A century of channel change caused by flow augmentation on Sixth Water Creek and Diamond Fork River, Utah, USA

The Diamond Fork River in central Utah, USA experienced extreme flow augmentation via transbasin flows. Beginning in 1916, irrigation water was delivered through a tributary, Sixth Water Creek, with daily summer flows regularly exceeding a 500‐year flood at the point of introduction. Flows were dramatically reduced by management action in 2004 but with mandated minimum flows. We examined the geomorphic response of Sixth Water and Diamond Fork using aerial imagery, lidar, topographic cross sections, and sediment transport measurements. River channel response varied with valley confinement and with position in the watershed, which determined the magnitude of augmented flow relative to natural floods and the amount of sediment supply. Confined, steep sections of Sixth Water incised many meters into bedrock, whereas partially confined and unconfined sections developed a braided form even under conditions of general incision. With the removal of flow augmentation, smaller natural floods on Sixth Water are unable to transport bed material and the channel remains static. On the alluvial lower Diamond Fork, a lower slope, upstream sediment supply, and larger natural floods produced a dynamic shifting channel that widened in response to natural floods. After flow augmentation, a coarsened bed is partially mobile and channel narrowing appears to be limited by artificial baseflow, which prevent vegetation establishment in the channel.

, a central topic in fluvial geomorphology is the flow that determines channel dimension. Hydraulic geometry, which describes the relationship between discharge and channel dimension, is a common method to characterize river geometry. Compilations of hydraulic geometry find that channel width typically scales with discharge to the one-half power (Dunne & Jerolmack, 2020;Leopold & Maddock, 1953;Pelletier, 2021). However, there is considerable variability around the average trend, as a number of factors, from hydrologic variability to morphologic complexity, complicate this relationship (Phillips et al., 2022). For example, relatively small floods of long duration or repeat occurrence can alter channel width (Andrews, 1980;Eagle et al., 2021;Gervasi et al., 2021;Slater et al., 2019;Wolman & Miller, 1960). And exceptionally large floods can produce either negligible change or extreme and long-lasting channel change, depending on the timing and geomorphic setting (Dethier et al., 2016;Pizzuto, 1994;Schumm & Lichty, 1963;Wolman & Eiler, 1958;Wolman & Gerson, 1978).
Large flows from transbasin flow augmentation provide a useful perspective on the connection between high flow and channel dimension. Rather than a disturbance and recovery pattern typical of floods, large augmented flows impose a new regime toward which the channel will adjust. Although the duration of high flow is the defining difference between floods and flow augmentation, there are other differences of note. Floods typically increase in magnitude with increasing drainage area, whereas augmented flow is generally introduced at one point in a drainage network such that the magnitude of augmentation relative to natural floods is larger upstream than downstream. Further, if the augmented flow is introduced without sediment, the upstream impact is likely to be erosional, thereby increasing sediment supply to downstream reaches. Downstream reaches thus experience augmented flows, increased sediment supply, and reduced influence of floods smaller than the augmented flows.
When floods larger than the augmentation do occur, recovery proceeds in the presence of increased flow and sediment transport capacity.
Channel response to high flows can be strongly influenced by local conditions, including lithology, valley confinement, valley slope, preexisting channel condition, and the prevalence of riparian vegetation (Brierley & Fryirs, 2004;Brunsden & Thornes, 1979;Dethier et al., 2016;Parida et al., 2017;Pfeiffer et al., 2019;Phillips, 2009;Schumm, 1973). Kellerhals et al. (1979) found that channel response to flow augmentation in bedrock systems was controlled by lithologic strength and valley geometry, whereas channels in unconsolidated terrace deposits rapidly established new channel forms as a function of the type of material and the preexisting channel slope. Response in alluvial channels varied from increased channel dimension for smaller augmentations to a transition from meandering to braided for larger augmented flows.
