Different data sources leveraged in Aardvark. The input data consist of observations from remote sensing instruments (top row), which we pre-grid before passing to the model, as well as in situ observations from land and marine observation platforms and radiosondes (bottom row). Each of these data modalities contains several observational variables, of which we selected a subset here for the purposes of illustration. Here we show remote sensing data 40, 41, 42, 43, 44, 45, after performing our gridding step, and raw in situ data 46, 47, 48. Note that the colours in all six plots are meant for illustration purposes. The remote sensing data also include a range of metadata about the measurements, omitted here for simplicity. White areas indicate regions of missing data, which must be handled by the encoder module of Aardvark.