Enhancing Clear-Air Turbulence Forecasts: A WTCNN-LSTM Framework Utilising Energy Dissipation Rate

Current Clear-Air Turbulence (CAT) forecast diagnostics from Numerical Weather Prediction (NWP) models are generated by estimating horizontal and vertical gradients of atmospheric parameters and are generally categorised into a few levels of turbulence intensity. This qualitativeevaluation approach mostly relies on the aircraft and focuses on the local area or a single air route. It is therefore significant to propose a quantitative prediction framework using a common indicator, the energy dissipation rate (EDR), across multiple pressure levels, over a large horizontal area, and over an extended time period. A CAT dataset of North America is built by collecting ECMWF Reanalysis v5 (ERA5) and Pilot Reports (PIREPs) data and converting them to typical CAT diagnostics with their corresponding EDR, plus the basic temporal and spatial information. Based on the dataset, a significant variation in EDR regarding pressure levels and different times is investigated. The prediction frameworkWavelet Transform Convolutional Neural Network Long Short-Term Memory (WTCNN-LSTM) is proposed toprovide a final forecast result with the correction of the real in situ EDR records. Compared to the cutting-edge frameworks, the proposed model demonstrates superior performance in the global region of the dataset.