Geomorphology 350 (2020) 106920
Contents lists available at ScienceDirect
Geomorphology
jou rnal homepage: www.elsevier.com/locate/geomorph
Riverscape mapping and hyperscale analysis of the sediment links concept
∗
Aaron Zettler-Mann , Mark Fonstad
University of Oregon, United States
a r t i c l e i n f o a b s t r a c t
Article history: Longitudinal patterns in channel complexity play an important role in our understanding of fluvial sys-
Received 19 August 2019
tems. The spatial scales over which external controls such as tributaries, landslides and debris flows,
Received in revised form 23 October 2019
alter channel form and influence the broader basin-wide trends in channel form. Channel complexity
Accepted 23 October 2019
across a basin has important implications for how we conceptualize basin-wide trends in channel form
Available online 13 November 2019
such as downstream hydraulic geometry and the sediment links concept. This study takes a hyperscale
approach to mapping channel morphology across 200 km of the Rogue River in Southwest Oregon - that
Keywords:
is high-resolution data over large spatial extents allowing analysis to be conducted simultaneously at and
Fluvial geomorphology
Structure-from-motion across multiple spatial scales. Using a sUAS and digital photosieving we measured gravels on exposed
gravel bars for the study area. We also measured thalweg depth in the field using a singlebeam echo
Sediment links
sounder. Slope and channel width were derived using existing GIS datasets. We used these data to create
a hyperscale view of the Rogue River examining the impact of tributary and non-tributary sediment links.
We used a statistical approach to determine if changes in channel form were truly unique in the context
of larger-scale channel variability. Our findings show inconsistent statistical significance of the identified
lateral sediment sources (LSS) between variables - e.g. a statistically significant change in slope would
not necessarily correlate with an increase in unit stream power. There were many LSS which triggered
statistically significant changes to channel form. We used hyperscale graphs to reveal the complexity
and spatial trends associated with the relationships between a given sediment link and channel form.
The results of this study illustrate the complexity with which sediment links impact channel form and
the distances below a given sediment link that a detectable signal exists, if a detectable signal exists.
© 2019 Elsevier B.V. All rights reserved.
1. Introduction scape evolution models also rely on smoothly varying trends. This
approach can have high accuracy for very generalized basin-wide
Fluvial geomorphology has a long history of using broad con- descriptions of channel form in fine-grained alluvial valleys where
ceptual frameworks to characterize watershed scale trends such autogenic processes dominate and external controls are minimal.
as downstream hydraulic geometry (Leopold and Maddock, 1953), However, these generalized trends do not accurately predict the
the river continuum concept (Vannote et al., 1980), channel-reach spatial heterogeneity seen across the diversity of channel forms in
morphology (Montgomery and Buffington, 1997) and the sediment existence at the resolution we can now quantify (Rice et al., 2001;
links concept (Rice, 1998). As described by Fonstad and Marcus Fonstad and Marcus, 2010). What is more frequently observed
(2010) and Carbonneau et al. (2012), increasing data resolution con- from measurements of channel width, depth, and particle size dis-
tinues to illustrate a far more complex geomorphic and hydrologic tribution is a far noisier signal. Local variability in channel form
landscape. Despite this, gradual, basin-wide trends continue to be dominates at small scales, making the gradual, longitudinal, basin-
referenced as the “expected” pattern against which morphologic wide trends more complicated than watershed scale conceptual
alterations are judged. The Rosgen classification scheme (Rosgen, models predict.
1994), a common basis for guiding fluvial restoration projects, Characterizations such as channel-reach morphology
relies on smoothly varying trends in channel form. Many land- (Montgomery and Buffington, 1997) and the river continuum
concept (Vannote et al., 1980) seek to address the heterogeneous
nature of rivers by identifying geomorphically unique sections
within the broader river. However, within these sections there ∗
Corresponding author.
remains the idealized downstream hydraulic geometry relation-
E-mail address: [email protected] (A. Zettler-Mann).
https://doi.org/10.1016/j.geomorph.2019.106920
0169-555X/© 2019 Elsevier B.V. All rights reserved.
2 A. Zettler-Mann and M. Fonstad / Geomorphology 350 (2020) 106920
ships. The sediment links concept (Rice, 1998; Rice and Church, conditions they still provide useful insight into the process-form
1998) is a variation on downstream hydraulic geometry (Leopold relationship.
and Maddock, 1953) that does not rely on characterizing the chan- This study integrates SfM, digital photosieving techniques,
nel type morphology as a function of its distance from the basin depth measured in the field, and remote sensing imagery to cre-
divide. Instead of a continuous downstream fining from the upper ate a hyperscale data set for the Rogue River, a gravel bed river
to lower watershed, the sediment links concept suggests that the in the Pacific Northwest, USA. We consider all potential sediment
pattern of downstream fining may be periodically interrupted by sources in the study area, recognizing that some LSS may not create
local sources of new material (e.g., an underlying glacial deposit, a quantifiable signal. In the context of the sediment links concept,
landslides, bank failure) or discharge from tributaries referred we examine whether and how LSS and basin geology control gravel
to as lateral sediment sources (LSS). The sediment links concept sizes and channel form by addressing the fundamental research
does not suggest that all LSS will trigger discontinuity in trends question: How do tributary and hillslope sediment contributions
of downstream hydraulic geometry. Instead, it seeks to identify influence channel form and particle size distributions? Specifically,
basic characteristics that may offer predictive capabilities for we evaluate the following hypotheses:
channel discontinuities. There are two processes by which new
material may cause an interruption in the process of downstream •
(H1) Tributaries will result in an increased gravel size at their
fining: (1) the new material deposited into the channel is of a
confluence with the Rogue River.
sufficient size as to disrupt the pattern of downstream fining in the •
(H2) Non-tributary sediment sources will produce larger gravel
mainstem (e.g., it is larger than that in the mainstem). The addition
sizes where they intersect the proximal channel compared with
of new material could be infrequent but deliver large quantities
those gravel bars upstream.
(e.g., landslide) or is delivered with a frequency such that it is a •
(H3) Tributary sediment sources will result in an increase in
continuous source that is constantly delivering new material (e.g.,
channel width and hillslope processes (e.g., debris flows and land-
tributaries or bank failure) or (2) in the case of a tributary, it has a
slides) will trigger a decrease in channel width.
sufficient discharge such that the transport capacity downstream •
(H4) Tributary and non-tributary sediment sources will generate
of the confluence increases with the additional discharge. The
an increase in slope.
increase in transport capacity is sufficient such that clast sizes •
(H5) Channel depth will decrease locally at tributary and non-
that could not be transported upstream of the confluence are
tributary sediment sources.
mobilized, leaving larger clasts, relative to those upstream, in the
channel. Either scenario will result in an increase in gravel sizes at
Our hypotheses come from previous studies and our under-
the confluence and potentially an increase in slope locally (Leopold
standing of the process-form relationship. The hypotheses
et al., 1992; Knighton, 1998).
concerning the size of gravels at LSS (H1 and H2) come from Rice
A primary challenge in reconciling basin-wide conceptual
(1998), with the understanding that not all tributaries will trig-
river models and more quantitative, location-specific observa-
ger morphologic discontinuities. Our prediction for the behavior of
tions is the fundamental disparity in the spatial extent and
channel width (H3) derive from a process-form based approach. An
resolution of the observation. As the spatial scale of obser-
increase in discharge at confluences would trigger increased erosiv-
vation increases, local variability in channel morphology is
ity within the active channel suggesting an increase in width. The
missed or “smoothed out” and generalized longitudinal trends
formation of small alluvial fans in the main stem decreases channel
become more obvious. However, as spatial scales decrease
depth and may also lead to an increase in channel width (Leopold
to the scale of individual gravels and near-continuous width
et al., 1992; Knighton, 1998). In contrast, the addition of large vol-
and depth measurements, local channel heterogeneity makes
umes of immobile colluvium at hillslope derived sediment sources
large-scale trends less and less apparent. A growing body of
would likely decrease channel width. Our final hypotheses regard-
literature establishes the importance of the conceptual frame-
ing slope and depth (H4 and H5) also follow from a process-form
work of riverscapes as holistic systems that exist simultaneously
understanding. The accumulation of alluvial and colluvial material
at scales from microhabitat to watershed (Fausch et al., 2002;
that are not transported downstream will eventually raise the ele-
Thorp et al., 2006, 2010; Carbonneau et al., 2012). Hyper-
vation of the channel bottom, creating a decrease in channel depth
scale analysis requires intensive, high-resolution data over large
(H5) and an increase in slope (H4).
spatial extents so that analysis can be conducted simultane-
How, and at what spatial scales we characterize longitudinal
ously at and across multiple spatial scales (Fonstad and Marcus,
trends in channel form is important for channel classification,
2010).
hydraulic modeling and stream restoration. If too large a spatial
Structure-from-Motion (SfM) and digital photosieving are tools
scale is characterized, important complexity is lost. If the scale is
that allow grain-by-grain measurement for entire gravel bars
too fine, it becomes challenging to relate longitudinal connectivity
(Buscombe et al., 2010; Millidine et al., 2010; Chang and Chung,
to morphologic connectivity. This research seeks to improve our
2012; Langhammer et al., 2017; Woodget et al., 2018). This
understanding across spatial scales of the impact of LSS as controls
approach to data collection allows us to more efficiently charac-
on channel form.
terize longitudinal trends in gravel sizes for all gravel bars in the
study watershed. If one travels via the river, it is possible to sam-
ple gravel bars continuously rather than being limited to those that 2. Study area
have road access, providing the opportunity to create a more com-
plete picture of the variations in sediment and local morphology. To accurately assess the impact of tributary and non-tributary
We can integrate field data with high-resolution remote sensing sediment sources on a fluvial system requires a river with a num-
datasets to create the spatially extensive and high-resolution data ber of tributaries and some knowledge of where hillslope activity
necessary to address questions concerning the longitudinal varia- has impacted the active channel. The Rogue River is a characteris-
tion in width, depth, slope and the spatial pattern of sediment sizes tic gravel bed river in the Pacific Northwest, USA (Fig. 1) and flows
within and between gravel bars. Observing patterns across spatial through alluvial, metamorphic and sedimentary provinces (Fig. 2).
scales is not possible with discrete data sets. Hyperscale data are Frequent gravel bars throughout the study area allowed us to map
critical to examining the conceptual models that have been a foun- longitudinal trends in gravel sizes. The study area is comprised of
dation of fluvial geomorphology for decades to evaluate under what an upper alluvial section with frequent gravel bars that flows over
A. Zettler-Mann and M. Fonstad / Geomorphology 350 (2020) 106920 3
Fig. 1. Rogue River watershed location in Oregon (upper left) and a detailed map of the study area. Gravel bar sample locations, tributaries and Grants Pass are also labeled.
The GIS portion of the analysis is comprised of the darker blue streamline. Field work is approximated by the location of gravel bar samples.
Fig. 2. General geologic units of the Rogue River study area and the locations of gravel bars (circles), tributaries and non-tributary sediment sources (triangles) (Archuleta et al., 2017a).
