Background

The closure of Flaming Gorge Dam had a substantial effect on the hydrology of the upper Green River. Annual peak discharges released from Flaming Gorge Dam are approximately 50 percent smaller than the pre-dam flood peaks (fig 2A), as measured at USGS Green River near Greendale, Utah, 09234500 gaging station (USGS, 2016), herein referred to as the “Greendale gaging station” (Grams and Schmidt, 2002; Topping and others, 2018). In addition, dam operations have substantially increased baseflow and have introduced daily fluctuations in discharge for power generation. Farther downstream in the middle Green River, the Yampa River partially enhances the reduced annual flood peaks; however, operation of Flaming Gorge Dam has caused an approximate 30 percent reduction in peak discharge in the middle Green River, as measured at the Jensen gaging station (Grams and Schmidt, 2002; Topping and others, 2018) (fig. 2B). This 30 percent reduction is solely caused by Flaming Gorge Dam operations because there has been no long-term decrease in the annual flood peaks within the Yampa River (Manners and others, 2014; Topping and others, 2018).

Beginning in the early 1990s, Flaming Gorge Dam releases were modified such that larger releases occurred in the spring (May and June) to more closely mimic the natural hydrology (U.S. Fish and Wildlife Service, 1992; Muth and others, 2000; U.S. Bureau of Reclamation, 2005, 2006; Bestgen and others, 2011; LaGory and others, 2012, 2019). Additionally, daily fluctuations were reduced, and baseflows were modified. Despite the post-1992 increases in Flaming Gorge Dam spring releases, the magnitude of annual flood peaks in the middle Green River has not changed substantially (fig. 2B) because peak Flaming Gorge Dam releases are generally offset from the annual peak of the Yampa River. Furthermore, there have been no statistically significant changes in the duration of floods above the 1.5-year recurrence interval discharge since 1992, also reflecting the offset between Flaming Gorge Dam releases and the annual peak discharge of the Yampa River. Thus, annual flood peaks in the middle Green River, which likely drive much of the channel change in the alluvial segments within DINO, remain similar to the annual flood peaks that occurred between the closure of Flaming Gorge Dam and the early 1990s, when much of the geomorphic data in the middle Green River were last collected and analyzed by Grams and Schmidt (2002).

Grams and Schmidt (2002) showed that the alluvial segments of the Green River within DINO narrowed substantially between the closure of Flaming Gorge Dam and 1993. In Echo Park, immediately downstream from the confluence of the Yampa and Green Rivers, the Green River narrowed by 10 percent. Farther downstream in the combined Island and Rainbow Parks segment, the river narrowed by 25 percent. This percentage of narrowing in the alluvial segments was greater than in the canyon-bound segments downstream from the confluence of the Yampa and Green Rivers; the channel in Whirlpool and Split Mountain Canyons narrowed by 11 and 12 percent, respectively (figure 1B). In the alluvial segments, channel narrowing occurred primarily by the abandonment of secondary channels around islands and the deposition of fine sediment along the channel margins, which formed a post-dam floodplain inset within pre-dam flood deposits (Grams and Schmidt, 2002). Because of the reduction in flood peak magnitude by Flaming Gorge Dam, the lowest pre-dam deposit, referred to as “the intermediate bench” (Grams and Schmidt, 2002) is rarely inundated. Gessler and Moser (2001) also showed that some secondary channels were continuing to fill with sediment and could be abandoned if sediment deposition continued. It was unclear whether channel narrowing in the alluvial segments within DINO has continued since 1993, and this study seeks to address this uncertainty.

Despite the long-term decrease in annual flood peaks caused by Flaming Gorge Dam, erosion and (or) deposition of sediment in the alluvial segments of the middle Green River in DINO is strongly dependent upon the streamflow and sediment dynamics of the Yampa River for two reasons: (1) the Yampa River supplies most of the sediment, and (2) annual peak floods from the Yampa River can substantially exceed the peak releases from Flaming Gorge Dam. Analyses of sediment budgets for the sediment budget reach that brackets Echo, Island, and Rainbow Parks (refer to USGS Grand Canyon Monitoring and Research Center at https://www.gcmrc.gov/discharge_qw_sediment/reaches/DINO [Sibley and others, 2015]) show that when the annual flood of the Yampa River is small (that is, annual peak discharge is less than 425 cubic meters per second [m³/s], and median March–July discharge is less than 99 m³/s), sand likely accumulates somewhere between the downstream end of Deerlodge Park on the Yampa River, and Green River near Jensen, Utah (fig. 1B) (Topping and others, 2025). However, when either the annual peak discharge or median March–July streamflow on the Yampa River exceed those thresholds, sand is likely eroded from this river segment.

