2013), (ii) operation & main-
focus will be on the spatial analysis of infrastructure defects and relations tenance (Sabatino et al., 2015; Ei-Mikawi and Mosallam, 1996), (iii)
to other factors for bridges in the contiguous US. Over the years, with planning & design (Ardeshir et al., 2014; Chen et al., 2001), and (iv)
improvement of new techniques and developments in technology, sub- demolition or recycling (Chen et al., 2013).
stantial progress has been made in improving bridge construction ma- Despite all the efforts mentioned above, it is an undeniable fact that
terials (Cheng and Karbhari, 2006). The idea here is to utilize materials bridges are aging, and for their continued safe operation, their health

* Corresponding author.
E-mail addresses: [email protected] (R.W. Amin), [email protected] (S. Kocak), [email protected] (H.E. Sevil), [email protected] (G.P. Peterson), jh150@
students.uwf.edu (J.T. Hamilton), [email protected] (B. Rivera), [email protected] (T. Barraco).

https://doi.org/10.1016/j.dibe.2020.100034
Received 2 August 2020; Received in revised form 5 October 2020; Accepted 23 October 2020
Available online 29 October 2020
2666-1659/© 2020 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
R.W. Amin et al. Developments in the Built Environment 4 (2020) 100034

should be monitored in terms of structural integrity (Brownjohn, 2006). method to identify and test the unusually high rates of SD bridges
There are many approaches for structural health monitoring presented in spatially. Moreover, a space-time interaction for the rate of SD bridges for
the literature, some of them being mentioned here: bridge waveform the period 2006–2017 was tested. The data on bridges were categorized
analysis (Heywood et al., 2000), visual inspection (Alampalli and Fu, based on Federal Highway Administration’s (FHWA) National Bridge
1994), dynamic analysis (Wenzel et al., 2002), pattern recognition (Liu, Inventory (NBI) recording and coding guide for the structural inventory
1995), Hilbert-Huang transform (Huang et al., 1998), as well as many and appraisal of the nation’s bridges (Office of Engineering Bridge Di-
different sensor types have been used in health monitoring (Ko and Ni, vision, 1995). This guide uses a three-digit coding system for the struc-
2005), such as cameras (Wong, 2004), seismometer (Wang, 2004), global ture type, which is divided into two groups under type of material, item
positioning system (Aiqun et al., 2003), accelerometer (Ou, 2004), and 43A, and type of design/construction, item 43B. The type of material is
strain gauges (Ni et al., 2004). Although these studies were presented, compiled under 10 categories while the type of design is gathered into 23
little attention has been given to the spatial analysis of bridge structure distinct groups. Table 1 illustrates the codes and corresponding de-
deficiency. In this effort, some of recent researches have focused on scriptions for kind of materials and type of design/construction as stated
spatial-clustering analysis (Li et al., 2007; Cheng and Leu, 2008; Hajia- in FHWA-NBI coding guide.
lizadeh et al., 2016). This will be a key to see cause-effect relations, and The scope included the data analyses on SD bridges for the 3108
may lead to better understanding of planning of infrastructure repair. counties and District of Columbia to produce the heat maps using GIS
The freely available SaTScan™ software is used all over world for the based software ArcMapTM following the descriptive data analysis of over
spatial, temporal and space-time scan statistics in many fields including 600,000 bridges available for the year of 2017. The Chi-square tests
criminology (Leitner and Helbich, 2011), economics (Helbich, 2012), between “SD or not SD” and “type of material”, and “SD or not SD” and
engineering (Kaza et al., 2013), agriculture (Cuadros et al., 2017), ge- “type of design/construction” were studied with the software SAS to find
netics (Enemark et al., 2002), forestry (Vega et al., 2012), geography, out how good or bad each type of material or type of construction
geology (Gao et al., 2014) and many others. The software is designed for affected the structural deficiency. It should be noted that it might not be
any of the following interrelated purposes. the materials performing poorly. However, it might be the impact of the
inspectors who rate the materials lower due to prudent consideration of
1 To test if the event is randomly distributed over space, over time or uncertainty related to the structural evaluation of the type of materials
over space and time. and structures. The clusters from the heat maps were tested for signifi-
2 To perform geographical surveillance, to detect spatial or space-time cance with a Monte Carlo study within a commercially available SaTS-
clusters, and to see if they are statistically significant. can™ software, which designated any cluster with p < 0.05 as a
