Applications of an Artificial Neural Network in Blowing Snow Detection from Depolarized Lidar Data

Blowing snow (BLSN) is a common hazard in many parts of the world, including the Northern Plains of the United States, with various societal and environmental impacts. To aid in nowcasting hazards from BLSN and to prepare for efforts to verify BLSN numerical weather models, a BLSN detection algorithm consisting of an artificial neural network (ANN) has been developed to identify layers of BLSN in ceilometer and lidar data. Input variables include normalized backscatter and linear depolarization ratio. Using normalized backscatter, the algorithm becomes agnostic, removing the necessity to re-train the algorithm for new lidar datasets that have varying units and intensities of attenuated backscatter. Training data consists of four instruments: Lufft CHM-15K and Vaisala CL-61 ceilometers located in Emerado, North Dakota, and a Vaisala CL-31 and Micropulse Lidar (MPL) from the Multidisciplinary drifting Observatory for the Study of Arctic Climate (MOSAiC) campaign in 2019-2020. Preliminary results on CL-61 data in North Dakota, and CL-31 and MPL data from both MOSAiC and McMurdo Station, Antarctica, show promise in successfully identifying BLSN. The ANN can struggle in complex environments with mixed precipitation and BLSN, scenes where lidars can suffer significant extinction. Provided that ground truth is unknown above the surface, future work involves collecting balloon-borne in situ data to verify algorithm output. The algorithm will also be applied to spaceborne lidar (e.g., CALIPSO) and polarimeter data to identify BLSN and other meteorological phenomena in remote parts of the world.