Substantial\nUnderestimation of Fine-Mode Aerosol Loading\nfrom Wildfires and Its Radiative Effects in Current Satellite-Based\nRetrievals over the United States

Wildfires generate abundant smoke\nprimarily composed of fine-mode\naerosols. However, accurately measuring the fine-mode aerosol optical\ndepth (fAOD) is highly uncertain in most existing satellite-based\naerosol products. Deep learning offers promise for inferring fAOD,\nbut little has been done using multiangle satellite data. We developed\nan innovative angle-dependent deep-learning model (ADLM) that accounts\nfor angular diversity in dual-angle observations. The model captures\naerosol properties observed from dual angles in the contiguous United\nStates and explores the potential of Greenhouse gases Observing Satellite-2’s\n(GOSAT-2) measurements to retrieve fAOD at a 460 m spatial resolution.\nThe ADLM demonstrates a strong performance through rigorous validation\nagainst ground-based data, revealing small biases. By comparison,\nthe official fAOD product from the Moderate Resolution Imaging Spectroradiometer\n(MODIS), the Visible Infrared Imaging Radiometer Suite (VIIRS), and\nthe Multiangle Imaging Spectroradiometer (MISR) during wildfire events\nis underestimated by more than 40% over western USA. This leads to\nsignificant differences in estimates of aerosol radiative forcing\n(ARF) from wildfires. The ADLM shows more than 20% stronger ARF than\nthe MODIS, VIIRS, and MISR estimates, highlighting a greater impact\nof wildfire fAOD on Earth’s energy balance.