Since 1996, when channel cross sections were last measured in Echo, Island, and Rainbow Parks, the peak discharge of the Yampa River annual flood was small (less than 425 m³/s) in 16 out of 26 years (figure 2) and the March–July median discharge was small (less than 99 m³/s) in 17 years. Together, there were 16 years when discharge conditions were less than both thresholds. Albeit the magnitude of channel change in any given year is unknown, the fact that the annual flood peak and the March–July median discharge failed to exceed the above thresholds in approximately 60 percent of the years since 1996 suggests that it is likely that sand has accumulated in Echo, Island, and Rainbow Parks since the mid- to late 1990s. Although changes in sediment storage can occur in both alluvial and canyon-bound segments, we hypothesize that most of the change in sediment storage within DINO occurs in the alluvial segments of Echo, Island, and Rainbow Parks, because those segments are generally wider than the canyons, and there is more space to accommodate increases in sediment storage compared to the canyon-bound segments.

Purpose and Scope

The purposes of this study are threefold: (1) determine whether changes in sediment storage have occurred in the middle Green River in Echo, Island, and Rainbow Parks between the 1990s and 2019, (2) determine the style of any resultant channel adjustment arising from net sediment deposition or erosion in these river segments, and (3) evaluate the Topping and others (2025) hypothesis that certain Yampa River hydrologic thresholds can prevent sand from accumulating between Deerlodge Park, Colorado, and Jensen, Utah (fig. 1). This report presents the results from aerial-photograph analyses and a June 2019 resurvey of the channel cross sections and bathymetry in Echo, Island, and Rainbow Parks that were originally surveyed in 1993 and 1994. Cross-section data collected by Grams (1997) in the 1990s and all cross section and bathymetric data collected during our 2019 resurvey are available from the USGS data release that accompanies this report (Dean and others, 2024).

Study Area

Our study area consists of the meandering segments within the alluvial valleys of the middle Green River within DINO, including Echo, Island, and Rainbow Parks (fig. 1B). Echo Park is the narrowest of these valleys, beginning just above the confluence of the Yampa and Green Rivers near river km (rkm) 106.0. River kilometers are measured downstream from Flaming Gorge Dam, Utah. The Green River within Echo Park is approximately 3.5 km long, extending downstream to approximately rkm 109.5. At low flow, the Green River in Echo Park is multi-threaded among migrating sand and gravel bars; at high flow, the channel is single threaded and is bounded by vegetated floodplains (fig. 3).

Downstream from Echo Park, the Green River flows through Whirlpool Canyon for approximately 14 km before entering Island Park. The Green River in Island Park is approximately 7 km long extending from rkm 123.8 to 130.8. The channel in Island Park is anabranching, woven among stable floodplain islands, although some single-threaded reaches exist (fig. 4). In the upstream half of Island Park, the bed of the Green River is composed of plane-bedded gravel alternating with migrating dune fields composed of sand; note that the term gravel includes all size classes of sediment coarser than 2 millimeters (mm) (Krumbein, 1934). The channel bed in the downstream half of Island Park is composed primarily of migrating dune fields of sand, although small, localized patches of gravel are also present.

The Green River in Rainbow Park, which is immediately downstream from Island Park, is approximately 4 km long and extends from rkm 130.8 to 134.8 (fig. 4). The Green River within Rainbow Park is also anabranching, and the channel bed primarily consists of sand dunes and lesser amounts of gravel, with active bars that are emergent at low flow. In this study, we collected data in the upstream-most 2.5 km of Rainbow Park, upstream from the Rainbow Park boat ramp.

Aerial-Photograph Analyses

We analyzed aerial photographs of Echo, Island, and Rainbow Parks taken in 1993 through 2019. For Echo Park, photographs used were collected in 1993, 2006, 2013, 2015, and 2019. For Island and Rainbow Parks, photographs used were collected in 1993, 2006, 2014, and 2018. Photographs also exist for 2009, 2011, 2016, and 2017, but were not used either because high flows at the time of the photographs limited the ability to accurately map in-channel features or because complete photograph coverage of the river in the study area did not exist. The 1993 photographs were acquired from the USGS National Aerial Photograph Program (NAPP; USGS, 1993), and all subsequent photographs were acquired from the U.S. Department of Agriculture (USDA) National Agriculture Imagery Program (NAIP; USDA, 2006, 2013, 2014, 2015, 2018, 2019). In all analyzed photographs, we mapped the boundaries of the active channel, which consisted of the wet parts of the channel, and the active sand deposits in the channel and on the banks that were generally devoid of vegetation. Although small amounts of sparse vegetation were present on some active sand deposits, these areas were included within the boundaries of the active channel. Active-channel width was determined by dividing the area of the active channel by the length of the channel centerline. For each photograph set in a given year, the percent change in the active-channel width was calculated relative to the 1993 width. We analyzed changes in width for each segment and for individual 1-km-long reaches roughly centered on the rkm.