4 A. Zettler-Mann and M. Fonstad / Geomorphology 350 (2020) 106920
mobile and cemented gravels with occasional bedrock outcrops. in channel form. Combined with gravel size measurements of all
The lower portion is characterized by geologic lateral confinement exposed gravel bars, we can examine the role LSS have on channel
with frequent bedrock outcrops controlling slope. Given the pres- morphology and sediment size distribution across spatial scales.
ence of in-channel bedrock and coarse gravels the majority of the
study area is likely supply-limited, having the stream competence
3.1. Methodological approach
to move more material than is available (Jones et al., 2011). Trib-
utary and non-tributary sediment sources occur throughout the
This study integrated aerial imagery, remote sensing, and field-
study area, allowing us to examine their relationship with chan-
based data. We used 10-m National Elevation Dataset (NED) and
nel form in the context of downstream hydraulic geometry and
1.2-m LiDAR data, aerial imagery from the National Agriculture
changing geologic landscape.
2 Imagery Program (NAIP), soil maps from the U.S. Department of
The watershed is 13,350 km and made up of four geologic
Agriculture (USDA), geologic maps from the Oregon Department of
provinces: The High Cascades (14%), the Western Cascades (16%),
Geology and Mineral industries (DOGAMI) and 100-yr flood inun-
the Coast Range (1%) and the Klamath Mountains (56%) (Jones et al.,
dation maps from the Federal Emergency Management Agency
2011). Fig. 2 is a geologic map of the study area. Much of the broad
(FEMA) (Archuleta et al., 2017a). We used these GIS data to digitize
valley floor the Rogue River flows through is comprised of Qua-
the active channel margin, extract channel slope, estimate the 2-yr
ternary sedimentary deposits and landslides. In many places it is
return interval discharge and compute unit stream power. We col-
broken up by a number of metamorphosed and intruded igneous
lected aerial imagery using a DJI Phantom 3 UAV and a Nikon D5200
plutons, dikes and sills. Upstream of the old Gold Ray Dam site is an
digital SLR camera. Depth data was collected using a Seafloor Sys-
area of lacustrine deposits from the former Gold Ray dam roughly
tems Hydrolite-TM single beam echo sounder paired with a Trimble
two river kilometers (RKM) long. The farther west one travels in
Geo7X handheld data collector.
the watershed, the less deformed and softer the Klamath terrain
becomes, which suggests the potential for the delivery of more
material from LSS. 3.2. GIS processing
In the eastern portion of the study area, the Rogue River val-
ley is predominately alluvial and is bordered by harder volcanic This study requires high resolution, spatially extensive data of
and metamorphosed rocks. Downstream of the confluence with multiple hydrogeomorphic variables to address the research ques-
the Applegate River, the broad alluvial valley begins to narrow and tion. Creating continuous maps of channel width and slope requires
the Rogue River becomes increasingly confined by ophiolite and an approach that can be implemented at a basin-wide scale using
sedimentary lithologies. The spatially varying river valley context all available data. GIS analysis for this study begins at the town of
exerts large-scale controls on channel morphology with impor- Shady Cove and ends downstream of the confluence with the Illi-
tant implications for watershed scale trends. There are eleven nois River. Owing to the spacing of river access sites, field work
perennial streams in the study area, seven of which we consider began at Touvelle State Park near White City and ends at Illahe,
2
large tributaries: the Illinois River (2550 km ), the Applegate River upstream of the confluence with the Illinois River. Mapping of the
2 2 2
(1994 km ), Bear Creek (930 km ), Little Butte Creek (917 km ), active channel, corresponding with the bankfull width, was based
2 2
Evans Creek (580 km ), Grave Creek (422 km ), and Jump-off predominately on NAIP imagery, relying on the presence of active,
2
Joe Creek (282 km ). There are an additional four small tribu- unvegetated gravels and the type and presence of vegetation along
2 2
taries: Reece Creek (55 km ), Mule Creek (77 km ), Snider Creek the channel margin. We supplemented this with breaks in bank
2 2
(60.6 km ), and Galice Creek (61.9 km ). Tributaries entering the slope from LiDAR and soils data, including those characterized by
Rogue River east of the Applegate River (Evans Creek, Bear Creek, the USDA as river wash (Harrelson et al., 1994). We used the FEMA
Snider Creek, and Little Butte Creek) originate in the older, more floodplain maps as a reference point. At no point does the bankfull
heavily metamorphosed portion of the Klamath Mountains sug- channel margin cross the 100-yr floodplain, but in certain con-
gesting less abrasion and therefore larger gravels. fined reaches they follow the same path. Distance downstream and
From field observations and GIS we mapped fourteen non- channel width were derived from the digitized bankfull channel
tributary sediment sources including landslides, active bank margin shapefile, converted to a four meter raster image, using
erosion, and leftover hydraulic mining debris. All non-tributary LSS the MATLAB program ChanGeom V0.3 (Fisher et al., 2012, 2013).
that intersect the active channel were included in this analysis. ChanGeom computes a cumulative distance downstream based on
Channel forming flows are generated from winter frontal systems pixel size centered within the rasterized active channel shapefile.
with base flows sustained by groundwater contributions from the Channel width is computed perpendicular to that center line at each
upper Rogue River basin, supplemented by the only remaining dam, pixel. The output of the ChanGeom program is a channel centerline
the William L. Jess Dam located 40 km upstream of the study area and width measurement at every pixel. We extracted elevation at
(Jones et al., 2011). each centerline point from the NED data that we used to compute
slope using a moving window. It is important that this window
be large enough to capture elevation change in low-slope reaches
3. Methods without being so large as to smooth over small riffles and rapids.
We built a semi-variogram to determine at what distance adja-
Downstream hydraulic geometry (Leopold and Maddock, 1953), cent elevation values were no longer related, termed the range
the river continuum concept (Vannote et al., 1980) and channel of a semi-variogram. This distance is derived from the range of
reach morphology (Montgomery and Buffington, 1997) all suggest the semi-variogram for elevation, where distances beyond 40 m no
certain downstream trends in channel form, drawing large-scale longer see spatial autocorrelation. We therefore used a 40-m win-
trends from a series of discrete data points. To accurately improve dow to compute slope. From our field observations we suspected
our understanding of the heterogeneity in channel morphology that changes in slope, as a result of LSS, may be smoothed over given
requires a near-continuous longitudinal dataset. The goals of our the window size for our slope computation and the subtle change
field work were to collect the data necessary for computing parti- in elevation at some riffles. Therefore, we digitized all named riffles
cle size distributions and channel depth, and to make observations and rapids in the study area based on Leidecker’s (2015) river guide
of unmapped, non-tributary sediment sources. We used a combina- book, which provides an independent source for the identification
tion of remote sensing and field data to plot longitudinal patterns of channel slope breaks.
A. Zettler-Mann and M. Fonstad / Geomorphology 350 (2020) 106920 5
To identify non-tributary sediment sources, we used a combi- Airspace and land management allow for the flight of UAVs from
nation of previously mapped landslides from DOGAMI and field the beginning of the field work portion of the study area at Touvelle
observations. GIS mapped landslides were included only where State Park to the confluence of Grave Creek. For all exposed gravel
they intersected the active channel. We added unmapped land- bars in this reach, aerial imagery was collected using a DJI Phantom
slides and debris flows from field observations where hillslope 3 UAV. From Grave Creek to the end of the study area at Illahe,
scars were present and colluvium was noted within the bankfull the Rogue River flows through a designated Wild and Scenic River
channel. An area with eroding mining debris (RKM 113) and the corridor that prohibits UAV operation. For gravel bars within this
two kilometers of bank erosion (RKM 50) through the lacustrine reach, we walked the bars taking photographs with a Nikon D5200
deposits upstream of the old Gold Ray Dam site also came from DSLR on a telescoping pole.
field observations. All gravel bars were photographed from heights between four
There are five USGS discharge gages on the Rogue River, two and ten meters above ground level, for both photographic plat-
of which are within the study area, two upstream of the study forms, giving a horizontal and vertical ground resolution of 0.4 cm
area, and one just downstream of the study area. We used the log- or better. Camera height for a given gravel bar was chosen based
Pearson Type III method to estimate the 2-yr return interval flood on visual estimations of the gravel sizes present, ensuring that
for all five gages (Bedient and Huber, 2002). We built a second- individual clasts were clearly visible based on the point of view
degree polynomial regression equation where drainage area at each screen. Each gravel bar was processed using Agisoft PhotoScan
gage is used to estimate the 2-yr return interval discharge at that Professional, 2018 (now Metashape), generating a sparse point
2
gage (R = 0.99). For this study, we take the 2-yr flood event to be cloud, dense point cloud and georeferenced orthophotograph for
the channel forming flow – that is, the flow that channel width, each gravel bar following well established best practice (Westoby
depth and gravel sizes are adjusted to (Wolman and Miller, 1960; et al., 2012; Fonstad et al., 2013; James and Robson, 2014;
Knighton, 1998; Stock and Montgomery, 1999). From the ten-meter Carbonneau and Dietrich, 2017). All gravel bars were exported as
NED data we computed a flow accumulation raster that we applied GeoTIFFs with one-centimeter resolution to standardize analytical
the polynomial equation to, generating a computed discharge at scale between gravel bars. For the UAV based imagery we relied
every pixel along the channel based on the size of the contribut- on direct georeferencing the orthophotographs as small ranges of
ing area at that pixel. From the above variables we computed unit uncertainty in absolute gravel bar location were deemed accept-
stream power (Eq. (1)) to provide a physically-based estimate of able in exchange for the more efficient data collection in the
sediment competence (Bagnold, 1966; Phillips, 1989), where field (i.e., not having to survey ground control points at each bar)
represents unit stream power, is the density of water, g is the (Carbonneau and Dietrich, 2017). The gravel bars photographed
gravitational constant, Q is discharge, S is slope and W is channel using camera-on-a-pole had ground control points taken using
width. the Trimble Geo7X so that they could be georeferenced. It took
between 30 and 200 photographs to capture a gravel bar depend-
gQS
ing on the size of the bar and camera height necessary to capture ω = (1)
W gravels.
We analyzed gravel bar orthophotographs using BASEGRAIN
The result of the GIS data processing is a spatial dataset with a
2.2, which is implemented in MATLAB. A number of alternative
point every four meters containing the northing and easting, dis-
approaches are available for digital photosieving including the
tance downstream (km), elevation (m), channel width (m), channel
3 2 Digital Grain Size Project (DGS) (Buscombe, 2013) and SediNet
slope (m/m), Q2 discharge (m /s) and unit stream power (W/m ).
