Improvement of Maximum Air Temperature Forecasts Using a Stacking Ensemble Technique

Due to the influence of complex factors such as atmospheric dynamic processes, physical processes and local topography geomorphology, prediction near-surface meteorological elements in numerical weather model often has deviation. The deep learning neural networks are more flexible but with high variance. Here, we proposed a stacking ensemble named FLT, which consists fully connected network embedded layers (ED-FCNN), long short-term memory (LSTM) temporal convolutional (TCN) overcome variance single improve maximum air temperature. case study daily temperature forecast evaluated observation almost 2400 stations shows substantial improvement over that model, ECMWF-IFS statistical post-processing model. FLT can effectively bias than any above RMSE reduced by 52.36% accuracy increased 43.12% compared average RMSEs decreases 8.39%, 1.50%, 2.96% 16.03%, respectively, ED-FCNN, LSTM, TCN decaying method.