We calculated uncertainty in the active channel width for each segment using a method adapted from Mount and others (2003), Swanson and others (2011), and Dean and Topping (2019, 2024), where active-channel width error (E_ω) is defined as follows:

The term $\sqrt{2\rho R}$is the error associated with delineation of the bank lines. A large source of uncertainty in active-channel width calculations involved the delineation of sparsely vegetated channel-margin deposits, because it was unclear whether these deposits resided within the active channel, or whether they were part of the floodplain. In this study, we included all the sparsely vegetated deposits within the extent of the active channel, because it appeared that the bare sand on these deposits was regularly reworked by floods. However, that decision was uncertain. Thus, to calculate $\rho$, we divided the area of active channel deposits with sparse vegetation by the length of the segment to get active channel width uncertainty for that segment. Since all active-channel delineations included the sparsely vegetated deposits within the active channel, $\rho$ was only applied to the negative error bars, because the active channel would become smaller if those deposits were instead mapped outside of the active channel.

To calculate $\text{Θ}$, we identified permanent, identifiable points (for example, the corners of large boulders) in the alluvial valleys using a 2015 light detection and ranging (LiDAR) surface collected by the State of Utah and others (2019). We identified seven points in or near Echo Park and six points in Island and Rainbow Parks. We then identified the same points in each year’s set of aerial photographs, and calculated the average offset for each set of images for each river segment.

Channel Cross-Section Installation and Initial Surveys

Between November 1993 and September–October 1994, Utah State University (Grams, 1997) established a network of 66 cross sections extending from approximately 1.5 km downstream from Flaming Gorge Dam to Split Mountain Canyon, within DINO. The endpoints of these cross sections were monumented with either rebar or fence posts. In some cases, existing fence posts were used as monuments. Topographic data were collected in a local coordinate system and were not georeferenced. The topography of the floodplain at these cross sections was measured using a total station, and the topography of the wetted portion of the channel was measured using a depth sounder. Resurveys were conducted in June 1995, August 1995, May 1996, and June 1996. Only a subset of the cross sections was measured during each resurvey trip (table 1). Cross sections 32–34 were established in Echo Park, 48a–53 in Island Park, and 54–56 in Rainbow Park, upstream from the boat ramp.

1998 Channel Bathymetry Collected Using an Acoustic-Doppler Current Profiler

In June 1998, Gessler and Moser (2001) measured channel bathymetry using an acoustic-Doppler current profiler (ADCP) over an approximate 3.2-km-long reach in Island Park (fig. 4).

Geographic positioning of the ADCP was acquired using a real-time kinematic (RTK) Global Positioning System (GPS) installed over known and surveyed control points, whose coordinates were transmitted to the ADCP software using a radio link. Horizontal geographic positions were acquired using the North American Datum of 1983 (NAD 83) and were subsequently converted to the State Plane Coordinate System (Utah central zone). Ellipsoid heights were acquired using the North American Vertical Datum of 1988 (NAVD 88). Geoid elevations were calculated post-hoc for analyses in this report by adding −15.93 meters (m) to the ellipsoid heights using the Geoid 18 model. The horizontal and vertical accuracy of the surveyed control points was reported to be ±0.09 m (Gessler and Moser, 2001). Gessler and Moser (2001) reported that their 1998 ADCP bathymetric measurements had a positive bias of 0.3048 m, thus we subtracted 0.3048 m from their ADCP elevations.

2019 Cross-Section Resurveys using Multibeam Sonar and Real-Time Kinematic Global Navigation Satellite System

In June 2019, the USGS resurveyed the cross sections within Echo, Island, and Rainbow Parks using multibeam sonar to measure channel bathymetry and an RTK global navigation satellite system (GNSS) to measure the topography of emergent bars and floodplains (Dean and others, 2024). In Echo Park, surveys were conducted at a discharge of 566 m³/s, as measured at the Jensen gaging station (USGS, 2016), which has a recurrence interval of approximately 2.4 years. In Island and Rainbow Parks, surveys were conducted at discharges between 507 and 518 m³/s as measured at the Jensen gaging station (USGS, 2016), which both have recurrence intervals of approximately 1.7 years. Survey control points were established, and static GNSS data were collected using GNSS base stations. Static data files were uploaded to the National Geodetic Survey Online Positioning User Service (https://www.ngs.noaa.gov/OPUS/) to obtain positional accuracy within 1–2 centimeters (cm) in the horizontal dimension and 4–5 cm in the vertical dimension.