Other studies of channel response to flow augmentation illustrate the combined effect of high flow, preexisting channel form, and local valley conditions. A 2-3 times increase in mean annual flow on the Kemano River, British Columbia promoted channel widening and a reduction in sediment mobility, making the channel more susceptible to a large flood that incised and narrowed the channel (Church, 1995).
Channel response to flow augmentation in Rocky Mountain streams in Colorado is controlled by slope, valley setting, and channel grain size. Widening, coarsening, incision, and loss of riparian vegetation occurred primarily in low slope, unconfined reaches (Abbott, 1976;Dominick & O'Neill, 1998;. Bradley and Smith (1984) documented some widening and an increase in meander migration rate on the Milk River, Alberta in response to a threefold increase in summer baseflow, even though high flows remained unchanged.
The Little Bow River, Alberta, experienced slight widening and changes in vegetation in response to a threefold increase in augmented flow magnitude (Hillman et al., 2016).
Though the influence of flood flows on channel geometry is wellstudied, fewer studies have explicitly considered the influence of low flows on channel geometry. However, there is evidence that elevated baseflows on regulated rivers can influence channel dimensions via feedbacks with vegetation. In arid environments, increased baseflows provide root water for plants, which promotes riparian growth and narrowing (Friedman, 2018;Kui et al., 2017;Sankey et al., 2015). At the same time, persistent inundation can drown out plants, creating a channel with bare sediment to the extent of inundation (Chin et al., 2002;Friedman & Auble, 1999;Lenhart et al., 2013;Sankey et al., 2015).
We report here on channel response to an exceptionally large flow augmentation of long duration on Diamond Fork River (Utah, USA). Starting in 1916, transbasin flows for irrigation were delivered May-September into Sixth Water Creek, a tributary of the Diamond Fork River. The relative magnitude of the introduced flow was larger than any natural flood high in the Sixth Water drainage, decreasing to a 5-to 10-year flood at the mouth of the Diamond Fork River. The augmented flows were highly erosional in Sixth Water Creek, incising the channel and delivering sediment to the Diamond Fork River. The magnitude of the augmentation largely eliminated any influence of natural floods on Sixth Water Creek. Diamond Fork experienced a complex change driven by upstream sediment supply, long-term flow alteration, and large magnitude floods. Augmented flows were removed from most of the Sixth Water drainage in 1997 and from the entire river network in 2004, with the exception of legally mandated minimum flows that are substantially larger than natural baseflows.
Diamond Fork River offers the opportunity to examine channel response to 87 years of exceptionally large augmented flows, followed by the removal of those flows, and the possible influence of elevated baseflows. Our goals in this paper are to document channel response to this flow history, to evaluate how key factors influence channel response, and to assess the efficacy of hydraulic geometry under extreme circumstances. These factors include along-stream variation in the relative magnitude of flow augmentation compared to natural floods, upstream sediment delivery, variations in lithology, and variations in valley slope and confinement.   Figure 2). We delineated eight distinct process domains using the River Styles Framework-a process-based method to distinguish river segments (Brierley & Fryirs, 2004      and the period with mandated baseflows (1998 up to the present). For Diamond Fork, the time periods represent pre-augmentation (1908)(1909)(1910)(1911)(1912)(1913)(1914)(1915), the period when high flows were transported in the channel , and the period with mandated baseflows (2004 up to the present).

