Science of the Total Environment 490 (2014) 538–544

Contents lists available at ScienceDirect

Science of the Total Environment
journal homepage: www.elsevier.com/locate/scitotenv

Challenges associated with projecting urbanization-induced
heat-related mortality
David M. Hondula a,b,⁎, Matei Georgescu b,c, Robert C. Balling Jr.
a
b
c

b,c

Center for Policy Informatics, Arizona State University, Phoenix, AZ 85004, USA
School of Geographical Sciences and Urban Planning, Arizona State University, Tempe, AZ 85287, USA
Global Institute of Sustainability, Arizona State University, Tempe, AZ 85287, USA

H I G H L I G H T S
•
•
•
•
•

Extreme heat is a public health hazard in metropolitan Maricopa County, Arizona.
Urbanization is a key driver of future regional temperature changes.
Projections of heat–health impacts for 2050 vary across urbanization scenarios.
The sign and magnitude of projections are also sensitive to exposure variable choice.
Consideration of urbanization effects is critical for heat–health policymaking.

a r t i c l e

i n f o

Article history:
Received 1 February 2014
Received in revised form 15 April 2014
Accepted 30 April 2014
Available online 28 May 2014
Editor: P. Kassomenos
Keywords:
Heat-related mortality
Urbanization
Climate change
Health
Extreme heat

a b s t r a c t
Maricopa County, Arizona, anchor to the fastest growing megapolitan area in the United States, is located in a hot
desert climate where extreme temperatures are associated with elevated risk of mortality. Continued urbanization in the region will impact atmospheric temperatures and, as a result, potentially affect human health. We
aimed to quantify the number of excess deaths attributable to heat in Maricopa County based on three future
urbanization and adaptation scenarios and multiple exposure variables. Two scenarios (low and high growth
projections) represent the maximum possible uncertainty range associated with urbanization in central Arizona,
and a third represents the adaptation of high-albedo cool roof technology. Using a Poisson regression model, we
related temperature to mortality using data spanning 1983–2007. Regional climate model simulations based on
2050-projected urbanization scenarios for Maricopa County generated distributions of temperature change, and
from these predicted changes future excess heat-related mortality was estimated. Subject to urbanization scenario and exposure variable utilized, projections of heat-related mortality ranged from a decrease of 46 deaths per
year (− 95%) to an increase of 339 deaths per year (+ 359%). Projections based on minimum temperature
showed the greatest increase for all expansion and adaptation scenarios and were substantially higher than
those for daily mean temperature. Projections based on maximum temperature were largely associated with declining mortality. Low-growth and adaptation scenarios led to the smallest increase in predicted heat-related
mortality based on mean temperature projections. Use of only one exposure variable to project future heatrelated deaths may therefore be misrepresentative in terms of direction of change and magnitude of effects.
Because urbanization-induced impacts can vary across the diurnal cycle, projections of heat-related health outcomes that do not consider place-based, time-varying urban heat island effects are neglecting essential elements
for policy relevant decision-making.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction
Extreme heat is the leading weather-related cause of death in the
United States (Luber et al., 2006), and urbanization, population aging,
⁎ Corresponding author at: 411 N. Central Avenue, Suite 400, Center for Policy
Informatics, Arizona State University, Phoenix, AZ, USA. Tel.: +1 908 472 9462.
E-mail address: david.hondula@asu.edu (D.M. Hondula).

http://dx.doi.org/10.1016/j.scitotenv.2014.04.130
0048-9697/© 2014 Elsevier B.V. All rights reserved.

and global-scale climate change will likely conspire in future years to
increase population vulnerability to heat. Despite a growing body of research advancing our understanding of spatial, temporal, environmental, social, and behavioral dimensions of heat-related health risks (e.g.,
Anderson and Bell, 2009; Harlan et al., 2013; Hondula et al., 2012), the
institution of warning systems and other intervention measures (e.g.,
Hondula et al., 2013; Sheridan and Kalkstein, 2010), and the fact that
relatively simple measures can prevent heat-related illnesses and

