ATC radar initiates a request to an aircraft for specific BDS data registers. If the aircraft is appropriately equipped, it will respond by broadcasting the requested register information. Each radar employs a distinct interrogation scheme that outlines the frequency of requests and the specific registers to be queried. Different countries or an Air Traffic Service Air Navigation Service Provider (ATS ANSP) may utilize varying interrogation protocols, including different frequencies and rates. The response sent by the aircraft is received by ATC radar but can also be received through a commercially available local receiver, as data are not encrypted. Unfortunately, it is more difficult to decode data received by these receivers as the type of register is not contained in the transmission, and hence fuzzy logic or other techniques are applied to decode the type transmitted properly . C-code software, developed in-house, performs this task (similar to the Python library developed by ).

Automatic Dependent Surveillance–Broadcast, or ADS-B, as its name suggests, allows an aircraft to broadcast aircraft state data using the transponder. Data are autonomously broadcast about every 0.5 s and contain the aircraft's onboard sensed position (through GPS and inertial systems), which is often more accurate than radar-derived position from ATC. These data are available to ATC but can also be displayed on the Navigation Display of newer aircraft for situational awareness. ADS-B does not broadcast wind and temperature, nor does it broadcast all required parameters to derive wind and temperature, although the difference between GNSS height and pressure altitude is transmitted frequently and could be used in data assimilation .

As outlined in the previous sections, EMADDC receives data in two ways: (1) aircraft data are collected by ATC, or (2) aircraft data are collected using a local ADS-B/Mode-S receiver. In both cases data are then forwarded to EMADDC through an FTP file transfer. New methods are currently being developed to enhance real-time data transfer, including NewPENS (the pan-European Network Service for real-time exchange of air traffic control data).

Data collected by ATC are of high quality as the contents are properly decoded since the content of each transmission is known to the interrogator. However for local receivers, the content of a received register is unknown, and logic is required to verify that a register is correctly decoded. All data have quality control and filtering applied.

ATC data can be of ASTERIX CAT48 format, which is mono-radar data, or CAT62 data from a radar tracker combining multiple radars. This latter data format uses filtering to sample all radar plots to a typical 4 s interval. The content of the formats CAT48 and CAT62 is similar, but the typical resolution of the Mach number in CAT62 is 0.008 (0.004 for CAT48), and as a consequence the derived temperatures are of lower quality (see Sect. ). EMADDC is working with EUROCONTROL Maastricht Upper Area Control Centre (MUAC) to develop a solution that provides the Mach number with a resolution of 0.004 and share this solution with other ATC providers.

An advantage of ATC radar data is that the individual BDS register messages corresponding to a single revolution of the radar are combined into a single “observation” message. For ATC radar, the position is determined by ATC radar. In contrast, for data received from local receivers, the position is decoded from the Compact Position Report (CPR) format, which is included in the ADS-B message (and hence not present in radar/tracker data as these data are not available in ASTERIX CAT48 or CAT62 data). The timestamp is not generated by the aircraft; instead, it is created by the radar or receiver at the time of reception. Local receivers utilize either a GPS antenna or a time server for time synchronization. Furthermore, the EMADDC system must combine the various BDS registers to derive observations, as previously outlined.

The techniques utilized by EMADDC are generating significant volumes of high-quality data from Mode-S EHS data. It is essential to implement various quality control that checks identify and address any observation discrepancies and assure that generated observations are of high quality.

A crucial measurement in any aircraft is the measurement of the airspeed, which can be obtained from the combination of a pitot-tube measurement and a temperature measurement. The pitot tube measures the static pressure ps and the total pressure pt . Both pressure observations suffer from inaccuracies related to, for example, a (small) angle between the flow and the probe . The Mach number is the true airspeed of the aircraft relative to the speed of sound. Let qt = pt-ps be the dynamic pressure, which is more accurately measured because the first-order error of pt and ps is cancelled. The Mach number M is defined as 1M=2γ-11+qtpsγ-1γ-1, where γ = cp/cv is the ratio of specific heats. Note that the dependence of M on the (inaccurate) ps remains.

