Research Design: GIS Mapping of Urban Heat Vulnerability in Seattle, Washington

Data and Methods

            This research is based in Seattle, Washington, a city located in the northwest U.S.A.  It is the largest city in the northwest, making it a pivotal one for analyzing heat vulnerability.  This is especially true as climate change continues to cause more extreme heat waves, which become exacerbated by the urban heat island (UHI) effect (Wolf and McGregor 2013, 59; Jalalzadeh et al 2021, 3).  Additionally, as more people move to urban areas like Seattle, risk analysis of heat waves will become more important as climate change worsens (Jalalzadeh et al 2021, 1).  The data and analysis for this study involves a variety of steps, including remote sensing to gather physical data about the city surface; secondary source research to gather socioeconomic data about the city’s neighborhoods; and Geographic Information Systems (GIS) analysis to map the results.  To map heat vulnerability in GIS, the physical and social data will be indexed to make representation easier.

The first component for our indexing is physical data.  Essential to measuring the UHI is remote sensing, which involves collecting data about the earth’s surface without being in direct contact with it (Gomez & Jones 2010, 156).  This is typically done by collecting observations from satellite or airborne instruments.  Sidiqui et al (2022, 265) state that UHI “can cause hotspots in city areas due to dense, impervious infrastructure and minimal vegetation cover”.  Built-up surfaces on land- such as concrete, asphalt, and steel- absorb more heat (Jalalzadeh et al 2021, 2).  Thus, Landsat data can reveal areas of Seattle that are higher in it by classifying each pixel according to land type (Clifford et al 2016, 428).  It can also reveal higher levels of thermal radiation in certain neighborhoods (Keeratikasikorn & Bonafoni S 2018, 5; Méndez-Lázaro et al 2018, 709; Sidiqui et al 2022, 273), especially when compared with higher levels of vegetation surrounding the city.  High resolution satellite imagers like Landsat are often used by government agencies and television broadcasters for planning responses to rapidly changing situations like natural disasters (Clifford et al 2016, 428).  As Gomez and Jones (2010, 386) point out, probability maps are highly useful in the study of risk because they provide a spatial reference for high-risk locations, making them easier for planners and officials to maintain.

Landsat data can be taken from the Earth Resources Observation and Science Centre (Sidiqui et al, 273), which is publicly available on USGS.gov.  Most studies that provide thermal mapping in urban areas use the Landsat Thermal and Infrared Sensor (TIRS), with a 100m spatial resolution (Sidiqui et al, 273; Keeratikasikorn & Bonafoni 2018, 2; Méndez-Lázaro et al 2018, 712).  For band selection, Keeratikasikorn & Bonafoni (2018, 5) state that “TIRS Band 10 (10.60 μm–11.19 μm) has the highest accuracy with respect to other methods”.  Images with 5-25% cloud cover or less are usually chosen (Sidiqui et al, 273; Méndez-Lázaro et al 2018, 712).  The range of dates to be covered are decided by extreme heat events, retrieved from local records and newspaper articles (Sidiqui et al, 273).  For Seattle, the notorious Heat Dome that lasted from June 26-July 2, 2021, is an ideal time frame for mapping urban heat vulnerability.  Images from Landsat TIRS can be downloaded from this time frame, provided they have the appropriate percentage of cloud cover.

Once the images are retrieved, land surface temperature can be calculated by converting digital values at the top of the atmosphere to radiance using gain and bias values (Méndez-Lázaro et al 2018, 712).  Gain and bias values measure the gradient of the calibration and the spectral radiance of the sensor, respectively (NOAA, n.d.).  Once this is adjusted, the radiance can then be converted to temperature (Méndez-Lázaro et al 2018, 712).  When doing this method, atmospheric correction needs to be made by using a coefficient parameter, provided by NASA (ibid. 712).  This step is not required for all thermal mapping; however, it may be convenient to use the more familiar units of temperature instead of digital values.

The second component for indexing is social data, which can be extracted from the U.S. Census.  Sisqui et al (2022, 273) took their data from meshblocks in the Australian Bureau of Statistics since the study was based on a city there called Geelong.  Meshblock units in Australia are analogous to census blocks in the U.S.  Wolf and McGregor (2013, 59) extracted their data from the 2001 London Census, which had 4,765 census districts.  Keeratikasikorn & Bonafoni (2018, 13) did not use social data on the city of Bangkok; they only used urban planning zones to delineate the boundaries of their remote sensing data.  Méndez-Lázaro et al (2018, 709) used the most relevant data collection method: U.S. Census tracts from San Juan, Puerto Rico.  Thus, it will be the most closely followed one for this study.  Seattle has as many as 131 census tracts (Seattle.gov) that contain socioeconomic data, which will provide the boundaries for our choropleth maps. 