| METHODS
We used multiple datasets to analyze geomorphic change of Sixth Water and Diamond Fork in response to long-term flow augmentation. These include historical aerial photographs, hillslope and valley bottom topography from airborne lidar (Jones, 2018), streambed elevation from field measurements at stream gauging stations, and field data including topographic cross sections, sediment transport measurements, bed grain size measurements, and hillslope grain size measurements.

| Channel width
Channel width is highly responsive to changes in flow regime, and width can be measured from historical images; therefore, it is an important and robust tool for measuring geomorphic response to flow alteration. We mapped the active channel of Sixth Water and lower Diamond Fork at 1:1000 scale in ArcGIS from historical aerial imagery (Table 2). Most years had complete coverage of the study area, and we mapped the active channel from the Strawberry Tunnel Outlet to the intersection of Diamond Fork and Highway 6 when possible. All digitizing was performed by the same operator. We defined the active channel as the wetted channel plus the adjacent area where vegetation is unable to colonize because of fluvial scour or frequent inundation (Gendaszek et al., 2012;Lauer et al., 2017). We measured active channel width at 10-m intervals for each set of aerial imagery using the Planform Statistics Toolbox in ArcGIS (Lauer & Parker, 2008).
Error in channel width measurements can obscure real trends in width change if the magnitude of the error exceeds the magnitude of change. Therefore, it is important to characterize the magnitude of error. Error in width measurements results from digitization error and co-registration error. We measured the first by repeat digitization of channel margins (Donovan et al., 2019;Toone et al., 2014). We re-digitized select reaches that encompassed a variety of edge typesoverhanging vegetation, areas covered by shadow, and clear banks.
We calculated the standard deviation of channel width for each edge type in each year of imagery. We then assigned a single error metric to each process domain by calculating the percentage of each process domain covered by each edge type and calculated a weighted average of the standard deviation based on the proportion of each edge type.
For most sets of images, the digitization error was less than the resolution of the images, so we assigned the resolution as the error metric (Supporting Information S1).
We calculated co-registration error by constructing a spatially variable error surface from a network of ground control points (Donovan et al., 2019;Lea & Legleiter, 2016). We used the 2016 imagery as our reference image and calculated the distance between ground control points for every historic image and the 2016 imagery.
X error and Y error were calculated for every ground control point, and X and Y error surfaces were created for each year of imagery by natural neighbor interpolation. We extracted the value of X and Y errors at the left and right bank of the active channel polygon at 10-m intervals. We added the X and Y errors to the end-point coordinates, calculated the channel width from the shifted end points, and then subtracted the new width from the original measurement. This method calculates overestimation or underestimation of channel width because of differential georeferencing error. For example, if X and Y errors were equal on the left and right bank, the channel width measurement would not change despite any co-registration errors. We calculated the average co-registration error for each set of images by averaging the absolute value extracted at each 10-m interval. Total error was defined as the root sum of squares of the digitization and co-registration error measurements, and further detail is provided in the Supporting Information (S1).
We measured active channel width and drainage area for other regional streams to compare current-day hydraulic geometry relations for the Diamond Fork watershed and neighboring watersheds. We T A B L E 2 Aerial photographs used for planform measurements. another for other regional streams.
An increase in channel width from flow increase can be approximated using the hydraulic geometry relation between channel width and discharge, B ¼ aQ 1=2 , where B is a channel width, a is a coefficient, and Q is a characteristic discharge (Ashmore & Church, 2001;Blench, 1969;Leopold & Maddock, 1953). Assigning a characteristic discharge and assuming that the coefficient a is constant, the ratio of which is an opportunity to evaluate the limits of this approach for a very large flow augmentation of very long duration. Channel width before flow augmentation is unknown, and B 1 is estimated for each process domain using the relation between channel width and drainage area for unaugmented regional streams described above. Preaugmentation Q 1 is estimated as the 2-year flood from USGS peak flow regional curves (Kenney et al., 2008). For Q 2 , we use the median annual peak augmentation release measured at the Strawberry Tunnel at West Portal gage (10.6 m 3 /s).

| Sixth Water incision and sediment sources
In addition to adjusting their width, rivers often adjust their elevation One approach for reconstructing pre-erosion topography is to use a 3-dimensional spatial interpolation algorithm across an eroded area. Spatial interpolation algorithms were originally designed for constructing gridded datasets from non-uniformly distributed point or line geometry data. However, these algorithms can also be used to reconstruct information over an area based on surrounding data. Bergonse and Reis (2015) assessed the performance of six common interpolation algorithms in quantifying gully and hillslope erosion.
Their validation procedure identified the "Topo to Raster" Tool (implemented in ArcGIS) as the optimal spatial interpolation algorithm for constructing pre-erosion terrain in their study areas. We followed the interpolation and validation methods of Bergonse and Reis (2015), implementing the "Topo to Raster" Tool in ArcGIS 10.3 using contour lines created from the airborne lidar dataset ( Figure 5). The primary difference in this study is that we interpolated the incised portion of a river bottom rather than hillslope gullies.
The greatest limitation of three-dimensional spatial interpolation is typically a lack of information, which creates unrealistic estimates of the interpolated value in certain areas. We reduced this source of error by using high-resolution topographic data for the contours sur- sheds and comparing the reconstructed and existing topography, as described by Bergonse and Reis (2015). The grain size of the sediments transported during incision was estimated using field measurements of active hillslope sediment (Section 3.6). More detail on the interpolation method and how it was applied in this study are provided in the Supporting Information (S1).
Sediment derived from hillslopes can also represent a significant sediment source to a river, and changes in hillslope morphology through time can be used to assess whether a river is incising, stable, or aggradational. To estimate the change in hillslope-derived sediment through time, we measured the length of active hillslopes along Sixth Water in ArcMap for four sets of air photos-1956, 1981, 2006, and 2016. We defined active hillslopes as those in contact with the active channel with no visible vegetation growing at the toe of the slope.
We validated the measurements from the 2016 air photos with field observations along the length of Sixth Water in October 2017. We identified active hillslopes in the field and compared with the length measured from air photos. To constrain uncertainty, we measured the length of hillslopes in the 2016 imagery that were identified in the field but not on the aerial photographs, because of shadows or image quality obscuring the slope toe. We calculated the relative uncertainty in the 2016 photos (i.e., length that would not have been identified without field reconnaissance Ä total length of active slopes) and applied it to the total length measured in each year. of approximately 100 trees that could be reliably identified was recorded for each set of imagery. We measured the offset of each tree in successive images and calculated uncertainty by extracting the co-registration error from the spatially variable error surfaces defined above. Aerial photographs prior to 1993 could not be rectified with high enough precision to estimate movement on the landslide.

