The International VLBI (Very Long Baseline Interferometry) Service for Geodesy and Astrometry (IVS) coordinates geodetic VLBI observing programs (Nothnagel et al., 2017). VLBI is important since it is the only geodetic technique capable of deriving the full set of Earth orientation parameters. The IVS has organized special measurement campaigns called “CONT” approximately every 3 years since 2002. These particularly intensive sessions cover 2 weeks of continuous network observations and must be distinguished from the routine observation program consisting of individual 24 and 1 h sessions. The main goal of CONT is to test the accuracy of the VLBI estimates of the Earth orientation parameters and to investigate possible network biases (Behrend et al., 2017). The CONT17 campaign started on 28 November 2017, with observations being carried out in three different networks (Behrend, 2017).

The Geodetic Observatory Wettzell (GOW) features two SLR (Satellite Laser Ranging) telescopes, several GNSS (Global Navigation Satellite System) reference stations, and a DORIS (Doppler Orbitography and Ranging Integrated by Satellite) beacon, as well as three VLBI telescopes (Schüler et al., 2015). All three radio telescopes participated in CONT17, each of them in one of the three different networks. VLBI, GNSS, and DORIS all operate in the microwave frequency domain. In this case, the atmosphere is a major complicating factor reducing the accuracy (Petit and Luzum, 2010). Consequently, the set of atmosphere sensors at the Geodetic Observatory Wettzell has been substantially enhanced in recent years to provide means to better deal with this problem. The propagation delays induced by the ionosphere can be compensated for with the help of measurements taken at at least two different frequencies.

However, the troposphere (and to a lesser extent also the stratosphere) remains a problem. The microwave signals are delayed when passing through these layers, and these effects are nondispersive; i.e., they are virtually identical at the various frequencies in use. As a consequence, pre-elimination of these propagation errors is not possible. One method to quantify tropospheric errors is to use models. Another one is to introduce tropospheric unknowns as nuisance parameters into the observation equations and to estimate these effects together with the set of target parameters. In practice, a combination of both approaches is usually accomplished. In any case, real measurements of the state of the atmosphere are very valuable to aid in tropospheric delay modeling and to interpret the results and residuals. This is the motivation to compile the atmosphere measurements collected during the CONT17 campaign, forming a comprehensive data set to understand the atmosphere over the Geodetic Observatory Wettzell, to aid VLBI analysis and to support studies dealing with the comparison of troposphere delays of microwaves derived from different techniques (e.g., Teke et al., 2013; Lu et al., 2015).

The temperature, humidity, and wind sensors of the local weather station are mounted on a concrete tower at 7 and 10 m height above the surface (Table 1). The air pressure sensor is inside the 20 m radio telescope control building, and the rain gauges are mounted on a platform as shown in Fig. 1. Data are continuously acquired, and averages are recorded once per minute. For wind direction and wind speed, minimum and maximum values measured within 1 min are also stored, indicated by “<” and “>”. The heated rain gauges measure snow as well and record the sum over 1 min.

As an addition to the meteorological station, global radiation is measured using a pyranometer, Thies CM 11. At the same place a net radiometer (Kipp & Zonen NR Lite) measures the difference between radiation from above, i.e., the sun and the sky, and from below, i.e., the soil surface. Both sensors are installed 1.5 m above the grass surface. The sampling rate is 10 min.

A quasi-continuous record of temperatures in the atmosphere up to 1000 m height is realized by a radio wave radiometer, MTP-5, from R.P.O. Attex. The microwave receiver measures the blackbody thermal radiation of the atmosphere at a frequency of 56.6 GHz. The intensity of the radiation is a function of the temperature. By scanning the atmosphere at different elevation angles, the operating software computes temperatures at different heights in 50 m steps up to 1000 m, under the assumption of a horizontal temperature layering. The basic principle and some field examples are described in Peña et al. (2013).

The temperature profiler is installed on a tower at 619 m a.s.l. and 10 m above ground. A complete profile is recorded every 5 min. The measurement uncertainty increases with height and is specified to be 0.2 to 1.2 ∘C, depending on the profile type and height.