3 To assess the statistical significance of cluster alarms. significant one. Additionally, a negative binomial regression model was
4 To perform prospective real-time or time-periodic surveillance for the conducted with SAS to adjust SD counts for several covariates or risk
early detection of outbreaks. factors. While the purely spatial analysis was based on data for 2017, the
space-time analysis used data for the years between 2006 and 2017.
The software can use discrete as well as continuous scan statistics data Subsequently, a Joinpoint analysis was performed to identify any sig-
types and methods. For discrete scan statistics, SaTScan uses either a nificant slope change for the annual SD rates between 2006 and 2017.
discrete Poisson-based model, where the number of events in a The special software for Joinpoint was downloaded from the National
geographical location is Poisson-distributed, according to a known un- Cancer Institute (nci.gov).
derlying population at risk; a Bernoulli model, with 0/1 event data such
as cases and controls; a space-time permutation model, using only case 3. Data and methods
data; a multinomial model for categorical data; an ordinal model, for
ordered categorical data; an exponential model for survival time data The mainstream data was provided by Federal Highway Adminis-
with or without censored variables; a normal model for other types of tration (FHWA), which included the total number of bridges in the 3108
continuous data; or a spatial variation in temporal trends model, looking counties of the contiguous US and the corresponding count of structurally
for geographical areas with unusually high or low temporal trends. A deficient bridges at them for the year 2017. Moreover, the data on the
common feature of all these discrete scan statistics is that the total area of the bridges in each US county and the total structurally
geographical locations where data can be observed are non-random and deficient area (measurements are recorded in meters squared) was ac-
fixed by the user. For the discrete scan statistics, the data may be either quired from FHWA’s online database. Additional data on bridges
aggregated at the census tract, zip code, county or other geographical regarding many associated factors were obtained from the National
level, or there may be unique coordinates for each observation. SaTScan Bridge Inventory (NBI), and Department of Homeland Security (DHS).
adjusts for the underlying spatial inhomogeneity of a background pop- The NBI data sets included information on the owner, the year built, the
ulation. It can also adjust for any number of categorical covariates pro- kind of material, the average daily traffic, and additional variables that
vided by the user, as well as for temporal trends, known space-time could be used in studying in more detail the distribution of bridges and
clusters and missing data. It is possible to scan multiple data sets simul- their associated structural conditions. It is worth to re-emphasize that
taneously to look for clusters that occur in one or more of them. For data available may have some limitations such as impact of inspectors,
continuous scan statistics, SaTScan uses a continuous Poisson model inspection methods set by states and training. In addition to the data on
(SaTScan 9.6.1, 2020). Even though this study was conducted on the data bridges, demographic data were initially utilized as covariates on each
available for the contiguous USA, such a surveillance study of the bridges county in the statistical analysis with SAS, such as the information on the
could be done in other parts of the world as long as such data are made rate of people living under the poverty line and the urban versus rural
available. The results of this study can assist the Governmental agencies score for each county. Data for these covariates were found on miscel-
and all other parties in charge of bridge maintenance and repair to focus laneous governmental websites. In order to begin any sort of analysis, the
attention on the geographic areas that are needed more. data had to be compressed and merged into one file. Although there is
data provided for 2006–2017 by the FHWA, only the year 2017 was used
2. Objectives and scopes when merging with the NBI data. The twelve-year span was kept to the
side to analyze separately in a space-time analysis and Joinpoint.
The main objective of this study was to identify significant spatial An epidemiology-based approach was selected to identify and test for
clusters of counties with high rates of structurally deficient (SD) bridges “bridges with poor structural rating” of counties with high rates of
in the contiguous U.S. for 2017. The motive behind this study was to fill bridges with structural problems, using the disease surveillance software
the gap in the literature on SD bridges using the epidemiology-based SaTScan™. This software uses a spatial scan statistic to first identify

