Exploring online aircraft metadata

Bruce Ingleby (ECMWF), Mickey Yun Chan (Latvia University of Life Sciences and Technologies), Mohamed Dahoui (ECMWF)


ECMWF is working towards improving the use of meteorological observations from aircraft by adding information such as aircraft type and airline. A matching between meteorological observations and online data was achieved using a method developed as part of the ECMWF Summer of Weather Code programme (ESoWC 2019, see ECMWF Newsletter No. 161).

How it works

Biases in aircraft observations of temperature are known to depend on both aircraft type and airline, due to different avionics systems employed by the various airlines and aircraft fleets. In a few cases, different avionics can also result in gross errors in wind observations. The idea behind the method developed during ESoWC 2019 is to use aircraft type and airline metadata information available online to improve the assimilation and monitoring of aircraft weather reports.

ECMWF currently receives about 840,000 aircraft‐based observations per day from around the world, most of which are AMDAR reports. The wind, temperature and (in some cases) humidity data provide valuable input to our forecasting system. When the main AMDAR programmes started several decades ago, the airlines and pilot associations insisted on anonymisation of aircraft identities. As a result, while each AMDAR report includes an ‘identifier’ unique to the aircraft which produced it, in most cases the type of aircraft and the airline cannot be deduced directly from the identifier. However, about 15 years ago online aircraft tracking systems started to provide information on flights using a mixture of crowdsourced radio messages and information from the airlines. The method developed during ESoWC 2019 matches aircraft metadata from those online resources to the aircraft identifiers included in AMDAR reports. Online data from the flightradar24 and flightaware websites essentially provide take‐off and landing times and the airports involved for each aircraft. AMDAR reports are sorted by identifier and time, the first and last report in each flight are identified and the position is used to find the nearest likely airport. The two data sources are then matched together in order to link AMDAR aircraft identifiers with aircraft type and airline metadata. AMDAR reports from European and other airlines sometimes start and stop in mid‐air, making the deduction of the airport where the aircraft came from or are flying to difficult or impossible. It is also more difficult where several airports are close together. A minority of AMDAR reports contain airport information, which makes matters easier. Despite the problems, useful progress has been made.

As an example, the figure shows schematically in black the movements of one aircraft over about six days, as available from flightradar24. This has been matched with the estimated movements of an aircraft associated with a particular AMDAR identifier. For non‐US AMDARs, about half of the AMDAR identifiers have been matched with aircraft metadata in this way with reasonable confidence. For the USA, the figure is about 15%. Interestingly, the South Korean AMDAR programme uses tail number as the AMDAR identifier, so all their AMDAR identifiers can easily be matched with aircraft type and airline.

Identifying an aircraft from its movements. The figure shows the movements of an aircraft for about six days. The red writing shows estimated landings and take-offs according to time and location data included in AMDAR reports for a specific aircraft (AMDAR identifier EU1006). The method developed during ESoWC 2019 made it possible to match this estimated trajectory with online information (from the website) on the trajectory of an Air France Boeing 777 (tail number F-GSQH), shown in black writing. The location discrepancies are a result of the fact that first/last AMDAR reports are not necessarily sent during ascents/descents but may be sent while the aircraft is closer to another airport, for example Belo Horizonte instead of São Paulo or Lyon/Brussels instead of Paris.


In view of the information available from flight tracking websites, the anonymisation of meteorological reports seems rather outdated. It may also make the forecasts available to airlines worse than if the metadata were available. While the matching process described above could be improved (and use of the OpenSky Network website investigated), a more long‐term solution may be to engage in dialogue with the airlines to try to obtain the required metadata directly from them. The US National Oceanic and Atmospheric Administration (NOAA) and EUMETNET (a consortium of European national meteorological services) would also be involved. However, it is likely that this process will take some years.

In 2020, the World Meteorological Organization and the International Air Transport Association (IATA), an association of the world’s airlines, are expected to start implementing a global AMDAR development programme (WICAP), which should boost airline participation and hence the number of AMDAR reports in the next few years. Another recent development is the use of air traffic management Mode‐S reports to provide very high‐density data (particularly winds) over parts of Europe and potentially elsewhere. Mode‐S data are not yet processed at ECMWF, but in collaboration with EUMETNET and our Member States we are starting to plan the steps towards usage. These issues and others will be discussed at a workshop on Aircraft Weather Observations and their Use to be held on 12 and 13 February 2020, organised by EUMETNET and ECMWF. More details are available on the ECMWF website: