Riverscape mapping with helicopter-based Structure-from-Motion photogrammetry

GEOMOR-05214; No of Pages 14 Geomorphology xxx (2015) xxx–xxx

Contents lists available at ScienceDirect

Geomorphology journal homepage: www.elsevier.com/locate/geomorph

Riverscape mapping with helicopter-based Structure-from-Motion photogrammetry James T. Dietrich ⁎ William H. Neukom Institute for Computational Science and Department of Geography, Dartmouth College, 6017 Fairchild, Hanover, NH 03755, USA

a r t i c l e

i n f o

Article history: Received 1 October 2014 Received in revised form 1 May 2015 Accepted 8 May 2015 Available online xxxx Keywords: Structure-from-Motion (SfM) Multi-view stereo photogrammetry Fluvial geomorphology Fluvial remote sensing Middle Fork John Day River

a b s t r a c t Recent developments in the remote sensing of fluvial systems have provided researchers with unprecedented views on the complexity of rivers. An aerial perspective is key to mapping and understanding the river at a variety of spatial scales. I employed a helicopter-mounted digital SLR camera and Structure-from-Motion (SfM) photogrammetry to bridge the gap between smaller scale aerial surveys from platforms like small unmanned aerial systems and larger scale commercial aerial photography or airborne LiDAR collections. This low-cost solution produced high spatial resolution aerial photography and digital elevation models for a 32-km segment of the Middle Fork John Day River in east central Oregon. Using these data, I extracted channel morphology data at 3-m intervals downstream and took an inductive approach to evaluating the controls on channel morphology and the human influences on the river using a combination of segment-scale and hyperscale analyses. The SfM process produced 10 cm/pixel orthophotographs and DEMs with submeter horizontal accuracy, but the DEMs suffered from a systematic distortion that resulted from the parallel camera geometry of the flight plan. The riverscape has been affected by human actions such as mining, cattle grazing, and restoration; however, differentiating a human signal from the natural patterns of channel morphology was difficult. The hyperscale analysis provided insight into several interesting downstream patterns in channel morphology that, with further analysis, could provide explanations on the physical controls of channel morphology. Overall, SfM has the potential to be a powerful, low-cost addition to the fluvial remote sensing toolkit. © 2015 Elsevier B.V. All rights reserved.

1. Introduction Remote sensing has proven to be an invaluable tool for collecting spatially continuous data sets of key river morphologic variables (i.e., width, depth, bed sediment size) at high spatial resolutions over entire river basins (Fonstad and Marcus, 2010; Carbonneau et al., 2012). Some studies have focused on continental scale mapping with satellite imagery (Yamazaki et al., 2014) and others have used aerial photography to map grain-size distributions (Dugdale et al., 2010). These measured morphologic variables, along with other derived hydraulic variables, can be used to produce detailed maps of the physical structure of rivers creating a holistic view of the riverscape (Fausch et al., 2002; Carbonneau et al., 2012). Collecting imagery and elevation data for this type of mapping is often an expensive endeavor, especially over large areas. New developments in computer vision-based Structure-from-Motion multi-view stereo photogrammetry (SfM or SfM-MVS) have made the collection of high-quality elevation data and accurate orthophotographs easily obtainable for anyone with a high-quality GPS and a modest digital camera (James and Robson, 2012; Fonstad et al., 2013; Javernick et al., 2014).

⁎ Tel.: +1 512 878 9596. E-mail address: [email protected].

Traditional sampling in rivers has been done either intensively over small areas or extensively over large areas with widely spaced measurements. Both of these methods have the potential to undersample river environments and miss critical elements of the river (Marcus and Fonstad, 2010). By employing remote sensing, researchers are able to collect spatially extensive measurements of primary physical variables (e.g., width, depth, slope, and sediment sizes) and calculate secondary hydraulic variables (e.g., velocity and stream power) (Carbonneau, 2005; Dugdale et al., 2010; Fonstad and Marcus, 2010; Walther et al., 2011; Carbonneau et al., 2012). In this study, I will focus on the physical variables, specifically the active channel width (an approximation of bankfull width), depth, and slope. These are important variables in hydrology and geomorphology because they form the basis for determining discharge, as well as defining hydraulic geometry relationships (Leopold and Maddock, 1953; Millar, 2005; Parker et al., 2007). Width, depth, and slope are also the major degrees of freedom that streams can adjust in response to natural fluctuations in discharge and disturbances (Mackin, 1948; Hey and Thorne, 1986; Knighton, 1998; Whittaker et al., 2007; Pizzuto, 2008). Theoretically, the only limiting factor for the precision of these measurements is the spatial resolution of the imagery. The measurements can be aggregated to provide spatially averaged data at any spatial scale. These multiscale data are challenging classic conceptual models of rivers such as downstream hydraulic geometry (Leopold and Maddock, 1953) or the river continuum concept (Vannote et al., 1980),

http://dx.doi.org/10.1016/j.geomorph.2015.05.008 0169-555X/© 2015 Elsevier B.V. All rights reserved.

Please cite this article as: Dietrich, J.T., Riverscape mapping with helicopter-based Structure-from-Motion photogrammetry, Geomorphology (2015), http://dx.doi.org/10.1016/j.geomorph.2015.05.008

2

J.T. Dietrich / Geomorphology xxx (2015) xxx–xxx

which view rivers as smooth downstream trends in morphology and energy. Alternative conceptualizations embrace the heterogeneity of the river as important to understanding geomorphic and ecological patterns (Frissell et al., 1986; Ward and Stanford, 1995; Montgomery and Buffington, 1998; Fausch et al., 2002; Poole, 2002; Fonstad and Marcus, 2010; Carbonneau et al., 2012). While photogrammetry is not a new tool in the toolkit for remote sensing of rivers (Lane et al., 1994, 2003), the recent introduction of SfM photogrammetry has given researchers new opportunities to collect aerial imagery and to create highly detailed digital elevation models at far lower costs than traditional aerial photo collection methods. Structure-from-Motion has already proven to be useful in rivers at a variety of spatial scales from smaller areas of tens and hundreds of meters (Fonstad et al., 2013; Woodget et al., 2014) to kilometers (Javernick et al., 2014). The original algorithms for SfM were developed in the field of computer vision in the 1970s (Ullman, 1979), and the newest iterations were synthesized in the late 2000s (Snavely et al., 2007; Agarwal et al., 2009). Structure-from-Motion provides an alternative to traditional photogrammetry that can be implemented with an offthe-shelf digital camera and requires a smaller number of ground control points for the final reconstruction (James and Robson, 2012; Westoby et al., 2012; Fonstad et al., 2013; Javernick et al., 2014). Structure-from-Motion uses multiple camera views to increase the accuracy of photogrammetrically derived, three-dimensional points and provides a dense point cloud data set, not unlike aerial or terrestrial LiDAR. By converting the point clouds into vector mesh or raster digital elevation models (DEMs), the input photographs can then be mosaicked and orthorectified to the DEM. This study illustrates the utility of SfM as a low-cost method for producing orthophotographs and DEMs for large sections of rivers and the multiscalar patterns in downstream channel morphology with data extracted from the SfM imagery and DEMs. 2. Study area I performed this research on the Middle Fork of the John Day River (hereafter the Middle Fork), in eastern Oregon, USA. The Middle Fork is one of three main branches of the John Day River, which drains to the Columbia River. The segment of interest is in the upper portion of the basin and covers 32 river kilometers from Bates State Park downstream to 700 m below the USGS stream gage (USGS 14043840) at the confluence with Camp Creek (Fig. 1). To aid in the analysis and

discussion, I divided the river into nine segments based on land use, valley width, and major tributary junctions (Fig. 1), with specific details for each segment outlined in Table 1. The Middle Fork is a gravel-bed river with predominantly alluvial riffle–pool morphology. The stream also has sections of steeper, plane-bed morphology that can interrupt downstream trends and lead to changes in channel morphology that are independent of larger scale controlling variables like drainage area (McDowell, 2001). The main human influences in the Middle Fork valley have been cattle grazing in the floodplain and riparian zones, channel modification to accommodate grazing, water diversion for irrigation, and placer mining in the channel. Grazing along the Middle Fork has resulted in a reduction of riparian vegetation and, in some areas, bank erosion caused by trampling, consistent with previous studies on the impact of grazing on river channels (Marston et al., 1995; Trimble and Mendel, 1995; Magilligan and McDowell, 1997). Channel modification reduces meandering and habitat diversity while increasing velocities and altering sediment transport. Irrigation diversions can decrease instream flows, which can affect sediment transport and aquatic habitat (Richards and Wood, 1977; Angelaki and Harbor, 1995; Ryan, 1997). Placer mining causes a complete reconfiguration of the valley and channel, which affects the entire local river system, but also alters the sediment and water supply to downstream reaches (Kondolf, 1994; Kondolf and Larson, 1995; Graf, 2000). The Middle Fork is home to populations of anadromous fish, Chinook salmon (Oncorhynchus tshawytscha) and steelhead (Oncorhynchus mykiss). Steelhead and bull trout (Salvelinus confluentus) are listed as threatened under the Endangered Species Act, while Chinook in the Middle Fork are listed as a species of concern. The Middle Fork, and the entire John Day River system, have been identified as critical habitat for the anadromous species because there are no barriers to fish passage after the three lower dams on the Columbia River (Bonneville, Dalles, and John Day) (NOAA, 2005; U.S. Fish and Wildlife Service, 2010). The critical habitat designation has led to a significant effort to rehabilitate and restore the Middle Fork to improve fish habitat for all life stages. The restoration activities have included riparian plantings to improve shading of the river, the installation of large wood structures to increase habitat in pools, and channel construction to reverse the effects of channel straightening and dredge mining (Cochran, 2013). Euro-American settlement in the area started in the mid-1800s. The Middle Fork watershed has been subject to a range of human impacts in that time. Beginning in the late 1800s, cattle ranchers used the valleys as

Fig. 1. Map of the upper Middle Fork John Day River, with study area highlighted by the labeled river segments.

Please cite this article as: Dietrich, J.T., Riverscape mapping with helicopter-based Structure-from-Motion photogrammetry, Geomorphology (2015), http://dx.doi.org/10.1016/j.geomorph.2015.05.008

J.T. Dietrich / Geomorphology xxx (2015) xxx–xxx

3

Table 1 River segment details (segments are mapped in Fig. 1). Segment

A B C D E F G H I

Length (km)

Drainage area (km2)

Downstream boundary

Valley width (m) Min

Max

Mean

2.44 4.5 7.19 1.78 3.01 4.79 3.31 4.9 0.38

345 404 462 467 538 579 640 667 829

Vinegar Cr. Caribou Cr. Tin Cup Cr. Butte Cr. Beaver Cr. Big Boulder Cr. Coyote Cr Camp Cr. -

45 107 26 104 257 51 112 66 68

354 495 170 451 573 318 384 695 132

164 305 85 198 447 165 219 437 85

Channel type

Present land ownership

Current land use

Human modifications

Restoration activity

Riffle/pool Riffle/pool Riffle/pool, plane Riffle/pool Riffle/pool Riffle/pool, plane Riffle/pool Riffle/pool Plane bed

State park Private Nat. forest Private Private Private/nat. forest Private Private Nat. forest

Rec Cons Rec / LG Cons Cons Cons / LG Cons Cons / IG IG

CS CS, G None None CS, DM CS, DM CS CS, G None

None Minor—RP, LW None Minor—LW Major—LW, RP, CM Minor—RP, LW Minor—LW Minor—RP, LW None

Current land use: Rec = recreation, Cons = conservation, LG = limited grazing, IG = intense grazing. Human modification: CS = channel straightening, DM = dredge mining, G = grazing. Restoration activity: RP = riparian planting, LW = large wood placement, CM = channel modification.

summer pasture. The primary effects of ranching have been a reduction in riparian vegetation, channel straightening, water diversion for irrigation, and bank trampling by cattle. Logging led to the establishment of a mill at the upper end of the study area, at Bates State Park, and the extension of a railroad spur down the valley. One of the most dramatic impacts on the river was placer gold mining operations that used dredges to overturn the valley floor sediment for gold extraction. Mining in the basin has been ongoing since the 1860s, but the primary dredging operations occured in the late 1930s to 1940s along the middle of the study area, section E, around Granite Boulder Creek. The geology of the area is volcanic bedrock, including Miocene basaltic-andesites of the Strawberry Volcanics at the higher elevations and mixed Eocene basaltic flows, tuffs, and conglomerates of the Clarno Formation at the lower elevations (Ferns and Brooks, 1995; Jett, 1998). The hillslope vegetation is predominantly Ponderosa (Pinus ponderosa) pine forest, while grass meadows cover the valley floodplains. The riparian vegetation consists mainly of grasses and sedges with sparse woody vegetation. Throughout the study area, clumps of in-channel vegetation, primarily torrent sedge (Carex nudata), can form significant obstacles to flow, especially at lower flows. The hydrology of the Middle Fork mirrors many rivers of

the Mountain West with peak flows coinciding with spring snowmelt in late April to early May and lower flows corresponding to the dryer summer months, August and September. At the USGS gage above Camp Creek, the median spring peak flows are ~ 17 m3/s and median summer low flows are ~0.5 m3/s. 3. Methods 3.1. Aerial photography collection I collected aerial photography for this research on 13 August 2013 using a Canon T5i digital SLR flown on a Robinson R44 Raven helicopter, chartered with a pilot through a local commercial aviation company (Fig. 2). The camera was mounted in a nadir position within the helicopter's cargo box using a custom vibration isolation mount to reduce the effect of aircraft vibration on image quality. A Garmin GPSMAP 60CSx handheld GNSS receiver recorded the camera's approximate location during the flight. The use of the GPS was strictly to speed up image processing in the SfM software. By geotagging each photo with an approximate X, Y, and Z locations the software can limit its

Fig. 2. Helicopter aerial photography setup. Canon T5i in the vibration-isolating mount with GPS and intervelometer (upper left). Robinson R44 helicopter at takeoff (upper right). Examples of ground control targets, 2-m tarps (lower left) and roadside painted targets (lower right).

Please cite this article as: Dietrich, J.T., Riverscape mapping with helicopter-based Structure-from-Motion photogrammetry, Geomorphology (2015), http://dx.doi.org/10.1016/j.geomorph.2015.05.008

4

J.T. Dietrich / Geomorphology xxx (2015) xxx–xxx

search for nearby camera locations, reducing the time needed for the image alignment stage of the SfM processing. The spatial resolution of the imagery and image overlap are two key variables in any SfM survey. Higher spatial resolutions allow for higher point cloud precision. Because SfM is a multi-view photogrammetric process, a minimum of 60% overlap is needed to guarantee at least four different perspectives of any area in the imagery (James and Robson, 2012; Fonstad et al., 2013). For this study, I planned for a ground resolution of 5 cm, which required a helicopter flight altitude of 200 m above ground level (AGL). To acquire the required 60% forwardlap in the images, the helicopter flew at 25 knots (12.8 m/s) and an intervelometer triggered the camera shutter to capture images at 4-s intervals. For narrow valley sections, the flight plan consisted of a single line of photographs that followed the river, while the wider valley sections used flight lines that covered the ground area with 60% sidelap. The total flight time, including time to and from the airport, was ~3 h. The total equipment costs for the flight were low compared to commercially contracted aerial photography or airborne LiDAR (Table 2). While the helicopter was the ideal platform for the required aerial photography, there was a lengthy pre-planning period with the pilot about the flight plan. Mounting the camera also needed to be negotiated with the pilot. The R44 did not have an internal camera port, and FAA rules about modifying an aircraft restricted mounting the camera to the exterior of the aircraft. I was fortunate that the pilot was willing to cut a hole in his cargo box to create a camera port. To georeference the SfM results I used a series of 66 ground control targets that were placed throughout the study area. Unlike traditional photogrammetry, SfM does not need control points in each photograph (see Fonstad et al., 2013). The control points have two functions in SfM: first, to provide georeferencing for the entire photogrammetric reconstruction and second, to provide constraints for optimizing photo locations and camera calibration parameters. The points were marked with paint on roadsides (18 points) or with 2-m blue tarp targets (48 points) (Fig. 2). I recorded the coordinates for each point using a Trimble GeoXH GNSS receiver with an external antenna. I performed a differential correction on all of the points to achieve decimeter (b20 cm) average error. This portion of the Middle Fork valley does not have a suitable network of previously surveyed benchmarks or control points, severely limiting the use of an RTK GNSS system. Performing the static occupations necessary to establish a network of accurate control points for an RTK survey was outside the scope of this research. I processed the photos with AgiSoft Photoscan Professional (version 1.0.4). The processing steps included the initial sparse reconstruction, applying ground control to the sparse reconstruction, optimizing the photo locations/sparse reconstruction, dense reconstruction, TIN surfacing, DEM production, and orthophotograph generation. The dense reconstruction step is the most memory intensive step of the processing. By grouping the photos into sets of 200 to 400 photos, it was possible to facilitate processing within the memory limitations of the computer (a Windows 7 workstation with dual, 4-core Celeron processors and 16 GB of RAM). The image groups roughly corresponded to changes in the valley morphology. Because SfM is an emerging technique, it is important to assess the horizontal and the vertical accuracy in the resulting orthophotographs and DEMs. I assessed the horizontal and vertical accuracy of the data set with 56 checkpoints in nonvegetated and short grass areas Table 2 Helicopter flight equipment costs. Equipment