2
R.W. Amin et al. Developments in the Built Environment 4 (2020) 100034

Table 1 4. Results
NBI coding description for structure type.
NBI Item 43- Structure Type 4.1. Chi-square test
43A- Kind of Material and/ 43B-Type of design and/or construction
or design Using SAS, the first Chi-square test for association was conducted
between the structural deficiency and type of materials for the bridges, as
Code Description Code Description Code Description
stated at NBI item 43A. Table 2 illustrates the finding of this test. Based
1 Concrete 01 Slab 12 Arch-Thru on the results, materials that lead to a higher rate of SD (with approxi-
2 Concrete 02 Stringer/Multi- 13 Suspension
mately SD > 20%) were wood, masonry and steel in the decreasing order.
continuous beam or Girder
3 Steel 03 Girder and 14 Stayed Firder Wood had the highest rate of SD with roughly 33%.
Floorbeam The second Chi-square test for association performed was between
System the structural deficiency and type of design/construction, as stated at NBI
4 Steel continuous 04 Tee Beam 15 Movable-Lift item 43B. The test provided information on how good or bad each con-
5 Prestressed 05 Box Beam or 16 Movable-
concrete * Girders- Multiple Bascule
struction type had an impact on the number of structurally deficient
6 Prestressed 06 Box Beam or 17 Movable- bridges. Table 3 demonstrates the results of Chi-square test for structural
concrete Girders- Single or Swing deficiency and type of bridge construction/design. While the blue
continuous * Spread colored row provided the total counts of bridge design type, the yellow
7 Wood or Timber 07 Frame (expect 18 Tunnel
row summarizes the total percentage of the bridges for that design type.
frame culverts)
8 Masonry 08 Orthotropic 19 Culvert The mostly used design type in the contiguous US is the stringer/
9 Aluminum, 09 Truss-Deck 20 Mixed types multi-beam or girder construction with almost 40%. The construction
Wrough Iron, or type that lead more to SD (with approximately SD > 20%) were con-
Cast Iron struction types 3, 9, 10, 13, 16 and 17, which corresponds to girder and
0 Other 10 Truss-Thru 21 Segmental
Box Girder
floor-beam system, truss-deck, truss-thru, suspension, movable-bascule
* Post-tension concrete 11 Arch-Deck 22 Channel and movable-swing, respectively. Truss-thru type of construction had
should be coded as Beam the highest SD with a little over 50%.
prestressed concrete 00 Other
4.2. Spatial analysis

A purely spatial cluster analysis was used with the disease surveil-
spatial clusters, followed by testing each cluster for significance. Basi- lance software SaTScan to identify 45 significant spatial clusters of
cally, we are asking the question whether clusters are existing due to counties with high counts of structurally deficient bridges based on the
random causes or not. The maximum spatial cluster size in our spatial Discrete Poisson model.
study has been set to 25% of the county population at risk, which means Fig. 1 shows a heat map with the normalized rates of structurally
that any identified cluster’s population may not exceed 25% of the deficient bridges for the 3109 counties in the contiguous states. The rates
population of the contiguous US. In this study, only areas with high rates range from a low of 1.91 to 1.24 standard errors away from the mean
were scanned. Using an epidemiologic mapping methodology, we test the (0) (deep green) to a high of 1.24–3.53 standard errors above the mean
null hypothesis that the rates of structurally deficient bridges are equal (0) (deep red). While the western half of the USA displays mostly green
across the 3108 counties in the contiguous US, versus the hypothesis that and yellow colored counties, red and orange counties were gathered
in some counties the rates of structurally deficient bridges is higher. The more frequently in the eastern half of the USA. The 45 significant spatial
Discrete Poisson model in SaTScan™ was used to identify clusters based clusters were numbered 1–45 according to the number of structurally
on the largest Likelihood Ratio observed, while the p-values are obtained deficient bridges in each cluster. Cluster 1 has 15,568 cases and relative
from SaTScan™ via a Monte Carlo study. More information on using risk 1.87, while Cluster 2 has 5878 structurally deficient bridges and
SaTScan™ is given at www.satscan.org. relative risk 1.74. Only clusters with p < 0.05 were kept as significant
For the statistical analysis, several key covariates which are as follows clusters. Table 4 lists information on 17 significant clusters with p < 0.05.
are investigated: 1. Average age of all bridges in a county 2. Average
inspection frequency for all bridges in a county, 3. Average total area of 4.3. Space-time analysis
all bridges in a county, 4. Poverty rate in a county, 5. Percent urban of
each county, 6. number of days with rain exceeding 1 inch in a county, 7. While the purely spatial analysis was done with the last available data
Number of days with temperature high exceeding 95oF in a county, 8. on SD bridges (2017), the space-time analysis used data for the years
Minimum temperature in a county, 9. Average Daily Travel per bridge in 2006–2017 as we tested for a space-time interaction. The main goal is to
a county. identify and then to test clusters of high rates of SD bridges at the county
Count data is used for structurally deficient bridges in a Poisson resolution each year in order to detect any existing space-time interac-
Regression model, and the total number of bridges per county is kept in tion. Fig. 2 shows the 25 significant space-time clusters (p < 0.005),
the model as a weight. A Poisson Regression model was used to predict while Table 5 lists detailed information including the Relative Risk (RR)
the count of structurally deficient bridges from these covariates to for each cluster on each of these 25 clusters. The 5 clusters with the
identify associations between each covariate used and the count of largest sizes (1,2,3,4,7) have space-time interaction (a surge) during
structurally deficient bridges per county. SaTScan™ provides a meth- 2006–2011. Smaller sized clusters may have large RR values.
odology for adjusting the response variable (the counts) for several
covariates by using the predicted values from the output from the Poisson 4.4. Covariates analysis
Regression and contrast it in SaTScan™ with the observed counts for the
structurally deficient bridges (without adjusting the counts for the Table 6 lists the effects of adjusting the structurally deficient count by
covariates) to identify the effect of each covariate on the predicted several covariates or risk factors based on a Negative Binomial regression
counts. This is then followed by creating heat maps and cluster maps with analysis with the statistical software SAS.
the GIS based software ArcMap™. While many covariates could be used here, we focused on 9 covariates
which are either directly associated with bridges or indirectly associated
as demographic factors on counties with bridges. The covariates are as

3
R.W. Amin et al. Developments in the Built Environment 4 (2020) 100034

Table 2
SD bridges versus NBI kind of material (043A).