The temperature is measured with a temperature probe . The measured total temperature Tt needs to be corrected to obtain the temperature T, 2T=Tt1+λγ-12M2-1. The true airspeed A can now be determined neglecting the effect of humidity, 3A=MγRdT, where Rd is the universal gas constant of dry air.

The (most relevant) parameters obtained through interrogation of Mode-S EHS radars are shown in Table . The timestamp is created at the moment of reception of the information. All parameters that originate from interrogation have an observation frequency depending on the radar; ADS-B information can have an observation frequency of twice per second. Table  provides information on the downlinked parameters. The information flow of the downlinked parameters is depicted in Fig. . Also shown in this figure are the corrections applied to the magnetic heading, true airspeed, and Mach number, discussed later.

The EMADDC quality procedure has been systematically developed and enhanced over the past decade. The first step in defining the quality is to check the input for obvious errors or measurements in conditions where calculation is not possible, as listed in Table . Measurements failing one of these checks are discarded from further processing.

Output control is necessary to obtain good-quality observations. The parameters for output quality control are related to the correction methods applied to the temperature and wind measurement.

Additionally, a safe list is created for wind speed and temperature independently, based on the difference between observations and forecast statistics. Aircraft with a 14 d wind standard deviation exceeding 4 kn are designated as ineligible, while for temperature, the standard deviation threshold is set at 1.23 K. EMADDC currently uses the operational ECMWF model with a minimal forecast lead time of 9 h for this comparison by collocating observations with numerical weather prediction (NWP).

Although the correction should be in the order of the (actual) declination, research showed that a simple correction is not enough . Each aircraft may use their own version of a declination lookup table, which implies that each aircraft corrects the true north to magnetic north in a different way. The correction method uses the assumption that the correction is determined by a geomagnetic reference table for a certain datum (or epoch) and is static until it is updated through aircraft maintenance. The optimal datum is found by minimizing a cost function, depending on datum, by comparing corrected winds from observations with NWP model forecast winds. The cost function is constructed by the vector length difference between the unit heading vector from the aircraft and the unit heading vector formed by the ground vector and NWP wind vector, that is 6δi(y)=cos⁡hNisin⁡hNi-cos⁡hmi+hcy,λi,ϕisin⁡hmi+hcy,λi,ϕi, where y is the date of the inclination table, i the index of an observation, hmi is the observed magnetic heading, and hcy(λi,ϕi) is the value of the declination table with datum y at location (λi,ϕi). NWP heading angle hNi is defined as 7hNi=atanGisin⁡ti-Visin⁡diGicos⁡ti-Vicos⁡di, with Gi the ground speed, ti the track angle, Vi the wind speed, and di the wind direction. The first vector in Eq. () is the NWP heading direction vector; the second is the corrected heading direction vector.

The cost function is defined as the sum of all vector length differences over all observations, that is 8C(y)=12∑iδi(y)29=12∑isin⁡hNi-sin⁡hmi+hci(y)2+cos⁡hNi-cos⁡hmi+hci(y)210=∑i1-cos⁡-hNi+hmi+hci(y). Next, magnetic declination is linearized with datum, in order to find a minimum, that is 11hci=H0i+y-yrefΔHi, where H0i is the value of magnetic declination for given location (λi,ϕi) on datum yref, and ΔHi is the value of the change in magnetic declination per year (this approximation is valid, as is discussed in Appendix ). The cost function is approximated by a quadratic function in the datum offset, which yields 12C(y)≈1-cos⁡H0i-hNi+hmi+δy∑iΔHisin⁡H0i-hNi+hmi+12δy2∑iΔHi2cos⁡H0i-hNi+hmi, where 13δy=y-yref. The (offset) datum value for which the cost function attains a minimum is found by setting the derivative to δy of the cost function to zero. Let x1i and x2i for observation i be defined by 14x1i=-ΔHisin⁡H0i-hNi+hmi15x2i=ΔHi2cos⁡H0i-hNi+hmi, then the datum minimizing the cost function is given by 16y=yref+∑ix1i∑ix2i. Using the value for an aircraft EMADDC calculates the corresponding magnetic declination tables at the location of an observation to find the declination and converts the reported magnetic heading to true heading and calculates the wind according to the equation above.

