River reach length and slope estimates for large-scale hydrological models based on a relatively high-resolution digital elevation model

SummaryApplication of regular grid-based distributed hydrological models requires information related to river drainage networks, such as flow directions, flow accumulated areas, basin delineation, and length and slope of river reaches for every grid cell. Most of those data may be extracted automatically from Digital Elevation Models (DEM). However, when dealing with large basins, available DEMs are usually in higher spatial resolution than the model grid. Upscaling procedures have therefore been developed to extract low-resolution flow directions and flow accumulated areas from relatively high-resolution DEMs. This paper presents a new methodology, which extends the upscaling method, to automatically extract length and slope of river reaches for large-scale grid-based hydrological models. The proposed method is new and is believed to be the first attempt to produce such information automatically from DEMs. The methodology was applied to parts of the Uruguay river basin in South-America, using the globally-wide available DEM produced by the Shuttle Radar Topography Mission (SRTM-90m). Quality of results was assessed by comparing calculated river lengths with distances measured over vectorized river networks, which were assumed to be correct. It was shown that the proposed methodology adequately assigns river lengths to every model cell, while ensuring that the whole length of the streams is considered. The effect of DEM resolution on calculated river length errors was analyzed by resampling the SRTM DEM to three different cell sizes, with best results obtained for the higher resolution DEM. The modification of the original high-resolution DEM through the process known as stream burning was also tested, largely improving the quality of the results. Finally, optimal distance transforms were used for the calculation of distance increments, instead of Euclidean local distances, giving results that were generally better. The quality of results for calculated slope values could not be assessed because no reliable data were available for comparison.