Hydraulic roughness is a fundamental property in river research, as it directly affects water levels, flow strength and the associated sediment transport rates. However quantification of roughness is challenging, as it is not directly measurable in the field. In lowland rivers, bedforms are a major source of hydraulic roughness. Decades of research has focused on dunes to allow parameterisation of roughness. This study aims to establish the predictive capacity of current roughness predictors, and to identify reasons for the unexplained part of the variance in roughness. We quantify hydraulic roughness based on the Darcy-Weisbach friction factor calculated from hydraulic field data of a 78 km long trajectory of the Lower Rhine and River Waal in the Netherlands. This is compared to predicted roughness values based on dune geometry, and to the spatial distribution of the local topographic leeside angle, both inferred from bathymetric field data. Results from both approaches show the same general trend and magnitude of roughness values (friction factor f=0.019-0.069, mean 0.035). Roughness inferred from dune geometry explains 42% of the variance, for the best performing predictor. Efforts to explain the remaining variance from statistics of the local topographic leeside angles, which supposedly control flow separation, were unsuccessful. Unexpectedly, multi-kilometer depth oscillations explain 34% of the total roughness variations. We suggest that flow divergence associated with depth increase causes energy loss, which is reflected in an elevated hydraulic roughness. Depth variations occur in many rivers worldwide, which may imply a cause of flow resistance that needs further study.