Machine learning-enabled real-time detection of cloud and aerosol layers using airborne lidar
Matthew J. McGill, Patrick Selmer, Andrew Kupchock, John E. Yorks
Abstract
Lidar profiling of the atmosphere provides information on existence of cloud and aerosol layers and the height and structure of those layers. Knowledge of feature boundaries is a key input to assimilation models. Moreover, identifying feature boundaries with minimal latency is essential to impact operational assimilation and real-time decision making. Using advanced convolution neural network algorithms, we demonstrate real-time determination of atmospheric feature boundaries using an airborne backscatter lidar. Results are shown to agree well with traditional processing methods and are produced with higher horizontal resolution than the traditional method. Demonstrated using airborne lidar, the algorithms and process are extendable to real-time generation of data products from a future spaceborne sensor.