Table 3
SD bridges versus NBI kind of design/construction (043B).

follows: 1. Average age of all bridges in a county, 2. Average inspection bridges, such as poverty rate in a county, even though this covariate is
frequency for all bridges in a county, 3. Average total area of all bridges statistically significant. As the bridge age increases or the minimum
in a county, 4. Poverty rate in a county, 5. Percent urban of each county, temperature decreases, the rate of structurally deficient bridges also in-
6. Number of days with rain exceeding 1 inch in a county, 7. Number of creases. As the amount of rainfall increases and as the daily high tem-
days with temperature high exceeding 95  F in a county, 8. Minimum peratures decreases, the rate of structurally deficient bridges increases.
temperature in a county, and 9. Average Daily Travel per bridge in a The sign of each effect (-, þ) indicates whether the rate of structurally
county. The effects are either positive (þ) or negative (), indicating a deficient bridges increases or decreases when this effect increases. The
higher rate of structurally deficient bridges for a positive effect and covariates indicate that rural counties with cold, rainy climate, less
indicating the opposite for a negative effect. The magnitude of Wald’s traveled bridges and high poverty rate are counties that tend to have a
Chi-square Test statistic can be used as a measure of relative importance high rate of structurally deficient bridges. Using a Poisson Regression
for each covariate. The larger the test statistic, the more important the model allows us to adjust the observed counts of structurally deficient
covariate is in the prediction of counties with a high rate of structurally bridges by all covariates to identify geographic regions in which these
deficient bridges. The regression analysis considers the joint contribu- covariates either were associated with the clusters or they had no effect
tions from all 9 covariates. The results show these contributions in order on the clustering of counties with structurally deficient bridges.
of importance, the top 5 covariates are (1) Average age of all bridges (þ, Fig. 3 uses three colors to make it very easy to identify all counties
W ¼ 171.12) (2), Minimum temperature (-, W ¼ 136.65) (3), Average that stayed inside a cluster even after the counts were adjusted for all
number of days with temperature exceeding 95  F (-, W ¼ 58.78) (4), covariates. These counties are shown in orange. The clusters are due to
Average Daily Travel per bridge (-, W ¼ 41.14), and (5) Number of Days some factors that were not used in this study, such as type of building
with Rain exceeding 1 inch (þ, W ¼ 35.36). All other covariates played a material or other. The red colored counties were inside some cluster, but
smaller role in the prediction of counties with structurally deficient after adjusting for the covariates, these counties were not anymore in a

4
R.W. Amin et al. Developments in the Built Environment 4 (2020) 100034

Fig. 1. Heatmap for rate of normalized structurally deficient bridges and significant spatial clusters in the contiguous USA in 2017.

Table 4
Structurally deficient bridges purely spatial clusters details.
CLUSTER OBSERVED EXPECTED O/E REL_RISK POPULATION P-Value

1 15,568 9635.64 1.62 1.87 108803.00 <0.00000000000000001

2 5878 3538.34 1.66 1.74 39954.00 <0.00000000000000001
3 2115 1268.10 1.67 1.70 14319.00 <0.00000000000000001
4 1052 505.95 2.08 2.10 5713.00 <0.00000000000000001
5 65 11.60 5.60 5.61 131.00 <0.00000000000000001
6 191 78.20 2.44 2.45 883.00 <0.00000000000000001
7 1583 1234.80 1.28 1.29 13943.00 0.0000000000000022
8 330 193.95 1.70 1.71 2190.00 0.00000000000078
9 244 138.51 1.76 1.77 1564.00 0.00000000029
10 262 163.04 1.61 1.61 1841.00 0.00000018
11 134 72.44 1.85 1.85 818 0.0000097
12 101 49.51 2.04 2.04 559 0.000013
13 74 32.06 2.31 2.31 362 0.000022
14 143 87.41 1.64 1.64 987 0.002
15 49 20.63 2.37 2.38 233 0.004
16 32 11.69 2.74 2.74 132 0.029
17 17 4.07 4.17 4.17 46 0.041