(Buscombe, 2019). Initial study design sought to examine spa-
tial trends in individual gravel sizes within a gravel bar – an
3.3. Field data output available in BASEGRAIN but not DGS. BASEGRAIN works
as a progressive edge detection algorithm that iteratively turns
Field work took place over nine days, in three trips, travel- clasts to white and the shadowed interstatial spaces black, and
ing a total of 200 km of the Rogue River. We traveled during converts the white raster area to individual vector polygons. It
3
dam-controlled summer low flows of approximately 32.5 m /s then computes the A axis, B axis, area, and orientation for each
(95% exceedance) to maximize the number and size of exposed identified gravel, exporting these measurements as tabular data
gravel bars. We stopped at each gravel bar and photographed all (Detert and Weibrecht, 2012; Detert and Weitbrecht, 2013). Small
2
unvegetated portions of the gravel bar that we used to generate gravel bars (<15 m ) were processed as a single image. Large
2
orthophotographs. Every attempt was made to travel in the chan- gravel bars (>15 m ) were broken into a series of adjacent, non-
nel thalweg so that depth data from the echo sounder represented overlapping 15 m by 15 m tiles. This was done to increase the
the deepest point in the channel. However, this was not always processing efficiency of BASEGRAIN. We used the parameterization
possible because of rapids and other obstacles. The echo sounder recommendations in the BASEGRAIN documentation concerning
and Geo7X recorded a point every five seconds comprised of chan- image processing thresholds at each of the five steps (Detert and
nel depth, northing, easting, and elevation. Initial study design Weibrecht, 2012; Detert and Weitbrecht, 2013). Before processing
anticipated using the field collected GPS elevation data for water each image, vegetation and fines were masked. After processing,
surface slope and channel elevation maps, but poor precision in the user can examine the partitioned image and mask, merge or
the Z direction (i.e., with elevation points changing by as many split clasts based on their visual analysis of the image (see Fig. 3).
as 100 m between adjacent measurements) made this impossible. After processing, the 15 m sample data tables were combined so
Depth measurements could not be accurately tied to an elevation that each gravel bar had a single table with all gravel data. From
datum; therefore, depth is a relative depth model (RDM). We can, the data tables containing the size of all identified and measured
however, still use the RDM to address our hypothesis that depth gravels at a single bar, we computed the D16, D50 and D84. As a
will decrease at LSS (H5). The RDM still illustrates the distinction of sensitivity analysis to the user-input portions of BASEGRAIN, two
pool-riffle sequences in the bathymetry even without being able to research assistants each processed 21 of the gravel bars. The gravel
tie the channel bottom to an absolute elevation above sea level. The bar metrics were also compared to hand sampling of gravel bars
lack of elevation datum also means that it impossible to compute conducted by the USGS in 2011 (Jones et al., 2011) on duplicated
water surface elevations, and therefore slope, based on the depth gravel bars. data.
6 A. Zettler-Mann and M. Fonstad / Geomorphology 350 (2020) 106920
Having traveled a number of rivers we have observed that rif-
fles and rapids frequently occur at tributary confluences and places
where hillslope activity interacts with the channel. Riffles and
rapids are a hydraulic indicator of an increase in channel slope and
or roughness (Leopold et al., 1992; Knighton, 1998;). We include
the point location of all named rivers and rapids in the study area
as an alternative indication of a break in channel slope that may
not be quantifiable from our statistical analysis.
4. Results
Our research on the Rogue River reveals a complex sediment
landscape. Commonly observed longitudinal trends like down-
stream hydraulic geometry (Leopold and Maddock, 1953) do not
appear in any of the variables measured (Fig. 5). For the entire study,
the downstream hydraulic geometry relation scaling exponent for
width is below the expected range of 0.3–0.5 for alluvial channels
Fig. 3. Comparison images of a portion of gravel bar ‘MidValley 14 . On the left is
(Gran and Montgomery, 2001). Relative depth and slope both vary
the partitioned image from BaseGrain. Blue lines show the location of the a and b
axes used in the computation. The right image shows the original photograph of the at short spatial scales with little to no lasting downstream impact
same area. due to LSS. Width and unit stream power both have sub-reaches
with higher spatial autocorrelation broken up by sub-reaches with
high spatial variability. Discharge increases smoothly between trib-
utaries, with abrupt jumps in discharge commensurate with the
size of the tributary. However, the increase in discharge does not
appear to correlate with changes to the other geomorphic variables
(Fig. 5). The downstream plot of D84 show some reaches between
LSS exhibit downstream fining (RKM 134 and 113) but these trends
are subtle and are not clear in the plot of D50 (Fig. 5). Many of the
tributary and non-tributary sediment sources we identified pro-
duced a statistically significant geomorphic signal when compared
to the study area median. However, statistical significance was not
always caused by the geomorphic change we hypothesized, such
as a significant high slope value at a landslide caused by a bedrock
outcrop rather than colluvial deposition. In many cases a significant
change in a variable that failed to reject a hypothesis did not indi-
cate that other related variables at that location were significant
(Fig. 6).
We photographed 60 gravel bars in the study area using
the UAV or camera-on-a-pole, each gravel bar requiring 30 to
200 photographs to capture. SfM processing generated usable
orthophotographs for 55 gravel bars. Of the five bars we could
Fig. 4. An example of the sampling method for the statistical analysis looking at not use, one had clast sizes that were too small to detect, two had
the confluence of Galice Creek and the Rogue River. The black rectangle shows the vegetation that prevented gravels from being clearly distinguish-
extent of an 80-meter sample, here sampling active-channel widths. The selection
able, and two suffered from poor photogrammetric alignment and
of width values here will be compared to all width values.
were therefore unreliable. Comparisons to gravels bars hand sam-
pled by the USGS in 2011 (Jones et al., 2011) suggests that the
3.4. Analysis BASEGRAIN analysis is an acceptably reliable method for analyzing
gravel sizes. Based on the single factor ANOVA test, the two sam-
Analysis of the hydrogeomorphic variables width, slope, unit pling approaches (USGS and BASEGRAIN) are statistically derived
stream power and relative depth relied on the non-parametric 1- from the same group as indicated by a large p-value. We computed
tailed Kruskal-Wallis H test. A test area consisted of the data for a p-value of 0.65 in comparing the D16, D50 and D84 computed by
a single variable adjacent to the identified sediment source, plus BASEGRAIN to the values reported by the USGS at the same gravel
those values extending one mean channel width (80 m) down- bars. When we only considered the D50 and D84 in the comparison,
stream of the source. An 80-m window was chosen as the average the p-value increased to 0.98, suggesting that the two size distri-
distance of the range from semi-variogram plots for each of the butions are statistically from the same population. Knowing that
variables except slope (as explained in Section 3.2). See Fig. 4 for a BASEGRAIN is limited in its ability to detect the smallest class sizes,
sampling example. A sample was then tested against all values for based on image resolution, we only considered the D50 and D84
the study area. in our analysis. Depth measurements exist for the 200 km of field-
As an external check of the accuracy of BASEGRAIN, we based study area except for a 20-km section beginning at RKM 104
compared the D16, D50 and D84 values we computed against hand- caused by a sensor error.
sampled USGS results from 2011 using single factor ANOVA. Gravel Not all tributary and non-tributary sediment sources will gener-
bar spatiality was analyzed using Getis-Ord General G to deter- ate a quantifiable change in channel morphology (Rice, 1998; Rice
mine whether gravel bars were randomly distributed within the and Church, 1998). Here we test all potential sediment sources in
study area. Size based clustering of gravel bars was analyzed using the study area for the presence of a geomorphic signal indicat-
Getis-Ord Gi* (Getis and Ord, 1992). ing a discontinuity in downstream hydraulic geometry. We also
A. Zettler-Mann and M. Fonstad / Geomorphology 350 (2020) 106920 7
Fig. 5. The hydrogeomorphic variables discharge, slope, width, unit stream power, depth, D84 and D50 for the Rogue River. Flow direction is from right to left. The red
dashed lines show the downstream hydraulic geometry best-fit equations. Width, depth and each gravel size are computed using a power function. Slope is computed as
an exponential function. Unit stream power is computed using the downstream hydraulic geometry equation prediction from the other variables. Vertical lines show the
location of potential sediment sources. Solid lines are tributary locations and dashed lines show non-tributary sediment sources. Tributaries are labeled right-to-left: (A)
Reese Creek, (B) Little Butte Creek, (C) Snider Creek, (D) Bear Creek, (E) Evans Creek, (F) Applegate River, (G) Jump-off Joe Creek, (H) Galice Creek, (I) Grave Creek, (J) Mule
Creek, (K) Illinois River.
Fig. 6. Conceptual flow chart for relationship between a lateral sediment source and transport capacity relative to time. A geomorphic signal (geomorphic discontinuity) can
be timeless – enduring regardless of how long it has been since the sediment input, or timebound – a transient geomorphic feature that will be removed shortly after the
input.
applied the empirical model (Eq. (2)) derived by Rice (1998) to tion by Rice (1998), which relates the relative drainage areas of the
provide some context for which tributaries are likely to produce tributaries and mainstem, there is a low probability of any trib-
a geomorphic discontinuity. utary producing a geomorphic discontinuity (Table 1). We then
. + . X + . X take a form-process approach in an attempt to better understand
e(8 68 6 08 1 10 04 2)
P = (2) when and why an LSS may or may not produce a quantifiable signal.
(8.68+6.08X +10.04X )
1 + e 1 2
We analyzed the impacts of tributary and non-tributary sediment
−0.51 sources on a variety of hydrogeomorphic variables. The results
In Eq. (2), X1 = log(At/Am) and X2 = log(0.14(At/Am) , where
of that analysis, in the context of our hypotheses, are presented
At refers to the area of the tributary watershed and Am refers to the
below.
watershed area of the mainstem. Based on the probability equa-
8 A. Zettler-Mann and M. Fonstad / Geomorphology 350 (2020) 106920
Table 1
The probability that a given tributary will result in a geomorphic discontinuity based on the empirical equation developed by Rice (1998). The Rogue River watershed area
2
(Am) is 13,350 km .
2
Tributary Name Watershed Area (km ) At/Am X1 X2 Probability
Illinois River 2550 0.19 −0.72 −0.49 0.36
Applegate River 1994 0.15 −0.83 −0.43 0.34
Bear Creek 930 0.07 −1.16 −0.26 0.27
Little Butte Creek 917 0.07 −1.16 −0.26 0.27
Evans Creek 580 0.04 −1.36 −0.16 0.23
− −
Grave Creek 422 0.03 1.50 0.09 0.21
Jump-off Joe Creek 282 0.02 −1.68 0.00 0.18
Reece Creek 55 0.00 −2.39 0.36 0.10
Mule Creek 77 0.01 −2.24 0.29 0.11
Snider Creek 60.6 0.00 −2.34 0.34 0.11
Galice Creek 61.9 0.00 −2.33 0.34 0.11
4.1. (H1) Tributaries will result in an increased gravel size at their are statistically lower than the median slope. Reece Creek, Little
confluence with the Rogue River Butte Creek, Evans Creek, Galice Creek, and Mule Creek are all sig-
nificantly steeper than the median slope. Snider Creek, Jump-off Joe
Only three of the tributaries have a gravel bar located at the Creek, and Mule Creek do not have a slope that is statistically dif-
confluence with the Rogue River and have gravel sizes larger than ferent than the median. In addition to the statistical test we used,
those at the nearest upstream gravel bars; the sign of a significant the presence or absence of named riffles and rapids can be used
LSS according to Rice (1998). Evans Creek, Galice Creek, and Mule as a method for identifying locations with a local increase in slope.
Creek all have D50 and D84 sizes larger than the closest upstream The only creek with a statistically high slope but no named riffle
gravel bar. The Applegate River, Bear Creek, and Grave Creek all at the confluence is Little Butte Creek. The remaining seven tribu-
have gravel bars at the confluence with the Rogue River but gravels taries with statistically significant slopes (high and low) all have a
are smaller than those upstream. Reece Creek, Little Butte Creek, named riffle or rapid at the confluence.
