Jurnal Internasional Menganalisis PREDIKTABILITAS OSCILLASI EL NINO SELATAN MENGGUNAKAN MODEL JANGKA PANJANG-JANGKA PENDEK – Huang – – Ilmu Bumi dan Luar Angkasa
El Niño Southern Southern Oscillation (ENSO) can have a global impact, affecting daily temperature and rainfall, and extreme weather, such as storms and tornadoes. Because of its importance, scientists are trying to understand the processes that govern ENSO and develop models to predict evolution and change their variability. Here, long-term long-term memory (LSTM) models are compared with linear regression models (LR) to explore the benefits of neural networks that are simple and deep in predicting ENSO, in addition to measuring the relative importance of ENSO predictability sources. The model uses central Pacific sea surface temperature (SST), equatorial Pacific warm water volumes, and western Pacific zone winds as predictors, individually and in combination, at monthly and daily resolutions, from leads to one to eleven months. Using these predictors, many characteristic time scales include – from day to week in the atmosphere, to month-to-season in pairing systems, and between seasons – inter-years in the subsurface ocean. The results show, with monthly input, predictions from LSTM are like predictions from LR. However, with daily SST on older leads, LSTM shows several advantages over LR in terms of correlation coefficients. This shows that daily SST can contain several non-linear elements that increase LSTM predictability compared to LR. In addition, this shows more information, such as data grids, additional variables, etc. It is likely to increase predictability using LSTM, but the results will be more difficult to interpret. Overall, LSTM may be interesting because after expensive LSTM training is computationally completed, predictions that use a trained model can be relatively inexpensive to do afterwards.