Introduction

The U.S. Geological Survey (USGS)’s Grand Canyon Monitoring and Research Center (GCMRC) is the primary science provider for the Glen Canyon Dam Adaptive Management Program (GCDAMP). In support of that mission, GCMRC periodically collects airborne image data for the Colorado River corridor within Arizona (fig. 1), enabling scientists to study the operational effects of Glen Canyon Dam on the corridor’s natural and cultural resources. The segment of the Colorado River that flows through Glen Canyon, Marble Canyon, and Grand Canyon is characterized by steep terrain and a narrow, sinuous riparian corridor, making many areas logistically difficult to study. Thus, remotely sensed data and derived maps are important logistical tools for field research and monitoring and for image-based, high-resolution change detection. These data are collected from Lake Powell near Page, Arizona (just upstream from Glen Canyon Dam), to Lake Mead at Pearce Ferry, Arizona, for a total length of 475 kilometers (km) at a width of about 500 meters (m) centered on the mainstem of the Colorado River and its seven primary tributaries: the Paria River, the Little Colorado River, Bright Angel Creek, Shinumo Creek, Tapeats Creek, Kanab Creek, and Havasu Creek (fig. 1). These data can be used for change detection by comparison with other data from the GCMRC image archive, which includes five dates of comparable full-corridor, digitally acquired, multispectral datasets since 2002, as well as a longer term record of analog and film-based aerial photography (see, for example, Durning and others, 2021; Sankey and others 2018, 2015).

Data Collection

Airborne data were acquired from May 29 to June 4, 2021, under contract by Fugro through the USGS Geospatial Products and Services Contracts, by using two fixed-wing aircraft at flight altitudes of 2,440 to 3,350 m above mean sea level. GCMRC worked with the U.S. Department of the Interior Bureau of Reclamation and other stakeholders of the GCDAMP to implement low steady release flows of about 227 cubic meters per second (m³/s) from Glen Canyon Dam during the entire period of data acquisition. The low steady release ensured that river discharge and the wetted shoreline of the river channel would be as stable and consistent as possible throughout the entire study period and resulting dataset. As is demonstrated in figure 2, the discharge recorded at Lees Ferry, 24 km downstream from Glen Canyon Dam, varied somewhat above the targeted discharge of 227 m³/s, and additional flow from tributaries added to the discharge as the steady-flow wave progressed downstream through the canyon. Given the speed of the wave moving through the canyon, it was important to closely project when the low steady flows would reach a given segment of the canyon for image acquisition to coincide with the targeted discharge; during the first few days of the mission, it was especially important that the aircraft did not proceed farther downstream than the low steady flow wave. The first flight was cleared to begin data collection at the dam at 10:30 mountain time (MT) on May 29, and the final data collection ended by 14:00 MT on June 4.

For any given section of the river corridor, 5 or 6 overlapping linear flightlines were acquired, allowing for the greatest probability of error-free and low-shadow imagery. The image data were acquired at a spatial resolution of 20 centimeters (cm) in four wavelength bands: red (0.619–0.651 micron [µm]; band 1), green (0.525–0.585 µm; band 2), blue (0.435–0.495 µm; band 3), and near infrared (NIR) (0.808–0.882 µm; band 4). The data were acquired utilizing two similar fixed-wing aircraft, each mounted with a Leica ADS100 digital push-broom multi-spectral sensor. The sensors collected multispectral data at the following viewing angles simultaneously along the flight track: red, green, blue, and NIR at +25.6° forward, 0° nadir, and −19.4° backward.

Image Processing

Aerotriangulation (AT) was performed by Fugro to assign ground-control values to points on three blocks of photographs. GCMRC provided Fugro with 112 ground-control points (GCPs) to use in the AT. Fugro processed the bundle adjustment for each block and generated stereo imagery to produce a photogrammetric digital surface model (DSM) which was filtered to produce a digital elevation model (DEM). The raw imagery from the ADS100 sensors was then orthorectified using the DEM to produce orthoimages. Fugro produced an orthomosaic from the orthoimages and delivered it to GCMRC in the required tiling scheme, projection, datum, and file format (see “Data Organization” section). In addition to the tiled orthomosaic product, much of the raw data used by Fugro to create the orthomosaic product were also delivered to GCMRC in 249 flightlines and subdivided into 4,417 tiles.

The orthomosaic data delivered by Fugro were checked by GCMRC scientists for smear, shadow extent, and water clarity, as described for previous image acquisitions by Durning and others (2016b) and Davis (2012). GCMRC staff then further corrected the orthomosaic primarily for localized instances of smear, shadow extent, and water clarity by replacing remaining blemishes with corresponding portions of orthoimages from error-free flight lines if available. The final orthomosaic product published by the USGS is subset into 105 files following the map-tile scheme shown in figure 3, and then further organized into 7 zones. The map-tile scheme used to subset the orthomosaic was modified from quarter quadrangles (QQ) subdivided from USGS 7.5-minute quadrangle map borders, to best fit the river corridor.

