Child pages
  • Carver - Final Project

Assessing Vulnerability to Heat Stress Among Boston Seniors


Project Description
Extreme weather events, including heat waves, are predicted to increase in frequency and severity due to climate change. Prolonged exposure to extreme heat increases the risk of heat stress, which can result in a variety of heat-related illnesses and death. The epidemiology literature identifies several socioeconomic, demographic, and environmental factors associated with increased sensitivity to heat (age over 65, racial/ethnic minority, poverty, disability, social and linguistic isolation) and increased exposure to heat (prevalence of heat island contributors such as impervious surfaces and lack of tree canopy). Populations at increased risk of heat stress include the elderly, young children, and persons with chronic disease.

Because heat stress prevention strategies can differ for vulnerable sub-populations, it is important for urban planners and public health professionals to identify and tailor planning responses to the unique needs of communities most at risk. A heat stress-prevention planning initiative targeting seniors, for example, would differ from a strategy focused on families with young children. To this end, I used ArcGIS to create a composite index of heat stress vulnerability among the elderly in Boston. The project goals were to 1) spatially identify those Boston block groups in which seniors are at highest risk of heat-stress-related morbidity and mortality, and 2) visually highlight the most heat-vulnerable census block groups that are beyond walking distance to an official Boston cooling center. While this analysis is preliminary, it can identify gaps in services and suggest areas of Boston that are most in need of additional heat-stress prevention efforts targeted to seniors.

The theoretical framework for this vulnerability assessment is a simple conceptual model proposed by Rinner et al. (2010), which views vulnerability to heat stress as a function of population sensitivity, exposure, and adaptive capacity. I used the census block group as the unit of analysis because this was the smallest unit of census geography for which data was available for several of my variables of interest (poverty status, disability, household status, and English-language skills). To examine heat stress sensitivity variables, I used data from the 2000 US Census to create a series of maps of heat-stress sensitivity indicators cross-tabulated with age 65 and up. I also mapped heat-stress exposure indicators associated with the urban heat island effect -- lack of tree canopy cover (from the National Land Cover Dataset) and prevalence of impervious surface (from MassGIS) -- then aggregated the raster data up to the census block group level. For each of the sensitivity/exposure maps, I classified the results into quintiles and assigned each quintile a score of one to five for increasing heat stress vulnerability. I added up all the heat sensitivity/exposure scores for each block group to create a cumulative vulnerability score, then mapped the results by quartiles to spatially identify differential cumulative heat- stress vulnerability across block groups. Finally, I geocoded the addresses of official Boston cooling centers for 2009 and mapped them with a ¼-mile buffer to visually highlight the most heat-vulnerable census block groups for Boston residents age 65+ that are beyond walking distance to a cooling center.


Data Sources

Variable of interest

Source/ Metadata/Source Scale/Date

Mass. census block groups


Source Scale:  1:100,000

Mass. towns (base map)


Source Scale:  1:100,000

Mass. roads (base map)


Source Scale:  1:5,000
Up to date through December 2007

Mass. water bodies (base map)


Source Scale:  1:25,000
Relevant for October 2004

Population age ≥ 65

Field descriptions:
(Same for next three items)

Households age ≥ 65, living alone

\MassGIS\Census_2000\ cen2K_bg_hh_age_fam_child.dbf

Population age ≥ 65,
income < 1999 federal poverty

\MassGIS\Census_2000\ income_poverty_levels_by_age.dbf


Population with income
< 1999 federal poverty level

\MassGIS\Census_2000\ income_poverty_levels_by_age.dbf

Population age ≥ 65, race/ethnicity other than white

Source   :
SF3 - Table:     P145I     Sex by Age (White Alone, Not Hispanic of Latino)

Population age ≥ 65, with English-language difficulties

Source   :
SF3 – Table P19:     Age by Language Spoken at Home by Ability to Speak English for the population 5 years and over

Population age ≥ 65
with any disability

Source   :
SF3 – Table P42     Sex by Age by Disability Status by Employment Status for the Civilian Noninstitutionalized Population
5 Years and Over

Mean tree canopy cover

National Land Cover Data Set: CanopyCover\Zone13\

Resolution: 30 meter cell
Data represents 07-05-1999 through 04-05-2001

Impervious surface

\MassGIS\Impervious surface raster layer

Resolution:  1 meter cell
Data acquired by MassGIS April 2005

City of Boston cooling centers
for 2009

List of addresses published by the Boston Centers for Youth and Families for summer 2009


Data Processing Steps

City of Boston cooling centers:   Data was geocoded by address and mapped.

