Spectral data processing
Spectrophotometric data were processed to reduce noise, remove technical
artifacts, and analyze reflectance spectra using the package pavo(Maia et al., 2013) in R v.3.6.1 (R Development Core Team, 2011). Raw
spectra were smoothed using the procspec function, which applies
the LOESS (Locally Weighted Scatterplot Smoothing) method with a
quadratic regression and a Gaussian distribution. To normalize spectra
and correct for negative values, we set the minimum value to zero and
scaled up other values accordingly using the procspec function.
We tested for repeatability of measurements taken at the same position
on the dewlap using the R package rptR (Stoffel et al., 2017).
Because all repeatability estimates were above 80%, we averaged
repeated measurements using the aggspec function in pavo .
We extracted four colorimetric variables: total brightness, cut-on
wavelength (i.e., hue), UV reflectance, and chroma. Total brightness was
calculated as the area under the “uncorrected” spectral curve from 300
to 700 nm. To determine the UV reflectance and cut-on wavelength (i.e.,
the midpoint between baseline and maximum reflectance; Cummings, 2007),
we corrected the spectra for brightness by making the area under each
curve equal to 1.0 (Endler, 1990). This correction allows for the
identification of differences in spectral shape that are unrelated to
brightness (Ng et al., 2013a).
Dewlap color composition and pattern
Each extended dewlap was photographed using a Nikon D3300 (24.2 MP)
digital camera on a white background under standardized room lighting
conditions in the lab. We included a color standard (i.e., X-rite Mini
ColorChecker® Classic) and a ruler for scale. Images were calibrated
using the colorChecker function in the R packagepatternize (Van Belleghem et al., 2018). This function calculates
a second order polynomial regression between the observed and expected
RGB (red, green, and blue) values and performs the calibration of the
image.
Because some dewlaps are a mixture of colors and the spectral data only
represent three points on the dewlap, we also wanted to quantify the
proportion of each color present in each dewlap. We determined the RGB
values of the colors present in each pixel (size ≈ 0.007 mm) of the
dewlap image using Color Inspector 3D (Barthel, 2006), a plugin for
ImageJ (Rasband, 2012). This plugin displays the distribution of colors
of an image within a 3D color space. We extracted the RGB values and
their frequency and imported them to R. We obtained a list of known
colors with their respective RGB values using the base R functionscolors and col2rgb . Then, we classified dewlap RGB values
for each pixel to color categories using Euclidean distance, which
determines the nearest known color in RGB-space. This quantitative
measure of color composition was calculated as the percent of red,
orange, and yellow present in each dewlap.
Brown anole dewlaps vary from a single color to some combination of red,
orange, and yellow-colored patches. A previous study of dewlap variation
in the native range of the brown anole categorized dewlap patterns into
two types (Driessens et al., 2017). ‘Solid’ dewlaps are uniformly
colored and may contain a distinct marginal color, such as a reddish
color covering most of the dewlap with a yellowish color along the outer
margin (Figure S2b). ‘Spotted’ dewlaps have yellowish spots scattered
across the reddish center and may also contain a yellow outer margin
(Figure S2c). We scored dewlaps from non-native populations in our study
using these same two categories to facilitate comparison to the native
range analysis in Driessens et al. (2017).