The first NARVAL campaign (NARVAL-South) took place between 10 December 2013 and 22 December 2013. This campaign and the campaign that followed directly after, NARVAL-North, were first and foremost planned as a demonstrator mission to assess the capabilities of HALO and HAMP to work as an airborne cloud remote sensing platform. Additionally, the focus of this campaign was to assess the representativeness of ground-based measurements on Barbados for a broader trade wind region over the tropical Atlantic and to evaluate satellite observations on collocated tracks . This was done by flying underneath the satellite, along the projected satellite track near the satellite overpass time. In total, eight flights amounting to about 67 flight hours were conducted. Of the eight flights, four flights were cross-Atlantic transects during transfers between Oberpfaffenhofen, Germany (OBF), and Grantley Adams International Airport, Barbados (BGI), and four were local flights over the tropical North Atlantic east (and upwind) of Barbados. The overall flight time for individual flights was around 10 h for transatlantic flights and around 7 h for local flights. Target areas of this campaign were the trade wind region east of Barbados as well as cross-Atlantic transects. The synoptic situation was relatively constant from flight to flight. Therefore, flight planning was done mainly in the interest of gathering statistically sound data and to compare with satellite data. A-Train collocations were achieved during seven flights. All flights and their aims are listed in Table , and flight tracks are shown in Fig. a.

Following directly after the first campaign, NARVAL-North took place from 7 to 22 January 2014. The center of operation was at Keflavik International Airport, Iceland (KEF). During this campaign, seven flights were conducted, overall amounting to 46 flight hours. The target area was the extratropical North Atlantic south and southwest of Iceland. The aim of NARVAL-North was to analyze postfrontal convective regimes of midlatitude cyclones over the extratropical North Atlantic and to investigate the accuracy of existing satellite precipitation climatologies . Flight planning was done taking into account current synoptic situations. The flights mainly sampled the cold sector of midlatitude cyclones as well as some occlusions. A-Train collocations were achieved during four flights. Figure b shows the flight tracks of NARVAL-North, and in Table  all flights and their aims are listed.

NARVAL2 followed up on NARVAL-South with a stronger focus on the local region around Barbados and over the tropical Atlantic between 8 and 30 August 2016. Ten research flights were flown for a total of about 84 flight hours (Table ) with different flight tracks (Fig. c). The center of operation was again BGI on Barbados. A-Train collocations were achieved during five flights as well as collocations with the satellites GPM (Global Precipitation Measurement) and Megha-Tropiques. The aim of the campaign was to assess the interaction between large-scale dynamics and the evolution of convective systems over the tropical Atlantic . Of the 10 flights during this campaign, 2 flights were transfers to and from Barbados. The remaining flights took place over the tropical North Atlantic east and southeast of Barbados. Three of these flights focused on clouds in and associated with the Intertropical Convergence Zone (ITCZ). The other five flights took place further north and mainly over shallow convection. Two flights focused on the surrounding atmosphere of Hurricane Gaston (2016). However, due to instrument failure, no radar measurements are available from these two flights. A special flight pattern, i.e., circles with a radius of about 110 km with frequent dropsonde launches, was performed in order to measure the large-scale vertical motion on six flights.

The fourth campaign of this set, NAWDEX, took place from 17 September until 18 October 2016. Thirteen flights amounted to about 96 flight hours. During this campaign, HALO, along with other research aircraft, was stationed at KEF on Iceland. Figure d shows the flight tracks of NAWDEX. A-Train collocation was achieved during one flight. Of the 13 flights during this campaign, two were transfers to and from Keflavik. The main target of the flights during this campaign was the clouds associated with the warm conveyor belt of midlatitude cyclones . Flight planning was done taking into account current synoptic situations. In Table , all flights and their aims are listed. The flights mainly sampled the warm sector and frontal systems of midlatitude cyclones.

The HALO aircraft is a Gulfstream G550 with a ceiling altitude up to 15 km and long endurance of up to 10 h . HALO's capabilities of high ceiling and long range enable researchers to cover a wide variety of atmospheric conditions during a single flight and to fly above the main airline traffic and most cloud systems. Thus, measurements on board HALO often provide a view of at least a large part of or even the entire troposphere through measurements. The median flight altitude was 13.1 km for the NARVAL-South campaign and 7.8 km during NARVAL-North, taking into account the lower tropopause height of the target region in winter and flight restrictions. During NARVAL2, the median flight altitude was 9.7 km, and during NAWDEX it was 12.4 km. The lower altitudes during NARVAL2 were flown to accommodate for the fact that dropsondes were launched in quick succession during some of these flights (see Sect. ).

