How color determines what we see.
Color plays an important role in the analysis and communication of scientific information.
Over a century ago, Albert H. Munsell drew on the work of Isaac Newton and Johann Wolfgang von Goethe to compose our modern concept of color mapping Munsel 2015. Munsell’s research produced the first perceptually ordered color space, a three-dimensional graph in which the axes represent hue (color), value (light or dark), and chroma (color intensity) Moreland 2016.
In the early 1900s, Hermann Grassmann’s theory of linear algebra deciphered abstract mathematics, revealing the origami-like properties of the higher dimensions. Grassmann thus created the concept of vector space, allowing the approximate calculation of the perceived color within a defined area (Grassmann, 1844). The study of color no longer depended on approximation, but could be coded numerically, plotted along a parabola. This level of precision is necessary because color perception is a subjective experience dependent on light, simultaneous contrast (the phenomenon of juxtaposed colors affecting the appearance of each color), and rod photoreceptors and cones within the viewer’s eyes (Albers, 2006; Itten, 1970).
“The language is inherently biased, but through visualization, we can let the data speak for itself,” Phillip Wolfram, a modeler of the Earth system and computational fluid dynamics at Los Alamos National Laboratory. At Los Alamos, data visualizations are as ubiquitous as the sage brush that stitches the nearby desert. Every day, teams of experts fight, represent and color-code strips of data for interpretation with Earth and the computer scientists in the lab. Choosing colors to represent various properties of your data — a step from responsive, iterative process to hasty post-thinking, is the final barrier between painstaking data collection and well-anticipated analysis and discovery.
Most visualization software comes equipped with color maps, a selection of standard color-coding gradients that researchers can, in seconds, apply to display and evaluate their data. But not all data visualizations are created equal, and despite the proliferation of literature denouncing standard maps such as the traditional rainbow color map [eg, Borland and Taylor, 2007], they permeate visualizations from basic bar charts to complex representations of biogeochemical data.
At AGU’s Fall 2019 Meeting in San Francisco, California, last December, row upon row of posters in the convention center’s great main hall featured the same standard, bright-colored maps that adorn displays of temperature scales, chlorophyll concentrations, land elevations, and a host of other data.
“The same colormap applied to a wide range of data gets monotonous and confusing,” said Rick Saltus, a senior research scientist with the Cooperative Institute for Research in Environmental Sciences (CIRES) at the University of Colorado Boulder. “You are trying to communicate effectively and efficiently, and that is impeded by presenting a variety of concepts to the viewer, all illustrated using identical color assignments.”
Color researchers and visualization experts around the world are working to change this status quo. Several groups are developing new tools to help scientists imagine increasingly complex data sets more accurately and intuitively, and with greater fidelity, using context as a guide to ensure a proper balance between hue, luminance, and saturation.
Topography of the Filchner-Ronne and Ross ice shelves in West Antarctica, the image illustrates the effects of simultaneous contrast within the traditional rainbow color map (left), increased details in a desaturated version of the traditional rainbow (center), and an analogous color palette for aesthetic quality as well as discriminatory power (right).
Francesca Samsel, a research scientist at the Texas Advanced Computing Center (TACC ) at the University of Texas at Austin, and her team describe the needs of scientists regarding data visualization using three categories (feature identification, exploration, and communication), as well as subcategories to identify outliers and determine relationships. When interpreting a large data set, scientists often look for specific characteristics (for example, the direction of flow of ocean currents or the water temperatures at certain locations) within known data ranges; o they are exploring the data holistically to make general observations; or they are looking to communicate specific properties of the data to colleagues, colleagues or the public. Occasionally, a researcher may be interested only in outliers, or in the way one variable affects another, for example, how does the water temperature change when two streams meet?
Visualization is how scientists interact with the quantitative results of their research and how they make arguments based on data, Wolfram said. He regularly uses visualizations, including aerial graphs, that represent the view from above (eg, from satellites) to explore Earth’s systems and climate data in his work at Los Alamos. “What I’m looking for is closely tied to the question of science,” he said. “I’m usually trying to understand geospatial relationships, particularly in airframe data, for surface features like (ocean) eddies.” Advanced color mapping tools help avoid significant loss of feature detail in the data that Wolfram analyzes, unlike standard maps like Jet.
