European Geosciences Union, Geoscientific Model Development, 8(13), p. 3731-3768, 2020
Full text: Download
Abstract. High-performance computing resources allow us to conduct numerical simulations with a horizontal grid spacing that is sufficiently high to resolve cloud systems on a global scale, and high-resolution models (HRMs) generally provide better simulation performance than low-resolution models (LRMs). In this study, we execute a next-generation model that is capable of simulating global aerosols using version 16 of the Nonhydrostatic Icosahedral Atmospheric Model (NICAM.16). The simulated aerosol distributions are obtained for 3 years with an HRM using a global 14 km grid spacing, an unprecedentedly high horizontal resolution and long integration period. For comparison, a NICAM with a 56 km grid spacing is also run as an LRM, although this horizontal resolution is still high among current global aerosol climate models. The comparison elucidated that the differences in the various variables of meteorological fields, including the wind speed, precipitation, clouds, radiation fluxes and total aerosols, are generally within 10 % of their annual averages, but most of the variables related to aerosols simulated by the HRM are slightly closer to the observations than are those simulated by the LRM. Upon investigating the aerosol components, the differences in the water-insoluble black carbon and sulfate concentrations between the HRM and LRM are large (up to 32 %), even in the annual averages. This finding is attributed to the differences in the aerosol wet deposition flux, which is determined by the conversion rate of cloud to precipitation, and the difference between the HRM and LRM is approximately 20 %. Additionally, the differences in the simulated aerosol concentrations at polluted sites during polluted months between the HRM and LRM are estimated with normalized mean biases of −19 % for black carbon (BC), −5 % for sulfate and −3 % for the aerosol optical thickness (AOT). These findings indicate that the impacts of higher horizontal grid spacings on model performance for secondary products such as sulfate, and complex products such as the AOT, are weaker than those for primary products, such as BC. On a global scale, the subgrid variabilities in the simulated AOT and cloud optical thickness (COT) in the 1∘×1∘ domain using 6-hourly data are estimated to be 28.5 % and 80.0 %, respectively, in the HRM, whereas the corresponding differences are 16.6 % and 22.9 % in the LRM. Over the Arctic, both the HRM and the LRM generally reproduce the observed aerosols, but the largest difference in the surface BC mass concentrations between the HRM and LRM reaches 30 % in spring (the HRM-simulated results are closer to the observations). The vertical distributions of the HRM- and LRM-simulated aerosols are generally close to the measurements, but the differences between the HRM and LRM results are large above a height of approximately 3 km, mainly due to differences in the wet deposition of aerosols. The global annual averages of the effective radiative forcings due to aerosol–radiation and aerosol–cloud interactions (ERFari and ERFaci) attributed to anthropogenic aerosols in the HRM are estimated to be -0.293±0.001 and -0.919±0.004 W m−2, respectively, whereas those in the LRM are -0.239±0.002 and -1.101±0.013 W m−2. The differences in the ERFari between the HRM and LRM are primarily caused by those in the aerosol burden, whereas the differences in the ERFaci are primarily caused by those in the cloud expression and performance, which are attributed to the grid spacing. The analysis of interannual variability revealed that the difference in reproducibility of both sulfate and carbonaceous aerosols at different horizontal resolution is greater than their interannual variability over 3 years, but those of dust and sea salt AOT and possibly clouds were the opposite. Because at least 10 times the computer resources are required for the HRM (14 km grid) compared to the LRM (56 km grid), these findings in this study help modelers decide whether the objectives can be achieved using such higher resolution or not under the limitation of available computational resources.