Spaceborne lidar measurement of global cloud properties through machine learning

With a large footprint size, multiple scattering measurements of clouds from spaceborne lidar provide useful information about cloud physical properties, such as cloud optical depths and cloud droplet size, both during daytime and nighttime. A neural network algorithm, with a subset of cloud backscatter profiles of dual-polarization and dual-wavelength Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) lidar measurements during daytime as input variables and cloud physical properties derived from collocated Moderate Resolution Imaging Spectroradiometer (MODIS) multi-spectral measurements as output, is developed and evaluated with an independent subset of the collocated CALIPSO and MODIS measurements. The study suggests that with a receiver footprint size of 110 m, CALIPSO lidar measurements are sensitive to liquid-phase cloud optical depth variations from 0 to 25. A larger footprint size, thus more multiple scattering, is required for lidar to have sensitivities to all liquid-phase clouds. The technique can be applied to all 17 years of CALIPSO daytime and nighttime measurements and, thus, provides useful information about global distributions of cloud physical properties both during day and night.