2.2 UAV Very High Resolution imagery acquisition and processing
Wetlands are difficult features to map from the ground because of the
complexity of the landscape, especially during months when they are
flooded. In order to map a wetland with accuracy it is necessary to use
an aerial perspective (orthogonal projection, angle of 90º), preferably
three dimensional and combined with field experiments to fully
understanding the character and spatial extent of this phenomenon
(Madden et al., 2015). Considering the hydrodynamic characterization of
the wetland surface, we used remote sensing and geoprocessing tools in
the study area, including photogrammetry by Unmanned Aerial Vehicle
(UAV) very high resolution (VHR) images, providing satisfactory results
for the calculation of the area and the volume of water that the wetland
could, theoretically, store.
The images were obtained with
fixed-wing UAV: model eBee Plus RTK/PPK; wingspan 110 cm; weight 1.1 kg;
powered by electricity; radio link range ~ 3 km;
cruising speed 40–110 km/h (24.8–68.3 mph); wind resistance up to 45
km/h (72.45 mph); maximum flight time ~ 50 minutes, with
no need for ground control points (GCP) in real-time kinematic (RTK) or
post-processed kinematic (PPK) mode; and absolute and relative accuracy
of X, Y, Z coordinates (RTK-/-PPK) of up to 3 cm, with a flight altitude
of 120 meters. For imaging, we used the optical sensor RGB SenseFly
S.O.D.A. (Sensor Optimised for Drone Applications), with 20 megapixels.
The flight plan and the correction of the images in PPK mode were done
in Emotion Software 3.0, by SenseFly (Figure 2 A). The predicted
cross-flight lateral and longitudinal overlap was 70%, ensuring high
data accuracy by obtaining a large number of images and covering the
target with overlaps in two different directions (d’Oleire-Oltmanns,
Marzolff, Peter & Ries, 2012), resulting in better digital surface
models (DSM) for the relief analysis. During processing, we were able to
overlap an average of nine images of the area where photogrammetry was
performed (Figure 2 B).
Next, we performed planialtimetric correction using GNSS data installed
in the field during flyovers, which allowed us to obtain post-processing
through the Precise Point Positioning service of the Brazilian Institute
of Geography and Statistics (IBGE-PPP, www.ibge.gov.br). This operation
performs a correction to PPP mode using GPS RINEX (Receiver Independent
Exchange Format) data (Matsuoka, de Azambuja, Souza & Veronez, 2009).
Our fieldwork equipment was the GNSS Topcon Hiper PRO RTK (GPS/GLONASS
dual receiver), used only for the base unit in operation for 04 hours of
point recording. The rover unit was onboard eBee Plus RTK/PPK (Figure
3).
The coordinates corrected by GNSS have significantly greater accuracy
than the conventional GPS onboard the UAV. The PPK processing greatly
increased the model’s reliability, resulting in a precision of 0.038 m,
against 6.220 m with the common GPS (Table 1).
The last step was to process all aerial photographs and GNSS data in the
Agisoft Metashape software and perform the digital photogrammetry. This
computational operation aligned all the images, creating an orthomosaic.
Processing proceeded with the tie point step (generated 5.716 points),
followed by the generation of dense cloud (15.699.250 points) and the
creation of the 3D model (1.046.616 faces). The digital surface model
(6677x5328) and the orthomosaic (19623x16048) were then processed and
readied for classification. Finally, after the classification of image
data, the contours of the wetland’s compartments were created
(shapefile) and their area and volume calculated. In order to identify
each compartment of the wetland, the digital surface model and the
orthomosaic data were processed in a GIS platform. The processing
consisted in tests of percentage declivity of the surface and by visual
interpretation of the relief. Thus, the Minimum bounding geometry,
Erase, Multipart to singlepart and dissolve tools were applied
to define the exact declivity of the murundus, which enabled the
differentiation of three declivity-density distinct compartments.
To calculate the volume, Agisoft Metashape performs interpolation and
value integration using the Best Fit Plane (BFP) tool, by the generation
of a plane between each of the compartments and the calculation of the
volume below that plane. The plane is generated from automatic analysis
of the topographic perimeter profiles of the compartments (Figure 4).
The generated plane minimizes the root mean square (RMS) of the
orthogonal distance between each vertex of its polygon.
Considering that soil saturation and the elevation of the water level
above the land surface occur first in the center of the topographic
depression and the cumulative saturation increases towards the border,
we used the Best Fit Plane tool to establish an average water height
limit for each compartment based in field observations, in order to
produce a more reliable volume estimate.
The vegetation does not influence the estimated volume considerably,
since wetlands consist largely of grassy area. Therefore, it was not
necessary to generate a digital terrain model; a digital surface model
was adequate for all calculations. The volume of each compartment is
directly influenced by the elevations of the perimeter, by the number of
mounds, and by the terrain’s unique morphology. In order to subtract the
space occupied by the mounds it was necessary to quantify them. The
processing is performed through the slope classes that better
represented the murundus contour of the study area. Percentage slope
tests were performed with indices of 10%, 15%, 20% and 25%. The
slope index that came closest to the correct mapping was 20%.