The TerraDT project develops a modular digital twin of the cryosphere and land systems for climate modelling. In collaboration with the Max Planck Institute for Meteorology (MPI-M), the Finnish IT Center for Science (CSC), and Uppsala University (UU), the German Climate Computing Center (DKRZ) contributes to the integration of advanced glacier and ice-sheet simulations into a global Earth system model (ESM). In this talk, I present how three independently developed simulation models were coupled using DKRZ’s Yet Another Coupler (YAC) library. In the coupled simulation, the global ESM ICON provides the atmosphere, ocean, and land components. Glacier and ice-sheet dynamics are simulated using CSC's finite-element software Elmer via the Elmer/Ice module. In addition, an energy balance and firn model (EBFM) developed at UU enables an accurate representation of energy and mass exchange between the glacier and other domains. YAC enables the combination of independently developed simulation models in a minimally invasive manner. This partitioned approach requires the implementation of dedicated "glue code" in ICON, Elmer/Ice, and EBFM. While code modifications are generally feasible because all components are open source, the scale of the projects ranges from small teams to large, distributed organizations. The resulting distribution of responsibilities and knowledge can pose an additional challenge for developing new coupling features and integrating them into upstream repositories. Beyond model coupling, the definition of an appropriate coupled test case is crucial. It requires careful configuration of multiple components---some of which are under active development---across different hardware architectures. In this context, strict compliance with open-source best practices, such as version control, formal releases, and software packaging, is invaluable. Finally, reproducibility demands that not only the software but also the required input files, runscripts, and expected results are archived---thereby enabling rapid detection of regressions.