arXiv Science⌕ Search

arXiv · 2611.15266

DaYu-TC: a global-regional nested AI system for tropical cyclone forecasting

Abstract

Accurate tropical cyclone intensity forecasts are essential for disaster preparedness, yet persistent intensity underestimation limits the value of artificial intelligence (AI) weather models for predicting these destructive storms. Here we present DaYu-TC, a global-regional nested system that links the preservation of strong-wind and low-pressure signals with high-resolution regional prediction. Its global and regional models share an architecture and extreme-aware training objectives. The regional model combines evolving local states with global forecasts valid at each target time, connecting large-scale environmental information with finer-scale cyclone evolution. In an independent evaluation of 16 western North Pacific tropical cyclones in 2023, the global model retains competitive conventional forecast skill while reducing five-day maximum-wind root-mean-square error (RMSE) by 6-22% relative to other evaluated global AI models. Relative to the global model, regional nesting reduces maximum-wind RMSE by 22.0-33.1% and minimum-pressure RMSE by 28.6-39.8% across 12-120 h. The complete system outperforms concurrent operational high-resolution forecasts from the European Centre for Medium-Range Weather Forecasts for minimum pressure at 12-120 h and maximum wind speed at 48-96 h, with respective RMSE reductions of 26% and 12% at 72 h. Without retraining, the same regional model enhances three other global AI models, reducing overall maximum-wind RMSE by 22.6-28.6%. Five-day forecasts take under one minute on a single GPU. These results establish a transferable multiscale approach to reducing cyclone intensity underestimation, extending advances in AI weather forecasting towards more accurate prediction of high-impact weather events.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Runzepu Hou, Feng Zhang, Xin Wu, Xujun Wei, Fenghua Ling, Kan Dai, Gaozhen Nie, Leiming Ma, Renhe Zhang. 2026-11-11. DaYu-TC: a global-regional nested AI system for tropical cyclone forecasting. https://arxiv.science/abs/2611.15266

Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite arXiv Science by Indiciae: BibTeX · RIS.

KEEP EXPLORING

Related papers

Twister After Dark: Diagnosing a Damaging Nocturnal Tornado in Iloilo, Philippines (08 August 2026)

At ~1430 UTC (2230 LST) on 08 August 2026, a significant tornado struck the Mandurriao and Jaro districts of Iloilo City, Philippines. Its parent supercell developed within the southwest monsoon enhanced by Typhoon Dolphin, under weak synoptic forcing, as a nocturnal low-level jet of ~40 kt near 1.5-3 km advected warm, moist marine air onshore. ERA5 depicted MUCAPE of ~2000 J kg$^{-1}$, negligible inhibition, an LCL below 300 m, and nearly purely streamwise low-level vorticity, but deep-layer shear (DLS, 0-6 km bulk wind difference) below 3 m s$^{-1}$. Hodographs from the newly operational Iloilo S-band dual-polarization radar (S-ILO), ~5-7 km from the storm, revealed low-level shear (LLS, 0-1 km bulk wind difference) of ~16 m s$^{-1}$ and 0-1 km storm-relative helicity of 124-211 m$^2$ s$^{-2}$. The storm exhibited a hook echo with a weak echo hole, a $Z_{DR}$ arc and ring, an elevated 55-60 dBZ rear-flank core consistent with a descending reflectivity core, and a ~50 dBZ ring interpreted as a debris ball. A survey of 31 damage indicators documented a ~2.9 km long and up to ~340 m wide path rated IF2 on the International Fujita Scale. Processor smoothing of the radial velocity and a $ρ_{HV}$ threshold for rainfall estimation masked any tornadic vortex or debris signature, although a radial velocity simulation shows that a narrow vortex (core radius ~50-100 m) with a rotational velocity of ~35-45 m s$^{-1}$ could have been concealed. This case shows that significant tornadoes can develop at night during the Philippine monsoon season and that operational radar quality control should be revised to retain severe-storm signatures.

physics.ao-ph↗

Projection of global daily CO2 emissions in 2026

Global fossil-fuel combustion and industrial-process CO2 emissions remain close to historical highs, yet the recent slowdown in growth is difficult to assess before conventional inventories become available. We project emissions for 2026 and examine daily departures during major climate and energy disruptions. Carbon Monitor activity-based estimates from 1 January 2019 to 30 September 2026 are combined with Prophet projections for 1 October to 31 December across 47 region-sector series. Global emissions are estimated at 38.1 Gt CO2 in 2026, 36.6 Mt CO2 (+0.10%) above 2025. The reference 68% range for the year-on-year rate is +0.011% to +0.18%. Five September-cutoff hindcasts show lower daily forecast errors for Prophet than for a seasonal-naive benchmark. Regional increases total 0.32 Gt and decreases total 0.29 Gt, so 94.0% of the 0.61 Gt gross movement cancels in the global sum. China contributes the largest reduction (-0.19 Gt), concentrated in industry and ground transport, whereas India contributes the largest increase (+0.19 Gt), led by power and industry. Heat-period power anomalies differ among countries and can be muted by summer averaging, while aviation and shipping show different disruption and recovery profiles. Resolving the near-zero global change reveals where reductions are concentrated and where continued growth offsets them, providing a timely basis for assessing the breadth of mitigation progress.

physics.ao-ph↗

Improving Eddy Detection from Wide-Swath Altimetry with Multimodal Data-Driven Models

Physically-based eddy detection algorithms usually rely on hard-coded thresholds and constraints that make the detected structures sensitive to small variations in the underlying fields. One way to relax such conditions is the use of machine-learning methodologies that transform the full problem into an optimization process. We present a multimodal, multi-task, data-driven eddy detection framework based on a shallow convolutional U-Net, trained to jointly segment eddy structures and reconstruct the co-located spatial sea surface temperature (SST) anomaly using a dynamically-based reconstruction as target. Two model variants are developed: NADIRnet, trained on conventional nadir altimetry, and SWOTnet, trained through a curriculum that incorporates SWOT-enriched maps. Both models achieve high pixel- and structure-wise detection scores against the ground-truth, with most residual disagreement attributable to boundary placement rather than missed or spurious detections. The coupled SST regression task, enforced through boundary-coupling and gradient-matching loss terms, is shown to sharpen and regularise the detected eddy boundaries. Compared to the dynamically-based reference, both models recover a substantially larger and finer eddy population; notably, SWOTnet's higher-resolution training signal transfers to conventional nadir altimetry thanks to the higher-resolution SST channel, detecting more eddies than a nadir-trained model on the same data. The framework further generalises, even without retraining, to altimetry products and processing levels unseen during training, including SWOT L3 wide-swath retrievals and an independent gridded product. Together with its low computational cost, this framework is thus promising for near-real-time operational deployment and further downstream applications.

physics.ao-ph↗