Pangu-Weather beats ECMWF on 2023 typhoon tracks in China's check

CMA's National Meteorological Centre found that Pangu-Weather and FengWu beat ECMWF's model on typhoon tracks beyond 36 hours in 2023, though not within 24 hours.

  • 2 an organization

    Averaged over all lead times, Pangu-Weather's track error was 140.8 against 169.7 for ECMWF's model in CMA's 2023 verification.

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

Status
Reported and intended benefits.
Evidence

foundstudiedreportedclaimed

Only the deployer supports the benefit.

Compared with

ECMWF's operational deterministic model (ECMWF-IFS) for the same 2023 typhoons. Measured.

Without the system
169.7 (ECMWF, mean absolute track error averaged over all lead times)
With it
140.8 (Pangu-Weather) and 141.1 (FengWu)
The system’s part
CMA compared each AI model's own track forecasts with ECMWF's operational model for the same 2023 typhoons, which isolates the models' skill. CMA describes them as data-driven models built on transformer architectures. CMA runs the models itself, and the makers are outside CMA.
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
Pangu-Weather, FengWu and FuXi, run in real time at CMA's National Meteorological Centre
Made by
Huawei (Pangu-Weather), Shanghai Artificial Intelligence Laboratory (FengWu) and Fudan University (FuXi)
Used by
CMA National Meteorological Centre
Where
Beijing, China (western North Pacific and South China Sea)
Stage
In routine use
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. Report, by the organization that used it.

    China Meteorological Administration, National Meteorological Centre. Preliminary Explorations on Data-driven AI-Models in Tropical Cyclone Prediction. Presentation by QIAN Qifeng to the 56th session of the ESCAP/WMO Typhoon Committee, 27 February 2024, slide 9.

    For forecast periods of 36 hours and beyond, Pangu-Weather and FengWu have errors smaller than the ECMWF-IFS.

Details

AI system

Workflow

CMA ran the three models in real time and made operational products from them on its forecasting platform. CMA verified each model's typhoon tracks for the 2023 season.

Follow-up

Conditions for replication

The advantage holds at lead times of 36 hours and beyond. Intensity comes from other models, since the AI models underestimate it.

Entry details

Sector

Weather forecasting / tropical cyclones.

Purpose

Huawei, the Shanghai Artificial Intelligence Laboratory and Fudan University built Pangu-Weather, FengWu and FuXi to forecast global weather in the medium range with models trained on past weather 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). Pangu-Weather beats ECMWF on 2023 typhoon tracks in China's check (AIBD-0755). Artificial Intelligence Benefits Database, edition of 31 October 2026. https://petervartanian.xyz/aibd/records/AIBD-0755.html (archived at https://doi.org/10.7910/DVN/AADSII, version 4.0)

BibTeX Data Field definitions