Google has an AI upgrade for anyone who has ever checked the forecast, left the umbrella at home and regretted trusting their phone.

The company announced WeatherNext 3 on September 3, saying the model would begin powering weather experiences globally across Google Search, Maps and the Gemini app that day. Its most immediately useful claim: precipitation forecasts can be up to 50% more accurate when planning a day or more ahead. Google says the biggest gains will be in regions where forecasts have historically been less reliable.

“Up to” is doing some work there. The figure describes the maximum claimed improvement; it does not guarantee the same gain for every forecast or location.

The underlying change is how often the model gets fresh evidence. According to the researchers‘ paper, WeatherNext 3 takes in recent satellite observations to generate a new forecast every hour. Earlier AI forecasting systems depended on analysis data that could pass its own biases into their predictions.

For someone repeatedly refreshing a weather app as clouds gather, the useful distinction is that the forecasting model itself gets another look at the atmosphere.

There is more local detail, too. Google’s developer documentation describes several resolutions: selected surface predictions at five kilometers, other surface grids at 10 kilometers and atmospheric levels at 25 kilometers. That finest, five-kilometer detail applies to particular outputs.

The researchers also trained the model to predict readings from individual weather stations, taking account of local geographical features. They report lower temperature and dewpoint errors than competing global models, including at stations the system had not seen during training. That matters when a regional forecast has to account for places with different elevations and surroundings.

Google lists a forecasting horizon of 15 days, using 64 ensemble members—multiple possible weather outcomes that help describe uncertainty. Those alternatives help indicate how confident a forecast should be.

For everyday users, the upgrade arrives through familiar Google products. The company is also making forecast data available through BigQuery, Earth Engine and Google Cloud Storage for developers and businesses.

The real test will come when the forecast says you can leave the umbrella at home.

The post Google’s new weather AI promises fewer “but it said sunny” moments appeared first on Search Engine Watch.

Source:: searchenginewatch.com