T he gradual migration of rivers across floodplains is punctuated by episodic shifts in river course called avulsions (1). Avulsions are natural phenomena; river relocation nourishes floodplains with water, sediment, and nutrients and results in vast plains of fertile farmlands and biodiverse ecosystems that support the most populous places on Earth (2)(3)(4). Avulsions are also responsible for devastating historical floods (1,5,6) and linked to the decline of early urban settlements (7). Avulsions occur quasi-periodically and persistently at the apex of alluvial fans and river deltas (1,8), resulting in their triangular shape. The time between successive avulsions can range from decades to millennia for different rivers (1,8,9), and direct observations of natural avulsions are rare. Consequently, the controls on where avulsions occur and how these sites will shift in response to climate change and human activity is poorly understood. Our understanding of the processes that control avulsion location are primarily based on physical experiments, numerical models, ancient river deposits (1,(8)(9)(10)(11)(12)(13)(14), and the analysis of river bifurcations (15), which can occur because of mechanisms other than avulsions (16). A time series of satellite imagery dating back to 1973 C.E. provides an opportunity to directly observe and characterize the global distribution of river avulsions to test classical and emerging theories (17).

We leveraged nearly 50 years of global satellite observations of river planform changes to locate avulsions. We focused on lobe-scale avulsions that occurred at the apex of river deltas and fans, including alluvial fans and fan deltas that build into a standing body of water from an adjacent highland. We defined avulsions as an abrupt and persistent change in the river course from the apex to the axial river or shoreline (Fig. 1). We identified avulsions from surface water maps derived from 30-m per pixel Landsat multispectral data from 1984 to 2020 C.E. (18), and from 60-m per pixel to 30-m per pixel Landsat imagery between 1973 and 2020 C.E. (19). We analyzed 113 historical river avulsions on fans and deltas, including 36 previously reported occurrences (Fig. 1 and table S1) (19).

Our results revealed 80 avulsions on coastal and inland river deltas and 33 avulsions on fans, captured by satellite imagery and historical maps (table S1). Snow and cloud cover in high-latitude regions and the spatial resolution of the satellite imagery affected avulsion documentation (19). Our compilation is therefore a representative-rather than exhaustiveglobal sample of avulsions. Avulsion sites covered 33°S to 54°N latitude, and were observed in tropical, temperate, and arid climates (Fig. 1). We found high avulsion density in the tropical islands of Papua New Guinea, Indonesia, and Madagascar (Fig. 1 and table S1) but did not observe avulsions in polar and snow climate zones. Rivers with avulsions covered a wide spectrum in modeled long-term water discharge (0.4 to 36,702 m 3 /s), suspended sediment discharge (2 to 38,101 kg/s), and estimated riverbed slopes (4 × 10 -5 to 2.6 × 10 -2 ) (Fig. 1 and table S1) (19). Study reaches on fans are steeper than on deltas, with median estimated riverbed slope (and interquartile range) of 4.3 × 10 -4 (0.001) on deltas and 2.7 × 10 -3 (0.011) on fans (Fig. 2A).

Avulsions on fans are thought to occur at the mouths of bedrock canyons or valleys, where rivers become unconfined (1,20,21) (fig. S4). Avulsions on deltas, however, lack a clear association with a canyon or valley and the processes that control avulsion location are debated (8,11,12,22,23). To assess the controls on avulsion sites, we extracted topographic swath profiles from 30-m spatial resolution global digital elevation models (Fig. 1) (19). Our results demonstrate that the avulsion sites on fans are always associated with at least a threefold slope drop in the topographic swath profiles (median of 6.5, interquartile range of 1.7; Fig. 2B), which is indicative of an abrupt valley-confinement change. This abrupt change can lead to a loss in fluvial sediment transport capacity (1, 9), causing focused sedimentation or present a steeper and more favorable path to the axial river, consistent with classical ideas (1,21). Enhanced riverbed aggradation perches the water surface above the surrounding floodplain (10,14,24), and later floods trigger an avulsion (1,10,14,24).

