Advancements made to NOAA’s FENGSHA dust emission scheme within the Unified Forecast System with applications to regional air quality and subseasonal to seasonal forecasting

This is the authors' abstract. We don't add key points for this paper.

NOAA is developing the next generation air quality  prediction system for the United States  and global aerosol predictions within the Unified Forecast System (UFS) framework.  A major goal of this effort is to better represent and forecast impacts of extreme events like wildfires and dust storms on air quality  and impacts of aerosols globally on weather ranging from hourly  to seasonal scales. The FENSGHA (English analog of the Mandarin term for wind-blown dust)  emissions scheme developed at NOAA Air Resources Laboratory is adapted to the UFS for applications in the high resolution Rapid Refresh Forecast System with Smoke and Dust (RRFS-SD), the coupled online full chemistry air quality model (UFS-AQM), and the Global Ensemble Forecast System (GEFS) with prognostic aerosols.  One of the key parameters to model aeolian emissions within weather, climate, and air quality models is the threshold friction velocity (TFV). While in many models the TFV is determined by knowledge of soil size distributions and characteristics often assumed on a global scale, the FENGSHA dust emission model was originally constrained by observations following the works of Dale Gillette (1980, 1982, 1988). In this way, FENGSHA is a data-driven model, constrained by observations of saltation with respect to soil characteristics. In this study, a new way to determine the TFV is developed through a machine learning approach  constrained by observations and high resolution soil characteristics. We first use observed dust points detected by satellites compiled in Hennen et al. (2022), soil parameters from the SOILGRIDS version 2.0 or Beijing Normal University database, and ERA5 Land reanalysis data to train the machine learning model for  the TFV based solely on the soil inputs, including composition and physical characteristics. The machine learning  model results compare well with observations from Gillette and provide a robust and high resolution method to accurately describe the saltation velocity in both time and scale. Next, we apply this TFV within the  suite of UFS applications described above from regional to global scale. Retrospective simulations are compared with surface and satellite observations during dust events.