Snider Creek, and Jump-off Joe Creek do not have gravel bars at the Only two non-tributary sediment sources have mainstem slopes
confluence. The Illinois River does have a gravel bar at the conflu- that are not statistically significant, both of which are landslides.
ence but gravel size measurements do not exist for those gravel Four non-tributary sediment sources have a significantly high
bars. Table 2 shows the sizes of each gravel bar in the study area. slope and eight have a significantly low slope, as compared to the
median slope (Table 3). Not all significant high slope, non-tributary
4.2. (H2) Non-tributary sediment sources will have larger gravel sediment sources are necessarily a result of channel-modifying col-
sizes at the exposed gravel bar where they intersect the active luvium (see Section 5.3).
channel
4.5. (H5) Channel depth will decrease locally at all sediment
Most non-tributary sediment sources do not have an associated sources
gravel bar. The active bank erosion at RKM 50, the debris at the
Flanagan Mine at RKM 113, the landslide at RKM 142, and the land- None of the depth measurements samples were statistically sig-
slide complex at RKM 203 all have associated gravel bars. Of those, nificantly different compared to the median depth. However, our
only the Flanagan Mine results in gravel sizes larger than those at more detailed exploration of this dataset in the Discussion sec-
upstream gravel bars. The remaining ten non-tributary sediment tion proposes a relationship between changes in depth and LSS in
sources do not have a gravel bar in close proximity (Table 2). certain locations.
4.3. (H3) Tributary sediment sources will result in an increase in 4.6. Hyperscale analysis
channel width and hillslope processes will trigger a decrease in
channel width Hyperscale graphs (Figs. 8–11) offer new ways to analyze
and interpret fluvial features at watershed to gravel bar extents
All tributaries that exhibited a statistically significant change (Fonstad and Marcus, 2010; Dietrich and Carbonneau, 2019).
in width did so as a predicted increase in width (Table 3). Evans Hyperscale graphs display the correlation between two variables
Creek, Jump-off Joe Creek, and Galice Creek confluences do not have using all possible window sizes to compute each correlation using
mainstem widths that are statistically different than the median for code implemented in Python (Dietrich and Carbonneau, 2019).
the study area. The statistical results of non-tributary LSS are less Here, we use color to show the Pearson’s correlation coefficient
clearly defined. The active cut bank and Flanagan Mine debris both between two variables. Along the x-axis is distance downstream.
show a significantly high width, as do the landslides at RKM 196 The y-axis represents the size of the moving window used to com-
and RKM 211. The remaining five statistically significant landslides pute the correlation coefficient (Dietrich, 2016a). Large window
are all significantly narrower than the median. An additional five sizes should reveal general, basin-scale trends in how the two
landslides do not have a significant impact on channel width. variables change through space relative to each other. At smaller
window sizes the impact of local controls on the channel domi-
4.4. (H4) Tributary and non-tributary sediment sources will nate the pattern. In the context of downstream hydraulic geometry
generate an increase in slope we would expect width to have a generally positive correlation
with discharge as the channel responds to increasing stream com-
Eight of the eleven tributary confluences have a slope that is petence, and slope to have a generally negative correlation as relief
significantly different than the median study area slope with 95% decreases (Leopold and Maddock, 1953).
confidence. However, not all of those eight are caused by slopes that Fig. 8 shows the correlation between channel width and esti-
are steeper than the study area median slope, as hypothesized. Bear mated discharge at the channel forming flow. At window sizes
Creek, Grave Creek, and the Illinois River all have slope values that above 120 km it exhibits the opposite trend as predicted by down-
A. Zettler-Mann and M. Fonstad / Geomorphology 350 (2020) 106920 9
Table 2
Gravel bar name, location and respective size thresholds. The D50 sizes are generally course gravels while D84 falls into the cobble classification. Normalized size classes are
the observation at a given gravel bar divided by the average of all observations for that size threshold.
Gravel Bar Name Location (RKM) D50(mm) D84(mm) NormalizedD50 NormalizedD84
MidValley GB02 46.36 62.40 112.10 1.08 1.13
MidValley GB03 47.73 62.59 113.28 1.09 1.15
MidValley GB04 49.07 59.07 96.31 1.02 0.97
MidValley GB05 49.73 56.21 93.54 0.97 0.95
MidValley GB06 50.49 59.06 101.64 1.02 1.03
MidValley GB07 51.55 59.53 99.47 1.03 1.01
MidValley GB08 65.31 55.85 98.28 0.97 0.99
MidValley GB01 65.31 56.45 99.88 0.98 1.01
MidValley GB09 74.15 62.62 108.00 1.09 1.09
MidValley GB11 77.23 60.02 106.19 1.04 1.07
MidValley GB12 77.41 66.06 116.89 1.15 1.18
MidValley GB13 79.04 58.69 99.26 1.02 1.00
MidValley GB14 82.80 55.01 93.30 0.95 0.94
MidValley GB15 83.81 53.00 91.86 0.92 0.93
MidValley GB16 85.76 57.32 95.91 0.99 0.97
MidValley GB17 85.96 58.69 99.59 1.02 1.01
MidValley GB18 86.38 57.41 97.56 1.00 0.99
MidValley GB19 91.73 58.91 97.82 1.02 0.99
MidValley GB21 95.11 64.33 118.74 1.12 1.20
MidValley GB22 96.49 60.02 105.00 1.04 1.06
MidValley GB23 101.83 53.63 93.72 0.93 0.95
MidValley GB24 103.41 54.62 89.18 0.95 0.90
MidValley GB25 103.78 52.19 66.24 0.90 0.67
RecSec GB01 108.96 52.41 89.45 0.91 0.90
RecSec GB02 111.14 56.56 98.98 0.98 1.00
RecSec GB04 113.96 62.37 111.99 1.08 1.13
RecSec GB05 115.91 59.07 104.25 1.02 1.05
RecSec GB06 116.50 53.94 93.63 0.94 0.95
RecSec GB08 117.73 52.19 71.24 0.90 0.72
RecSec GB09 118.95 52.19 71.24 0.90 0.72
RecSec GB11 124.11 55.30 97.52 0.96 0.99
RecSec GB12 128.41 53.91 103.68 0.93 1.05
RecSec GB13 128.50 57.64 99.54 1.00 1.01
RecSec GB15 130.25 51.05 72.52 0.89 0.73
RecSec GB16 131.29 52.68 82.97 0.91 0.84
RecSec GB17 134.60 60.02 120.50 1.04 1.22
RecSec GB18 135.09 53.91 109.80 0.93 1.11
RecSec GB19 138.48 54.04 97.48 0.94 0.99
RecSec GB20 139.73 58.69 110.18 1.02 1.11
W SR GB01 142.30 53.30 89.54 0.92 0.91
W SR GB02 145.89 53.24 72.58 0.92 0.73
W SR GB03 148.01 52.19 71.26 0.90 0.72
W SR GB04 149.68 57.32 93.68 0.99 0.95
W SR GB05 153.11 54.83 91.65 0.95 0.93
W SR GB06 164.91 67.31 133.73 1.17 1.35
W SR GB07 166.64 0.00 0.00 0.00 0.00
W SR GB09 174.10 56.45 105.41 0.98 1.07
W SR GB11 177.39 52.19 64.70 0.90 0.65
W SR GB12 181.75 55.29 104.42 0.96 1.06
W SR GB13 187.30 54.16 96.03 0.94 0.97
W SR GB18 193.03 85.70 151.35 1.49 1.53
W SR GB19 193.88 65.54 121.61 1.14 1.23
W SR GB20 194.28 61.85 116.26 1.07 1.18
W SR GB21 195.98 55.60 95.75 0.96 0.97
W SR GB22 197.91 59.91 103.61 1.04 1.05
W SR GB24 204.49 57.32 96.58 0.99 0.98
stream hydraulic geometry, a general decrease in channel width nel width. As the Rogue River flows out of the alluvial section,
with increasing discharge expressed as a weak negative correla- channel width decreases downstream (RKM 100) as the river flows
tion between the two. This is a function of the generally narrow into and through the geologically confining Klamath Mountains.
channel as the Rogue River passes through the Klamath Mountains As the Rogue River approaches the coast (RKM 180), the valley
and Coast Range in the lower half of the study area. At window begins to widen moving downstream. Within these window sizes
sizes between 30 km and 120 km changes in valley geology appear. the relationship between channel width and discharge appears to
The upper portion of the watershed displays the trend in correla- be independent of tributary contributions, suggesting that geology
tion we would expect from a confined reach where volcanic rocks is still the primary control.
confine the channel (Fig. 2) and is represented by a weak nega- At window sizes under 10 km (Fig. 9) we would expect to see the
tive correlation shaded green. At around RKM 60 the Rogue River influences of LSS and single-point geologic controls. Where chan-
enters the wider alluvial valley where width generally increases nel form is a function of downstream hydraulic geometry we should
with increasing discharge, displayed in red, indicating a positive also see some evidence at these spatial scales. Tributaries that join
correlation coefficient. Above this point, geologic confinements the Rogue River in the alluvial portion of the study area tend to be
prevent the predicted trend of a downstream increase in chan- associated with a positive correlation between channel width and
10 A. Zettler-Mann and M. Fonstad / Geomorphology 350 (2020) 106920
Table 3
Statistical output table for all of the tributaries and landslides for all hydrogeomorphic variables. We take a p-value < 0.05 to indicate a statistically significant result as
compared to the study area median value.