D.M. Hondula et al. / Science of the Total Environment 490 (2014) 538–544

deaths (e.g., Luber and McGeehin, 2008; O'Neill et al., 2009), heat persists as a public health burden. There has been an increase in research
efforts to apply climate change projections to health outcomes, with
the greatest emphasis on predicting future heat-related mortality.
From a global perspective, increasingly frequent and severe hot days
are expected to lead to an increase in corresponding negative health
outcomes (McMichael et al., 2006). There is now interest in quantifying
the related health burdens to inform planning and policy (Ebi et al.,
2006; Gosling et al., 2009; Sheridan et al., 2012). The level of an investment a particular municipality would make in mitigation and adaptation programs aimed at reducing thermally stressful situations in the
future would likely be influenced by the magnitude of the expected
changes in undesirable heat-attributable health outcomes.
Populated cities in hot climates offer an interesting case for analysis
of resident sensitivity to heat under current and future climatic conditions. The population of Maricopa County, Arizona, home of the Phoenix
metropolitan area, the hottest major urban area in the United States, is
not immune to the dangerous effects of heat (Chuang et al., 2013;
Harlan et al., 2013). Unlike many other large metropolitan areas in the
United States where episodic heat is associated with elevated risk of illness and death (Anderson and Bell, 2009, 2011; Barnett et al., 2010;
Sheridan and Kalkstein, 2004), dangerously high temperatures persist
in Maricopa County for much of the warm season. Whereas cities in
the southeastern United States exhibit decreased (or no) sensitivity to
heat compared with mid-Atlantic and northeastern cities (Curriero
et al., 2002; Davis et al., 2003), the temperature extremes present for
such an extended period of time result in high heat sensitivity among
residents of Maricopa County.
A wide range of studies has documented a significant heat effect in
Maricopa County and its associated cities (e.g., Phoenix, Tempe,
Glendale, Scottsdale, Mesa, Chandler, Gilbert, and Peoria) (e.g., Harlan
et al., 2013; Yip et al., 2008). Most recently, the Maricopa County
Department of Public Health reported 106 heat-related deaths in the
county in 2011 and at least 102 deaths in 2012 (with 13 cases pending
review) (MCDPH, 2013). Heat-related mortality in the region occurs
among young and elderly populations alike: a majority of indoor deaths
during the period 2000–2005 occurred among the elderly while a majority of outdoor deaths occurred among those less than five years old
(Yip et al., 2008). Analysis of a different time period (2002–2009)
showed that men in agricultural and construction/extraction occupations were at particularly high risk (Petitti et al., 2013). National-scale
studies have reached mixed conclusions regarding the sensitivity of
the region's population to extreme heat compared to other cities, ranging from little to no effect of heat (Kalkstein and Greene, 1997; Sheridan
and Kalkstein, 2010) to statistically significant effects equal or greater
than those in other major U.S. cities (Anderson and Bell, 2009, 2011;
Saha et al., 2014).
Here, we compare projections of heat-related deaths for Maricopa
County using regional urbanization scenarios associated with high
growth, low growth, and adaptation-oriented growth (Georgescu
et al., 2013). In Maricopa County, urbanization and urban heat island intensity have been linked to higher heat vulnerability (Chow et al., 2012;
Golden et al., 2008; Harlan et al., 2006; Jenerette et al., 2011; Ruddell
et al., 2009). There has been minimal research specifically examining
the extent to which urbanization (versus greenhouse gas-forced climate
change) may contribute to future increases or decreases in heat-related
mortality, which is important to consider because urbanization rates
can be managed at the local and regional levels. Urban development
modifies regional climate primarily through reduced evening and nighttime longwave energy loss to the overlying atmosphere, relative to rural
areas, based on the geometry and higher heat capacity of the built environment (Georgescu et al., 2011). We also explore the extent to which
choice in exposure variable (e.g., maximum, mean, or minimum temperature) impacts the projection, which has been largely unaddressed
to date (Huang et al., 2011). Exposure variable is especially important
to consider in semiarid environments where urbanization-induced

539

changes can lead to substantial differences in projected minimum and
maximum temperatures.