On the same tower as the temperature profiler, a water vapor radiometer, Radiometrics WVR-1100, is installed. It is a microwave receiver measuring the intensity of atmospheric radiation at 23.8 and 31.4 GHz. The water vapor dominates the 23.8 GHz observations, whereas the cloud liquid in the atmosphere dominates the power in the 31.4 GHz channel. This allows the simultaneous determination of integrated water vapor and liquid water along the line of sight. From the measured brightness temperatures at both frequencies, Tb23 and Tb31, the frequency-dependent atmospheric opacities τ23 and τ31 are calculated. The water vapor and liquid water content and the path delay are obtained using the following relationships: Vap=c0vap+c1vap⋅τ32+c2vap⋅τ31Liq=c0liq+c1liq⋅τ32+c2liq⋅τ31Del=c0del+c1del⋅τ32+c2del⋅τ31.

The retrieval coefficients c0, c1, and c2 are site-dependent and have to be determined from a history of radiosonde observations from a representative site. The retrieval coefficients used in this work are valid for Munich and displayed in Table 2. The blackbody temperature, TkBB, as given in col. 4 of the data file, is only used to establish the temperature coefficient of the instrument gain. A description of the determination of atmospheric water vapor using microwave radiometry is given, e.g., in Elgered et al. (1982).

The instrument performs about one measurement per minute in one particular direction. In azimuth steps of 30∘, elevation scans between 20∘ and 160∘ are carried out; i.e., the scan passes over the zenith direction. For a complete scan of the entire sky, it takes about 90 min. In order to obtain the zenith delay only, all lines with 90∘ elevation have to be extracted from the data files. This results in 198 zenith data points per day.

The accuracy of the brightness temperature measurement is specified with 0.5 K. The accuracy of the resulting water vapor and liquid water contents and phase delays strongly depends on the instrument calibration, i.e., the retrieval coefficients used.

The cloud detector or nubiscope measures the thermal radiation of the sky in one particular direction. Since clouds absorb radiation from the sun and reflected infrared radiation from the ground, the temperature of the cloud base is significantly higher than the clear sky. By scanning the entire sky, a map of the cloud coverage can be generated. As low clouds generally yield higher temperatures than high clouds, additional information regarding the height of the clouds is obtained. Taking into account the horizon effect, that is, the temperature increase from zenith to horizon, the processing software determines the fraction of low-, medium- and high-level clouds, the coverage, temperature, and height of the main cloud base, and the temperature and height of the lowest clouds. Further information is given on the manufacturer's website (Sattler, no year).

The cloud detector is installed on an observation platform on the roof of the Twin Telescope operation building at 625 m a.s.l. and 9 m above the surface. The recorded heights of the cloud base refer to the instrument height. A complete scan of the sky is done once every 10 min.

Every day during the CONT17 experiment, radiosonde balloons were launched at 8:00 and 14:00 UTC at the launch site depicted in Fig. 1. We used Graw DFM-09 radiosondes and helium-filled Totex 350 balloons with 300 g buoyancy. The transmission rate is one data set per second. The radiosondes are equipped with a GPS receiver, permitting an absolute localization with an accuracy of 5 m in horizontal and 10 m in vertical position. The tracking allows precise measurements of wind speed and wind direction at different heights, with an accuracy of 0.2 m s-1, and ascent and descent rates. The air pressure is computed from the surface pressure at the station, the geopotential height, and the temperature, with an accuracy of 0.3 hPa. The accuracy of the temperature and relative humidity sensors is specified with 0.2 ∘C and 4 %, respectively. The relative humidity hrel can be expressed as water vapor pressure e using e=es⋅hrel100 and the Magnus formula according to Sonntag (1990) for the saturation vapor pressure for water in hPa es=6.112⋅e17.62⋅T243.12+T, with the temperature T in ∘C.

Each radiosonde launch yields two files, a profile data file with measured and derived meteorological quantities and a position data file from the GPS receiver. Both files were merged to one file using time interpolation when necessary (see Table 5).