The above method uses NWP wind forecasts extracted from the operational ECMWF forecast (IFS model). The forecast lead time is at least 9 h, such that observations from EMADDC are not used as input for assimilation and correction simultaneously.

Figure  shows the results of the heading correction for all 19 006 aircraft operational in 2021–2023. Each aircraft is represented by a vertical line in the top panel. The white colour indicates that no correction was found and aircraft with a lot of white pixels are not flying regular in the EMADDC domain. The top panel shows the offset with the maximum reported heading correction in 2021–2023. In general, the change in heading correction over 2023 is small, except for aircraft that have large corrections. The most constant correction values (offsets smaller than 2 years, red- to brown-coloured) are found for values close to 2023, while higher offsets are only found with high maximum values, which gives us reason to believe that for these aircraft, the datum correction algorithm may not perform optimally. Note that aircraft near the right of Fig.  have declination tables decades into the future and are obviously not correctly estimated, which is also reflected in the noisy pattern for these aircraft. The corrections for these aircraft are invalidated, and the aircraft are marked as bad and will not be present in dissemination.

The magnetic heading is calibrated using NWP wind vector information. Consequently, the obtained correction depends on the quality of the NWP information. The magnitude of this dependence is rather small and of the order of 1 over the magnitude of ground speed, which is explained by the following.

Suppose we have a biased NWP wind direction, that is the true wind direction d is biased by β, then 17h̃N=atanGsin⁡(t)-Vsin⁡(d+β)Gcos⁡(t)-Vcos⁡(d+β)≈hN+βV-Gcos⁡(d-t)+VG2-2GVcos⁡(d-t)+V2, which implies that 18h̃N-hN⪅βV(V+G)(G-V)2⪅|β|G≪|β|. The offset from the true heading using wind-direction-biased information is substantially smaller than the actual bias.

The difference in heading from an unbiased and a biased NWP model is smaller than the bias divided by the ground speed (G). The ground speed is in general around 100 to 200 m s⁻¹, and thus the bias in heading is around 100 times smaller than the bias in the wind vector.

Similarly, the offset from the true heading based on wind speed biased (by α) is given by 19h̃N-hN⪅αGsin⁡(d-t)G2-2GVcos⁡(d-t)+V2⪅|α|G≪|α|.

When the model contains a wind speed bias, the bias in the heading is smaller than the wind speed bias divided by the ground speed and thus will be a factor of 10 to 100 smaller than the wind speed bias in the model.

Since the heading correction is based on many observations, over a large period (at least 15 d) it can be regarded as independent of the NWP information. Moreover, the EMADDC system is fitting a declination table over a large domain with a single parameter; the effect of locally biased or erroneous NWP estimates will be small. Aircraft visiting this region will have deviating quotients from which the datum is calculated (see Eq. ). This deviation is detected by checking the standard deviations of the quotients separately.

The measurement of true airspeed (TAS) depends on the temperature and Mach number; see Eq. (). Since the observed Mach number is corrected , the true airspeed measurement might be improved likewise.

The current algorithm uses an aircraft-dependent constant correction value. However, in the next update of the EMADDC processing (EMADDC 3.0), a true airspeed-dependent correction will be investigated. This improvement is reasoned by comparing the observed true airspeed with a NWP-model-equivalent true airspeed, thus creating a correction that depends on the observed value of true airspeed. The model true airspeed is obtained by the vector difference between the observed ground vector and wind vector from the NWP model. Figure  displays the average difference of the observed true airspeed versus the model-based true airspeed. The differences are averaged over true airspeed bins for all data points observed in the EMADDC domain in 3 d. Note that the difference in wind speed is of the order of a few tenths of a metre per second. Clearly there is a relation between mean airspeed difference and true airspeed itself.

As a first-order approach, the EMADDC system currently applies a true airspeed bias correction depending on aircraft and phase of flight. Future research will study physical methods of true airspeed correction.

The aircraft measurements are improved by algorithms on board the aircraft. The applied corrections are not available and may be aircraft-dependent, or aircraft-type-dependent, or both. It is known that the measurement of the static pressure ps suffers from airflow instabilities and/or angle of attack . The static pressure is corrected, which consequently results in a correction of the Mach number M and temperature T.