cluster. This means that the used covariates “explained” the high counts is also useful to consider annual rates of structurally deficient bridges at
or rates of structurally deficient bridges. The yellow counties are counties the national level of resolution. In particular, the nonparametric regres-
that were not inside any cluster before we adjusted for the covariates. sion methodology Joinpoint is used by the National Cancer Institute to
Such counties have counts or rates of structurally deficient bridges that check whether annual cancer rates significantly change from year to year.
are inside “new clusters” after adjusting for the covariates. They need to The software can be retrieved from nci.gov.
be inspected to see if the counts or rates structurally deficient are “too A permutation test is used in which the slope of the regression line is
high” when covariates are considered. tested for each subsequent year to see if the slope is changing signifi-
cantly or not. The year for the significant slope change would then be
called a Joinpoint. Fig. 4 demonstrates a declining rate of structurally
4.5. Joinpoint analysis
deficient bridges for the period 2006–2017, but a Joinpoint is identified
in 2011. The regression slope was 1.00 for 2006–2010, and it changed
While this study uses the county level resolution to identify clusters, it

5
R.W. Amin et al. Developments in the Built Environment 4 (2020) 100034

Fig. 2. A space-time interaction cluster map of structurally deficient bridges in the contiguous USA in 2006–2017.

Table 5
Structurally deficient bridges space-time clusters details (p < 0.005).
CLUSTER OBSERVED EXPECTED O/E REL RISK POPULATION START DATE END DATE

1 113,961 65882.65 1.73 1.85 100133.43 2006 2011

2 48,668 25905.43 1.88 1.94 39781.38 2006 2011
3 16,578 9681.52 1.71 1.73 14716.92 2006 2011
4 10,293 6084.00 1.69 1.70 9457.96 2006 2011
5 3109 1615.02 1.93 1.93 2484.35 2006 2011
6 1198 429.31 2.79 2.79 659.67 2006 2011
7 15,673 12316.68 1.27 1.28 18957.04 2006 2011
8 2016 1009.41 2.00 2.00 1558.10 2006 2011
9 316 55.95 5.65 5.65 170.67 2013 2015
10 195 35.86 5.44 5.44 310.57 2006 2006
11 462 173.52 2.66 2.66 265.42 2006 2011
12 705 339.33 2.08 2.08 521.59 2006 2011
13 2178 1479.61 1.47 1.47 2269.76 2006 2011
14 261 99.19 2.63 2.63 152.92 2007 2012
15 2137 1627.02 1.31 1.31 2502.19 2006 2011
16 488 269.44 1.81 1.81 427.51 2006 2011
17 838 554.20 1.51 1.51 848.17 2010 2015
18 413 238.58 1.73 1.73 363.00 2006 2011
19 744 506.19 1.47 1.47 1168.17 2006 2009
20 1177 897.21 1.31 1.31 1381.44 2007 2012
21 152 79.01 1.92 1.92 119.33 2006 2011
22 175 102.33 1.71 1.71 154.01 2007 2012
23 596 458.38 1.30 1.30 697.50 2006 2011
24 255 170.32 1.50 1.50 260.67 2012 2017
25 64 27.08 2.36 2.36 42.00 2011 2016

6
R.W. Amin et al. Developments in the Built Environment 4 (2020) 100034

Table 6
Negative binomial regression analysis of structurally deficient bridges, adjusted for all covariates.
Analysis of Maximum Likelihood Parameter Estimates

Parameter DF Estimate Standard Error Wald 95% Confidence Limits Wald Chi-Square Pr > ChiSq

Intercept 1 3.0955 0.1521 3.3936 2.7974 414.19 <.0001

Avg_Age 1 0.0246 0.0019 0.0209 0.0283 171.12 <.0001
Min_Temp 1 0.0232 0.002 0.0271 0.0193 136.65 <.0001
Days_Temp_Over_95 1 0.0036 0.0005 0.0045 0.0027 58.78 <.0001
New_Avg_ADT 1 2.1747 0.3391 2.8393 1.5102 41.14 <.0001
Days_Rain_Over_1_Inc 1 0.0169 0.0028 0.0113 0.0224 35.36 <.0001
PctPoverty 1 1.2507 0.2565 0.7479 1.7534 23.77 <.0001
PctUrban 1 0.1382 0.0566 0.2491 0.0273 5.96 0.0146
New_Total_Area 1 0.0162 0.0077 0.0012 0.0313 4.49 0.0341
Avg_Inspect_Freq 1 0.0015 0.0041 0.0096 0.0065 0.14 0.7101

Fig. 3. Cluster map for rate of normalized structurally deficient bridges and selected covariates in the USA in 2017.