The endpoints of each cross section were georeferenced using RTK GNSS, thus tying the local coordinates of the cross-section endpoints to the Universal Transverse Mercator (UTM) coordinate system (zone 12 north). The emergent topography along each cross section was measured by surveying along the line connecting the two endpoints of the cross section using the “stake out” procedure in Trimble Access data collection software. Submerged bathymetry was measured using a Norbit iWBMSh multibeam echosounder with an integrated Applanix OceanMaster GNSS-aided inertial navigation system. Multibeam data were acquired using RTK positioning in Hypack 2017, and post-processed using PosPac MMS v8.0 to compute the smoothed best estimate of boat trajectory (SBET) for the surveys. Raw data were edited by hand and by using Hypack filters to remove bad soundings, and the SBET files were applied to the edited data. Edited soundings were filtered within Hypack to a uniform density of 1 point per 0.25 m² in Echo and Island Parks and 1 point per 0.5 m² in Rainbow Park, using the mean elevation for each grid cell.

In Echo Park, the entire wetted channel bed was surveyed using multibeam sonar (fig. 3). In some areas, obstruction of the sky by canyon walls limited the number of satellites useable for RTK GNSS; thus, the endpoints and floodplains of cross sections 32 and 33 (fig. 3) could not be surveyed. Although the entire channel bed was surveyed with sonar, it was impossible to determine exactly how the channel bed had changed at cross sections 32 and 33 because the endpoints of the cross sections were not initially georeferenced to a projected coordinate system, and they could not be accurately located due to poor RTK GNSS coverage. Using geographic information system (GIS) software, bathymetry data along cross section 34 were extracted from the sonar data and combined with the RTK GNSS data to obtain topography for the entire cross section.

In Island Park, swaths of the channel bed that encompassed each cross-section line were surveyed using multibeam sonar for cross sections 48a–52. A longitudinal profile of the channel bed was also surveyed using the sonar between the cross-section swaths. Starting approximately 850 m downstream from cross section 52, we mapped the entire channel bed downstream to the Rainbow Park boat ramp (fig. 4). Using GIS software, bathymetry data along each cross-section line were extracted from the sonar data and combined with the RTK GNSS data of the above-water topography to obtain the full cross-section topography.

Changes in Cross-Sectional Area, Channel Width, and Bed Elevation

Changes in cross-sectional area indicate whether sediment has deposited or eroded within each cross section. Deposition may result in channel narrowing, bed aggradation, or vertical floodplain accretion with no change in channel width. Conversely, erosion may result in channel widening, bed scour, or vertical floodplain scour with no change in channel width. Thus, to determine changes in cross-sectional area, the method used must be able to detect lateral changes in width, and vertical changes in the bed and floodplain. To account for this, we calculated cross-sectional area as the area between the top of the highest, most stable floodplain and the channel bed. The top of the floodplain was defined by the inflection point between the relatively vertical channel bank and the relatively flat floodplain top (fig. 5). Variations in accuracy between surveys can result in slight differences (<5 cm) in the elevations of the top of the floodplains. To ensure that changes in area were only a result of topographic change, and not because of survey uncertainty and (or) error in the elevations of the floodplain, we set the elevations of the top of the floodplain to be the same among all years of surveys at an individual cross section (fig. 5). Note that for cross sections 51 and 52, which exist in multi-threaded parts of the channel, the change in cross-sectional area was measured separately for each channel because the channels are separated by large, stable islands. Thus, the total cross-sectional area involved in channel change is a small proportion of the total cross-sectional area between the endpoints of these cross sections.

Changes in channel width may occur through the lateral accretion of sediment on the channel banks, lateral bank erosion, or the vertical accretion of relatively low-elevation surfaces as they are converted to floodplains. Since changes in channel width may occur by both lateral and vertical depositional and (or) erosional processes, there is often an array of geomorphic surfaces and (or) floodplains that exist at different elevations above the channel bed. Thus, changes in channel width were measured at the cross-section locations by identifying the width of the channel at the water surface elevation surveyed in 2019 (and not simply the line that was previously used to calculate changes in cross-sectional area, as that line was delineated from the highest, relatively static floodplain surface, which does not typically intersect actively changing surfaces at lower elevations). The discharges during the 2019 surveys were equivalent to an approximate 2.4-year flood in Echo Park and an approximate 1.7-year flood in Island and Rainbow Parks. This operation allowed us to directly compare the changes in channel width determined from the surveys to the changes in active-channel width inferred from aerial photographs at the location of each cross section. In general, these two measurements should be highly correlated; however, they may differ slightly because of the way they were quantified. The channel width measured from the cross-section surveys represents the wetted width of the channel at the specific surveyed discharge, whereas the active-channel width measured on the aerial photographs, was primarily based on the presence or absence of vegetation. Demonstrating that channel widths measured on aerial photographs are similar to channel widths measured on the cross-section surveys provides the ability to then relate changes in channel width on the aerial photographs to changes in sediment storage at the cross sections.

