Assessing Clouds Using Satellite Observations Through Three Generations of Global Atmosphere Models
Multiple satellite products exist that report cloud cover and other cloud properties over the past two decades or more. Satellite simulators provide a forward calculation that can translate a climate model’s simulation into an estimate of what a satellite would observe in the model. The Cloud Feedback Model Intercomparison Observation Simulator Package (COSP) provides a set of satellite simulators that have been incorporated into climate models over the past decade. Using these long observational records and climate model simulation that include COSP output, this study provides a detailed evaluation of the cloud climatology through three generations of global atmosphere models that share a lineage: CAM4, CAM5, CAM6, and E3SMv1.
To provide a fair comparison across the models, simulations with prescribed sea-surface temperature and sea ice based on observations are used. Spatial errors in cloud radiative effect and cloud cover are quantified using the Normalized Mean Squared Error, and that metric is decomposed to expose the contributions from mean bias versus errors in the pattern. Errors in cloud type are quantified by applying the Earth Mover’s Distance to cloud-top pressure versus cloud optical thickness histograms; this is a novel method for evaluating these histograms that incorporates the two-dimensional information into a single scalar value. These errors are evaluated within the context of the large-scale circulation by conditional sampling based on the vertical velocity.
There is an unambiguous improvement in clouds from the oldest model (CAM4) to the more recent ones. This improvement is seen in the cloud radiative effect and cloud cover and is present across dynamical regimes. In the more recent models (CAM5, CAM6, and E3SMv1), the distinctions are less clear-cut. E3SMv1 tends to show the smallest errors by most measures but has slightly larger shortwave errors (depending on spatial sampling). All the models show too little cloud cover over most of the tropics, and both CAM6 and E3SMv1 show too much cloud cover over the Southern Ocean. These biases are partially compensated by biases in cloud optical properties.