to a steeper decline to a slope of 4.26 for the years 2010. This is an outbreaks.
indication that “something” was done around 2010 that resulted in this Based on the results, the most significant outcome was that counties
sharp decline afterwards. in rural areas with cold weather, small size, and less traveled bridges
were the ones with high counts or rates of SD bridges with the strongest
5. Discussions impact to be cold weather followed by effect of traffic on the bridge. This
phenomenon may be explained with the weathering process. Bridges
This study focused on the geographical analysis of structurally defi- loses heat energy from top, bottom and sides compared to only top for the
cient bridges in US counties from a spatial and spatio-temporal point of roadways in cold weather. As a result, bridges cool down quickly and
view using cluster analysis. The corresponding risk factors selected for freezes along with the water due to rain or snow in the bridge openings,
structurally deficient bridges were age of the bridge, temperature, such as joints and cracks. While water freezes, it expands in volume and
poverty, area of the bridge, daily traffic. The analyses were based on the exerts pressures in all directions. This can result in existing openings to
approach for epidemiological methodology for detecting disease enlarge and new ones to be formed. Infiltration of aggressive media, such

7
R.W. Amin et al. Developments in the Built Environment 4 (2020) 100034

Fig. 4. Joinpoint graph for annual percent change of structurally deficient bridges ratio in the USA 2006 to 2017.

as deicing agents and salts, in those openings can further increase the uplift force observed in cold weathers and may minimize the number of
deterioration of bridge structural members, such as truss members, SD bridges. As any other material, aging may increase the number of SD
gusset plates, beam ends at expansion joints and piers, and reinforcing bridges.
bars through corrosion. Although all bridges meet bridge design codes, Another perspective to the analysis can be added considering the data
design requirements of the low volume rural bridges are less complex on maintenance and repair expenditure of bridges per county. Although
than the ones of more heavily used bridges since the design parameters the authors contacted necessary agencies and attempted to access such
such as average annual daily truck traffic, axle load spectra, axle weight data, it was not made available to the authors during manuscript prep-
and distance, number of maximum axles allowed by each states are aration. The expenditure part is reserved for the future work.
different. As the bridges deteriorate over time, the resources for main-
tenance and rehabilitation may be shared unevenly between low volume 6. Conclusions
and heavily used bridges. This may result in urbanized heavily used
bridges receiving more maintenance than the low traffic bridges. As a This paper investigated the effect of various risk factors on the rate of
result, rural bridges will deteriorate faster with a little or no maintenance structurally deficient bridges in the contiguous US counties for 2017 by
and this can out them into a lower structural rating category. In such using an epidemiologic methodology, which detected the significant
cases, to keep the bridges open to the traffic, the load and speed limita- spatial clusters of counties with high rates of SD bridges. The importance
tions may be enforced. of each risk factor was ranked based on regression analysis and subse-
This study investigated the link between SD bridges and risk factors, quently used in cluster analysis to identify the counties with unusually
as well. Based on the findings, the most important five covariates were high rates of SD bridges. Although this research was performed for the
average age of the bridges, minimum temperatures, number of days with contiguous USA, such a surveillance studies can be conducted any part of
temperatures over 95oF, average daily traffic, number of days with rain the world as long as such data are made available. Based on the results of
exceeding 1-inch, and poverty rate of the county. The other covariates the analyses, the following conclusions can be made.
played a smaller role in the prediction of SD bridges. The relation be-
tween SD bridges and risk factors can follow the same rationale as ● The highest rate of SD occurred on wood bridges, with almost 34% of
explained for significant outcomes. The inverse relation between SD all the wood bridges were structurally deficient. Masonry and steel
bridges and average daily traffic may be due to the precise and better bridges followed the wood bridges with 24% and 19%, respectively.
design parameters required for heavy traffic volume bridges. Alterna- ● In terms of construction type, bridges with truss-thru, movable swing
tively, the days with temperatures over 95oF may reverse the impact of and girder-floor-beam systems had the highest three rates of SD,

