On Lower-Tropospheric Arctic Ice Particle Properties Retrieved Using Machine Learning Applications to Remote Sensing Data, and their representation in Global Model Simulations

Ice particles and their properties (shape, size, etc.) have great potential for influencing cloud lifecycles, from their formation through their growth, and precipitation. Coupled with liquid-phase hydrometeors and associated processes in mixed-phase clouds, ice processes can be critical for understanding the extent of aerosol-cloud interactions and their causal links, which require model simulations across scales.  Robust estimates of observed ice properties, therefore, can support the evaluation and rectification of model physics, which could ultimately increase model fidelity. However, the entanglement of various parameterized processes that affect modeled cloud characteristics poses challenges for direct comparisons of model state variables with cloud observations. The use of consistent cloud process metrics can help address some of these difficulties and enable direct comparisons between observations and models. In the case of Arctic mixed-phase clouds, the representation of ice precipitation rates at and below cloud base, which serve as a key cloud water sink, could impact simulated cloud lifecycles and alter cloud feedbacks. These metrics can be robustly retrieved from ground-based measurements, which are considerably less susceptible to uncertainties associated with accurately locating cloud base, tropospheric profiling limitations, and spatial footprint size.

Here, we describe the main approaches of ice property retrievals. We then focus on cloud-base ice particle property retrievals using a Markov Chain Monte Carlo (MCMC) algorithm applied to radar and high-spectral-resolution lidar observations from the U.S. Department of Energy Atmospheric Radiation Measurement (ARM) User Facility site at the North Slope of Alaska (NSA). We describe the analysis results, including a qualitative distinction between ice-number- and size-dominated reflectivity regimes, number-concentration enhancements in certain temperature ranges, and a weak, typical Arctic cloud-base vertical motion. After examining insights derived from those retrievals, they are used in a brief bulk evaluation of mixed-phase cloud ice representation in three different global models: the NASA Goddard Institute for Space Studies ModelE3, the NCAR Community Earth System Model Version 2 (CESM2), and the DOE Energy Exascale Earth System Model Version 1 (E3SMv1). To perform a robust evaluation of Arctic cloud precipitation rates against observations, we process regional model output using the Earth Model Column Collaboratory (EMC²) instrument simulator and subcolumn generator, and compare them with corresponding cloud-base precipitation statistics calculated from the long-term ground-based remote-sensing dataset collected at the ARM NSA site. This brief analysis demonstrates key differences between the models and examines the agreement between model output and observations. Finally, we describe our current effort to generate sub-mixed-phase cloud ice precipitation profiles using a deep neural network emulator of the computationally intensive MCMC algorithm and discuss future plans.