Describing the Global Distribution and Health Response of Particulate Matter Using Modern Computational Tools

Abstract

Atmospheric aerosols have important impacts on climate and human health. This thesis advances our knowledge of the 3-dimensional distribution of atmospheric aerosols by using state of the art tools and methods to new fields or new sources of data in three applications. First, I use adjoint modeling to provide the global health response to fine particulate matter with a diameter < 2.5 μm (PM2.5) by implementing the Global Burden of Disease Project's Integrated Response Function. The response of global mortality to changes in local unit-mass anthropogenic emissions varied spatially by several orders of magnitude; the largest reductions in mortality for a 1 kg km-2yr-1 decrease in emissions were for ammonia and carbonaceous aerosols in Eastern Europe. When comparing relative responses, the greatest reductions in mortality for a 10% decrease in emissions were found for secondary inorganic sources in East Asia. Next, I applied the same adjoint to assimilate CALIOP Lidar profiles. Comparing mean extinction height (Zα) between the baseline simulation and two optimised simulations to CALIOP observed Zα, both optimisations improved the slope and offset of a linear fit across AEROCOM regions, but optimising initial conditions improved agreement (R = 0.78) compared to the baseline (R = 0.72) while optimising emissions decreased agreement (R = 0.65). In a global comparison of AOD, the optimised emissions greatly improved agreement with observations in the Sahara in January, but failed to capture the overall underestimate of AOD seen almost everywhere. Finally, I used the modern machine learning framework Tensorflow to compute a high resolution (0.01°x0.01°) map of probability distributions of PM2.5 fitted to ground monitoring data from the World Health Organization's cities database. This model achieved an average correlation R2 of 0.93 and an average RMSD = 5.00 μgm-3 in 10-fold cross validation. Because the model outputs probability distributions at each grid location, I was able to calculate a global probability density function for population exposure to PM2.5. Based on the global PDF, 83% of the world’s population exceeds the WHO guideline of 10 μgm-3.