Application of deep learning to estimate stratospheric gravity wave potential energy

One of the most important dynamic processes in middle and upper atmosphere, gravity waves (GWs) play a key role determining global atmospheric circulation. Gravity wave potential energy (GW Ep) is an parameter that characterizes GW intensity, so it critical to understand its distribution. In this paper, deep learning algorithm (DeepLab V3+) used estimate stratospheric Ep. The model inputs are ERA5 reanalysis datasets GMTED2010 terrain data. Ep averaged over 20−30 km from 60°S−60°N, calculated by COSMIC radio occultation (RO) data, as measured value corresponding output. results show (1) method can effectively zonal trend However, errors between estimated larger low-latitude regions than mid-latitude regions, possibly due large number convolution operations model. Additionally, has associated with interpolation grid; tends be amplified because RO data relatively sparse, affecting training accuracy. (2) shows seasonal variations, which stronger winter hemisphere weaker summer hemisphere. (3) effect quasi-biennial oscillation (QBO) clearly observed monthly variation Ep, QBO amplitude may less measured Ep.