Information on water depth in river channels is important for a number of applications in water resource management but can be difficult to obtain via conventional field methods, particularly over large spatial extents and with the kind of frequency and regularity required to support monitoring programs. Remote sensing methods could provide a viable alternative means of mapping river bathymetry (i.e., water depth). The purpose of this study was to develop and test new, spectrally based techniques for estimating water depth from satellite image data. More specifically, a neural network-based temporal ensembling approach was evaluated in comparison to several other neural network depth retrieval (NNDR) algorithms. These methods are described in a manuscript titled "Neural Network-Based Temporal Ensembling of Water Depth Estimates Derived from SuperDove Images" and the purpose of this data release is to make available the depth maps produced using these techniques. The images used as input were acquired by the SuperDove cubesats comprising the PlanetScope constellation, but the original images cannot be redistributed due to licensing restrictions; the end products derived from these images are provided instead. The large number of cubesats in the PlanetScope constellation allows for frequent temporal coverage and the neural network-based approach takes advantage of this high density time series of information by estimating depth via one of four NNDR methods described in the manuscript: 1. Mean-spec: the images are averaged over time and the resulting mean image is used as input to the NNDR. 2. Mean-depth: a separate NNDR is applied independently to each image in the time series and the resulting time series of depth estimates is averaged to obtain the final depth map. 3. NN-depth: a separate NNDR is applied independently to each image in the time series and the resulting time series of depth estimates is then used as input to a second, ensembling neural network that essentially weights the depth estimates from the individual images so as to optimize the agreement between the image-derived depth estimates and field measurements of water depth used for training; the output from the ensembling neural network serves as the final depth map. 4. Optimal single image: a separate NNDR is applied independently to each image in the time series and only the image that yields the strongest agreement between the image-derived depth estimates and the field measurements of water depth used for training is used as the final depth map. MATLAB (Version 24.1, including the Deep Learning Toolbox) source code for performing this analysis is provided in the function NN_depth_ensembling.m available on the main landing page for the data release of which this is a child item, along with a flow chart illustrating the four different neural network-based depth retrieval methods. To develop and test this new NNDR approach, the method was applied to satellite images from the Colorado River near Lees Ferry, AZ, acquired in March and April of 2021. Field measurements of water depth available through another data release (Legleiter, C.J., Debenedetto, G.P., and Forbes, B.T., 2022, Field measurements of water depth from the Colorado River near Lees Ferry, AZ, March 16-18, 2021: U.S. Geological Survey data release, https://doi.org/10.5066/P9HZL7BZ) were used for training and validation. The depth maps produced via each of the four methods described above are provided as GeoTIFF files, with file name suffixes that indicate the method employed: Colorado_mean-spec.tif, Colorado_mean-depth.tif, Colorado_NN-depth.tif, and Colorado-single-image.tif. In addition, to assess the robustness of the Mean-spec and NN-depth methods to the introduction of a large pulse of sediment by a flood event that occurred partway through the image time series, depth maps from before and after the flood are provided in the files Colorado_Mean-spec_after_flood.tif,

