I’m one of the creators of Meteosource Weather App. We have just released an update to our weather app and I would love to get your feedback.

Most weather apps are simple wrappers around a single API. We wanted to see if we could actually improve accuracy by using multiple models.

Instead of relying on a single provider, we built an engine that pulls from a multi-model ensemble (combining GFS, ECMWF, HRRR, etc.). We then apply machine learning to post-process these outputs — essentially training models on historical observations to identify and correct the systematic biases of each underlying numerical model for specific locations.

Key Technical Features: - ML-driven Nowcasting: We use real-time radar data and neural networks to predict precipitation to the minute - Bias Correction: Our models learn from past errors to improve local accuracy - Hyper-local resolution: We downscale global models to provide data for any specific coordinate

The App: - Precise hourly forecasts and interactive radar - Activity planning based on custom weather conditions - Minute-cast notifications and beautiful animated maps (in the pro version)