RKM Name Tributaries - Width Tributaries - USP Tributaries - Slope Tributaries - RDM
f-test p-value f-test p-value f-test p-value f-test p-value
30.1720 Reece Creek 42.4648 0.0000 34.8996 0.0000 10.3433 0.0013
40.8470 Little Butte Creek 32.1138 0.0000 21.6222 0.0000 6.4899 0.0108
46.9198 Snider Creek 63.7348 0.0000 30.9205 0.0000 3.3870 0.0657 −3210112 1.0000
49.8309 Bear Creek 13.1801 0.0003 34.3002 0.0000 31.9361 0.0000 −3204587 1.0000
77.2360 Evans Creek 2.0568 0.1515 35.1559 0.0000 41.3773 0.0000 −3215240 1.0000
103.7830 Applegate River 187.3628 0.0000 12.2517 0.0005 74.6202 0.0000 −3227247 1.0000
122.8440 Jump-off Joe Creek 3.7009 0.0544 2.2563 0.1331 0.1989 0.6556
134.9150 Galice Creek 0.2738 0.6008 15.0866 0.0001 15.1132 0.0001 −3205673 1.0000
148.0730 Grave Creek 30.8438 0.0000 71.8310 0.0000 83.7539 0.0000 −3211041 1.0000
181.4670 Mule Creek 21.2692 0.0000 11.3884 0.0007 1.4064 0.2357 −3203885 1.0000
216.8100 Illinois River 48.5746 0.0000 39.0889 0.0000 95.9994 0.0000
RKM Name Other Sed Sources - Width Other Sed Sources - USP Other Sed Sources - Slope Other Sed Sources - RDM
f-test p-value f-test p-value f-test p-value f-test p-value
50.2730 Bank Erosion 133.0299 0.0000 42.7646 0.0000 54.9947 0.0000 −3248931 1.0000
52.7470 LS1 3.1970 0.0738 21.6456 0.0000 89.6859 0.0000 −3218668 1.0000
58.3200 LS1B 0.7297 0.3930 34.7757 0.0000 264.9154 0.0000 −3317631 1.0000
71.2000 LS2 54.2229 0.0000 69.4422 0.0000 105.4161 0.0000 −3224778 1.0000
82.1730 LS3 15.4510 0.0001 12.5206 0.0004 3.7440 0.0530 −3237431 1.0000
87.8630 LS4 0.0927 0.7607 3.0668 0.0799 2.3042 0.1290 1.0000
113.6360 Mine Tailings 118.4365 0.0000 264.0668 0.0000 239.3876 0.0000 1.0000
121.9380 LS5 0.7923 0.3734 2.8251 0.0928 50.9051 0.0000 −3230671 1.0000
142.1640 LS6 3.1304 0.0768 33.8489 0.0000 31.8670 0.0000 −3211833 1.0000
148.6270 LS7 53.9994 0.0000 59.6556 0.0000 59.8362 0.0000 −3205991 1.0000
153.6470 LS8 52.5275 0.0000 7.2049 0.0073 34.9529 0.0000 −3219850
203.7500 LS9 96.6879 0.0000 0.1484 0.7001 44.3352 0.0000 −3283365
209.5890 LS10 101.7981 0.0000 8.2935 0.0040 172.3356 0.0000
218.4410 LS11 211.6071 0.0000 1.8321 0.1759 117.7772 0.0000
discharge. These include Evans Creek (RKM 77), the Applegate River (such as bedrock grade control) are apparent. This is most notable
(RKM 103) and Jump-off Joe Creek (RKM 122). Of these, only the at locations where landslides interact with the channel such as at
Applegate River confluence was statistically wider than the study RKM 87, 121, 142, 148 and 203. Some, but not all of the tributaries
area. The bank erosion at RKM 50 also shows a positive correlation exhibit an increase in slope while others have the opposite relation-
between width and discharge. Landslides should result in a neg- ship with slope. Evans Creek (RKM 77), the Applegate River (RKM
ative correlation coefficient, driving a decrease in channel width 103) and Galice Creek (RKM 134) all exhibit a positive correlation
regardless of discharge. Many of the landslides occur where there indicating a local increase in slope. Jump-off Joe Creek (RKM 122)
is a statistically insignificant correlation between width and dis- and Mule Creek (RKM 181) exhibit a negative correlation; a local
charge in the hyperscale graph indicated by a gap in the graph. decrease in slope. At the scale of single pixels there is pattern of
Some, such as the landslides at RKM 82, 87 and 203 occur at places alternating positive and negative values. Carbonneau et al. (2012)
with a clear negative correlation coefficient between width and dis- suggest that this is potentially a function of the pool-riffle sequence.
charge indicating a localized narrowing of the channel. In general, The black lines in Fig. 11 show the location of named riffles and
the impact of this narrowing on the correlation does not extended rapids in the study area. Some of these agree with the sequence of
beyond window sizes of 10 km. Additional notable signals in Fig. 9 positive (high slope) and negative (low slope) correlations in the
occur at RKM 5 and 17. The increase in width at RKM 5 is attributable hyperscale graph, but not all. In some locations such as between
to an unvegetated mid-channel bar. At RKM 17 an unvegetated RKM 50 and 110 there is good overall agreement between loca-
point bar in an area with agricultural use suggests the increase in tions with named riffles and rapids and local increases in slope. In
active channel width generating the large, positive values. other areas, between RKM 150 and the end of the study area, there
The graph of slope and discharge (Fig. 10) also exhibits a trend appears to still be a correlation, but it is not strong enough to be
running opposite to that of downstream hydraulic geometry at win- visible in more than the first few rows – the smallest window sizes
dow sizes greater than 130 km. At window sizes between 30 km of 10–100 m.
and 130 km, the middle alluvial portion of the study area shows a
weak positive correlation between slope and discharge. This runs
5. Discussion
counter to downstream hydraulic geometry and the trends seen in
the hyperscale graph of width and discharge. In the upper portion of
The Rogue River presents a hydrogeomorphic landscape with
the study area the mid-sized correlation coefficient window sizes
high spatial variability that runs opposite the conceptual model
show a patchwork of the expected negative correlation between
for changes in geomorphic properties proposed by downstream
slope and discharge. The impact of particularly low slope areas,
hydraulic geometry (Leopold and Maddock, 1953), represents a
such as at RKM 50, influence average slopes at these larger window
sediment landscape with different external controls than the
sizes. In the downstream portion of the study area, average slope
sediment links concept (Rice, 1998), and is less conducive to
decreases as the Rogue River leaves the Coast Range generating the
reach-specific characterization than channel-reach morphology
expected negative correlation between slope and discharge.
(Montgomery and Buffington, 1997). For the entire study, the
At window sizes under 10 km (Fig. 11) the influence of tribu-
downstream hydraulic geometry relation scaling exponent for
tary and non-tributary sediment sources and local geologic control
width is below the expected range of 0.3-0.5 for alluvial channels
A. Zettler-Mann and M. Fonstad / Geomorphology 350 (2020) 106920 11
(Gran and Montgomery, 2001). When dividing the study area into 5.1. Digital photosieving calibration
geologically similar reaches, those that are characterized by alluvial
processes do exhibit scaling exponents in-line with those expected We show that BASEGRAIN does an acceptable job comput-
by downstream hydraulic geometry: specifically, the area upstream ing gravel sizes, as compared to the 2011 study by the USGS
of Bear Creek, and the area beginning just downstream of Grants (Jones et al., 2011), which used traditional hand-sampling tech-
Pass and ending at Jump-off Joe Creek. This suggests that the rea- niques and produces a far more complete data set requiring less
son for the lack of characteristic downstream hydraulic geometry is time in the field as compared to hand counting. Our statistical
those reaches that are geologically confined. These alluvial reaches analysis showed no difference between the D50 and D84 popula-
are also where we see the clearest signals of geomorphically sig- tions generated using the different methods. However, the use of
nificant LSS expressed in gravel sizes. The geologically confined BASEGRAIN does present some challenges. Any given clast must be
sections where the expected patterns of downstream hydraulic comprised of a certain number of pixels for the processing algo-
geometry are not present also have more narrow channels and rithm to positively identify it and compute its size. Additionally,
higher slopes. there must be sufficient contrast between individual clasts and
In the Rogue River, LSS appear to exert a quantitative control on the interstitial spaces for accurate distinction between any two
many of the hydrogeomorphic variables we examined. However, adjacent grains. Poor contrast could have the effect of over esti-
variability within the channel is not entirely explained through the mating gravel sizes because unique particles would be lumped
LSS. Rice (1998) and others (Ferguson et al., 2006) suggest that trib- together. For all but five gravel bars, we were able to control
utary area relative to the main channel is a good predictor for the both potential sources of error with proper study design. Overlap-
presence of a statistically meaningful LSS based on the metric of an ping clasts, such as imbrication, are not possible to control with
increase in gravel sizes. Nevertheless, we find that gravel size does proper study design. Overlapping clasts would cause a system-
not act as a reliable indicator of a sustaining source of sediment, atic decrease in the size of any given gravel. We recognize that
regardless of tributary watershed area. In the Rogue River different this has the potential to create inaccurate data, but our compar-
LSS seem to be highlighted by different hydrogeomorphic variables ison with the USGS study suggests that it was not an issue on
– width in some places, slope in others – and some do not appear to the Rogue River. We believe that for many applications accept-
exert a quantifiable control on channel form. Our findings show that ing a marginal decrease in the level of precision in exchange for
the identification of geomorphically meaningful LSS in the Rogue the clear increase in data resolution and spatial scale is desirable
River must be considered in the context of the multiple hydrogeo- in those studies where environmental factors are appropriate for
morphic variables that can be used to quantify channel form. This implementation of BASEGRAIN or other digital photosieving tech-
suggests that to successfully identify important LSS for the Rogue niques.
River, and likely other similar rivers, measuring a suite of hydrogeo-
morphic variables is important. No single variable was a consistent 5.2. Spatial patterns of gravel bars
indicator of a meaningful LSS. Instead, different hydrogeomorphic
variables exhibited a statistically meaningful change at likely sed- Gravel bars in the Rogue River do not exhibit clustering in their
iment sources inconsistently, suggesting that in the Rogue River spatial distribution throughout the watershed. However, both D50
enduring sediment sources do not create consistent geomorphic and D84 sizes reveal some statistically significant size-based clus-
signals. We also suggest that the historic anthropogenic context tering of large gravel sizes at the gravel bars between RKM 187
is important to understand the presence or absence of a geomor- and 198 (Fig. 5), an area with no identified LSS. Within the context
phic signal associated with an LSS (James, 1999; James and Marcus, of the sediment links concept, we would expect that at important
2006). Human activity has the potential to increase the size and vol- sediment sources there would be a lack of size-based clustering as
ume of sediment contributions to the channel and modify channel the sudden increase in gravel size would be much different than
width such that there is not suitable room for gravel bar devel- the gravels upstream. Immediately downstream of a new LSS some
opment. The historical role of anthropogenic activity is discussed size-based clustering would be evident as the gravel sizes between
further in Section 5.5. In the Rogue River some of the heterogeneity the adjacent bars would likely be similar because any size decrease
in channel form and gravel sizes can be explained in the context of would be a function of the gradual in-channel fining processes.
downstream hydraulic geometry, LSS, anthropogenic activity and However, this expectation was not met. Patterns of downstream
geology, but much of it remains unexplained. fining along the Rogue River are not present at the reach
We found statistical evidence of geomorphic discontinuities scale, nor are they reliably present between adjacent sediment
at certain tributaries and for some geomorphic variables, despite sources.
the low probability of a geomorphic discontinuity as predicted in The D84 and D50 values for the gravel bar at RKM 193 are
Table 1. A number of reasons likely exist for the lack of agree- notably higher than those upstream and downstream. This gravel
ment between the predicted model and statistical analysis found bar immediately follows a short, narrow reach and the large size
here. We view the variable basin geology and the history of anthro- might be related to the sudden drop in unit stream power associ-
pogenic activity as primary reasons why the predictive model did ated with an increase in width. However, this direct relationship is
not perform better. The original paper by Rice (1998) explores not present at other locations in the study area. At RKM 164 the D50
two study areas with relatively low anthropogenic disturbance and and D84 values also appear anomalously large. This bar consists of
similar geology. As such, they find watershed area to be the domi- predominately sands with interspersed cobbles. It is possible that
nate predictor for geomorphic discontinuity, where watershed area the lack of measurable grains across the smaller size thresholds has
serves as a proxy for discharge, stream power, and therefore the resulted in a grain size distribution that is positively skewed, there-
ability of a tributary to deliver sediment. The mixed geology of our fore giving anomalously large D50 and D84 sizes. In the Rogue River,
study area and history of mining suggest that watershed area is gravel sizes appear to be more a function of local conditions such
likely not the best predictor for the presence of a geomorphic dis- as sediment supply, geologic composition, and the local channel
continuity. The discrepancy at Galice Creek highlights this, with the form as it relates to shear stresses and transport rates. The down-
lowest probability of a geomorphic discontinuity and one of the stream patterns of gravel sizes in the Rogue River are not generally
most statistically significant geomorphic discontinuities in slope of explained by continuous downstream fining, discontinuous LSS, or
the study area. some combination of these principles (Fig. 5).