Cost

Canon T5i with battery grip Intervelometer Garmin GPSMAP 60CSx (used) Vibration isolating mount (custom construction) Helicopter Charter (~3 h × $480 per hour) Total

$1300 $50 $200 $40 $1400 $2900

according to the APSRS Positional Accuracy Standards for Digital Geospatial Data (ASPRS, 2014). The horizontal and vertical positions of the checkpoints came from individual points in a 2008 LiDAR point cloud. The horizontal positions of the LiDAR points were compared to the corresponding features on the SfM orthophotographs and the elevations of these points in the SfM DEM were compared to the elevation in the LiDAR data set. The second measure of accuracy was a spatially extensive comparison of the elevation values between the SfM DEM and the 2008 LiDAR DEM that allowed me to test for systematic errors that can occur in SfM-derived data (James and Robson, 2014). While the LiDAR data is several years old, it is the highest resolution comparable data set available; and there have not been any large-scale changes in the study area that would adversely affect an accuracy assessment. 3.2. Riverscape data extraction and analysis Hand digitizing the SfM orthophotographs produced the major geomorphic units of the river, which included the wetted channel, active channel (an approximation of bankfull), bars, islands, and in-channel vegetation. Digitizing was a significant time investment. Because the imagery was only natural color (RGB), without a near-infrared band, automated techniques, like image classification or RivWidth (Pavelsky and Smith, 2008), did not produce reliable results. The basis for the downstream measurements was the active channel's geometric centerline, which was extracted by using a skeletonization algorithm applied to the active channel boundary polygon (Haunert and Sester, 2008). A Savitzky–Golay filter (Legleiter and Kyriakidis, 2006) was applied to the centerline to smooth the line and remove angular joints that did not accurately follow the natural curvature of the channel. By using a Python script in ArcMap (Ferreira, 2014), I created cross sections at 3-m intervals along the centerline. The cross sections were orthogonal to the smoothed centerline, and I chose the 3-m interval so that in the most sinuous sections of the stream there were no overlapping cross section lines and each cross section was a spatially unique sample. Intersecting these cross sections with the digitized channel polygons produced cross sectional width data, which were converted to point features at the intersection of each cross section and the centerline. The point features created a master data set at 3-m increments used for later downstream comparisons. Downstream distances were calculated by transforming the XY coordinates of the sample points to a channel-centered streamwise, normal coordinate system (Legleiter and Kyriakidis, 2006). Channel slope was calculated using the LiDAR elevations and a custom moving window mean filter that removed areas in the longitudinal profile data that resulted in negative (uphill) slopes. The filtering produced smoothed slopes and elevation data for each downstream sample point. To calculate valley width, I intersected valley cross sections with a previously mapped valley floor polygon (Bandow, 2004). The larger valley cross sections were created using the same procedure as the channel cross sections. To avoid intersecting cross sections, the sampling distance was increased to 100 m. Using a linear interpolation between the 100-m samples allowed me to add valley width measurements to each 3-m downstream sample point. The drainage area for each downstream sample point was calculated using the hydrologic tools in ArcGIS and the 1/3 arc-second National Elevation Dataset for the Middle Fork John Day watershed. I calculated the sinuosity index for the active channel in a 500-m window. Several categorical variables were added to the sample points using the SfM orthophotographs as reference: channel units (riffle/pool/glide), land ownership, grazing intensity, presence of restoration activity, affected by dredge mining, and underlying geology. The minimum mapping unit for these variables was one sample point, or 3 m (1.5 m upstream and 1.5 m downstream). I classified channel units visually, guided by my field observations of the different channel segments. Areas of consistent white water were classified as riffles, calm water was classified as pools, and areas of intermittent white water were classified as glides.

Please cite this article as: Dietrich, J.T., Riverscape mapping with helicopter-based Structure-from-Motion photogrammetry, Geomorphology (2015), http://dx.doi.org/10.1016/j.geomorph.2015.05.008

J.T. Dietrich / Geomorphology xxx (2015) xxx–xxx

Extracting stream depth from the spectral information in the SfM orthophotographs (Fonstad and Marcus, 2010; Carbonneau et al., 2012; Legleiter and Fonstad, 2012) was one of the goals of this project. Unfortunately, the radiometric resolution, water turbidity, and image exposure were not conducive to using spectral depth techniques (see Legleiter et al., 2004, 2009; Legleiter and Roberts, 2005, for a discussion on the limitations on spectral depth measurements). In place of actual depths, average depths were estimated using an inverted form of Manning's equation, substituting Jarret's n into Manning's equation (Jarrett, 1984; Marcus et al., 1992) and replacing the hydraulic radius with average depth: Q ¼Av

ð1Þ

where A ¼ w  d; v ¼
d¼

25  S

ð2Þ

0:547

8Q 0:12

2 1 d =3  S =2 −0:16 ; n ¼ 0:32  S0:38  d n

w

Q 2 ¼ 0:508  DA0:82  P 1:36  ð1 þ F Þ−0:27 :

ð3Þ

ð4Þ

In Eqs. (1)–(3), Q is discharge (m3/s), A is the cross-sectional area of the channel (m 2), v is the average velocity (m/s), w is the cross section width (m), d is the average depth (m), n is roughness (Jarret's n in this case), and S is the channel slope. Average depths for each sample point (across the active channel width) were calculated using the 2-year peak discharge (Q2) as an approximation of bankfull flow, calculated with the regional regression equations for the northeast region of eastern Oregon (Eq. (4)). In Eq. (4), DA is the drainage area (sq. miles) at each sample point, P is the basinaveraged mean annual precipitation (inches), and F is the percent forest cover in the basin (Jennings et al., 1994). All of the variables except drainage area and channel slope were obtained from the USGS StreamStats program (U.S. Geological Survey, 2012). By creating boxplots of the grouped categorical variables, I examined relationships between several of the river variables and the geomorphic and land use variables. These boxplots can be used to identify broad relationships across the whole study area. To analyze multiscale downstream patterns in the river, I employed hyperscale graphs (Fonstad and Marcus, 2010; Carbonneau et al., 2012). These pyramidal graphs illustrate the statistical relationship of two river variables across multiple spatial scales using a moving window approach. At the top of each graph, the window size encompasses all sample points from the river (n). The window size of the second row is n − 2, so the row has three data points. The third row has a window size of n − 4, generating five data points. This continues with an ever-shrinking window size to the bottom row (window size = 2) where pairs of sample points are being compared (see Fig. 3 for an illustration of this process). The Yaxis of each graph shows the window sizes from largest at the top to smallest at the bottom. The X-axis provides the relative position of each sample in the downstream direction, upstream on the left. A variety of statistical measures/tests can be incorporated into the hyperscale framework, including the coefficient of determination (R2) for a given regression equation or a variety of correlation measures (e.g., Pearson product–moment, Spearman rank-order, or Kendall's τ). For this analysis, I examined the correlation between pairs of variables using the Pearson correlation coefficient, testing for statistical significance at a 99% confidence interval (p = 0.01). This type of analysis can illustrate patterns not always visible with other types of reach or segment scale analysis.

