Estimating Fine-Scale Heat-Related Vulnerability Using Principal Components Analysis: Are We Answering the Wrong Question?

Abstract

Estimating Fine-Scale Heat-Related Vulnerability Using Principal Components Analysis: Are We Answering the Wrong Question?Abstract Number:2728 Kathryn Conlon*, Evan Mallen, Carina Gronlund, Veronica Berrocal, Larissa Larsen, and Marie O'Neill Kathryn Conlon* National Center for Atmospheric Research, United States, E-mail Address: [email protected] Search for more papers by this author , Evan Mallen University of Michigan, Taubman College of Architecture and Urban Planning, United States, E-mail Address: [email protected] Search for more papers by this author , Carina Gronlund University of MIchigan, School of Public Health, Department of Epidemiology, United States, E-mail Address: [email protected] Search for more papers by this author , Veronica Berrocal University of Michigan, School of Public Health, Department of Biostatistics, United States, E-mail Address: [email protected] Search for more papers by this author , Larissa Larsen University of Michigan, Taubman College of Architecture and Urban Planning, United States, E-mail Address: [email protected] Search for more papers by this author , and Marie O'Neill University of MIchigan, School of Public Health, Departments of Environmental Health Sciences and Epidemiology, United States, E-mail Address: [email protected] Search for more papers by this author AbstractEfforts to reduce heat-related morbidity and mortality sometimes use maps of population and community-level characteristics shown to contribute to heat vulnerability. Because heat vulnerability involves components of exposure, sensitivity and adaptive capacity, composite vulnerability indices have been integrated into maps, often using methods like principal components analysis (PCA). These analyses and maps may have influential roles in the development of heat preparedness, response, and mitigation policies. Evaluation of the methodology that produces these measures and provision of guidance on how to interpret the results is crucial. We evaluated the process of calculating and visualizing fine-scale composite heat vulnerabilities for the City of Detroit, MI, USA. We used PCA to create heat vulnerability indices (HVI) for census block groups and census tracts by two methods. The ‘un- supervised’ method used variables selected a priori including socio-demographic characteristics used in Reid et al (2009) plus percent tree canopy, percent non-canopied and distance to water. The ’supervised’ method used variables correlated with the log- transformed ratio of all-cause mortality occurring on a hot day in the City of Detroit (2000 – 2009), where hot days were defined as days above the month-specific 95th percentile of apparent temperature. The two methods produced spatially different patterns of vulnerability across Detroit, and the supervised PCA included only half of the variables used in the un-supervised method. These results suggest that PCA-based analyses intended for data reduction and manageability may not be optimal for creation of city- specific HVIs that accurately depict heat vulnerability. We propose partnerships with local users and validation of HVI’s with health data as a potentially more fruitful approach for vulnerability mapping.