12 A. Zettler-Mann and M. Fonstad / Geomorphology 350 (2020) 106920
5.3. Sediment links concept noticed that in-channel morphology does not always align with
the sediment links hypothesis. The landslide at RKM 58.32 has a
In his original paper Rice (1998) describes tributaries as con- named hydraulic feature, a statistically steeper slope, and higher
tinuously depositing new material in the mainstem resulting in unit stream power. However, at this location the cause of the rapid
a persistent geomorphic signal from the LSS. These systems must is bedrock, not colluvium from field observations. Similarly, field
therefore be transport limited, with more material of a large enough observations suggest that the landslide at RKM 148.63 is caused by
size that not all of it can be mobilized during channel forming flow a combination of bedrock and hillslope deposits. This study cannot
events. Conversely, landslides are interpreted as discrete events positively attribute a given gravel bar’s presence to the adjacent
with the geomorphic signal diminishing over time. In supply lim- hillslope. Some landslides, such as the one at RKM 148.63 and RKM
ited systems such as the Rogue River, a geomorphic signal caused 142.16, have large angular midstream boulders with no evidence
by deposition at tributary confluences is less likely to occur because of alluvial transportation supporting the observation of a hillslope
a flow event in a tributary of sufficient magnitude to deposit mate- process LSS. Other bars adjacent to landslides do not.
rial into the Rogue River is likely to occur when discharge in the Slope and the presence or absence of riffles and rapids should be
Rogue River is sufficient to immediately transport the material linked, as riffles and rapids are defined, in part, by a local increase
downstream. This suggests that tributaries generate temporally in slope. All of the LSS that have a named riffle have a significantly
transient, timebound geomorphic signals. This is supported by our steep slope except for Little Butte Creek and the landslide at RKM
findings at most of the tributary confluences. The most notable 82.17, neither of which have a significant slope. However, not all
geomorphic evidence of LSS in the Rogue River were at landslide LSS reveal a significantly high slope value as expected. The incon-
deposits where the size of hillslope material greatly exceeded the sistency between on-the-water identified hydraulic features and
transport capacity of the Rogue River. While some finer material the statistical slope analysis is potentially related to how the USGS
associated with landslides and debris flows is transported down- generates its 10-m elevation data and therefore how slopes were
stream, the largest clasts remain in-channel and create many of computed. Elevation data for the study area is based on the 1/3 arc-
the geomorphic signals expected at an LSS, most notably a local second DEM from the USGS. Vertical accuracy is based on the best
increase in slope. The hillslope sediment sources therefore exist as available source data that, for the study area, is usually based on
enduring, timeless geomorphic signals – the geomorphic evidence LiDAR. For a small portion of the study area beginning just down-
enduring through time. Fig. 6 illustrates a conceptual model of the stream of Mule Creek and extending to the end of the field study
relationship between a lateral sediment source, transport capacity, area LiDAR data is unavailable and the elevation data was likely
and time. derived by the USGS using the 1:24,000 contour lines (Archuleta
We enriched our analytical approach to identifying statistically et al., 2017b). The rapids at LSS that either do not have a signifi-
high slopes through the inclusion of named riffles and rapids. This cant slope value compared to the median, or have a significantly
is an alternative method for independently identifying the pres- low value, all have low relief through the riffle and tend to be less
ence of high slopes that can help identify which sediment sources than 40 m long. Thus, the slope computation used in this study
may be triggering an increase in slope; either through high magni- smooths the riffle to a low slope value that is not statistically dif-
tude, low frequency events or as a continuous source of sediment. ferent than the median, despite the fact that they are observable,
This is important given the inconsistent relationship between the named features.
statistical variables in our study and challenges associated with
relying purely on statistical analysis (Ziliak and McCloskey, 2008). 5.4. Relative depth
One would expect that the tributary sediment sources that have a
named feature and a local increase in slope value larger than the The samples for the depth analysis were conducted in the same
median are likely indicators of an enduring sediment source. The manner as the other hydrogeomorphic variables. With depth, these
Applegate River, Evans Creek, and Galice Creek are good examples included the range of depths before, at, and after an LSS. The ana-
of this, with named hydraulic features associated with a statistically lytical approach was such that this range of values was tested for
meaningful increase in channel slope. Despite the channel form its statistical relationship with all observations, but within-sample
evidence for active LSS, the gravel sizes at the Applegate River con- variation was not analyzed. How depth changes at an LSS should
fluence are smaller than the reach average and represent a decrease reveal the presence of a pool-tail crest, suggesting a submerged
in gravel size compared to gravel bars upstream and downstream. delta deposit that may not be revealed based on channel slope
We propose that the lack of a sediment size spike at Bear Creek, or unit stream power. At Mule Creek there is a rapid decrease in
Jump-off Joe Creek, and Grave Creek is owing to a combination of depth at the confluence, indicative of a pool-tail crest (Heitke et al.,
sediment storage occurring upstream of the confluence with the 2010). The Mule Creek signal is important because it does not show
Rogue River and the downstream transport of deposited material up as a significant sediment source in any of the other hydrogeo-
by the Rogue River, thus removing geomorphic evidence of the morphic variables. At the Mule Creek confluence depth decreases
LSS. Aerial imagery from NAIP show extensive portions of the trib- by five meters compared to the pool upstream of the confluence
utary channel as alluvial and unvegetated. We hypothesize that (Fig. 5). The clear alluvial fan at the Rogue River’s edge suggests
these gravels are not entering the Rogue River because of deposi- that the change in depth is an indication of a submerged delta and
tion upstream of their confluence, with infrequent mobilization to a potentially important source of sediment in the formation of the
the Rogue River occurring at flows sufficiently large that deposition downstream gravel bar, which has a D50 of 55.3 mm and a D84 of
does not occur in the main Rogue River. It is also possible that abra- 104.4, both of which are larger than the nearest upstream bar. This
sion prior to entering the Rogue River results in gravels that are of indicates that there are some sediment contributions that are not
a similar enough size that there is no discernable signal. Tributary discernable from our methods.
sediment sources are therefore timebound – the presence of a geo-
morphic signal is transient with respect to time and therefore may 5.5. Historic anthropogenic context
or may not be observable.
All the non-tributary sediment sources except four have a The Galice Creek confluence has statistically significant high unit
named riffle or rapid adjacent to where they intersect the active stream power and high slope in the mainstem despite the small
channel, which is confirmed with the statistical analysis of slope. watershed area, which runs counter to the predictive model pro-
However, when considering non-tributary sediment sources, we posed by the sediment links concept. Hydraulic mining started at
A. Zettler-Mann and M. Fonstad / Geomorphology 350 (2020) 106920 13
Fig. 7. Historical imagery from 1939 (left) and current NAIP imagery (right) showing the shift in the confluence of the Applegate River (entering from the south) and the
Rogue River (flowing right to left). The riffle associated with the confluence of the two rivers is outlined in black.
Fig. 8. Hyperscale correlation coefficients for width and the computed Q2 discharge. Gaps in the data occur where the correlation is not statistically significant (p = 0.01).
the Old Channel Mine on Galice Creek in 1860 with written accounts unit stream power here are likely the legacy of the volume and size
claiming massive amounts of material being washed down daily of material washed into the Rogue River during mining operations.
(“Galice Creek - Oregon Gold Locations, 2019). Hydraulic mining Our analysis suggests that Galice Creek is a potentially important
likely delivered an abundance of material, oversized for the trib- sediment source; however, the history of land use in the area calls
utary, to the Rogue River. The notable rapid, high slope, and high the geomorphic evidence into question. The only section with an
14 A. Zettler-Mann and M. Fonstad / Geomorphology 350 (2020) 106920
Fig. 9. A close-up view of the 10 km and below window size examining the correlation between width and discharge. The study area begins in the upper right, and flows
right to left through each row. Solid lines represent non-tributary sediment sources, dashed lines are tributaries.
apparent trend in downstream fining begins at Galice Creek (RKM slope signal because of human activity moving the channel. We
134.9) with gravel sizes fining consistently before increasing at cannot know what the gravel sizes at the historic confluence of the
the two landslides downstream of Grave Creek (RKM 148.62 and Applegate River were, but at the bar upstream of the modern con-
153.64) (Fig. 5). Given the historical context of Galice Creek, we fluence adjacent to the historic confluence, D50 size is the same,
believe the downstream fining could be a function of the deliv- and the D84 is slightly larger than the D50 and D84 at the modern
ery of especially large material to the Rogue River as a byproduct confluence (Table 2).
of hydraulic mining (James, 1999). Galice Creek and the Flanagan The exaggerated signal at Gravel Creek, the buried alluvial
Mine debris at RKM 113.63 (Fig. 5) have two of the largest D50 and deposit at Mule Creek, and the spatially disjointed signal at the
D84 values of the LSS with an adjacent gravel bar. Galice Creek has Applegate River all suggest that observing physical evidence of
an important history of hydraulic mining that is likely a key source meaningful LSS can be problematic. Channel modification and the
of the size of material at its confluence with the Rogue River and overlying hydraulic conditions can enhance or hide a signal. The
the associated rapid. Likewise, the material from the Flanagan Mine majority of the sediment sources in this study could not be consis-
are also comprised of larger material than would be frequently tently identified across all hydrogeomorphic variables measured.
transported and appear to result in an in-channel increase in gravel This challenge in signal identification is likely closely related to the
size. challenges arising from readily available hyperscale data sets. As
Based on the watershed area and geology of the Applegate River, heterogeneity in a channel appears to increase with data resolution,
we would expect it to produce a clear signal with an increase in what we think of as a geomorphic signal becomes harder to distin-
gravel sizes and likely a riffle or rapid at its confluence with the guish from natural variability, or noise. Frequently, observations of
Rogue River owing to the local increase in slope from the contribu- fluvial systems identify a variety of processes interacting at mul-
tion of new alluvium. The statistically high slope and named riffle tiple spatial scales including sediment sources, sediment breaks,
at the Applegate River ends just upstream of the modern conflu- evidence of legacy events, and an overall patchwork of channel-
ence. Examining aerial imagery from 1939 (georectified using NAIP forming process links (Carbonneau et al., 2012).
imagery), we see that the Applegate River used to meet the Rogue
River just upstream of the riffle in question (Fig. 7). Since the 1939
5.6. Additional geomorphic controls
imagery anthropogenic modification forced the Applegate River
mouth downstream to its current location. This suggests that his-
The Applegate River and Grave Creek are the only large trib-
torically the Applegate River contributed enough material to form
utaries that have a gravel bar at the confluence and D84 and D50
an alluvial deposit sufficient to create a riffle. Unfortunately, we do
values that are smaller than the reach average. Given the findings
not have a good time constraint on when this shift occurred. Unlike
of Rice (1998) and the relative watershed sizes of these tributaries,
at Galice Creek, which has a sediment signal because of human
we would expect both tributaries to trigger an increase in gravel
activity, the Applegate River may be missing the statistical high-
sizes. The sub-basin geology of these tributaries is comprised pre-
A. Zettler-Mann and M. Fonstad / Geomorphology 350 (2020) 106920 15
Fig. 10. Hyperscale correlation coefficients for slope and the computed Q2 discharge. Gaps in the data occur where the correlation is not statistically significant (p = 0.01).
