Heterogeneous Convective Weather Forecast Translation into Airspace Permeability with Prediction Intervals

Abstract

The translation of convective weather information into capacity metrics such as airspace permeability or flow rate is important for more effective air traffic management. Also important is understanding and conveying the uncertainty in future convective weather forecasts. Although some prior methods compute a confidence score, they do not translate the varying skills of different forecasts directly into uncertainty in the capacity metric itself. This paper describes a supervised machine learning approach leveraging multiple heterogeneous weather sources to forecast permeability, including prediction intervals to help guide the selection of appropriate traffic management initiatives. Computed permeability metrics are compared to observed traffic flow rates in 57 regions in the United States over 122 case days, and they are found to correlate well in high-demand airspace. The flight-path length through the impacted region, which is a partial surrogate for workload, also correlates well with permeability. Features from single and ensemble storm-resolving forecasts combined with two different probabilistic forecasts are used to generate 0–12 h estimates of airspace permeability including prediction intervals. The skills of the combined forecast and each contributing forecast are quantified across varying forecast horizons. The algorithms are implemented in a real-time decision-support system that was evaluated at several facilities in 2014 and 2015.