An investigation for the deep-learning-based determination of convective boundary layer basing on coherent Doppler wind lidar

In this study, the estimation of Convective Boundary Layer Height (CBLH) using coherent Doppler lidar with a novel deep learning approach is presented. A modified stacked hourglass network, a convolutional neural network architecture is employed to automate the retrieval of CBLH from aerosol and wind products measured by the Doppler lidar. The model is trained using a comprehensive dataset collected over one year in central Taiwan, comprising over 30,000 lidar maps. Ground truth CBLH is determined from the variance of vertical velocity, and the dataset is divided into subsets to evaluate the minimum training requirements. The results demonstrate the effectiveness of the deep learning model in accurately predicting CBLH and the possibility of deriving CBLH from the aerosol backscatter profile.