EMPIRIC_TC: A Site‐Specific AI‐Based Emulator for Tropical Cyclone Hazards Using Fourier Neural Operators.
Saved in:
| Title: | EMPIRIC_TC: A Site‐Specific AI‐Based Emulator for Tropical Cyclone Hazards Using Fourier Neural Operators. |
|---|---|
| Authors: | Winkelman, Eli1 (AUTHOR), McCrystall, Michelle2 (AUTHOR), Forbes, Callum3 (AUTHOR), Dresdner, Gideon4,5 (AUTHOR), Kafoa, Berlin6 (AUTHOR), Natuzzi, Eileen7 (AUTHOR), Taylor, Subhashni8 (AUTHOR), McLeod, Elizabeth9 (AUTHOR), Horvat, Christopher1 (AUTHOR) christopher_horvat@brown.edu |
| Source: | Journal of Advances in Modeling Earth Systems. Apr2026, Vol. 18 Issue 4, p1-21. 21p. |
| Subject Terms: | *Islands, *Climate change, Tropical cyclones, Deep learning, Emulation software, Cyclone forecasting, Artificial intelligence |
| Geographic Terms: | South Pacific Ocean, Islands of the Pacific |
| Abstract: | Tropical Cyclones (TCs) pose significant and evolving threats to Pacific Island Countries and Territories (PICTs) healthcare systems. Changes in cyclone intensity, frequency, and spatial distribution due to anthropogenic climate change are likely to increase these threats. Accessible information on future uncertainty of high intensity TCs can enable adaptation that reduces the severity of systemic disruptions. However, the infrequency of high‐intensity TC events in historical and GCM data and the small and dispersed geometries of PICTs make modeling the statistics of TC impacts challenging. In this paper, we try to close this gap by developing EMPIRIC_TC, a deep‐learning emulator of a statistical‐dynamical TC model, to provide local, accessible, ensemble projections of high‐intensity TCs in the South Pacific. We implement EMPIRIC_TC using a Fourier Neural Operator and show that it outperforms a UNet and Nearest Neighbors regression model at predicting facility relevant TC hazard and reproducing ensemble means. We then explore fundamental limits to the predictability of infrequent extreme events from climatic conditions and show their affect on deep learning model performance. The emulator allows for high‐skill, adaptable, and near‐instantaneous projections of TC hazards, estimation of climate model and internal variability‐related uncertainty, and evaluating projection uncertainty. The emulation system, EMPIRIC_TC, is available through a web portal for the use of PICT collaborators. Plain Language Summary: Healthcare facilities in Pacific Island countries and territories are especially exposed to tropical cyclones (TCs) and the associated impacts. Climate change is affecting TC frequency, intensity, and landfall locations. However, due to the rarity and uncertainty of high‐intensity TCs, it is difficult to assess these changes at the local scale required for proactive climate adaptation decisions. In this paper, we generate large data sets of TCs for many different future climate scenarios. This data is used to train and test deep‐learning emulators that produce estimates of TC hazards. The model, which we call EMPIRIC_TC, enables a better understanding of the uncertainty introduced by climate change and locally relevant information for affected parties. The emulation system is available through a web portal for use by PICT collaborators. Key Points: We develop a deep learning model that predicts 0.5°‐resolution Tropical Cyclone (TC) hazards from cyclogenesis frequency mapsEMPIRIC_TC skillfully estimates local ensemble means and generalizes to out‐of‐domain inputs with minor losses in facility relevant errorComparison of TC hazard variability due to internal stochasticity and initial conditions shows spatially varied limits to predictability [ABSTRACT FROM AUTHOR] |
| Copyright of Journal of Advances in Modeling Earth Systems is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) | |
| Database: | GreenFILE |
|
Full text is not displayed to guests.
Login for full access.
|
|
| Abstract: | Tropical Cyclones (TCs) pose significant and evolving threats to Pacific Island Countries and Territories (PICTs) healthcare systems. Changes in cyclone intensity, frequency, and spatial distribution due to anthropogenic climate change are likely to increase these threats. Accessible information on future uncertainty of high intensity TCs can enable adaptation that reduces the severity of systemic disruptions. However, the infrequency of high‐intensity TC events in historical and GCM data and the small and dispersed geometries of PICTs make modeling the statistics of TC impacts challenging. In this paper, we try to close this gap by developing EMPIRIC_TC, a deep‐learning emulator of a statistical‐dynamical TC model, to provide local, accessible, ensemble projections of high‐intensity TCs in the South Pacific. We implement EMPIRIC_TC using a Fourier Neural Operator and show that it outperforms a UNet and Nearest Neighbors regression model at predicting facility relevant TC hazard and reproducing ensemble means. We then explore fundamental limits to the predictability of infrequent extreme events from climatic conditions and show their affect on deep learning model performance. The emulator allows for high‐skill, adaptable, and near‐instantaneous projections of TC hazards, estimation of climate model and internal variability‐related uncertainty, and evaluating projection uncertainty. The emulation system, EMPIRIC_TC, is available through a web portal for the use of PICT collaborators. Plain Language Summary: Healthcare facilities in Pacific Island countries and territories are especially exposed to tropical cyclones (TCs) and the associated impacts. Climate change is affecting TC frequency, intensity, and landfall locations. However, due to the rarity and uncertainty of high‐intensity TCs, it is difficult to assess these changes at the local scale required for proactive climate adaptation decisions. In this paper, we generate large data sets of TCs for many different future climate scenarios. This data is used to train and test deep‐learning emulators that produce estimates of TC hazards. The model, which we call EMPIRIC_TC, enables a better understanding of the uncertainty introduced by climate change and locally relevant information for affected parties. The emulation system is available through a web portal for use by PICT collaborators. Key Points: We develop a deep learning model that predicts 0.5°‐resolution Tropical Cyclone (TC) hazards from cyclogenesis frequency mapsEMPIRIC_TC skillfully estimates local ensemble means and generalizes to out‐of‐domain inputs with minor losses in facility relevant errorComparison of TC hazard variability due to internal stochasticity and initial conditions shows spatially varied limits to predictability [ABSTRACT FROM AUTHOR] |
|---|---|
| ISSN: | 19422466 |
| DOI: | 10.1029/2025MS004998 |