Census data:

  1. I used the Join/Relates tool to create a Table Join between the individual tabular data sets and the census block group polygons GIS layer using the LOGRECNO field. For data downloaded from American Factfinder, I joined the GEO_ID2 field in the census data tables with the BG_ID field in the census block groups shapefile.
  2. I created new fields for the heat sensitivity variables of interest, then used the Field Calculator to populate the new field. For example, for the data table showing poverty status by age group by block group, I used the Field Calculator tool to sum all relevant fields for age > 65 in poverty, divided by the total sample population, then multiplied the quotient by 100 to create a new field for percentage of population ≥ 65 living below the poverty level.
  3. For each variable, I mapped the newly created field, classified by quintiles, then created a new field for quintile scores and used the Select by Attribute and Field Calculator tools to populate the new field with quintile scores of 1-5 (for ascending vulnerability).

Canopy cover and impervious surfaces raster data:

  1. I downloaded and mosaicked the impervious surface files that included Boston.
  2. For both raster data layers, I created a smaller dataset by using the Data Export tool to clip the resulting file down to slightly larger than the Boston town boundary, setting the extent and the spatial reference to the chosen data frame to get my data into a common coordinate system (in my case, the Mass. State Plane Mainland).
  3. I used the Zonal Statistics as Table tool in Spatial Analyst to aggregate the raster data up to the block group level and calculate the mean canopy cover/impervious surface cover within each block group.
  4. I mapped the newly created fields by quintiles, then created a new field for quintile scores and use the Select by Attribute and Field Calculator tools to populate the new field with quintile scores of 1-5 for each heat exposure variable, with a higher score indicating higher vulnerability.

Cumulative Heat-Score Index: I added up all the heat sensitivity/exposure scores for each block group and created a new field for cumulative vulnerability score, then mapped the resulting field by quintiles, as described above, to spatially identify the highest vulnerability block groups in Boston. I used the Proximity: Buffer tool in ArcToolbox to create a ¼-mile (walkable distance) buffer around the Boston Cooling Centers.

Additional analysis steps: In addition to visually analyzing the proximity of high-vulnerability census blocks to the cooling center buffers, I also mapped the population age 65+ with graduated symbols to visually compare the absolute population of seniors in each block group to the percentage of seniors to total block group population. I also mapped the longterm-care data from MassGIS to get a sense of how residents of longterm-care facilities such as nursing homes contributed to the overall population of seniors in each block group. This would be important information for planners crafting a strategy for heat-stress prevention among seniors, since community-dwelling seniors would not have access to services/resources available to seniors in longterm care facilities.


  • I had planned to include a data layer from the Boston Assessing Dept. to show air conditioner ownership by parcel as an indicator of heat exposure, but after 1) mapping this data layer for all of Boston and seeing the prevalence of areas with no air conditioning data, 2) considering that this data layer likely shows only central air conditioning (not window units, which would be as useful as central AC for my analysis), and 3) comparing this data to the figures for air conditioner ownership in the American Housing Survey (for the Boston Metropolitan Area) and the US Energy Information Administration’s Residential Energy Consumption Survey (for the Northeast), I decided that this data was not complete enough to be useful for my analysis, so I chose not to include it.
  • One of the challenges I encountered in working with the census data was the issue of choosing the correct universe when working with cross-tabulated data. At times, I found it difficult to determine which universe would give me the most accurate results. For example, when calculating percentage of seniors for a particular cross-tabbed variable within a census block group, I found that choosing sample population rather than total population as the denominator for my calculation gave me very different results. At other times, there wasn’t single right way to do the calculation, but the choice of denominator really influenced the analysis. So I had to think through carefully what information I was trying to capture.

Similar Analyses with Useful Methods

  1. Vescovi, Luc et al. Assessing Public Health Risk Due to Extremely High Temperature Events: Climate and Social Parameters. Climate Research 30, 71-78, 2005.

This study analyzes two separate risk factors for heat-wave-related morbidity and mortality – climate hazard and social vulnerability – then synthesizes the results of the analysis using GIS mapping. To measure climate hazard, the authors used data from Environment Canada stations in southern Quebec to produce two indices: mean number of days with maximum temperature > 30 C and mean number of annual episodes of at least 3 consecutive days of maximum temperature > 30 C and minimum temperature > 22 C. Climate modeling was then conducted to create climate projections for 2039-2060, and current and projected climate hazard maps were created using GIS. For social vulnerability, the authors used Canadian census data to create four sub-indices of social vulnerability to high-temperature events: Age (frequency of age > 65), Poverty (frequency of low-income earners), Social isolation (frequency of single-person households), and Education (frequency of people older than 20 with less than 13 years of education). The sub-indices were then summed to create a single social vulnerability index, which was also mapped. Finally, the authors overlaid their social vulnerability and climate hazard maps to produce two “risk maps” that identify the geographic areas of southern Quebec where public health risk is highest for current and projected high temperature events.