5

4. Results and interpretation 4.1. Aerial photography During the flight, 1483 photographs were collected, all of which were within the target ground resolution of 5 to 7 cm/pixel. The planned flight lines and camera timing provided sufficient forward-lap and sidelap for an SfM reconstruction with only a few small gaps in the coverage. The seven sections of photographs took ~60 h to process through the Photoscan workflow to the orthophoto and DEM outputs. Of the 60 h, ~20 h required direct interaction and 40 h consisted of unattended processing time. The output data sets had uniform pixel resolutions of 10 cm after accounting for the effects of the mosaicking and orthorectification processes in Photoscan (Fig. 4). The SfM orthophoto resolution had sufficient resolution to identify individual logs in engineered logjams along the river but was too coarse to identify sediment sizes (e.g., Dugdale et al., 2010). The exceptions were a few areas where the helicopter drifted from its flight lines and created gaps in the photo coverage (Fig. 5). Most of these gaps were in the floodplain areas of the valleys, so there was little impact on river data extraction. Two small areas (b10 m each) had additional gaps in the photo coverage. Because these areas were in narrow, constrained canyon sections, where there has not been a significant amount of change, I used the 2009 National Agricultural Imagery Program (NAIP) orthophotos to fill in gaps in the SfM orthophotos. The orthorectification and mosaicking process created artifacts in several locations (Fig. 5), resulting in blurry or noisy sections that were caused by an incomplete and noisy DEM underlying the orthophotos. These areas only slightly affected the visual interpretation of the imagery and in total, accounted for b1% of the photographed river length. The accuracy statistics reported by Photoscan suggested that the georeferencing had root mean squared error (RMSE) values of less than 2 cm for all three axes (X = 1.4 cm, Y = 1.8 cm, and Z = 1.5 cm). However, these statistics only show how well Photoscan is fitting the data to the ground control points, which individually had errors ranging from 10 cm to 1.1 m in the horizontal direction and in elevation. The checkpoint statistics (Table 3) show that the SfM data has submeter RMSE accuracy compared to the 2008 LiDAR. The horizontal and vertical accuracies are both outside the desired accuracy for this resolution imagery and DEM based on the ASPRS standards. The major contributing factor was the accuracy of the GPS measurements of the GCPs. Because of the nature of the georeferencing and optimization steps in Photoscan, the GPS error is propagated directly to the output data sets. When compared to the LiDAR elevations, the GCP elevation errors are closely correlated (r = 0.93) to the SfM elevation errors at the same locations (Fig. 6). Another compounding factor in the accuracy of the SfM elevation data was a systematic error in the DEMs, discovered by comparing the SfM DEMs to the bare earth LiDAR DEM over the entire study area. An example of this systematic error, an alternating pattern of positive and negative differences, is visible along the valley floor Fig. 7). James and Robson (2014) have shown that systematic error in SfM similar to that documented in this study can be the result of the parallel geometry of photographs along flight lines. Parallel geometry (Fig. 8) allows the SfM algorithms to accumulate error from radial lens distortions, affecting the accuracy of the reconstruction. Two other problems affected the vertical accuracy of the DEMs: errors where photo overlap was low and inconsistency in capturing vegetation. Low overlap and incomplete coverage have the effect of creating abrupt edges in the DEMs, producing a stair-step pattern in the DEM (see left side of Fig. 7.). While some researchers have had success mapping vegetation with SfM (Dandois and Ellis, 2013; Mathews and Jensen, 2013), I found that the DEMs did not consistently reflect accurate vegetation heights. In one area along the riparian corridor of a tributary, the DEM contained a dense stand of Alder bushes 1–2 m tall, but adjacent 15–20 m conifers were absent from the DEM.

Please cite this article as: Dietrich, J.T., Riverscape mapping with helicopter-based Structure-from-Motion photogrammetry, Geomorphology (2015), http://dx.doi.org/10.1016/j.geomorph.2015.05.008

6

J.T. Dietrich / Geomorphology xxx (2015) xxx–xxx

Fig. 3. Illustration of the construction of the first three lines of a hyperscale graph for a data set of distance downstream and channel width with 10 values. CC is the correlation coefficient for each subset of the data.

4.2. Riverscape mapping The largely automated process produced a data set with 10,776 sample points along the river that has no obvious outliers or anomalous results (Fig. 9). The elevation profile has one large break at 10 km, where

the river traverses the steepest section of the river through a boulderlined cascade. A smaller break occurs at 18 km in section E, where the stream transitions from its natural floodplain into the dredge-mined portion of the valley. These two steep sections are shown as peaks in the channel slope plot. The plot of channel slope also shows a high

Fig. 4. Sample orthophotograph and DEM located in segment E. The Middle Fork is at the bottom of the photo, and the tributary is the recently reconstructed Granite Boulder Creek. Flow is right to left.

Please cite this article as: Dietrich, J.T., Riverscape mapping with helicopter-based Structure-from-Motion photogrammetry, Geomorphology (2015), http://dx.doi.org/10.1016/j.geomorph.2015.05.008

J.T. Dietrich / Geomorphology xxx (2015) xxx–xxx

7

Fig. 5. Examples of the effects of missing photo coverage (left) and orthorectification artifacts (right).

degree of local variability, likely an expression of riffle-pool sequences. Overall, the active channel width tends to increase downstream with abrupt increases in width at major tributary junctions (increases in drainage area). However, the variability in channel width along the entire length of the river is substantial. The plot of sinuosity has a distinct imprint of low sinuosity where major channel modifications have occurred: the upstream portion of segment B (3 km), downstream end of sections E (18–19 km), and segments G and H. The regional regression equation for Q2 provided a reasonable estimate of 2-year peak discharge (33 m3/s) when compared to the limited record (water year 2011 to present) of the USGS gage at the downstream end of segment G, which has recorded peak discharges ranging from ~25 to 31 m3/s. Estimated depth is also highly variable, which results from the variability in active channel width and slope, but does show some expected patterns such as decreasing depth with increasing width and vice versa. Aggregating the data using the classified variables (Fig. 10) eliminated some of the noise from the plots of the variables over distance downstream (Fig. 9). Most river segments exhibit a long-tailed distribution of active channel widths, but the median values show increased active channel width with downstream distance. The distributions of active channel widths across the three channel units also exhibit long-tailed distributions, and there are small differences in the median values. The median riffle width is 11.91 m, median pool width is 11.11 m, and the median glide width is 10.69 m. The amount of stream restoration has only a small influence on channel width; sections of the river with no restoration have a median width of 11.14 m, sections with minor restoration had the highest median (12.41 m), and sections with major restoration had the lowest median width (10.74 m). Current cattle grazing intensity illustrates that intensely grazed segments have wider active channels (median = 15.50 m) than those with moderate (median = 10.51 m) or no grazing (median = 11.39 m). I also looked at the influence of the underlying geology on active channel width and on valley

Fig. 6. Residual plot of ground control GPS (GCP) and SfM elevation errors versus bare earth LiDAR (n = 66).

width. The weaker rocks of the Clarno Formation (segment C to segment I) give rise to wider channels and valleys. All of the categorical relationships are statistically significant (p = 0.01, using a Kruskal–Wallis test) when comparing each category against all others. The results were significant when the comparisons used the actual active channel widths and when active channel width was normalized by drainage area to account for the downstream increases in width. Hyperscale graphs were used to explore scale-related changes in the relationships between active channel widths and variables such as downstream distance, slope, and valley width. Active channel width versus downstream distance (Fig. 11) is an illustration of downstream hydraulic geometry. At the larger spatial scales, above an 18-km window, width and downstream distance show a moderate positive correlation (r ≈ 0.2–0.6). Below the 18-km window, the pattern of correlations becomes more complex, with positive and negative correlation coefficients, indicating that there may be other factors influencing width. Active channel width versus slope shows small positive correlations (r ≈ 0–0.2) at the large and intermediate scales (10–32 km) (Fig. 12). At the local scale (1 km or less), there are stronger positive (r N 0.5) and negative (r b − 0.5) relationships with width and a large number of statistically insignificant correlations (white areas), suggesting that slope is not influencing width uniformly in the stream. The relationship between active channel width and valley width shows weak correlation (r ≈ −0.3 to 0.2) throughout most scales (Fig. 13). The alternating pattern of positive and negative correlations in the 2–4 km range reflects the alternating variation in valley width at this scale (see Fig. 9). 5. Discussion: Limitations and implications

Table 3 Accuracy statistics by axis for the 56 checkpoints.

