Assessing the cloud radiative bias at Macquarie Island in the ACCESS-AM2 model

Abstract. As a long-standing problem in climate models, large positive shortwave radiation biases exist at the surface over the Southern Ocean, impacting the accurate simulation of sea surface temperature, atmospheric circulation, and precipitation. Underestimations of low-level cloud fraction and liquid water content are suggested to predominantly contribute to these radiation biases. Most model evaluations for radiation focus on summer and rely on satellite products, which have their own limitations. In this work, we use surface-based observations at Macquarie Island to provide the first long-term, seasonal evaluation of both downwelling surface shortwave and longwave radiation in the Australian Community Climate and Earth System Simulator Atmosphere-only Model Version 2 (ACCESS-AM2) over the Southern Ocean. The capacity of the Clouds and the Earth’s Radiant Energy System (CERES) product to simulate radiation is also investigated. We utilise the novel lidar simulator, the Automatic Lidar and Ceilometer Framework (ALCF) and all-sky cloud camera observations of cloud fraction to investigate how radiation biases are influenced by cloud properties.

Overall, we find an overestimation of +9.5 ± 33.5 W m⁻² for downwelling surface shortwave radiation fluxes and an underestimation of -2.3 ± 13.5 W m⁻² for downwelling surface longwave radiation in ACCESS-AM2 in all-sky conditions, with more pronounced shortwave biases of +25.0 ± 48.0 W m⁻² occurring in summer. CERES presents an overestimation of +8.0 ± 18.0 W m⁻² for the shortwave and an underestimation of -12.1 ± 12.2 W m⁻² for the longwave in all-sky conditions. For the cloud radiative effect (CRE) biases, there is an overestimation of +4.8 ± 28.0 W m⁻² in ACCESS-AM2 and an underestimation of -7.9 ± 20.9 W m⁻² in CERES. An overestimation of downwelling surface shortwave radiation is associated with an underestimation of cloud fraction. The associated biases in cloud occurrence are less clear and we suggest that modelled cloud phase is also having an impact on the radiation biases. Our results show that the ACCESS-AM2 model and CERES product require further development to reduce these radiation biases, not just in shortwave and in all-sky conditions, but also in longwave and in clear-sky conditions.