2. Methods
2.1. Data sources
Maricopa County is located in central Arizona in the northern portion of the Sonoran Desert (see Fig. 1). The urbanized area is contained
largely in a valley surrounded by mountains ranging from 300 m to
2000 m above the valley floor. Summer afternoon temperatures regularly exceed 40 °C, and occasionally approach 50 °C. The metropolitan
area continues to be one of the fastest growing in the United States,
growing from just over 3,000,000 residents in 2000 to over 4,000,000
in 2010, despite a downturn in the economy during that time period.
The County is the anchor of the ever-growing Arizona Sun Corridor
that extends from Tucson and Nogales to the south and Prescott to the
north and west.
All-cause daily mortality data for residents of Maricopa County were
obtained from the Office of Vital Records at the Arizona Department of
Health Services for the time period 1983–2007. Over the 25-year period
of record there were 472,327 deaths in the study region. The increasing
daily mortality counts over the period of record are primarily reflective
of the rapid population growth in Maricopa County (see Supplemental
Material Fig. S1). Seasonal variability is also evident in the data with
mortality peaking in January and at a minimum in August. We elected
to use all-cause mortality as the variable of interest for this study
because exposure to high ambient temperatures can exacerbate a
range of health problems and lead to premature death (Sheridan and
Kalkstein, 2004; Kovats and Hajat, 2008). The use of all-cause mortality
also facilitates comparison with other studies projecting future heatrelated health impacts (e.g., Greene et al., 2011; Petkova et al., 2013;
Sheridan et al., 2012). The use of human subject data in this research
is exempt from the IRB per Title 45 Part 46 Exemption Category 4.
These data were preserved by public agencies for research purposes
and personal identifiers were removed, eliminating the requirement
for consent.
Daily minimum and maximum temperature data for the geographic
centroid of Maricopa County were obtained from Daymet (Thornton
et al., 2012). The Daymet model provides daily values for a suite of
meteorological parameters over a 1 km by 1 km continuous surface spanning the United States and portions of Canada and Mexico based on meteorological observations and a digital elevation model (Thornton et al.,
1997). We used Daymet data to facilitate the future extension of this
work to other locations where ground-based monitors may not be available. Daily mean temperatures were calculated as the average of daily
minimum and maximum temperatures.
Urbanization-induced climate change scenarios were based on
Maricopa Association of Governments (MAG)-projected high- and lowdevelopment Sun Corridor expansion scenarios for 2050 (Georgescu
et al., 2012, 2013). The expansion scenarios considered included a maximum (both geographical urban extent and density) and minimum
(both geographical urban extent and density) expansion, hereafter
referred to as SunCorrHi and SunCorrLo, respectively. These scenarios account for the largest degree of uncertainty associated with differing Sun
Corridor urbanization paths, whose actual development trajectory
remains unknown. Finally, widespread adoption of cool (i.e., highly reflective) roofs was utilized as an adaptation approach corresponding to
the SunCorrHi expansion scenario (hereafter SunCorrAdapt). These scenarios were incorporated into and used as surface boundary conditions
for the Weather Research and Forecasting (WRF) Model (Skamarock
and Klemp, 2008; Georgescu et al., 2013). WRF simulation output was
extracted for 0500 and 1700 local standard time (LST), consistent with
observed average daily minimum and maximum temperatures
(Georgescu et al., 2011). The average of the 0500 and 1700 LST data

540

D.M. Hondula et al. / Science of the Total Environment 490 (2014) 538–544

a

b

c

Fig. 1. A map highlighting the study region, Maricopa County, Arizona, United States. The insets show projected changes in summer mean daily temperatures (°C) in Maricopa County
under (a) a high-growth scenario (SunCorrHi), (b) a low-growth scenario (SunCorrLow), and (c) an adaptation scenario (SunCorrAdapt) for year 2050.

was a proxy for daily mean temperature (typically calculated as the
average of the daily minimum and daily maximum temperatures).
2.2. Statistical modeling: historical impacts of heat
To determine the specific relationship between temperature and
mortality in Maricopa County, we employed a time series Poisson regression model that accounts for long-term time trends in the mortality
records. The model is expressed as:
EðMÞ ¼ sðtrend; k ¼ n  yÞ þ sðtemperature; k ¼ mÞ

ð1Þ

where E(M) is the expected value of daily mortality and s represents a
natural penalized smoothing spline with k knots. For the trend term
we used five knots (n) for each of the 25 years (y) and for the temperature term we used a total of five knots (m). Sensitivity analyses tested
the impact of varying n and m. The trend term is included in the model
to minimize confounding of the temperature–mortality relationship
that might result from long-term changes in the mortality rate (associated with population growth and/or demographic changes) as well as
seasonal variability in mortality believed to be related to influenza and
other factors that are not directly related to temperature (e.g., Davis
et al., 2003). The dependent variable used throughout the study is the
daily count of all-cause mortality. We tested a number of different independent variables to examine potential differences between the current
and future climate (defined in this manuscript as impacts due to urbanization alone, separate from greenhouse gas forcing) with respect to
daily temperatures and periods of sustained high temperatures (“heat
waves”) relative to normal summer conditions in Maricopa County.
Specifically, nine different independent variables were considered:
daily maximum temperature, daily mean temperature, and daily minimum temperature, as well as the three-day and five-day trailing moving

averages of each. A separate temperature–mortality relationship was
developed for each of these independent variables. Statistical modeling
was performed using the mgcv package in the R computing environment (version 2.13.2) (R Development Core Team, 2012; Wood, 2006).
Subsequently, we quantified the number of excess deaths attributable to heat during the ten-year period 2003–2012 period (henceforth,
present day). We restricted the analysis to the months of June, July, and
August for consistency with climate model output. To calculate excess
mortality, we first identified ten different seasonal baseline mortality
patterns by generating predictions from the model represented by
Eq. (1) for the ten most recent years of mortality data while holding
temperature constant at the summer mean over the same time period.
Baseline mortality (MB) is a theoretical value representing what the expected mortality count would be on every single day assuming that the
temperature on each day was equal to the summer mean—it solely
captures seasonal variability unrelated to temperature. Next, we used
ten years of summer temperature observations (2003–2012) and
applied the temperature–mortality relationships calculated previously
to the MB time series, shifting each day's expected mortality up or
down based on the modeled temperature–mortality curve. For example, if the model associates a 40 °C daily maximum temperature with
a relative risk of mortality of 1.10, a day on which the daily maximum
temperature was 40 °C had its MB multiplied by 1.10 to generate the
predicted mortality count with the temperature effect included (MT).
Heat-attributable mortality (MH) was calculated as the sum of the positive differences between MT and MB. An illustration of this process is
shown in the Supplemental Material.
2.3. Projecting future impacts
For each urbanization scenario and exposure variable (27 different
combinations) we generated an empirical distribution of temperature

