A 40-year Climatology and New Sub-Millimeter Retrievals: Two Novel Datasets for Observational Constraints on Ice Cloud Mass

Constraining frozen cloud particles remains a key challenge for improving global climate models. Current estimates of atmospheric ice mass have significant limitations. The spaceborne radar-lidar missions CloudSat-CALIPSO and EarthCARE offer high-quality data but with sparse sampling and limited mission duration. Passive satellite products provide better spatiotemporal coverage but have traditionally exhibited strong biases compared to CloudSat-based measurements. These observational gaps limit our ability to evaluate and validate simulations of ice clouds.

We present two complementary datasets to address this challenge: the Chalmers Cloud Ice Climatology (CCIC) and the Chalmers Hydrometeor Inversion Product from the Arctic Weather Satellite (CHIP-AWS). Both datasets provide a number of quantities; here we focus on vertically integrated atmospheric ice mass: frozen water path (FWP). They provide estimates with regular global coverage between ±60° latitude and are accompanied by per-retrieval uncertainty. Though both use neural networks, they have contrasting training approaches: CCIC employs empirical training on CloudSat-retrieved data, while CHIP-AWS uses physics-based radiative transfer simulations. For average values, both datasets agree with CloudSat-based retrievals.

CCIC provides quasi-global coverage of FWP estimates at high temporal resolution. The inputs are geostationary infrared images to a neural network model trained on 3.5 years of CloudSat-CALIPSO data. Once trained, the model can be applied to archived and future imagery. Two variants are available: a 0.07°/3-hour product spanning 1980-present and a higher resolution 0.036°/30-minute product spanning 2000-present. These 40+/20+ year climatologies enable analysis of both long-term trends and diurnal variations in ice cloud properties and have been applied to evaluate global storm-resolving models and identify regional trends.

CHIP-AWS uses novel sub-mm passive microwave radiances from the polar-orbiting Arctic Weather Satellite (launched 2024), providing more direct sensitivity to ice mass compared to previous passive instruments. The retrieval model is trained on a database of radiative transfer simulations that use defined particle models and scattering data. This approach allows assessment of the underlying microphysical assumptions. The dataset covers 2025 and onward with an 800 km swath and ~10 km nadir resolution. This provides high spatial coverage compared to a satellite cloud radar but low compared to CCIC. On the other hand, CHIP-AWS offers higher spatial resolution and much higher accuracy at local scales.

Together, the different strengths of these datasets provide observational constraints for evaluating and improving ice cloud processes in climate models across scales from individual cloud systems to multi-decadal trends.