Advancements and continued challenges in global modelling and observations of atmospheric ice masses

We assess the current status of atmospheric ice mass estimates by comparing global circulation models and global storm-resolving models with satellite observations. The analysis focuses on the frozen water path, which offers a more consistent measure across modelling and observational datasets than cloud ice or other partial quantities. As a reference, we use three retrievals derived from the CloudSat mission. Despite being based on the same input data, these retrievals exhibit a significant spread, and we estimate the uncertainty in overall means to be as high as 40 %. A recently developed machine learning product based on passive observations highly extends spatial and temporal coverage for comparisons, but its accuracy is limited by biases inherited from its training dataset. The latest generation of global circulation models systematically underestimates frozen water paths compared to the observational benchmark. While the spread among models has narrowed relative to earlier assessments, they still fail to provide consistent representations of regional temporal changes or the annual cycle. Storm-resolving models, which operate at finer grid spacing and resolve convective dynamics explicitly, show a better representation of total ice masses, with a variation among them that's similar to the observational uncertainty. However, several issues were noted, such as apparent deviations from the observations in the spatial structures of tropical deep convection, and that they differ significantly in their relative amounts of cloud ice, snow, and graupel. Together, these findings reveal progress but highlight continuing uncertainties that limit confidence in projections of cloud-related climate feedbacks.