D.M. Hondula et al. / Science of the Total Environment 490 (2014) 538–544

for each of the four model runs as well as the mean of the four model
runs. We then found the difference between each scenario/exposure
variable distribution and the control run distribution at each 1/10th of
one percentile of the respective distributions. Collectively, these differences capture the change in temperature expected in the region in
2050. Gosling et al. (2009) demonstrated the importance of accounting
for changes in temperature variability as well as changes in the mean
when projecting future heat-related mortality and our percentilebased method fully captures the projected temperature changes across
the entire temperature distribution.
We applied the percentile-specific temperature changes to the
2003–2012 temperature observations derived from Daymet to generate
a synthetic daily temperature time series for the future. Each observation
in the historical record was assigned to a specific percentile, and the corresponding change from the distributional analysis was applied to that
observation to generate the synthetic data. For example, the 90th percentile summer daily minimum temperature in the historical observations was 30.3 °C, and the distributional analysis for the SunCorrHi
scenario indicated that the 90th percentile daily minimum temperature
would increase by 6.9 °C by 2050; all historical observations of 30.3 °C
were replaced with values of 37.2 °C in the synthetic time series. This
calculation is illustrated in the Supplementary Material (Fig. S4).
To estimate future excess mortality, we repeated the calculation described above for estimating excess mortality for the present-day period
(2003–2012) but replaced the historical temperature observations with
the new synthetic temperature time series. Ultimately, excess heatrelated mortality for different future urbanization scenarios was compared to that found for present-day conditions.
3. Results
The historical relationship between temperature and mortality was
found to be statistically significant (p b 0.001) for daily mean, minimum, and maximum temperatures, as well as the three- and five-day
moving averages thereof. The strongest relationships were found for
single-day maximum and mean temperature. All relationships demonstrated the familiar “U-shaped curve” for the temperature–mortality relationship also found for other cities in the U.S. (e.g., Curriero et al.,
2002; Li et al., 2013). Fig. 2 shows examples for daily maximum and
minimum temperatures (all models are shown in Supplemental Material Fig. S5). For maximum temperature, there is a positive association
between temperature and mortality for temperatures above 20 °C, but

541

as temperatures approach and exceed 40 °C the slope of the relationship increases substantially. A similar shape is observed for the minimum temperature–mortality relationship with a minimum mortality
temperature near 5 °C and an upper inflection point near 25 °C. Estimates of excess mortality attributable to high temperatures during the
ten year present-day period ranged from an average of 43.1 deaths
per year (five-day average maximum temperature) to 94.5 deaths per
year (three-day average minimum temperature). The mean across all
nine temperature metrics was 63.6 deaths per year (results for all
model runs and scenarios can be found in the Supplemental Material,
Tables S2–S4).
Future temperature changes in the region varied across the three
urbanization scenarios and exposure metrics considered (Fig. 3). Daily
minimum temperatures increased substantially in all urbanization
scenarios while daily maximums showed little change except for the
SunCorrAdapt scenario in which daily maximums decreased.
Estimates of heat-related excess mortality varied considerably
across temperature metrics and scenarios examined (Fig. 4 and Supplemental Material Fig. S6 and Tables S2–S4). In terms of the absolute and
relative change in the number of heat-related deaths compared to
present day, the largest increases in heat-related mortality were associated with minimum temperature projections for the SunCorrHi and
SunCorrAdapt scenarios. All projections based on maximum temperature showed a decline in heat-related mortality in the future.
Comparing projections based on absolute changes (Fig. 4, left panel),
the largest increases were associated with minimum temperatures for
each scenario. Three-day minimum temperature projections for the
SunCorrHi and SunCorrAdapt scenarios led to the highest heat-related
mortality increases of 339 and 280 deaths per year, respectively (the
number reported represents the mean of all four model runs). Oneday and five-day minimum temperature projections for these scenarios
also led to greater increases in mortality than any minimum temperature projections associated with the SunCorrLow scenario. Mean temperature projections led to the highest mortality increases with the
SunCorrHi scenario, ranging from 117 to 151 additional deaths per
year. The SunCorrLow and SunCorrAdapt mean temperature projections
led to comparable mortality increases on the order of 50 additional
deaths per year. Each model run using all possible combinations of urbanization scenarios and temperature metrics led to a projected decline
in heat-related mortality with future urbanization. Maximum temperature projections for the SunCorrAdapt scenario led to the largest declines in heat-related mortality, the sharpest of which was a reduction
of 46 deaths per year.
Evaluating the projections based on changes relative to the current period (Fig. 4, right panel) reveals a similar pattern. Again, the largest increases were associated with minimum temperature projections for the
SunCorrHi and SunCorrAdapt scenarios. In terms of a relative increase,
the difference between the one-, three-, and five-day temperature projections was much smaller than observed for the absolute numbers. The
highest relative increase was found for the three-day average minimum
temperature projection with the SunCorrHi scenario, with 359% more annual deaths associated with heat. Twelve of the 27 projection combinations were linked to a more than 100% increase in heat-related deaths
in 2050, including all nine projections based on minimum temperature
and three projections based on mean temperature with the SunCorrHi
scenario. The relative decline in heat-related mortality using maximum
temperature projections was largest for the SunCorrAdapt scenario,
with reductions of 94–95% compared to present day.
4. Discussion

