In recent years, machine learning (ML) models have been used for improving physical parameterizations of general circulation models (GCMs). A significant challenge of integrating ML models into GCMs is the online instability when they are coupled for long-term simulation. In this study, we present a new strategy that demonstrates robust online stability when the entire physical parameterization package of a GCM is replaced by a deep ML algorithm. The method uses a multistep training scheme of the machine learning model with experience replay in which the memory of physical tendencies from the training dataset and the ML algorithm’s own output at the previous time step are used in the training. The physics memory improves the accuracy of the machine learning model, while the experience replay constrains the amplification of cumulative errors in the online coupling. The method is used to train the whole physical parameterization package for the Community Atmosphere Model version 5 (CAM5) with data from its Multi-scale Modeling Framework (MMF) high resolution simulations. Three 6-year online simulations of the CAM5 with the ML physics package at operational spatial resolution with real-world geography are presented. The simulated spatial distributions of precipitation, surface temperature and zonally averaged atmospheric fields demonstrate overall better accuracy than that of the standard CAM5 and benchmark model even without the use of additional physical constraints or tuning. This work is the first to demonstrate a solution to address the online instability problem in climate modeling with ML physics by using experience replay.

A reasonable representation of orographic anisotropy in earth system models is vital for improving weather and climate modeling. In this study, we implemented the orographic drag scheme, including 3-D orographic anisotropy (3D-AFD), into the Chinese Academy of Sciences Earth System Model version 2 (CAS-ESM 2.0). Three groups of simulations named sensitivity run, medium-range forecast, and seasonal forecast respectively were conducted using the updated CAS-ESM model together with the original 2-D isotropic scheme (2-D) and the 3-D orographic anisotropy for the eight-direction scheme (3D-8x) to validate its performance. Sensitivity runs indicated that the simulated drag using the original 2-D scheme did not change with the wind directions, while the simulated drag using the updated 3D-AFD showed a smoother transition than that using 3D-8x. The 3D-AFD and 3D-8x had also about 80% larger drag and smaller wind speed of 1m/s than that of the 2-D scheme. Enhanced drag in the medium range and seasonal forecast using the updated CAS-ESM both alleviated the bias of the overestimated wind speed and the cold bias over mountain regions in the 2-D scheme. This was more apparent in winter (0.4-0.5 m/s and ~1K) than that in summer (0.1 m/s and ~0.1K) for the northern hemisphere region, such as the Tibetan Plateau. The vertical wind profile was also improved in the seasonal forecast. The results suggested that a reasonable representation of the orographic anisotropy was important in climate modeling, and the updated model of CAS-ESM with 3D-AFD alleviated the bias of the mountain wind.