Abstract Cold-air outbreaks over high latitude oceans typically include mixed-phase clouds and precipitation, in particular supercooled liquid clouds that support snow and graupel through ice growth processes. The partitioning of the total water into the liquid and ice phases impacts both weather and climate prediction, but accurate measurements on the phase partitioning remain difficult to acquire, especially near-real-time. Here we present a machine learning approach to retrieve liquid water path (LWP) using passive microwave measurements combined with vertically-integrated radar reflectivities. The approach is an extension of Cadeddu et al. (2009), with the novel addition of radar reflectivity. The machine learning models are trained using the Passive and Active Microwave radiative TRAnsfer (PAMTRA) code applied to output from numerical simulations of three independent cold-air outbreaks sampled during the Cold-Air Outbreaks in the Marine Boundary Layer Experiment (COMBLE) campaign. Brightness temperatures corresponding to the four sidebands of an upward-looking G-band (183 GHz) Vapor Radiometer, along with the vertically-integrated reflectivity from a zenith-pointing 95 GHz Wyoming Cloud Radar, are simulated from the perspective of a near-surface aircraft track. The radar reflectivity helps discriminate the snow contribution to the brightness temperatures. The machine learning models are thereafter tested on a simulation of an independent cold-air outbreak during COMBLE, and against measurements from the US Department of Energy Atmospheric Radiation Measurement North Slope of Alaska observatory. This machine learning approach is shown to provide robust, computationally-efficient, near-real-time measurements of LWP and water vapor path during the Cold Air Outbreak Experiment in the Sub-Arctic Region (CAESAR) campaign in February-April 2024.
Ephraim et al. (2025) studied this question.