Numerical Weather Prediction (NWP) models and reanalysis systems possess a number of key attributes for the comprehensive assessment of observational datasets. These models routinely ingest large volumes of observations within the framework of data assimilation and, combined with model data, produce optimal estimates of the global atmospheric state. The model fields are constrained to be physically consistent, and have continuous coverage in time and space. NWP fields exhibit sufficient accuracy in their representation of temperature and humidity fields to enable the characterisation of subtle biases in monitored satellite data. Examples include the evaluation of SSMIS (Bell et al., 2008), FY-3A sensors (Lu et al., 2011) and AMSU-A (Lupu et al., 2016).

However, robust uncertainty estimates for NWP fields are still lacking. Space agencies and instrument teams, as well as key climate users, are sometimes slow (or reluctant) to react to the findings of NWP-based analyses of satellite data due to the current lack of traceable uncertainties. Reliable estimates for the uncertainty of NWP fields, and modelled TOA radiances, would allow an assessment of absolute radiometric errors in satellite instruments. The aim is to assess uncertainties in NWP fields, through systematic monitoring, using reference-quality data,

The aim of GAIA-CLIM activities is to assess uncertainties in NWP fields through systematic monitoring, using data from the GCOS Reference Upper-air Network (GRUAN) radiosonde network. Difference statistics evaluated by Noh et al. (2016) for three institutes’ models indicated good agreement with GRUAN profiles for temperature (biases not exceeding 0.1-0.2 K throughout the troposphere, with root-mean-square (RMS) differences within 1 K). Models were found to be less skilful at representing relative humidity (RH) fields, with biases cf. GRUAN sondes of up to 5% RH and RMS differences up to 15% RH. This illustrates the particular need to quantify NWP humidity uncertainties, as a means of improving the assessment of satellite EO data, which are sensitive to atmospheric water vapour.

GAIA-CLIM has developed a ‘GRUAN processor’ as a software tool, which enables the routine comparisons of NWP fields with reference radiosonde data. Importantly, these comparisons can be conducted both in terms of geophysical variables (temperature, humidity) and TOA radiances or brightness temperatures. It is estimated that significant progress can be made in establishing this routine monitoring within the timescale of GAIA-CLIM, although maintenance of the processor is not guaranteed beyond the lifetime of the project.

The complexity of NWP and reanalysis systems is such that a complete error budget is unattainable. However, progress can be made in accounting for spatial, seasonal, diurnal, and weather regime factors that affect uncertainties. This can be achieved through comparisons with recognised reference measurements, such as GRUAN radiosondes, complemented by ‘near-reference’ measurements with greater global coverage.