A system for airport weather forecasting based on circular regression trees

This paper describes a suite of tools and a model for improving the accuracy of airport weather forecasts produced by numerical weather prediction (NWP) products, by learning from the relationships between previously modelled and observed data. This is based on a new machine learning methodology that allows circular variables to be naturally incorporated into regression trees, producing more accurate results than linear and previous circular regression tree methodologies. The software has been made publicly available as a Python package, which contains all the necessary tools to extract historical NWP and observed weather data and to generate forecasts for different weather variables for any airport in the world. Several examples are presented where the results of the proposed model significantly improve those produced by NWP and also by previous regression tree models.