Fig. 2. The modeled relationship between daily mortality and temperature in Maricopa
County, Arizona, United States, 1983–2007. The red line shows the relationship for daily
maximum temperature and the blue line shows the relationship for minimum temperature. 95% confidence intervals are expressed with dashed lines. A penalized thin plate
smoothing spline with a maximum of five degrees of freedom was used to estimate the
temperature effect in generalized additive model that also accounted for seasonality and
long-term time trends.

The climate of the desert southwest and Maricopa County in particular is expected to change in the coming decades as a result of urbanization and greenhouse gas forcing (Georgescu et al., 2013). Although the
majority of heat-related mortality projections to date have focused on
impacts owing to greenhouse gas-induced climate change (e.g., Hayhoe

542

D.M. Hondula et al. / Science of the Total Environment 490 (2014) 538–544

Fig. 3. Probability density functions of (left panels) daily summer temperatures at 1700 local standard time (used as a proxy for daily maximum temperature), (center panels) daily summer temperatures at 0500 local standard time (used as a proxy for daily minimum temperature), and (right panels) daily mean temperatures (average of 0500 and 1700 temperatures) for
Maricopa County, Arizona, United States. The black lines show the density functions for a present-day control period. Density functions for year 2050 are shown for three different urbanization scenarios for Arizona's Sun Corridor: High growth (red lines), low growth (blue lines), and adaptation (green lines). Each thin line represents one of four separate model runs; the
thick lines show the average of all four runs.

et al., 2004; Knowlton et al., 2007; Ostro et al., 2011; Petkova et al., 2013),
recent results suggest that urban-induced climate change has similar
order-of-magnitude climate effects on the spatial scales of human significance (Georgescu et al., 2013, 2014; Argüeso et al., 2014). Accounting for
impacts due to large-scale climate change, while neglecting regional
climate impacts owing to rapidly expanding urban areas, will therefore
provide incorrect quantitative assessment of climate change consequences for heat-related mortality.
Projections of urbanization-induced heat-related mortality for
Maricopa County for the middle of the 21st century based on minimum
and mean temperatures show considerable increases relative to present
day; even the most conservative estimate using these metrics was associated with more than a 50% increase in annual heat-related deaths. The
most extreme estimates were associated with mortality increases

exceeding 300%. The general direction and magnitude of these estimates is similar to those reported for other large metropolitan areas in
the United States (e.g., Li et al., 2013; Sheridan et al., 2012).
A sharp contrast from most existing literature was found for projection based on maximum temperature. Results indicate that maximum
temperature changes under all urbanization scenarios considered are
associated with declining future mortality. Maximum temperature (or
maximum apparent temperature) is among the exposure metrics
most commonly used in heat-related mortality projection studies
(Huang et al., 2011). With a few city-specific estimates in multi-city
studies as exceptions (e.g., Sheridan et al., 2012), the bulk of the evidence to date suggests that changes in maximum temperature will be
associated with increasing heat-related mortality. We did not find this
to be the case in semiarid Maricopa County, because urbanization

Fig. 4. Projected annual change in summer (June–August) excess all-cause mortality associated with heat in Maricopa County. The values shown reflect the (left panel) absolute differences
and (right panel) relative differences between future heat-related mortality (based on the period 2045–2055) and calculated present-day heat-related mortality (based on the period
2002–2012). Red symbols represent the SunCorrHi urbanization scenario, blue symbols represent SunCorrLow, and green symbols represent SunCorrAdapt. Upward-pointing triangles
represent daily maximum temperature, squares represent daily mean temperature, and downward-pointing triangles represent daily minimum temperature. The error bars show the
range of changes across four different model runs; the symbol corresponds to the mean of four model runs.

