Modern computer displays and printers enable the widespread use of color in scientific communication, but the expertise for designing effective graphics has not kept pace with the technology for producing them. Historically, even the most prestigious publications have tolerated high defect rates in figures and illustrations [Cleveland, 1984], and technological advances that make creating and reproducing graphics easier do not appear to have decreased the frequency of errors. Flawed graphics consequently beget more flawed graphics as authors emulate published examples. Color has the potential to enhance communication, but design mistakes can result in color figures that are less effective than gray scale displays of the same data.
Empirical research on human subjects can build a fundamental understanding of visual perception [Ware, 2004] and scientific methods can be used to evaluate existing designs, but creating effective data graphics is a design task and not fundamentally a scientific pursuit. Like writing well, creating good data graphics requires a combination of formal knowledge and artistic sensibility tempered by experience: a combination of “substance, statistics, and design” [Tufte, 1983, p. 51].
Unlike writing, however, proficiency in creating data graphics is not a main component of secondary or postsecondary education. This article provides some concrete suggestions to help geoscientists use color more effectively when creating data graphics. The article explains factors to consider when designing for color-blind viewers, offers some example color schemes, and provides guidance for constructing and selecting color schemes in the form of design patterns. Because scientific authors frequently misuse spectral color, the particular focus here is on better alternatives to such schemes.
Designing for Color-blind Viewers
One of the commonly overlooked considerations in scientific data graphics is perception by individuals with color-deficient vision. The significance of “color-blindness” increases for geoscience publications whose readers are disproportionately male because, while 0.4% of women exhibit some form of color-vision deficiency, the figure is approximately 8% for Caucasian men. Among the predominantly male readership of Eos, as many as one in fifteen may have difficulty interpreting the rainbow hues frequently used in maps, charts, and graphs. Figure 1 simulates how one spectral color scheme, and a better alternative, may appear to color-blind readers.
Color-blind individuals see some colored data graphics quite differently from the general population. The human visual system normally perceives color through photosensitive cones in the eye that are tuned to receive wavelengths in the red, green, and blue portions of the visible spectrum. People who lack cones sensitive to one of the three wavelengths are called dichromats [Fortner and Meyer, 1997]. Individuals whose receptors are shifted toward one or the other end of the spectrum are called anomalous trichromats. The term “color deficient” encompasses dichromats and anomalous trichromats, as well as those who exhibit rarer forms of impaired color vision. Because a sex-linked recessive gene is implicated in the condition, color-vision deficiency is far more common in men than in women.
Algorithms based on psychophysical observations make it possible to simulate the appearance of colored images to color-deficient viewers [Brettel et al., 1997]. Data maps of the type shown here serve two main purposes: detection of large-scale patterns and determination of specific grid-cell or point values. The saturated spectral scale (Figure 1a) creates a region of confusion centered on the North American continent where achieving either purpose becomes nearly impossible for dichromat readers. Large negative temperature anomalies keyed to violet and blue appear on the map adjacent to large positive temperature anomalies depicted in red and orange; but to the color-deficient viewer, the hues form a continuous progression of “blue”such that they can not distinguish large positive anomalies from large negative anomalies (Figure 1b).The two-hued red and blue image (Figure 1c) displays the same map region such that both color-deficient (Figure 1d) and normally sighted readers can detect patterns and look up values.
Improving Color Schemes While Accomodating Color Deficiency
Designing effective color schemes demands attention to the needs of readers who are unable to perceive certain colors. Color schemes that accommodate red or green-blind dichromats will accommodate most other forms of color deficiency [Rigden, 2002]. By designing with the severest forms of red and green colorblindness in mind, authors can create data graphics that work for all readers.
The following suggestions can help authors make rainbow-colored graphics accessible to more of their readers and can be used to improve both spectral and nonspectral color schemes.
- Avoid the use of spectral schemes to represent sequential data because the spectral order of visible light carries no inherent magnitude message. Readers do not automatically perceive violet as greater than red even though the two colors occupy opposite ends of the color spectrum. Rainbow color schemes are therefore not appropriate if the data to be mapped or graphed represent a distribution of values ranging from low to high. With suitable modification, however [Brewer, 1997], spectral schemes can work for continuously distributed diverging data, such as anomalies and residuals (Figure 2c) and for the display of categorical data (Figure 2e).
- Use yellow with care and avoid yellow-green colors altogether in spectral schemes. Readers with color-deficient vision often confuse yellow-green with orange colors. Yellow appears brightest among the primary colors and stands out visually for color-impaired and normally sighted readers alike. The yellow portion of the scheme should therefore be aligned with the midpoint of the data distribution if emphasizing the midpoint of a diverging distribution is a clear goal of the presentation. However, yellow may lead to misperceptions when the midpoint critical value of the data is not significant to the presentation or is not precisely known.