Méndez-Lázaro et al (2018, 713) cited five census variables as being the most prevalent indicators of vulnerability: age, income, education, social isolation, and healthcare access.  Data for these variables were incorporated into their analysis by setting minimum and maximum values that transformed different units into indices (ibid. 713).  This type of method is called factor analysis, which “reduces a large number of variables into fewer numbers of factors” (Statistics Solutions, n.d.).  For physical data, Méndez-Lázaro et al (2018, 713) added two additional variables to their index from the Landsat rasters: temperature and built-up areas.  Their index was a “sum of the scores for the seven variables” (ibid. 713), which calculated a heat vulnerability index for each census tract.

The same method was used in Jalalzadeh et al’s study (2021, 5) on heat vulnerability in Nebraska.  Additional variables for this study included poverty level and race (ibid. 4).  However, instead of remote sensing, this study was based on different urbanization levels in the state provided by the National Land Cover Database (ibid. 3), making their approach slightly different.  After standardizing their data, each urbanization level needed to be weighed by putting the variables in a matrix before doing factor analysis (ibid. 5).  Since the study did not use remote sensing data, the matrices added an extra dimension of urbanization that allowed the factor analysis to be indexed through social variables.  For the Seattle study, this step is unnecessary.

With all the variables gathered, it then becomes necessary to enter them on an Excel spreadsheet(s) for processing in GIS.  GIS provides the necessary tools for spatial analysis on variables that require mapping (Gomez & Jones 2010, 377-380).  Once the data is entered, they need to be added to a geodatabase that has the appropriate spatial coordinate system for the census tracks.  A geodatabase is a collection of information that can be used to create layers on a map in programs like ArcGIS Pro (ibid. 377).  To do this, census boundaries from ESRI’s Living Atlas (Gilbert, n.d.) will be added to the geodatabase, with the parameters being the Seattle city limits.  The raster file from the Landsat data will be added and classified by census tract.  Finally, census data on a spreadsheet will be added.  Once these have been uploaded as layers, they will be integrated with the Landsat raster data using the join tool (ibid.). 

There are several ways to calculate the index values for each census tract in GIS.  ArcGIS Pro has a tool that can automatically standardize each variable used in the joined table, called the Standardized Field geoprocessing tool (ibid.).  Sidiqui et al (2022, 275) elected to use CommunityViz, an urban planning program with a GIS plugin.  Variables were segregated into layers using the stability wizard tool, which weighed them to be merged into an output map that was composited in ArcGIS to construct a final urban heat vulnerability map (ibid. 275).  However, in case any errors need to be traced, it is more reliable to calculate the index values in Excel for easy reference.  The values can then be uploaded into ArcGIS Pro, as demonstrated by Méndez-Lázaro et al (2018, 713).

Wolf and McGregor (2013, 61) used an entirely different approach.  Their method was a statistical one involving GIS.  They wanted to see whether any evidence of clustering in areas with high or low vulnerability would be found using social variables only (ibid. 61).  They used a hot spot analysis tool within ArcGIS to identify “spatial clusters of statistically significant areas with high or low attribute values” (ibid. 61).  A statistic with a z-score was calculated for each census in the study to represent the significance of clustering, which were then mapped (ibid. 61).  Z-scores are a measure of standard deviation from the mean; they are helpful for indicating whether a variable has a strong relationship with another (Gomez & Jones 2010, 288).  Only hot spots with a z-score greater than |1.96| were considered in their study (Wolf and McGregor 2013, 61), showing a 95% confidence level.  While this method won’t be used in our Seattle study, other researchers may find it more convenient or valuable than creating an index.  Additionally, as the London study did not include any physical variables that measured actual heat in its neighborhoods, it was not determined to be as relevant as the other articles.

Jalalzadeh et al (2021, 5) also used a statistical approach involving hot spot analysis, with the added value of indexing from factor analysis.  First, the values of their variables were categorized into five groups for classifying census tracts on a choropleth map of Nebraska (ibid. 5).  They were then given a factor score to be indexed based on vulnerability level, which was mapped in ArcGIS Pro (ibid. 5).  Moran’s I analysis, which detects local clusters of similar values (Clifford et al 2016, 542), was then used to find “hot spots, cold spots, and outlier census tracts of total vulnerabilities in each urban class of Nebraska”.  Combining these methods may be useful for other researchers, particularly if their study is about larger geographic areas than cities.  As raster data from remote sensing is more accurate at the local level, hot spot analysis may be more beneficial for mapping larger areas like states and regions.