For the time span covering the CONT17 campaign, a data set was extracted from the ICON-EU model from the German Weather Service (Deutscher Wetterdienst, DWD) containing pressure, temperature, and humidity data at different height levels. The ICON-EU model is a refined domain (local nest) of the global ICON (ICOsahedral Nonhydrostatic) model, whose grid is made up by a set of nearly equal spherical triangles spanning the entire Earth (Reinert et al., 2018). The ICON-EU nest is refined by dividing each triangle into four subtriangles, resulting in a grid spacing of ∼6.5 km. It includes 60 height levels up to 22.5 km. The physical parameters at the top of the model are controlled by the global model reaching a height of 75 km.

The extracted subset covers a radius of 4∘ (∼445 km) around the GOW. The structure of the grid file “we_iconeu_4deg.grd” is given in Table 3, where each line represents one of the 13 941 grid points. The data files are named “we_iconeu_4deg_yyyymmddhh.xxx”, where yyyy denotes the year, mm the month, dd the day, hh the hour, and xxx the physical quantity.

pre: air pressure (hPa)

The model data represent the atmospheric analysis fields at the beginning of each forecast run and are computed every 3 h using assimilated observed data.

As a comparative data set, both zenith hydrostatic and wet delays (ZHDs and ZWDs) from the NCEP (National Center for Environmental Prediction) global numerical weather model are provided. This data set is derived from GDAS (Global Data Assimilation System) and GSF (Global Forecasting System) weather fields. The derivation of these tropospheric path delay data requires some explanation because only one dimensional output file from the GDAS numerical weather model (so-called “surfaces fluxes”) was used. From our experience, zenith total delays are expected to reveal a standard deviation approaching 1 cm for the region of Wettzell. This is slightly less accurate than the estimation of tropospheric delays using GNSS permanent stations (see Fig. 9) but still useful for a number of applications.

The original weather model output data can be found on the ftp server at http://ftp.ncep.noaa.gov/ (last access: 22 February 2019) in the directory “/data/nccf/com/gfs/prod”, all available in standard grib2 format. Note that this is a rolling real-time archive. Regions of interest are routinely extracted at our observatory and converted into a tailored format, addressing the specific needs of space geodesy. Analysis fields are used whenever possible (every 6 h) with one 3 h prediction in between.

The necessary information is horizontally interpolated and vertically reduced to the central GNSS station WTZR at the observatory. The horizontal interpolation approach is depicted in Schüler (2001, p. 197ff) using the four nearest neighbors, but as a modification, bilinear functions of type a0+a1⋅ϕ+a2⋅λ+a3⋅ϕ⋅λ are employed for interpolation of the surface flux data, where a0,…,3 are the interpolation coefficients determined from the four nearest neighbors, ϕ is the latitude of the interpolation site, and λ is its longitude. Vertical reduction to the target height is important. The TropGrid2 model (Schüler, 2014) is used for this purpose. TropGrid2 is a global gridded 1∘ × 1∘ model containing reduction coefficients for all quantities needed. The coefficients of these reduction functions were derived using 9 years of numerical weather model data.

The determination of ZHD (zenith hydrostatic delay) from GDAS and GSF surface fields is straightforward: surface pressure is horizontally interpolated and vertically reduced and then converted into ZHD using the Saastamoinen model (Saastamoinen, 1972): ZHD=0.0022767⋅p1-0.00266⋅cos⁡(2ϕ)-0.00028⋅h, with the pressure p (hPa), the ellipsoidal height h (km), and the geographic latitude ϕ of the station. The derivation of ZWD (zenith wet delay) requires more effort, but GDAS/GSF surface fluxes are a very attractive resource since these weather fields already contain the total column atmospheric water vapor (IWV, integrated water vapor). These values are converted into ZWD with knowledge of the weighted mean temperature of the atmosphere TM (see Schüler, 2001, p. 184ff). TM itself is substituted in the standard product by a surface temperature conversion function available on the TropGrid2 data grid. After conversion, ZWD is vertically reduced and horizontally interpolated to the target height.