Machine Learning Techniques for Determining Cloud Phase Using Backscatter Lidar: Observations from IMPACTS and Implications for AOS

[09-Jan-2023] Abstract  Clouds play important roles in Earth's climate system and that role is heavily dependent on the height, thickness, and particle properties of clouds in the atmosphere. Liquid water clouds closer to the Earth's surface tend to reflect incoming sunlight, cooling Earth's surface. However, ice clouds can absorb heat emitted from the surface and re-radiate it back down, warming Earth's surface. Thus, determining cloud phase (liquid or ice) is important to understanding cloud impacts on the radiation budget. Cloud-aerosol lidars are important tools for estimating cloud height, thickness, and cloud phase. The Cloud Physics Lidar (CPL) is a multi-wavelength (355, 532, 1064 nm) elastic backscatter lidar built for use on the high-altitude ER-2 aircraft. CPL fundamentally measures the total (aerosol plus Rayleigh) attenuated backscatter as a function of altitude (30 m vertical resolution) at each wavelength. Additional cloud properties provided by CPL data include the volume depolarization ratio, extinction coefficient, and optical depth. CPL has participated in over two dozen field campaigns since 2000. One such field campaign is the Investigation of Microphysics and Precipitation in Atlantic Coast Threatening Snowstorms (IMPACTS). IMPACTS consists of three 6-week deployments (2020, 2022, 2023) utilizing a complementary suite of remote-sensors on the ER-2 and in-situ instruments on the NASA P-3 (within the clouds). Traditionally, lidar cloud phase algorithms rely on hard temperature and depolarization ratio thresholds or statistical relationships between attenuated backscatter and depolarization ratio. In this talk, we will present a machine learning technique that determines cloud phase from CPL data during the IMPACTS deployments. Results will be verified using the collocated in situ cloud measurements from the P-3 aircraft and preliminary attempts to estimate supercooled liquid water using CPL will be presented. These machine learning techniques could be applicable to future space-based lidar sensors, such as the lidars as part of the NASA Earth System Observatory (ESO) Atmosphere Observing System (AOS).