Hello Ventusky team!
Based on analysis, it appears that both AI GFS and AI ECMWF are superior to their respective progenitors (GFS/ECMWF), relying on historical maps and not on atmospheric physics. I think GFS also has a hybrid version combining both approaches.
Does Ventusky have plans to incorporate these AI models?
Thank you for all you do,
Aydin Michael, USA
A. M. Scienceman
They're not superior yet. AI models currently provide far fewer parameters, and they still aren't more accurate than traditional models for near-surface weather variables or precipitation in medium or long-term forecast. As soon as they offer a real advantage for users, we'll implement them. On the contrary, traditional physics-based models provide a wide range of parameters because weather is governed by the laws of physics, not just patterns. The future lies in combining the two. However, hybrid versions of the ECMWF and GFS models are not yet available.
I sure appreciate your feedback Ventusky team. It seems, as per below, we are both correct.
This is what I could ascertain from, ironically, a Google Gemini AI report on comparison between models, the effectiveness of the Hybrid model is also summarized.
Predictive Skill: Pure AI-based prediction systems like AIGFS and GraphCast generally outperform traditional numerical weather prediction (NWP) models in baseline metrics, such as global root-mean-square error (RMSE) and anomaly correlation. AIGFS extends medium-range forecast skill by roughly 18 to 24 hours.
Precipitation Forecasting: While AIGFS excels at lower-to-moderate rainfall thresholds, research comparing AI and NWP indicates that the GFS preserves physical precipitation distributions much better during heavy downpours. The AI tends to smooth out heavy rain spikes.
Tropical Cyclones: AI forecasting offers a significant upgrade in predicting the track of tropical cyclones. However, it often misrepresents or degrades the peak intensity (wind speed and central pressure) of the storm.
The "Hybrid" Solution: Forecasters have found that the most accurate results stem from HGEFS (Hybrid Global Ensemble Forecast System), a pioneering grand ensemble that blends AIGFS with conventional GFS ensembles. The HGEFS consistently outperforms both the AI-only and physics-only systems, mitigating the blind spots of both approaches.
A. M. Scienceman
It is true that HGEFS already exists (it combines GFS and AI GFS), but it currently offers only a limited number of layers and doesn't provide much added value for users. At this stage, it's mainly intended for testing purposes. In the future, GFS is expected to use a similar hybrid approach, and additional parameters may be added over time.
Ventusky
Ok, thank you!
Please log in to add a post to the discussion.
