On Friday, 14th November 2025, Arnaud Monges, PredictWind meteorologist, introduced our three new global weather models —PWAi, AIFS, and ICON —and provided a high-level technical overview of AI forecasting, validation, model physics, and initial conditions.
📺 Watch the Recording
You can now watch the full session on YouTube here:
👉 YouTube Recording – New Weather Models Webinar
Below, we answer the seven most commonly asked questions:
1. How are weather models built, and what are initial conditions?
Traditional weather models begin with initial conditions—a detailed snapshot of the atmosphere from satellites, weather stations, aircraft, buoys and radar. Models like GFS, ECMWF, ICON, UKMO and PW then use physics-based equations to simulate how the weather will evolve. AI models such as AIFS work differently: instead of solving equations, they learn patterns from huge historical datasets and apply that knowledge to today’s initial conditions to generate a forecast.
PredictWind’s proprietary PWAi model goes a step further by combining the strengths of ECMWF, AIFS, Fengwu and GraphCast into a single optimised forecast. It offers 1-hour time steps for the first three days and performs extremely well in validation testing, especially for short- to medium-range forecasts over larger-scale areas. PWAi is a key model to compare within the PredictWind suite when looking for consensus across different forecasts.
2. What’s the best model for my area?
There’s no golden rule for which model will be most accurate, as forecast accuracy varies with weather patterns, terrain, location, and time of year. The most reliable method is to use the highest-resolution models available and compare their results to see where they agree. It’s also crucial to verify real-world conditions by observing what you see on the water, checking live weather stations, and using the PredictWind Validation Tool. A comparison of multiple models, supported by actual observations, will always give you the most confidence.
3. Is PWAi better than AIFS?
In many regions, PWAi is performing exceptionally well and often outperforms AIFS in the short range, thanks to its blend of ECMWF, AIFS, Fengwu, and GraphCast to create a single optimised forecast. AIFS is still a strong model, particularly for medium-range forecasting and storm-structure detail, and both models offer valuable insights. Using them together provides the most reliable overall picture.
4. Do AI models struggle with rare or extreme events?
AI models learn from past weather patterns, so very unusual or unprecedented events can be harder for them to handle. For typical extreme systems—such as most cyclones and hurricanes—AI models such as AIFS can perform better than traditional physics-based models like the ECMWF. But if a truly unusual, out-of-distribution event occurs, then AI models are more likely to struggle. PWAi helps reduce this risk by blending information from multiple trusted models, resulting in stronger performance during intense weather. And as always, when extreme conditions are possible, it’s best to compare the AI guidance with ECMWF, GFS, UKMO, ICON, and the PredictWind GMDSS tools to get the clearest overall picture.
5. Where can I see the forecast update times for all the models?
You can view all model update times directly in both the PredictWind website and app. On the website, the next update time appears below your forecast locations, and if you hover over it, you’ll see the full list of update times for each model. In the PredictWind App, the update times are shown in the same place—below your forecast locations on the main menu and on each forecast page—and tapping this area will show all model update times. For a deeper explanation, you can also refer to the Forecast Updates Times article here.
6. Will you be doing wave and current AI models, too?
We are considering developing an AI wave model; however, it is in very early stages, and we haven't yet done a proper feasibility study, so we can't say if or when it will happen. It is definitely something we are interested in, though.
7. What do you mean by “high resolution”?
When we say “high resolution”, we’re talking about how detailed the forecast model is in space and time. A high-resolution model uses a much finer grid (for example, around 1–8 km between data points instead of tens of kilometres), and shorter time steps, so it can “see” smaller-scale features like sea-breezes, coastal wind bends, funnelling in channels and localised squalls that coarser models tend to smooth out. In practice, that means more precise predictions of wind, rain, waves and currents in the areas you actually sail in, especially near the coast. You can learn more and watch a short explainer video in this Help Centre article here.
Want to Learn More?
Explore the complete Marine Meteorology Course with in-depth articles, visual guides, and expert tips. Perfect for cruisers, racers, and weather enthusiasts looking to sharpen their forecasting skills.