DRAFT NOTE as of February 4, 2000

A very strong Atlantic jet stream (Fig. 1) preceded strong storms that affected France and southern England around 26 December 1999 (the 250 mb wind speed contours range from 30-100 m/s). Observations from four flights of the Global Aircraft Data Set (GADS) experiment are available that did not appear on the GTS. The situation at 1200 UTC 25 December 1999 is particularly interesting.

Because the bulk of the eastbound transatlantic flights seek out the westerly jet, the analysis for 0600 UTC 25 December 1999 has plentiful data (Fig 2a) and the strength of the jet is reasonably accurate (Table 1). But as is typical, the 1200 (Fig. 2b) and 1800 UTC 25 December 1999 and 0000 UTC 26 December 1999 analyses (not shown) have far fewer AIREP and AMDAR observations near the core of the jet. In these quality control maps, green dots mark data that have been used, light blue for used but flagged (less weight for these observations), and red for rejected. In comparison with the 0600 UTC situation, the data is much more sparse, the analyzed jet is too weak (Table 1), and the ECMWF quality control has rejected two observations in what will turn out to be a key location (red dots near 45N, 35W in Fig. 2b).

GADS observations from flight data recorders, though not available in real time, provide information from 57 British Airways 747-400 aircraft which include routes that are outside the peak North Atlantic flow in time and space. For this case (1200 UTC on 25 December), Fig. 3 combines the ECMWF analysis, first guess minus regular aircraft observations vector wind errors (green, blue, and red), and a thinned (every third observation) version of the off-time GADS observations (black). Because the GADS flight (Mexico City to London) departs after the typical eastbound Atlantic flight, it reached the jet core during the 1200 UTC observational window and provides an independent check of the accuracy of the analyses.

The GADS observations (Fig. 4) clearly show that the hints of two separate peaks in the jet streak are much too smooth in the analysis with -7% to +10% amplitude discrepancies. The analyses in Fig. 4 have been interpolated to the exact aircraft height (37 000 ft, 217 mb) and are close enough to 1200 UTC to be insensitive to time interpolations. They come from GADS mode C observations which are slightly different than the mode A observations shown in Table 1. Note that the discrepancies are not local but extend over 25 degrees of longitude (about 2000 km at these latitudes),

In addition to problems with the jet stream strength there are also difficulties with a displacement of the jet axis and problems with the quality control limits when the wind veers slightly from the first guess. Two results lead to these conclusions. First, note in Fig. 3 the 8 northerly error vectors along the strongest upstream portion of the jet core between 45W and 35W compared with 2 southerly error vectors (12 to 7 between 50W and 30W). By contrast note the full angular range of error vectors near 58N. Second, note that the rejection of the two isolated aircraft reports with the strongest vector errors near 45N, 35W. A brief check of the actual observations shows that they appear legitimate. They were presumably rejected because at these wind speeds a relatively small error in wind direction leads to vector wind errors that are beyond the acceptance threshold.

Efforts at rerunning the analyses and forecasts with the GADS observations included are currently under way at (at least) ECMWF and NASA Goddard.

If these results and associated studies on the impact of the GADS winds on ECMWF forecast accuracy hold up, there would be real value in seeing that some of these off-peak time and off-peak route flights are added to the GTS. This suggestion is the commercial aircraft equivalent of generating flow dependent additional data via dropsondes. The essence of the idea would be to activate additional AMDAR transmissions via flight number and synoptic situation rather than just aircraft registration number. For reasons of cost, capturing all data from the 57 British Airways (~1 Million wind and temperature observations per day) 747-400's would not be feasible nor particularly useful. But "adaptive" activation of key flights may have a tangible effect on operational forecasts, especially over Europe.