Discussion and Conclusion

Our results suggest that previous regional studies indicating that size is the first order control on riverbank erosion at large scales (Hooke, 1980; Nanson and Hickin, 1986; Ielpi and Lapôtre, 2020) are largely correct (e.g. Figure 4). However, upon closer inspection, we reveal the limits of the global applicability of such relationships. We show large regional differences in this first order relationship that arise from the unique blend of secondary controls in each basin. For example, riverbank erosion in the Congo basin, which is characterized by varied drainage patterns due to localized geologic controls (Flügel et al., 2015), is notably lower than the global average. The world’s largest basin by discharge, the Amazon, has a lot of spread in this first order relationship, a result of the varied inputs of sediment load as shown by Constantine et al. (2014) and seen in the REAL data in Figure 4b. The Ob’ basin is also biased low, however, likely due to low flows in the winter and a discontinuous and sporadic permafrost setting. On the other hand, riverbank erosion in the Ganges-Brahmputra-Meghna basin is higher than the global average, which may be because of high relief and sediment yield in the tectonically active headwaters and thick alluvium downstream as well as frequent high discharge during monsoon season. As shown in these examples, river size does provide a good first order approximation of riverbank erosion for most basins globally, however these specific relationships can vary greatly between regions. REAL provides a starting point for analyzing these second order relationships at the global scale, and further research may be able to show the relative importance of geologic, hydrologic, and human influences on the rate of riverbank erosion in different basins.
We have revealed and clarified some limitations in quantifying river planform change using remotely sensed data products. First, the resolution of Landsat (~30 meters) requires decades between images to reliably detect the rate of change for all but the most dynamic rivers. This long timeframe comes with an added challenge, in that any changes in hydrology, human influence, or data quality over the intervening years can result in biased estimates of planform change. Higher resolution satellites like Sentinel-2 and the Planet constellation will become increasingly viable for global riverbank migration analysis as the length of record grows, and at the very least would be useful for multi-source remote sensing data fusion with the Landsat record. Second, the uncertainty in water classifications is a well-known problem in remote sensing of planform change, however the severity of the problem is in relation to the magnitude of actual planform change. We found that classification uncertainties propagated through our change calculations were greater than the observed planform change for only about 5.7% of rivers. Improving the representation of all sources of uncertainty and the impact on change calculations is a necessary next step remote sensing of river morphology change detection.
We have also provided the first global accounting of highly mobile rivers that may present a problem to remote sensing applications which treat rivers as immobile. Our earth observation satellite record is now several decades long, yet most river morphology data are static in time. As the satellite record grows, dynamic methods that track the movement and change of rivers in time will be critical. Even satellite missions over relatively short periods, like the SWOT satellite (Biancamaria et al., 2016), which is currently planned to launch in 2022 and will be operational for 3 years, will require some flexibility to accommodate the mismatch between SWORD centerline locations and highly mobile rivers. The REAL dataset provides the historical data to predict where and how soon these problems may arise in the future.