The growth of data used by data-intensive computations, e.g. Geographical Information Systems (GIS), has far outpaced the growth of the power of a single processor. The increasing demand of data-intensive applications calls for distributed computing. In this paper, we propose a high performance workflow system MRGIS, a parallel and distributed computing platform based on MapReduce clusters, to execute GIS applications efficiently. MRGIS consists of a design interface, a task scheduler, and a runtime support system. The design interface has two options: a GUI-based workflow designer and an API-based library for programming in Python. Given a GIS workflow, the scheduler analyzes data dependencies among tasks, then dispatches them to MapReduce clusters based on the current status of the system. Our experiment demonstrates that MRGIS can significantly improve the performance of GIS workflow execution.