Paraglidable.com uses deep learning to analyse standard weather forecast models.
Paraglidable.com uses an artifical neural network to compute flyability and crossability scores by analysing ~200 weather parameters for each day.
The neural network is trained over ground truth data spanning the last 10 years.
The ground truth data is composed of weather archives and available online flights databases (~2 000 000 flights).
During the training, the neural network learns the correlation between weather conditions and reported flights (their number, distance, max vario, max altitude...).
For computing forecast, paraglidable.com downloads the last weather forecast from multiple sources to feed the neural network. The neural network outputs a prediction of the flying condition.