An Experimental Evaluation of Kernel Density Estimation to Choose Categorical Map Colours

Abstract

ABSTRACT When selecting categorical map colours, colour conventions should be respected to leverage semantic–colour resonance to facilitate cartographic communication. Given a set of sample colours, kernel density estimation (KDE) can be used to estimate each colour's probability density (appropriateness) to represent the category. How to couple bandwidth and kernel to estimate better appropriateness remains unknown. To fill this gap, an experiment was designed to explore best pairs of bandwidth and kernel capturing users' assessments. We gathered six groups of colour samples from 10 well-accepted land use atlases and 30 randomly sampled test colours; we then applied KDE to estimate the appropriateness of test colours using all possible pairs of bandwidth and kernel, and invited participants to score each test colour. Results show that pair of rule-of-thumb bandwidth and Gaussian kernel yields the best estimates. Our findings are generalizable to diverse colours and can serve as a complement to design colours.