The temperature is calculated from the Mach number and the true airspeed (TAS) on board the aircraft. Actually, there are two types available of the Mach number, one being the downlinked data and one determined using indicated airspeed (IAS) information. The downlinked Mach number is of worse accuracy than the Mach number determined from the indicated airspeed, as will be discussed next. An estimate of the formal error of the temperature T calculated using Eq. () is constructed following 22σT2≈∂T∂A2σA2+∂T∂M2σM2=4T2A2σA2+4T2M2σM2. Rounding leads to an additional error of r/12, where r is the rounding (see Appendix ). One of the downlinked parameters is indicated airspeed, which is defined as the airspeed measured as if the aircraft were flying at mean sea level (p0 = 1013.25 hPa, T0 = 288.15, and ρ0 = 1.225 kg m⁻³, the standard pressure, temperature, and density at mean sea level) 23AI=2γ-1p0ρ0qtp0+1γ-1γ-1⇒qt=p0γ-12ρ0p0AI2+1γγ-1-1 The dynamic pressure qt can be calculated from AI, which in turn can be used to recalculate M, according to Eq. (), so an (first-order) estimate of the error in Mach number becomes 24σM2≈∂M∂qt∂qt∂AI2σAI225=AI2M2ρ02p21+qtp-2γAI2ρ0γ-12p0+12γ-1σAI2. Figure  shows the formal temperature errors as calculated with the downlinked Mach number (σT2(M), red line) and the temperature error based on the indicated airspeed (σT2(AI), blue line) and the formal error term related to the A. The effect on introducing the Mach indicated airspeed is a reduction in formal error of a factor of 4. The largest part in the formal error is (with the AI) now related to the true airspeed error. This implies that to further improve the temperature observation, reduction of true airspeed error needs to be accomplished.

The temperature correction is performed by first averaging, followed by correction. An average over 20 s is determined to reduce the noise in the temperature observation, that is 26T‾=1N∑iNTiI, and average observations are marked as bad when the standard deviation of the observations TiI exceeds 5 K.

The temperature correction applied is based upon correcting the static pressure measurements and recalculating subsequently the Mach number, the total temperature, and finally the temperature. The correction used is described in detail in . The corrected static pressure correction pc depends on the static pressure itself and the static pressure divided by the square of true airspeed, that is 27pc=a+psb+cA2, where a, b, and c are determined by fitting NWP temperatures. These coefficients are aircraft-dependent. The corrected temperature is determined as follows: first the corresponding total temperature is calculated, 28T‾t=1+λγ-12M2T‾, and then the corrected temperature calculated using the corrected pressure pc becomes 29T‾c=1+λγ-12Mc2-1T‾t, where 30Mc2=2γ-11+qtpcγγ-1-1.

Figure  displays the effect of the three steps (green “raw” temperature, improving M by the indicated airspeed (red), followed by the averaging (blue) and pressure correction (black)). Applying a simple smoothing of adjacent points, the standard deviation is reduced to values between 1 K in the mid-troposphere and up to 1.5 K at higher altitudes and near the surface (Fig.  compares red and black dashed lines); The bias (and standard deviation to a lesser extent) is then further reduced by applying the raw pressure correction method (Fig.  black lines).

Figure  displays the necessary steps needed in the processing to turn Mode-S EHS observations into meteorological information.

Wind and temperature observations are corrected using NWP information, as was discussed above. The heading correction applied may introduce some (NWP) correlation but not directly to wind components and temperature and when correlations exists it will be small (see Sect. ). Furthermore, corrections, as well as parameters for the safe list, are based on past NWP data, while the comparison is made using current (forecast) data.

