For me, the most important consideration when choosing a colour palette is which one will best help reinforce the message I want to convey through my visualisation? The Seaborn documentation on choosing color palettes puts it like this:
... color palette choices are about more than aesthetics: the colors you
choose can reveal patterns in your data if used effectively or hide them if
used poorly. There is not one optimal palette, but there are palettes that
are better or worse for particular datasets and visualization approaches.
There are three main categories of colour palettes, choosing the correct one is important:
- qualitative - good for showing categorical data.
- sequential - good for showing ranges of data that follow a sequential scale.
- divergent - good for showing ranges of data that span a midpoint and both end points are interesting.
Where possible, it's also good to pick colours that are related to the data you are visualising - e.g. use blue for water, green for vegetation, etc.
As well as the Seaborn documentation mentioned previously, this series of articles from NASA's Earth Observatory is worth a read. It contains a lot of information about how we perceive colour and how to use colour effectively in visualisations.