HYDROLOGICAL PROCESSES Hydrol. Process. 21, 1026–1044 (2007) Published online 21 February 2007 in Wiley InterScience (www.interscience.wiley.com) DOI: 10.1002/hyp.6277 Comparison of the performance of flow-routing algorithms used in GIS-based hydrologic analysis John P. Wilson*, Christine S. Lam and Yongxin Deng Department of Geography, University of Southern California, USA Abstract: Flow direction and specific catchment area were calculated for different flow-routing algorithms using TAPES-G and TauDEM. A fuzzy classification was used along with eight topo–climatic attributes to delineate six landscape classes from a 10-m USGS DEM. A series of maps and tabular outputs were produced to compare flow-routing predictions in different parts of the study area in the Santa Monica Mountains of southern California. The matched pair t-test was used to compare the performance of pairs of specific catchment area grids across six user-defined fuzzy landscape classes. The results show that (1) the ‘source’ cells predicted with the D1, DEMON, and FD8 algorithms were confined to hilltops; (2) two single flow-routing algorithms (Rho8, D8) produced poor results; and (3) the choice of flow-routing algorithm has potentially important consequences for the calculation of upslope contributing areas, sediment transport capacity, topographic wetness, and several other topographic indices. Copyright  2007 John Wiley & Sons, Ltd. KEY WORDS digital elevation models; flow-routing algorithms; GIS; terrain analysis Received 22 March 2004; Accepted 19 December 2005 INTRODUCTION The shape of the earth’s surface plays a fundamental role in the hydrologic, geomorphic, and ecological processes operating at the earth’s surface because it influences the movement of water, sediment, and other constituents within a landscape. As a consequence, terrain shape is also fundamental to the prediction of various surface and subsurface flow characteristics such as soil moisture and stream flow depth and velocity (Moore et al., 1988, 1991; Sulebak et al., 2000). Most of the patterns and processes that create and shape these characteristics operate at the meso- and topo-scales, and many of the solutions to envi- ronmental problems, such as accelerated soil erosion and nonpoint source pollution, require management strategies that are implemented at these scales as well (Moore and Hutchinson, 1991). Many primary and secondary topographic attributes have been computed from square-grid, triangulated irreg- ular and contour-based elevation networks, and incorpo- rated in numerous environmental classification schemes and modeling frameworks during the past two decades (e.g. Dikau, 1989; Band et al., 2000; Burrough et al., 2001). The provision of gridded elevation data sets by many national mapping agencies coupled with the devel- opment and wide distribution of methods for converting spot height and contour elevation data to square grids (see Hutchinson (1989) for one such method) has con- tributed to the tremendous growth in the popularity of * Correspondence to: John P. Wilson, Department of Geography, Univer- sity of Southern California, Los Angeles, CA 90089-0255, USA. E-mail: jpwilson@usc.edu gridded elevation data sets and grid-based algorithms for calculating topographic attributes. Three of the most popular attributes used in hydrologic models—specific catchment area (SCA), topographic wetness index, and sediment transport capacity index— rely on some form of flow-routing algorithm to calculate the upslope contributing area and transfer flow (water, sediment, nutrients) to lower adjacent points or areas in a landscape (Desmet and Govers, 1996; Mitasova and Mitas, 2002). Grid-based flow-routing algorithms allocate the outflow from a given cell to one or more downs- lope cells. There are two basic terrain features that may be responsible for directing flow from one digital eleva- tion model (DEM) pixel to multiple downslope neigh- bors. The first is the presence of submeter terrain fea- tures that direct flow to neighboring cells other than the cell marking the path of maximum descent and the sec- ond is the presence of two or more neighboring cells that are lower in elevation. Few, if any, DEM-based flow-routing algorithms can resolve submeter features, but several can direct flow to two or more downslope cells (Endreny and Woods, 2003). Several studies have compared the performance of two or more flow-routing algorithms using a variety of criteria but they generally stop short of describing their coincidence with observed runoff behavior and/or their impact on runoff prediction (Peters et al., 1995). Wolock and McCabe (1995) compared topographic wetness index distributions computed with single and multiple flow-routing algorithms using topographic maps and surveys spread across ten states. The single flow- routing algorithm was similar to D8 (O’Callaghan and Mark, 1984) and the multiple-flow direction algorithm Copyright  2007 John Wiley & Sons, Ltd.