|
|||||
|
|
A novel deep neural network model approach to predict Indian Ocean dipole and Equatorial Indian Ocean oscillation indices |
|
| Institution: | 1. Ocean Analysis and Modelling Laboratory, School of Earth, Ocean and Climate Sciences, Indian Institute of Technology Bhubaneswar, Odisha, India;2. Ocean Observation Systems, National Institute of Ocean Technology, Chennai, India;1. St. Petersburg State University, 7–9, Universitetskaya nab., St. Petersburg 199034, Russia;2. NIERSC, Nansen International Environmental and Remote Sensing Centre, 14th line, 7, St. Petersburg 199034, Russia;3. Arctic and Antarctic Research Institute, Bering str., 38, St. Petersburg 199397, Russia;4. Pacific Oceanological Institute of the Russian Academy of Sciences, 43 Baltiiskaya St., 690041 Vladivostok, Russia;1. University of Monastir, Faculty of Sciences of Monastir (FSM), UR13ES64, Analysis and Control of PDEs, 5019 Monastir, Tunisia;2. University of Carthage, National Institute of Marine Sciences and Technologies (INSTM), LR16INSTM04, Marine Environment Laboratory, 2025 Salammbô, Tunisia;3. Environmental Modeling Center (EMC), National Oceanic and Atmospheric Administration, MD, USA;1. Department of Earth Sciences, Science and Research Branch, Islamic Azad University, Tehran, Iran;2. Department of Space Physics, Institute of Geophysics, University of Tehran, Iran |
| Abstract: | Indian Ocean Dipole (IOD) and Equatorial Indian Ocean oscillation (EQUINOO) are important climatic system oscillation events in the Indian Ocean region that affects the Indian summer monsoon rainfall (ISMR). The prime focus of this study is to deliberate the influence of these events on ISMR and an attempt has been made to predict these events for future time scales using a Long short term memory (LSTM) deep learning model. LSTM is a special kind of recurrent neural network (RNN) which specializes in learning long-term dependencies and extracting important features. The features learnt by the model is then ranked using correlational analysis (linear and nonlinear). This approach helps in selecting decisive and imperative set of relevant predictors, which can be employed to predict IOD and EQUINOO. Nonlinear correlational identified predictors are found to forecast with greater precision as to their linear counterparts. The model-calibrated correlation coefficient for IOD and for EQUNIOO was 0.90 and 0.88 respectively at a lead of 5 months. Our proposed model was observed to work at par with the other existing models in terms of various statistical evaluation measures. |
| Keywords: | Deep learning Equatorial Indian Ocean oscillation Feature extraction Indian Ocean dipole Modeling Prediction |
| 本文献已被 ScienceDirect 等数据库收录！ | |