Average error (m) Standard deviation (m) Mean absolute error (m) RMSE (m) Radial RMSE (m) Horizontal accuracy at 95% confidence Vertical accuracy at 95% confidence qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Radial RMSE ¼ RMSEx 2 þ RMSEy 2

5.1. Aerial photography

X

Y

Z

0.08 0.43 0.39 0.50 0.62 ±1.08 m ±1.44 m

0.03 0.36 0.30 0.37

−0.08 0.74 0.53 0.73

Horizontal Accuracy at 95 % = 1.7038 × Radial RMSE Vertical Accuracy at 95 % = 1.96 × RMSEz (ASPRS, 2014).

One planned data set was the extraction of bathymetric data, either by direct SfM measurements (Woodget et al., 2014) or spectral depth mapping techniques (Marcus and Fonstad, 2008; Walther et al., 2011; Legleiter and Fonstad, 2012). The SfM elevations in the river did not match depth data measured at several cross sections throughout the study area, making the data set unusable. Instead, I used the cross section data to create color–depth regression curves for several band ratio combinations, but none of these regressions had adequate fits to enable spectral depth mapping. I can attribute the failure of both of these methods partly to the shutter speed, water turbidity, and the

Please cite this article as: Dietrich, J.T., Riverscape mapping with helicopter-based Structure-from-Motion photogrammetry, Geomorphology (2015), http://dx.doi.org/10.1016/j.geomorph.2015.05.008

8

J.T. Dietrich / Geomorphology xxx (2015) xxx–xxx

Fig. 7. Map of SfM vs. Bare Earth LiDAR showing systematic errors in the SfM data from the parallel camera geometry and abrupt edges from incomplete coverage. Flow is right to left.

radiometric resolution of the camera. The shutter speed (1/800 s) was optimized to produce evenly exposed images that balanced the brighter areas (like dry floodplain vegetation) and darker areas (such as conifers, water, and shadows). A slower shutter speed would shift the exposure to better capture the darker areas but increase the risk of overexposing brighter areas and causing motion blur in the photographs. To conserve disk space on the camera for the 2-h flight, the photographs were

Fig. 8. Illustration of parallel versus convergent camera orientations.

captured in JPEG format with 8-bit radiometric resolution (256 shades of gray per band), which limits the camera's ability to capture the true dynamic range of the scene. In principle, to increase the radiometric resolution I could have saved the images as RAW format images, though this could have led to memory issues in collection and processing. On the Canon T5i, the RAW format provides ~14-bit (16,384 shades of gray per band) radiometric resolution, which could have provided more color information over the darker areas on the water and shadowed areas. This extra color information may have provided enough color depth to create accurate spectral depth regressions, assuming ideal weather and water conditions (Legleiter et al., 2009). Slight modifications to the setup and execution of future aerial photography for SfM should solve many, if not all, of the structural and accuracy issues with the orthophotos and DEMs. To remedy coverage gaps and low overlap, I recommend increasing the camera interval and using closer flight lines. Collecting RAW imagery could reduce any difficulties with exposure, and more extensive testing could ensure that all scenes have the correct exposure. The accuracy of the ground control used to georeference is paramount to the accuracy of SfM reconstructions. The nonlinear optimization step in Photoscan adjusts the camera positions and point cloud to minimize the georeferencing errors. The optimization results in an almost one-to-one transfer of the error in the GCPs to the output data sets (Fig. 6). The accuracy required in any elevation data set is dependent on the research question; however, because SfM is capable of producing extremely dense and precise measurements of the landscape, I would recommend that SfM surveys be georeferenced only with high-accuracy surveyed ground control

Please cite this article as: Dietrich, J.T., Riverscape mapping with helicopter-based Structure-from-Motion photogrammetry, Geomorphology (2015), http://dx.doi.org/10.1016/j.geomorph.2015.05.008

J.T. Dietrich / Geomorphology xxx (2015) xxx–xxx

9

Fig. 9. Primary morphologic variables for the study area. Dotted lines represent river segment breaks from Fig. 1; labels are at the top and bottom.

points. Collecting ground control points with a total station or RTK-GPS will reduce errors introduced into the data set through the georeferencing step. The greatest problem to overcome is the systematic error (Fig. 7) caused by the parallel camera geometry (Fig. 8). The simplest solution is to create convergent camera geometry, and James and Robson (2014) suggest several solutions to achieve this geometry in small unmanned aerial systems. One solution is the use of a single camera mounted in a gimbal that can be pointed off-nadir and adjusted to create convergent geometries. Another option would be to use multiple off-nadir cameras, which would not only create convergent geometry but also increase coverage and reduce overlap problems. 5.2. Riverscape mapping The heterogeneity of the Middle Fork is apparent in the downstream data (Fig. 9). With these raw data, it is difficult to differentiate the natural variations in the river from those that are the direct result of human influence. One exception is the lower portion of segment E, where the active channel width becomes narrow compared to the surrounding channel. This section of the channel is affected by dredge mining activity that produced an almost straight channel about 4 m wide and approximately 2 m deep. This straight section is visible in the plot of sinuosity, where the increase in slope around 18 km marks the transition from the natural channel to the dredged channel and the sinuosity index drops to almost one. Low sinuosity values in segments G and H also highlight channel-straightening modifications connected to cattle grazing, although the width changes are not as dramatic as the dredged channel.

The estimated depth calculation was an attempt to recover crosssectionally averaged depths for each sample point. The general pattern of the depths, deeper in the wider valley sections and shallower in the narrow valleys, matches with field observations. However, the estimated depths do not capture the depth variations associated with channel unit (riffle–pool) transitions and the overall variability in average depth between adjacent points is higher than was expected. The uncertainty in these data is likely the result of the assumptions that go into inferring depth from both the Manning and Jarrett equations. In addition, the smoothly varying Q2 discharge is unlikely to equal the actual bankfull conditions for all points on the stream, which accounts for some of the uncertainty. Ideally, recovering depths from the imagery and/or photogrammetry would provide more accurate depth measurements for all points in the stream. Accurate estimates of depth from imagery can be used to extend the riverscape analysis into the crossstream direction in addition to the downstream direction. The legacy of human impacts on the channel has complicated the riverscape of the Middle Fork. Despite these impacts or because of them, there are several definite relationships between the distance downstream and classified variables. Comparing the active channel widths across the nine different segments (Fig. 10), we see the general downstream trend of increased width with downstream distance, typical of downstream hydraulic geometry. Two areas that do not fit that general trend are segments E and H. Segment E consists of the valley section that has been the focus of intense restoration activity in an attempt to rehabilitate the stream from the historic dredge mining activity. The most recent phase of restoration, completed in the summer of 2012, filled in a secondary channel created by the dredging operation

Please cite this article as: Dietrich, J.T., Riverscape mapping with helicopter-based Structure-from-Motion photogrammetry, Geomorphology (2015), http://dx.doi.org/10.1016/j.geomorph.2015.05.008

10

J.T. Dietrich / Geomorphology xxx (2015) xxx–xxx

Fig. 10. Boxplots for active channel width aggregated by river segment (A), channel unit (B), restoration treatment (C), and grazing intensity (D). Boxes represent the median, 25th and 75th percentiles and whiskers are ≈2.97σ. N-values are the number of samples in each class (total = 10,776).

