SPHERE: A Synthetic Population-Based Heat Exposure Estimation Platform

Methods to estimate the effects of heat exposure on health generally use weather station or satellite-derived temperature data to estimate exposures, which are then regressed against health outcome data such as emergency department visits. This ignores the role of population mobility. Since a significant proportion of people generally aren’t at home during the hottest parts of the day, it is important to account for activity patterns and activity locations to properly estimate heat exposure. Since representative samples of such data aren’t directly available, we turn to the use of detailed population models, known as synthetic populations, to fill this gap. We describe the SPHERE platform, which uses a synthetic population generated by integrating data from multiple sources to model demographics and mobility at the individual level, in combination with high resolution temperature data to provide improved estimates of heat exposure. Hourly heat estimates are generated using Parameter-elevation Regressions on Independent Slopes (PRISM) Model daily data and MetSim to fit a diurnal curve. The synthetic population is generated via a software pipeline that integrates multiple data sets as follows. First, blockgroup level data from the US Census (demographic distributions and microsamples) are used to generate disaggregated individuals and households using a method known as iterative proportional fitting. Second, households are assigned residences using home location data from Black Knight. Third, each individual is assigned a demographic-appropriate activity sequence using the National Household Travel Survey. Finally, an appropriate activity location is assigned for each activity of each individual using location data from multiple sources (HERE Maps, the National Center for Education Statistics, the Microsoft Building Database, and more). The resulting synthetic population gives a model of mobility of the population of a chosen region that is deeply grounded in data about the demographics and built environment. The SPHERE platform runs a scalable parallel simulation of population movements, aggregating exposures to varying heat conditions throughout a chosen time period. These exposures can be further aggregated by regions and/or demographic groups as desired, thus facilitating comparisons and downstream analysis, such as health effect estimation or intervention analysis. SPHERE can also run a “live” simulation, where the platform can be coupled with weather data APIs to automatically update exposure estimates as temperature data/forecasts are updated.