dominately of softer, mixed sedimentary rock. Additionally, they the section of the Rogue River where this trend is clearest is in the
both flow through their own depositional valley before entering alluvial portion downstream of the Applegate River.
the Rogue River. Jump-off Joe Creek is also large enough that we In coarse alluvial rivers sediment tends to disperse down-
would expect an increase in gravel size associated with the sed- stream rather than translate downstream maintaining the form
iment source, but the closest downstream gravel bar is roughly of a wave (Sklar et al., 2009; Venditti et al., 2010). The deposi-
a kilometer away and the gravel sizes there are smaller than the tion attributed to an LSS would increase bed elevation locally with
study area average. Both suggest that the depositional environment the same effect on channel form as discussed at the end of Section
before the confluence, and softer geology in the tributary water- 1. Between the spatially episodic hillslope and continuous tribu-
shed more prone to fining result in a sediment source that is not tary sediment delivery, processes of dispersion and translation of
expressed in gravel size. However, given the similarity between the sediment will diminish the initial signal – potentially render-
tributary lithologies and mainstem lithology we do not believe ing the wave (sediment source) undetectable. In-channel storage
that tributary sub-basin geology is playing a major role in channel where channel width increases or slope decreases, and the poten-
morphology. tial of (de)synchronization at tributaries, could also complicate the
The previous discussion has examined the role of tributary sediment links concept (Gran and Czuba, 2017). This study showed
and non-tributary sediment sources as external controls on chan- that at many of the confluences channel width was high and slope
nel morphology. However, autogenic processes also play a role in was low, relative to the study area, suggesting that local storage of
channel form. We see potential evidence of self-organized pool- sediment may be occurring. In-channel storage may also account
riffle sequences in alluvial reaches. Sediment waves, or pulses, are for the subtle clustering of larger gravel sizes between RKM 187 and
also potentially playing a role in channel form (Sklar et al., 2009; 198. In addition to the dispersion and storage of sediment, the com-
Venditti et al., 2010). While not explicitly tested in this study, these bined effects of multiple sediment contributions may also serve to
processes likely have some explanatory role in the portion of chan- enhance or depress the LSS signal.
nel heterogeneity not previously explained. Depth, particularly
in the alluvial sub-reach downstream of the confluence with the 5.7. Hyperscale graphs
Applegate River, appears to exhibit a semi-regular trend between
relatively deep and shallow reaches with a spacing of 400–1000 m, The hyperscale graphs reveal the importance of the varying spa-
roughly five to seven times the mean channel width (Leopold et al., tial extent of a given LSS relative to larger spatial scale geologic
1992; Knighton, 1998). The hyperscale graph of slope (Fig. 9) shows controls. The hyperscale graphs displaying the full range of corre-
a similar pattern for slope. At the smallest window sizes alternating lation windows (Figs. 8 and 10) illustrate the relationship between
positive and negative correlation values, indicate alternating high larger-scale controls and their respective variables. Changes in the
and low slope, have a similar spatial frequency as depth. Like depth, longitudinal relationship between any two variables appears to be
16 A. Zettler-Mann and M. Fonstad / Geomorphology 350 (2020) 106920
Fig. 11. A close-up view of the 10 km and below window size examining the correlation between slope and discharge. The study area begins in the upper right, and flows
right to left through each row. The black diagonal lines show the location of named riffles and rapid.
related to the broad scale geologic controls, a similar finding and coefficient is computed using a smaller sample size and is therefore
interpretation to that by Dietrich (2016). These broad-scale trends more sensitive to any change in width.
provide useful insight into the spatial scales at which trends such
as downstream hydraulic geometry persist.
5.8. Future work
At smaller scales of correlation (Figs. 9 and 11), the impact of
local controls on channel form are more apparent. In Fig. 11 we
There are currently a number of promising digital photosiev-
would expect the presence of a pool-riffle channel form to be most
ing techniques including BASEGRAIN (used in this study) (Detert
apparent as the window size is small enough to not smooth over
and Weitbrecht, 2012), two by Buscombe (2013, 2019) and one
these channel forms. In some reaches such as between RKM 100
by Carbonneau et al. (2004). As the application of digital photo-
and RKM 160 there does appear to be a strong co-occurrence of
sieving becomes more prevalent it is increasingly important that
named riffles and rapids (black lines) and local increases in slope
direct comparisons between all digital photosieving techniques are
(positive correlations). The distance over which the local increase
made. A series of direct comparisons between analytic approaches
in slope influences broader trends can be estimated by the ver-
will help illustrate any strengths or weaknesses in the different
tical extent (y-axis) of the signal. The majority of the riffles and
analytic approaches to digital photosieving. Understanding the dif-
rapids with a positive correlation only exert a signal that is quan-
ferent strengths and weaknesses of the various techniques, such as
tifiable at the smallest window sizes. We believe that these are
dealing with complicated lighting, vegetation, and wood, is impor-
likely true indicators of the pool-riffle sequence in sections where
tant to ensure high quality data. A series of deliberate tests of these
alluvial processes dominate. High slope areas, such as at RKM 150
systems against the same dataset is therefore needed.
(Figs. 10 and 11), which have a signal persisting beyond the 10 km
Gravels on exposed gravel bars are only mobilized at high flow
window, indicate very high slope riffles and rapids that are more
events and it is therefore possible that they do not accurately repre-
frequently a function of bedrock control. Because these rapids are
sent the subtle variations in gravel size at LSS. Therefore, it would be
of such high slope, their presence generates positive correlation
advantageous to measure gravel sizes for the portion of the channel
coefficients at larger window sizes than those lower slope riffles
at, and adjacent to, the channel thalweg. Continuous measurement
and rapids. This interpretation of the pool-riffle sequence is similar
of gravel sizes would allow their inclusion in hyperscale graphs
to that by Carbonneau et al. (2012). At correlation windows below
to help illustrate changes in gravel size moving downstream, pro-
10 km, the relationship between width and discharge appears dom-
viding a continuous picture of longitudinal changes in gravel sizes.
inated by large, seemingly isolated, changes in width. In some
In shallow, clear water streams this is potentially possible using
areas such as between RKM 65–80, width does not vary down-
underwater cameras, or by using aerial photography and adjust-
stream (Fig. 5) so a relatively abrupt, small increase or decrease in
ing for the refraction of light (Dietrich, 2016b). Direct measures
width (RKM 68 and 72, respectively) would have a disproportion-
of water surface elevations would also prove valuable. This would
ate impact on the hyperscale graph. This is because the correlation
allow slope computations to occur at the same spatial scale as mea-
A. Zettler-Mann and M. Fonstad / Geomorphology 350 (2020) 106920 17
sures of other variables such as depth and width. At locations with Acknowledgments
LiDAR water surface slope is readily available; however, in narrow
river canyons and where LiDAR data is not available alternative We would like to thank research assistance Laura Gravelle,
measures of water surface are important. Additional sensors could James Major and Josh Spector for help collecting and processing
also prove valuable in detecting geomorphic variables that were data. James Dietrich for help with the hyperscale graphing. The
not well characterized given the resources available at the time of three reviewers who provided valuable feedback. And, The National
this study. Improved depth measurements such as the inclusion of Science Foundation Doctoral Dissertation Research Improvement
a side-scan sonar would be better suited to characterizing features grant (1734840) and the University of Oregon Department of Geog-
such as the potentially buried alluvial fan at Mule Creek. raphy for their financial support of this research.
The data collection techniques presented in this paper are rela-
tively straight forward and require very little specialized training.
References
As such, they lend themselves to broad applications in a diver-
sity of fluvial environments. Taking the approach discussed in this
Agisoft PhotoScan Professional (Version 1.4.x) (Software). (2018). Retrieved from
paper and applying to additional field locations has the potential http://www.agisfot.com/downloads/installer.
to help develop a more nuanced understanding of the relationship Archuleta, C.M., Constance, E.W., Arundel, S.T., Lowe, A.J., Mantey, K.S., Phillips, L.A.,
2017a. U.S. Geological Survey, The National Map, 3DEP products and services:
between LSS, geology, and channel form across spatial scales. Our
The National Map [WWW Document]. https://www.doi.org/10.3133/tm11B9.
tests were conducted in supply-limited locations with a history
Archuleta, C.M., Constance, E.W., Lowe, A.J., Mantey, K.S., Phillips, L.A., 2017b. The
of anthropogenic impacts on channel from. Expanding the num- national map seamless digital elevation model specifications. In: U.S. Geological
Survey Techniques and Methods, Book 11.
ber of watersheds, including geologic and anthropogenic context,
Bagnold, R.A., USGS Prof. Pap. 422-J 1966. An Approach to the Sediment Transport
will prove valuable to improving the field methods for this type
Problem From General Physics.
of study and further developing theories examining longitudinal Bedient, P.B., Huber, W.C., 2002. Hydrology and Floodplain Analysis. Prentice-Hall,
Inc, Upper Saddle River.
trends in channel morphology. In addition, this research focused
Buscombe, D., 2013. Transferable wavelet method for grain-size distribution from
on the natural processes controlling channel form. Future research
images of sediment surfaces and thin sections, and other natural granular pat-
will examine the role of on-going anthropogenic impacts. terns. Sedimentology 60, 1709–1732, http://dx.doi.org/10.1111/sed.12049.
Buscombe, D., Rubin, D.M., Warrick, J.A., 2010. A universal approximation of grain
size from images of noncohesive sediment. J. Geophys. Res. Earth Surf., http://
dx.doi.org/10.1029/2009jf001477.
Buscombe, D., In review 2019. SediNet: a configurable deep learning model for mixed
6. Conclusion
qualitative and quantitative optical ganulometry. In: Submitted to Earth Surface
Processes and Landforms.
This study found that the impact of tributary and non-tributary Carbonneau, P.E., Dietrich, J.T., 2017. Cost-effective non-metric photogrammetry
from consumer-grade sUAS: implications for direct georeferencing of structure
sediment sources on channel morphology was highly varied,
from motion photogrammetry. Earth Surf. Process. Landforms 486, 473–486,
depending on sediment source type and hydrogeomorphic variable http://dx.doi.org/10.1002/esp.4012.
being examined. The lack of consistent correlation between any Carbonneau, P.E., Lane, S., Bergeron, N., 2004. Catchment-scale mapping of surface
grain size in gravel bed rivers using airborn digital imagery. Water Resour. Res.
given variable and a change in channel morphology suggests that
40, http://dx.doi.org/10.1029/2003WR002759, W07202.
for the Rogue River, a predictive model of LSS as primary controls of
Carbonneau, P.E., Fonstad, M.A., Marcus, A.W., Dugdale, S.J., 2012. Making river-
channel morphology is not appropriate, although some sediment scapes real. Geomorphology 137, 74–86, http://dx.doi.org/10.1016/j.geomorph.
2010.09.030.
sources are evident. In the low-disturbance field sites used by Rice
Chang, F.J., Chung, C.H., 2012. Estimation of riverbed grain-size distribution using
(1998), gravel size plays a key role in the identification of channel
image-processing techniques. J. Hydrol. 440–441, 102–112, http://dx.doi.org/
forming sediment sources. Local controls on the Rogue River and its 10.1016/j.jhydrol.2012.03.032.