The HALO aircraft is equipped with the Basic HALO Measurement and Sensor System (BAHAMAS). This instrument system provides data about the aircraft's location and attitude (position, altitude, heading, roll and pitch angle) and atmospheric measurements at aircraft level (e.g., temperature, humidity, pressure) . During the first two campaigns, data were sampled at 1 Hz, while, in 2016, 100 Hz data were also available. In this data set, however, the 1 Hz data are provided for all flights, since these correspond best to the sampling rates of the other instruments. Measurements of aircraft location (latitude, longitude, altitude above WGS84 ellipsoid) and attitude (angles of roll, pitch and heading) are provided in this data set. Note that most measurements were performed in straight legs or large circles with only 6.72 % of the measurements having a roll angle of more than 5∘.

The median speed above ground was 224 m s-1, and therefore measurements with a resolution of 1 s roughly represent a distance of about 200 m. Typically, it took 35 min to reach flight level after takeoff and 30 min for the final descent.

The HALO Microwave Package HAMP, comprises microwave radiometers with 26 channels in the range between 20 and 183 GHz and a 35 GHz cloud radar. The radiometer modules and radar antenna are mounted below the fuselage inside the belly pod (see Fig. 1 of ) with a nadir viewing direction. illustrate the sensitivity of the different measurements with respect to hydrometeors. Besides the HAMP data, profiles from dropsonde measurements and auxiliary data from the aircraft's basic data system are also included in the published data set described here. The specifications of these instruments are listed in Table  and are described in the following sections.

The radiometers were calibrated on the ground before almost every flight using the manufacturer's warm and cold load calibration method. In this method, each radiometer was pointed successively on a warm black-body target at ambient temperature and on a cold black-body target cooled down to the boiling point of liquid nitrogen (LN2). Targets with an open air–LN2 interface were used during the NARVAL1 campaigns. Later, these were exchanged with targets embedded in a foam box that is transparent to microwaves, avoiding reflections at the LN2 interface. The new targets were vertically oriented, and a metal mirror was used to redirect the view from nadir to horizontal. During flight, the radiometers were continuously calibrated using two reference loads. More details are given by .

The quality of the original brightness temperature measurements was evaluated by comparing them with synthetic ones simulated from dropsonde profiles of clear sky. For this comparison, the sondes were filtered for clear-sky conditions with low atmospheric variability, and their thermodynamic profiles were fed into the Passive and Active Microwave TRAnsfer model (PAMTRA). Random discrepancies between both can be attributed to the matching of HAMP nadir measurements and the drifting sondes. The systematic differences could be identified and cannot be explained by systematic errors of the drop sonde measurement nor the microwave absorption model. Most likely they result from changes within the belly pod during takeoff. Because some differences in the biases between different flights could be found, a bias correction based on the mean differences between synthetic and measured brightness temperatures was performed. However, for some flights only very few or no drop sondes are available. In order to arrive at a robust correction, those flights having three or less dropsondes were corrected using the campaign mean. For further information, the offset corrections used for each channel are included in the data files.

Some obvious errors in radiometer measurements occurred during the campaigns. During the NARVAL-South and NARVAL2 campaigns, the 183 GHz radiometer module suffered from instabilities during the initial phase of some flights. Because of strong drifts, the regular gain calibration (performed approximately every 5 min) caused a sawtooth pattern whose amplitude decreased with time as the module became more stable. After NARVAL2, a broken dielectric resonator oscillator (DRO) was replaced. Furthermore, in some instances, unreasonably high or low values were recorded by different modules likely due to instabilities in data transfer. This is also thought to be the reason that time stamps sometimes did not increase continuously but jumped ahead or backwards in time.

All data were inspected thoroughly by eye, and errors of brightness temperature from sawtooth patterns or spikes in data were not corrected but removed from the data set. The erroneous time stamps were reconstructed where possible. If the interval with faulty time stamps was too long and reconstruction was not possible, the data were removed. Additionally, measurements during turns (roll angle >5∘) or when the aircraft was below 6 km altitude were removed.

By comparison with airborne and spaceborne radar measurements at 95 GHz, it has been observed that HAMP radar reflectivity was too low by 8–10 dBZ. To assess the exact value of this offset, derived a calibration value for the HAMP cloud radar reflectivity. To this end, calibration maneuvers with constant bank angles were flown during NARVAL2 and NAWDEX while the radar emitting power was reduced to use the returned sea surface signal. Additionally, individual instrument components were measured to calibrate the cloud radar. This calibration was validated with other airborne and spaceborne measurements. concluded that the resulting bias of 7.6 dBZ originated from differences in software configuration and instrument calibration. This value is constant over the entire measurement range and was added to all reflectivity measurements in this data set.

