A new mixing length scale formalism for a TKE-l turbulence closure will be presented. It will be shown that many of the present day length scale formulations fail in one of following limits: schemes based on a local stability measure (e.g. the Richardson number) display unrealistic behavior and instabilities in the convective limit, whereas a non-local parcel method (like Bougeault & Lacarrere) has difficulties with surface layer scaling. The new length scale formulation combines local and nonlocal stability characteristic in a new way; it uses vertical integrals over the stability (the Richardson number) in a simple "parcel" framework. As such, it can be matched to surface layer similarity close to neutral conditions and also displays a good behavior in the convective limit away from the surface. The length scale formulations is numerically very stable and computationally cheap. It can also be extended well to moist (cloudy) conditions.

The behavior of length scale is evaluated in a single column model and in a limited area model on high resolution. The single column model showed good (numerically stable and physically realistic) results for the usual convective cases and a diurnal cycle of a dry convective boundary layer with shear. The prediction of the near surface wind and temperature of the limited area model is compared to tower measurements at Cabauw (the Netherlands).