  1. Reid, Colleen E. et al. Mapping Community Determinants of Heat Vulnerability. Environmental Health Perspectives 117:11, 1730-1736, 2009.

This study mapped and analyzed 10 variables shown in the epidemiology literature to increase vulnerability to heat-related morbidity and mortality in the US. Using data from US Census 2000, the study examined six demographic/socioeconomic variables related to age, poverty, education, and race/ethnicity. Other variables examined included home air conditioner ownership, land cover, and diabetes prevalence. The study unit of analysis was the US census tract. The authors conducted a factor analysis of these 10 variables to obtain four factors that explained > 75% of the variance in the 10 original variables examined: social/environmental vulnerability (combined education, poverty, race, green space), social isolation, AC prevalence, and proportion elderly/diabetes.  As the authors note in their background section, the published literature on mapping heat vulnerability is scant and includes analyses conducted at different spatial scales and with different variables. This study attempted to expand heat vulnerability mapping to a national scale, and the authors suggest that their methodology could serve as a template for future regional and local heat vulnerability maps, which may include additional variables for which regional- or local-level data is available.

  1. Rinner, Claus et al. The Role of Maps in Neighborhood-Level Heat Vulnerability Assessment for the City of Toronto. Cartography and Geographic Information Science 37:1, 31-44(14), 2010.

This study uses statistical methods and GIS mapping to create a spatially explicit preliminary assessment of population vulnerability to heat-related illness in Toronto, Canada. To characterize heat vulnerability, the authors create a simple conceptual framework that considered indicators of risk related to exposure (including outdoor temperature, lack of tree canopy/green space, old or high-density dwellings without air conditioning), sensitivity (including age, low income, chronic illness, social isolation, recent immigration/non-English speaking, race), and adaptive capacity (including access to cooling centers, common cooling space, malls/libraries). Using ArcGIS and interactive mapping tools, the authors created several types of maps: maps of individual risk indicators, index maps “in which multiple indicators were weighted and combined into a composite measure of heat vulnerability” (p35), and cluster maps that highlighted geographic concentrations of vulnerable populations. The authors then compared the various types of maps to assess geographic aggregation and scale effects, the impact of variable selection, and the impact of cluster analysis input parameters.


  1. Clark, George E. et al. Assessing the Vulnerability of Coastal Communities to Extreme Storms: the Case of Revere, MA, USA.   Mitigation and Adaptation Strategies for Global Change 3: 59-82, 1998.

For this vulnerability analysis, the author created a series of GIS maps based on a conceptual framework of vulnerability as a function of exposure to hazard and coping ability. The author used a floodplain map as a simplified measure of exposure, and examined 34 variables from the 1990 Census to measure coping ability. The study unit of analysis was the census block group, which is the smallest unit of census data for which relatively complete socioeconomic data is available. Employing factor analysis, the author reduced the 34 census variable to the 5 factors that accounted for the greatest variance in the data. In descending order, the factors considered were poverty (included variables related to low income, minority race/ethnicity, low educational attainment, and lack of access to cars); transience (relatively new to the Revere area, low mortgage value); disability (immobility, low-self care, work disability); immigrant status; and young family status (many children < 5, few elders, working parents). The author first mapped each factor by census group for the Revere area, indicating factor load by color-coded sextile, then used two different statistical methods (averaging and data envelopment analysis) to combine the factors into a single composite scaled index of coping ability. Finally, the author overlaid the floodplain (physical exposure) map with the overall coping ability map to show spatial distribution of vulnerability to extreme storms based on physical and socioeconomic vulnerability. Interestingly, the author notes that overlaying floodplain maps with maps of individual factors such as disability or poverty can highlight the causes of differential social vulnerability, thus informing choice of mitigation strategy.

I think this approach constitutes a reasonable first pass at a heat-stress vulnerability analysis for Boston seniors, but I think future analyses would benefit from more up-to-date census data, as well as accurate data on air conditioner ownership and usage. Also, I think it would be very informative to re-do this analysis adding satellite-derived neighborhood climate data to better get at differential heat exposure issues. Future analyses could also examine additional contributors to adaptive capacity beyond official city cooling centers, such as other public air-conditioned spaces accessible to seniors. Additionally, I think future versions of this analysis could benefit from weighting sensitivity/exposure/adaptive capacity indicators in the final cumulative index based on their relative contribution to heat-stress vulnerability.