the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 24 Jun 2025
Enhancing Extended Weather Forecasts in the TCWAGFS Model Using Deep Learning Method for SST Bias Correction
Abstract. The extended weather (> 7 days) and the seasonal climate predictions are highly dependent on the status of Madden Julian Oscillation (MJO) and El Niño-Southern Oscillation (ENSO). Both the evolutions of MJO and ENSO are found to be correlated to the anomalies of the global sea surface temperature (SST). To decrease the predicting SST bias (BiasSST) in a coupled ocean-atmosphere global model, we evaluate nine well-developed machine learning algorithms. By using the Bi-directional Long Short-Term Memory (Bi-LSTM) algorithm, it is found the bias can be reduced significantly. For example, the Root Mean Squared Error on Day 10 forecast (denoted as D10) is reduced to 0.01 K by the Bi-LSTM algorithm while the original bias is 0.38 K by the Taiwan Central Weather Administration Global Forecast System (TCWAGFS), of which the error is reduced by 97 %.
- Preprint
(3554 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2025-142', Anonymous Referee #1, 21 Jul 2025
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-142/egusphere-2025-142-RC1-supplement.pdfCitation: https://doi.org/
10.5194/egusphere-2025-142-RC1 -
CEC1: 'Comment on egusphere-2025-142', Astrid Kerkweg, 22 Jul 2025
Dear authors,
please note that GMD's policy is to ensure that the results published in an article are fully reproducible.
To achieve this goal, the information about the input data used to feed the ML algorithm is still missing in your Code / Data availability section.
In order to fulfill all necessary requirements for publication, please provide this information as soon as possible (here in the discussion) and, upon revision, in the revised paper.
Best regards,
Astrid Kerkweg (GMD Executive Editor)
Citation: https://doi.org/10.5194/egusphere-2025-142-CEC1 -
RC2: 'Comment on egusphere-2025-142', Anonymous Referee #2, 26 Mar 2026
This study evaluates the performance of nine machine learning algorithms for Sea Surface Temperature (SST) prediction, concluding that the Bi-LSTM model achieves the best performance. However, while the comparison is comprehensive, the manuscript in its current form lacks sufficient innovation and contains several methodological and presentation weaknesses. I do not believe the manuscript is suitable for publication based on the following concerns.
- Innovation and Novelty
The primary concern regarding this manuscript is the limited scientific innovation.
- Redundancy of Research: The application of Bi-LSTM for SST forecasting has been extensively studied in recent years. This manuscript appears to be a repetition of existing work without providing new insights or methodological breakthroughs. Specifically, several published works have already established the efficacy of Bi-LSTM and deep learning in this domain, such as:
Xiao, C., Chen, N., Hu, C., Wang, K., Xu, Z., Cai, Y., ... & Gong, J. (2019). A spatiotemporal deep learning model for sea surface temperature field prediction using time-series satellite data. Environmental Modelling & Software, 120, 104502.
Zrira, N., Kamal-Idrissi, A., Farssi, R., & Khan, H. A. (2024). Time series prediction of sea surface temperature based on BiLSTM model with attention mechanism. Journal of Sea Research, 198, 102472.
Xu, T., Zhou, Z., Li, Y., Wang, C., Liu, Y., & Rong, T. (2023). Short-term prediction of global sea surface temperature using deep learning networks. Journal of Marine Science and Engineering, 11(7), 1352.
The authors claim that ‘Bi-LSTM is a powerful tool for improving the accuracy of SST bias correction in the TCWAGFS model.’ However, the study does not actually implement or demonstrate a bias correction framework. It merely compares the errors between Bi-LSTM and TCWAGFS. The title and the stated objectives are therefore misleading, as the actual work does not reflect the promised ‘bias correction’ application.
- Selection of Research Areas
The authors selected five major areas, but failed to provide the specific geographic coordinates (latitude/longitude degrees) for these areas. What’s more, the ‘major’ areas do not necessarily equate to regions with the highest prediction errors. The authors should clarify the logic behind this selection. The results show that Bi-LSTM provides the most significant improvement in SST bias prediction in the Southern Ocean, which was not emphasized as a primary focus area. This discrepancy between the selection criteria and the results needs to be addressed.
- Visualization and Presentation
The figures are repetitive and lack diversity. Almost all figures in the manuscript utilize the same chart type. This makes the manuscript feel monotonous and hinders readability. The authors should employ a variety of figures to better illustrate their findings. For instance, time-series line graphs could be used to show SST prediction performance over different lead times in the key study regions, which would provide more intuitive insights than static error comparisons.
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 1,998 | 131 | 61 | 2,190 | 54 | 57 |
- HTML: 1,998
- PDF: 131
- XML: 61
- Total: 2,190
- BibTeX: 54
- EndNote: 57
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1