Changes in the minimum bed elevation were also quantified at the channel cross sections between each successive survey. For a more spatially robust analysis of bed-elevation change, we analyzed changes in the channel bed elevation between the 1998 ADCP survey (Gessler and Moser, 2001) and the 2019 multibeam-sonar survey. The 1998 ADCP data were not collected at a spatial resolution sufficient to build a three-dimensional digital elevation model (DEM), thus it was not possible to simply compare changes of the overlapping parts of the 1998 and 2019 DEMs. Instead, changes in bed elevation were evaluated using the following five-step process:

To analyze dune heights, we used the Bedform Tracking Tool (Van der Mark and Blom, 2007) to determine the mean dune height at the time of the 2019 survey. This height was used to define the error bars around the mean bed-elevation change. Thus, to be considered significant, the change in bed elevation between 1998 and 2019 would have to exceed the error bars associated with mean dune height.

Lastly, the texture of the 2019 DEM was used to describe general channel-bed characteristics. Migrating dunes appear as quasi-linear features extending across the channel. Areas of the bed consisting of sand or fine-gravel bars (that is, <128 mm) without dunes appear smooth, and areas of the bed composed of large gravel (that is, >128 mm) appear rough (size-class divisions based on Wentworth, 1922 and Pettijohn, 1975). Thus, based on image resolution, it can be difficult to determine the difference between plane-bedded sand and gravel that is less than 128 mm, but can be relatively more straightforward to determine the difference between dune fields and parts of the bed composed of gravel coarser than 128 mm because the grain size begins to approach the size of the DEM grid cell (0.25m).

Longitudinal Trends in Bed-Sediment Grain Size

During our 2019 resurvey, bed-sediment samples were collected in most of the surveyed cross sections. At most cross sections, a minimum of three bed-sediment samples were collected using a US BMH–60 sampler at the channel center and at one-quarter and three-quarters of the distance across the width of the wetted channel (measured from the left to the right bank, as viewed facing downstream). Samples were collected at cross sections 32–34, 49, 50, and 52–56 (figs. 3, 4). In cross section 32, only one sample was collected, at one-quarter the distance from the left to right bank, and no samples were retrieved from the middle or three-quarters of the distance from the left to right bank. This is likely because the bed was composed of gravel that was too coarse for the US BMH–60 to sample. In cross section 51, no samples were retrieved, also likely because of the coarse bed. At cross section 52, there were two channels, and three samples were collected in each channel according to the sampling scheme described above. At cross section 55, there was a main channel and a small secondary channel; three samples were collected in the main channel, and one sample was collected in the center of the secondary channel. Samples were sieved at ¼-ϕ (phi unit of sediment grain size [D], defined as ϕ = −log₂ D) intervals (Krumbein, 1934) to determine the grain-size distributions.

The median grain size (D₅₀) of the total sediment sample and of the sand fraction was measured, as well as a further subdivision of sand into percent of very fine sand (PVFS), to determine if there were different grain-size characteristics between Echo Park and Island and Rainbow Parks. The PVFS was analyzed because changes in bed-sand grain size most relevant for sand transport occur mainly through changes in the fine tail of the bed-sand grain-size distribution (Topping and others, 2018). Two-sided t-tests were conducted on the populations of total-sediment D₅₀, sand D₅₀, and PVFS to determine if the means of these distributions were significantly different from each other at the 95-percent confidence level. Prior to carrying out the t-tests, F-tests were used to determine whether to conduct the t-tests assuming equal or unequal variances.

We also analyzed the bed-sediment data to determine if there were longitudinal trends in grain size downstream through each river segment. We thus regressed the mean values of the total-sediment D₅₀, sand D₅₀, and PVFS among the samples collected across each cross section against the rkm downstream from Flaming Gorge Dam. F-tests were conducted to determine whether the slopes of the best-fit linear regression lines were statistically significant at the 95-percent confidence level.

