Background

Verification of model forecasts with conventional scores far from being satisfactory
        traditional verification done only partially (stations of the country only, not continuously in time ...)
        scores computed but not really evaluated, nor exchanged
        different tools employed to compute the differences between model forecasts and observations

Clean the conventional scores first, going to the more sophisticated verification after

Not easy : cf. failure of the previous attempt
        lack of manpower and specific control teams
        hardly any exchange of local observations
        no standard verification procedure defined
        lack of motivation ?

But important
        verification to help tracking problems in operational suites
        verification to check the impact of increased resolution, for nested models.

Going on subjective verification and case studies