Samsel et al. Argue that data non-dependent color mapping strategies can, in fact, perpetuate bias if the hues are not arranged in a familiar order (e.g. rainbow order), if the luminance is not specifically accounted for, or if several gradients are not organized within a map for a specific data set [Borland and Taylor, 2007]. The importance of visualization throughout the investigation process requires high-level tools that maximize information and minimize obstruction caused by any of these color-related issues. Researchers Samsel, Schloss, and Danielle Szafir, assistant professor and director of VisuaLab at the University of Colorado Boulder, have taken concrete steps to create these types of tools, while maintaining the goal of intuitive operation.
Starting in the 1990s, perceptual researchers established [Ware et al., 2019; Rogowitz, 2013] that the human eye shows greater sensitivity to differences in luminance (perceived brightness) than hue when distinguishing patterns within densely packed data points, findings that Samsel and colleagues have reaffirmed, in part through Theory of color. Using this information in conjunction with hue and saturation gradations, Samsel created linear (a hue gradient), divergent (two hue gradients meeting in the middle), and what she calls “wave” color maps, maps that traverse the luminance distribution across many shades. “This creates a higher density of contrast throughout the map, which resolves many more features on continuous data,” said Samsel.
Samsel, who was trained and worked for 25 years as an artist before pivoting to visualization, uses her knowledge of color theory to address the perceptual challenges of color mapping. “We have discovered in the course of our research that presenting scientists with perceptually matched color maps is not always the most beneficial solution” for providing the kind of resolution that scientists need within their data, she said Samsel. There are nuances and complexity in the interaction of color that affect a viewer’s associative response and how they derive the relative importance of different characteristics within the data, she explained.
Samsel’s research prompted the creation of ColorMoves [Samsel et al., 2018], an applied tool to interactively adjust color maps to suit the needs of different data sets. The online interface of the tool provides sets of maps focused on achieving greater discriminatory power [Ware et al., 2019] while reflecting the palette of the natural world, something called “semantic association”. For example, many people associate blue with ocean data and green with land data.
The ColorMoves interface allows users to place multiple colormaps on their data and adjust the result interactively, seeing the changes in their data in real time and assigning hue and contrast whenever appropriate. “It was not necessarily intended as a tool for data exploration, but scientists have identified it as a priority” and have been using it for exploration, Samsel said.
Schloss, who directs the Visual Reasoning Laboratory at the University of Wisconsin-Madison, focuses on making visual communication more effective and efficient through the study of color cognition, targeting trade-offs in color mapping between high contrast and aestheticism. “People see personalization as a great asset for creating visualizations,” she said, “and I think making that ability easily accessible can encourage people to be more careful in how they are color-coding and presenting their data.”
The Colorgorical tool features a series of adjustable sliders to customize a palette based on the perceptual distance between tones, the differences between the tone names, how close the tones are on the color wheel, and the specificity of the tone names ( for example, peacock or sapphire versus blue) (Heer and Stone, 2012). Users can also select a hue and luminance range within which the custom palette should fall, and they can build a color palette around a “seed color”.
Scientific visualizations serve multiple functions to help people quantify, interpret, evaluate, and communicate information. Its importance in data exploration and discovery is immense and it is growing with advances in computational power, yet the single most efficient encoder of visualization, color, remains very little studied. While the macroscale effects of such neglect on public perception of issues like climate change are still unknown, many groups are beginning to devote greater resources to designing their displays with the public in mind.
Ed Hawkins, lead author of the upcoming sixth report from the Intergovernmental Research Group on Climate Change (IPCC) and creator of the now-viral “Warming Stripes” graphic, said the IPCC pays multiple graphic designers to transform complex visualizations into simplified graphics. “We have to be able to reach a very wide audience,” Hawkins added. “Not only for general communication, but also to inform political decisions and help people respond to risks that threaten their way of life.” Hawkins and his team spend “a lot of time” focusing on issues of color blindness, as well as the readability and semantic understanding of color.
Considering the difference in perception between published reports and those who read them is imperative, according to Lyn Bartram, professor at the School of Interactive Art and Technology at Simon Fraser University and director of the Vancouver Institute for Visual Analytics. “Instead of engaging people in the experience of understanding big data, we just throw the facts out there and wash our hands of it,” she said. “Data visualization is a language, and fundamentally you are trying to tell a story. Color is a big part of it.”
References.
Albers, J. (2006), Interaction of Color, Yale Univ. Press, New Haven, Conn.
Ware, C., et al. (2019), Measuring and modeling the feature detection threshold functions of colormaps, IEEE Trans. Visualization Comput. Gráficos, 25(9), 2.777–2790
