A GPU-Enabled Aerosol Model for Global Kilometer-Scale Earth System Modeling

JEROME FAST, Balwinder Singh, Oscar Diaz-Ibarra, James Overfelt, Michael Schmidt, Brian Gaudet, Meng Huang, Chandru Dhandapani, Kai Zhang, Mingxuan Wu, Shuaiqi Tang, Kyle Pressel, Jaelyn Litzinger, Po-Lun Ma, Pacific Northwest National Laboratory

     Abstract Number: 271
     Working Group: Aerosols, Clouds and Climate

Abstract
The lifecycle of aerosols is a key component of global Earth System Models since aerosols directly and indirectly affect atmospheric processes through their interactions with radiation and clouds. However, global models still contain relatively large uncertainties in the prediction of aerosol properties, such as number, size, and composition, and their interactions with the atmosphere. These uncertainties can be attributed to both imperfect parameterizations of aerosol processes and inadequate resolution. Global models typically have grid spacings of 50 km or more that cannot resolve the observed high spatial variability of aerosol properties as well as overlapping aerosols and cloud fields. Consequently, global Earth System Models have recently been developed to simulate atmospheric processes using grid spacings on the order of a few kilometers to better represent spatial variabilities in aerosols and clouds. One such model is DOE’s next-generation Energy Exascale Earth System Model (E3SM) Atmosphere Model in C++ (EAMxx), formerly known as the Simple Cloud Resolving E3SM Atmosphere Model (SCREAM), that uses C++/Kokkos to take advantage of GPUs on exascale computers. Here we describe how the four-mode version of the Modal Aerosol Model (MAM4) has been ported to C++/Kokkos and integrated into EAMxx.  Extensive tests were performed to ensure the ported code behaves in the same way as the original Fortran code used in earlier versions of E3SM. Predictions are compared with observations and simulations with coarser resolution and the computational cost is quantified to demonstrate the potential benefits of this new modeling framework. In addition, we discuss how this modeling framework will influence the development and testing of new aerosol parameterizations, including how new findings from laboratory and field measurements are integrated into the next generation of models.