Pangu, FengWu and other AI models foresaw Typhoon Danas's turn

In a case study, AI models placed Typhoon Danas's formation and sharp turn toward Taiwan better than physics-based models in forecasts from 30 June to 4 July 2025.

  • Not rated an organization

    The team compares the forecast tracks in figures without error measures and does not report how forecasters used them.

Depth runs from 1, a convenience or time saved, to 5, a life or a livelihood preserved.

Status
Reported benefit.
Evidence

foundstudiedreportedclaimed

Only the deployer supports the benefit.

Compared with

ECMWF, JMA, NCEP and UKMO global models run from the same start times. Measured.

The system’s part
The figure sets the AI models' own forecast tracks beside those of four physics-based global models from the same start times, which isolates the models' skill for this storm. HKO runs the models itself, and the makers are outside HKO, except CMA for FengQing.
Limits
Not yet checked.
Costs and harms
Not yet checked.
Later
Not yet checked.

Download data Cite Suggest a correction

Benefit type
Improved Task Accuracy
System
AI weather models Pangu, FengWu, FuXi, GraphCast, AIFS, FengQing and Aurora, as named in the case study figure
Made by
Huawei (Pangu), Shanghai Artificial Intelligence Laboratory (FengWu), Fudan University (FuXi), Google DeepMind (GraphCast), ECMWF (AIFS), CMA (FengQing) and Microsoft (Aurora)
Used by
Hong Kong Observatory, which runs the models and provided the figure
Where
Northern South China Sea and seas near Taiwan, with forecasts from Hong Kong, China
Stage
In routine use
Date
Kind
Benefit created
Broad benefit
Estimating the Quantifiable Characteristics of Products, Events, or Information
Last revised
31 October 2026

Sources

  1. Report, by the organization that used it.

    ESCAP/WMO Typhoon Committee, Expert Team on AI Applications in Tropical Cyclone Analysis and Forecasting (ET-AITC). Annual Operating Plan 13 Activity Report 2025. April 2026, section 3.4, p. 19, Figure 3.4.1 (figure provided by the Hong Kong Observatory).

    In this case, the genesis and movement of Danas were not captured well by the conventional physics-based global NWP models in their medium-range forecasts, while AI models were able to suggest more accurately the genesis position and its earlier movement, highlighting AI models’ value in particular for forecasts of longer lead times.

Details

AI system

Workflow

AI models and physics-based global models were run from 12 UTC starts each day from 30 June to 4 July 2025. The team compared pre-genesis and early tracks with the operational analysis positions.

Follow-up

Conditions for replication

The pattern holds across many storms, since this is one case. Forecasters weigh AI tracks together with physics-based guidance.

Entry details

Sector

Weather forecasting / tropical cyclones.

Purpose

Huawei, the Shanghai Artificial Intelligence Laboratory, Fudan University, Google DeepMind, ECMWF, CMA and Microsoft built these AI models to forecast global weather from data, including the tracks of tropical cyclones.

Not yet checked: model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.

Cite this record

Peter H. Vartanian. (2026). Pangu, FengWu and other AI models foresaw Typhoon Danas's turn (AIBD-0757). Artificial Intelligence Benefits Database, edition of 31 October 2026. https://petervartanian.xyz/aibd/records/AIBD-0757.html (archived at https://doi.org/10.7910/DVN/AADSII, version 4.0)

BibTeX Data Field definitions