Thawing of permafrost potentially affects the global climate system through the mobilization of greenhouse gases, and poses a risk to human infrastructure in the Arctic. The response of ice-rich permafrost landscapes to a changing climate is particularly uncertain, and challenging to be addressed with numerical models. A main reason for this is the rapidly changing surface topography resulting from melting of ground ice, which is referred to as thermokarst. It is expressed in characteristic landforms which alter the hydrology, the surface energy balance, and the redistribution of snow of the entire landscapes. Polygonal patterned tundra which is underlain by massive ice-wedges, is a prototype of a sensitive permafrost system which is increasingly subjected to thermokarst activity throughout the Arctic. In this talk I will present a scalable modeling approach, based on the CryoGrid land surface model, to investigate the degradation of ice-wedges. The numerical model takes into account lateral fluxes of heat, water, and snow between different topographic units of polygonal tundra and simulates topographic changes resulting from melting of excess ground ice (i.e., thermokarst), and from lateral erosion of sediment. We applied the model to investigate the influence of hydrological conditions on the development of different types of ice-wedge polygons in a study area in northern Siberia. We further used projections of future climatic conditions to confine the evolution of ice-wedge polygons in a changing climate, and assessed the amount of organic matter which could thaw under different scenarios. In a related study for a study site in northern Alaska, we demonstrated that the model setup can be used to study the effect of infrastructure on the degradation of ice-wedges. Altogether, our modeling approach can be seen as a blueprint to investigate complexly inter-related processes in ice-rich permafrost landscapes, and marks a step forward towards an improved representation of these landscapes in large-scale land surface models.