Information on water depth in river channels is important for a number of applications in water resource management but can be difficult to obtain via conventional field methods, particularly over large spatial extents and with the kind of frequency and regularity required to support monitoring programs. Remote sensing methods could provide a viable alternative means of mapping river bathymetry (i.e., water depth). The purpose of this study was to develop and test new, spectrally based techniques for estimating water depth from satellite image data. More specifically, a neural network-based temporal ensembling approach was evaluated in comparison to several other neural network depth retrieval (NNDR) algorithms. These methods are described in a manuscript titled "Neural Network-Based Temporal Ensembling of Water Depth Estimates Derived from SuperDove Images" and the purpose of this data release is to make available the depth maps produced using these techniques. The images used as input were acquired by the SuperDove cubesats comprising the PlanetScope constellation, but the original images cannot be redistributed due to licensing restrictions; the end products derived from these images are provided instead. The large number of cubesats in the PlanetScope constellation allows for frequent temporal coverage and the neural network-based approach takes advantage of this high density time series of information by estimating depth via one of four NNDR methods described in the manuscript: 1. Mean-spec: the images are averaged over time and the resulting mean image is used as input to the NNDR. 2. Mean-depth: a separate NNDR is applied independently to each image in the time series and the resulting time series of depth estimates is averaged to obtain the final depth map. 3. NN-depth: a separate NNDR is applied independently to each image in the time series and the resulting time series of depth estimates is then used as input to a second, ensembling neural network that essentially weights the depth estimates from the individual images so as to optimize the agreement between the image-derived depth estimates and field measurements of water depth used for training; the output from the ensembling neural network serves as the final depth map. 4. Optimal single image: a separate NNDR is applied independently to each image in the time series and only the image that yields the strongest agreement between the image-derived depth estimates and the field measurements of water depth used for training is used as the final depth map. MATLAB (Version 24.1, including the Deep Learning Toolbox) source code for performing this analysis is provided in the function NN_depth_ensembling.m available on the main landing page for the data release of which this is a child item, along with a flow chart illustrating the four different neural network-based depth retrieval methods. To develop and test this new NNDR approach, the method was applied to satellite images from the Colorado River near Lees Ferry, AZ, acquired in March and April of 2021. Field measurements of water depth available through another data release (Legleiter, C.J., Debenedetto, G.P., and Forbes, B.T., 2022, Field measurements of water depth from the Colorado River near Lees Ferry, AZ, March 16-18, 2021: U.S. Geological Survey data release, https://doi.org/10.5066/P9HZL7BZ) were used for training and validation. The depth maps produced via each of the four methods described above are provided as GeoTIFF files, with file name suffixes that indicate the method employed: Colorado_mean-spec.tif, Colorado_mean-depth.tif, Colorado_NN-depth.tif, and Colorado-single-image.tif. In addition, to assess the robustness of the Mean-spec and NN-depth methods to the introduction of a large pulse of sediment by a flood event that occurred partway through the image time series, depth maps from before and after the flood are provided in the files Colorado_Mean-spec_after_flood.tif,

Information on water depth in river channels is important for a number of applications in water resource management but can be difficult to obtain via conventional field methods, particularly over large spatial extents and with the kind of frequency and regularity required to support monitoring programs. Remote sensing methods could provide a viable alternative means of mapping river bathymetry (i.e., water depth). The purpose of this study was to develop and test new, spectrally based techniques for estimating water depth from satellite image data. More specifically, a neural network-based temporal ensembling approach was evaluated in comparison to several other neural network depth retrieval (NNDR) algorithms. These methods are described in a manuscript titled "Neural Network-Based Temporal Ensembling of Water Depth Estimates Derived from SuperDove Images" and the purpose of this data release is to make available the depth maps produced using these techniques. The images used as input were acquired by the SuperDove cubesats comprising the PlanetScope constellation, but the original images cannot be redistributed due to licensing restrictions; the end products derived from these images are provided instead. The large number of cubesats in the PlanetScope constellation allows for frequent temporal coverage and the neural network-based approach takes advantage of this high density time series of information by estimating depth via one of four NNDR methods described in the manuscript: 1. Mean-spec: the images are averaged over time and the resulting mean image is used as input to the NNDR. 2. Mean-depth: a separate NNDR is applied independently to each image in the time series and the resulting time series of depth estimates is averaged to obtain the final depth map. 3. NN-depth: a separate NNDR is applied independently to each image in the time series and the resulting time series of depth estimates is then used as input to a second, ensembling neural network that essentially weights the depth estimates from the individual images so as to optimize the agreement between the image-derived depth estimates and field measurements of water depth used for training; the output from the ensembling neural network serves as the final depth map. 4. Optimal single image: a separate NNDR is applied independently to each image in the time series and only the image that yields the strongest agreement between the image-derived depth estimates and the field measurements of water depth used for training is used as the final depth map. MATLAB (Version 24.1, including the Deep Learning Toolbox) source code for performing this analysis is provided in the function NN_depth_ensembling.m and the figure included on this landing page provides a flow chart illustrating the four different neural network-based depth retrieval methods. As examples of the resulting models, MATLAB *.mat data files containing the best-performing neural network model for each site are provided below, along with a file that lists the PlanetScope image identifiers for the images that were used for each site. To develop and test this new NNDR approach, the method was applied to satellite images from three rivers across the U.S.: the American, Colorado, and Potomac. For each site, field measurements of water depth available through other data releases were used for training and validation. The depth maps produced via each of the four methods described above are provided as GeoTIFF files, with file name suffixes that indicate the method employed: X_mean-spec.tif, X_mean-depth.tif, X_NN-depth.tif, and X-single-image.tif, where X denotes the site name. The spatial resolution of the depth maps is 3 meters and the pixel values within each map are water depth estimates in units of meters.

