Recent wildfire activity in Canada has been rising the public demand for smoke forecasts. This however comes with a challenge for national weather centers to provide fast, yet accurate forecasts of these strong unexpected events. This study explores the potential of a novel approach for operational air quality forecasting during wildfire smoke episodes. It not only enables an improved use of sparse observation signals, but also provides theoretical insights how uncertainties evolve over time.

While atmospheric chemical forecasts rely on uncertain model parameters, their huge dimensions hamper an efficient uncertainty estimation. This study presents a novel approach to efficiently sample these uncertainties by extracting dominant dependencies and correlations. Applying the algorithm to biogenic emissions, their uncertainties can be estimated from a low number of dominant components. This states the capability of an efficient treatment of parameter uncertainties in atmospheric models.