To analyze the data, principal component analysis (PCA) was used to find possible relationships between variables with a varimax rotation (Méndez-Lázaro et al 2018, 713; Jalalzadeh et al 2021, 5; Wolf & McGregor 2013, 61).  Varimax rotation minimizes the number of original variables by loading highly on more significant ones and lowly on less significant ones (Statistics How To, n.d.).  Méndez-Lázaro et al (2018, 713) used SPSS, a statistical software program, to highlight a small number of important variables in their index, which made it easier to interpret their results.  They found that increased heat vulnerability in San Juan was associated with highly built areas, higher rates of solitary living, people with disabilities, advanced age, and a lack of health insurance; the coolest areas corresponded to vegetated areas and urban water bodies (ibid. 709).  Jalalzadeh et al (2021, 6) found that poverty, race, education, and language were the most important risk factors in Nebraskan urban areas.  Wolf & McGregor (2013, 62-63) found that old age, being in a minority, living alone, and crowded living conditions were key factors in London.  Information like this is useful for city planners who wish to ensure a balance between high built-up areas and vegetation in cities, which may reduce the effects of heat related illness.

Based on these methods, we will follow a similar pattern in the research design for Seattle, one that most closely follows the San Juan study.  Census variables to add are race and poverty level, which probably aren’t as relevant in San Juan because most of its population is poor.  Seattle has a very stratified urban population, so class differentiation will be essential.  It is also a city high in diversity, which is why we will include race.  Using the methods above, we will be able to generate the first urban heat vulnerability map for Seattle, WA.

Bibliography

 

Clifford, Nicholas, Meghan Cope, Thomas Gillespie and Shaun French. Key Methods in Geography. Third edition. London: SAGE Publications Ltd, 2016.

Gilbert, Mark.  “Build A Heat Risk Index for Local Climate Planning.”  ESRI.  https://www.esri.com/arcgis-blog/products/arcgis-pro/imagery/heat-resilience-planning-part-1/

Gomez, Basil, and John Paul Jones III. Research Methods in Geography. Chichester: Blackwell Publishing Ltd, 2010.

Jalalzadeh Fard, Babak, Rezaul Mahmood, Michael Hayes, Clinton Rowe, Azar M. Abadi, Martha Shulski, Sharon Medcalf, Rachel Lookadoo & Jesse E. Bell. Mapping Heat Vulnerability Index Based on Different Urbanization Levels in Nebraska, USA. Geohealth 5, no. 10 (2021): e2021GH000478-n/a.

Keeratikasikorn, Chaiyapon & Stefania Bonafoni. Urban Heat Island Analysis Over the Land use Zoning Plan of Bangkok by Means of Landsat 8 Imagery. Remote Sensing (Basel, Switzerland) 10, no. 3 (2018): 440.

Méndez-Lázaro, Pablo, Frank E. Muller-Karger, Daniel Otis, Matthew J. McCarthy, and Ernesto Rodríguez. A Heat Vulnerability Index to Improve Urban Public Health Management in San Juan, Puerto Rico. International Journal of Biometeorology 62, no. 5 (2018): 709-722.

NOAA.  “Lesson 3. Radiometric Correction of Satellite Images: When and Why Radiometric Correction is Necessary”.  Accessed March 27, 2024, https://cwcaribbean.aoml.noaa.gov/bilko/module7/lesson3/

Seattle.gov.  Office of Planning & Community Development.  Accessed March 27, 2024, https://www.seattle.gov/opcd/population-and-demographics/geographic-files-and-maps

Sidiqui, Paras, Phillip B. Roös, Murray Herron, David S. Jones, Emma Duncan, Ali Jalali, Zaheer Allam, et al. Urban Heat Island Vulnerability Mapping using Advanced GIS Data and Tools. Journal of Earth System Science 131, no. 4 (2022): 266-279.

Statistics How To.  “Varimax Rotation: Definition”.  Accessed March 27, 2024, https://www.statisticshowto.com/varimax-rotation-definition/

Statistics Solutions.  “Factor Analysis.”  Accessed March 27, 2024, https://www.statisticssolutions.com/free-resources/directory-of-statistical-analyses/factor-analysis/

Wolf, Tanja & Glenn McGregor. The Development of a Heat Wave Vulnerability Index for London, United Kingdom. Weather and Climate Extremes 1, (2013): 59-68.