Bibliographic citations
Gomez, K., (2023). Pronóstico enos en las regiones Niño 3.4 y Niño 1+2, utilizando redes neuronales profundas con secuencias espacio temporales [Universidad Nacional Agraria La Molina]. https://hdl.handle.net/20.500.12996/5619
Gomez, K., Pronóstico enos en las regiones Niño 3.4 y Niño 1+2, utilizando redes neuronales profundas con secuencias espacio temporales []. PE: Universidad Nacional Agraria La Molina; 2023. https://hdl.handle.net/20.500.12996/5619
@mastersthesis{renati/247072,
title = "Pronóstico enos en las regiones Niño 3.4 y Niño 1+2, utilizando redes neuronales profundas con secuencias espacio temporales",
author = "Gomez Tunque, Kennedy Richard",
publisher = "Universidad Nacional Agraria La Molina",
year = "2023"
}
El Niño-Southern Oscillation (ENSO) forecasting is one of the most discussed and challenging tasks in ocean and atmospheric sciences, since the observed ENSO events show large differences between events, in their amplitude, temporal evolution and spatial pattern. The objective of this research is to propose a deep neural network model that integrates the convolutional structure of short and long term memory (ConvLSTM) to forecast sea surface temperature (SST) patterns that represent the evolution of ENSO in different horizons simultaneously in time and space in the Niño 3.4 and Niño 1+2 regions. For this purpose, gridded SST data from the equatorial Pacific Ocean (latitude 10°S–10°N and longitude 140°E–80°W) for the period 1854-2022 at the monthly level were used. This information was obtained from the Oceanic and Atmospheric Administration (NOAA). In summary, the methodology includes the normalization of the tropical Pacific SST, the creation and selection of the deep neural network model architecture, the training, validation, and finally, the forecast and evaluation (test) of the SST anomaly spatiotemporal of ENSO in the Niño 3.4 and Niño 1+2 regions six months in advance. The results indicate a good performance of the model for El Niño 1982/83 (training) and El Niño 1997/98 (validation) in space and time, in the test stage it was forecast spatially and temporally for El Niño 2015/16, El Niño Costero 2017 and La Nina 2022. This performance was evaluated using statistical metrics from observed and forecast data over six months in the equatorial Pacific Ocean and subsequently focusing on the Niño 3.4 and Niño 1+2 region, resulting in average statistical metrics for El Niño 2015/16 (Niño 3.4) of RMSPE at 0.35 percent and MAPE at 0.30 percent, El Niño Costero 2017 (Niño 1+2) of RMSPE at 2.89 percent and MAPE at 2.01 percent and La Niña 2022 (Niño 3.4) of RMSPE at 1.24 percent and MAPE at 0.83 percent. Likewise, based on space-time forecasts, it was compared with climatic indices of the Niño 3.4 and Niño 1+2 regions with global dynamic and statistical models. Therefore, it is concluded that a better forecast is obtained in the Niño 3.4 region than in the Niño 1+2 region; with very good results in the first three months, reducing its forecasting efficiency as the forecast month increases.
This item is licensed under a Creative Commons License