the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 24 Nov 2023
Advances and Prospects of Deep Learning for Medium-Range Extreme Weather Forecasting
Abstract. In recent years, deep learning models have rapidly emerged as a standalone alternative to physics-based numerical models for medium-range weather forecasting. Several independent research groups claim to have developed deep learning weather forecasts which outperform those from state-of-the-art physics-basics models, and operational implementation of data-driven forecasts appears to be drawing near. Yet, questions remain about the capabilities of deep learning models to provide robust forecasts of extreme weather. This paper provides an overview of recent developments in the field of deep learning weather forecasts, and scrutinises the challenges that extreme weather events pose to leading deep learning models. Lastly, it argues for the need to tailor data-driven models to forecast extreme events, and proposes a foundational workflow to develop such models.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(917 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(917 KB) - Metadata XML
- BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-2490', Anonymous Referee #1, 31 Jan 2024
In this article, the authors provide an overview of recent developments in the field of deep-learning weather forecasts and point out the challenges that extreme weather events pose to leading deep-learning models. The authors identify two principal constraints hindering current state-of-the-art deep learning forecasts of extreme weather: suboptimal utilization of limited training samples for extreme values within existing architectures and simplistic assumptions regarding the distribution of forecasting errors for extreme events.
Furthermore, the absence of rigorous validation of extreme weather forecasts by leading global Deep Learning Weather Prediction (DLWP) models exacerbates these challenges.
The authors advocate for a targeted DLWP workflow tailored to extreme weather forecasts, wherein deep learning models specifically engineered to address extreme events should complement those maximizing average forecast skills. The authors recommend adapting existing deep learning architectures rather than pursuing entirely novel and untested methodologies. The authors emphasize that this endeavour should be augmented by prioritizing the evaluation of model performance within the tails of forecasted variable distributions.Aligned with the aforementioned recommendations, this article outlines a foundational workflow aimed at advancing deep learning-based extreme weather forecasts. The choice of methodology hinges on the meteorological inquiry—whether probabilistic or deterministic—and the return period associated with the extreme events under scrutiny. Leveraging recent architectural advancements in deep-learning weather forecast models, the proposed workflow envisions robust deep-learning forecasts of extreme weather becoming attainable in the foreseeable future.
I recommend the publication of the paper since it is clear, well-written, and provides a clear and thorough summary of the efforts that have been carried out in the field of Deep Learning for Medium-Range Extreme Weather Forecasting.
Citation: https://doi.org/10.5194/egusphere-2023-2490-RC1 -
AC1: 'Reply on RC1', Leonardo Olivetti, 06 Feb 2024
Dear Referee #1,
Thank you for your valuable feedback and positive remarks. We appreciate the time you spent reviewing our work!
Citation: https://doi.org/10.5194/egusphere-2023-2490-AC1
-
AC1: 'Reply on RC1', Leonardo Olivetti, 06 Feb 2024
-
RC2: 'Comment on egusphere-2023-2490', Anonymous Referee #2, 04 Feb 2024
This review paper highlights the need for deep learning weather prediction models to improve their extreme event prediction skill and provides a roadmap to do so. This is a timely and important work that will be of great interest to those in meteorology and climate. It is well-written and referenced. I highly recommend that it be published in GMD.
Citation: https://doi.org/10.5194/egusphere-2023-2490-RC2 -
AC2: 'Reply on RC2', Leonardo Olivetti, 06 Feb 2024
Dear Referee #2,
Thank you for your valuable feedback and positive remarks. We appreciate the time you spent reviewing our work!
Citation: https://doi.org/10.5194/egusphere-2023-2490-AC2
-
AC2: 'Reply on RC2', Leonardo Olivetti, 06 Feb 2024
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-2490', Anonymous Referee #1, 31 Jan 2024
In this article, the authors provide an overview of recent developments in the field of deep-learning weather forecasts and point out the challenges that extreme weather events pose to leading deep-learning models. The authors identify two principal constraints hindering current state-of-the-art deep learning forecasts of extreme weather: suboptimal utilization of limited training samples for extreme values within existing architectures and simplistic assumptions regarding the distribution of forecasting errors for extreme events.
Furthermore, the absence of rigorous validation of extreme weather forecasts by leading global Deep Learning Weather Prediction (DLWP) models exacerbates these challenges.
The authors advocate for a targeted DLWP workflow tailored to extreme weather forecasts, wherein deep learning models specifically engineered to address extreme events should complement those maximizing average forecast skills. The authors recommend adapting existing deep learning architectures rather than pursuing entirely novel and untested methodologies. The authors emphasize that this endeavour should be augmented by prioritizing the evaluation of model performance within the tails of forecasted variable distributions.Aligned with the aforementioned recommendations, this article outlines a foundational workflow aimed at advancing deep learning-based extreme weather forecasts. The choice of methodology hinges on the meteorological inquiry—whether probabilistic or deterministic—and the return period associated with the extreme events under scrutiny. Leveraging recent architectural advancements in deep-learning weather forecast models, the proposed workflow envisions robust deep-learning forecasts of extreme weather becoming attainable in the foreseeable future.
I recommend the publication of the paper since it is clear, well-written, and provides a clear and thorough summary of the efforts that have been carried out in the field of Deep Learning for Medium-Range Extreme Weather Forecasting.
Citation: https://doi.org/10.5194/egusphere-2023-2490-RC1 -
AC1: 'Reply on RC1', Leonardo Olivetti, 06 Feb 2024
Dear Referee #1,
Thank you for your valuable feedback and positive remarks. We appreciate the time you spent reviewing our work!
Citation: https://doi.org/10.5194/egusphere-2023-2490-AC1
-
AC1: 'Reply on RC1', Leonardo Olivetti, 06 Feb 2024
-
RC2: 'Comment on egusphere-2023-2490', Anonymous Referee #2, 04 Feb 2024
This review paper highlights the need for deep learning weather prediction models to improve their extreme event prediction skill and provides a roadmap to do so. This is a timely and important work that will be of great interest to those in meteorology and climate. It is well-written and referenced. I highly recommend that it be published in GMD.
Citation: https://doi.org/10.5194/egusphere-2023-2490-RC2 -
AC2: 'Reply on RC2', Leonardo Olivetti, 06 Feb 2024
Dear Referee #2,
Thank you for your valuable feedback and positive remarks. We appreciate the time you spent reviewing our work!
Citation: https://doi.org/10.5194/egusphere-2023-2490-AC2
-
AC2: 'Reply on RC2', Leonardo Olivetti, 06 Feb 2024
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 339 | 154 | 15 | 508 | 6 | 8 |
- HTML: 339
- PDF: 154
- XML: 15
- Total: 508
- BibTeX: 6
- EndNote: 8
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Leonardo Olivetti
Gabriele Messori
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(917 KB) - Metadata XML