on the north side of the valley floor. The meandering channel on the south side of the valley has not yet adjusted to the increased flow, causing a discontinuity in the downstream trend. Segments G and H exhibit the largest range of active channel widths in all of the segments, where landowners straightened and relocated sections of the river to the side of the valley to facilitate cattle grazing. The broad range of width in these sections appears to be a reflection of these modifications to the river channel. The large increase in width in segment I is the result of the input of flow from Camp Creek, the largest tributary in the upper part of the basin, and channel transitioning to a plane-bed with decreased depth and increased width. Richards (1976) demonstrated that rivers show a difference in the width of the channel between riffles and pools. The data from the Middle Fork suggest that all three of the mapped channel units have small, yet statistically significant, differences in width. These data agree with field observations that width does vary, but not greatly between adjacent riffles and pools at the reach scale. Another possible control on width could be the underlying geology of the valley floor and adjacent hill slopes, which may act as a control on valley width and channel width (McDowell, 2001; May et al., 2013). The parts of the river in the Clarno formation have a significantly higher (p = 0.01) median valley and channel width than the Strawberry Volcanics in the upstream portions of the study area. Because these data also incorporate the downstream hydraulic geometry signal (Fig. 10), it is difficult to determine how much the geology is influencing the width throughout the study area. From my field observations of the intensely grazed portions of the study area, I hypothesized that the intensity of cattle grazing would increase the active channel width when compared to other sections of the river. In the downstream plot of active channel width (Fig. 9), it is

difficult to identify any reaches that are anomalously wider than others based on the current grazing intensities. Aggregating the data by grazing intensity confirms that the intensely grazed portion of the study area does have a significantly higher (p = 0.01) median width compared to areas with moderate or no grazing. The different restoration treatments appear to have an effect on the width of the river. However, most of the restoration work on this river is b10 years old and any width signal may result from pre-restoration conditions and is not a result of the restoration itself. It will be important to monitor these sites into the future to examine how the river adjusts to the restoration treatments. In the hyperscale analysis of active channel width versus downstream length, the positive correlations coefficients in the larger window sizes, 18–32 km, are indicative of the expected DHG relationship of a channel that widens as drainage area increases with distance downstream. From the 18-km window down to the 8-km window, the river divides into two zones, with the lower half of the river exhibiting a clear DHG relationship, and the upper half not conforming to a distinct DHG trend. This deterioration in the DHG trend appears to be triggered by the steep, narrow section of the river in segment C at 10 km downstream, where the river narrows slightly. The narrowing of the river over a relatively short section has a large effect on the relationship at intermediate scales (10–18 km windows). In the 2–6 km window, the frequency of the alternating pattern of positive and negative correlations between channel width and valley width reflects the alternating wide and narrow valleys at this scale (Fig. 9). Below the 2-km window, the oscillation of positive and negative correlations becomes more frequent. Carbonneau et al. (2012) suggested that the patterns at this scale could relate to riffle–pool sequences, but an initial examination of the pattern in relation to the classified channel unit data does not support that

Please cite this article as: Dietrich, J.T., Riverscape mapping with helicopter-based Structure-from-Motion photogrammetry, Geomorphology (2015), http://dx.doi.org/10.1016/j.geomorph.2015.05.008

J.T. Dietrich / Geomorphology xxx (2015) xxx–xxx

11

Fig. 11. Hyperscale graph of Pearson correlation coefficients of active channel width as a function of downstream distance. White areas within the triangle are portions of the analysis that did not meet the significance criteria (p = 0.01).

Fig. 12. Hyperscale graph of Pearson correlation coefficients of active channel width as a function of slope. White areas within the triangle are portions of the analysis that did not meet the significance criteria (p = 0.01).

Please cite this article as: Dietrich, J.T., Riverscape mapping with helicopter-based Structure-from-Motion photogrammetry, Geomorphology (2015), http://dx.doi.org/10.1016/j.geomorph.2015.05.008

12

J.T. Dietrich / Geomorphology xxx (2015) xxx–xxx

Fig. 13. Hyperscale graph of Pearson correlation coefficients of active channel width as a function of valley width. White areas within the triangle are portions of the analysis that did not meet the significance criteria (p = 0.01).

relationship. It will require additional investigation to determine what drives these patterns. The other two hyperscale analyses do not have the same welldefined patterns as active channel width and downstream distance. The graph of active channel width versus slope shows weak positive correlations at most scales, suggesting that slope has little relation to width. At the finer window sizes of 1 km or less, there is again a high frequency switching of the coefficients, interspersed with areas of insignificant correlation. This pattern suggests that slope is related to width in localized sections of the river. The irregular nature of the pattern could also be an artifact of the sampling distance or smoothing that was applied to the slope data. In the comparison of active channel width versus valley width, the notable relationships are the weak negative correlations, which suggest that stream width decreases in wide valleys. In this case, large changes in valley width compared to small fluctuations in channel width are influencing the strength of this relationship. Most of the large-scale patterns in these hyperscale analyses can be attributed to physical aspects of the river such as geology or the downstream hydraulic geometry. By comparison, the small-scale patterns are more difficult to interpret, and an important next step for this type of analysis would be to export the hyperscale results into a GIS environment where hyperscale patterns could be better visualized in the context of other GIS data and imagery. This could reveal what other variables might be related to variations in width, such as bank features, instream vegetation, or position in relation to meander bends. 6. Conclusions I have shown that a consumer-grade, off-the-shelf digital SLR camera mounted to a helicopter platform is capable of producing high spatial resolution aerial imagery using SfM. This instrument/platform combination bridges a gap between the aerial perspective afforded by small unmanned aerial systems and higher-cost commercial aerial

photography. The outputs from the SfM process produced highquality orthophotographs for a 32-km segment of the Middle Fork John Day River with only a few gaps in coverage. These data provided a solid foundation for extracting high-resolution measurements of the river at 3-m increments downstream that provide the basis for analyzing the morphologic complexity of the river at different scales. While the impacts of human activity on the Middle Fork are visible in the field, the collected data show that it is difficult, with few exceptions, to differentiate impact of human activity from the natural patterns in channel morphology. Both the boxplots of classified data and the hyperscale analysis show that there are differences in the downstream patterns at several different spatial scales, but these different patterns will need to be investigated further to determine causal relationships. One explanation could be that enough time has passed for the river to adjust partially to the historic human disturbances and that the river has not had enough time to adjust to the recent restoration activities. Structure-from-Motion has the potential to be a powerful and inexpensive tool for fluvial remote sensing with a few refinements. These refinements include correcting the camera geometry from parallel to convergent, collecting control points with high-accuracy GPS, improving flight planning to avoid gaps in the imagery, and correcting the exposure and increasing the radiometric resolution of the imagery to facilitate spectral or SfM-derived bathymetric measurements. With accurate DEMs and bathymetric measurements, the analysis of the riverscape can be extended to examine the complexities in the crossstream dimension in addition to the downstream component. The river morphology can also be coupled with other spatial environmental data and various aquatic habitat quality metrics and models to assess, in detail, the spatial distribution of aquatic habitat. In rivers like the Middle Fork, where human activity has altered the river and the restoration of habitat for threatened and endangered species has become a priority, high-resolution morphology and habitat data can identify areas of the

Please cite this article as: Dietrich, J.T., Riverscape mapping with helicopter-based Structure-from-Motion photogrammetry, Geomorphology (2015), http://dx.doi.org/10.1016/j.geomorph.2015.05.008

