Timeliness of Fire Observations as a Skill Factor for Aerosol Prediction

Near-real-time predictions of atmospheric conditions in areas affected by smoke rely on dynamic estimates of smoke production. Fires are widely understood to be a weather-driven phenomenon, and current atmospheric models employ a wide range of methods to account for the impact of weather on fire growth and smoke plume behavior. However, the factors affecting fire ignition and development to the point of self-sustained combustion operate at very fine scales and are difficult to model over regional domains. We first describe the factors affecting timeliness of fire information in operational smoke prediction systems. We describe and quantify the wide gap between best-case and worst-case fire timeliness, and discuss modifications of prediction workflows that can maximize timeliness of fire information. We then quantify the effects of data timeliness on the skill of smoke predictions by comparing retrospective runs with complete fire observations to reforecast simulations limited by the timeliness constraints of different near-real-time operational use cases. We conclude with qualitative insights about the relative importance of different phases of fire growth for smoke prediction skill.