We detail the verification of the WALE large eddy simulation turbulence model for application in cell-based lattice Boltzmann methods, as implemented in our generic Cartesian structured adaptive mesh refinement framework AMROC. We demonstrate how to effectively apply the test case of decaying homogeneous isotropic turbulence to verify the core WALE implementation against higher resolved direct numerical simulations and the constant-coefficient Smagorinsky turbulence model. Both standard and regularised single relaxation collision models are analysed systematically. While our results confirm the established observation that the standard collision model yields less dissipative energy spectra, novel quantitative evidence is given that this positive behaviour comes at the cost of unphysical perturbations in high wavenumbers. In order to allow unaltered application of the finite-difference stencils intrinsic to the WALE approach in real-world flow situations, a new method is presented for ensuring consistent boundary conditions in microscopic distribution functions as well as in macroscopic variables. The benefit of the proposed technique is shown for dynamically adaptive simulations of flow around a sphere at Reynolds number 1000 and compared to a large eddy simulation using the constant-coefficient Smagorinsky model.