I'm interested in effective methods to visualize and subsequently analyze long-term acoustic datasets, including those from disparate datasets. Are there preferred color schemes that work well for color blindness and intuitive interpretation?
$\begingroup$
$\endgroup$
Add a comment
|
3 Answers
$\begingroup$
$\endgroup$
A labmate of mine has had good luck with long-term false color spectrograms, following the protocol of this paper from Michael Towsey et al.
I've also done a fair bit of simply showing loess-smoothed time series of various acoustic indices, using the geom_smooth() function from ggplot() in R. It can take a bit of fussing to find a span to apply to the smoother to get it looking good but it can be pretty useful.
$\begingroup$
$\endgroup$
I'd also recommend another paper from the QUT group that @dtsavage mentions: Phillips et al (2018) "Revealing the ecological content of long- duration audio-recordings of the environment through clustering and visualisation" PLoS ONE https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0193345
They show a range of interesting options for visualising long-duration recordings.
I really like long-duration false-colour spectrograms, especially for data exploration, but unfortunately they may not always be colour-blind friendly. Most rely on 3 channel Red-Green-Blue combinations to depict the spectro-temporal differences. You might need to create custom functions to force use of a colour-blind friendly palette, especially if you are hoping to directly compare spectrograms from different sources. The colours are assigned through relative rather than absolute scores of the 3 indices. So two e.g. red areas in a pair of spectrograms may not necessarily reflect the same soundscape patterns.
$\begingroup$
$\endgroup$
As the OP indicates there are two aspects
- compression of long data sets into a smaller dataset that is easier to analyse
- 'color-scheme' that support analysis and are suited for color blindness.
Adapting from spectrogram, one could have time on x-axis, and relevant categories (may be in the order of frequency) on y-axis For The compression, you must decide if the parameter can be smoothed and you are only interested in mean value (background noise, wind, rain, ships, etc) than you smooth/average, or you are interested in occurrence of events (calls, clicks, etc.) then you keep always the maximal value.
Concerning color and color-blindness, it still seems appropriate to use gray-scale or at least diverging colomaps that work both on screen and on paper.
