In this paper a hierarchical evaluation is proposed. At a first step the ability of a given model to produce a cloud at the right place and at the right time (i.e. cloud occurrence) is evaluated. The differences in instrumentation (i.e. radar sensitivity or lidar versus ceilometer) appear crucial if one wants to compare the results from one site to another. The differences in cloud occurrence are also studied on a seasonal basis.
The variables of cloud parametrisations (diagnostic of prognostic depending of the models) : cloud fraction and ice water content are then compared between model and observations when model and observations agree on the cloud occurrence. A statistical comparison is proposed in order to evidence potential biases in models like for instance the inability to produce large cloud fraction values as well as particular skills depending of the cloud levels. Concerning the ice water content it has been chosen to first derive this parameter from the observations and to compare it with the model variable. In this case a site dependency related to the radar sensitivity appears but highlights the fact that the models seem unable to produce a wide range of ice water content values.