From 1 January to 1 April 2024, a total of 4.5 billion observations were derived by the EMADDC system. From these observations 2.8 billion unique and allowed wind observations are made available to the users, and in total nearly 1.8 billion temperature observations are disseminated in these 3 months. The quality of wind observations compared to ECMWF-IFS is around 2.5 m s⁻¹ in standard deviation, with a small bias of 0.3 m s⁻¹; see Table . Table  also shows the statistics of wind observations for different height levels: the error in wind speed with respect to the model increases with height from 2.2 m s⁻¹ near the surface to 2.8 m s⁻¹ at a height of 11 km. The wind direction statistics show a different signal; near the surface the wind direction error is nearly 15°, with a minimum at around flight level 350, increasing again to an error of 10° at around 11 km. Note that only observations with wind speed larger than 4 m s⁻¹ are used for wind direction. Wind is in general more variable near the surface. These values are all within the acceptable range for use in for example data assimilation. According to the WMO Rolling Review of Requirements, the requirements for temperature used in high-resolution NWP are 0.5 K (goal), 1 K (threshold), and 3 K (breakthrough) and for wind speed 1, 2, and 5 m s⁻¹, respectively .

The temperature statistics are shown in Table . The temperature error in total is slightly smaller than 1 K, with a minimum error of 0.8 K around flight level 250. The maximum error is found at cruising level (11 km). Note that the bias with the model is around zero.

Radiosondes are regarded as anchor observations for meteorology and are generally launched 45 min prior to the main synoptic hours 00:00 and 12:00 UTC to achieve a profile centred around 00:00 UTC (12:00 UTC). Note that a few sites also launch at 06:00 and 18:00 UTC. Due to budget restrictions some radiosondes are only launched once a day. Table  shows collocated observations statistics of wind (speed and wind components). Table  contains temperature statistics. Aircraft and radiosonde observations will never be exactly collocated in both space and time. Nevertheless, collocations are made by having the maximum distance between aircraft and radiosondes of at most 50 km, a maximum height difference of 100 m, and a time difference of 1800 s.

For all wind parameters, the comparison between radiosonde and EMADDC mostly has a standard deviation lower than that of the comparison model and EMADDC or radiosonde. Furthermore, the difference between model and radiosonde or model and EMADDC is similar and of the order of 0.3 to 0.5 m s⁻¹, while the mean difference between aircraft and balloon is small. The reason for this is the model representation error due to the model grid size. The temperature statistics show that all three systems have the same average temperature (all biases are small and near zero). Not surprisingly, the temperature observations of EMADDC are clearly of less quality than radio soundings, although above 850 hPa the quality is reasonable.

Finally, a comparison is made between EMADDC observation and AMDAR observations. AMDAR observations are extracted from the onboard computer and should have a large resemblance to EMADDC observation, and thus the mutual statistics are expected to be small. Figure  shows the statistics of EMADDC, AMDAR, and NWP for temperature and vector RMSE. A lookup table is used to connect AMDAR aircraft to an ICAO aircraft identification. This lookup table is partially based on E-AMDAR information and is completed with results from collocation with EMADDC observations. Due to rounding of position and time, exact collocations of AMDAR and Mode-S can be tedious. Moreover, the reporting observation frequencies are different, resulting in about half of the AMDAR observations being collocated. Here an AMDAR observation is collocated with an EMADDC observation when the time difference is less than 2 min, the distance is less than 1 km, and most importantly the height difference is at most 25 ft. The height discrimination is strong because in general temperature and wind tend to change more with height than with horizontal displacements.

The temperature of AMDAR has a positive bias when compared to NWP, which is known in the literature . The bias of EMADDC is around zero. The AMDAR temperature standard deviation is clearly better than EMADDC. The increase in standard deviation near the ground is due to atmospheric turbulence being more present near the surface . The EMADDC temperature error increases at lower levels because aircraft land generally with a low airspeed and Mach number, which results in a larger error due to the rounding of the airspeed observation given the relatively low resolution of these parameters.

For wind the vector RMSE shows that both AMDAR and EMADDC perform equally well; there are small differences but not significant. However, a gross error check on the zonal wind is needed to detect systematic errors in B787 aircraft .

The EMADDC team tries to improve the quality of derived wind and temperature continuously. The team has the following items to investigate or implement:

The current heading correction algorithm does not detect changes in the heading table datum effectively, especially when high datum correction values are detected. A revisit of the heading correction algorithm is foreseen in the near future.

In the future ADS-B is foreseen to broadcast meteorological information creating enormous data volumes which are expected to require quality control of some kind for use in meteorological applications. The EMADDC system could fulfil the future quality assurance function.