Direct Assimilation of Satellite Visible and Near-Infrared Radiances to Improve Aerosol Simulations

Abstract

Existing satellite-based aerosol data assimilation approaches mostly rely on assimilating retrieved aerosol products, which suffer from inconsistencies between retrieval assumptions and model representations, as well as uncertainties in the retrieved aerosol parameters. Here we develop a direct shortwave radiance assimilation system for improving aerosol simulations in the Weather Research and Forecasting model coupled with Chemistry (WRF-Chem). The system integrates AI-based observation operator, trained on 1,000,000 samples from forward radiative transfer simulations, and a three-Dimensional Variational Data Assimilation (3DVAR) assimilation module to simultaneously constrain multi-band aerosol optical depth (AOD) and surface albedo using radiance measurements from five shortwave bands of Moderate Resolution Imaging Spectroradiometer (MODIS) under clear sky. Observing System Simulation Experiment (OSSE) results demonstrate that the system can effectively disentangle aerosol and surface signals, with forward simulated radiance from the analysis field being closer to pseudo-observations than those from the background field. Experiments using MODIS radiance measurements show substantial improvements. In the February 2016 case study, assimilation reduces the domain-mean spatial AOD error relative to MODIS retrievals by 42.19%, and reduces the absolute AOD bias relative to independent AERONET observations by 53.46%–79.76% across four validation sites. The continuous assimilation experiment spanning December 2016 to February 2017 demonstrates robust long-term performance, reducing AOD errors by 52.77% overall. The assimilation scheme further improves simulated surface albedo and surface downward shortwave radiation, demonstrating the potential to better represent radiative transfer process in numerical weather prediction and climate models.