By contrast, the 80 avulsions on deltas were not coincident with an abrupt topographic change (median slope break in swath profiles of 1.28, interquartile range of 0.65; Figs. 1 and 2B and table S1). We used these observations to test emerging ideas about the controls of avulsion location on deltas. Theory, physical experiments, and limited field observations indicate that avulsions on deltas cluster within the backwater zone (8,22,25)-the downstream reach of rivers characterized by nonuniform flows (26,27). We estimated the backwater length scale, which approximates the upstream extent to which nonuniform flows prevail in alluvial rivers, defined as L b ¼ h bf =S, where h bf and S are the bankfull flow depth and riverbed slope upstream of the avulsion site, respectively (19,27,28). We also estimated the avulsion length (L A )-the streamwise distance from the avulsion site to the river mouth of the parent channel-from the satellite image that best captured the time of avulsion for each event (Fig. 1 and table S1) (19). Our measured L A on deltas ranged from 0.5 to 490 km globally (Figs. 1 and 3A and table S1). Our results reveal that 62.5% of avulsions on deltas (n = 50) have a backwater-scaled avulsion node with L A ≈ L b (Fig. 3A). For these cases, the dimensionless avulsion length, defined as 11), is 0.87 ± 0.38 [mean ± standard deviation (SD)], which is consistent with backwater-controlled avulsions (11,12,24). We interpreted that deltaic avulsions occur within the backwater zone because rivers during low flow decelerate in approach to the receiving basin, which causes sedimentation in the upstream part of the backwater zone ( 26 (fig. S4). Sedimentation is periodically interrupted by large floods, which cause an erosional wave to propagate from the river mouth to a distance upstream that is generally within the backwater zone (27, 29) (fig. S4). Over time, delta-lobe progradation causes riverbed aggradation and the sedimentation within the backwater zone during low flows-coupled with preferential riverbed scouring in the downstream accelerating part of the backwater zone during floods-causes a peak in riverbed sedimentation in the upstream part of the backwater zone (fig. S4), resulting in avulsions there (11,12).

Our results also reveal a separate class of deltaic avulsions (n = 30) that correlate with neither the backwater length scale (L A ≫ L b ; Fig. 3A) nor a valley-confinement break (Figs. 1 and2B). These avulsions have L Ã A ≫ 1 (mean and SD of 13.4 ± 13.0) and correspond with steep, sediment-laden rivers in deserts and tropical islands (figs. S9 and S10). In accordance with emerging theory (30), we hypothesized that the longitudinal extent of flood-driven scours in these rivers is more pronounced than in backwater-scaled deltas (fig. S4), which diminishes sedimentation within the backwater reach and thus causes L A ≫ L b (31). To test this hypothesis, we estimated the dimensionless flood duration, defined as T Ã e ¼ t scour =t adj , where t scour is the typical bankfull-overtopping flood duration and t adj is a bed-adjustment time scale (11) The dimensionless flood duration (T Ã e ) separated the data into two distinct classes of deltaic avulsions (Fig. 3B and fig. S7) (19).

A ≫ 1 are associated with T Ã e ≫ 1 (mean and SD of 168 ± 505). These rivers had t adj on the order of days to weeks such that flood-driven scours propagated beyond the backwater zone during typical floods and caused avulsions with L Ã A ≫ 1 (Fig. 3B and figs. S4 and S8). In addition, for rivers with T Ã e ≳ 1, the theoretical flood-driven scour length scale, L b ffiffiffiffiffi ffi T Ã e p ; appears to control the avulsion length, L A (Fig. 3B), rather than the backwater length scale. In comparison, backwater-scaled avulsions occurred only if T Ã e ≲ 1, which is typical of lowland rivers with a t adj of months to centuries (figs. S7 to S9). For these rivers, flood-driven scours were limited to the backwater zone during typical floods, resulting in backwater-scaled avulsion sites (L A ≈ L b ) (Fig. 3, A and B, and fig. S4).