Information on water depth in river channels is important for a number of applications in water resource management but can be difficult to obtain via conventional field methods, particularly over large spatial extents and with the kind of frequency and regularity required to support monitoring programs. Remote sensing methods could provide a viable alternative means of mapping river bathymetry (i.e., water depth). The purpose of this study was to develop and test new, spectrally based techniques for estimating water depth from satellite image data. More specifically, a neural network-based temporal ensembling approach was evaluated in comparison to several other neural network depth retrieval (NNDR) algorithms. These methods are described in a manuscript titled "Neural Network-Based Temporal Ensembling of Water Depth Estimates Derived from SuperDove Images" and the purpose of this data release is to make available the depth maps produced using these techniques. The images used as input were acquired by the SuperDove cubesats comprising the PlanetScope constellation, but the original images cannot be redistributed due to licensing restrictions; the end products derived from these images are provided instead. The large number of cubesats in the PlanetScope constellation allows for frequent temporal coverage and the neural network-based approach takes advantage of this high density time series of information by estimating depth via one of four NNDR methods described in the manuscript: 1. Mean-spec: the images are averaged over time and the resulting mean image is used as input to the NNDR. 2. Mean-depth: a separate NNDR is applied independently to each image in the time series and the resulting time series of depth estimates is averaged to obtain the final depth map. 3. NN-depth: a separate NNDR is applied independently to each image in the time series and the resulting time series of depth estimates is then used as input to a second, ensembling neural network that essentially weights the depth estimates from the individual images so as to optimize the agreement between the image-derived depth estimates and field measurements of water depth used for training; the output from the ensembling neural network serves as the final depth map. 4. Optimal single image: a separate NNDR is applied independently to each image in the time series and only the image that yields the strongest agreement between the image-derived depth estimates and the field measurements of water depth used for training is used as the final depth map. MATLAB (Version 24.1, including the Deep Learning Toolbox) source code for performing this analysis is provided in the function NN_depth_ensembling.m available on the main landing page for the data release of which this is a child item. To develop and test this new NNDR approach, the method was applied to satellite images from the Potomac River near Brunswick, MD, acquired in July and August of 2021. Field measurements of water depth available through another data release (Duda, J.M., Greise, A.J., and Young, J.A., 2020, Potomac River ADCP Bathymetric Survey, October 2019: U.S. Geological Survey data release, https://doi.org/10.5066/P9GOZZYX) were used for training and validation. The depth maps produced via each of the four methods described above are provided as GeoTIFF files, with file name suffixes that indicate the method employed: Potomac_mean-spec.tif, Potomac_mean-depth.tif, Potomac_NN-depth.tif, and Potomac-single-image.tif. The spatial resolution of the depth maps is 3 meters and the pixel values within each map are water depth estimates in units of meters.

