Detection and nowcasting of convective turbulence using artificial intelligence techniques

A recent spate of commercial aircraft encounters with dangerous turbulence in or near convective storms has underscored the need for turbulence detection and nowcast products that can provide timely tactical information to airline dispatchers and to en-route aircraft via uplinks to cockpit electronic flight bags. This paper describes two products developed at the National Center for Atmospheric Research to address this need. The first, the NEXRAD Turbulence Detection Algorithm (NTDA), uses Doppler weather radar data to detect regions of turbulence in clouds and storms, employing artificial intelligence techniques to perform quality control and aggregate measurements. The second, Diagnosis of Convectively-Induced Turbulence (DCIT), uses data mining to analyze a database of many millions of turbulence observations and collocated data from observations, numerical weather prediction models, and derived features. This analysis is used to create skillful predictive functions that can be used for real-time convective and convectively-induced turbulence nowcasting. Important elements of this approach include development and incorporation of relevant derived features and predictor combinations that enhance the algorithm's skill, as well as the automated selection of a small, skillful set of predictors for different regimes (e.g., altitude levels and proximity to storms). In addition to statistical performance analyses, examples of the products' output for three turbulence accident cases are presented and implications for tactical decision-making are discussed.