The proper definition of environmental conditions that cause heat-related morbidity and mortality is an unsettled question in human biometeorological research. Various studies have employed different threshold variables, such as maximum temperature, afternoon AT, and morning dew point temperature [7]. Here, we attempt to address this shortcoming by examining AT diurnality for Philadelphia County. Although there are significant mortality elevations when ATs exceed the threshold at any hour, we found three periods with exceedances greater than 10%. Mortality rates were highest when thresholds were exceeded in the morning or evening hours on the day immediately prior to death and in the afternoon two days prior. Examining effects by hour, rather than using more conventional metrics like daily maximum, minimum, or mean temperature allows for a more specific identification of hazardous periods. These patterns may arise in part from the threshold chosen for each hour: the threshold temperature might be expected to follow an even smoother pattern than that shown in Figure 3a. In particular, the threshold temperature seems to increase rapidly on lag 1 between roughly 6:00 and 10:00 a.m., and thus the relatively high values here might be leading to the spike in the response at the same time in Figure 3b. The other peaks in the response curve (Figure 3b) seem less likely to be influenced by variations in the threshold curve. Future work might examine the mortality response above various percentiles of hourly temperature rather than a mortality-based threshold. The lack of an especially high response on the day of death (Figure 3b) may arise from the absence of time for exposure and resultant physiological stress (i.e., the response is not immediate.)
Mortality following days with high ATs in Philadelphia County is not randomly spatially distributed but is concentrated in several distinct regions. Certain ZCTAs exhibit mortality that is more than 30% above the daily citywide average for particular AT-time combinations. Intra-County variability in heat-related mortality has been observed or suggested elsewhere [10, 13–18], but the majority of studies to date have focused on a larger spatial scale, single heat events, hot summers, or did not consider the actual mortality response. This study is among the first to quantify local-scale mortality responses over a multi-decadal period.
Several of the variables associated with higher local-scale mortality are consistent with observations and hypotheses in the literature, including high-density housing, low socioeconomic status, high surface temperatures, and elderly populations [13, 16]. The spatial distribution of heat-related mortality in Philadelphia County during the 1993 heat wave was previously examined and the same variables were associated with elevated risk [10, 17]. The lack of a strong relationship with recreational zoning is surprising because we expected places with more parks and green space to have lower surface temperatures, thereby reducing heat and heat-related mortality. Recreational zoning is highest in two ZCTAs along the Schuylkill River in the western portion of the County, one of which also has a high percentage of high density residential zoning. However, the two zoning types are not interspersed, and where green space is not intermingled amongst residential areas, the mitigating effect on temperature in dense residential areas may be diminished. Although a large body of research points to the advantages of adding green space to lower temperatures in the urban environment, we are not able to conclude that ZCTAs with more parkland are associated with lower mortality rates. This does not indicate that green space is not beneficial, but rather that many other variables may confound the signal, especially at the scale of this analysis. We are continuing to investigate the relationship between zoning types, air and surface temperatures, and mortality outcomes.
This study also incorporates the relatively recent approach of including remotely-sensed measurements of surface temperature in the study of heat-related mortality. Individuals living in areas with higher surface temperatures are at greater risk following hot days. This finding is consistent with the expectation that individuals living in hotter places are under greater physiological stress [10]. We are encouraged that the results from a remote sensing approach are similar to those using other sensors or models of the UHI.
We did not directly identify race as a key factor in the spatial distribution of heat-related deaths. Principal component loadings for the racial variables were only high in one significant component (PC 1), but loadings for other variables (income, surface temperature, educational attainment, and density of development) were higher. As previously documented for Phoenix [5], minority populations in Philadelphia County live in areas that are associated with higher surface and air temperatures. We directly observed the relationship with surface temperature and can infer the relationship with air temperature because of the high density of residential development in these locales. Racial variable loadings are very small for the other two components included in the model (PCs 5 and 6). Thus, we cannot conclude that race alone is a key factor in the spatial distribution of heat-related deaths in Philadelphia.