Information on water depth in river channels is important for a number of applications in water resource management but can be difficult to obtain via conventional field methods, particularly over large spatial extents and with the kind of frequency and regularity required to support monitoring programs. Remote sensing methods could provide a viable alternative means of mapping river bathymetry (i.e., water depth). The purpose of this study was to develop and test new, spectrally based techniques for estimating water depth from satellite image data. More specifically, a neural network-based temporal ensembling approach was evaluated in comparison to several other neural network depth retrieval (NNDR) algorithms. These methods are described in a manuscript titled "Neural Network-Based Temporal Ensembling of Water Depth Estimates Derived from SuperDove Images" and the purpose of this data release is to make available the depth maps produced using these techniques. The images used as input were acquired by the SuperDove cubesats comprising the PlanetScope constellation, but the original images cannot be redistributed due to licensing restrictions; the end products derived from these images are provided instead. The large number of cubesats in the PlanetScope constellation allows for frequent temporal coverage and the neural network-based approach takes advantage of this high density time series of information by estimating depth via one of four NNDR methods described in the manuscript: 1. Mean-spec: the images are averaged over time and the resulting mean image is used as input to the NNDR. 2. Mean-depth: a separate NNDR is applied independently to each image in the time series and the resulting time series of depth estimates is averaged to obtain the final depth map. 3. NN-depth: a separate NNDR is applied independently to each image in the time series and the resulting time series of depth estimates is then used as input to a second, ensembling neural network that essentially weights the depth estimates from the individual images so as to optimize the agreement between the image-derived depth estimates and field measurements of water depth used for training; the output from the ensembling neural network serves as the final depth map. 4. Optimal single image: a separate NNDR is applied independently to each image in the time series and only the image that yields the strongest agreement between the image-derived depth estimates and the field measurements of water depth used for training is used as the final depth map. MATLAB (Version 24.1, including the Deep Learning Toolbox) source code for performing this analysis is provided in the function NN_depth_ensembling.m available on the main landing page for the data release of which this is a child item. To develop and test this new NNDR approach, the method was applied to satellite images from the Potomac River near Brunswick, MD, acquired in July and August of 2021. Field measurements of water depth available through another data release (Duda, J.M., Greise, A.J., and Young, J.A., 2020, Potomac River ADCP Bathymetric Survey, October 2019: U.S. Geological Survey data release, https://doi.org/10.5066/P9GOZZYX) were used for training and validation. The depth maps produced via each of the four methods described above are provided as GeoTIFF files, with file name suffixes that indicate the method employed: Potomac_mean-spec.tif, Potomac_mean-depth.tif, Potomac_NN-depth.tif, and Potomac-single-image.tif. The spatial resolution of the depth maps is 3 meters and the pixel values within each map are water depth estimates in units of meters.

Field measurements of water depth were acquired from a reach of the American River at Sailor Bar, near Fair Oaks, California, October 19-21, 2020, to support research on remote sensing of water depth from satellite images. The depth measurements included in this data release were obtained via two different methods: 1) By wading the shallow channel margins with RTK GPS receivers and measuring water surface elevations along the water's edge and bed elevations within the channel; depths were calculated by subtracting bed elevations from the nearest water surface elevation. 2) For the deeper areas representing most of the channel, depths were recorded along a series of cross-sections by a SonTek RiverSurveyor S5 acoustic Doppler current profiler (ADCP) and a SonarMite single beam echo sounder deployed from a boat. The spatial location of each measurement was obtained via Trimble R10 RTK GPS receivers. The map projection and datum for these data are UTM Zone 10 N and WGS84, respectively. The ADCP- and echo sounder-based depth measurements were cross-calibrated to one another using collocated observations from the two instruments and then combined with the wading measurements for the purposes of this data release, which consists of a comma-delimited (*.csv) text file with three columns: Easting_m, Northing_m, and Depth_m; the units of the spatial coordinates and the depths are meters. This ground-based depth data set was used to calibrate (i.e., train) models for inferring water depths from passive optical image data and to assess the accuracy of image-derived depth estimates.

Field measurements of water depth were acquired from a reach of the American River at Sailor Bar, near Fair Oaks, California, October 19-21, 2020, to support research on remote sensing of water depth from satellite images. The depth measurements included in this data release were obtained via two different methods: 1) By wading the shallow channel margins with RTK GPS receivers and measuring water surface elevations along the water's edge and bed elevations within the channel; depths were calculated by subtracting bed elevations from the nearest water surface elevation. 2) For the deeper areas representing most of the channel, depths were recorded along a series of cross-sections by a SonTek RiverSurveyor S5 acoustic Doppler current profiler (ADCP) and a SonarMite single beam echo sounder deployed from a boat. The spatial location of each measurement was obtained via Trimble R10 RTK GPS receivers. The map projection and datum for these data are UTM Zone 10 N and WGS84, respectively. The ADCP- and echo sounder-based depth measurements were cross-calibrated to one another using collocated observations from the two instruments and then combined with the wading measurements for the purposes of this data release, which consists of a comma-delimited (*.csv) text file with three columns: Easting_m, Northing_m, and Depth_m; the units of the spatial coordinates and the depths are meters. This ground-based depth data set was used to calibrate (i.e., train) models for inferring water depths from passive optical image data and to assess the accuracy of image-derived depth estimates.

