This very basic question occurred to me during a talk on reanalysis data sets. Often we know the observations produced by a particular instrument, say rain radar, are biased from one year to the next because of changes in how the instrument has been calibrated, or because methodologies for processing the raw data it produces have changed.
How is this dealt with in long reanalysis data sets like ERA-Interim? Are the observational data sets made as consistent as possible before being fed into the reanalyses model, or are biases somehow smoothed out through complex data-assimilation methods or something along those lines?
EDIT: To provide a specific example, how are scatterometer data sets incorporated into ERA-Interim when we know there are biases between them (for example Wu and Chen (2015))?