Deep Kernel Dictionary Learning for detection of wave breaking features in Atmospheric Gravity Waves

Abstract

Atmospheric Gravity Waves play a significant role in the Middle Atmosphere Dynamics and breaking of Gravity Waves leads to turbulence. To confirm the breaking of Atmospheric Gravity Waves, we require Wind Velocity profiles with high accuracy and the problem in the existing scenario is the Wind velocity measuring instruments are not accurate at the altitude of interest and are very sparse. We came up with a solution by taking advantage of Dictionary Learning and Deep Learning approach for the detection of the Wave Breaking events from atmospheric temperature perturbations instead of looking for wind velocity profiles.In the present work, we discuss incorporating Kernels into Deep Dictionary Learning. The Deep Dictionary Learning algorithm, introduced recently, shows an improvement in feature detection in comparison to k-means Singular Value Decomposition (KSVD) Dictionary Learning, Convolution Neural Network, and LSTM Auto-Encoders. Incorporating Kernels into Deep Dictionary Learning increases the effectiveness of the detection of the Wave Breaking events. The potential of the proposed method is demonstrated with a case study on the detection of atmospheric Gravity Wave Breaking event leading to turbulence in the atmosphere using satellite data (Aura-Microwave Limb Sounder) and is validated using data available from the ground-based instruments.