Wind Profile Estimation from Aircraft Derived Data Using Kalman Filters and Gaussian Process Regression

Accurate wind information is crucial in air traffic management, for instance, to improve trajectory predictability and precision in controlled time of arrival. Nowadays, air traffic management relies on Numerical Weather Prediction, which usually has a low resolution and low update rate. A potential approach for improving the resolution and accuracy of the weather predictions consist in using airborne aircraft as meteorological sensors. Aircraft surveillance systems such as ADS-B and Mode S, transmit data related to weather conditions, automatically or in response to interrogation by air traffic control surveillance radars. In this paper, three different methods for constructing wind profiles from surveillance data have been applied and a comparison between them carried out. The first two methods being modifications of the Kalman filter have been referred to as the Adapted Kalman Filter and Smooth Adapted Kalman Filter. The third one is based on Gaussian process regression. The Kalman filter based methods are able to assimilate nearby data in a straightforward way and update the wind speed estimation in real time. Gaussian process regression is a very flexible and general regression model that can smoothly interpolate in space and extrapolate in time. These three methods have been validated using a test data set, achieving a 50% reduction of the prediction uncertainty in comparison with a baseline model. In addition, the Gaussian process methodology has been applied to reconstruct and forecast the wind field.