Kīlauea volcano, located on the island of Hawaii, is one of the most active volcanoes in the world. It was in a state of nearly continuous eruption from 1983 to 2018 with copious emissions of sulfur dioxide (SO2) that affected public health, agriculture, and infrastructure over large portions of the island. Since 2010, the University of Hawaiʻi at Mānoa provides publicly available vog forecasts that began in 2010 to aid in the mitigation of volcanic smog (or “vog”) as a hazard. In September 2017, the forecast system began to produce operational ensemble forecasts. The months that preceded Kīlauea’s historic lower east rift zone eruption of 2018 provide an opportunity to evaluate the newly implemented air quality ensemble prediction system and compare it another approach to the generation of ensemble members. One of the two approaches generates perturbations in the wind field while the other perturbs the sulfur dioxide (SO2) emission rate from the volcano. This comparison has implications for the limits of forecast predictability under the particularly dynamic conditions at Kīlauea volcano. We show that for ensemble forecasts of SO2 generated under these conditions, the uncertainty associated with the SO2 emission rate approaches that of the uncertainty in the wind field. However, the inclusion of a fluctuating SO2 emission rate has the potential to improve the prediction of the changes in air quality downwind of the volcano with suitable postprocessing.