D.M. Hondula et al. / Science of the Total Environment 490 (2014) 538–544

patterns are expected to maintain, or slightly lower, daytime maximum
urban temperatures within semi-arid regions (Imhoff et al., 2010;
Georgescu et al., 2011).
Urbanization-induced climate change depends on the biome upon
which the built environment resides (Georgescu et al., 2011). For example, mid-latitude urban areas exhibit a small positive urban heat island
(UHI) effect during daytime hours followed by a large positive UHI
effect during nighttime hours (Brazel et al., 2000; Imhoff et al., 2010),
illustrating the classic warmer urban environment relative to rural
(and consequently less developed) surrounding areas throughout the
diurnal cycle. This is in contrast with the small, but negative, UHI effect
during daytime hours followed by a significant positive UHI effect
during the nighttime period, a prominent feature of semi-arid urban
regions (Brazel et al., 2000; Georgescu et al., 2011; Imhoff et al., 2010).
Choice of exposure variable has a profound impact on estimates of
future urbanization-induced heat-related mortality in the region we examined. Differences were more pronounced between minimum, mean,
and maximum temperatures than between single day, three-day
moving average, and five-day moving average temperatures. The sign
of the change in heat-related mortality was different for maximum
temperature projections than for minimum or mean, and considerable
differences in magnitude were found among projections with the
same sign. Barnett et al. (2010) reported that choice in exposure variable for temperature–mortality modeling should be guided by practical
concerns because no one metric was found to be superior to others with
respect to fitting the data. These results led Huang et al. (2011) to suggest that, when projecting future heat-related mortality, choice in temperature–mortality model is likely a more important determinant of
variability in projections than the exposure metric used. Our findings
offer a different perspective. Like Barnett et al. (2010), we found little
difference in the statistical performance of different exposure metrics
in relating temperature to mortality. All nine models developed were
statistically significant and similarly fit the data. However, differing
projected future changes in the exposure metrics we used – minimum
temperatures generally increase by several degrees while maximum
temperatures slightly decrease – results in large differences in heatrelated mortality projections. Our analysis only tested differences between minimum, mean, and maximum dry bulb temperatures, but it
could very well be the case that future changes in dry bulb temperature
differ from those other metrics like apparent temperature or Humidex.
Thus, we recommend that even if different exposure variables similarly
fit mortality data, their impact on future mortality outcomes also be
tested because the manner in which the exposure variables themselves
evolve in the future may differ from one to another. In arid locations,
home to over two billion people (Confalonieri et al., 2007), this is especially important.
The projections provided herein do not account for demographic
changes, greenhouse-gas forced climate change, or population acclimatization to high temperatures. Population growth is accounted for in the
urban expansion scenarios (determining future temperature changes)
but mortality increases (both in absolute and relative terms) are provided for the current population size to provide a meaningful comparison
with respect to risk. These variables are held constant so that we can
specifically examine variability arising from urbanization and exposure
variable choice. Population growth, demographic changes, greenhousegas forced climate change, and population acclimatization are all important elements of a comprehensive projection of heat-related mortality.
This work highlights the need to include built environment expansion
and its impacts on regional climate in such work.
5. Conclusions
Our results bring to light several dimensions of the challenge of
projecting heat-related mortality that have been largely neglected to
date. Heat-related mortality in the United States is widely projected to increase in the future as the climate warms, urban areas continue to grow,

