Digital Surface Models for the South Saskatchewan River, Canada

DSMs were constructed to provide information on the morphology or the river, and changes in morphology through time (e.g. due to erosion and deposition of sediment). DSMs were constructed using imagery obtained on four occasions (13th May 2015; 2nd Sept 2016; 8th June 2017; and 12th June 2017). The dataset consists of eight DSMs; one for each of the two river sections on each of the four dates.

The DSM was constructed from photogrammetric quality aerial plane images with a 0.06m ground resolution that were captured at a height of c. 1500 m from a fixed-wing aeroplane with an UltraCamXp sensor. The DSM was constructed from the images using a three stage process:

(1) An orthomosaic was generated from the aerial photography (using up to 160 images) by automated image matching using the SfM program Pix4D (https://pix4d.com/) with the 'large frame' supplementary software package. Bundle adjustment was undertaken using Pix4D’s key point matching system optimised for aerial nadir imagery. Camera properties were specified using calibration certificates and were not allowed to vary during the calculation. A point cloud was then derived following the bundle adjustment.

(2) The point cloud was checked for residual tilt using a network of Ground Control Points with coordinates that were determined by RTK differential GPS survey. The corrected point cloud density was reduced to a resolution of 0.5 m and the resulting dataset was filtered in MATLAB using a Chauvenet-type criterion. This procedure was implemented to derive the DSM in dry (non-submerged) areas of the river.

(3) In wet (submerged) areas, bed elevations were derived using a depth-brightness model as follows. First, water surface elevations at the interface between wet and dry areas of the river bed were extracted from the corrected point cloud and interpolated to generate a water surface elevation field for the submerged bed areas. Second, water depth measurements made at the time of image acquisition were located on the orthomosaic of the river, and pixel brightness values were extracted at these locations. A log-linear regression between water depth and pixel brightness was then derived, and this relationship was used to model water depth for all sub-merged bed areas. Bed elevations in these areas were determined by subtracting the predicted depth from the water surface elevation. The final DSM was derived by combining elevation values for the dry and wet bed areas.