Geophysical variables are seldom measured directly by satellites. Rather, satellites measure properties such as radiances, backscatter, or reflectance, which are quantities not often simulated by, for example, Earth System Models. Therefore, the geophysical variables of interest are typically derived from satellite measurements through retrieval algorithms. The retrieved variables are then used as observations, representing the “real” state against which models may be evaluated, assimilated, and trained using neural networks in data fusion exercises. However, in many cases, such retrieval algorithms require assumptions to derive the geophysical variable, and the measurement error associated with these assumptions is difficult to quantify. This leads to uncertainty in the application using the set of retrieved “observations”. One possible approach to circumvent the use of retrieval algorithms is to apply satellite simulators or observation operators, which translate the simulated system state in an observable quantity, directly comparable to satellite measurements.
In this session, we invite contributions from studies developing or applying satellite simulators or observation operators for geophysical model evaluation, assimilation and machine learning, including atmospheric, oceanic, sea ice, terrestrial and ice sheet models, and fully coupled earth system models that include them. As this approach can be applied across a range of research fields, we also welcome submissions from a broader class of geophysical models, including of planets and moons, to discuss and exchange common challenges and opportunities that might be solved in one field of research but still remain unsolved in another.