Probabilistic Ensemble Weather Forecasting with Machine Learning
Machine-learning models that produce many possible weather futures, giving forecast uncertainty instead of one answer.
Tier Grades
Tier A: Frontier AI labs
feasible
Google DeepMind built GenCast, which beats ECMWF ENS on 97.4% of 1320 targets.
Tier B: Indonesian AI labs
stretch
Ensembles multiply inference cost, though one 15-day ensemble takes 8 minutes; training compute is not stated.
Tier C: Institution units such as ITB
stretch
Open GenCast code allows study, but training compute is not stated in the abstract.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
feasible
Using released ensemble forecasts for risk ranges fits energy and agriculture proofs of value.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeGenCast: a 15-day global ensemble at 0.25 degree for over 80 variables in 8 minutes; training compute not stated in the abstract.
- DataGenCast: decades of reanalysis data; skill above ECMWF ENS on 97.4% of 1320 targets.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 12 authors on the GenCast paper (arXiv 2312.15796).
- Time horizonYears.
- Shelf-life signalsGenCast already beats ECMWF ENS on 97.4% of 1320 targets, so the operational physics ensemble is no longer the bar.
- Pipeline roleTier A trains ensembles; Tier B and Tier C verify them over Indonesia; Tier D turns forecast ranges into decision support.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
Linked models
WeatherNext 2 (FGN)weatherexemplar
FourCastNet 3weatherexemplarGenCast (WeatherNext 1 Gen)weatherexemplar
ECMWF AIFS (Single / ENS)weatherproduced by
ArchesWeatherGenweatherproduced byWeatherNext 1 Gen (GenCast)weatherproduced by
Stormer (ENS)weatherproduced by
NeuralGCM (0.7 and ENS)weatherproduced byIFS ENSweatherused byLinks
R&D TopicProbabilistic Ensemble Weather Forecasting with Machine Learning
ModelsFourCastNet 3NVIDIA, weatherGenCast (WeatherNext 1 Gen)Google DeepMind, weatherWeatherNext 2 (FGN)Google DeepMind, weatherArchesWeatherGenINRIA, weather+5 more
FourCastNet 3NVIDIA, weatherGenCast (WeatherNext 1 Gen)Google DeepMind, weatherWeatherNext 2 (FGN)Google DeepMind, weatherArchesWeatherGenINRIA, weather+5 moreUse CasesWeather ForecastingAgriculture
Models 9
WeatherNext 2 (FGN)Google DeepMind, Google, weatherexemplarECMWF AIFS (Single / ENS)ECMWF, weatherproduced byFourCastNet 3NVIDIA, weatherexemplarNeuralGCM (0.7 and ENS)Google, weatherproduced byWeatherNext 1 Gen (GenCast)Google DeepMind, Google, weatherproduced byGenCast (WeatherNext 1 Gen)Google DeepMind, weatherexemplarArchesWeatherGenINRIA, weatherproduced byIFS ENSECMWF, weatherused byStormer (ENS)UCLA, weatherproduced byPapers
- GenCast: Diffusion-based ensemble forecasting for medium-range weatherIlan Price et al., arXiv preprint, 2023 founding
- FourCastNet 3: A geometric approach to probabilistic machine-learning weather forecasting at scaleBoris Bonev et al., arXiv preprint, 2025 state of the art
- FuXi-ENS: A machine learning model for medium-range ensemble weather forecastingXiaohui Zhong et al., arXiv preprint, 2024 state of the art
- ArchesWeather & ArchesWeatherGen: a deterministic and generative model for efficient ML weather forecastingGuillaume Couairon et al., arXiv preprint, 2024 state of the art
- Neural General Circulation Models for Weather and ClimateDmitrii Kochkov et al., arXiv preprint, 2023 state of the art
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2312.15796 https://arxiv.org/abs/2312.15796
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Probabilistic Ensemble Weather Forecasting with Machine Learning", https://aurion.semai.id/topics/probabilistic-ensemble-weather-forecasting/, records of October 2026.