Field measurements of water depth were acquired from a reach of the Colorado River near Lees Ferry, Arizona, March16-18, 2021, to support research on remote sensing of water depth from satellite images. The depth measurements included in this data release were obtained along a series of cross-sections using a SonTek RiverSurveyor M9 acoustic Doppler current profiler (ADCP) deployed from a boat. The spatial location of each measurement was obtained using a differential GPS included as part of the RiverSurveyor M9 ADCP instrument package. The map projection and datum for these data are UTM Zone 12S and WGS84, respectively. The USGS Qrev software program was used to ingest and process the raw ADCP data. The Qrev data file was then imported into MATLAB, which was used to create a comma-delimited (*.csv) text file with three columns: Easting_m, Northing_m, and Depth_m; the units of the spatial coordinates and the depths are meters. This ground-based depth data set was used to calibrate (i.e., train) models for inferring water depths from passive optical image data and to assess the accuracy of image-derived depth estimates.

Field measurements of water depth were acquired from a reach of the Colorado River near Lees Ferry, Arizona, March16-18, 2021, to support research on remote sensing of water depth from satellite images. The depth measurements included in this data release were obtained along a series of cross-sections using a SonTek RiverSurveyor M9 acoustic Doppler current profiler (ADCP) deployed from a boat. The spatial location of each measurement was obtained using a differential GPS included as part of the RiverSurveyor M9 ADCP instrument package. The map projection and datum for these data are UTM Zone 12S and WGS84, respectively. The USGS Qrev software program was used to ingest and process the raw ADCP data. The Qrev data file was then imported into MATLAB, which was used to create a comma-delimited (*.csv) text file with three columns: Easting_m, Northing_m, and Depth_m; the units of the spatial coordinates and the depths are meters. This ground-based depth data set was used to calibrate (i.e., train) models for inferring water depths from passive optical image data and to assess the accuracy of image-derived depth estimates.

This data release includes multispectral images and field measurements of water depth from the Sacramento River near Glenn, California, used to evaluate the potential for efficient reach-scale mapping of river bathymetry using Uncrewed Aircraft Systems (UAS). The images were acquired by a MicaSense RedEdge-MX Dual Camera deployed from a Trinity F90 vertical take-off and landing (VTOL) UAS. The 4 km long study area along the Sacramento River was subdivided into three distinct but adjacent areas of interest (AOIs) and image data were collected from one AOI each day between September 14 and 16, 2021. The image data were ortho-rectified using Quantum-Systems QBase 3D and Agisoft Metashape software and saved as GeoTIFF files. A header text file is also included with each of the images and includes information on the 10 bands of the MicaSense RedEdge-MX Dual Camera, which include: Band 1 (Coastal blue): 444 nm Band 2 (Blue): 475 nm Band 3 (Green): 531 nm Band 4 (Green): 560 nm Band 5 (Red): 650 nm Band 6 (Red): 668 nm Band 7 (Red Edge): 705 nm Band 8 (Red Edge): 717 nm Band 9 (Red Edge): 740 nm Band 10 (Near Infrared): 842 nm Each image has an associated ENVI format header text file that also contains this spectral information. The pixel values for these images are digital numbers between 0 and 65535 for all 10 spectral bands. To convert these digital numbers to reflectance values between 0 and 1, divide each pixel value in each band by 32768. To display the images as a familiar true-color composite, band 5 can be displayed as red, band 4 as green, and band 2 as blue. Water depths were measured directly in the field to calibrate relationships between depth and reflectance and to assess the accuracy of image-derived depth estimates. In shallow areas along the margins of the channel, depth measurements were made while wading with a Trimble R10 real-time kinematic (RTK) global navigation satellite system (GNSS) receiver to record spatial coordinates and riverbed elevations. Additional points along the edge of the water were surveyed to capture the water surface elevation and depths for points within the channel were calculated by subtracting the local bed elevation from the nearest water surface elevation. For deeper areas that could not be waded safely, another Trimble R10 RTK GNSS receiver was mounted on a boat with an outboard motor and connected to a Seafloor Systems SonarMite echo sounder that measured the water depth. The boat-based data were collected along a series of 25 channel-spanning cross sections and one longitudinal profile that extended over the full length of the study area. The map projection and datum for the shapefiles and GeoTIFF images contained in this data release are UTM Zone 10 N and NAD83, respectively. Each shapefile has coordinates in units of meters and the depth measurements are also in meters.