J.T. Dietrich / Geomorphology xxx (2015) xxx–xxx

river that suffer from impaired habitat quality and help prioritize future restoration sites. Acknowledgments This research was funded in part by a grant from the Oregon Watershed Enhancement Board with additional funding from the U.S. Bureau of Reclamation. I would like to thank Mark Fonstad and Bruce Rhoads for inviting this submission to the special issue. I would also like to thank Bruce and an anonymous reviewer for their comments that greatly improved this paper. I would also like to thank Patricia McDowell at the University of Oregon, Mark Croghan of the Bureau of Reclamation, and the Confederated Tribes of the Warms Springs Reservation of Oregon for providing field and equipment support for this research. Appendix A. Supplementary data Supplementary data associated with this article can be found in the online version, at http://dx.doi.org/10.1016/j.geomorph.2015.05.008. These data include Google map of the most important areas described in this article. References Agarwal, S., Snavely, N., Simon, I., Seitz, S.M., Szeliski, R., 2009. Building Rome in a day. 2009 IEEE 12th International Conference on Computer Vision, pp. 72–79 http://dx. doi.org/10.1109/iccv.2009.5459148. Angelaki, V., Harbor, J.M., 1995. Impacts of flow diversion for small hydroelectric power plants on sediment transport, Northwest Washington. Phys. Geogr. 16, 432–443. http://dx.doi.org/10.1080/02723646.1995.10642564. ASPRS, 2014. ASPRS positional accuracy standards for digital geospatial data. Photogramm. Eng. Remote Sens. 81 (3), a1–a26. http://dx.doi.org/10.14358/pers.81. 3. Bandow, J.R., 2004. Holocene Alluvial History of the Middle Fork John Day River, Oregon (MA Thesis). University of Oregon. Carbonneau, P.E., 2005. The threshold effect of image resolution on image-based automated grain size mapping in fluvial environments. Earth Surf. Process. Landf. 30, 1687–1693. http://dx.doi.org/10.1002/esp.1288. Carbonneau, P., Fonstad, M.A., Marcus, W.A., Dugdale, S.J., 2012. Making riverscapes real. Geomorphology 137, 74–86. http://dx.doi.org/10.1016/j.geomorph.2010.09.030. Cochran, B., 2013. Oxbow Conservation Area Dredge Tailings Restoration Project, Phase 2 Completion Report. Confederated Tribes of the Warm Springs Reservation of Oregon Fisheries Department – Habitat Program. Dandois, J.P., Ellis, E.C., 2013. High spatial resolution three-dimensional mapping of vegetation spectral dynamics using computer vision. Remote Sens. Environ. 136, 259–276. http://dx.doi.org/10.1016/j.rse.2013.04.005. Dugdale, S.J., Carbonneau, P.E., Campbell, D., 2010. Aerial photosieving of exposed gravel bars for the rapid calibration of airborne grain size maps. Earth Surf. Process. Landf. 35, 627–639. http://dx.doi.org/10.1002/esp.1936. Fausch, K.D., Torgersen, C.E., Baxter, C.V., Li, H.W., 2002. Landscapes to riverscapes: bridging the gap between research and conservation of stream fishes. Bioscience 52, 483–498. http://dx.doi.org/10.1641/0006-3568(2002)052[0483:LTRBTG]2.0.CO;2. Ferns, M.L., Brooks, H.C., 1995. The Bourne and Greenhorn subterranes of the Baker Terrane, Northeastern Oregon: implications for the evolution of the Blue Mountains Island-Arc System. In: Vallier, L.T., Brooks, H.C. (Eds.), Geology of the Blue Mountains Region of Oregon, Idaho, and Washington; Petrology and Tectonic Evolution of PreTertiary Rocks of the Blue Mountains Region. USGS Professional Paper 1438. United States Geological Survey. Ferreira, M., 2014. Transect Tool Available at. http://gis4geomorphology.com/streamtransects-partial/. Fonstad, M.A., Marcus, W.A., 2010. High resolution, basin extent observations and implications for understanding river form and process. Earth Surf. Process. Landf. 35, 680–698. http://dx.doi.org/10.1002/esp.1969. Fonstad, M.A., Dietrich, J.T., Courville, B.C., Jensen, J.L., Carbonneau, P.E., 2013. Topographic structure from motion: a new development in photogrammetric measurement. Earth Surf. Process. Landf. 38, 421–430. http://dx.doi.org/10.1002/esp.3366. Frissell, C.A., Liss, W.J., Warren, C.E., Hurley, M.D., 1986. A hierarchical framework for stream habitat classification: viewing streams in a watershed context. Environ. Manag. 10, 199–214. http://dx.doi.org/10.1007/BF01867358. Geological Survey, U.S., 2012. The StreamStats program for Oregon [WWW Document]. URL http://water.usgs.gov/osw/streamstats/oregon.html. Graf, W.L., 2000. Locational probability for a dammed, urbanizing stream: Salt River, Arizona, USA. Environ. Manag. 25, 321–335. http://dx.doi.org/10.1007/s002679910025. Haunert, J.-H., Sester, M., 2008. Area collapse and road centerlines based on straight skeletons. Geoinformatica 12, 169–191. http://dx.doi.org/10.1007/s10707-007-0028-x. Hey, R., Thorne, C., 1986. Stable channels with mobile gravel beds. J. Hydraul. Eng. 112, 671–689. http://dx.doi.org/10.1061/(ASCE)0733-9429(1986)112:8(671).

13

James, M.R., Robson, S., 2012. Straightforward reconstruction of 3D surfaces and topography with a camera: accuracy and geoscience application. J. Geophys. Res. 117. http:// dx.doi.org/10.1029/2011JF002289 F03017. James, M.R., Robson, S., 2014. Mitigating systematic error in topographic models derived from UAV and ground-based image networks. Earth Surf. Process. Landf. 39, 1413–1420. http://dx.doi.org/10.1002/esp.3609. Jarrett, R., 1984. Hydraulics of high‐gradient streams. J. Hydraul. Eng. 110, 1519–1539. http://dx.doi.org/10.1061/(ASCE)0733-9429(1984)110:11(1519). Javernick, L., Brasington, J., Caruso, B., 2014. Modeling the topography of shallow braided rivers using Structure-from-Motion photogrammetry. Geomorphology 213, 166–182. http://dx.doi.org/10.1016/j.geomorph.2014.01.006. Jennings, M.E., Thomas, W.O., Riggs, H., 1994. Nationwide Summary of US Geological Survey Regional Regression Equations for Estimating Magnitude and Frequency of Floods for Ungaged Sites, 1993 (No. 94-4002), Water Resources Investigations Report. US Geological Survey. Jett, S.M., 1998. Alluvial Fan Development in a Confined Montane Valley, Middle Fork John Day River, Eastern Oregon (MS Thesis). University of Oregon. Knighton, D., 1998. Fluvial Forms and Processes: A New Perspective, Rev. and updated ed. Arnold, London; New York. Kondolf, G.M., 1994. Geomorphic and environmental effects of instream gravel mining. Landsc. Urban Plan. 28, 225–243. http://dx.doi.org/10.1016/0169-2046(94)90010-8. Kondolf, G.M., Larson, M., 1995. Historical channel analysis and its application to riparian and aquatic habitat restoration. Aquat. Conserv. Mar. Freshwat. Ecosyst. 5, 109–126. http://dx.doi.org/10.1002/aqc.3270050204. Lane, S.N., Richards, K.S., Chandler, J.H., 1994. Developments in monitoring and modelling small-scale river bed topography. Earth Surf. Process. Landf. 19, 349–368. http://dx. doi.org/10.1002/esp.3290190406. Lane, S.N., Westaway, R.M., Murray Hicks, D., 2003. Estimation of erosion and deposition volumes in a large, gravel-bed, braided river using synoptic remote sensing. Earth Surf. Process. Landf. 28, 249–271. http://dx.doi.org/10.1002/esp.483. Legleiter, C.J., Fonstad, M.A., 2012. An introduction to the physical basis for deriving river information by optical remote sensing. Fluvial Remote Sensing for Science and Management. John Wiley & Sons, Ltd, pp. 43–69 http://dx.doi.org/10.1002/9781119940791.ch3. Legleiter, C.J., Kyriakidis, P.C., 2006. Forward and inverse transformations between cartesian and channel-fitted coordinate systems for meandering rivers. Math. Geol. 38, 927–958. http://dx.doi.org/10.1007/s11004-006-9056-6. Legleiter, C.J., Roberts, D.A., 2005. Effects of channel morphology and sensor spatial resolution on image-derived depth estimates. Remote Sens. Environ. 95, 231–247. http:// dx.doi.org/10.1016/j.rse.2004.12.013. Legleiter, C.J., Roberts, D.A., Marcus, W.A., Fonstad, M.A., 2004. Passive optical remote sensing of river channel morphology and in-stream habitat: physical basis and feasibility. Remote Sens. Environ. 93, 493–510. http://dx.doi.org/10.1016/j.rse.2004.07. 019. Legleiter, C.J., Roberts, D.A., Lawrence, R.L., 2009. Spectrally based remote sensing of river bathymetry. Earth Surf. Process. Landf. 34, 1039–1059. http://dx.doi.org/10.1002/esp. 1787. Leopold, L.B., Maddock, T., 1953. The hydraulic geometry of stream channels and some physiographic implications. United States Geological Survey, p. 252. Mackin, J.H., 1948. Concept of the graded river. Geol. Soc. Am. Bull. 59, 463–512. http://dx. doi.org/10.1130/0016-7606(1948)59[463:COTGR]2.0.CO;2. Magilligan, F.J., McDowell, P.F., 1997. Stream channel adjustments following elimination of cattle grazing. J. Am. Water Resour. Assoc. 33, 867–878. http://dx.doi.org/10. 1111/j.1752-1688.1997.tb04111.x. Marcus, W.A., Fonstad, M.A., 2008. Optical remote mapping of rivers at sub-meter resolutions and watershed extents. Earth Surf. Process. Landf. 33, 4–24. http://dx.doi.org/10. 1002/esp.1637. Marcus, W.A., Fonstad, M.A., 2010. Remote sensing of rivers: the emergence of a subdiscipline in the river sciences. Earth Surf. Process. Landf. 35, 1867–1872. http://dx.doi. org/10.1002/esp.2094. Marcus, W.A., Roberts, K., Harvey, L., Tackman, G., 1992. An evaluation of methods for estimating Manning's n in small mountain streams. Mt. Res. Dev. 12, 227–239. http:// dx.doi.org/10.2307/3673667. Marston, R.A., Girel, J., Pautou, G., Piegay, H., Bravard, J.-P., Arneson, C., 1995. Channel metamorphosis, floodplain disturbance, and vegetation development: Ain River, France. Geomorph. Biogeomorph. Terrestrial Freshw. Syst. 13, 121–131. http://dx. doi.org/10.1016/0169-555X(95)00066-E. Mathews, A.J., Jensen, J.L.R., 2013. Visualizing and quantifying vineyard canopy LAI using an unmanned aerial vehicle (UAV) collected high density structure from motion point cloud. Remote Sens. 5, 2164–2183. http://dx.doi.org/10.3390/rs5052164. May, C., Roering, J., Eaton, L.S., Burnett, K.M., 2013. Controls on valley width in mountainous landscapes: the role of landsliding and implications for salmonid habitat. Geology 41, 503–506. http://dx.doi.org/10.1130/G33979.1. McDowell, P.F., 2001. Spatial Variations in Channel Morphology at Segment and Reach Scales, Middle Fork John Day River, Northeastern Oregon. In: Dorava, J.M., Montgomery, D.R., Palcsak, B.B., Fitzpatrick, F.A. (Eds.), Geomorphic Processes and Riverine Habitat. American Geophysical Union, pp. 159–172 http://dx.doi.org/10. 1029/ws004p0159. Millar, R.G., 2005. Theoretical regime equations for mobile gravel-bed rivers with stable banks. Geomorphology 64, 207–220. http://dx.doi.org/10.1016/j.geomorph.2004.07. 001. Montgomery, D.R., Buffington, J.M., 1998. Channel processes, classification, and response. In: Naiman, R.J., Bilby, R.E. (Eds.), River Ecology and Management: Lessons from the Pacific Coastal Ecoregion. Springer, New York. NOAA, 2005. Endangered and threatened species; designation of critical habitat for 12 evolutionarily significant units of West Coast Salmon and Steelhead in Washington, Oregon, and Idaho (50 CFR Part 226). Fed. Regist. 70.