8
R.W. Amin et al. Developments in the Built Environment 4 (2020) 100034

respectively. While a little more than 50% of the truss-thru type of Gao, J., Zhang, Z., Hu, Y., Bian, J., Jiang, W., Wang, X., 2014. Geographical distribution
patterns of iodine in drinking-water and its associations with geological factors in
bridges were structurally deficient, the rates were about 31% for
Shandong Province, China. Int. J. Environ. Res. Publ. Health 11 (5), 5431–5444.
movable swing and 30% for girder-floor-beam type of construction. https://doi.org/10.3390/ijerph110505431.
● Disease surveillance software SatScan with a Discrete Poisson model Gillies, T., 2019. US Infrastructure Is Crumbling, and it Needs Lots of Money to Fix it: Civil
showed on a heap map that there were 17 significant spatial clusters Engineer Group. CNBC. https://www.cnbc.com/2019/02/03/us-infrastructure-is-cru
mbling-and-it-needs-big-money-to-fix-it-asce.html. (Accessed 31 July 2020).
of the counties in the contiguous states with normalized rates of SD Hajializadeh, D., Stewart, M.G., Enright, B., Obrien, E., 2016. Spatial time-dependent
bridges. Most of the spatial clusters were in the eastern half of the reliability analysis of reinforced concrete slab bridges subject to realistic traffic
States. loading. Structure and Infrastructure Engineering 12 (9). https://doi.org/10.1080/
15732479.2015.1086385.
● The space-time analysis of SD bridges in 3108 counties between 2006 Hayes, M.D., Ohanehi, D., Lesko, J.J., Cousins, T.E., Witcher, D., 2000. Performance of
and 2017 illustrated that there were 25 space-time clusters and 5 of tube and plate fiberglass composite bridge deck. J. Compos. Construct. 4 (2), 48–55.
them (1, 2, 3, 4, 7) had space-time surge during 2006–2011. Helbich, M., 2012. Beyond potsuburbia? Multifunctional service agglomeration in
Vienna’s urban fringe. Journal of Economic and Social Geography 103 (1), 39–52.
● Based on negative binomial regression analysis results of joint con- Heywood, R., Roberts, W., Taylor, R., Andersen, R., 2000. Fitness-for-purpose evaluation
tributions of 9-risk factors directly or indirectly associated with of bridges using health monitoring technology. Transport. Res. Rec. 1696 (1),
bridges, older bridges in colder climates with rainy days and less 193–201.
Huang, N.E., Shen, Z., Long, S.R., Wu, M.C., Shih, H.H., Zheng, Q., et al., 1998. The
amount of average daily traffic had the higher rate of structural empirical mode decomposition and the Hilbert spectrum for nonlinear and non-
deficiencies. stationary time series analysis. Proc. R. Soc. Lond. - Ser. A: Math., Phy. Eng. Sci. 454
● Joinpoint analysis revealed that the rate of SD bridges significantly (1971), 903–995.
Kaza, N., Lester, T.W., Rodriguez, D.A., 2013. The spatio-temporal clustering of green
changed in the year 2011, with a sharp decline.
buildings in the United States. Urban Stud. 50, 3262–3282.
Ko, J.M., Ni, Y.Q., 2005. Technology developments in structural health monitoring of
Declaration of competing interest large-scale bridges. Eng. Struct. 27 (12), 1715–1725.
Layne, R., 2019. America’s Aging Infrastructure: what Would $2 Trillion Actually Buy?
CBS News. https://www.cbsnews.com/news/infrastructure-what-will-2-trillion-buy-
none. and-how-will-america-pay/. (Accessed 31 July 2020).
Leitner, M., Helbich, M., 2011. The impact of hurricanes on crime: a spatio-temporal
References analysis in the city of Houston, TX. Cartogr. Geogr. Inf. Sci. 37, 214–222.
Li, R., Ye, L.Y., Chen, Y., 2007. Static and dynamic spatial analysis of super-wide arch
bridge using finite element method. J. Highw. Transp. Res. Dev. 10.
Aiqun, L., Changqing, M., Zhaoxia, L., 2003. Health monitoring system for the Runyang Liu, P.L., 1995. Identification and damage detection of trusses using modal data. J. Struct.
Yangtse river bridge. J. Southeast Univ.: Natural Sci. Edition 33 (5), 544–548. Eng. 121 (4), 599–608.
Alampalli, S., Fu, G., 1994. Instrumentation for remote and continuous monitoring of Lopez-Anido, R., GangaRao, H.V., Vedam, V., Overby, N., 1997. Design and Evaluation of
structure conditions. Transport. Res. Rec. (1432), 59–67. a Modular FRP Bridge Deck, 3-E. Institutes International Composites Expo, Nashville,
Ardeshir, A., Mohseni, N., Behzadian, K., Errington, M., 2014. Selection of a bridge TN, pp. 1–6, 1997.
construction site using fuzzy analytical hierarchy process in geographic information Ni, Y.Q., Ko, J.M., Zhou, H.F., Hua, X.G., 2004. Challenges in developing structural health
