The OpenMP-parallel model TsunAWI for the simulation of tsunami propagation and inundation discretizes the shallow water quations on an unstructured linear conforming-nonconforming finite element grid. The data access to the variables on the unstructured grid is crucial for the computational performance. A reordering of the unknowns at elements, nodes, and edges along a space filling curve (SFC) guarantees data locality on all levels of the memory hierarchy, thus reducing cash misses and false sharing. The SFC resorting algorithm is presented and its influence on the serial and OpenMP parallel computation times of TsunAWI is compared to other common resorting algorithms like reverse Cuthill-McKee and minimum degree ordering.