FengWu, FuXi and GraphCast beat ECMWF's model in Taiwan's test

Taiwan's weather agency ran five AI models for June to November 2023. FengWu, FuXi and GraphCast had lower errors than ECMWF's IFS, and FengWu tracked Typhoon Haikui best.

  • Not rated an organization

    The authors compare errors of forecasts run after the 2023 season, and no forecaster or agency is reported to have acted on them.

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

Status
Reported and intended benefits.
Evidence

foundstudiedreportedclaimed

An independent peer-reviewed study supports the benefit.

Compared with

ECMWF's operational IFS forecasts at 9 km resolution. Measured.

Without the system
IFS average track error exceeding 1000 km at 96 h for Typhoon Haikui
With it
FengWu average track error of 41 km at 96 h
The system’s part
The authors ran each AI model from identical ERA5 initial conditions and scored its own forecasts against ERA5 and best tracks. The IFS started from a different analysis, so its comparison with the AI models rests on error trends.
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
FengWu, FuXi, GraphCast, FourCastNet v2 (FCN2) and Pangu-Weather
Made by
Shanghai Artificial Intelligence Laboratory (FengWu), Fudan University (FuXi), Google DeepMind (GraphCast), NVIDIA (FourCastNet v2) and Huawei (Pangu-Weather)
Where
East Asia and the western North Pacific (evaluation domain), Taiwan
Stage
Tested in a laboratory
Date
, when the source was published
Kind
Aggregate result
Broad benefit
Estimating the Quantifiable Characteristics of Products, Events, or Information
Last revised
31 October 2026

Sources

  1. Peer-reviewed study, by authors with no tie to the system.

    Liu CC, Hsu K, Peng MS, et al. Evaluation of five global AI models for predicting weather in Eastern Asia and Western Pacific. npj Climate and Atmospheric Science 7, 221 (2024).

    Based on RMSE, the best model is FengWu, followed by FuXi and GraphCast, with FCN2 trailing behind, and Pangu-Weather ranking last. IFS with a resolution of 9 km, lies in the middle of the group.

Details

AI system

Workflow

The authors ran all five models from identical ERA5 initial conditions with 13 levels and a 6-hour step. Forecasts ran to 168 hours and were checked against ERA5 and the agency's best tracks.

Follow-up

Conditions for replication

Starting conditions must come from a reanalysis or analysis of good quality. Typhoon intensity needs other guidance, since all five models showed weak intensity biases.

Entry details

Sector

Weather forecasting / tropical cyclones.

Purpose

Shanghai Artificial Intelligence Laboratory, Fudan University, Google DeepMind, NVIDIA and Huawei built FengWu, FuXi, GraphCast, FourCastNet v2 and Pangu-Weather to forecast global weather with models trained on reanalysis data.

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). FengWu, FuXi and GraphCast beat ECMWF's model in Taiwan's test (AIBD-0759). Artificial Intelligence Benefits Database, edition of 31 October 2026. https://petervartanian.xyz/aibd/records/AIBD-0759.html (archived at https://doi.org/10.7910/DVN/AADSII, version 4.0)

BibTeX Data Field definitions