Climate Time Series Prediction with Deep Learning and LSTM

Authors:

Abstract
Climate time series prediction is a critical task for understanding and forecasting complex climate patterns and trends. Traditional methods have shown limited accuracy and struggle to capture the intricacies of climate data. In this study, a deep learning-based approach is proposed to improve the accuracy of climate time series prediction and address the challenges posed by previous works. The main objective of this work is to train Deep Learning models from collected data to generate climate predictors. For this public data on climate indicators is collected; Deep Learning models are trained; climate forecasts from the models trained from the collected data is carried out. The innovation in proposed approach lies in the integration of a novel temporal attention mechanism that allows the model to focus on relevant time steps, improving interpretability and prediction accuracy. Furthermore, transfer learning is adopted with pretrained models to capitalize on their knowledge and accelerate convergence on climate-specific tasks. The comparison with most recent related work showcases the superiority of the proposed approach in capturing complex climate patterns and delivering precise long-term forecasts. The model's interpretability enables researchers to gain valuable insights into the underlying climate dynamics, facilitating informed decision-making and policy planning.

This publication has 10 references indexed in Scilit: