An algorithm for retrieving nighttime aerosol optical depth (AOD) from the Visible Infrared Imaging Radiometer Suite (VIIRS) Day-Night Band (DNB) observations of reflected moonlight is presented for rural areas during the western U.S. fire seasons. The algorithm uses the UNified and Linearized Vector Radiative Transfer Model (UNL-VRTM) with newly developed capabilities for considering lunar illuminations. Cloud and fire pixels are screened out by utilizing the radiance from the VIIRS Moderate-resolution Bands (M-Band) and the DNB. Rural and city pixels are classified based on a pre-calculated city light database. The surface spectral reflectance for DNB ranging from 342 to 1107 nm is estimated by a random forest approach, which is trained using the surface spectral reflectance from the existing spectral libraries. For the fire seasons of 2017 and 2020, the nighttime AOD retrieval is shown to play an indispensable role in describing the nonlinear diurnal movement of smoke transport and discerning the source of smoke plumes heretofore observable only in the daytime. The retrieved AOD values show good agreement with spatiotemporally collocated Aerosol Robotic NETwork (AERONET) and Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) AOD values, with linear correlation coefficient values of ~0.96/0.95 and ~86%/69% of the AOD pairs falling in an uncertainty envelope of ±(0.085 + 0.10AOD), which is superior to AOD reanalysis from Modern-Era Retrospective analysis for Research and Applications, version 2 (MERRA-2). These results affirm the significant potential of nighttime AOD to improve the analysis and forecast of regional to global biomass-burning aerosol distributions, filling a critical gap in the diurnal description of a key element of Earth's climate system.