Spatiotemporal exposure modeling and epidemiological analyses of the 2008 northern California wildfires

Projections under likely climate change scenarios demonstrate that the area burned from wildfires and the length of the wildfire season will continue to increase in the western US and in many other parts of the world. Wildfires emit many air pollutants of concern for public health. In this dissertation, I systematically reviewed the scientific literature on which health effects were associated with wildfire smoke exposure, piloted the most sophisticated exposure model to date to estimate PM2.5 exposures during the 2008 northern California wildfires, and assessed the effect of these exposures on respiratory and cardiovascular hospitalizations. The scientific literature documents clear evidence of respiratory health effects from wildfire smoke, most specifically for exacerbations of asthma and chronic obstructive pulmonary disease (COPD). Only inconsistent evidence exists for cardiovascular effects. Several high-quality studies demonstrate significant associations between wildfire smoke exposure and all-cause mortality, but not for specific causes of mortality. To study the 2008 northern California wildfires, I created a spatiotemporal exposure model that used the latest in machine learning methods to combine information from satellites, output from a chemical transport model, and meteorological and other spatial and temporal variables to create an optimal exposure model. The best model had an out-of-sample cross-validated R2 of 0.80. Using this model, I could estimate exposures at any location on any day within the study area. To analyze the health effects of the 2008 northern California wildfires, I estimated exposure at all ZIP codes within the study region and found evidence of increased asthma, COPD, pneumonia, and all-cause respiratory hospitalizations associated with PM2.5 during the fire period. Only for asthma hospitalizations was this effect significantly different from effects for PM2.5 outside of the fire period. I also found some evidence of increased rates of hospitalization for asthma associated with PM2.5 during the wildfire period among women compared to men, and among adults compared to children and the elderly. I also found higher rates of hospitalizations for COPD and pneumonia in ZIP codes with lower levels of owner-occupied housing, but I did not find consistent differences along gradients of other measures of socio-economic status by ZIP code. My research piloted the first spatiotemporal exposure model to use machine learning methods to improve estimation of PM2.5 exposures during wildfires using a combination of satellite data, chemical transport model output, meteorological parameters, and land use information. The results from my epidemiological analysis confirm previous evidence of exacerbation of respiratory health effects from exposure to wildfire PM2.5 and highlight evidence of groups that are more vulnerable to this exposure