Effective LSTMs with seasonal-trend decomposition and adaptive learning and niching-based backtracking search algorithm for time series forecasting

Wu, Yuhan; Meng, Xiyu; Zhang, Junru; He, Yang; Romo, Joseph A.; Dong, Yabo; Lu, Dongming

doi:10.1016/j.eswa.2023.121202

Cited by 13 publications

(1 citation statement)

References 81 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…With a memory cell capable of storing and retrieving information over extended periods, LSTMs can crucially remember features and patterns from earlier time steps, enhancing the understanding of climate variable dynamics. Their ability to handle diverse input types simultaneously is vital for climate prediction, where various meteorological variables contribute to the overall system (Wu et al, 2024). LSTMs are robust against noise and uncertainties in climate data, contributing to improved prediction accuracy.…”

Section: Discussionmentioning

confidence: 99%

Predicting climate change using an autoregressive long short-term memory model

Chin,

Lloyd

2024

Front. Environ. Sci.

View full text Add to dashboard Cite

Climate change is a pressing global issue. Mathematical models and global climate models have traditionally been invaluable tools in understanding the Earth’s climate system, however there are several limitations. Researchers are increasingly integrating machine learning techniques into environmental science related to time-series data; however, its application in the context of climate predictions remains open. This study develops a baseline machine learning model based on an autoregressive recurrent neural network with a long short-term memory implementation to predict the climate. The data were retrieved from the ensemble-mean version of the ERA5 dataset. The model developed in this study could predict the general trends of the Earth when used to predict both the climate and weather. When predicting climate, the model could achieve reasonable accuracy for a long period, with the ability to predict seasonal patterns, which is a feature that other researchers could not achieve with the complex reanalysis data utilized in this study. This study demonstrates that machine learning models can be utilized in a climate forecasting approach as a viable alternative to mathematical models and can be utilized to supplement current work that is mostly successful in short-term predictions.

show abstract

Section: Discussionmentioning

confidence: 99%