Physics-based numerical models demonstrate that historical avulsion lengths are a diagnostic indicator of future avulsion sites even in the face of anthropogenic climate change and interference (30), which suggests that our global database provides a first-order prediction of future avulsion sites on deltas. However, numerical and field studies also indicate that considerable shoreline encroachment from accelerated sea level rise can shift the deltaic avulsion nodes upstream (34,35). In addition, our analysis implies that T Ã e ≈ 1 is a transition point between two scaling regimes for avulsions on deltas (Fig. 3B and fig. S8), where a further reduction in t adj can drive avulsion sites upstream. Agriculture and land use have enhanced the sediment loads of most global coastal rivers (36), and extensive dam infrastructure can reduce sediment caliber-changes likely to increase T Ã e . These changes can result in coastal rivers to transition beyond T Ã e ≈ 1, causing an upstream shift of their avulsion nodes. Our analysis indicates that inland river deltas and small, low-gradient coastal deltas in tropical islands (e.g., Indonesia) are most susceptible to transitioning from the backwater-scaled avulsion regime to the high-sediment-load modulated avulsion regime with changes in magnitude and duration of floods as well as sediment supply (19) (fig. S8), which may expose previously unaffected upstream communities to the risks of avulsion hazards.

Our results also highlight that changes in flood frequency caused by differing climates (37) or engineering (e.g., dams) can substantially affect the avulsion location on lowland deltas (Fig. 3C). The backwater-scaled avulsions in our dataset are variable with L Ã A ∈ 0:24; 1:62 ½ (Fig. 3, A andB). We investigated the causes of this variability on four deltas with multiple recorded avulsions: the Huanghe (n = 7) (9, 13) and Sulengguole rivers (n = 6) (35) in China, and the Cisanggarung (n = 3) and Cipunagara (n = 3) rivers in Java, Indonesia (Fig. 4, A andB these rivers is statistically different (one-way ANOVA test, P = 0.001), indicating that the observed L Ã A variability is not driven by chance. Instead, mean L Ã A increases with the frequency of bankfull-overtopping floods (Fig. 3C). This trend is consistent with numerical models (11), which show that more frequent floods enhance the longitudinal extent of flood-driven scours within the backwater zone, thereby increasing L Ã A . River deltas with multiple recorded avulsions also demonstrate the mobility of backwaterscaled avulsion sites with river mouth evolution. These examples provide a template for how backwater-scaled avulsion sites on deltas may respond to future changes in sediment supply and relative sea level. Backwater avulsion theory indicates that avulsions should shift upstream or downstream in tandem with river mouth evolution such that the avulsion node remains within the backwater zone (9,12,30,34). Our data support this idea wherein progradation of the Cisanggarung and the Cipunagara river deltas caused downstream shifts in their avulsion sites (Fig. 4) (19), similar to previous observations on the Huanghe (9, 13); further, the Sulengguole river mouth retreat caused the avulsion sites to migrate inland (Fig. 4C) (35). In all cases, the episodic avulsion-site migration was commensurate with the direction and magnitude of river mouth evolution such that L Ã A for each delta remained consistent with time (Fig. 4C). These results suggest that accelerated delta progradation associated with global agriculture and deforestation (36) will lead to the downstream migration of avulsion sites; however, drowning of river deltas from accelerated relative sea-level rise will shift avulsion sites upstream (30,34).

Our global analysis reveals distinct controls on avulsion locations on fans and deltas: fan avulsions are tied to an abrupt valleyconfinement change; however, the longitudinal extent of flood-driven scours sets the avulsion location on deltas, and dimension-less flood duration separates deltaic avulsions between those that are backwater-scaled and those that occur farther upstream. For most rivers, avulsions reoccur on time scales longer than the 50-yr record we analyzed, such that most avulsion locations have not been documented historically, and the devastating consequences of flooding following an avulsion have not been realized. Our work provides a predictive framework to assess future avulsion locations on fans and deltas, and their response to land use and climate change.