Quality assessment of Automatic Dependent Surveillance Contract ( ADS-C ) wind and temperature observation from commercial aircraft

Aircraft observations of wind and temperature are very important for upper air meteorology. In this article, the quality of the meteorological information of an Automatic Dependent Surveillance-Contract (ADS-C) message is assessed. The ADS-C messages broadcast by the aircraft are 5 received at air traffic control centres for surveillance and airline control centres for general aircraft and dispatch management. A comparison is performed against a global numerical prediction (NWP) model and wind and temperature observations derived from Enhanced Surveillance (EHS) air-traffic 10 control radar which interrogates all aircraft in selective mode (Mode-S EHS). Almost 16,000 ADS-C reports with meteorological information were compiled from the Royal Dutch Airlines (KLM) database. The length of the data set is 76 consecutive days and started on 2011/01/01. The wind and 15 temperature observations are of good quality when compared to the global NWP forecast fields from the European Centre for Medium-Range Weather Forecasts (ECMWF). Comparison of ADS-C wind and temperature observations against Mode-S EHS derived observations in the vicinity of Ams20 terdam Airport Schiphol shows that the wind observations are of similar quality and the temperature observations of ADS-C are of better quality than those from Mode-S EHS. However, the current ADS-C data set has a lower vertical resolution than Mode-S EHS . High vertical resolution can 25 be achieved by requesting more ADS-C when aircraft are ascending or descending, but could result in increased data communication costs.


Introduction
Aircraft related observations are widely used for numerical weather prediction (NWP).Aircraft Meteorological Data Re-Correspondence to: Siebren de Haan (Siebren.de.Haan@knmi.nl)lay (AMDAR) and Mode-S Enhanced Surveillance (EHS) have shown to be beneficial when used for initializing an NWP model (Benjamin et al., 2010;de Haan and Stoffelen, 35 2012).AMDAR observations are generated especially for the meteorological community and are downlinked when the aircraft is in the vicinity of a ground station.Wind information is derived from the information in the flight management system, and temperature is measured by on-board equipment 40 (Painting, 2003).Designated aircraft provide AMDAR observations created using software (and in some instance by hardware) which provides additional quality and enhancement of the observations (Painting, 2003).The meteorological messages are send to the meteorological community 45 through ground stations (e.g.located at aerodromes).An advantage of AMDAR is that the coverage includes data sparse areas such as over the oceans.However, intercontinental flights are using almost the same routes, leaving some areas unsampled.Moreover, these long-haul flights will pro-50 vide observations at high flight levels only (approximately at 10 km altitude).Mode-S EHS wind observations are gathered by using the selective mode (Mode-S) of an enhanced surveillance air traffic control (ATC) radar (de Haan, 2011).All aircraft in view of the radar are interrogated individu-55 ally very frequently (at Amsterdam Airport Schiphol every 4 seconds) and reply to the request by downlinking information from which wind and temperature information can be inferred.Using the ATC radar at Schiphol airport, de Haan (2011) showed that the wind information from this source 60 has a quality comparable to AMDAR wind observations.The downlinked information consists of Automatic Dependent Surveillance Broadcast (ADS-B) messages.The ADS-B system is initially developed to prevent aircraft collisions by constantly broadcast of the aircraft position.As such, 65 ADS-B is FAA's satellite-based successor to radar positioning.ADS-B makes use of Global Navigation Satellite System technology to determine and share precise aircraft location information, and streams additional flight information to : cockpits of other properly equipped aircraft.Additionally, more information on flight path can be requested by other aircraft or ATC radars.
In this paper a different type of surveillance technique is used to extract meteorological information from commercial aircraft.ADS-C (Automatic Dependence Surveillance Contract) differs from ADS-B in that it can only be initiated by registered users.Moreover, ADS-C is a surveillance technique in which aircraft automatically provide on request, via a data link, data derived from on-board navigation and position-fixing systems, including aircraft identification, four-dimensional position and additional data as appropriate.Specific ADS-C messages may contain meteorological information obtained from the Flight Management System (FMS), when asked for.
ADS-C messages are different from AMDAR and Mode-S but contain the same type of information.In case of Mode-S the information can be identical to an ADS-C message, because Mode-S is related to ADS-B.ADS-B, ADS-C and AMDAR wind and temperature observations are based on direct read outs from the FMS, while AMDAR uses its own algorithms (Painting, 2003).Mode-S, on the other hand requires an additional calibration and correction step (De Haan, 2011).Figure 1 shows schematically the information flow of  equipped aircraft.This data block consists of aircraft identification, position, time and flight level.Additional ADS data blocks can be included as necessary.Table 1 shows the different data blocks of an ADS message.
In addition to the transmission of the ADS-C data block a) and b) for air traffic surveillance purposes, the Meteorological information data block may also be requested.The Meteorological information group f) may be requested to satisfy conditions specified in ICAO.Annex3 (July 2010), Section 5.3.1, air traffic management applications, or airline monitoring systems, etc.
The ADS-C data contains a large number of parameters; here attention is paid to atmospheric parameters wind and temperature.This information is made available from the flight management computer.The temperature is measured directly but the wind speed and direction is inferred from the ground track and the speed (and direction) of the aircraft relative to air.In total 71832 ADS-C messages were collected in the period from 2011/01/01 00:13 UTC to 2011/03/17 14:30 UTC.In total 15995 ADS-C messages contained meteorological information and 5818 messages air vector information; 4934 messages contained both ADS-C types d) and f). Figure 2 shows the coverage of the data set used in this study.

