Development of JEDI-Based Satellite Aerosol Optical Depth Assimilation for the RRFS Smoke and Dust Model

Wildfire activity has been increasing across parts of the western United States and Canada in recent years, and with the increased wildfire activity comes increased smoke in the atmosphere. Wildfire smoke poses air quality hazards, limits visibility, and impacts atmospheric radiative transfer. To better forecast air quality and visibility, and to account for the radiative effects of smoke on forecasts in general, NOAA Global Systems Laboratory has been developing smoke forecasting components for its regional models beginning with the High Resolution Rapid Refresh - Smoke (HRRR-Smoke) model. This development continues in the Rapid Refresh Forecast System - Smoke and Dust (RRFS-SD) model with the addition of dust forecasting, improved model physics, and improved data assimilation. Smoke is introduced into the RRFS-SD model using Fire Radiative Power (FRP) to drive smoke-injection algorithms, while dust emissions are dependent on near-surface winds and surface characteristics. Forecasts of atmospheric smoke and dust can deviate from the true evolution of the atmosphere due to uncertainties and/or errors in FRP, injection algorithms, and modeled atmospheric flow. To improve forecasts, satellite retrievals of smoke- and dust-affected aerosol optical depth (AOD) can be assimilated to mitigate errors in the initial state estimates of smoke and dust fields used in models like the RRFS-SD.

This presentation will cover the development of a JEDI based system for VIIRS AOD data assimilation in the RRFS-SD which separately assimilates smoke- and dust-affected satellite AOD observations. The work to be presented includes the separation of smoke- and dust-affected observations from the full AOD dataset using an enhanced IODA converter, configuration and optimization of a CRTM- and GOCART-based observation operator to simulate AOD from modeled smoke and dust fields, development of static background error covariances, and examples of cycled tests of the assimilation system in the RRFS-SD workflow. The final product of this work will be an EnVar data assimilation system, but this presentation will mostly focus on development of the variational portion of the system.