Aerial-Photograph Analyses

Aerial-photograph analyses indicate that channel narrowing has continued within Echo, Island, and Rainbow Parks since 1993. Between 1993 and 2018–2019, the channel has narrowed on average by 4 percent in Echo Park, 16 percent in Island Park, and 15 percent in Rainbow Park (table 2; fig. 6). Although narrowing was the general pattern in each park’s river segment, there was substantial variability in the magnitude and direction of channel width changes over shorter reaches. For example, although the Green River narrowed by 4 percent on average over the entire length of the Echo Park segment, one individual 1-km-long reach widened by 2 percent, whereas another one 1-km-long reach narrowed by 9 percent (fig. 6). In Island Park, all of the 1-km-long reaches narrowed; the 1-km-long reach at rkm 126 only narrowed by 4 percent, but the 1-km-long reach at rkm 124 narrowed by 35 percent (fig. 6). In Rainbow Park, the 1-km-long reach at rkm 131 widened by 2 percent, whereas the 1-km-long reach at rkm 132 narrowed by 26 percent. These results indicate that the alluvial reaches of Echo, Island, and Rainbow Parks remain dynamic, but with a dominant trend of channel narrowing.

In all river segments, narrowing occurred at faster rates between the 1993 and 2006 aerial photographs (0.6 m/year in Echo Park, and 1.9 to 2.4 m/year in Island and Rainbow Parks, respectively) compared to post-2006 aerial photographs (0 to 0.5 m/year in all segments) (fig. 6). This is likely because the annual floods from the Yampa River were relatively small between 1993 and 2006, when the hydrologic thresholds defined by Topping and others (2025) were infrequently met; during that period, 66 percent of the years did not exceed the peak discharge threshold of 425 m³/s, and 70 percent of the years did not exceed the median March–July discharge of 99 m³/s. Topping and others (2025) indicate that sand accumulation is likely to occur within the Yampa River and (or) the middle Green River when those thresholds are not exceeded. Conversely, between the 2006 and 2018–2019 aerial photographs, those discharge thresholds were exceeded approximately 50 percent of the time, and the rates of channel narrowing were smaller. These results indicate that hydrologic thresholds defined by Topping and others (2025) are important for limiting sediment deposition within the sediment-budget reach between Deerlodge Park and Jensen, Utah.

Changes in Cross-Sectional Area, Active-Channel Width, and Bed Elevation

Comparisons of cross-section measurements show that a variety of styles of geomorphic change have occurred within Echo, Island, and Rainbow Parks, arising from both net depositional and erosional processes. Depositional processes include vertical floodplain accretion (figs. 7, 8; figs. 1.6, 1.7 in appendix 1), lateral accretion (fig. 9; figs. 1.1–1.4 in appendix 1), and bed aggradation (fig. 1.5 in appendix 1). Erosional processes include channel widening (fig. 1.9 in appendix 1), cutbank retreat (fig. 8; fig. 1.9 in appendix 1), and minor bed scour (fig. 1.9 in appendix 1). In multi-threaded reaches, some channel migration has occurred (fig. 8; figs. 1.7, 1.9 in appendix 1). Elsewhere, the cross sections were stable, with very little change in cross-sectional area and no channel migration (figs. 1.6, 1.8 in appendix 1). Cross-section measurements thus illustrate the dynamic channel behavior that continues to occur within Echo, Island, and Rainbow Parks, in agreement with the results of aerial-photograph analyses.

Repeat measurements of channel width among the cross sections also indicate that the channel has predominantly narrowed between 1993–1994 and 2019 (fig. 10A; table 3). The changes in channel width calculated from the repeat cross-section surveys are significantly correlated with changes in channel width measured on the aerial photographs at the cross-section locations (level of significance [p] = 0.004, as indicated by an F-test conducted on the linear regression). On average, the cross-section measurements show that the channel has narrowed by approximately 9 percent since the earliest measurements were made in 1993–1994, compared to narrowing by approximately 15 percent on the aerial photographs at the cross-section locations from 1993 through 2018–2019. The narrowing measured on the cross sections is less than the narrowing inferred from the aerial photographs because on the photographs, we delineated where narrowing had been caused by vegetation encroachment, even if there has been little or no associated topographic change to the channel itself. Note that cross section 55 (fig. 1.9 in appendix 1) is noted by an asterisk because cutbank erosion has resulted in channel widening that is visible in the surveyed cross sections; however, a vegetated mid-channel island had developed as of 2019, and was visible in the aerial photographs, indicating channel narrowing.

