Addressing Information Loss and Discontinuous Boundary Conditions in Cycled Assimilation Systems

This study reports on preparatory work towards the assimilation of radar observations in a Pan-Canadian rapid-refresh assimilation system. Before the first assimilation experiments could be performed, it was necessary to address two issues specific to cycled assimilation systems.

First is the loss of information occurring every time the model is halted to allow the update of the control vector.  Since control vectors typically contain only a subset of all the variables necessary to a model's integration, some information is lost at the analysis step.  In systems with hourly cycling frequencies, this loss may become significant.  As a mitigating measure, it is possible to ``recycle'' a set of variables in addition to those already contained in the control vector.  Choosing the variables to be recycled is not straightforward as it is difficult to establish what may constitute a tolerable level of disruption.

A second difficulty arises from the fact that convective-scale assimilation systems usually consist of Local Area Models (LAMs) whose boundary conditions come from another cycled assimilation system.  The disturbance of model-balance in the driver model (caused by imperfect analyses and information loss) can also affect the nested LAM.

The two issues are examined in experiments conducted using the Canadian Global Environmental Multi-scale (GEM) model run with two configurations, hydrostatic and non-hydrostatic, at resolutions of 10 and 2.5~km respectively.
Preliminary results identify the prognostic Turbulent Kinetic Energy (TKE) as an important variable to recycle in addition to the control vector.  At a resolution of 2.5~km, its omission has a significant impact on the amount of liquid condensates and the development precipitation.  It is also shown that the use of a cycled driver can lead to the complete decorrelation of small-scale precipitation features in the nested LAM after a few hours of model integration.