Google's WeatherNext 3 is the best weather model yet. You too will get a more accurate forecast.

Google has introduced WeatherNext 3 – a new artificial intelligence model designed to predict weather more accurately and, most importantly, with much greater detail than its predecessor. The company is immediately enabling this new feature in its most used services, so you too will see a better forecast without installing anything.

What can WeatherNext 3 do better than before?

The biggest leap is in detail. While last year’s model calculated forecasts on a 25-kilometer grid and updated them every six hours, the new version gets much closer to the ground. Key variables like temperature and humidity are displayed at a resolution of up to 5 kilometers, other surface values at ten, and atmospheric phenomena like wind speed at twenty-five kilometers. In its announcement, Google talks about an overall roughly five times sharper weather picture than WeatherNext 2 offered.

The difference will be most noticeable in rugged terrain. Near coastlines, in valleys, or in mountains, temperatures can change significantly within just a few kilometers, and a coarse grid simply smoothed over such jumps. Furthermore, the new model also learns from ground station measurements, so according to the company, the result better fits local conditions – even in parts of Latin America, Africa, or Asia, where detailed forecasts were previously almost unavailable due to supercomputing costs.

Google has paid special attention to precipitation, which global models traditionally struggle with. The model was trained on NASA satellite data and its own analysis, and for medium-range forecasts, it reports an improvement of up to 60 percent compared to one of the common metrics. However, these are figures predominantly from the company’s own tests, so take them more as a promise than an independently confirmed verdict. Nevertheless, independent evaluations by the Brightband organization also indicate that it is currently the most accurate global model.

Where will you see a more accurate forecast?

Since last week, WeatherNext 3 has been powering weather forecasts in Google Search, the Gemini app, Maps, the Google Maps Platform Weather API, and Google Earth Engine. When planning a day or more in advance, precipitation forecasts are expected to be up to fifty percent more accurate, according to Google, with the biggest improvement in areas where reliability previously lagged.

However, the model is not only aimed at ordinary users. It also predicts wind speed at an altitude of around 100 meters, i.e., roughly at the blades of wind turbines, as well as cloud cover and solar radiation. This is intended to help grid operators and renewable energy producers estimate how much energy they will generate. Furthermore, developers and researchers can retrieve forecasts via the BigQuery service or download them in bulk from Google Cloud storage.

Do you trust AI weather forecasts more than traditional ones?

Sources: Google, Brightband