The algorithm implementation involved reading the header information from the input ERDAS/Lan map file, determining partition size, reading the input map, performing cluster identification, and tabulating the results. In the sequential implementation (SI), reading the header information and map data were performed using the C++ language file input/output functions. The map header contains sufficient information to determine the appropriate size of the data structures for the csize and matrix working arrays. The number of rows read and processed by the SI was limited to 1024. This limitation is due to limited physical memory resources, i.e., random access memory and virtual memory swap space on the Sun SPARCstation's local hard disk drive. If the number of rows exceeds 1024, the algorithm will partition the map and perform cluster identification on each of the partitions in sequential order. After the algorithm completes, the csize array on contains the appropriate number of pixels in each cluster and the matrix array contains the temporary cluster label for each non-zero pixel in the input map. Final cluster statistics are obtained by traversing the csize array and tabulating the positive non-zero values.