Direct Wind observations from ADS-C
The ADS-C messages of the meteorology-group contain information on wind speed and direction and temperature (see Table 1, Report f) ).The vector difference between the ground track and the direction and speed relative to the air is the wind vector; the aircraft has to correct for the wind to fly along a desired ground track.In reverse, when the air vector V t and ground speed vector V g are known the wind vector V can be calculated: The vector V is reported in the ADS-C meteorological information group.These wind observations will be called Direct ADS-C wind observations, in the following.

Derived Wind observations from ADS-C
There are also ADS-C message (of the type 'Ground Vector' and 'Air Vector', Table 1, Report c) and d) ) which contain the ground track information, the heading and the Mach number but no direct atmospheric information.Wind infor-215 mation can not be inferred directly because an estimate of the airspeed is missing.However, with additional temperature information, wind information can be obtained since the Mach number is the quotient of the airspeed and the speed of sound and the latter is dependent on the temperature through, where γ = 1.4, the adiabatic index, R = 8.3145 J mol −1 K −1 molar gas constant, T temperature and M = 0.0289645 kg mol −1 molar mass of dry air.V can be expressed as where φ h is the heading of the aircraft with respect to true north.Errors due assumptions in T are Suppose that the error in T is 1 K this will result in an 230 error of less than 0.5% in airspeed.Because of the linear :  relationship between wind and airspeed, this temperature related error is thus also small.The wind observations obtained using ground track vector, the air vector and additional temperature information from NWP are called Derived ADS-C observations, in the following.

ECWMF NWP Data
The   and Mode-S will differ.A match between ADS-C and Mode-S is found when the distance between the observation locations is less than 20 km.
The ECMWF temperature is used to calculate the wind vector when the ADS-C report contains Mach and heading.

260
The estimated ECMWF temperature error in the upper air is less than 1K.

Temperature
Table 3 shows the statistics of the comparison of temperature of ECMWF and ADS-C and the statistics of the triple 265 comparison of ADS-C, Mode-S and ECMWF.In total 15995 direct ADS-C observations are used for the global comparison, while only 67 direct ADS-C observations were reported in the vicinity of Amsterdam Schiphol airport for triple comparison with Mode-S and ECMWF.

270
The global ADS-C temperature data set has a bias of around -0.5 K and a standard deviation of less than 1 K when compared to ECMWF.The mean ECMWF temperature is 224 K, indicating that the average observation height is around 200 hPa (see also Figure 2).These statistics are similar to those found for AMDAR observations (Schwartz and Benjamin, 1995;Drüe et al., 2007Drüe et al., , 2010;;de Haan, 2011).
Nearly the same statistics are found when 67 ADS-C observations near Schiphol airport are compared to ECMWF.The Mode-S temperature observations are known to be more 280 noisier due to the method of derivation of temperature from a Mach number (De Haan, 2011).Both ADS-C and Mode-S have a bias of around 0.7K with ECMWF, while between each other almost no bias is present.This is most likely related to the fact that the observations, although derived dif-285 ferently, are based on the same measurements.Note that the mean ECMWF temperature is around 241K, which is at approximately at 500hPa.The ADS-C temperature observations are of good quality, comparable to AMDAR, and better then Mode-S temperatures.

