Image analysis
To remove non-linearity of the camera response, we converted all raw
photos to linearised TIFF format images using Dcraw (Coffin 2008). Then
we rescaled each colour channel to ensure that the 99 % reflectance
standard had the corresponding value. After this process, RGB values of
each pixel (ranging from 0 – 255 for each colour channel) scale
linearly with reflectance of 0 - 100 %. We used the B channel for
ultraviolet photos, R, G, and B channels for visible photos, and R and B
channels for near-infrared photos because the camera sensors have
sensitivity in the corresponding spectral range.
For all images, we measured mean channel values of three body regions:
thorax-abdomen (both of which are important for thermoregulation due to
haemolymph circulation between them (Rawlins 1980); we refer to thorax
for brevity), basal wings, and entire wings. Thus, we measured six
different body regions for each specimen: dorsal thorax (DT), dorsal
basal wings (DB), dorsal entire wings (DE), ventral thorax (VT), ventral
basal wings (VB), and ventral entire wings (VE). Because butterfly wings
were symmetrical, we measured wing regions only from one side (left).
For each region, we averaged the pixel values of all measured channels
to get a mean reflectance value. Then we calculated the mean reflectance
of each body part over 1) VIS (320 – 700 nm; animal-visible range), 2)
NIR (700 – 1050 nm; animal-invisible near-infrared range), and 3)
VIS-NIR (320 – 1050 nm) ranges. For example, VIS reflectance was
calculated by (60u + 360v )/(60 + 360), where u andv are the mean reflectance of ultraviolet and visible-range
images respectively, and the numbers 60 and 360 are wavebands of the
spectral images. We measured the entire wing area for each species
(averaged among specimens) and used this as a size index for each
species.