Starting this year, Tomorrow.io is launching a constellation of radars with planned average hourly revisit 
globally. This new observational platform stands to revolutionize both scientific applications as well as
realtime nowcasting and forecasting. The data assimilation of spaceborne radar observations remains 
challenging: convection is caused by thermodynamics, so directly updating the hydrometeors, which radar 
measures, in the NWP model has a tendency to be rejected by the model - the so-called spin-down problem. 
In this talk we will present the use of statistical operators for hybrid ensemble-variational assimilation
in OSSEs and OSEs. Classification of observations, such as with Gaussian Mixture Models, is a prerequisite 
to understand the statistical relationships. Once classified, we demonstrate the use of kernels methods
- a class of methods that includes thinning, superobbing, principal component analysis, canonical
correlation analysis, and neural networks. We present preliminary results of the OSSE impact of adding 
these radar observations to the global observing system.