Multi-step change of scale approach for deriving coarse-resolution flow directions

• This paper proposes a method of producing a low-resolution river drainage network from high-resolution elevation data.
• Performing multi-step changes of scale may improve the upscaling of flow directions.
• Improvement depends on the relation between basin size and grid element area.
• Improvements are beneficial for achieving a more physically sound hydrological model.
• The proposed approach does not eliminate the need to correct the river network manually.

When applying a large-scale grid distributed hydrological model, it is necessary to establish the connection of cells through a low-resolution river drainage network (RDN). The state-of-the-art approach to carry out this task starts at determining fine-resolution flow directions and then upscaling this information to the low-resolution grid. Several upscaling algorithms have been proposed recently and very reasonable results have been achieved. This paper aims at testing the following hypothesis: would the low-resolution RDN better resemble the finer-resolution RDN if the flow direction upscaling were applied iteratively with gradual changes in scale, instead of a direct jump from the finer to the coarser resolution? To test this hypothesis, we apply one of the recent upscaling algorithms proposed in the literature in two different ways: 1) following the conventional approach, and 2) through a multi-step procedure. Four large-scale South American basins are taken as study cases (drainage areas ranging from 20,000 to 642,000 km²). High-resolution (0.001° or 0.002°) RDN is extracted from SRTM-DEM and used to obtain low-resolution RDN of 0.1–0.5°, according to each basin. The results presented here show that the multi-step upscaling approach outperformed the conventional direct approach regarding the reproduction of high-resolution network patterns. Although this improvement often affects only a small number of cells (5–7% of the total), the impact of the changes can be significant when factors such as river flow paths, junctions and contributing areas are considered. This implies that the proposed approach may be useful in achieving a more physically-sound hydrological model, though it does not eliminate the need to analyze the results carefully and make occasional corrections.