Assessment of The Graphical Turbulence Guidance, Version 3 (GTG3)

Abstract

The QA PDT was tasked to assess the quality of the Graphical Turbulence Guidance, version 3, (GTG3) algorithm developed by the National Center for Atmospheric Research (NCAR). This product is to replace the current GTG2.5 algorithm currently being used for operational aviation turbulence decisions. Changes between GTG2.5 and GTG3 include: 1) an extension of the forecast domain down to 100-ft altitude (from 10,000 ft), 2) an increase in forecast leads from 12 to 18 hours, 3) the addition of an explicit mountain-wave (MW) turbulence component, and 4) an upgrade to the conversion of the raw algorithm output to the Eddy Dissipation Rate (EDR). The assessment has five main areas of investigation and incorporates output from the operational GTG2.5 algorithms, the GTG3, and the National Weather Service (NWS)-produced Graphical Airmen's Meteorological Advisories (G-AIRMETs), as well as, PIREPs and EDR values derived from in situ measurements. The forecasts were analyzed using output generated from 1 January - 31 March 2013 and 1 July - 30 September 2013 over the CONUS. Primary findings include: GTG3 distributions are noticeably different than the distributions for GTG2.5--the GTG3 distribution is more constrained (i.e., lower variance, weaker tails) and the peak of the distribution is shifted from near-zero values to around 0.1. GTG3 is consistently better at discriminating events from non-events than GTG2.5, at all observed thresholds. When the forecast threshold is constrained to match the observed threshold (i.e., no calibration), GTG3 is more skillful than GTG2.5 for only a small range of thresholds; however, this range can be expanded with proper calibration. With calibration, GTG3 outperforms GTG2.5 for events with an EDR greater than about 0.14, while GTG2.5 outperforms GTG3 for events with an EDR less than about 0.14. GTG3 captures more Moderate-or-greater (MOG) events than G-AIRMETs for the same forecast volume, or by choosing a different forecast threshold, GTG3 captures the same number of events as G-AIRMETs while using only one-third of the volume. Performance of GTG3 in the Near-surface layer (below 10,000 ft) is not as skillful as other layers, but GTG3 outperforms G-AIRMET in this layer. Mountain-wave component: Very effective (99%) at capturing Light-or-greater intensity explicit MW PIREPs, but with a higher number of false alarms (60%). Captures 70% of Moderate-or-greater MW PIREPs, with very few false alarms (6%). Using all reports (MW and others), forecasts using the clear-air (CAT) algorithm with the MW component) are equally skillful as forecasts using only the CAT algorithm