Detert, M., Weibrecht, V., 2013. User guide to gravelometric image analysis by
tributaries, in concert with the history of mining and land use, gen-
basegrain. In: Adv. Sci. Res., ISBN 978-1, 1789–1795.
erate a scenario where sediment sources may be expressed through
Detert, M., Weitbrecht, V., 2012. Automatic object detection to analyze the geometry
some, but not necessarily all, of the hydrogeomorphic variables of gravel grains–a free stand-alone tool. River Flow 2012, 595–600.
used here. Thus, measuring as suite of hydrogeomorphic variables Dietrich, J.T., 2016a. Riverscape mapping with helicopter-based Structure-from-
motion photogrammetry. Geomorphology 252, 144–157, http://dx.doi.org/10.
is crucial to effectively characterizing the impact of LSS on channel
1016/j.geomorph.2015.05.008.
form. Despite gravel size and tributary drainage area not being con-
Dietrich, J.T., 2016b. Bathymetric Structure from Motion: extracting shallow stream
sistent indicators of meaningful LSS, as suggested by Rice (1998), bathymetry from multi-view stereo photogrammetry. Earth Surf. Process. Land-
forms, http://dx.doi.org/10.1002/esp.4060.
we do find several interesting signals such as the role that historic
Dietrich, J.T., Carbonneau, P.E., 2019. Hyperscale Rivers v1.0., http://dx.doi.org/10.
anthropogenic activity appears to have in enhancing LSS signals, 5281/zenodo.3382267.
and submerged alluvial fan-like features as evidence of otherwise Fausch, K., Torgersen, C., Baxter, C., Li, H., 2002. Landscapes to riverscapes: bridg-
ing the gap between research and conservation of stream fishes. Bioscience 52, unapparent LSS.
483–498, http://dx.doi.org/10.1641/0006-3568(2002)052[0483:ltrbtg]2.0.co;2.
As our ability to measure rivers over larger distances and at
Ferguson, R.I., Cudden, J.R., Hoey, T.B., Rice, S.P., 2006. River system discontinuities
higher resolutions improves, this spatially extensive and intensive due to lateral inputs: generic styles and controls. Earth Surf. Process. Landforms
31, 1149–1166, http://dx.doi.org/10.1002/esp.1309.
data can be used to deepen our understanding of channel mor-
Fisher, G.B., Amos, C., Bookhagen, B., Burbank, D., Godard, V., 2012. Channel widths,
phology across spatial scales. The hyperscale approach taken here
landslides, faults, and beyond: the new world order of high-spatial resolution
allows for the simultaneous study of basin-scale patterns and local Google Earth imagery in the study of earth surface processes. Geol. Soc. Am. 492,
1–22, http://dx.doi.org/10.1130/2012.2492(01).
processes. The relationship between localized, autogenic process
Fisher, G.B., Bookhagen, B., Amos, C.B., 2013. Channel planform geometry and slopes
on channel form can be observed in the context of basin-wide
from freely available high-spatial resolution imagery and DEM fusion: Impli-
trends in channel morphology. This offers the potential to deepen cations for channel width scalings, erosion proxies, and fluvial signatures in
our understanding of how a single sediment wave may behave tectonically active landscapes. Geomorphology 194, 46–56, http://dx.doi.org/
10.1016/j.geomorph.2013.04.011.
and influence pool-riffle sequences, and what that means for larger
Fonstad, M.A., Dietrich, J.T., Courville, B.C., Jensen, J.L., Carbonneau, P.E., 2013.
spatial-scale patterns such as downstream fining between LSS.
Topographic structure from motion: a new development in photogrammetric
measurement. Earth Surf. Process. Landforms 38, 421–430, http://dx.doi.org/
10.1002/esp.3366.
Fonstad, M.A., Marcus, A.W., 2010. High resolution, basin extent observations and
implications for understanding river form and process. Earth Surf. Process. Land-
Declaration of Competing Interest
forms 35, 680–698, http://dx.doi.org/10.1002/esp.1969.
Galice Creek - Oregon Gold Locations [WWW Document], n.d. Oregon Gold. http:// None. www.oregongold.net/tag/galice-creek/.
18 A. Zettler-Mann and M. Fonstad / Geomorphology 350 (2020) 106920
Getis, A., Ord, J.K., 1992. The analysis of spatial association by use of distance statistics Rice, S.P., 1998. Which tributaries disrupt downstream fining along gravel-
(Vol 23, Pg 189, 1991). Geogr. Anal. bed rivers? Geomorphology 22, 39–56, http://dx.doi.org/10.1016/S0169-
Gran, K., Montgomery, D.R., 2001. Downstream variations in the width of 555X(97)00052-4.
bedrock channels. Water Resour. Res. 37, 1841–1846, http://dx.doi.org/10.1029/ Rice, S.P., Church, M., 1998. Grain size along two gravel-bed rivers: statistical varia-
2000WR900393. tion, spatial pattern and sedimentary links. Earth Surf. Process. Landforms 23,
Gran, K.B., Czuba, J.A., 2017. Sediment pulse evolution and the role of network struc- 345–363, http://dx.doi.org/10.1002/(SICI)1096-9837(199804)23:4<345:AID-
ture. Geomorphology 277, 17–30, http://dx.doi.org/10.1016/j.geomorph.2015. ESP850>3.0.CO;2-B.
12.015. Rice, S.P., Greenwood, M.T., Joyce, C.B., 2001. Tributaries, sediment sources, and the
Harrelson, C.C., Rawlins, C.L., Potyondy, J.P., 1994. Stream channel reference longitudinal organisation of macroinvertebrate fauna along river systems. Can.
sites: an illustrated Guide to Field technique. In: U.S. Forest Service Rocky J. Fish. Aquat. Sci. 58, 824–840, http://dx.doi.org/10.1139/f01-022.
Mountain Forest and Range Experiment Station General Technical Report RM Rosgen, D.L., 1994. A classification of natural rivers. Catena 22, 169–199, http://dx.
-245. doi.org/10.1016/0341-8162(94)90001-9.
Heitke, J.D., Archer, E.K., Roper, B.B., 2010. Effectiveness Monitoring for Streams and Sklar, L.S., Fadde, J., Venditti, J.G., Nelson, P., Aleksandra Wydzga, M., Cui, Y., Dietrich,
Riparian Areas: Sampling Protocol for Stream and Channel Attributes. W.E., 2009. Translation and dispersion of sediment pulses in flume experiments
James, L.A., 1999. Time and the persistence of alluvium: river engineering, fluvial simulating gravel augmentation below dams. Water Resour. Res., http://dx.doi.
geomorphology, and mining sediment in California. Geomorphology. org/10.1029/2008WR007346.
James, L.A., Marcus, A.W., 2006. The human role in changing fluvial systems: retro- Stock, J.D., Montgomery, D.R., 1999. Geologic constraints on bedrock river incision
spect, inventory and prospect. Geomorphology 79, 152–171, http://dx.doi.org/ using the stream power law. J. Geophys. Res. Solid Earth, http://dx.doi.org/10.
10.1016/j.geomorph.2006.06.017. 1029/98jb02139.
James, M.R., Robson, S., 2014. Mitigating systematic error in topographic models Thorp, J.H., Flotemersch, J.E., Delong, M.D., Casper, A.F., Thoms, M.C., Ballantyne, F.,
derived from UAV and ground-based image networks. Earth Surf. Process. Land- Williams, B.S., O’Neill, B.J., Haase, C.S., 2010. Linking ecosystem services, rehabil-
forms 39, 1413–1420, http://dx.doi.org/10.1002/esp.3609. itation, and river hydrogeomorphology. Bioscience, http://dx.doi.org/10.1525/
Jones, K.L., O’Connor, J.E., Keith, M.K., Mangano, J.F., Wallick, R.J., No. 2011-1280 bio.2010.60.1.11.
2011. Preliminary Assessment of Channel Stability and Bed-Material Transport Thorp, J.H., Thoms, M.C., Delong, M.D., 2006. The riverine ecosystem synthesis: bio-
in the Rogue River Basin, Southwestern Oregon. US Geological Survey. complexity in river networks across space and time. River Res. Appl., http://dx.
Knighton, A.D., 1998. Fluvial Forms and Processes: A New Perspective. Hodder Edu- doi.org/10.1002/rra.901.
cation. Vannote, R.L., Minshall, G.W., Cummins, K.W., Sedell, J.R., Gushing, C.E., 1980. The
Langhammer, J., Lendzioch, T., Mirijovskˇ y,` J., Hartvich, F., 2017. UAV-based optical river continuum concept. Can. J. Fish. Aquat. Sci.
granulometry as tool for detecting changes in structure of flood depositions. Venditti, J.G., Dietrich, W.E., Nelson, P.A., Wydzga, M.A., Fadde, J., Sklar, L.S., 2010.
Remote Sens. (Basel) 9, http://dx.doi.org/10.3390/rs9030240. Effect of sediment pulse grain size on sediment transport rates and bed mobil-
Leidecker, M., 2015. Idaho River publications LLC, Sacramento. In: The Rogue River ity in gravel bed rivers. J. Geophys. Res. Earth Surf., http://dx.doi.org/10.1029/
- A Comprehensive Guide From Prospect to Gold Beach, 2nd ed. 2009JF001418.
Leopold, L.B., Maddock, T.J., 1953. The hydraulic geometry of stream channels and Westoby, M.J., Brasington, J., Glasser, N.F., Hambrey, M.J., Reynolds, J.M., 2012.
some physiographic implications. Geol. Surv. Prof. Pap. 252, 57. “Structure-from-Motion” photogrammetry: a low-cost, effective tool for geo-
Leopold, L.B., Wolman, M.G., Miller, J.P., 1992. Fluvial Processes in Geomorphology. science applications. Geomorphology 179, 300–314, http://dx.doi.org/10.1016/
Dover Publications, http://dx.doi.org/10.1016/0022-1694(65)90101-0. j.geomorph.2012.08.021.
Millidine, K.J., Malcolm, I.A., Gibbins, C.N., 2010. The potential of digital pho- Wolman, M.G., Miller, J.P., 1960. Magnitude and frequency of forces in geomorphic
togrammetry for characterising streambed grain-size distributions in fish processes. J. Geol. 68, 54–74, http://dx.doi.org/10.1086/626637.
habitat studies: a feasibility and limitations report. Mar. Scotl. Sci. Rep., Woodget, A.S., Fyffe, C., Carbonneau, P.E., 2018. From manned to unmanned aircraft:
1–20. adapting airborne particle size mapping methodologies to the characteristics of
Montgomery, D.R., Buffington, J.M., 1997. Channel-reach morphology in moun- sUAS and SfM. Earth Surf. Process. Landforms 43, 857–870, http://dx.doi.org/10.
tain drainage basins. Bull. Geol. Soc. Am., http://dx.doi.org/10.1130/0016- 1002/esp.4285.
7606(1997)109<0596:CRMIMD>2.3.CO;2. Ziliak, S., McCloskey, D.N., 2008. The Cult of Statistical Significance: How the Stan-
Phillips, J.D., 1989. Fluvial sediment storage in wetlands. Water Resour. Bull. 25, dard Error Costs Us Jobs, Justice, and Lives. University of Michigan Press. 867–873.