Wind speed and direction
Wind observations from ADS-C can be obtained in two different ways.Either it is observed directly or it is derived from the track vector and air vector of the aircraft (with additional temperature information, see Section 2.2.2).The number 295 of direct wind observations are 15995 (the same number as for temperature observations), while the number of derived ADS-C wind observations is 5818.From these 5818, in total 4934 have also direct wind measurements.In total 67 direct ADS-C wind observations are in the vicinity of Schiphol air-300 port from 13 ascending or descending aircraft; the number of derived ADS-C wind observations near Schiphol airport is 35 (7 profiles).
Wind speed biases from direct measurements are of the order of 0.5 m/s and standard deviation is around 2.8 m/s as 305 presented in Table 4. Derived wind speed biases and standard deviations are of the same order, however, the data sets sample different parts of the globe and atmosphere as can be seen from the difference in mean ECMWF wind speed and wind direction for the data sets.

310
The wind direction statistics are calculated on a subset of the data sets by excluding observations for which the ECMWF wind speed was less than 4 m/s.For wind direction the direct measurement wind direction bias against ECMWF is small with a standard deviation of less than 10 degrees 315 (Tab.4).Note that the mean wind direction is south-west.The statistics for derived wind measurements also show a small mean difference with ECMWF.The standard deviation however, is 13 degrees which is larger than the standard deviation of the direct wind direction standard deviation.The   In total 4934 ADS-C observations reported both direct wind and temperature as well as ground track vector and air vector.With temperature information, the wind vector can be derived when ground track and heading are present.In Table 5 the statistics are shown for these 4934 observations.Clearly, the quality of both types are very close.The biases and standard deviations between the observations and ECWMF are almost equal.The mean wind speed and direction difference between the two ADS-C observations types are very small, with small standard deviations.

Profiles of Wind and Temperature
In Figure 4 the profiles of all ADS-C reports with both direct and derived wind observations are shown for 7 profiles of 5 360 different aircraft.Mach number and heading are shown for each profile in the left panel; wind speed and direction are depicted in the right panel.Also shown are ECMWF data (solid lines) and Mode-S data (dashed lines).
In some instances Mach number and heading compare 365 well.This is not surprising since both observations are observed by the same instruments but can be a few seconds apart, since the observation frequency of Mode-S is 4 seconds.Consequently, the derived wind observations match the Mode-S wind observations.Also, the direct ADS-C re-370 ports of wind are close to the Mode-S and derived ADS-C wind observations.Note that the ECMWF profile is very smooth compared to Mode-S wind observations.The Mode-S profile matches ECMWF closely when the vertical wind variability is small (panels c) and e)).The other panels show 375 more wind variability.For example panel g) shows a very smooth Mach number and heading profile while the wind speed shows more small scale variability.Note the large difference in wind speed with ECMWF below FL50, observed by ADS-C and Mode-S.ilar to AMDAR and Mode-S EHS observations.However, the vertical sampling rate in the present data set is less than Mode-S.Note that the vertical sampling rate is highly correlated with the temporal sampling during ascent or descent of an aircraft.

Conclusions
In this article the quality of meteorological information inferred from ADS-C reports is assessed by comparison with ECMWF and Mode-S wind and temperature information.
The data set contained 16,000 temperature and wind data points and nearly 6,000 Mach number and heading data points.From the latter data set, using additional temperature information, from for example ECMWF, wind vectors can be derived.
The direct temperature and wind observations are of the same quality as AMDAR when compared to ECMWF.The ADS-C temperature observations are of better quality (lower standard deviation) than Mode-S.Wind observations from direct ADS-C reports and derived ADS-C reports have simi-400 lar quality.Both types of ADS-C wind observations compare reasonably well to Mode-S wind observations, although the number of comparisons is small.
The general quality of ADS-C wind and temperature observations is comparable to AMDAR observations; the mea-405 surement uncertainty of AMDAR temperature is approximately 1K (Schwartz and Benjamin, 1995;Drüe et al., 2007Drüe et al., , 2010)).Benjamin et al. (1999) found an observation error wind component 1.1 ms −1 and 0.5 K for temperature above the boundary layer.Within the boundary layer larger values 410 were found.In Ballish and Kumar (2008) it was shown that temperature observation from AMDAR exhibit a considerable variation with aircraft model and are on average warmer than radiosondes.Drüe et al. (2010) found in a comparison of AMDAR with a wind profiler radar a wind vector differ-415 ence of approximately 2.5 ms −1 .Furthermore, Drüe et al. (2007) also showed that systematic deviations in AMDAR wind measurements can be regarded as an error vector, which is fixed to the aircraft reference system.atic deviations in wind measurements from different aircraft 420 types (more than 0.5 ms −1 ) parallel to the flight direction.A intensive ADS-C comparison study over a longer period is needed to investigate whether this aircraft dependecy is also present for ADS-C measurements.
Mode-S wind information is available with a temporal resolution of 4 seconds, while ADS-C reports are less frequent.
Because of this difference in temporal resolution profile in formation from ADS-C is limited.
In conclusion, the accuracy of ADS-C meteorological information as observed in this article is of good quality and can be valuable source of wind and temperature information for operational weather forecasting and assimilation in NWP models.

