Aries: A Proprietary Medium-Range Weather Prediction Model for the Energy Industry
Lukas Hedegaard Morsing, Arian Bakhtiarnia, Jonas Lynge Olesen, Tómas Bragi Björnsson Leth, Christian Gøbel Bach
physics.ao-ph · cs.LG
Quick verdict
InCommodities' 2.4B-param Swin MLWP beats ECMWF HRES/AIFS on 10m wind RMSE to ~4 days (2025, vs ERA5); credible, but narrow benchmark and no code.
Summary
The paper presents Aries, a proprietary machine-learned weather prediction (MLWP) model developed at the energy trading firm InCommodities. Aries uses a non-hierarchical SwinV2-based transformer (14 layers, 16 heads, 3,584-dim embeddings, 2.4B parameters), trained on ERA5 reanalysis at 0.25° resolution (1979–2022, validation 2023, test 2024) to advance 74 prognostic and 11 diagnostic variables in 6-hour steps. Training combines single-step latitude-weighted MSE with multi-step autoregressive fine-tuning using L1 loss up to 10 rollout steps (60 h), plus a lightly weighted power-spectral-density penalty to counter spectral blurring. Evaluation covers the full calendar year 2025, initialized from ECMWF operational analyses at 00Z/12Z and verified against ERA5 over Europe and the USA. The headline result: latitude-weighted RMSE of 10 m wind speed is lower than both ECMWF HRES and AIFS operational up to roughly 96 h (4 days), after which Aries degrades faster; on 2 m temperature Aries matches AIFS, with HRES's higher RMSE attributed largely to a constant negative bias from its calibration to ECMWF Analysis rather than ERA5. Limitations, acknowledged in Section VI, include ERA5-only verification, a single evaluation year, absent station-level checks, and no evaluation of radiation or precipitation variables. No code or weights are released, and training details such as optimizer and compute budget are not reported.
Review complete
Assessment at a glance
5
risk flags identified
● high 0
● medium 3
● low 2
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 18 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- Aries achieves lower latitude-weighted RMSE than ECMWF HRES and AIFS operational for 10 m wind speed over Europe and the USA up to about 96 hours, on 2025 operational initializations verified against ERA5.
- On 2 m temperature, Aries matches AIFS operational RMSE through the first 96 hours, with HRES's larger RMSE largely a constant negative bias traceable to its calibration to ECMWF Analysis rather than ERA5.
- The loss of wind-skill advantage beyond ~4 days follows from the 10-step (60 h) autoregressive training depth; extending the training horizon would push skill further into the medium range.
- The results establish that proprietary, industry-side development of competitive medium-range MLWP models is technically viable for energy applications.
Strengths
Strictly out-of-sample, operational-mode evaluation
All 2025 initializations post-date training (1979–2022) and even the internal 2024 test year (Sec. III.C, IV); all three models run in operational mode at 00Z/12Z, and the authors note Aries was never fine-tuned toward ECMWF analysis inputs unlike HRES/AIFS (Sec. IV).
Unusually honest treatment of baseline artifacts
Sec. V.B correctly attributes HRES's higher 2 m RMSE to a roughly constant negative bias from its calibration to ECMWF Analysis rather than ERA5 — an interpretation that undercuts an overclaim the authors could have made about beating HRES.
Mechanistic explanation of the skill crossover
Energy-relevant, literature-grounded design
Candid limitations section
Weaknesses
Benchmark limited to two ECMWF systems
Train-truth circularity unexamined for the headline variable
No inferential statistics
Reproducibility-poor methods description
Minor figure-text inconsistency
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
The paper introduces Aries, a proprietary machine-learned weather prediction (MLWP) model built at the energy trading firm InCommodities, motivated by renewable generation and demand forecasting (renewables supplied 32% of global electricity in 2024 per Sec. I). Four claims are made: (1) lower latitude-weighted RMSE than ECMWF HRES and AIFS operational on 10 m wind speed up to ~96 h in Europe and the USA (Figs. 1–2); (2) 2 m temperature RMSE on par with AIFS, with HRES's deficit attributed to a constant negative bias from its calibration to ECMWF Analysis rather than ERA5 (Sec. V.B); (3) the 4-day wind-skill crossover is a direct consequence of the 10-step (60 h) autoregressive training depth (Sec. V.A, VI); (4) the meta-claim that proprietary industry development of competitive weather models is now technically viable. Scope is deliberately narrow: two near-surface variables (10si, 2t), two regions, deterministic verification against ERA5 only, and a single evaluation year (2025).
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 5 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What is the Aries weather model?
Aries is a proprietary machine-learned weather prediction model developed by the energy trading firm InCommodities. It is a 2.4-billion-parameter Swin Transformer trained on the ERA5 reanalysis at 0.25° resolution that advances the global atmospheric state in 6-hour steps, predicting 74 prognostic and 11 diagnostic variables.
Is Aries more accurate than ECMWF forecasts?
In the paper's evaluation, verified against ERA5 over Europe and the USA on 2025 operational runs, Aries shows lower latitude-weighted RMSE than ECMWF HRES and AIFS for 10-metre wind speed up to about 4 days ahead; beyond that horizon AIFS leads. For 2-metre temperature Aries is on par with AIFS and better than HRES, whose deficit is largely a constant negative bias.
Is the Aries model code or data publicly available?
No. The paper describes Aries as a proprietary, independently trained in-house model; no code, weights, or forecast archives are released, and key training details such as optimizer and compute budget are not published, so independent reproduction is currently not possible.
Why does Aries lose its advantage after four days?
The authors attribute the crossover to autoregressive training depth: the model was fine-tuned over only 10 rollout steps of 6 hours each (60 hours total). They state that extending this training horizon would be expected to push competitive skill further into the medium range.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- All quantitative RMSE/bias values exist only inside figure images (Figs. 1–2); the extracted text provides only qualitative descriptions, so exact margins between models could not be verified numerically.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.