The GAIA-CLIM user survey highlighted the need for a readily accessible radiative-transfer capability available as part of the Virtual Observatory to allow the transfer of reference measurements into the measurement space of satellite instruments. Such a tool would enable a more direct characterisation of the satellite measurements. The validation of satellite measurements in terms of the measured radiance is more straightforward than a validation of retrieved (or analysed) quantities. This is because the forward calculation from the geophysical profile is unique, whereas solutions to the inverse problem are non-unique in that several distinct geophysical profiles can be consistent with a given radiance measurement. As part of this, the uncertainty information in reference measurements needs to be appropriately transformed in the mapping (e.g. from reference measurements to top-of-atmosphere (TOA) brightness temperatures). In turn, this requires knowledge of the vertical and / or horizontal correlation structures present in the reference measurement.

The GAIA-CLIM project realised the development and demonstration of a GRUAN-processor, which is able to monitor Numerical Weather Prediction (NWP) model temperature and humidity fields relative to GRUAN radiosonde observations, and to monitor the differences in computed TOA radiances for a wide range of meteorological satellite sensors from both measured (GRUAN) and modelled (NWP) state estimates. The GRUAN-processor is built around several core capabilities that are likely to be supported longer-term by EUMETSAT (the fast RT modelling capability [RTTOV] and the flexible interface to NWP model fields [the Radiance Simulator]), nevertheless there is a foreseen governance gap beyond the term of GAIA-CLIM regarding the ongoing development priorities and support for the GRUAN-processor.

The key stakeholders include: satellite agencies (engaged in assessing the quality of long term satellite datasets and implementing Cal/Val plans for forthcoming missions); NWP centres (with an interest in determining traceable uncertainties in model fields); GRUAN governance groups and site operators (with an interest in assessing the value of NWP for cross–checking GRUAN-data quality); and the wider climate-research community (also with an interest in assessing the quality of long term satellite datasets). The future governance of the processor would ideally take account of the priorities of this group of stakeholders.

Associated with this top-level requirement for a flexible observation operator is a specific requirement, related to the need for comprehensive information on the error characteristics of reference measurements. In the context of reference radiosonde measurements, this includes estimates of the error correlations between measurements. Other ground-based data sources such as microwave radiometers and Lidar systems could be developed into reference measurements, including the full assessment of uncertainty.

GRUAN was established with the goal of creating a network of sites around the world where reference measurements of atmospheric vertical profiles are performed (Seidel et al., 2009). Data processing for GRUAN sondes attempts to account for all known sources of systematic and random error affecting the temperature and humidity sensors (Dirksen et al., 2014). However, although vertically resolved best-estimate uncertainties are available, the error correlation structure (i.e. between vertical levels) in the sonde measurements is not presently available, constituting a current gap.

Many applications of reference radiosonde measurements require an estimate of error correlations. For example, as part of the comparison of reference-sonde measurements and NWP fields in terms of TOA brightness temperatures, it is necessary to have realistic estimates of these error covariances. Only then is it possible to estimate realistically, using a radiative-transfer model, the uncertainty in TOA brightness temperature that propagates from sonde profile uncertainty.

Calbet et al. (2017) performed a study into the calibration-traceability chain for forward modelling of the Infrared Atmospheric Sounding Interferometer (IASI), using collocated GRUAN sondes and the LBLRTM radiative transfer model. They found the propagation of uncertainties from sonde profiles was hampered by the lack of covariance information between levels. They resorted to analysing two extreme cases: where the level-by-level sonde profile uncertainties are perfectly correlated or perfectly uncorrelated. The uncertainty in modelled TOA radiances was assumed to lie between the two extremes.

The vertical error correlation structure in GRUAN-sonde profiles is the subject of current research. Such uncertainties are envisaged to be reported in the version 3 GRUAN product (correlated, partially correlated and random terms) being developed by the GRUAN Lead Centre.

A tractable means of representing vertical error covariances is by parametrisation. If the measurement variance at each vertical level is known, the correlated errors between levels can be represented by Gaussian statistics assuming a characteristic correlation length (see e.g. Haefele and Kämpfer, 2010). The correlations should be based on physical constraints where these are known.