Stable Boundary Layer Height Parameterization: Learning from Artificial Neural Networks

Abstract

Artificial neural networks (ANN) are employed using different combinations among the surface friction velocity u*, surface buoyancy flux Bs, free-flow stability N, Coriolis parameter f, and surface roughness length z0 from large-eddy simulation data as inputs to investigate which variables are essential in determining the stable boundary layer (SBL) height h. In addition, the performances of several conventional linear SBL height parameterizations are evaluated. ANN results indicate that the surface friction velocity u* is the most predominant variable in the estimation of SBL height h. When u* is absent, the secondly important variable is the surface buoyancy flux Bs. The relevance of N, f, and z0 to h is also discussed; f affects more than N does, and z0 shows to be the most insensitive variable to h