During severe wildfire seasons, smoke can have a significant impact on air quality in Australia. Our study demonstrates that characterization of the smoke plume injection fractions greatly affects estimates of surface smoke PM_2.5. Using the plume behavior predicted by the machine learning method leads to the best model agreement with observed surface PM_2.5 in key cities across Australia, with smoke PM_2.5 accounting for 5 %–52 % of total PM_2.5 on average during fire seasons from 2009 to 2020.