Back to Search
Start Over
ILF-LSTM: enhanced loss function in LSTM to predict the sea surface temperature.
- Source :
-
Soft Computing - A Fusion of Foundations, Methodologies & Applications . Sep2023, Vol. 27 Issue 18, p13129-13141. 13p. - Publication Year :
- 2023
-
Abstract
- Globe's primary issue is global warming, water temperatures have accompanied it as the sea surface temperature, and it is the primary attribute to balance the energy on the earth's surface. Sea surface temperature prediction is vital to climate forecast. Downwelling currents carry some of this heat to the ocean's bottom layers, which are also heating, covering far behind the increase in sea surface temperature. In deep learning models, the correct loss function will try to reduce the error and converge fast. The proposed improved loss function correctly estimates how close the predictions made by the long short-term memory match the observed values in the training data. This research considers location-specific sea surface temperature predictions using the improved loss function in the long short-term memory neural network at six different locations around India for daily, weekly, and monthly time horizons. Most existing research concentrated on periodic forecasts, but this paper focused on daily, weekly, and monthly predictions. The improved loss function—long short-term memory, achieved 98.7% accuracy, and this improved loss function overcomes the limitations of the existing techniques and reduces the processing time to ~ 0.35 s. In this research, the sea surface temperature prediction using the improved loss function in the long short-term memory neural network gives better results than the standard prediction models and other existing techniques by considering the long-time dependencies and obtaining features from the spatial data. [ABSTRACT FROM AUTHOR]
Details
- Language :
- English
- ISSN :
- 14327643
- Volume :
- 27
- Issue :
- 18
- Database :
- Academic Search Index
- Journal :
- Soft Computing - A Fusion of Foundations, Methodologies & Applications
- Publication Type :
- Academic Journal
- Accession number :
- 167308064
- Full Text :
- https://doi.org/10.1007/s00500-022-06899-y