Algorithmic skeletons are polymorphic higher-order functions that represent common parallelization patterns. They can be used as the building blocks of parallel applications by integrating them into a sequential language. In this paper, we consider the design and implementation of skeletons for the management of distributed dynamic data. Such skeletons are used by grid-managers of numerical solvers like multigrid algorithms with adaptive refinement techniques. We present three mechanisms to encapsulate the necessary communication in parallel implementations of those solvers. Further, we have integrated our skeletons in a single-grid example solver and have run it on a PC-Cluster against a pure C implementation. Run-time measurements show that the speedups and efficiency of the skeleton-based program are comparable to those obtained for the C implementation.