Deep Learning Model for Global Spatio-Temporal Image Prediction

Mathematical methods are the basis of most models that describe natural phenomena around us. However, well-known conventional mathematical for atmospheric modeling have some limitations. Machine learning with Big Data is also based on mathematics but offers a new approach modeling. There two methodologies to develop deep spatio-temporal image prediction. On these bases, were built—ConvLSTM and CNN-LSTM—with types predictions, i.e., sequence-to-sequence sequence-to-one, in order forecast Aerosol Optical Thickness sequences. The input dataset training was NASA satellite imagery MODAL2_E_AER_OD from Terra/MODIS satellites, which presents global an 8 day temporal resolution 2000 present. obtained results show ConvLSTM sequence-to-one model had lowest RMSE error highest Cosine Similarity value. advantages developed DL they can be executed milliseconds PC, used global-scale Earth observations, serve as tracers study how Earth’s atmosphere moves. transfer similar time-series forecasting models.