Most publicly available weather forecasts are based on one of the many meteorological models. In practice, it cannot be said that one model is better than another. Each is differently specialized, and it holds true that a different model is best for different types of weather. Some are better at predicting storms, others excel at handling winter inversions, while others correctly forecast nighttime temperatures in certain meteorological situations.
Weather forecasts for this website are created using machine learning - they learn from historical data and past predictions, which are compared to actual outcomes. Subsequently, they intelligently combine many different models to minimize prediction errors. Thanks to this unique approach, we achieve the most accurate forecasts possible.
Combining Models Using Machine Learning
When creating our forecasts, we utilise our extensive database of past weather and archives of past runs of various meteorological models in the first phase. Based on this data, our machine learning generates a unified global output that eliminates most biases and errors. Our machine learning algorithms evaluate the past error of individual models in various meteorological situations and locations.
The output provides the most likely scenario for the weather's development, based on a combination of all available models. This foundational forecast is computed four times a day.
Our Proprietary Local Meteorological Forecasting Model
The global multimodel then enters our regional models, which further increase resolution and consider local geographic specifics and orography. Meteorologists in service at this stage can also adjust the input based on their expertise.
AI Model for Weather Correction in Cities
For locations where measurements from meteorological stations are available - typically cities and more populated areas - we use an additional model layer that uses artificial intelligence (AI) to correct forecasts based on the specific area and the current and predicted meteorological conditions.
Given the high variability of weather, in some situations, even the best models need adjustment based on the current real-time developments. Using all available measurements from our partners and our own measurements, we update our outputs to reflect the latest weather conditions. These minor corrections are continuously calculated and refined in real-time based on the actual weather developments. This allows us to provide the most up-to-date weather forecast - hourly and by the minute.
See for yourself the accuracy of our weather forecasts, or check out ratings from our users.