- Use color intensity (or value) to reinforce hue as a visual indicator of magnitude. Hue is what we typically refer to as color; red, blue, green, and orange are all hues. Intensity or value may also be referred to as lightness, brightness, or luminosity. While it is possible to select hue sequences that are distinguishable by individuals with either color-deficient or normal vision, intensity readily provides perceptual ordering for all readers. Using intensity as well as hue also makes the quality of color reproduction less critical to the presentation and helps make even gray scale photocopies somewhat legible.
The results of human-subject studies may inform designers, but design skill comes primarily from experience and example. Design patterns [Alexander, 1979; Gamma et al., 1995] for data graphics can communicate experience to nonexpert designers and may help integrate scientific knowledge into the design process. Design patterns make well-proven design knowledge explicit using a specific literal form: “Each pattern is a three-part rule, which expresses a relation between a certain context, a problem, and a solution” [Alexander, 1979, p. 247]. Table 1 documents two design patterns for the use of color in data graphics. Every design pattern carries a descriptive name (in bold italic in Table 1) and contains instructions to the designer for creating a specific instance of the pattern. Collectively, the names form a shared vocabulary for communicating about designs.
Example Color Schemes
Computer graphics tools make changing color palettes easy, but devising new color schemes for effective data graphics remains a challenging design task. The color schemes in Figure 2 have been tried and tested for print and online display and evaluated for legibility by readers with color-deficient vision. They should work well with a range of Earth science data. Accurate reproduction of color for print and computer displays is a complex problem in its own right. Computer displays typically reproduce colors using an additive (RGB) color model, while print reproduction usually uses a subtractive (CMYK) color model [Fortner and Meyer, 1997]. Detailed specifications for reproducing these and other color schemes using both RGB and CMYK models are available on the Web site: http://geography.uoregon.edu/datagraphics/. Others are provided by Brewer et al. .
The authors gratefully acknowledge the contributions of Aileen Buckley, Peter Killoran, Sarah Shafer, and Jacqueline Shinker to this work, and the improvements to the manuscript suggested by Cindy Brewer and Henry Shaw. National Centers for Environment Prediction (NCEP) reanalysis data were provided by the National Oceanic and Atmospheric Administration- Cooperative Institute for Research and Environmental Sciences’ Climate Diagnostics Center, Boulder, Colorado (http://www.cdc.noaa.gov/). Research was supported by U.S. National Science Foundation grant ATM-9910638.
Alexander, C. (1979), The Timeless Way of Building, Oxford Univ. Press, New York.
Brettel, H., F. Viénot, and J. D. Mollon (1997), Computerized simulation of color appearance for dichromats, J. Opt. Soc. Am.,14(10), 2647–2655.
Brewer, C. A. (1996), Guidelines for selecting colors for diverging schemes on maps, Cartogr. J., 33(2), 79-86.
Brewer, C. A. (1997), Spectral schemes: controversial color use on maps, Cartogr. Geogr. Inf. Syst., 24(4), 203–220.
Brewer, C. A., G. W. Hatchard, and M. A. Harrower (2003), Colorbrewer in print: A catalog of color schemes for maps, Cartogr. Geogr. Inf. Sci., 30, 5–32. (http:// www.ColorBrewer.org).
Cleveland, W. S. (1984), Graphs in scientific publications, Am. Stat., 38(4), 261–269.
Fortner, B., and T. H. Meyer (1997), Number by Colors: A Guide to Using Color to Understand Technical Data, 349 pp., Springer-Verlag, New York.
Gamma, E., R. Helm, R. Johnson, and J. Vlissides (1995), Design Patterns: Elements of Reusable Object-Oriented Software, 395 pp., Addison-Wesley-Longman, Reading, Mass.
Rigden, C. (2002), Now you see it, now you don’t, IEEE Comput., 35(7), 104–105.
Tufte, E. R. (1983), The Visual Display of Quantitative Information, 197 pp., Graphics Press, Cheshire, Conn.
Ware, C. (2004), Information Visualization: Perception for Design, 486 pp., Morgan Kaufmann, Burlington, Mass.
Adam Light and Patrick J. Bartlein, University of Oregon, Eugene
For additional information, contact P. J. Bartlein; E-mail: firstname.lastname@example.org.
Citation: Light, A., and P. J. Bartlein (2004), The end of the rainbow? Color schemes for improved data graphics, Eos, 85(40), doi:10.1029/2004EO400002. Published on 5 October 2004.
© 2004. American Geophysical Union. All rights reserved.
© 2004. American Geophysical Union. All rights reserved.