A retrieval-based approach to obs-space localization of solar satellite channels
Observation space localization of vertically integrated quantities is a considerable challenge related to Ensemble Kalman Filters.
A key problem in this context is the vertical positioning and localization of solar satellite radiances sensitive to cloud water and ice. Through phase transitions and latent heating these quantities are directly linked to temperature and relative humidity. Due to model error and few observational constraints to microphysical properties in data assimilation, cloud processes in state-of-the art NWP models tend to have limited predictability. Given the uncertainty in the occurrence and vertical distribution of clouds combined with their discontinuous nature, it is obvious that comparing and correcting the first guess ensemble with cloud-sensitive satellite radiances in the wrong vertical region may have a detrimental effect on the accuracy of the analyzed upper air profiles.
To deal with that, obs-space localization requires detailed information about cloud layers and their vertical extent. This is not provided by single narrowband channels from satellite imagers which exclusively represent 2-dimensional integrated information. Therefore, for clear-sky satellite observations, a typical approach is to use sensitivity functions based on information (i.e. the Jacobian) from a model state, e.g. an arbitrary first guess ensemble member. However, due to the often large uncertainty in the first guess cloud profiles, we consider this less suitable for radiances sensitive to cloud water and ice and develop a retrieval-based approach to obs-space localization.
Using NWC-SAF cloud type and cloud top height products, we derive vertical meta-data for the all-sky assimilation of the 0.6 micron solar channel of the SEVIRI instrument onboard MSG. Key concepts we discuss are superobbing in the presence of cloud borders, vertically piecewise application of the MFASIS forward operator (Scheck et. al, 2016) and the assimilation of so-called “derived” observations. We exemplify the approach by means of single observation experiments using the German regional NWP-system COSMO-KENDA (Schraff et. al, 2016). Further, we discuss potential advantages and disadvantages compared to new model-space localization approaches like the GETKF (Bishop, 2017).
Dr. Lilo Bach Deutscher Wetterdienst