This article describes the development of tools for routine 4-dimensional variational data assimilation of Global Positioning System Slant Total Delay (STD) data in the framework of the MM5 system at the Institute of Physics and Meteorology of the University of Hohenheim. The Slant Total Delay forward operator is introduced which allows model validation and the assimilation in the Message-Passing Interface environment. An experiment is conducted which highlights the importance of accurate model physics in the variational assimilation system. We demonstrate that the model minus observation statistics of STD data crucially depends on the convection scheme and the implementation of horizontal diffusion. A set of modifications to the existing non linear, tangent linear and adjoint model is presented. These include an improvement of the horizontal diffusion scheme and the implementation of the Grell cumulus convective scheme in order to eliminate the observed systematic tendency in the model minus observation statistics of the STD data and precipitation in mountainous terrain. A first assimilation experiment with the improved MM5 variational assimilation system shows promising results.