Network Systems Science and Advanced Computing | Resilient Societies and Interdependent Infrastructures

From Space to Front Porch: Connecting Earth Observations to Health Outcomes with an Environmental Exposure Modeling System



How does weather (and climate) affect human health? Every year, we are seeing increasingly harsher weather, including bigger and more frequent hurricanes, stronger heatwaves, wildfires, and more. Are some people more vulnerable than others? How do geography, the built environment, and human activity patterns affect vulnerability?

In this project, we are developing methods and tools to combine Earth Observations data, population models, and health data to create better estimates of vulnerability, using Hurricane Harvey as our case study. The project team includes collaborators at Virginia Tech (the team lead) and Johns Hopkins University.

Project Overview

We are working with partners at the CDC, who already provide a widely used static measure of vulnerability known as the Social Vulnerability Index (SVI). Our goal is to augment the SVI by extending it to include population dynamics. Synthetic populations and disaster simulations developed at UVA are essential to this goal since they allow modeling population mobility in typical as well as atypical situations (such as during a hurricane). For example, by simulating evacuation before and during a hurricane, we can estimate which people are at greatest risk of exposure to various hazards, such as flooding and chemical toxicant spills. Earth Observations data provide information about hazards and health data provide information about the effects of exposure. Population modeling allows us to link the two.


Research in this project is ongoing. Early activities and results include,

  • Analysis of an association between flooding and reported illness during and following Hurricane Harvey.
  • An analysis of the relationship between the SVI and FEMA aid applications, while controlling for physical conditions such as inundation, surface imperviousness, and power outage at the Census tract level.

In both cases above, we found that the SVI helps improve predictions. We are now working on computing a dynamic version of the SVI based on population mobility derived from synthetic populations. Our goal is to see if dynamic measures of vulnerability can further improve predictions.

The figure below shows a comparison for the state of Texas, between CDC SVI (Theme 1 is based on socioeconomic factors) and our Dynamic SVI based on weekly activity patterns computed from our synthetic population. The lower left panel shows how the correlation between the two changes over the course of a week. The correlation is highest at night, when most people are at their home locations, and lowest in the late afternoon, when people are out for work and other activities. We also see that the daytime discrepancy is smaller on weekend days, when people do not travel as far from home as on weekdays.

In order to simulation population mobility during Hurricane Harvey, we are assembling data from multiple sources to develop an integrated model and simulation of evacuation and other behaviors. This will help generate estimates of population exposures to hazards that can be correlated with health outcomes.


Research Associate Professor

Student Researcher