Accurate delineation of drainage networks (DNs) is crucial for hydrological or hydraulic modelling, and the com-24 prehension of fluvial processes. This task presents challenging aspects in complex lowland terrains with subtle 25 relief and particularly for data poor-areas like the Cuvelai river basin (CRB), Namibia, where the present study 26 takes place. In the CRB standard methods of drainage network extraction from low resolution gridded digital 27 elevation models (DEMs) are unsuitable, hence airborne Light Detection and Ranging (LiDAR) solutions have 28 been utilized. However, LiDAR also presents challenges to large areal applications, especially with a surface 29 roughness exceeding the capacity of numerous algorithms. Indeed, LiDAR-based DEMs (2 and 50 m resolutions) 30 need to be hydrologically corrected and smoothed to enable the extraction of scale-relevant geomorphologic 31 features such as DNs. In the present contribution, channels from topographic maps (blue lines) were compared 32 to those from hydrologically corrected and uncorrected LiDAR DEMs, heads-up digitized channels from high-33 resolution digital aerial orthophotographs, field-mapped channels and auxiliary data. The 'maximum gradient 34 deterministic eight (D8)' GIS algorithm was applied to the corrected and uncorrected LiDAR DEMs using two 35 network extraction methods: area threshold support and curvature/drop analysis. Different progressive flow 36 accumulation threshold values (12) were used to delineate channels with these methods. Validation was per-37 formed between the field-mapped channels, the modelled channels and those derived from multiple sources. 38 Additionally, spatial and quantitative analyses were performed on geomorphologic parameters and indices.39 The results have shown that hydrologically corrected LiDAR DEMs offer useful details for identifying low order 40 stream segments in headwaters, while blue lines derived from the national hydrography datasets for watersheds, 41 located in elevated and low-lying areas of the study area, underestimated total stream lengths for field-mapped 42 channels by −15.3% and −88.5%, respectively. This study also confirmed that DNs can be extracted from com-43 plex low-terrain areas with standard GIS algorithms and accurate field data. The results will aid national mapping 44 agencies in data-poor regions to modernize their national hydrography datasets and to account for changing land 45 surface conditions that can affect channel spatial arrangements over time. Angola and Namibia (Fig. 1), encompasses these and many other 64 challenges related to flood risk management.
65The CRB, where this research has taken place, displays a series of