The radar data include some features that might not be desirable to use: in the beginning of flights and sometimes also around measurement interruptions when data were recorded but no radiation was emitted, the data include noise. Further, on a couple of flights, calibration maneuvers for the radar were executed (Sect. ). During these maneuvers, the transmitting power of the radar was changed and thus the received signal also changed. Additionally, especially when the aircraft overflew land, the unified radar data contain pixels that are at or below the surface. All of these cases – surface and subsurface, sea surface, noise, intervals with calibration maneuvers – were marked in an additional data flag variable indicating the state of the radar data: data ok (0), noise (1), surface or subsurface (2), sea surface (3), radar calibration (4). For additional information, a variable that indicates a turning of the aircraft (roll angle >5∘) was added to the data set.

To ensure comparability of measurements of a certain time interval, a good temporal collocation of the instruments is needed. Temporal collocation of HAMP measurements with HALO's onboard instrumentation (BAHAMAS) was performed independently for radar and microwave radiometers. A good synchronization with BAHAMAS lets us then assume that also radar and radiometer measurements are synchronized well.

Temporal collocation between the cloud radar and BAHAMAS was checked by using the aircraft attitude data from BAHAMAS to correct the radar data for this attitude. The idea is that, if time stamps between both systems match perfectly, the surface return signal from the ocean surface should be a straight line, even during turns of the aircraft. To test for the time difference, the time series were shifted and the quality of attitude correction was assessed by looking at the variance of the height of the surface (taken as the maximum signal in each profile). The accuracy of this procedure is estimated to be 2 s. The analyzed temporal differences mainly ranged between 0 and 2 s. The maximum difference was 19 s. Measurements were corrected with these found offsets.

Temporal collocation between radiometer modules and BAHAMAS was investigated by looking at the transition between land and sea. This happens shortly after takeoff or before landing since both airports, Barbados Grantley Adams Airport and Keflavik Airport, are located close to the coast. This becomes possible since the microwave emissivity at low frequencies strongly differs between ocean (about 0.5) and land (about 0.9). The NAVO/GHRSST global 1 km land–sea mask (https://www.ghrsst.org/ghrsst-data-services/tools/, last access: 8 June 2018) was used to identify locations where the aircraft passed the shore. The footprint of the radiometers at an altitude of 900 m during takeoff is approximately 70 m long, which corresponds to 0.7 s at a groundspeed of 105 m s-1. The aircraft flies with a pitch of about 11∘, and therefore the instruments look ahead of the aircraft position. Due to this looking ahead and the calculated footprint at this altitude, the accuracy of the estimated land–sea transition is about 2.5 s. The land–sea mask resolution is approximately 0.0083∘. Using this, the crossing of the shore can be determined within 3 to 12 s, depending on position (0.0083∘ longitude is less distance in the midlatitudes than it is in the tropics) and aircraft speed. After a review of the measurements, all time series stayed in this interval and are therefore deemed correct.

As mentioned in Sect.  and Table , the instruments have different sampling rates, and, in the case of radar and dropsondes, also different vertical resolutions. To create a self-contained data set where variables from different instruments are easily comparable in time and height, the data were transformed onto a uniform grid for each flight.

Radar measurements have additionally been corrected for aircraft attitude. After this correction, the vertical coordinate no longer corresponds to the range from an instrument but instead to height above the surface. This makes the vertical profiles of radar data from different flights comparable with each other. After that, all data were interpolated to the new grid with 30 m vertical and 1 s temporal resolution using the nearest-neighbor value. The resulting data set consists of radar data with 1 s temporal and 30 m vertical resolution, radiometer data with 1 s temporal resolution and dropsonde data with 30 m vertical resolution.

In addition to the quality control described in Sect. , spikes in dropsonde and BAHAMAS data were filtered out. Spikes in dropsonde profiles were identified if data between two measurements jumped by more than half of the data range of the entire profile. Spikes in BAHAMAS data were identified by eye. Data gaps in dropsonde profiles were interpolated if the gap was shorter than 10 s. With an average falling velocity of about 12 m s-1, this corresponds to roughly 120 m. BAHAMAS and radiometer time series were interpolated if the gap was not longer than 3000 s (BAHAMAS) and 30 s (radiometers), which at an average aircraft velocity of 200 m s-1 corresponds to 6000 and 6 km respectively. The reasoning behind these different thresholds is that BAHAMAS data fluctuate very little in the upper troposphere and lower stratosphere. Therefore, it is reasonable to interpolate longer intervals than for radiometer data. A flag was added to radiometer data indicating which values were interpolated.