2.2 Drone survey and geospatial analysis
To develop a topographic digital surface model (DSM) of the study area
using photogrammetry, a drone photographic survey was undertaken using a
Phantom Potensic T25 drone equipped with internal GPS and a 1080p GPS
digital RGB camera with an optimized 120° wide-angle lens. The
flightpath was set to take photographs from an altitude of 400m, reduced
to 150 m over hill top areas that were obscured by cloud cover. The
flightpath was set so that images had greater than 65% overlap. Twelve
ground-based GPS reference points were made along the boundaries of the
designated area to permit vertical alignment of relative elevations
derived from the drone survey. A Digital Elevation Model (DEM) was
created using photogrammetric routines within Agisoft Metashape whereby
building and tree interference was removed from the DSM by excluding
points that had a steep (>80°) angle with adjacent point.
The red, green and blue bands of a compile orthomosaic aerial image of
the study area were used as the input for a maximum likelihood
classification using the QGIS 3.4 saga tool Supervised Classification
for Grids’. Training data were generated through heads up digitisation
of that aerial photograph. The output of this process
successfully differentiated gullies and rills from other forms of ground
cover but could not distinguish between the different vegetation
communities present within the catchment. This high-resolution mapping
of soil erosion features was therefore supplemented with a
conventional aerial photograph interpretation exercise carried out at a
scale of 1:3000. In addition to the classification of broad land cover
types within the study area, cultivated land use was subject to a
further level of classification according to the direction of
cultivation. This categorisation was undertaken as a simple aerial image
interpretation (at a scale 1:500) with discontinuities in land
management practices being recorded as separate features (land parcels)
within the GIS layer.
The 1m DEM of the site was analysed in QGIS 3.10 to produce a model of
potential surface flow, slope aspect and slope degree. In order to
simplify subsequent analyses, the aspect of each 1m pixel within the
elevation model were categorised into one of the 8 cardinal directions
(N, NE, E, SE, S, SW, W, NW) These DEM derivatives were used as inputs
in the following analyses. Mean slope degree and modal aspect were
calculated for each parcel of cultivated land identified in the aerial
photograph interpretation using QGis’s Zonal statistics tool. The
direction of cultivation recorded in the aerial imagery interpretation
was then compared with the modal aspect for each land use parcel and the
relationship between the two described using these qualitative classes:
cultivation along contours, cultivation with slope and intermediate
cultivation – where the direction of cultivation had no strong
relationship the aspect of the field. The modelled surface flow was used
for two purposes, firstly to create a Strahler ordered model of
potential overland flow convergence within the study area, and secondly
to select those bare ground (interpreted as eroded) pixels within the
landscape that were probably generated by surface flow i.e. rills and
gully features (in the main visible in high resolution photography). The
selected pixels were then converted into point data and their density
expressed as area (m2) per m2calculated to derive an independent index of observed erosion feature
density.