| Identification of alluvial deposits
Streambed elevation is an important metric to determine whether a river is incisional, stable, or aggradational. Historic measurements of streambed elevation are often spatially limited, but dated alluvial deposits can be used to determine past elevations of the streambed. DEM. This analysis does not account for subsequent erosion and deposition on these deposits.

| Historical streambed elevation
Each field measurement of discharge at a USGS stream gage is accompanied by a gage height and cross-section measurement that includes wetted channel width, area, and velocity. These measurements can be used to reconstruct streambed elevations, following the methods of Jacobson (1995) and Smelser and Schmidt (1998). This a robust mea-  (Wolman, 1954). We compared these measurements to the grain size measured in sediment sources to constrain the amount of bed material delivered to the channel via incision of the Sixth Water valley. We also compared bed grain size to the grain size of sediment in transport to characterize the transport regime.
We measured the grain size of potential sediment sources on Sixth Water and lower Diamond Fork in October 2017. Potential sources were identified in the field as either tributaries or hillslopes in contact with the channel. We collected tributary samples from the bed of tributaries and hillslope samples no more than 1 m above the channel. To determine grain size, we divided samples in the field using a gravelometer for grains larger than 64 mm and sieves for material larger than 22, 8, and 2 mm. The fraction of each size class was weighed to generate a mass-based grain size distribution. Samples weighed at least 5 kg, and the largest grain rarely represented more than 5% of the sample.

| Sediment transport measurements
We measured sediment entrainment and transport at eight locations along Sixth Water and Diamond Fork (Figure 1a). Diamond Fork was also generally wider than other regional rivers with similar drainage area and had weaker scaling with drainage area, though there were some regional rivers with similar channel width. F I G U R E 1 2 Historical channel width and channel width estimated from hydraulic geometry. Heavy central line is median channel width with light lines representing the 10th and 90th percentile widths. Numbers in each panel correspond to process domain numbers. Solid black lines are approximate channel widths estimated from drainage area for regional streams (B 1 ) and approximated for augmented flow (B 2 ). Major floods in 1952Major floods in , 1983Major floods in , and 1984  F I G U R E 1 3 The 2018 channel width as a function of drainage area for the Sixth Water/Diamond Fork system and for stream channels in neighboring watersheds. Location map shows watersheds used for analysis and location where width data were measured. The Diamond Fork watershed is outlined in thicker black.

| Incision of Sixth Water Creek
( Table 3). Measured bed grain size was rarely finer than 8 mm, so this volume represents bed material sediment that was contributed to the channel.

| Timing of hillslope erosion in Sixth Water
Channel incision on Sixth Water steepened adjacent hillslopes, which should increase lateral supply of sediment to the river. For available aerial imagery, the extent of unvegetated toe-slopes along Sixth Water was at a maximum in 1956 and consistently decreased over time to the last photo measured in 2016 ( Figure 15). This suggests that the amount of sediment supplied to the channel from hillslopes decreased over time.  F I G U R E 1 5 Length of active hillslopes along Sixth Water as measured from aerial photographs. Error was estimated based on field identification of active hillslopes that were covered by shadow in photographs.

