Accelerating TauDEM as a Scalable Hydrological Terrain Analysis Service on XSEDE
Terrain analysis using Digital Elevation Models (DEM) has been widely used in hydrology to derive topographic information for hydrologic analysis and modeling. Finer resolutions on DEMs have been shown to have significant impact on hydrologically important variables and improve the accuracy and reliability of results . However, as high-resolution DEMs become increasingly available, e.g., LiDAR-based DEMs (http://opentopography.org) and the high-resolution DEMs in the National Elevation Dataset produced by the U.S. Geological Survey (http://ned.usgs.gov), the computation cost of DEM-based hydrological modeling significantly increases  and, thus, makes desktop computer-based analyses extremely difficult. TauDEM is a parallel computing solution to tackle this computational challenge. A set of parallel hydrological terrain processing algorithms was developed in TauDEM to leverage the Message Passing Interface (MPI, http://mpi-forum.org) and MPI IO for efficient handling of the processing and input/output (I/O) of DEM data and TauDEM results on multiple processors. A multi-institutional effort has been ongoing to leverage expertise in multiple disciplines (i.e., hydrology, computational science, geographic information science, and geography) to scale TauDEM for analyzing large DEMs on national cyberinfrastructure such as XSEDE (http://xsede.org) and publish TauDEM as a widely accessible cyberinfrastructure-empowered science gateway application service. This is a collaboration among the TauDEM team at Utah State University (http://hydrology.usu.edu/taudem), XSEDE Extended Collaborative Support Services (ECSS ), the National Science Foundation (NSF)-funded CyberGIS project (http://cybergis.org), and the NSF OpenTopography data facility (http://opentopography.org). This paper presents the findings and experience in enhancing TauDEM for large-scale terrain analysis on massive computing resources provided on XSEDE through rigorous computational performance profiling and analysis. As a result, computational bottlenecks that were not observed on local clusters were identified and successfully eliminated. The improved TauDEM software scales to thousands of processors and is capable of handling 36GB DEM data, which was not possible on small-scale clusters. The software is now deployed on XSEDE and paves the way for the development of TauDEM science gateway services in order to provide our communities with TauDEM capabilities on high-resolution DEMs.