Developing High-Resolution Descriptions of Urban Heat Islands: A Public Health Imperative
GIS SNAPSHOT — Volume 13 — September 15, 2016
Jackson Voelkel, BS; Vivek Shandas, PhD; Brendon Haggerty, MURP
Suggested citation for this article: Voelkel J, Shandas V, Haggerty B. Developing High-Resolution Descriptions of Urban Heat Islands: A Public Health Imperative. Prev Chronic Dis 2016;13:160099. DOI: http://dx.doi.org/10.5888/pcd13.160099.
Extreme heat events affect the most vulnerable human populations and are a lethal health hazard to urban dwellers globally; in the United States, extreme heat causes more deaths annually than all other weather events and natural hazards combined (1). Previous studies described urban heat islands as isolated, static, monolithic areas of cities. We challenged this contention by hypothesizing that diurnal temperature cycles and diverse landscape features create variation in places that amplify heat (2). A temporal description of urban heat islands would identify populations that are susceptible to heat stress, particularly at night, when most people are asleep and unable to regulate internal body temperatures. If public health agencies are to prevent illness and death caused by heat, they will need to know which populations are most vulnerable to heat stress, particularly at night; such information can guide timely interventions (3). Researchers lack high-resolution tools for identifying neighborhoods and households where extreme weather events might have profound and fatal effects on human health. The objective of this study was to use spatial analytics at previously unattained resolutions to answer the following research question: to what extent can we observe temporal variation in urban heat islands and the physical features that induce heat stress?
Following an established protocol (4), we collected approximately 60,000 temperature readings during 1 day of an extreme heat event on August 25, 2014, in Portland, Oregon, when the average temperature during the hottest hour of the day was in the 75th percentile of 30-year historic daily temperatures for the study region. We sampled temperatures for 1 hour at 3 times during the day (6 AM, 3 PM, and 7 PM) using vehicle traverses (cars with a mounted temperature sensor and global positioning system [GPS]) in 6 predetermined sections of the city. The temperature sensor consisted of a type T fine (30 gauge) thermocouple in a plastic shade tube (12 cm in length and 2.5 cm in diameter) mounted on the passenger-side window approximately 25 cm above the roof of each of 5 vehicles deployed. Each temperature sensor was connected to a data-logging device with an estimated system accuracy of ±0.5°C and a 90% response time of less than 60 s in 1 m/s airflow. A GPS unit on each vehicle paired temperature measurement and location.
On the basis of a sensitivity analysis and research on landscape features that mediate urban heat, we selected 6 variables as predictors: 1) building heights, 2) standard deviation of building heights, 3) building volume, 4) canopy cover, 5) low-lying vegetation, and 6) canopy biomass. Data on the first 3 variables were derived from 3-dimensional point cloud data acquired through Light Detection and Ranging (LiDAR). LiDAR combines a laser ranging device with a GPS system to provide highly resolved terrain measurements. The LiDAR data were extracted from the 2014 Oregon Metro Regional Land Information System (RLIS) (5). Data on canopy cover and low-lying vegetation were created from 2014 Portland LiDAR/photography flight data (5). The biomass metric was created by multiplying tree height by tree density determined by LiDAR-beam tree penetrability. Using a moving window analysis at 15 spatial extents (also known as “buffer distances”) from 50 m to 1 km, we tested the effective distances of each variable on the urban heat island, wherein each pixel represented the amount of each variable within a specified distance.
We analyzed modeling techniques (Appendix) and determined that random forest modeling (a machine-learning model available in the statistical package R [The R Foundation]) was more accurate than standard linear modeling. The values of the 90 new buffer-distance grids were spatially assigned to a randomly selected 70% of the traverse points; we tested the validity of our model by predicting the remaining 30%. This 70%–30% training model predicted a temperature for each of the 1.034 billion 1-m pixels. The models generated data on changes in mean standard error, which represents the effect of each on local temperature. Our statistical technique was run for each of the 3 one-hour data-collection periods, resulting in 3 temperature-prediction grids.
