Supporting weather forecasting performance management at aerodromes through anomaly detection and hierarchical clustering

Abstract

Weather forecasting is a critical factor for aerodrome and enroute flight operations. Airport decision-makers rely on assessments made by forecasters to ensure operations safety and optimize flight schedule despite potential adverse weather conditions. This manuscript suggests a novel methodology based on Machine Learning to detect forecasting anomalies in historic data, and to rely on them for anticipating potential threats in aerodrome future forecasts. The methodology is fed with historic bulletins from radars and with previous forecasts, which are then processed via an anomaly detection algorithm, and a hierarchical clustering algorithm. While the former algorithm spots anomalous data points, the latter is used to group sets of similar forecasts. The joint usage of the results allows calculating an error propensity metric, which can predict the expected tendency of a certain forecast to be inaccurate. The methodology is meant to enhance decision makers in managing aerodrome weather forecasting, understanding criticalities related to their accuracy levels.