543

and the population ages. Research to date has not comprehensively examined the role that urbanization-induced climate change and choice in
exposure variable play in these projections. Integrating historical weather
and mortality data with climate projections for Maricopa County, Arizona,
we have shown that both of these factors introduce large uncertainties.
Substantial differences in heat-related mortality projections were evident
across three different urbanization scenarios and choice of minimum,
mean, or maximum temperature. All projections based on minimum
and mean temperatures were associated with increasing heat-related
mortality, with the largest projected increase exceeding 350% above
present day. Conversely, all projections based on maximum temperatures
were associated with declining heat-related mortality, and under an
adaptation-oriented scenario these declines reduced mortality by more
than 94%. The scale of these differences shows that critical evidence for
policymakers is absent when projecting the health effects of climate
change without considering the important role of urbanization and exposure variable choice.
Acknowledgments
Thanks to the Office of Vital Statistics at the Arizona Department of
Health Services for providing the mortality data used in this study.
David Hondula was supported by the Virginia G. Piper Charitable Trust
Health Policy Informatics Initiative at Arizona State and Matei
Georgescu was supported by the United States National Science Foundation under grant EAR-1204774.
Appendix A. Supplementary data
Supplementary data to this article can be found online at http://dx.
doi.org/10.1016/j.scitotenv.2014.04.130.
References
Anderson BG, Bell ML. Weather-related mortality: how heat, cold, and heat waves affect
mortality in the United States. Epidemiology 2009;20(2):205.
Anderson GB, Bell ML. Heat waves in the United States: mortality risk during heat waves
and effect modification by heat wave characteristics in 43 US communities. Environ
Health Perspect 2011;119:210–8.
Argüeso D, Evans JP, Fita L, Bormann KJ. Temperature response to future urbanization and
climate change. Clim Dyn 2014;42(7–8):2183–99.
Barnett AG, Tong S, Clements ACA. What measure of temperature is the best predictor of
mortality? Environ Res 2010;110:604–11.
Brazel AJ, Selover N, Vose R, Geisler G. The tale of two climates—Baltimore and Phoenix
urban LTER sites. Clim Res 2000;15:123–35.
Chow WT, Chuang WC, Gober P. Vulnerability to extreme heat in metropolitan Phoenix:
spatial, temporal, and demographic dimensions. Prof Geogr 2012;64(2):286–302.
Chuang WC, Gober P, Chow WT, Golden J. Sensitivity to heat: a comparative study of
Phoenix, Arizona and Chicago, Illinois (2003–2006). Urban Clim 2013;5:1–18.
Confalonieri U, Menne B, Akhtar R, Ebi KL, Hauengue M, Kovats RS, et al. Human health.
In: Canziani OF, Palutikof JP, van der Linded PJ, Hanson CE, Parry ML, editors. Climate
Change 2007: impacts, adaptation and vulnerability. Contribution of Working Group
II to the fourth assessment report of the Intergovernmental Panel on Climate Change.
Cambridge: Cambridge University Press; 2007. p. 391–431.
Curriero FC, Heiner KS, Samet JM, Zeger SL, Strug L, Patz JA. Temperature and mortality in
11 cities of the eastern United States. Am J Epidemiol 2002;155:80–7.
Davis RE, Knappenberger PC, Michaels PJ, Novicoff WM. Changing heat-related mortality
in the United States. Environ Health Perspect 2003;111:1712–8.
Ebi KL, Kovats RS, Menne B. An approach for assessing human health vulnerability and
public health interventions to adapt to climate change. Environ Health Perspect
2006;114(12):1930.
Georgescu M, Moustaoui M, Mahalov A, Dudhia J. An alternative explanation of the
semiarid urban area “oasis effect”. J Geophys Res 2011;116:D24113.
Georgescu M, Mahalov A, Moustaoui M. Seasonal hydroclimatic impacts of Sun Corridor
expansion. Environ Res Lett 2012;7(3):034026.
Georgescu M, Moustaoui M, Mahalov A, Dudhia J. Summer-time climate impacts of
projected megapolitan expansion in Arizona. Nat Clim Change 2013;3(1):37–41.
Georgescu M, Morefield PE, Bierwagen BG, Weaver CP. Urban adaptation can roll back
warming of emerging metropolitan regions. Proc Natl Acad Sci 2014;111(8):
2909–14.
Golden JS, Hartz D, Brazel A, Luber G, Phelan P. A biometeorology study of climate and
heat-related morbidity in Phoenix from 2001 to 2006. Int J Biometeorol 2008;
52(6):471–80.
Gosling SN, McGregor GR, Lowe JA. Climate change and heat-related mortality in six cities
Part 2: climate model evaluation and projected impacts from changes in the mean

544

D.M. Hondula et al. / Science of the Total Environment 490 (2014) 538–544

and variability of temperature with climate change. Int J Biometeorol 2009;53(1):
31–51.
Greene S, Kalkstein LS, Mills DM, Samenow J. An examination of climate change on
extreme heat events and climate–mortality relationships in large US cities. Weather
Clim Soc 2011;3(4):281–92.
Harlan SL, Brazel AJ, Prashad L, Stefanov WL, Larsen L. Neighborhood microclimates and
vulnerability to heat stress. Soc Sci Med 2006;63(11):2847–63.
Harlan SL, Declet-Barreto JH, Stefanov WL, Petitti DB. Neighborhood effects on heat
deaths: social and environmental predictors of vulnerability in Maricopa County,
Arizona. Environ Health Perspect 2013;121(2):197.
Hayhoe K, Cayan D, Field CB, Frumhoff PC, Maurer EP, Millner NL, et al. Emissions pathways, climate change, and impacts on California. Proc Natl Acad Sci U S A 2004;
101(34):12422–7.
Hondula DM, Davis RE, Leisten MJ, Saha MV, Veazey LM, Wegner CR. Fine-scale spatial
variability of heat-related mortality in Philadelphia county, USA, from 1983–2008:
a case-series analysis. Environ Health 2012;11:1–11.
Hondula DM, Vanos JK, Gosling SN. The SSC: a decade of climate–health research and
future directions. Int J Biometeorol 2013. http://dx.doi.org/10.1007/s00484-0120619-6. [Online 29 January 2013].
Huang C, Barnett AG, Wang X, Vaneckova P, FitzGerald G, Tong S. Projecting future heatrelated mortality under climate change scenarios: a systematic review. Environ
Health Perspect 2011;119:1681–90.
Imhoff ML, Zhang P, Wolfe RE, Bounoua L. Remote sensing of the urban heat island effect
across biomes in the continental USA. Remote Sens Environ 2010;114:504–13.
Jenerette GD, Harlan SL, Stefanov WL, Martin CA. Ecosystem services and urban heat
riskscape moderation: water, green spaces, and social inequality in Phoenix, USA.
Ecol Appl 2011;21(7):2637–51.
Kalkstein LS, Greene JS. An evaluation of climate/mortality relationships in large US
cities and the possible impacts of a climate change. Environ Health Perspect 1997;
105(1):84.
Knowlton K, Lynn B, Goldberg RA, Rosenzweig C, Hogrefe C, Rosenthal JK, et al. Projecting
heat-related mortality impacts under a changing climate in the New York City region.
Am J Public Health 2007;97(11):2028–34.
Kovats RS, Hajat S. Heat stress and public health: a critical review. Annu Rev Public Health
2008;29:41–55.
Li T, Horton RM, Kinney PL. Projections of seasonal patterns in temperature-related
deaths for Manhattan, New York. Nat Clim Change 2013. http://dx.doi.org/10.1037/
nclimate1902. [Online 19 May 2013].
Luber G, McGeehin M. Climate change and extreme heat events. Am J Prev Med 2008;
35(5):429–35.
Luber GE, Sanchez C, Conklin L. Heat-related deaths: United States, 1999–2003. Morb
Mortal Wkly Rep 2006;55(29):796–8.

