The direct assimilation of microwave and infrared temperature sounders into Numerical Weather Prediction (NWP) and reanalysis systems improves estimates of the atmospheric state, directly improving both the NWP weather forecasts, as well as the long-term monitoring of atmospheric temperature by reanalysis systems, such as Copernicus C3S reanalysis (ERA-5 and later). When data from new temperature sounders (e.g. ATMS, AMSU-A, IASI, AIRS, CrIS) become available, it is important to assess the quality of the observations before they can be assimilated. Short-range temperature forecasts from NWP systems provide a good reference for validating new temperature-sounding satellite instruments due to the high accuracy of these forecasts, particularly in the troposphere. For example, estimates of the uncertainties of tropospheric temperature forecasts for the ECMWF system indicate that they are around 0.1K in radiance space (Bormann et al, 2010). Using NWP forecasts as a reference also facilitates the inter-comparison of satellite data, since differences in time and space of the measurements can be accounted for with the use of a forecast model. This allows to estimate inter-satellite biases (e.g. Bormann et al 2013; Lu et al 2015).

While NWP temperature fields are very useful as a reference for satellite Cal/Val, this method does not currently lead to fully traceable estimates of uncertainty (see gap G4.01), since the uncertainties in the NWP background, the uncertainties in the radiative-transfer model, and the spatial-mismatch uncertainties are not known to fully traceable standards. This first point can be addressed by using reference in-situ data such as from GRUAN for assessing the uncertainties in the ECMWF and Met Office NWP short-range forecasts of temperature and humidity. To do this, a tool known as the ‘GRUAN processor’ has been developed based on the EUMETSAT NWP Satellite Application Facility (NWP SAF) Radiance Simulator (see https://www.nwpsaf.eu/GProc_test/ins.shtml). This tool can be used to calculate the differences between GRUAN-temperature measurements and NWP forecasts in both geophysical space (temperature and humidity as a function of height) and radiance space (radiances as a function of channel for different satellite instruments) and compare these differences to the GRUAN uncertainties.

GRUAN reference temperature measurements are available from the surface to an atmospheric height of up to 5 hPa. However, less radiosonde data are available in the stratosphere than the troposphere and none above 5 hPa. In the upper reaches balloon-burst propensity leads to potentially biased sampling of solely warmer tail conditions. The lack of reference data in the upper stratosphere and mesosphere affects the assessment of uncertainties in NWP temperature fields to reference standards, leading to a poorer assessment at heights around 40 – 5 hPa and no assessment being possible above 5 hPa. In turn, this affects the calibration/validation of new temperature sounding data, which are sensitive to this portion of the atmosphere. This is particularly true of AMSU-A channel 14, whose weighting function peaks around 2 – 3 hPa, but it also affects channels 12 – 13 (peaking at 10 and 5 hPa respectively). The equivalent channels on ATMS are also affected, and there are also a number of infra-red temperature sounding channels on hyperspectral infrared sounders which are affected, including CrIS channels at 667.500 cm-1, 668.125 cm-1, and 668.750 cm-1, IASI channels at 648.500 - 669.750 cm-1 and AIRS channel numbers 54 - 83. Furthermore, the weighting functions for most satellite sounding channels have a stratospheric tail with some small sensitivity to the stratospheric temperature, so that this will contribute to the uncertainty of the Cal/Val for all channels, although with less of an impact for the channels peaking lower in the atmosphere.

The gap identified here is twofold – a lack of reference observations at 40 – 5 hPa, and no reference observations above 5 hPa. The first part could be solved by supplementing the GRUAN-reference dataset with GNSS Radio Occultation (GNSS-RO) observations and products, including sets of bending angles and temperature retrievals. GNSS-RO bending angles have a high vertical resolution and uncertainties have been calculated both for these observations and for the derived temperature profiles with a high accuracy (Kursinski et al 1997). This makes GNSS-RO observations potentially very valuable as references for the calibration/validation of new satellite temperature-sounding data. We propose, therefore, including both the bending angles and derived temperature profiles, along with their estimated uncertainties, in the GRUAN processor in future work. This requires efforts to co-locate GRUAN profiles and GNSS-Radio Occultations. Such work will benefit where GRUAN sites in future make use of an EUMETSAT simulator that predicts up to two weeks in advance coincidence of polar orbiter overpasses and GNSS-RO occultations.

It should be noted that there are some known drawbacks to using GNSS-RO temperature profiles as a reference, however. Firstly, since the observations are directly sensitive to pressure/temperature rather than temperature, there is a so-called null space, in which the observations are blind to combined mean errors in temperature and pressure, which cancel each other out. Because of this, it is important to keep using reference radiosondes such as the GRUAN observations. Secondly, the temperature profiles at higher altitudes are less accurate since the observations rely on the bending by the atmosphere and in thin atmosphere the signal-to-noise ratio becomes very low. This makes it difficult to use GNSS-RO observations as a reference at altitudes above around 5 hPa (Healy and Eyre, 2000; Collard and Healy, 2003). The use of GNSS-RO measurements would therefore not help the lack of observations about 5hPa, but it would increase global coverage, improving the cal/val of new satellite temperature sounding data at heights of 40 – 5 hPa.

There is a clear need to develop instrumentation capable of measuring temperature routinely above 40 hPa (and in particular above 5hPa) in a traceable manner with metrologically well characterised uncertainties. The remedy defined here (using GNSS-RO temperature profiles as a reference dataset) only partially closes this gap and does not obviate the need for technological developments in upper atmosphere profiling.