Dealing with unresolved scales of motion and systematic errors in the assimilation of cloud-affected satellite radiances

Hydrometeor formation and cloud processes occur at very small spatial scales and cannot be explicitly resolved on the numerical grids of weather prediction models. Parameterizations of these processes are a necessary component of forecast models and are known to be a major source of forecast error. Two main challenges arise. First, the mismatch between the effective resolution of prediction models and the resolution of satellite observations leads to representativeness errors. Second, sub-optimal parameterizations induce systematic errors in hydrometeor fields and cloud properties. In the presence of such model biases, the assimilation of cloud-scale observations can be detrimental to the analysis. It remains an open question how to properly account for the scale mismatch and for systematic errors when assimilating small-scale observations into a larger-scale numerical model.

In this work, we compare different approaches for assimilating radiance observations that contain unresolved scales, such as data thinning, use of superobservations, and a multi-scale decomposition of the two-dimensional cloud field. We study the problem using observing system simulation experiments (OSSEs) performed with the WRF model and the DART EAKF assimilation system. The nature run is a high-resolution large eddy simulation (dx=250 m) of deep moist convection developing in moderate wind shear, which supports organization into multicell storms. Synthetic satellite imagery is computed from the nature run using operators available through RTTOV. Several 40-member km-scale ensemble experiments (dx=2 km) evaluate the impact of assimilating thinned, averaged, or multi-scale radiance observations.