MCDPH (Maricopa County Department of Public Health). Maricopa County Department
of Public Health health statistics and reports: heat reports. Available http://www.
maricopa.gov/publichealth/Services/EPI/Reports/heat.aspx, 2013. [accessed 10
October 2013].
McMichael AJ, Woodruff RE, Hales S. Climate change and human health: present and
future risks. Lancet 2006;367(9513):859–69.
O'Neill MS, Carter R, Kish JK, Gronlund CJ, White-Newsome JL, Manarolla X, et al.
Preventing heat-related morbidity and mortality: new approaches in a changing
climate. Maturitas 2009;64(2):98–103.
Ostro B, Rauch S, Green S. Quantifying the health impacts of future changes in temperature in California. Environ Res 2011;111(8):1258–64.
Petitti DB, Harlan SL, Chowell-Puente G, Ruddell D. Occupation and environmental heatassociated deaths in Maricopa County, Arizona: a case-control study. PLoS One 2013;
8(5):e62596.
Petkova EP, Horton RM, Bader DA, Kinney PL. Projected heat-related mortality in the US
urban northeast. Int J Environ Res Public Health 2013;10(12):6734–47.
R Development Core Team. R: a language and environment for statistical computing.
Vienna: R Foundation for Statistical Computing; 2012 [www.R-project.org].
Ruddell DM, Harlan SL, Grossman-Clarke S, Byuntajev A. Risk and exposure to
extreme heat in microclimates of Phoenix, AZ. In: Lu Y, Showalter PS, editors.
Geospatial techniques in urban hazard and disaster analysis. New York: Springer;
2009. p. 179–202.
Saha MV, Davis RE, Hondula DM. Mortality Displacement as a Function of Heat Event
Strength in 7 US Cities. American journal of epidemiology 2014;179(4):467–74.
Sheridan SC, Kalkstein LS. Progress in heat watch-warning system technology. Bull Am
Meteorol Soc 2004;85:1931–42.
Sheridan SC, Kalkstein AJ. Seasonal variability in heat-related mortality across the United
States. Nat Hazards 2010;55(2):291–305.
Sheridan SC, Allen MJ, Lee CC, Kalkstein LS. Future heat vulnerability in California, Part II:
projecting future heat-related mortality. Clim Change 2012;115(2):311–26.
Skamarock WC, Klemp BJ. A time-split nonhydrostatic atmospheric model for weather
research and forecasting applications. J Comput Phys 2008;227:3465–85.
Thornton PE, Running SW, White MA. Generating surfaces of daily meteorological
variables over large regions of complex terrain. J Hydrol 1997;190(3):214–51.
Thornton PE, Thornton MM, Mayer BW, Wilhelmi N, Wei Y, Cook RB. DAYMET: daily
surface weather on a 1 km grid for North America; 2012 [1980–2008. Acquired online
[http://daymet.ornl.gov/] on 1 August 2013 from Oak Ridge National Laboratory
Distributed Active Archive Center, Oak Ridge, TN, USA].
Wood SN. Generalized additive models: an introduction with R. CRC Press; 2006.
Yip FY, Flanders WD, Wolkin A, Engelthaler D, Humble W, Neri A, et al. The impact of
excess heat events in Maricopa County, Arizona: 2000–2005. Int J Biometeorol
2008;52(8):765–72.