Parallelized Two-Dimensional Dam-Break Flood Analysis with Dynamic Data Structures

Numerical solution of shallow water equations using an explicit scheme are generally implemented using array type data structures. The computation is performed by sweeping the entire two-dimensional computational domain in row or column-wise order, starting with the cell in one corner and ending with the cell in the opposite corner. Although much simpler to program, this technique presents an important drawback due to the fact that it does not distinguish between wet (cells containing water) and dry cells. During a simulation, a considerable amount of computational time is spent sweeping over dry cells and skipping them once they are detected to be dry. The present study introduces a new solution algorithm which computes only those cells of the computational domain that are already wet or are becoming wet. This algorithm is especially useful for accelerating computations over very large Cartesian grids when only a small portion of cells are wet. The tracking of wet cells is achieved using dynamic data structures. This implementation takes advantage of the CFL (Courant-Friedrichs-Lewy) condition for explicit schemes, which requires that the flood propagates a fraction of the cell size during a time step. The CFL condition ensures that only the dry cells that are direct neighbors of wet cells can become wet cells. A multi-threaded parallelized version of this algorithm is implemented and tested in the CCHE2D-FLOOD TM , which is the solver of the DSS-WISE TM software developed by the National Center for Computational Hydroscience and Engineering, the University of Mississippi. The paper describes the implementation of the algorithm and discusses the performance gain achieved for different ratios of wet cells to the total number of cells.