Between the 1993–1994 surveys and the 2019 surveys, the average direction of channel-width change at the cross sections in Echo and Island Parks agreed with the direction of the average channel-width change measured on the aerial photographs (table 4). In Rainbow Park, however, the average channel width at the cross sections widened by approximately 2 percent, whereas the aerial photographs showed an average narrowing by approximately 15 percent. This disparity is because only two cross sections were used to calculate the average change in channel width of the cross sections, and one cross section had substantial cutbank erosion (cross section 55, fig 1.9 in appendix 1).

In 9 of the 15 cross sections, the narrowing measured on the aerial photographs is associated with a loss in cross-sectional area (fig. 10B; table 4). This indicates that the deposition associated with channel narrowing and (or) floodplain aggradation has generally outpaced erosional processes, such as bed scour or erosion of low-elevation benches. Net deposition of sediment and associated declines in cross-sectional area occurred in 10 of the 15 cross sections between the initial 1993–1994 surveys and our 2019 survey (figs. 10B, 11); the average decline in cross-sectional area was 6.3 percent for those 10 cross sections, with a standard deviation of 3.8 percent. Only 5 of the 15 cross sections showed sediment erosion and increases in cross-sectional area: 0.8 percent for the left channel of cross section 51 (fig. 7), 2.2 percent for the right channel of cross section 52 (fig. 1.7 in appendix 1), 15.0 percent for cross section 54 (fig. 1.8 in appendix 1), 8.4 percent for cross section 55 (fig. 1.9 in appendix 1), and 44.9 percent for cross section 56 (fig. 1.10 in appendix 1). The increase in area of the left channel of cross section 51 was negligible. The increases in area of the right channel of cross sections 52 and 54 were a result of bed scour during the 2019 annual flood. The increase in area of cross section 55 was caused by cutbank retreat. The large increase in cross-sectional area from cross section 56 includes substantial uncertainty, because the right-bank endpoint was not found; thus, the exact location of the original cross section could only be estimated during the 2019 survey, and the result cannot be verified.

Analyses of cross sections and channel bathymetry data in the study area suggest that no major changes in bed elevation occurred between the 1990s and 2019. At the cross-section locations, the minimum bed elevation in the 2019 channel was slightly lower on average than in the most recent measurements from the 1990s, but the variability between cross sections greatly exceeded the mean change (fig. 11). The mean change in minimum bed elevation between the first cross-section measurements in 1993–1994 and the 2019 measurements was −0.30 m (standard deviation of 0.76 m), and the mean change between the last cross-section measurements from the 1990s and the 2019 measurements was −0.08 m (standard deviation of 0.91 m). Similarly, analyses of bed-elevation change between the 1998 ADCP measurements and the DEM, created from the 2019 multibeam-sonar survey, showed that the channel bed was approximately 0.30 m lower on average in 2019 than in 1998 (fig. 12). However, the spatial distribution of bed-elevation change between the 1998 ADCP data and the 2019 multibeam-sonar data substantially overlaps zero (fig. 12). Moreover, the uncertainty in bed elevation associated with the mean dune height in 2019 is similar in magnitude to the absolute value of the approximately 0.30 m change in bed elevation between 1998 and 2019. Thus, although the data indicate that slight mean scour likely occurred between the late 1990s and 2019, this scour was much smaller than the spatial variability in bed-elevation change, indicating insignificance in the context of this study.

Longitudinal profiles of bed-elevation change, calculated from the 1998 and 2019 surveys, further illustrate the large spatial variability in bed-elevation change that occurred within Island Park (fig. 13). Upstream from rkm 128.5, where the bed alternates between a plane bed and sand dunes, the variability in bed-elevation change is rather small (fig. 13A); there was as much as 2.5 m of scour within the sandy secondary channel at approximately rkm 127.8 (downstream from cross section 51), but changes elsewhere consisted of smaller amounts of scour and fill that were generally less than 0.75 m (fig. 13A). Downstream from rkm 128.5, where the bed is largely sand, changes in bed elevation were much more variable, with as much as 4.9 m of fill and 3.8 m of scour in some areas. Some of these larger changes occurred in nearly the same longitudinal location, indicating that large amounts of local scour were offset by large amounts of local deposition (for example, at rkm 130.4, fig. 13A).

Between approximately rkm 129 and 130.5, longitudinal changes along the channel centerlines of both the left and right channels show different patterns of scour and fill. In the left channel, there are short reaches, approximately half a kilometer in length, of alternating scour and fill along the centerline (fig. 13B), but the moving average shows that the bed-elevation change is approximately 0 at the start of Rainbow Park at rkm 131. In the right channel, much of the bed along the channel centerline in 2019 was scoured compared to the 1998 data. However, the scour does not persist into Rainbow Park, and the bed-elevation change downstream from rkm 130.7, as shown by the moving average, is approximately 0 (fig. 13C). The longitudinal profiles in figure 13 thus show both the longitudinal and cross-stream variability in scour and fill that occurred between 1998 and 2019.

