The wunderground.com proprietary forecasting model, BestForecast, provides the world's most localized, reliable weather forecasts on the web. The model combines data from our network of over 20,000 personal weather stations as well as historical data and data from over 12,000 airports around the world. After running our BestForecast model, we publish our hourly weather forecasts.
Weather is not an exact science. To enhance the accuracy of forecasting, we need to understand the existing error base from which to improve. To do this you need to cross-verify forecasts with actual conditions. We here at wunderground.com have more data points than any other weather source and are therefore better positioned to provide reliable and accurate hourly forecasts.
For every location in the world, BestForecast analyzes the accuracy of 75 different forecasting models over the previous 2-3 weeks, identifying which model has historically performed the best. BestForecast analyzes every weather variable at every 3-hour time interval going back 14 days. From this data we publish our hourly forecasts.
The BestForecast model verifies eight different variables: hourly temp, high temp, low temp, humidity, wind speed, wind direction, cloud cover and pressure.
By multiplying all our data points: 31,000 (stations) x 75 (models) x 112 (time intervals) x 8 (data variables) one gets over 2 billion variables that the model analyzes each time we run the model. The model is run twice a day, and in the future we plan to run it four times a day.
Our 20,000+ personal weather stations essentially fill in the gaps between the airport stations. Most other weather sources have to use data assimilation to fill in these gaps.
The new proprietary forecasting system allows wunderground.com to provide our users with unprecedented levels of transparency by publishing the recent accuracy of its weather forecast hourly for each location.