The creation of level-3 data by averaging non-uniformly distributed level-2 measurements inevitably leads to representativeness errors, see e.g. Coldewey-Egbers et al., (2015) for the case of a level-3 (gridded monthly means) total ozone data set. The resulting representativeness uncertainty can be larger than the formal uncertainty on the mean. In the best case this would represent an additional random uncertainty term. If the sampling pattern of the sounder changes in time, this may give rise to systematic, time-dependent representativeness errors that affect for example trend analyses for climate research (see e.g. Damadeo et al., 2014).  However, estimates of these representativeness uncertainties are rarely included with the data product. Also, the representativeness of the ground-based network should be taken into account when validating such data sets, i.e. the sparse spatial and temporal sampling of the ground network leads to significant representativeness uncertainties in for instance derived monthly (zonal) means.

Note that also in the context of validation of level-2 data, measurements are sometimes averaged after co-location (e.g. Valks et al., 2011; Schneising et al.,2012) without explicit calculation of the representativeness errors and resulting uncertainty.