system. Arabian J. Sci. Eng. 39 (6), 4405–4420. monitoring system for a long-span cable-stayed bridge. In: Proceedings of the US-
Brownjohn, J.M., 2006. Structural health monitoring of civil infrastructure. Phil. Trans. Korea Joint Seminar/Workshop on Smart Structures Technologies, Seoul, Korea, 2-4
Math. Phys. Eng. Sci. 365 (1851), 589–622. September, pp. 150–157.
Chen, T.Y., 2014. The extended linear assignment method for multiple criteria decision Ou, J., 2004. The state-of-the-art and application of intelligent health monitoring systems
analysis based on interval-valued intuitionistic fuzzy sets. Appl. Math. Model. 38 for civil infrastructures in mainland of China. Prog. Struct. Eng., Mech. Comput.
(7–8), 2101–2117. 599–608.
Chen, S.Y., Fu, G.T., Wang, J.M., Liu, G., 2001. Fuzzy optimum model of semi-structural Penades-Pla, V., García-Segura, T., Martí, J., Yepes, V., 2016. A review of multi-criteria
decision for lectotype optimization of offshore platforms. China Ocean Eng. 15 (4), decision-making methods applied to the sustainable bridge design. Sustainability 8
453–466. (12), 1295.
Chen, Z., Abdullah, A.B., Anumba, C.J., Li, H., 2013. ANP experiment for demolition plan Sabatino, S., Frangopol, D.M., Dong, Y., 2015. Sustainability-informed maintenance
evaluation. J. Construct. Eng. Manag. 140 (2), 06013005. optimization of highway bridges considering multi-attribute utility and risk attitude.
Cheng, L., Karbhari, V.M., 2006. New bridge systems using FRP composites and concrete: Eng. Struct. 102, 310–321.
a state-of-the-art review. Prog. Struct. Eng. Mater. 8 (4), 143–154. SaTScanTM version 9.6.1. Last visited on 10/01/2020. https://www.satscan.org.
Cheng, Y.M., Leu, S.S., 2008. Constraint-based clustering model for determining contract Salim, H.A., Davalos, J.F., 1999. FRP composite short-span bridges: analysis, design and
packages of bridge maintenance inspection. Autom. ConStruct. 17 (6), 682–690. testing. J. Adv. Mater. 31 (1), 18–26.
https://doi.org/10.1016/j.autcon.2007.12.001. Thompson, C., Matousek, M., 2019. America’s Infrastructure Is Decaying — Here’s a Look
Cheng, L., Zhao, L., Karbhari, V.M., Hegemier, G.A., Seible, F., 2005. Assessment of a at How Terrible Things Have Gotten. Business Insider. https://www.businessinsider.
steel-free fiber reinforced polymer-composite modular bridge system. J. Struct. Eng. com/asce-gives-us-infrastructure-a-d-2017-3. (Accessed 31 July 2020).
131 (3), 498–506. Vega, O.C., Tonini, M., Conedera, M., Kanveski, M., 2012. Cluster recognition in spatial-
Chou, J.S., Pham, A.D., Wang, H., 2013. Bidding strategy to support decision-making by temporal sequences: the case of forest fires. GeoInformatica 16, 653–673.
integrating fuzzy AHP and regression-based simulation. Autom. ConStruct. 35, Wang, M.L., 2004. State-of-the-art applications in health monitoring. In: Invited
517–527. Presentation to Workshop on Basics of Structural Health Monitoring and Optical
Cuadros, D.F., Hernandez, A., Torres, M.F., Torres, D.M., Branscum, A.J., Rincon, D.F., Sensing Technologies in Civil Engineering. National Central University, Taiwan,
2017. Vector transmission alone fails to explain the potato yellow vein virus epidemic pp. 13–14.
among potato crops in Colombia. Front. Plant Sci. 8, 1654. Wenzel, H., Geier, R., Engineers, V.V.C., 2002. Bridge classification based upon ambient
Dumlao, C., Lauraitis, K., Abrahamson, E., Hurlbut, B., Jacoby, M., Miller, A., Thomas, A., vibration Measurements. In: Proc. First European Conference on Structural Health
1996. Demonstration low-cost modular composite highway bridge. In: First Monitoring, Paris, pp. 981–988.
International Conference on Composites in Infrastructure, Tucson, AZ, 1996, Wong, K.Y., 2004. Instrumentation and health monitoring of cable-supported bridges.
pp. 1141–1155. Struct. Contr. Health Monit. 11 (2), 91–124.
Ei-Mikawi, M., Mosallam, A.S., 1996. A methodology for evaluation of the use of Yost, J.R., Schmeckpeper, E.R., 2001. Strength and serviceability of FRP grid reinforced
advanced composites in structural civil engineering applications. Compos. B Eng. 27 bridge decks. J. Bridge Eng. 6 (6), 605–612.
(3–4), 203–215.
Enemark, H.L., Ahrens, P., Juel, C.D., Petersen, E., Petersen, R.F., Andersen, J.S., Lind, P.,
Thamsborg, S.M., 2002. Molecular characterization of Danish Cryptosporidium
parvum isolates. Parasitology 125, 331–341.