Quantifying vegetation structure
Vegetation structure was recorded in the field by taking photographs of
the vegetation against a whiteboard applying standardised settings. The
whiteboard was made of plexiglass of size 0.25 m (width) ×1 m (height),
installed in a vertical position on its shorter edge at all viper
presence and random locations. The vegetation against the whiteboard was
photographed with a digital SLR camera (55 mm focal length and maximum
f/11 aperture) fixed at a height of 0.5 m in a distance of 4 m from the
whiteboard (Fig. 2) as in Volesky et al. (1999).
The resulting photographs were pre-processed (cropping, white
adjustment, retouching) with the GIMP 2.8.18. image editing software.
Next, we applied image processing using an automated for loopwritten in the R statistical environment (version 3.6.1., R Core Team,
2019). The script is available in Supplementary Material (SM). The
script first retrieved the images by the ‘load.image’ function of the
‘imager’ package (Barthelme, 2019), converted it to a black and white
image by the ‘grayscale’ function, and then to a binary (0-1) image by
the ‘threshold’ function of the ‘imager’ package. The resulting image
was converted to a data frame using the ‘as.data.frame’ function and the
coordinates of every image pixel covering the whiteboard were calculated
(0.25×1 m, average resolution: 1 to 1.5 megapixels
image-1). The resulting data frame had three columns
for each image, the x and y pixel coordinates (in cm) and the pixel
value (0 = white, 1 = black).
We used the data frame to calculate four variables to quantify
particular attributes of the vegetation structure. At first, leaf area,
referred to as LA hereafter, a frequently used quantity in vegetation
characterisation (Volesky, 1999), was calculated as the count of black
pixels rescaled to cm2 units. At second, we calculated
visual obstruction readings (VOR), developed primarily for prairie
vegetation based on the Robel pole method (Benkobi et al., 2000;
Vermeire & Gillen, 2001). This method takes two readings by eye at a
height of 1 m from a distance of 4 m from the pole with height
tick-marks: (i) the height at which the pole is first visible, i.e., not
obstructed by vegetation (low reading) and (ii) the maximum height
reached by the vegetation (high reading). The average of the two
readings strongly correlates with prairie phytomass (Benkobi et al.,
2000; Vermeire & Gillen, 2001). In our study, we modified the lower
reading by calculating the maximum height at which 95% of the
whiteboard is mantled by the vegetation, and we refer to this as the
height of closed vegetation (HCV) (Fig. 2) to avoid confusion with the
VOR reading terminology. We chose 95% as a threshold because glint on
some leaves in the image could return white cells and could thus reduce
the true coverage. At third, the high reading was calculated as the
maximum height of the vegetation (MHV) regardless of its width, cover,
or surface area (Fig. 2). Finally, to characterise the vertical
distribution of vegetation, we calculated foliage height diversity (FHD)
(Karr & Roth, 1971) as the Shannon diversity of the number of black
cells in each pixel row using the ‘diversity’ function of the ‘vegan’
package (Oksanen et al., 2019). Calculating FHD from values in each
pixel row circumvents the problem of arbitrarily choosing counting
intervals, e.g. ten 10-cm height intervals in each of which cover is
estimated or measured for the calculation of FHD (MacArthur et al.,
1962; Karr & Roth, 1971).