Comparison of remote sensing based approaches for mapping bathymetry of shallow, clear water rivers

Shallow rivers provide important habitat for various aquatic and terrestrial species. The bathymetry of such environments is, however, difficult to measure as devices and approaches have been traditionally developed mainly for deeper waters. This study addresses the mapping of shallow water bathymetry with high spatial resolution and accuracy by comparing three remote sensing (RS) approaches: one based on echo sounding (active RS) and two on photogrammetry (passive RS): bathymetric Structure from Motion (SfM) and optical modelling. The tests were conducted on a 500 m long and ~30 m wide reach of sand-bedded meandering river: (1) during a rising spring flood (Q = 10–15 m3/s) with medium turbidity and high water color and; (2) during autumn low discharge (Q = 4 m3/s) with low turbidity and color. Each method was used to create bathymetric models. The models were compared with high precision field measurements with a mean point spacing of 0.86 m. Echo sounding provided the most accurate (ME~−0.02 m) and precise (SDE = ± 0.08 m) bathymetric models despite the high degree of interpolation needed. However, the echo sounding-based models were spatially restricted to areas deeper than 0.2 m and no small scale bathymetric variability was captured. The quality of the bathymetric SfM was highly sensitive to flow turbidity and color and therefore depth. However, bathymetric SfM suffers less from substrate variability, turbulent flow or large stones and cobbles on the river bed than optical modelling. Color and depth did affect optical model performance, but clearly less than the bathymetric SfM. The optical model accuracy improved in autumn with lower water color and turbidity (ME = −0.05) compared to spring (ME = −0.12). Correlations between the measured and modelled depth values (r = 0.96) and the models precision (SDE = 0.09–0.11) were close to those achieved with echo sounding. Shadows caused by riparian vegetation restricted the spatial extent of the optical models.