There are a few limitations we faced in creating our model for Philadelphia that merit discussion. The sociodemographic and zoning variables were derived from data available at a fixed point in time (e.g., the year 2000 census). However, the underlying demographics and zoning ordinances both change over time, a process we were unable to capture using this approach. This introduces some uncertainty into the results, and future research should explore local-scale mortality patterns over both space and time.
We were especially interested in exploring the relationship between the complex temperature patterns present in the metropolitan area and heat-related mortality. Satellite imagery has become much more accessible and makes this type of analysis possible using surface temperature measurements. The surface temperature field may be much different from the air temperature field over the same area, and we are not claiming that the two are identical, although some research indicates a high degree of spatial correlation between the two fields during daylight hours [24]. There are many aspects of the urban heat island worthy of consideration in the context of urban health, including day/night variability and the contrast between the surface heat island and that of the canopy layer. We are investigating if residents in places with higher morning surface temperatures on hot days are at greater risk.
We are currently implementing a cloud-masking scheme that will increase the number of available images and extending our sample prior to 2004. We believe that the use of remote imagery in our study, and others, could be greatly enhanced if more surface temperature images were used. In just the two used in our study, there is variability in the surface temperature pattern that may be linked to seasonal differences, synoptic-scale conditions, or other environmental controls. An additional concern with the satellite imagery is that many of the image pixels are measuring rooftop temperature, which may not be representative of the surface conditions experienced where individuals might be living, working, or spending time outdoors.
We observed a high correlation between the surface temperature field and several socioeconomic variables, as evidenced by the high loadings on the first principal component. Although principal components allows for the examination of potential effects of a large suite of variables believed to be associated with risk, one tradeoff can be difficulty in interpreting the results. We can definitively say that places with higher surface temperatures are associated with higher mortality risk, but those places also have a high percentage of residents living in poverty and a high percentage of residents without a high school diploma. This pattern has been observed for other cities in the United States [5] and makes it difficult to pinpoint a causal relationship between the individual predictor variables and the health response. Even if it is difficult to separate the effects of individual variables, identifying characteristics of places associated with higher heat-related mortality can lead to improvements in the allocation of medical resources during dangerous conditions. Our future analysis in other cities in the United States where socioeconomic status and surface temperatures may not be as highly correlated may shed light on the relative impact of exposure, education, and income on heat-related risk.
The role of air quality in leading to increased mortality during heat waves is a topic of continued debate in the literature and is beyond the scope of this study. As heat waves are commonly associated with clear skies and stagnant air, conditions are ideal for the rapid buildup and accumulation of various pollutants. It is likely that during heat waves a portion of the excess deaths are attributable to the thermal stress whereas others might be linked to high concentrations of unhealthy atmospheric constituents. We did not incorporate air quality data into this study but encourage future study of the interactive effects of heat and air quality on summertime mortality as well as the potential for differential mortality over space as a result of local-scale air quality variability. Both are topics of active ongoing investigation by the authors and many others. We also note that we were unable to locate air conditioning use data at an appropriate resolution for this study. Air conditioning has become widely adopted in the United States and increases in availability have been linked to decreases in heat-related mortality [2]. However, we believe that air conditioning availability and usage is likely highly correlated with measures of socioeconomic status, and thus may be implicitly included in our analysis. Access and willingness to use medical care is a potential confounder at the individual level that we were not able to represent at the scale of this study, although it may be highly correlated with the socioeconomic variables included. Finally, the use of AT may not identify all of the critical physiological factors in evaluating the heat-mortality relationship and we intend to adopt this approach with other physiological indices in the future.
Intra-city variability in the response to high heat and humidity conditions indicates an opportunity for the improvement of heat-health watch-warning systems (HHWWS) that have been deployed in cities across the globe. When a dangerous event is forecast, for example, emergency managers might reprioritize allocation of medical resources to those geographic areas responsible for the largest portion of the heat-related deaths in the past. A more thorough effort to build and validate a predictive model of both the timing and placement of heat-related deaths is recommended prior to operational changes in any HHWWS. Longer-term strategies to reduce the heat stress and health burden in these localities might be considered as well, such as the implementation of building weatherization programs, adding green space to the city landscape, adoption of low-albedo and/or green building practices, and location of future healthcare facilities.