Introduction

Humans have long relied on rivers as a source fresh water, as a means of transportation, and because of their role in shaping the landscape. Our attraction to rivers consequently means people, buildings, and agriculture are at risk from riverbank erosion. For example, in Bangladesh, an estimated average of 60,000 people per year are displaced by riverbank erosion of the Ganges-Brahmaputra-Jamuna system (Mutton and Haque, 2004). In this context, it is easy to understand why the United States Army Corps of Engineers spends more than $200 million per year managing the Mississippi River (James, 2019). However, sediment deposition and riverbank accretion create natural riparian habitats and fertile floodplains (Florsheim et al., 2008). Humans are dramatically modifying river sedimentation, both directly through river management, infrastructure, and land use development (Shields et al., 2000; Kondolf et al., 2002), as well as indirectly through climate change (East and Sankey, 2020). Land development has greatly increased the total sediment transport of rivers, but trapping in reservoirs has resulted in a net reduction in sediment export from land (Syvitski et al., 2005). These changes in water and sediment supply alter riverbank migration rates as the morphology adjusts to the new conditions (Brandt, 2000). The deposition and erosion of material on the banks acts as a temporary store of sediment, attenuating the impact of changing sediment supply, and as such is an important input to sediment flux models (Kronvang et al., 2013). As we think about the consequences of human impacts on climate and the landscape, assessing large-scale patterns in geomorphology is becoming increasingly valuable (Grill et al., 2019).
The many applications and broad importance of riverbank erosion has resulted in diverse measurement methods. On annual timescales and local spatial scales, bank migration and channel sedimentation can be measured volumetrically with bank and bathymetry surveying techniques (e.g., De Rose and Basher, 2011). Tectonic geomorphologists collect sediment cores and use radiometric dating to relate sediment accumulation rates (which are largely transported and deposited by river processes) to mountain-building uplift processes over geologic time scales (e.g., Walling, 1999). Over annual to decadal time scales and large spatial scales, fluvial geomorphologists rely on remote sensing methods like repeat lidar and optical remote sensing (e.g., De Rose and Basher, 2011). Several methods have been developed to allow efficient analysis of large domains from satellite or aerial imagery (Peixoto et al., 2009; Monegaglia et al., 2018). Many of these methods find and track the movement of the river centerline, which works well for well-behaved single-threaded channels, but have difficulty representing the complexity of anabranching or braided rivers (Parker et al., 2011; Schwenk et al., 2017). Other approaches quantify bank migration by the area that changes between river and non-river pixels (Rowland et al., 2016; Nagel et al., 2022). Because these methods calculate change on a pixel-by-pixel basis, they can more easily capture complex morphologies and irregular change that are not well represented by a river centerline. Among these many methods, there are also many ways to define riverbank migration; some consider only the erosion of stable, vegetated banks and bars (Boruah et al., 2008), while others include the shifting of channel bars (Lane et al., 2010). None of these methods have been used at global scales to produce a consistent dataset of riverbank migration across all climates and geographies. As a result, our existing geomorphological understanding of riverbank erosion and accretion are limited to, and biased by, the patchwork of studies and locations where we have observations.
Predicting and modeling bank erosion as a means of understanding sediment processes has a long history in geomorphology. Over that history, variables that describe the overall size of a river, such as catchment area, discharge, and width have all been found to be highly correlated with bank erosion (Hooke, 1980; Nanson and Hickin, 1986). River size is considered a “first order” control on bank migration (i.e. big rivers move faster than small rivers), and many researchers now investigate “second order” influences on bank erosion by first normalizing bank migration by the river width (Jarriel et al., 2021). Regardless of scale, we know from physical modelling that the rate of bank erosion is a balance between shear stress and bank resistivity (Ikeda et al., 1981). This balance points to second order controls that affect the interaction between flow and the banks. However, unlike the first order controls described above, the results of these studies are varied. For example, Constantine et al. (2014) found that suspended sediment flux is the primary control of width-normalized bank migration rate for 20 river reaches in the Amazon Basin. Ielpi and Lapôtre (2020) showed that the presence of vegetation slowed down width-normalized bank migration by an order of magnitude using a global sample of 983 meanders. Both of these variables can be heavily modified by human interventions, such as damming (Shields et al., 2000) and bank stabilization with concrete or vegetation (Grizzetti et al., 2017). These examples show the range of second order controls in the literature which arise from the different subsets of rivers studied. The interactions and hierarchies of these controls at the global scale have not yet fully been evaluated.
Here, we present the first global observations of decadal-scale average riverbank erosion and accretion for all rivers wider than 150 meters. We use a surface water occurrence dataset derived from Landsat satellite imagery, a river centerline dataset, and cloud computing to efficiently estimate decadal-scale average bank erosion and accretion rates. We analyze these data in the context of previous literature to show that existing methods of estimating riverbank erosion and accretion accurately capture large-scale patterns. However, the specific form of these relationships between geomorphic predictors and riverbank erosion and accretion varies from basin to basin. These data importantly confirm geomorphic scaling theories and present a new, uniform dataset that can help advance geomorphic modeling efforts.