Impacts of tropical forecast errors on two extreme precipitation events: insights from relaxation experiments using machine-learning weather prediction models

Li, Siyu; Dias, Juliana; Moore, Benjamin; Quinting, Julian

doi:10.5194/wcd-7-787-2026

Articles | Volume 7, issue 2

https://doi.org/10.5194/wcd-7-787-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/wcd-7-787-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 7, issue 2

Research article

|

19 May 2026

Research article |

| 19 May 2026

Impacts of tropical forecast errors on two extreme precipitation events: insights from relaxation experiments using machine-learning weather prediction models

Siyu Li, Juliana Dias, Benjamin Moore, and Julian Quinting

Data sets

ERA5 hourly data on single levels from 1940 to present Copernicus Climate Change Service https://doi.org/10.24381/cds.adbb2d47

Model code and software

google-research/weatherbench2: v0.2.0 S. Rasp et al. https://doi.org/10.5281/zenodo.11376271

stefantulich/ufs-weather-model-hr1: UFS-HR1-MOORE.etal.2025 (HR1.MOORE.2025) S. Trahan et al. https://doi.org/10.5281/zenodo.17109574

Short summary

Weather forecasts weeks to months ahead, called subseasonal forecasts, help communities prepare for floods or droughts but are hard to make accurately. We tested a method called relaxation, which nudges parts of a model to see how different regions affect predictions. Using two machine learning models and a traditional model, we found the machine learning models performed better. Relaxation offers a simple, low-cost way to improve future forecasts.