A Method for Estimating Distributions of Near-surface Eddy Dissipation Rates (EDR) at Sample Airports Across the National Airspace System

Currently, separations between aircraft include protections against wake vortices produced by leading or nearby aircraft. Near-surface separations are primarily based on analyses of historical LIDAR observations, the amassing of which is both costly and difficult to maintain. An alternative path is to utilize physics-based wake vortex models to assess this risk. These models require reliable climatological information on the joint probabilities of meteorological variables, including winds, stability and eddy dissipation rate (EDR). This last value is not regularly observed or archived, so there is no direct way of generating these joint distributions. This is further complicated by the fact that the wake vortex model assessments are sensitive to the very low EDR portion of the probability density function – a region which is less well studied than the larger EDR portion.

This talk will discuss a potential methodology for producing these joint distributions at airports across the National Airspace System, by leveraging existing datasets, including a few with EDR observations, and a set of machine-learning models to overcome this obstacle. A key to the method is recognition that the EDR distributions can be viewed in two parts: the main portion of the distribution and the very low EDR tail. The approach is to generate models of the main portion of EDR by using machine learning techniques to relate the EDR distribution to other regularly observed and archived variables. This will provide the seasonal and regional contributions to the distributions. The shape of the tail will be modeled separately via a second set of machine learning models and stitched to the main EDR distribution to provide a full estimate. These estimates can then be used to provide the desired joint probabilities with other variables such as winds and stability.