Please cite this article as: Dietrich, J.T., Riverscape mapping with helicopter-based Structure-from-Motion photogrammetry, Geomorphology (2015), http://dx.doi.org/10.1016/j.geomorph.2015.05.008

14

J.T. Dietrich / Geomorphology xxx (2015) xxx–xxx

Parker, G., Wilcock, P.R., Paola, C., Dietrich, W.E., Pitlick, J., 2007. Physical basis for quasiuniversal relations describing bankfull hydraulic geometry of single-thread gravel bed rivers. J. Geophys. Res. 112. http://dx.doi.org/10.1029/2006JF000549 F04005. Pavelsky, T.M., Smith, L.C., 2008. RivWidth: a software tool for the calculation of river widths from remotely sensed imagery. IEEE Geosci. Remote Sens. Lett. 5, 70–73. http://dx.doi.org/10.1109/LGRS.2007.908305. Pizzuto, J.E., 2008. Streambank erosion and river width adjustment. Sedimentation Engineering. American Society of Civil Engineers, pp. 387–438. Poole, G.C., 2002. Fluvial landscape ecology: addressing uniqueness within the river discontinuum. Freshw. Biol. 47, 641–660. http://dx.doi.org/10.1046/j.1365-2427. 2002.00922.x. Richards, K.S., 1976. Channel width and the riffle-pool sequence. Geol. Soc. Am. Bull. 87, 883–890. http://dx.doi.org/10.1130/0016-7606(1976)87b883:CWATRSN2.0.CO;2. Richards, K.S., Wood, R., 1977. Urbanization, water redistribution, and their effect on channel processes. In: Gregory, K.J. (Ed.), River Channel Changes. Wiley, Chichester. New York, pp. 369–388. Ryan, S., 1997. Morphologic response of subalpine streams to transbasin flow diversion. J. Am. Water Resour. Assoc. 33, 839–854. http://dx.doi.org/10.1111/j.1752-1688. 1997.tb04109.x. Snavely, N., Seitz, S.M., Szeliski, R., 2007. Modeling the world from internet photo collections. Int. J. Comput. Vis. 80, 189–210. http://dx.doi.org/10.1007/s11263-007-0107-3. Trimble, S.W., Mendel, A.C., 1995. The cow as a geomorphic agent—a critical review. Geomorph. Biogeomorph. Terrestrial Freshw. Syst. 13, 233–253. http://dx.doi.org/ 10.1016/0169-555X(95)00028-4. U.S. Fish & Wildlife Service, 2010. Bull Trout Final Critical Habitat Justification: Rationale for Why Habitat is Essential, and Documentation of Occupancy. Oregon, Portland.

Ullman, S., 1979. The Interpretation of Visual Motion. MIT Press, Cambridge, Mass. Vannote, R., Minshall, G., Cummins, K., Sedell, J., Cushing, C., 1980. The river continuum concept. Can. J. Fish. Aquat. Sci. 37, 130–137. Walther, S.C., Marcus, W.A., Fonstad, M.A., 2011. Evaluation of high-resolution, truecolour, aerial imagery for mapping bathymetry in a clear-water river without ground-based depth measurements. Int. J. Remote Sens. 32, 4343–4363. http://dx. doi.org/10.1080/01431161.2010.486418. Ward, J.V., Stanford, J.A., 1995. The serial discontinuity concept: extending the model to floodplain rivers. Regul. Rivers Res. Manag. 10, 159–168. http://dx.doi.org/10.1002/ rrr.3450100211. Westoby, M.J., Brasington, J., Glasser, N.F., Hambrey, M.J., Reynolds, J.M., 2012. “Structurefrom-Motion” photogrammetry: a low-cost, effective tool for geoscience applications. Geomorphology 179, 300–314. http://dx.doi.org/10.1016/j.geomorph.2012.08.021. Whittaker, A.C., Cowie, P.A., Attal, M., Tucker, G.E., Roberts, G.P., 2007. Bedrock channel adjustment to tectonic forcing: implications for predicting river incision rates. Geology 35, 103–106. http://dx.doi.org/10.1130/G23106A.1. Woodget, A.S., Carbonneau, P.E., Visser, F., Maddock, I., 2014. Quantifying submerged fluvial topography at hyperspatial resolutions with UAS imagery and SfMphotogrammetry. Earth Surf. Process. Landf. http://dx.doi.org/10.1002/esp.3613. Yamazaki, D., O'Loughlin, F., Trigg, M.A., Miller, Z.F., Pavelsky, T.M., Bates, P.D., 2014. Development of the global width database for large rivers. Water Resour. Res. 50, 3467–3480. http://dx.doi.org/10.1002/2013WR014664.

Please cite this article as: Dietrich, J.T., Riverscape mapping with helicopter-based Structure-from-Motion photogrammetry, Geomorphology (2015), http://dx.doi.org/10.1016/j.geomorph.2015.05.008