Fig. 1 .
Fig. 1.Information flow for a) ADS-C and AMDAR using ACARS network, and b) Mode-S EHS observations only available in the vicinity of an Mode-S EHS tracking radar. 210

Fig. 2 .
Fig. 2. Horizontal and vertical distribution of ADS-C messages.
Direct and Derived ADS-C observations are compared to the operational global NWP model from ECMWF.Satel-lite, radiosonde and aircraft observations are the main input 240 for upper air initializtion at analysis time.The resolution of ECMWF-model was reduced to 1 degree due to computational limitations, with 91 vertical levels.Because the operational model is started every 12 hours, observations are compared to at least a 12 hour forecast.ECMWF wind and 245 temperature from the model are linearly interpolated bilinear in position and linear in time between two successive forecast.These forecasts are 3 hours apart, with a maximum

3
Quality Evaluation of ADS-C observations by comparison with ECMWF and Mode-S The quality of the ADS-C messages is compared to ECWMF model data and to Mode-S observations in the vicinity of the Amsterdam Schiphol airport.Because the time of observation differ for Mode-S and ADS-C, the positions of ADS-C 290 320 mean ECMWF wind direction for the derived wind data set is north-west which differs by 30 degrees from the direct observation data set; the data sets sample different region and times.The 67 direct ADS-C observations in the vicinity of 325 Schiphol airport show nearly the same wind statistics as the global direct data set.Mode-S versus ECWMF has a similar bias and a slightly larger standard deviation for wind speed than ADS-C.Figure 3 (top row) shows the scatter plots of temperature and wind for direct ADS-C and Mode-S versus ECMWF.The statistics for the 35 data points show that the bias and standard deviation of the ADS-C and Mode-S wind speed observations compared to ECMWF are similar, with Mode-S having a slightly smaller standard devia-: tion.The wind speed standard deviation of the difference between ADS-C and Mode-S is around 1.6 m/s, approximately half the standard deviation of observation versus model.The Mode-S and ADS-C observations are not exactly at the same position and therefore part of the error of the difference is related to difference in position.The statistics of the wind direction are similar.Mode-S versus ECMWF wind direction standard deviation is slightly larger that that of ADS-C versus ECMWF.The wind direction observations are close to each other indicated by the small standard deviation.The bottom row in Figure3shows the scatter plots of wind speed and direction of the derived ADS-C and Mode-S observations versus ECMWF.

380Fig. 3 .
Fig. 3. Scatter plots showing ECMWF temperature and wind versus ADS-C (direct and derived) and Mode-S.

Fig. 4 .
Fig. 4. Profiles of Mach number, heading, wind speed and wind direction for ADS-C reports (direct and derived) in the vicinity of Schiphol airport.Left panel of each sub graph shows the mach number and heading (solid and open squares, resp.);right panel shows wind speed and direction (solid and open triangles, resp.).Also shown are Mode-S heading, Mach number and wind speed and direction (dashed line) and ECMWF wind speed and direction (solid line).

Table 2 .
Example of an ADS-C message.

Table 3 .
Statistics of the comparison of temperature observations from ADS-C versus ECMWF for the whole set, and triple comparison for ADS-C observations, Mode-S and ECMWF in the vicinity of Amsterdam Schiphol airport.

Table 4 .
Statistics of the comparison of wind observations from ADS-C versus ECMWF.

Table 5 .
Statistics of the comparison of wind (direct and derived) observations from ADS-C versus ECMWF.