To demonstrate the corridor-wide spectral variability in the orthomosaic of images, we summarized digital-number (DN) radiance values by distance along the Colorado River upstream or downstream from Lees Ferry, Ariz., for four different landcover types (fig. 4). The four landcover types were sampled from interpretation of the 2013 and 2021 imagery and represent overlapping areas in each image of deep water, dry sand, geologic (which included a variety of lithologies, rock colors, and spectral characteristics found in Grand Canyon), and vegetation. Figure 4 shows summaries of DN radiance values by the four landcover types for both the 2021 orthomosaic and the 2013 orthomosaic published by Durning and others (2016a) for the same landcover samples throughout the river corridor. The 2021 imagery data span the 16-bit unsigned-integer range (0–65,535), whereas the 2013 imagery data were published by Durning and others (2016a) as 16-bit signed-integer files with DN values that primarily (0–99th percentile) spanned the 12-bit unsigned-integer range (0–4,095) but included positive integer outlier values in the larger 16-bit signed-integer range. Thus, for presentation purposes in figures 4 through 6, the 2013 DN values from imagery published by Durning and others (2016a) were multiplied by a factor of 16. In addition to illustrating spectral variability by four common landcover types in the river corridor, figure 4 also highlights variability by river distance throughout the river corridor. That variability is indicative of variation in landcover composition over those distances along the river in Grand Canyon. That variability is also indicative of variations in factors such as atmospheric water-vapor content from changing weather over time during the overflight missions, in which reflectance is dominated by atmospheric scattering or shadowing caused by interactions between solar illumination angles and topography.

Figure 5 summarizes the DN variation by spectral band in the 2021 orthomosaic and the 2013 orthomosaic published by Durning and others (2016a) for the same cover types and areas shown in figure 4. The shapes of the spectral curves plotted in figure 5 are similar between 2013 and 2021 datasets, but the 2021 image data occupy different and often larger or higher DN values in the 16-bit unsigned-integer range as explained above.

Figure 6 demonstrates the corridor-wide spectral variability in the largest (99th percentile), smallest (1st percentile), and mean DN values in the orthomosaic of images, and highlights variability between the 2013 and 2021 orthomosaics.

Some shadows remained in the final mosaic dataset following work by GCMRC and Fugro. Locations of shadows that could pose an issue to users working with the imagery in areas of the river channel and riparian zone were identified and are published in a table in the data release associated with this report (Sankey and others, 2024). Shadows that obscured riparian vegetation, sandbars, or the river itself within the shoreline terrace were digitized into a point shapefile with attributes including the tile name, river kilometer, river mile, side of the river the shadow was on, and the latitude and longitude position of the approximate centroid of the shadow. The tile name was unique to the file. The river kilometer (km) is the distance upstream (−) or downstream (+) of Lees Ferry (km 0) to the hundredth of a kilometer, and the river mile (using the same system) is represented to the tenth of a mile. The side of the river is the riverbank where the shadow is located, either river-left or river-right, as viewed in the downstream direction. The interpretation of shadows can be subjective, and the approximate centroid point locations, but not areal extents, of shadows are identified to reduce bias in how the shadow effects could or should be interpreted.

Accuracy and Error

A total of 159 GCPs were marked during image acquisition with diagonally alternating black and white plastic panels centered on control points throughout the river corridor in the GCMRC survey control network (Hazel and others, 2008). GCMRC provided coordinates of 112 of these GCPs to Fugro to use for AT. Results of the AT reported by Fugro were root mean square errors (RMSEs) of 0.10 m easting and 0.12 m northing, or smaller, depending on the image block.

The GCMRC independently assessed the horizontal accuracy of the final published orthomosaic using three horizontal accuracy assessments that are relevant for the 2021 dataset. First, GCMRC staff compared 2021 orthomosaic image coordinates (“image”) to GCMRC survey-control network control-point coordinates (“control”) for 47 GCPs that were not provided to, or used by, Fugro in the AT (fig. 7). Second, GCMRC staff compared 2021 orthomosaic image coordinates to GCMRC survey-control network control-point coordinates for 107 GCPs. Although 112 GCPs were provided to, and used by, Fugro in the AT, 5 of those GCPs had large observed errors and visible, localized smear in the vicinity of the panel in the 2021 orthomosaic and were not included in this analysis. Third, GCMRC staff compared 2021 orthomosaic image coordinates to 2013 orthomosaic image coordinates using 74 GCPs that coincided in both dates of image acquisitions. A total of 79 GCPs coincided between 2009 and 2013; thus, not included in this analysis were one point that had a large observed error and visible, localized smear in the 2021 orthomosaic and four points in the 2013 orthomosaic which were believed to have been erroneously placed or disturbed (see Durning and others, 2016b). The accuracy assessments were completed by following the Federal Geographic Data Committee geospatial-positioning accuracy standard for calculating RMSE and accuracy at the 95-percent confidence level (Federal Geographic Data Committee, 1998). First GCMRC staff calculated the square error for each coordinate pair independently. The square error ($\text{SE}$) was determined for the easting ($X$), northing ($Y$), and $X$ and $Y$ combined, respectively, using the following equations:

where A sum of square errors ($\text{SSE}$) was calculated for the $X$ coordinates, $Y$ coordinates, or the combined $X$ and $Y$ coordinates, as follows: where Next, the mean square error ($\text{MSE}$) was calculated from the $\text{SSE}$ for the $X$ coordinates, $Y$ coordinates, or combined $X$ and $Y$ coordinates: where Then the $\text{RMSE}$ was calculated for the $X$ coordinates, $Y$ coordinates, or the combined $X$ and $Y$ coordinates as follows: Alternatively, the $\text{RMSE}$ calculation can be accomplished in one step for the $X$ or $Y$ coordinates with the following equation: where Then accuracy at the 95-percent confidence level was calculated separately for the $X$ or $Y$ coordinates as follows: The combined horizontal RMSE for the $X$ and $Y$ coordinates can be calculated in one step with the following equation: Then the combined horizontal accuracy at the 95-percent confidence level for the $X$ and $Y$ coordinates was calculated as follows: In comparing the 2021 and 2013 image orthomosaics, 2021 was considered the “control” and 2013 the “image” dataset.

Comparing the 2021 orthomosaic to the GCMRC survey-control network using the 47 GCPs that were not provided to Fugro, GCMRC staff determined RMSEs of 0.208, 0.212, and 0.297 m for easting, northing, and $XY$ combined, respectively (table 1.1 in appendix 1), with 95-percent-confidence-level accuracy estimates of 0.408, 0.416, 0.514 m, respectively (table 1). Comparing the 2021 mosaic to the GCMRC survey-control network using the 107 GCPs provided to Fugro for AT, GCMRC staff determined RMSEs of 0.253, 0.195, and 0.320 m for easting, northing, and $XY$ combined, respectively (table 1.2 in appendix 1), with 95-percent-confidence-level accuracy estimates of 0.495, 0.383, and 0.553 m, respectively (table 1). Comparing the 2021 mosaic with the 2013 mosaic, GCMRC staff determined RMSEs of 0.310, 0.297, and 0.430 m for easting, northing, and $XY$ combined, respectively (table 1.3 in appendix 1), with 95-percent confidence level accuracy estimates of 0.609, 0.583, 0.744 m, respectively (table 1). A comparison of the 2013 orthomosaic to the GCMRC survey-control network was published by Durning and others (2016a, b).

Data Organization

The 16-bit unsigned-integer image data are stored as four-band images in embedded GeoTIFF format, which can be read and used by most geographic information systems (GIS) and image-processing software. The bands are organized as follows: band 1, red; band 2, green; band 3, blue; and band 4, NIR. The image files are projected in the State Plane Coordinate System, using the central Arizona zone (202) with the North American Datum of 1983 National Adjustment of 2011 (NAD 83(2011)). The map-tile scheme used to segment the image mosaic was modified from quarter quadrangles (QQ) and quarter-quarter quadrangles (QQQ) subdivided from USGS 7.5-minute quadrangle map borders to best fit the river corridor. To minimize file storage and processing time, each image tile was subset to include only the area of interest following the riparian corridor and main tributaries. A total of 105 image tiles are in the final mosaic dataset. The 105 tiles are labeled (and associated data files are named) using QQQ names. However, tile boundaries do not strictly adhere to the QQQ or QQ boundary associated with their name. The image-tile files are placed in seven file folders based on the half-degree geographic boundaries within the study area (fig. 3). The file folders are sequentially referred to as zones 1 through 7, beginning at Glen Canyon Dam and proceeding downriver. The image tiles contained within each folder or zone are shown on the index map for each respective zone (fig. 3).

Companion Data

This May 2021 image mosaic dataset (Sankey and others, 2024) is the most recent in a series of high-resolution datasets collected for the Colorado River in the Grand Canyon. The latest comparable dataset, which was acquired in May 2013, has the same band wavelengths and a spatial resolution of 20 cm (Durning and others, 2016a, b). Comparable datasets were also previously acquired in 2009 (Davis, 2013) and 2005 and 2002 (Davis, 2012). During the May 2021 mission, data for a DSM and a DEM were also acquired. The DSM and DEM data, which were processed independently of the spectral imagery, are available from Sankey and others (2025) (available online at https://doi.org/10.5066/P93Y4FMJ).