The apparent bed erosion between the 1990s and 2019 could be an artifact of the timing of the surveys (figs. 11–13). Throughout the Green River system, from Flaming Gorge Dam to the confluence with the Colorado River in Canyonlands National Park, it is common for the bed to scour and fill during spring snowmelt floods, although these patterns are highly spatially variable (Allred and Schmidt, 1999; Grams and Schmidt, 2005; Alexander, 2007; Dean and others, 2020; Grams and others, 2020; Walker and others, 2020). Because of how the flow interacts with nonuniform planform geometry and bed topography in any given reach, some cross sections fill whereas others scour during the flood rise and recession (Colby, 1964; Andrews, 1979; Howard and Dolan, 1981; Topping and others, 2000). Thus, it is difficult to detect long-term changes in bed elevation because of the natural processes of scour and fill that occur every year, especially because the local magnitudes of scour and fill can greatly exceed the possible 0.2–0.3 m of long-term scour detected in our study. At the locations of the cross sections in Island Park, cross-section surveys show that the channel bed often scours on the rising limb of the snowmelt flood, at times as much as 2 m (fig. 9), and subsequently fills during the falling limb of the flood (fig. 11; figs. 1.7, 1.8 in appendix 1). Survey frequency in Echo and Rainbow Parks was not sufficient to analyze scour and fill dynamics during individual floods in those segments. Our 2019 survey data was collected during the rising limb of the snowmelt flood, when the channel bed usually scours at the Island Park cross sections. Conversely, the 1998 ADCP data was collected at a lower discharge during the falling limb, when the channel bed usually fills at the Island Park cross sections. Thus, the apparent erosion of the channel bed between the 1998 and 2019 surveys (figs. 12, 13C) cannot be differentiated beyond the scour and fill processes that occur annually. The timing of scour and fill also likely influenced the increase in cross-sectional area for the right channel of cross sections 52 and 54 (fig. 10B); the initial surveys were conducted in September–October of 1994, whereas the 2019 survey was conducted near the peak of the annual flood, when the bed is usually scoured at those cross sections.

Bed-Sediment Grain Size

Within Echo Park, the channel bed consisted of relatively smooth sand and gravel, and of dunes composed of sand. Bed-sediment samples at cross sections 33 and 34 indicate the bed at those locations was predominantly gravel (fig. 14A). Multibeam-sonar data show that in the upstream part of Island Park, between cross sections 48a (rkm 123.8) and 49 (rkm 125.5), the bed was primarily composed of gravel (fig. 15A). Between cross section 49 (rkm 125.5) and approximately rkm 127, the bed surface transitioned from gravel to sand dunes. Between river km 127 and cross section 52 (rkm 128.7), the bed was likely composed of a mixture of gravel and sand dunes; however, in some patches of the channel, no multibeam-sonar data were collected. Downstream from cross section 52, nearly the entire bed was composed of migrating sand dunes (fig. 15B). Isolated patches of gravel were present among the dune fields, such as near the left bank at cross section 54 (rkm 130.9; fig. 15B). No bed-sediment sample was collected along the left part of the channel at that cross section because that part of the channel bed was likely composed of gravel too coarse to be retrieved by the sampler. However, some gravel was sampled in the center of the channel at cross section 55 (rkm 132.1) in Rainbow Park (fig. 14A).

Although bed-sediment samples indicate the bed in Echo Park was primarily gravel, and the bed in Island Park was primarily sand, t-tests indicate that there are no statistically significant differences in the sampled bed sediment between Echo Park and Island and Rainbow Parks. This result arises from the small sample size (n=7) and the large variance in grain size among samples in Echo Park. The p-values associated with the t-tests were 0.06 for total-sediment D₅₀, 0.40 for sand D₅₀, and 0.84 for PVFS; thus, the differences were not statistically significant at the 95-percent confidence level. Linear regression showed that there were no longitudinal trends in bed-sediment grain size in either Echo Park or Island and Rainbow Parks (fig. 14); p-values were all greater than 0.05, indicating that the slopes of the linear regressions were not statistically significant at the 95-percent confidence level. Although there were no longitudinal trends in bed-sediment grain size among the bed-sediment samples in Island Park, there is a distinct transition from gravel to sand downstream from cross section 49 (rkm 125.5); however, no samples were collected at cross sections 48a–d because the US BMH–60 sampler could not retrieve the coarse gravel from the bed.