Erasures in communication channels play an important role in data transmission in real networks. It has also been well establised that linear network coding is enough to achieve the upper bound in multicast problems. In this report, we investigate the problem of minimizing the overall erasure probability in a multicast scenario at multiple levels of hierarchy in a network. Firstly, we use ant colony optimization at the upper level as a technique to obtain an optimal set of nodes for network coding. Based on the result, we also analyze the degree distribution distortion of LT-coded symbols at each of the optimal coding nodes. After that, we focus our investigation at the next lower hierarchy level to identify the nodes inside the logical node where the logical summation of symbols (or simply network coding) takes place. For this, we construct minimum spanning trees on the logical nodes which takes into consideration the erasure probability at each edge. Finally, we discuss the effect of change in topology of the logical node and the edges associated with the nodes inside the logical node, and provide efficient algorithms to update the minimum spanning tree.