The models for the morning and evening data-collection periods predicted upwards of 98% (r2 = 0.98) and 97% (r2 = 0.97) of the temperature variation across the study region; the afternoon model had a predicting power of 83% (r2 = 0.83). Although the afternoon model was weaker, possibly because of atmospheric mixing and surface convective processes that we were unable to detect, it performed remarkably well.
In contrast with previous research findings, our models suggest that each data-collection period had unique land-use and land-cover factors that helped to explain variation (Table). In the morning (6 AM), low-lying vegetation cover had the strongest effect on temperature, and in the afternoon (3 PM), the 2 variables for building height (mean building heights and variation in building heights) had the strongest effect. In the evening (7 PM), temperatures were most strongly affected by the variation in building heights. Material science and computational fluid dynamics processes suggest that buildings absorb incoming solar radiation during the day and re-radiate it as heat at night (2) and that variation in building heights helps to circulate air. Indeed, diurnal patterns of extreme heat, documented in studies of heat mortality and urban heat islands, show that the re-radiation of heat by buildings peaks at night and places vulnerable individuals at greatest risk of death from heat. The amount of canopy cover may moderate temperatures at night because this variable is the strongest predictor in the morning.
By using these empirically derived heat measures, local land-use and land-cover variables, and spatial machine learning techniques, we described and explained variation in the distribution of urban heat islands in Portland, Oregon. High-resolution data sets and analysis such as those used here can inform preparation for extreme heat events and public health interventions (eg, information campaigns, cooling centers, tree planting programs, and surveillance) for vulnerable communities in local and regional areas. Our results also suggest that policy and environmental interventions should deploy temperature-mitigation strategies at night, when heat stress is greatest for vulnerable communities.
We thank the Institute for Sustainability Solutions at Portland State University, the US Forest Service (no. 2011-DG-11062765-016), and the U.S. National Science Foundation (no. 1444755) for providing financial support for this project. We also thank Anandi van Diepen-Hedayat for providing conceptual and grammatical clarity on an early draft of this manuscript.
Corresponding Author: Vivek Shandas, PhD, Urban Studies and Planning, Portland State University, PO Box 751, Portland, Oregon 97201-0751. Telephone: 503-725-5222. Email: firstname.lastname@example.org.
Author Affiliations: Jackson Voelkel, Nohad A. Toulan School of Urban Studies and Planning Portland State University, Portland, Oregon; Brendon Haggerty, Multnomah County Health Department, Portland, Oregon.
- Luber G, McGeehin M. Climate change and extreme heat events. Am J Prev Med 2008;35(5):429–35. CrossRef PubMed
- 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. CrossRef PubMed
- Balbus JM, Malina C. Identifying vulnerable subpopulations for climate change health effects in the United States. J Occup Environ Med 2009;51(1):33–7. CrossRef PubMed
- Hart MA, Sailor D. Quantifying the influence of land-use and surface characteristics on spatial variability in the urban heat island. Theor Appl Climatol 2009;95(3):397–406. CrossRef
- Metro Data Resource Center. RLIS Live: Metro’s Regional Land Information System. http://www.oregonmetro.gov/rlis-live. Accessed August 21, 2014.
- Ryan PH, LeMasters GK. A review of land-use regression models for characterizing intraurban air pollution exposure. Inhal Toxicol 2007;19(Suppl 1):127–33. CrossRef PubMed
- Lee L, Chen L, Wang X, Zhao J. Use of Landsat TM/ETM data to analyze urban heat island and its relationship with land use/cover change. International Conference on Remote Sensing, Environment and Transportation Engineering. 2011 Jun 24–26; Nanjing, China.
- Makido Y, Shandas V, Ferwati S, Sailor D. Daytime variation of urban heat islands: the case study of Doha, Qatar. Climate 2016;4(2):32. CrossRef