WeatherNext 3 is a global AI weather forecasting model designed to generate forecasts every hour of the day, every day of the year, and to expose local weather variables at a resolution that is useful for real decisions. Instead of waiting for a few daily model cycles, you get a continuously refreshed picture of precipitation, temperature, humidity, wind, radiation, and cloud cover.
That single design choice, hourly generation, changes the kinds of questions a forecast can answer. This guide explains what WeatherNext 3 produces, where the inputs come from, what the resolution numbers mean in practice, and how to start exploring it yourself.
Why hourly generation changes the questions you can ask
Most familiar forecasts move in six or twelve hour steps. That is fine for "will it rain tomorrow", but it hides the part of the day that usually drives the decision: the two-hour window when a rain band crosses a site, the morning when humidity collapses, the afternoon when wind ramps up and then fades.
When a forecast is regenerated every hour, you can also watch the forecast itself evolve instead of treating one run as final. A storm track that shifts noticeably between runs is telling you something important about uncertainty, and that signal is invisible if you only look at a single daily output.
What WeatherNext 3 produces
WeatherNext 3 covers the surface variables that most operational decisions depend on, and it exposes them together so you can read the conditions as a set rather than one number at a time.
| Variable group | What you get | Typical question |
|---|---|---|
| Precipitation | Rain and snow timing, intensity, and movement | When does the rain band arrive? |
| Surface temperature | Targeted temperature at 5 km | How warm will the afternoon get? |
| Humidity | Local moisture levels read alongside temperature | Will the air dry out overnight? |
| Wind | Speed and direction for energy and logistics planning | Is the wind window usable? |
| Radiation and cloud cover | Solar-relevant inputs for production planning | How much sunlight reaches the site? |
| Tropical systems | Global forecast layers for large-scale events | Where is the system tracking? |
Where the inputs come from
WeatherNext 3 draws directly on raw satellite imagery rather than relying only on pre-processed analysis fields. That matters for two reasons.
First, it widens coverage. Large parts of the planet have sparse ground observations, and satellite-first inputs let the model keep producing detailed forecasts where station networks are thin.
Second, it preserves detail. Every processing step between a raw observation and a model input has the chance to smooth out the small-scale structure that makes a local forecast useful. Starting closer to the raw imagery keeps more of that structure intact.
Resolution: what 5 km and 10 km mean in practice
Targeted temperature and humidity outputs are available at 5 km, while other surface variables are provided at 10 km. Those numbers are easy to read past, so it helps to translate them.
At 5 km, you are working at roughly the scale of a city district, a valley, or a coastal strip. That is the difference between "the region will be windy" and "the ridge above the site will be windy while the valley stays calm". For energy planning, agriculture, and any operation tied to a specific piece of ground, that gap is often the whole decision.
At 10 km, the other surface variables still deliver far more local structure than a coarse global grid, and they are the right level of detail for regional planning, routing, and comparative analysis across many sites.
Who benefits most
- Renewable energy. Wind, radiation, and cloud-cover forecasts feed directly into generation planning and into the question of whether a maintenance window is actually usable.
- Logistics and field operations. Hourly precipitation and wind make it possible to schedule around a window instead of cancelling a whole day.
- Agriculture. Local precipitation, heat, humidity, and wind variables determine irrigation timing, spraying windows, and frost risk.
- Cities and public planning. Hourly temperature and humidity support heat response, snow operations, and event planning.
- Research. Comparing hourly operational forecasts with historical output is a practical way to study how forecast behaviour changes run to run.
How to start exploring
- Choose a location: a city, region, or a specific operational area.
- Select the variables that match your decision rather than everything at once.
- Describe the decision you are actually making, including the time horizon.
- Explore the resulting forecast and compare the hourly progression.
You can walk through this end to end in How to use WeatherNext 3, and the documentation covers variables, resolution, and forecast credits in more detail.
Practical limitations worth remembering
- Forecasts are probability distributions, not promises. Read the trend across several hourly runs before committing to a plan.
- Higher resolution is not the same as a station measurement. Local terrain still shapes conditions in ways a gridded field will smooth over.
- For safety-critical decisions, official national weather service warnings stay authoritative. WeatherNext 3 is a decision-support input, not a replacement.
Frequently asked questions
What is WeatherNext 3?
WeatherNext 3 is a global AI weather model that generates forecasts every hour and provides high-resolution local weather variables, including targeted temperature and humidity at 5 km and other surface variables at 10 km.
Do I need my own model or compute to use it?
No. You can explore WeatherNext 3 forecasts here without setting up, training, or operating a forecasting model yourself. Plans include monthly forecast credits, which are covered on the pricing page.
Where can I compare it with earlier WeatherNext models?
See WeatherNext 3 vs WeatherNext 2 for a focused comparison of what changed.