| Streambed elevation
The mean streambed elevation at USGS stream gages reveals that the   (Abbott, 1976;Bradley & Smith, 1984;Church, 1995;Hillman et al., 2016;Kellerhals et al., 1979). We expand this body of literature by detailing geomorphic response over a long time period, by explicitly evaluating the influence of along-stream change in flow magnitude and by considering along-stream sediment dynamics caused by the flow alteration.
Augmented flows on Sixth Water were much larger and more persistent than natural floods and produced widespread incision and widening, but channel response to the sediment deficit varied with channel setting. Confined, steep reaches have limited capacity for lateral adjustment, so incision was the primary mode of adjustment. We document at least 6 m of incision in the headwaters where the magnitude of flow augmentation relative to natural flows was the greatest and the shale and mudstone bedrock is erosionally weak. This echoes the findings of Kellerhals et al. (1979), who found that augmentation in steep bedrock rivers caused "rapid and complete removal of overburden and erosion of soft bedrock." There is evidence of incision throughout Sixth Water, and we estimate an average of nearly 3 m of incision for the entirety of the Sixth Water valley (Table 3).
In partially confined reaches with lower slope, the response was more complex, as the channel was able to both incise and widen by cutting into adjacent valley fill. A broad braided channel is observed in the 1956 aerial images of the partially confined sections (Figure 8).
Though the magnitude of flow alteration at Sixth Water was greater than previous studies, several studies from Colorado found that low slope, lower confinement reaches experienced greater width change than confined sections, in line with our findings (Abbott, 1976;Dominick & O'Neill, 1998;. Although an active braid channel is often associated with a surplus of sediment supply over transport capacity, there is good evidence that the channel had incised by several meters by this time. The observation of incision in a wide, active braided channel may be attributed to the abnormally large river discharge and the need to transport a large upstream sediment supply (Erwin et al., 2011;Leonard et al., 2017;Major et al., 2019).
Valley incision and hillslope erosion decreased over the decades of the project, changing the character of the river. Decreased incision and hillslope erosion is evident from channel stabilization by vegetation and the decrease in active hillslope length over the last 40 years of the augmentation period ( Figure 15). With decreasing sediment supply, the persistent diversion flows cut a single channel into the braid plain ( Figure 18). As the channel incised, overbank flows were no longer able to move sediment on the former braid plain, whereas development of an armored boulder bed limited the extent of incision, causing the channel to stabilize and allowing vegetation to establish.
These changes led the channel to narrow below the width that would be predicted from hydraulic geometry ( Figure 14). Wohl and Dust (2012) Schumm & Lichty, 1963;Wolman & Gerson, 1978).
Throughout the Sixth Water/Diamond Fork drainage, the modern channel is significantly wider than nearby streams at the same drainage area (Figure 13). The legacy of the large augmentation likely contributes to this trend, but the influence of elevated baseflows may also support the wider channel. Sixth Water is mandated to carry at least 0.91 m 3 /s from May 1 to September 30 and 0.71 m 3 /s from October 1 to April 30; these flows exceed the natural baseflow by 7-9 times and exceed the natural 2-year flood for channel segments with drainage area less than 25 km 2 . On lower Diamond Fork, the mandated flows are 2.27 m 3 /s in summer and 1.70 m 3 /s in winter.
These flows are 5-8 times larger than natural baseflow. In the field, we observed that the mandated baseflows generally cover the entire channel bottom for most of the year, which may prevent vegetation from establishing within the channel.
Previous studies have noted the potential for baseflows to maintain channel width via interactions with vegetation. Sankey et al. (2015) found that heightened baseflows increased vegetation growth along the Colorado River in Grand Canyon, but that areas inundated during 10% of days had suppressed vegetation establishment. Similarly, ecogeomorphic model results suggest that increased base flows decrease the areal coverage of vegetation at the Yampa and Green River, Colorado (Diehl et al., 2020). Lenhart et al. (2013) observed that elevated May-July baseflow on the Minnesota River played a role in reducing vegetative cover of bars and lower banks, which enhanced river widening over 70 years. Yegua Creek, Texas, maintained its channel width 35 years after the closure of the Sommerville Dam. The dam operations decreased flood flows and increased baseflows. This led to bed aggradation and increased vegetation growth on the river banks.
The vegetation increased channel stability but was unable to establish in the active channel because of persistent inundation (Chin et al., 2002;Jennings, 1999). Though they did not have a mechanistic explanation, Bradley and Smith (1984) found that channel width increased with a 3 times increase in mean annual flow despite consistent high flows.
Coupled with these studies, our observations suggest that elevated baseflows commonly associated with flow regulation deserve further study for their role in determining the extent of possible narrowing.  (Abbott, 1976;Church, 1995;Dominick & O'Neill, 1998;Kellerhals et al., 1979;. Alteration for Sixth Water was so large that the river incised until mobile sediment was exhausted and the channel had very little capacity for morphologic adjustment and no longer conformed to the predictions of hydraulic geometry. For lower Diamond Fork, natural floods were larger than the augmentation flows, so the channel remained active throughout the period of augmentation. Under these conditions, confinement influenced the capacity for change, with the most change recorded in the least confined areas.

| CONCLUSION
For the case of flow decrease, the initial state is an important determinant of channel function. For Sixth Water, the channel is so altered that there is very little sediment transport or morphologic change. Lower Diamond Fork has more geomorphic change because natural floods are still capable of transporting sediment. But legacy alluvial deposits limit the potential for geomorphic change. Finally, we contribute to a growing body of literature about the influence of baseflows on channel width, as we illustrate that Sixth Water and Diamond Fork remain wider than regional streams despite a restoration of the natural flood regime. Elevated baseflows cover the channel bottom throughout the summer growing season and may prevent vegetation establishment, thereby prohibiting the channel from narrowing further. A similar scenario may be present in other regulated rivers with elevated baseflows, warranting further investigation.